From dafebc5bb70303f0b5baf0b087cf4d9a64b5c7f0 Mon Sep 17 00:00:00 2001 From: marha Date: Mon, 12 Sep 2011 11:27:51 +0200 Subject: Synchronised line endinge with release branch --- mesalib/src/mesa/program/arbprogparse.c | 432 +- mesalib/src/mesa/program/hash_table.c | 418 +- mesalib/src/mesa/program/hash_table.h | 314 +- mesalib/src/mesa/program/ir_to_mesa.cpp | 6914 ++++++++++++------------ mesalib/src/mesa/program/prog_instruction.h | 908 ++-- mesalib/src/mesa/program/prog_optimize.c | 2732 +++++----- mesalib/src/mesa/program/prog_print.c | 2192 ++++---- mesalib/src/mesa/program/prog_statevars.c | 2416 ++++----- mesalib/src/mesa/program/prog_statevars.h | 296 +- mesalib/src/mesa/program/program.c | 2160 ++++---- mesalib/src/mesa/program/program_parse.y | 5612 +++++++++---------- mesalib/src/mesa/program/program_parse_extra.c | 530 +- mesalib/src/mesa/program/program_parser.h | 602 +-- mesalib/src/mesa/program/programopt.c | 1370 ++--- mesalib/src/mesa/program/programopt.h | 110 +- mesalib/src/mesa/program/register_allocate.c | 1116 ++-- mesalib/src/mesa/program/register_allocate.h | 148 +- mesalib/src/mesa/program/sampler.cpp | 274 +- 18 files changed, 14272 insertions(+), 14272 deletions(-) (limited to 'mesalib/src/mesa/program') diff --git a/mesalib/src/mesa/program/arbprogparse.c b/mesalib/src/mesa/program/arbprogparse.c index bca033477..dffc8abf7 100644 --- a/mesalib/src/mesa/program/arbprogparse.c +++ b/mesalib/src/mesa/program/arbprogparse.c @@ -1,216 +1,216 @@ -/* - * Mesa 3-D graphics library - * Version: 7.1 - * - * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - -#define DEBUG_PARSING 0 - -/** - * \file arbprogparse.c - * ARB_*_program parser core - * \author Karl Rasche - */ - -/** -Notes on program parameters, etc. - -The instructions we emit will use six kinds of source registers: - - PROGRAM_INPUT - input registers - PROGRAM_TEMPORARY - temp registers - PROGRAM_ADDRESS - address/indirect register - PROGRAM_SAMPLER - texture sampler - PROGRAM_CONSTANT - indexes into program->Parameters, a known constant/literal - PROGRAM_STATE_VAR - indexes into program->Parameters, and may actually be: - + a state variable, like "state.fog.color", or - + a pointer to a "program.local[k]" parameter, or - + a pointer to a "program.env[k]" parameter - -Basically, all the program.local[] and program.env[] values will get mapped -into the unified gl_program->Parameters array. This solves the problem of -having three separate program parameter arrays. -*/ - - -#include "main/glheader.h" -#include "main/imports.h" -#include "main/context.h" -#include "main/mtypes.h" -#include "arbprogparse.h" -#include "programopt.h" -#include "prog_parameter.h" -#include "prog_statevars.h" -#include "prog_instruction.h" -#include "program_parser.h" - - -void -_mesa_parse_arb_fragment_program(struct gl_context* ctx, GLenum target, - const GLvoid *str, GLsizei len, - struct gl_fragment_program *program) -{ - struct gl_program prog; - struct asm_parser_state state; - GLuint i; - - ASSERT(target == GL_FRAGMENT_PROGRAM_ARB); - - memset(&prog, 0, sizeof(prog)); - memset(&state, 0, sizeof(state)); - state.prog = &prog; - - if (!_mesa_parse_arb_program(ctx, target, (const GLubyte*) str, len, - &state)) { - /* Error in the program. Just return. */ - return; - } - - if (program->Base.String != NULL) - free(program->Base.String); - - /* Copy the relevant contents of the arb_program struct into the - * fragment_program struct. - */ - program->Base.String = prog.String; - program->Base.NumInstructions = prog.NumInstructions; - program->Base.NumTemporaries = prog.NumTemporaries; - program->Base.NumParameters = prog.NumParameters; - program->Base.NumAttributes = prog.NumAttributes; - program->Base.NumAddressRegs = prog.NumAddressRegs; - program->Base.NumNativeInstructions = prog.NumNativeInstructions; - program->Base.NumNativeTemporaries = prog.NumNativeTemporaries; - program->Base.NumNativeParameters = prog.NumNativeParameters; - program->Base.NumNativeAttributes = prog.NumNativeAttributes; - program->Base.NumNativeAddressRegs = prog.NumNativeAddressRegs; - program->Base.NumAluInstructions = prog.NumAluInstructions; - program->Base.NumTexInstructions = prog.NumTexInstructions; - program->Base.NumTexIndirections = prog.NumTexIndirections; - program->Base.NumNativeAluInstructions = prog.NumAluInstructions; - program->Base.NumNativeTexInstructions = prog.NumTexInstructions; - program->Base.NumNativeTexIndirections = prog.NumTexIndirections; - program->Base.InputsRead = prog.InputsRead; - program->Base.OutputsWritten = prog.OutputsWritten; - program->Base.IndirectRegisterFiles = prog.IndirectRegisterFiles; - for (i = 0; i < MAX_TEXTURE_IMAGE_UNITS; i++) { - program->Base.TexturesUsed[i] = prog.TexturesUsed[i]; - if (prog.TexturesUsed[i]) - program->Base.SamplersUsed |= (1 << i); - } - program->Base.ShadowSamplers = prog.ShadowSamplers; - program->OriginUpperLeft = state.option.OriginUpperLeft; - program->PixelCenterInteger = state.option.PixelCenterInteger; - - program->UsesKill = state.fragment.UsesKill; - - if (program->Base.Instructions) - free(program->Base.Instructions); - program->Base.Instructions = prog.Instructions; - - if (program->Base.Parameters) - _mesa_free_parameter_list(program->Base.Parameters); - program->Base.Parameters = prog.Parameters; - - /* Append fog instructions now if the program has "OPTION ARB_fog_exp" - * or similar. We used to leave this up to drivers, but it appears - * there's no hardware that wants to do fog in a discrete stage separate - * from the fragment shader. - */ - if (state.option.Fog != OPTION_NONE) { - static const GLenum fog_modes[4] = { - GL_NONE, GL_EXP, GL_EXP2, GL_LINEAR - }; - - /* XXX: we should somehow recompile this to remove clamping if disabled - * On the ATI driver, this is unclampled if fragment clamping is disabled - */ - _mesa_append_fog_code(ctx, program, fog_modes[state.option.Fog], GL_TRUE); - } - -#if DEBUG_FP - printf("____________Fragment program %u ________\n", program->Base.Id); - _mesa_print_program(&program->Base); -#endif -} - - - -/** - * Parse the vertex program string. If success, update the given - * vertex_program object with the new program. Else, leave the vertex_program - * object unchanged. - */ -void -_mesa_parse_arb_vertex_program(struct gl_context *ctx, GLenum target, - const GLvoid *str, GLsizei len, - struct gl_vertex_program *program) -{ - struct gl_program prog; - struct asm_parser_state state; - - ASSERT(target == GL_VERTEX_PROGRAM_ARB); - - memset(&prog, 0, sizeof(prog)); - memset(&state, 0, sizeof(state)); - state.prog = &prog; - - if (!_mesa_parse_arb_program(ctx, target, (const GLubyte*) str, len, - &state)) { - _mesa_error(ctx, GL_INVALID_OPERATION, "glProgramString(bad program)"); - return; - } - - if (program->Base.String != NULL) - free(program->Base.String); - - /* Copy the relevant contents of the arb_program struct into the - * vertex_program struct. - */ - program->Base.String = prog.String; - program->Base.NumInstructions = prog.NumInstructions; - program->Base.NumTemporaries = prog.NumTemporaries; - program->Base.NumParameters = prog.NumParameters; - program->Base.NumAttributes = prog.NumAttributes; - program->Base.NumAddressRegs = prog.NumAddressRegs; - program->Base.NumNativeInstructions = prog.NumNativeInstructions; - program->Base.NumNativeTemporaries = prog.NumNativeTemporaries; - program->Base.NumNativeParameters = prog.NumNativeParameters; - program->Base.NumNativeAttributes = prog.NumNativeAttributes; - program->Base.NumNativeAddressRegs = prog.NumNativeAddressRegs; - program->Base.InputsRead = prog.InputsRead; - program->Base.OutputsWritten = prog.OutputsWritten; - program->Base.IndirectRegisterFiles = prog.IndirectRegisterFiles; - program->IsPositionInvariant = (state.option.PositionInvariant) - ? GL_TRUE : GL_FALSE; - - if (program->Base.Instructions) - free(program->Base.Instructions); - program->Base.Instructions = prog.Instructions; - - if (program->Base.Parameters) - _mesa_free_parameter_list(program->Base.Parameters); - program->Base.Parameters = prog.Parameters; - -#if DEBUG_VP - printf("____________Vertex program %u __________\n", program->Base.Id); - _mesa_print_program(&program->Base); -#endif -} +/* + * Mesa 3-D graphics library + * Version: 7.1 + * + * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +#define DEBUG_PARSING 0 + +/** + * \file arbprogparse.c + * ARB_*_program parser core + * \author Karl Rasche + */ + +/** +Notes on program parameters, etc. + +The instructions we emit will use six kinds of source registers: + + PROGRAM_INPUT - input registers + PROGRAM_TEMPORARY - temp registers + PROGRAM_ADDRESS - address/indirect register + PROGRAM_SAMPLER - texture sampler + PROGRAM_CONSTANT - indexes into program->Parameters, a known constant/literal + PROGRAM_STATE_VAR - indexes into program->Parameters, and may actually be: + + a state variable, like "state.fog.color", or + + a pointer to a "program.local[k]" parameter, or + + a pointer to a "program.env[k]" parameter + +Basically, all the program.local[] and program.env[] values will get mapped +into the unified gl_program->Parameters array. This solves the problem of +having three separate program parameter arrays. +*/ + + +#include "main/glheader.h" +#include "main/imports.h" +#include "main/context.h" +#include "main/mtypes.h" +#include "arbprogparse.h" +#include "programopt.h" +#include "prog_parameter.h" +#include "prog_statevars.h" +#include "prog_instruction.h" +#include "program_parser.h" + + +void +_mesa_parse_arb_fragment_program(struct gl_context* ctx, GLenum target, + const GLvoid *str, GLsizei len, + struct gl_fragment_program *program) +{ + struct gl_program prog; + struct asm_parser_state state; + GLuint i; + + ASSERT(target == GL_FRAGMENT_PROGRAM_ARB); + + memset(&prog, 0, sizeof(prog)); + memset(&state, 0, sizeof(state)); + state.prog = &prog; + + if (!_mesa_parse_arb_program(ctx, target, (const GLubyte*) str, len, + &state)) { + /* Error in the program. Just return. */ + return; + } + + if (program->Base.String != NULL) + free(program->Base.String); + + /* Copy the relevant contents of the arb_program struct into the + * fragment_program struct. + */ + program->Base.String = prog.String; + program->Base.NumInstructions = prog.NumInstructions; + program->Base.NumTemporaries = prog.NumTemporaries; + program->Base.NumParameters = prog.NumParameters; + program->Base.NumAttributes = prog.NumAttributes; + program->Base.NumAddressRegs = prog.NumAddressRegs; + program->Base.NumNativeInstructions = prog.NumNativeInstructions; + program->Base.NumNativeTemporaries = prog.NumNativeTemporaries; + program->Base.NumNativeParameters = prog.NumNativeParameters; + program->Base.NumNativeAttributes = prog.NumNativeAttributes; + program->Base.NumNativeAddressRegs = prog.NumNativeAddressRegs; + program->Base.NumAluInstructions = prog.NumAluInstructions; + program->Base.NumTexInstructions = prog.NumTexInstructions; + program->Base.NumTexIndirections = prog.NumTexIndirections; + program->Base.NumNativeAluInstructions = prog.NumAluInstructions; + program->Base.NumNativeTexInstructions = prog.NumTexInstructions; + program->Base.NumNativeTexIndirections = prog.NumTexIndirections; + program->Base.InputsRead = prog.InputsRead; + program->Base.OutputsWritten = prog.OutputsWritten; + program->Base.IndirectRegisterFiles = prog.IndirectRegisterFiles; + for (i = 0; i < MAX_TEXTURE_IMAGE_UNITS; i++) { + program->Base.TexturesUsed[i] = prog.TexturesUsed[i]; + if (prog.TexturesUsed[i]) + program->Base.SamplersUsed |= (1 << i); + } + program->Base.ShadowSamplers = prog.ShadowSamplers; + program->OriginUpperLeft = state.option.OriginUpperLeft; + program->PixelCenterInteger = state.option.PixelCenterInteger; + + program->UsesKill = state.fragment.UsesKill; + + if (program->Base.Instructions) + free(program->Base.Instructions); + program->Base.Instructions = prog.Instructions; + + if (program->Base.Parameters) + _mesa_free_parameter_list(program->Base.Parameters); + program->Base.Parameters = prog.Parameters; + + /* Append fog instructions now if the program has "OPTION ARB_fog_exp" + * or similar. We used to leave this up to drivers, but it appears + * there's no hardware that wants to do fog in a discrete stage separate + * from the fragment shader. + */ + if (state.option.Fog != OPTION_NONE) { + static const GLenum fog_modes[4] = { + GL_NONE, GL_EXP, GL_EXP2, GL_LINEAR + }; + + /* XXX: we should somehow recompile this to remove clamping if disabled + * On the ATI driver, this is unclampled if fragment clamping is disabled + */ + _mesa_append_fog_code(ctx, program, fog_modes[state.option.Fog], GL_TRUE); + } + +#if DEBUG_FP + printf("____________Fragment program %u ________\n", program->Base.Id); + _mesa_print_program(&program->Base); +#endif +} + + + +/** + * Parse the vertex program string. If success, update the given + * vertex_program object with the new program. Else, leave the vertex_program + * object unchanged. + */ +void +_mesa_parse_arb_vertex_program(struct gl_context *ctx, GLenum target, + const GLvoid *str, GLsizei len, + struct gl_vertex_program *program) +{ + struct gl_program prog; + struct asm_parser_state state; + + ASSERT(target == GL_VERTEX_PROGRAM_ARB); + + memset(&prog, 0, sizeof(prog)); + memset(&state, 0, sizeof(state)); + state.prog = &prog; + + if (!_mesa_parse_arb_program(ctx, target, (const GLubyte*) str, len, + &state)) { + _mesa_error(ctx, GL_INVALID_OPERATION, "glProgramString(bad program)"); + return; + } + + if (program->Base.String != NULL) + free(program->Base.String); + + /* Copy the relevant contents of the arb_program struct into the + * vertex_program struct. + */ + program->Base.String = prog.String; + program->Base.NumInstructions = prog.NumInstructions; + program->Base.NumTemporaries = prog.NumTemporaries; + program->Base.NumParameters = prog.NumParameters; + program->Base.NumAttributes = prog.NumAttributes; + program->Base.NumAddressRegs = prog.NumAddressRegs; + program->Base.NumNativeInstructions = prog.NumNativeInstructions; + program->Base.NumNativeTemporaries = prog.NumNativeTemporaries; + program->Base.NumNativeParameters = prog.NumNativeParameters; + program->Base.NumNativeAttributes = prog.NumNativeAttributes; + program->Base.NumNativeAddressRegs = prog.NumNativeAddressRegs; + program->Base.InputsRead = prog.InputsRead; + program->Base.OutputsWritten = prog.OutputsWritten; + program->Base.IndirectRegisterFiles = prog.IndirectRegisterFiles; + program->IsPositionInvariant = (state.option.PositionInvariant) + ? GL_TRUE : GL_FALSE; + + if (program->Base.Instructions) + free(program->Base.Instructions); + program->Base.Instructions = prog.Instructions; + + if (program->Base.Parameters) + _mesa_free_parameter_list(program->Base.Parameters); + program->Base.Parameters = prog.Parameters; + +#if DEBUG_VP + printf("____________Vertex program %u __________\n", program->Base.Id); + _mesa_print_program(&program->Base); +#endif +} diff --git a/mesalib/src/mesa/program/hash_table.c b/mesalib/src/mesa/program/hash_table.c index 3e9b9d4d3..877a9e2ff 100644 --- a/mesalib/src/mesa/program/hash_table.c +++ b/mesalib/src/mesa/program/hash_table.c @@ -1,209 +1,209 @@ -/* - * Copyright © 2008 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER - * DEALINGS IN THE SOFTWARE. - */ - -/** - * \file hash_table.c - * \brief Implementation of a generic, opaque hash table data type. - * - * \author Ian Romanick - */ - -#include "main/imports.h" -#include "main/simple_list.h" -#include "hash_table.h" - -struct node { - struct node *next; - struct node *prev; -}; - -struct hash_table { - hash_func_t hash; - hash_compare_func_t compare; - - unsigned num_buckets; - struct node buckets[1]; -}; - - -struct hash_node { - struct node link; - const void *key; - void *data; -}; - - -struct hash_table * -hash_table_ctor(unsigned num_buckets, hash_func_t hash, - hash_compare_func_t compare) -{ - struct hash_table *ht; - unsigned i; - - - if (num_buckets < 16) { - num_buckets = 16; - } - - ht = malloc(sizeof(*ht) + ((num_buckets - 1) - * sizeof(ht->buckets[0]))); - if (ht != NULL) { - ht->hash = hash; - ht->compare = compare; - ht->num_buckets = num_buckets; - - for (i = 0; i < num_buckets; i++) { - make_empty_list(& ht->buckets[i]); - } - } - - return ht; -} - - -void -hash_table_dtor(struct hash_table *ht) -{ - hash_table_clear(ht); - free(ht); -} - - -void -hash_table_clear(struct hash_table *ht) -{ - struct node *node; - struct node *temp; - unsigned i; - - - for (i = 0; i < ht->num_buckets; i++) { - foreach_s(node, temp, & ht->buckets[i]) { - remove_from_list(node); - free(node); - } - - assert(is_empty_list(& ht->buckets[i])); - } -} - - -void * -hash_table_find(struct hash_table *ht, const void *key) -{ - const unsigned hash_value = (*ht->hash)(key); - const unsigned bucket = hash_value % ht->num_buckets; - struct node *node; - - foreach(node, & ht->buckets[bucket]) { - struct hash_node *hn = (struct hash_node *) node; - - if ((*ht->compare)(hn->key, key) == 0) { - return hn->data; - } - } - - return NULL; -} - - -void -hash_table_insert(struct hash_table *ht, void *data, const void *key) -{ - const unsigned hash_value = (*ht->hash)(key); - const unsigned bucket = hash_value % ht->num_buckets; - struct hash_node *node; - - node = calloc(1, sizeof(*node)); - - node->data = data; - node->key = key; - - insert_at_head(& ht->buckets[bucket], & node->link); -} - -void -hash_table_remove(struct hash_table *ht, const void *key) -{ - const unsigned hash_value = (*ht->hash)(key); - const unsigned bucket = hash_value % ht->num_buckets; - struct node *node; - - foreach(node, & ht->buckets[bucket]) { - struct hash_node *hn = (struct hash_node *) node; - - if ((*ht->compare)(hn->key, key) == 0) { - remove_from_list(node); - free(node); - return; - } - } -} - -void -hash_table_call_foreach(struct hash_table *ht, - void (*callback)(const void *key, - void *data, - void *closure), - void *closure) -{ - int bucket; - - for (bucket = 0; bucket < ht->num_buckets; bucket++) { - struct node *node, *temp; - foreach_s(node, temp, &ht->buckets[bucket]) { - struct hash_node *hn = (struct hash_node *) node; - - callback(hn->key, hn->data, closure); - } - } -} - -unsigned -hash_table_string_hash(const void *key) -{ - const char *str = (const char *) key; - unsigned hash = 5381; - - - while (*str != '\0') { - hash = (hash * 33) + *str; - str++; - } - - return hash; -} - - -unsigned -hash_table_pointer_hash(const void *key) -{ - return (unsigned)((uintptr_t) key / sizeof(void *)); -} - - -int -hash_table_pointer_compare(const void *key1, const void *key2) -{ - return key1 == key2 ? 0 : 1; -} +/* + * Copyright © 2008 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +/** + * \file hash_table.c + * \brief Implementation of a generic, opaque hash table data type. + * + * \author Ian Romanick + */ + +#include "main/imports.h" +#include "main/simple_list.h" +#include "hash_table.h" + +struct node { + struct node *next; + struct node *prev; +}; + +struct hash_table { + hash_func_t hash; + hash_compare_func_t compare; + + unsigned num_buckets; + struct node buckets[1]; +}; + + +struct hash_node { + struct node link; + const void *key; + void *data; +}; + + +struct hash_table * +hash_table_ctor(unsigned num_buckets, hash_func_t hash, + hash_compare_func_t compare) +{ + struct hash_table *ht; + unsigned i; + + + if (num_buckets < 16) { + num_buckets = 16; + } + + ht = malloc(sizeof(*ht) + ((num_buckets - 1) + * sizeof(ht->buckets[0]))); + if (ht != NULL) { + ht->hash = hash; + ht->compare = compare; + ht->num_buckets = num_buckets; + + for (i = 0; i < num_buckets; i++) { + make_empty_list(& ht->buckets[i]); + } + } + + return ht; +} + + +void +hash_table_dtor(struct hash_table *ht) +{ + hash_table_clear(ht); + free(ht); +} + + +void +hash_table_clear(struct hash_table *ht) +{ + struct node *node; + struct node *temp; + unsigned i; + + + for (i = 0; i < ht->num_buckets; i++) { + foreach_s(node, temp, & ht->buckets[i]) { + remove_from_list(node); + free(node); + } + + assert(is_empty_list(& ht->buckets[i])); + } +} + + +void * +hash_table_find(struct hash_table *ht, const void *key) +{ + const unsigned hash_value = (*ht->hash)(key); + const unsigned bucket = hash_value % ht->num_buckets; + struct node *node; + + foreach(node, & ht->buckets[bucket]) { + struct hash_node *hn = (struct hash_node *) node; + + if ((*ht->compare)(hn->key, key) == 0) { + return hn->data; + } + } + + return NULL; +} + + +void +hash_table_insert(struct hash_table *ht, void *data, const void *key) +{ + const unsigned hash_value = (*ht->hash)(key); + const unsigned bucket = hash_value % ht->num_buckets; + struct hash_node *node; + + node = calloc(1, sizeof(*node)); + + node->data = data; + node->key = key; + + insert_at_head(& ht->buckets[bucket], & node->link); +} + +void +hash_table_remove(struct hash_table *ht, const void *key) +{ + const unsigned hash_value = (*ht->hash)(key); + const unsigned bucket = hash_value % ht->num_buckets; + struct node *node; + + foreach(node, & ht->buckets[bucket]) { + struct hash_node *hn = (struct hash_node *) node; + + if ((*ht->compare)(hn->key, key) == 0) { + remove_from_list(node); + free(node); + return; + } + } +} + +void +hash_table_call_foreach(struct hash_table *ht, + void (*callback)(const void *key, + void *data, + void *closure), + void *closure) +{ + int bucket; + + for (bucket = 0; bucket < ht->num_buckets; bucket++) { + struct node *node, *temp; + foreach_s(node, temp, &ht->buckets[bucket]) { + struct hash_node *hn = (struct hash_node *) node; + + callback(hn->key, hn->data, closure); + } + } +} + +unsigned +hash_table_string_hash(const void *key) +{ + const char *str = (const char *) key; + unsigned hash = 5381; + + + while (*str != '\0') { + hash = (hash * 33) + *str; + str++; + } + + return hash; +} + + +unsigned +hash_table_pointer_hash(const void *key) +{ + return (unsigned)((uintptr_t) key / sizeof(void *)); +} + + +int +hash_table_pointer_compare(const void *key1, const void *key2) +{ + return key1 == key2 ? 0 : 1; +} diff --git a/mesalib/src/mesa/program/hash_table.h b/mesalib/src/mesa/program/hash_table.h index 3910d0624..e715bb1cc 100644 --- a/mesalib/src/mesa/program/hash_table.h +++ b/mesalib/src/mesa/program/hash_table.h @@ -1,157 +1,157 @@ -/* - * Copyright © 2008 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER - * DEALINGS IN THE SOFTWARE. - */ - -/** - * \file hash_table.h - * \brief Implementation of a generic, opaque hash table data type. - * - * \author Ian Romanick - */ - -#ifndef HASH_TABLE_H -#define HASH_TABLE_H - -struct hash_table; - -typedef unsigned (*hash_func_t)(const void *key); -typedef int (*hash_compare_func_t)(const void *key1, const void *key2); - -#ifdef __cplusplus -extern "C" { -#endif - -/** - * Hash table constructor - * - * Creates a hash table with the specified number of buckets. The supplied - * \c hash and \c compare routines are used when adding elements to the table - * and when searching for elements in the table. - * - * \param num_buckets Number of buckets (bins) in the hash table. - * \param hash Function used to compute hash value of input keys. - * \param compare Function used to compare keys. - */ -extern struct hash_table *hash_table_ctor(unsigned num_buckets, - hash_func_t hash, hash_compare_func_t compare); - - -/** - * Release all memory associated with a hash table - * - * \warning - * This function cannot release memory occupied either by keys or data. - */ -extern void hash_table_dtor(struct hash_table *ht); - - -/** - * Flush all entries from a hash table - * - * \param ht Table to be cleared of its entries. - */ -extern void hash_table_clear(struct hash_table *ht); - - -/** - * Search a hash table for a specific element - * - * \param ht Table to be searched - * \param key Key of the desired element - * - * \return - * The \c data value supplied to \c hash_table_insert when the element with - * the matching key was added. If no matching key exists in the table, - * \c NULL is returned. - */ -extern void *hash_table_find(struct hash_table *ht, const void *key); - - -/** - * Add an element to a hash table - */ -extern void hash_table_insert(struct hash_table *ht, void *data, - const void *key); - -/** - * Remove a specific element from a hash table. - */ -extern void hash_table_remove(struct hash_table *ht, const void *key); - -/** - * Compute hash value of a string - * - * Computes the hash value of a string using the DJB2 algorithm developed by - * Professor Daniel J. Bernstein. It was published on comp.lang.c once upon - * a time. I was unable to find the original posting in the archives. - * - * \param key Pointer to a NUL terminated string to be hashed. - * - * \sa hash_table_string_compare - */ -extern unsigned hash_table_string_hash(const void *key); - - -/** - * Compare two strings used as keys - * - * This is just a macro wrapper around \c strcmp. - * - * \sa hash_table_string_hash - */ -#define hash_table_string_compare ((hash_compare_func_t) strcmp) - - -/** - * Compute hash value of a pointer - * - * \param key Pointer to be used as a hash key - * - * \note - * The memory pointed to by \c key is \b never accessed. The value of \c key - * itself is used as the hash key - * - * \sa hash_table_pointer_compare - */ -unsigned -hash_table_pointer_hash(const void *key); - - -/** - * Compare two pointers used as keys - * - * \sa hash_table_pointer_hash - */ -int -hash_table_pointer_compare(const void *key1, const void *key2); - -void -hash_table_call_foreach(struct hash_table *ht, - void (*callback)(const void *key, - void *data, - void *closure), - void *closure); - -#ifdef __cplusplus -} -#endif -#endif /* HASH_TABLE_H */ +/* + * Copyright © 2008 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +/** + * \file hash_table.h + * \brief Implementation of a generic, opaque hash table data type. + * + * \author Ian Romanick + */ + +#ifndef HASH_TABLE_H +#define HASH_TABLE_H + +struct hash_table; + +typedef unsigned (*hash_func_t)(const void *key); +typedef int (*hash_compare_func_t)(const void *key1, const void *key2); + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * Hash table constructor + * + * Creates a hash table with the specified number of buckets. The supplied + * \c hash and \c compare routines are used when adding elements to the table + * and when searching for elements in the table. + * + * \param num_buckets Number of buckets (bins) in the hash table. + * \param hash Function used to compute hash value of input keys. + * \param compare Function used to compare keys. + */ +extern struct hash_table *hash_table_ctor(unsigned num_buckets, + hash_func_t hash, hash_compare_func_t compare); + + +/** + * Release all memory associated with a hash table + * + * \warning + * This function cannot release memory occupied either by keys or data. + */ +extern void hash_table_dtor(struct hash_table *ht); + + +/** + * Flush all entries from a hash table + * + * \param ht Table to be cleared of its entries. + */ +extern void hash_table_clear(struct hash_table *ht); + + +/** + * Search a hash table for a specific element + * + * \param ht Table to be searched + * \param key Key of the desired element + * + * \return + * The \c data value supplied to \c hash_table_insert when the element with + * the matching key was added. If no matching key exists in the table, + * \c NULL is returned. + */ +extern void *hash_table_find(struct hash_table *ht, const void *key); + + +/** + * Add an element to a hash table + */ +extern void hash_table_insert(struct hash_table *ht, void *data, + const void *key); + +/** + * Remove a specific element from a hash table. + */ +extern void hash_table_remove(struct hash_table *ht, const void *key); + +/** + * Compute hash value of a string + * + * Computes the hash value of a string using the DJB2 algorithm developed by + * Professor Daniel J. Bernstein. It was published on comp.lang.c once upon + * a time. I was unable to find the original posting in the archives. + * + * \param key Pointer to a NUL terminated string to be hashed. + * + * \sa hash_table_string_compare + */ +extern unsigned hash_table_string_hash(const void *key); + + +/** + * Compare two strings used as keys + * + * This is just a macro wrapper around \c strcmp. + * + * \sa hash_table_string_hash + */ +#define hash_table_string_compare ((hash_compare_func_t) strcmp) + + +/** + * Compute hash value of a pointer + * + * \param key Pointer to be used as a hash key + * + * \note + * The memory pointed to by \c key is \b never accessed. The value of \c key + * itself is used as the hash key + * + * \sa hash_table_pointer_compare + */ +unsigned +hash_table_pointer_hash(const void *key); + + +/** + * Compare two pointers used as keys + * + * \sa hash_table_pointer_hash + */ +int +hash_table_pointer_compare(const void *key1, const void *key2); + +void +hash_table_call_foreach(struct hash_table *ht, + void (*callback)(const void *key, + void *data, + void *closure), + void *closure); + +#ifdef __cplusplus +} +#endif +#endif /* HASH_TABLE_H */ diff --git a/mesalib/src/mesa/program/ir_to_mesa.cpp b/mesalib/src/mesa/program/ir_to_mesa.cpp index 29975150b..69a84de39 100644 --- a/mesalib/src/mesa/program/ir_to_mesa.cpp +++ b/mesalib/src/mesa/program/ir_to_mesa.cpp @@ -1,3457 +1,3457 @@ -/* - * Copyright (C) 2005-2007 Brian Paul All Rights Reserved. - * Copyright (C) 2008 VMware, Inc. All Rights Reserved. - * Copyright © 2010 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER - * DEALINGS IN THE SOFTWARE. - */ - -/** - * \file ir_to_mesa.cpp - * - * Translate GLSL IR to Mesa's gl_program representation. - */ - -#include -#include "main/compiler.h" -#include "ir.h" -#include "ir_visitor.h" -#include "ir_print_visitor.h" -#include "ir_expression_flattening.h" -#include "glsl_types.h" -#include "glsl_parser_extras.h" -#include "../glsl/program.h" -#include "ir_optimization.h" -#include "ast.h" - -extern "C" { -#include "main/mtypes.h" -#include "main/shaderapi.h" -#include "main/shaderobj.h" -#include "main/uniforms.h" -#include "program/hash_table.h" -#include "program/prog_instruction.h" -#include "program/prog_optimize.h" -#include "program/prog_print.h" -#include "program/program.h" -#include "program/prog_uniform.h" -#include "program/prog_parameter.h" -#include "program/sampler.h" -} - -class src_reg; -class dst_reg; - -static int swizzle_for_size(int size); - -/** - * This struct is a corresponding struct to Mesa prog_src_register, with - * wider fields. - */ -class src_reg { -public: - src_reg(gl_register_file file, int index, const glsl_type *type) - { - this->file = file; - this->index = index; - if (type && (type->is_scalar() || type->is_vector() || type->is_matrix())) - this->swizzle = swizzle_for_size(type->vector_elements); - else - this->swizzle = SWIZZLE_XYZW; - this->negate = 0; - this->reladdr = NULL; - } - - src_reg() - { - this->file = PROGRAM_UNDEFINED; - this->index = 0; - this->swizzle = 0; - this->negate = 0; - this->reladdr = NULL; - } - - explicit src_reg(dst_reg reg); - - gl_register_file file; /**< PROGRAM_* from Mesa */ - int index; /**< temporary index, VERT_ATTRIB_*, FRAG_ATTRIB_*, etc. */ - GLuint swizzle; /**< SWIZZLE_XYZWONEZERO swizzles from Mesa. */ - int negate; /**< NEGATE_XYZW mask from mesa */ - /** Register index should be offset by the integer in this reg. */ - src_reg *reladdr; -}; - -class dst_reg { -public: - dst_reg(gl_register_file file, int writemask) - { - this->file = file; - this->index = 0; - this->writemask = writemask; - this->cond_mask = COND_TR; - this->reladdr = NULL; - } - - dst_reg() - { - this->file = PROGRAM_UNDEFINED; - this->index = 0; - this->writemask = 0; - this->cond_mask = COND_TR; - this->reladdr = NULL; - } - - explicit dst_reg(src_reg reg); - - gl_register_file file; /**< PROGRAM_* from Mesa */ - int index; /**< temporary index, VERT_ATTRIB_*, FRAG_ATTRIB_*, etc. */ - int writemask; /**< Bitfield of WRITEMASK_[XYZW] */ - GLuint cond_mask:4; - /** Register index should be offset by the integer in this reg. */ - src_reg *reladdr; -}; - -src_reg::src_reg(dst_reg reg) -{ - this->file = reg.file; - this->index = reg.index; - this->swizzle = SWIZZLE_XYZW; - this->negate = 0; - this->reladdr = reg.reladdr; -} - -dst_reg::dst_reg(src_reg reg) -{ - this->file = reg.file; - this->index = reg.index; - this->writemask = WRITEMASK_XYZW; - this->cond_mask = COND_TR; - this->reladdr = reg.reladdr; -} - -class ir_to_mesa_instruction : public exec_node { -public: - /* Callers of this ralloc-based new need not call delete. It's - * easier to just ralloc_free 'ctx' (or any of its ancestors). */ - static void* operator new(size_t size, void *ctx) - { - void *node; - - node = rzalloc_size(ctx, size); - assert(node != NULL); - - return node; - } - - enum prog_opcode op; - dst_reg dst; - src_reg src[3]; - /** Pointer to the ir source this tree came from for debugging */ - ir_instruction *ir; - GLboolean cond_update; - bool saturate; - int sampler; /**< sampler index */ - int tex_target; /**< One of TEXTURE_*_INDEX */ - GLboolean tex_shadow; - - class function_entry *function; /* Set on OPCODE_CAL or OPCODE_BGNSUB */ -}; - -class variable_storage : public exec_node { -public: - variable_storage(ir_variable *var, gl_register_file file, int index) - : file(file), index(index), var(var) - { - /* empty */ - } - - gl_register_file file; - int index; - ir_variable *var; /* variable that maps to this, if any */ -}; - -class function_entry : public exec_node { -public: - ir_function_signature *sig; - - /** - * identifier of this function signature used by the program. - * - * At the point that Mesa instructions for function calls are - * generated, we don't know the address of the first instruction of - * the function body. So we make the BranchTarget that is called a - * small integer and rewrite them during set_branchtargets(). - */ - int sig_id; - - /** - * Pointer to first instruction of the function body. - * - * Set during function body emits after main() is processed. - */ - ir_to_mesa_instruction *bgn_inst; - - /** - * Index of the first instruction of the function body in actual - * Mesa IR. - * - * Set after convertion from ir_to_mesa_instruction to prog_instruction. - */ - int inst; - - /** Storage for the return value. */ - src_reg return_reg; -}; - -class ir_to_mesa_visitor : public ir_visitor { -public: - ir_to_mesa_visitor(); - ~ir_to_mesa_visitor(); - - function_entry *current_function; - - struct gl_context *ctx; - struct gl_program *prog; - struct gl_shader_program *shader_program; - struct gl_shader_compiler_options *options; - - int next_temp; - - variable_storage *find_variable_storage(ir_variable *var); - - function_entry *get_function_signature(ir_function_signature *sig); - - src_reg get_temp(const glsl_type *type); - void reladdr_to_temp(ir_instruction *ir, src_reg *reg, int *num_reladdr); - - src_reg src_reg_for_float(float val); - - /** - * \name Visit methods - * - * As typical for the visitor pattern, there must be one \c visit method for - * each concrete subclass of \c ir_instruction. Virtual base classes within - * the hierarchy should not have \c visit methods. - */ - /*@{*/ - virtual void visit(ir_variable *); - virtual void visit(ir_loop *); - virtual void visit(ir_loop_jump *); - virtual void visit(ir_function_signature *); - virtual void visit(ir_function *); - virtual void visit(ir_expression *); - virtual void visit(ir_swizzle *); - virtual void visit(ir_dereference_variable *); - virtual void visit(ir_dereference_array *); - virtual void visit(ir_dereference_record *); - virtual void visit(ir_assignment *); - virtual void visit(ir_constant *); - virtual void visit(ir_call *); - virtual void visit(ir_return *); - virtual void visit(ir_discard *); - virtual void visit(ir_texture *); - virtual void visit(ir_if *); - /*@}*/ - - src_reg result; - - /** List of variable_storage */ - exec_list variables; - - /** List of function_entry */ - exec_list function_signatures; - int next_signature_id; - - /** List of ir_to_mesa_instruction */ - exec_list instructions; - - ir_to_mesa_instruction *emit(ir_instruction *ir, enum prog_opcode op); - - ir_to_mesa_instruction *emit(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, src_reg src0); - - ir_to_mesa_instruction *emit(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, src_reg src0, src_reg src1); - - ir_to_mesa_instruction *emit(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, - src_reg src0, src_reg src1, src_reg src2); - - /** - * Emit the correct dot-product instruction for the type of arguments - */ - ir_to_mesa_instruction * emit_dp(ir_instruction *ir, - dst_reg dst, - src_reg src0, - src_reg src1, - unsigned elements); - - void emit_scalar(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, src_reg src0); - - void emit_scalar(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, src_reg src0, src_reg src1); - - void emit_scs(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, const src_reg &src); - - bool try_emit_mad(ir_expression *ir, - int mul_operand); - bool try_emit_mad_for_and_not(ir_expression *ir, - int mul_operand); - bool try_emit_sat(ir_expression *ir); - - void emit_swz(ir_expression *ir); - - bool process_move_condition(ir_rvalue *ir); - - void copy_propagate(void); - - void *mem_ctx; -}; - -src_reg undef_src = src_reg(PROGRAM_UNDEFINED, 0, NULL); - -dst_reg undef_dst = dst_reg(PROGRAM_UNDEFINED, SWIZZLE_NOOP); - -dst_reg address_reg = dst_reg(PROGRAM_ADDRESS, WRITEMASK_X); - -static int -swizzle_for_size(int size) -{ - int size_swizzles[4] = { - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y), - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z), - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W), - }; - - assert((size >= 1) && (size <= 4)); - return size_swizzles[size - 1]; -} - -ir_to_mesa_instruction * -ir_to_mesa_visitor::emit(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, - src_reg src0, src_reg src1, src_reg src2) -{ - ir_to_mesa_instruction *inst = new(mem_ctx) ir_to_mesa_instruction(); - int num_reladdr = 0; - - /* If we have to do relative addressing, we want to load the ARL - * reg directly for one of the regs, and preload the other reladdr - * sources into temps. - */ - num_reladdr += dst.reladdr != NULL; - num_reladdr += src0.reladdr != NULL; - num_reladdr += src1.reladdr != NULL; - num_reladdr += src2.reladdr != NULL; - - reladdr_to_temp(ir, &src2, &num_reladdr); - reladdr_to_temp(ir, &src1, &num_reladdr); - reladdr_to_temp(ir, &src0, &num_reladdr); - - if (dst.reladdr) { - emit(ir, OPCODE_ARL, address_reg, *dst.reladdr); - num_reladdr--; - } - assert(num_reladdr == 0); - - inst->op = op; - inst->dst = dst; - inst->src[0] = src0; - inst->src[1] = src1; - inst->src[2] = src2; - inst->ir = ir; - - inst->function = NULL; - - this->instructions.push_tail(inst); - - return inst; -} - - -ir_to_mesa_instruction * -ir_to_mesa_visitor::emit(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, src_reg src0, src_reg src1) -{ - return emit(ir, op, dst, src0, src1, undef_src); -} - -ir_to_mesa_instruction * -ir_to_mesa_visitor::emit(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, src_reg src0) -{ - assert(dst.writemask != 0); - return emit(ir, op, dst, src0, undef_src, undef_src); -} - -ir_to_mesa_instruction * -ir_to_mesa_visitor::emit(ir_instruction *ir, enum prog_opcode op) -{ - return emit(ir, op, undef_dst, undef_src, undef_src, undef_src); -} - -ir_to_mesa_instruction * -ir_to_mesa_visitor::emit_dp(ir_instruction *ir, - dst_reg dst, src_reg src0, src_reg src1, - unsigned elements) -{ - static const gl_inst_opcode dot_opcodes[] = { - OPCODE_DP2, OPCODE_DP3, OPCODE_DP4 - }; - - return emit(ir, dot_opcodes[elements - 2], dst, src0, src1); -} - -/** - * Emits Mesa scalar opcodes to produce unique answers across channels. - * - * Some Mesa opcodes are scalar-only, like ARB_fp/vp. The src X - * channel determines the result across all channels. So to do a vec4 - * of this operation, we want to emit a scalar per source channel used - * to produce dest channels. - */ -void -ir_to_mesa_visitor::emit_scalar(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, - src_reg orig_src0, src_reg orig_src1) -{ - int i, j; - int done_mask = ~dst.writemask; - - /* Mesa RCP is a scalar operation splatting results to all channels, - * like ARB_fp/vp. So emit as many RCPs as necessary to cover our - * dst channels. - */ - for (i = 0; i < 4; i++) { - GLuint this_mask = (1 << i); - ir_to_mesa_instruction *inst; - src_reg src0 = orig_src0; - src_reg src1 = orig_src1; - - if (done_mask & this_mask) - continue; - - GLuint src0_swiz = GET_SWZ(src0.swizzle, i); - GLuint src1_swiz = GET_SWZ(src1.swizzle, i); - for (j = i + 1; j < 4; j++) { - /* If there is another enabled component in the destination that is - * derived from the same inputs, generate its value on this pass as - * well. - */ - if (!(done_mask & (1 << j)) && - GET_SWZ(src0.swizzle, j) == src0_swiz && - GET_SWZ(src1.swizzle, j) == src1_swiz) { - this_mask |= (1 << j); - } - } - src0.swizzle = MAKE_SWIZZLE4(src0_swiz, src0_swiz, - src0_swiz, src0_swiz); - src1.swizzle = MAKE_SWIZZLE4(src1_swiz, src1_swiz, - src1_swiz, src1_swiz); - - inst = emit(ir, op, dst, src0, src1); - inst->dst.writemask = this_mask; - done_mask |= this_mask; - } -} - -void -ir_to_mesa_visitor::emit_scalar(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, src_reg src0) -{ - src_reg undef = undef_src; - - undef.swizzle = SWIZZLE_XXXX; - - emit_scalar(ir, op, dst, src0, undef); -} - -/** - * Emit an OPCODE_SCS instruction - * - * The \c SCS opcode functions a bit differently than the other Mesa (or - * ARB_fragment_program) opcodes. Instead of splatting its result across all - * four components of the destination, it writes one value to the \c x - * component and another value to the \c y component. - * - * \param ir IR instruction being processed - * \param op Either \c OPCODE_SIN or \c OPCODE_COS depending on which - * value is desired. - * \param dst Destination register - * \param src Source register - */ -void -ir_to_mesa_visitor::emit_scs(ir_instruction *ir, enum prog_opcode op, - dst_reg dst, - const src_reg &src) -{ - /* Vertex programs cannot use the SCS opcode. - */ - if (this->prog->Target == GL_VERTEX_PROGRAM_ARB) { - emit_scalar(ir, op, dst, src); - return; - } - - const unsigned component = (op == OPCODE_SIN) ? 0 : 1; - const unsigned scs_mask = (1U << component); - int done_mask = ~dst.writemask; - src_reg tmp; - - assert(op == OPCODE_SIN || op == OPCODE_COS); - - /* If there are compnents in the destination that differ from the component - * that will be written by the SCS instrution, we'll need a temporary. - */ - if (scs_mask != unsigned(dst.writemask)) { - tmp = get_temp(glsl_type::vec4_type); - } - - for (unsigned i = 0; i < 4; i++) { - unsigned this_mask = (1U << i); - src_reg src0 = src; - - if ((done_mask & this_mask) != 0) - continue; - - /* The source swizzle specified which component of the source generates - * sine / cosine for the current component in the destination. The SCS - * instruction requires that this value be swizzle to the X component. - * Replace the current swizzle with a swizzle that puts the source in - * the X component. - */ - unsigned src0_swiz = GET_SWZ(src.swizzle, i); - - src0.swizzle = MAKE_SWIZZLE4(src0_swiz, src0_swiz, - src0_swiz, src0_swiz); - for (unsigned j = i + 1; j < 4; j++) { - /* If there is another enabled component in the destination that is - * derived from the same inputs, generate its value on this pass as - * well. - */ - if (!(done_mask & (1 << j)) && - GET_SWZ(src0.swizzle, j) == src0_swiz) { - this_mask |= (1 << j); - } - } - - if (this_mask != scs_mask) { - ir_to_mesa_instruction *inst; - dst_reg tmp_dst = dst_reg(tmp); - - /* Emit the SCS instruction. - */ - inst = emit(ir, OPCODE_SCS, tmp_dst, src0); - inst->dst.writemask = scs_mask; - - /* Move the result of the SCS instruction to the desired location in - * the destination. - */ - tmp.swizzle = MAKE_SWIZZLE4(component, component, - component, component); - inst = emit(ir, OPCODE_SCS, dst, tmp); - inst->dst.writemask = this_mask; - } else { - /* Emit the SCS instruction to write directly to the destination. - */ - ir_to_mesa_instruction *inst = emit(ir, OPCODE_SCS, dst, src0); - inst->dst.writemask = scs_mask; - } - - done_mask |= this_mask; - } -} - -src_reg -ir_to_mesa_visitor::src_reg_for_float(float val) -{ - src_reg src(PROGRAM_CONSTANT, -1, NULL); - - src.index = _mesa_add_unnamed_constant(this->prog->Parameters, - (const gl_constant_value *)&val, 1, &src.swizzle); - - return src; -} - -static int -type_size(const struct glsl_type *type) -{ - unsigned int i; - int size; - - switch (type->base_type) { - case GLSL_TYPE_UINT: - case GLSL_TYPE_INT: - case GLSL_TYPE_FLOAT: - case GLSL_TYPE_BOOL: - if (type->is_matrix()) { - return type->matrix_columns; - } else { - /* Regardless of size of vector, it gets a vec4. This is bad - * packing for things like floats, but otherwise arrays become a - * mess. Hopefully a later pass over the code can pack scalars - * down if appropriate. - */ - return 1; - } - case GLSL_TYPE_ARRAY: - assert(type->length > 0); - return type_size(type->fields.array) * type->length; - case GLSL_TYPE_STRUCT: - size = 0; - for (i = 0; i < type->length; i++) { - size += type_size(type->fields.structure[i].type); - } - return size; - case GLSL_TYPE_SAMPLER: - /* Samplers take up one slot in UNIFORMS[], but they're baked in - * at link time. - */ - return 1; - default: - assert(0); - return 0; - } -} - -/** - * In the initial pass of codegen, we assign temporary numbers to - * intermediate results. (not SSA -- variable assignments will reuse - * storage). Actual register allocation for the Mesa VM occurs in a - * pass over the Mesa IR later. - */ -src_reg -ir_to_mesa_visitor::get_temp(const glsl_type *type) -{ - src_reg src; - - src.file = PROGRAM_TEMPORARY; - src.index = next_temp; - src.reladdr = NULL; - next_temp += type_size(type); - - if (type->is_array() || type->is_record()) { - src.swizzle = SWIZZLE_NOOP; - } else { - src.swizzle = swizzle_for_size(type->vector_elements); - } - src.negate = 0; - - return src; -} - -variable_storage * -ir_to_mesa_visitor::find_variable_storage(ir_variable *var) -{ - - variable_storage *entry; - - foreach_iter(exec_list_iterator, iter, this->variables) { - entry = (variable_storage *)iter.get(); - - if (entry->var == var) - return entry; - } - - return NULL; -} - -void -ir_to_mesa_visitor::visit(ir_variable *ir) -{ - if (strcmp(ir->name, "gl_FragCoord") == 0) { - struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog; - - fp->OriginUpperLeft = ir->origin_upper_left; - fp->PixelCenterInteger = ir->pixel_center_integer; - - } else if (strcmp(ir->name, "gl_FragDepth") == 0) { - struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog; - switch (ir->depth_layout) { - case ir_depth_layout_none: - fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_NONE; - break; - case ir_depth_layout_any: - fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_ANY; - break; - case ir_depth_layout_greater: - fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_GREATER; - break; - case ir_depth_layout_less: - fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_LESS; - break; - case ir_depth_layout_unchanged: - fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_UNCHANGED; - break; - default: - assert(0); - break; - } - } - - if (ir->mode == ir_var_uniform && strncmp(ir->name, "gl_", 3) == 0) { - unsigned int i; - const ir_state_slot *const slots = ir->state_slots; - assert(ir->state_slots != NULL); - - /* Check if this statevar's setup in the STATE file exactly - * matches how we'll want to reference it as a - * struct/array/whatever. If not, then we need to move it into - * temporary storage and hope that it'll get copy-propagated - * out. - */ - for (i = 0; i < ir->num_state_slots; i++) { - if (slots[i].swizzle != SWIZZLE_XYZW) { - break; - } - } - - variable_storage *storage; - dst_reg dst; - if (i == ir->num_state_slots) { - /* We'll set the index later. */ - storage = new(mem_ctx) variable_storage(ir, PROGRAM_STATE_VAR, -1); - this->variables.push_tail(storage); - - dst = undef_dst; - } else { - /* The variable_storage constructor allocates slots based on the size - * of the type. However, this had better match the number of state - * elements that we're going to copy into the new temporary. - */ - assert((int) ir->num_state_slots == type_size(ir->type)); - - storage = new(mem_ctx) variable_storage(ir, PROGRAM_TEMPORARY, - this->next_temp); - this->variables.push_tail(storage); - this->next_temp += type_size(ir->type); - - dst = dst_reg(src_reg(PROGRAM_TEMPORARY, storage->index, NULL)); - } - - - for (unsigned int i = 0; i < ir->num_state_slots; i++) { - int index = _mesa_add_state_reference(this->prog->Parameters, - (gl_state_index *)slots[i].tokens); - - if (storage->file == PROGRAM_STATE_VAR) { - if (storage->index == -1) { - storage->index = index; - } else { - assert(index == storage->index + (int)i); - } - } else { - src_reg src(PROGRAM_STATE_VAR, index, NULL); - src.swizzle = slots[i].swizzle; - emit(ir, OPCODE_MOV, dst, src); - /* even a float takes up a whole vec4 reg in a struct/array. */ - dst.index++; - } - } - - if (storage->file == PROGRAM_TEMPORARY && - dst.index != storage->index + (int) ir->num_state_slots) { - linker_error(this->shader_program, - "failed to load builtin uniform `%s' " - "(%d/%d regs loaded)\n", - ir->name, dst.index - storage->index, - type_size(ir->type)); - } - } -} - -void -ir_to_mesa_visitor::visit(ir_loop *ir) -{ - ir_dereference_variable *counter = NULL; - - if (ir->counter != NULL) - counter = new(mem_ctx) ir_dereference_variable(ir->counter); - - if (ir->from != NULL) { - assert(ir->counter != NULL); - - ir_assignment *a = - new(mem_ctx) ir_assignment(counter, ir->from, NULL); - - a->accept(this); - } - - emit(NULL, OPCODE_BGNLOOP); - - if (ir->to) { - ir_expression *e = - new(mem_ctx) ir_expression(ir->cmp, glsl_type::bool_type, - counter, ir->to); - ir_if *if_stmt = new(mem_ctx) ir_if(e); - - ir_loop_jump *brk = - new(mem_ctx) ir_loop_jump(ir_loop_jump::jump_break); - - if_stmt->then_instructions.push_tail(brk); - - if_stmt->accept(this); - } - - visit_exec_list(&ir->body_instructions, this); - - if (ir->increment) { - ir_expression *e = - new(mem_ctx) ir_expression(ir_binop_add, counter->type, - counter, ir->increment); - - ir_assignment *a = - new(mem_ctx) ir_assignment(counter, e, NULL); - - a->accept(this); - } - - emit(NULL, OPCODE_ENDLOOP); -} - -void -ir_to_mesa_visitor::visit(ir_loop_jump *ir) -{ - switch (ir->mode) { - case ir_loop_jump::jump_break: - emit(NULL, OPCODE_BRK); - break; - case ir_loop_jump::jump_continue: - emit(NULL, OPCODE_CONT); - break; - } -} - - -void -ir_to_mesa_visitor::visit(ir_function_signature *ir) -{ - assert(0); - (void)ir; -} - -void -ir_to_mesa_visitor::visit(ir_function *ir) -{ - /* Ignore function bodies other than main() -- we shouldn't see calls to - * them since they should all be inlined before we get to ir_to_mesa. - */ - if (strcmp(ir->name, "main") == 0) { - const ir_function_signature *sig; - exec_list empty; - - sig = ir->matching_signature(&empty); - - assert(sig); - - foreach_iter(exec_list_iterator, iter, sig->body) { - ir_instruction *ir = (ir_instruction *)iter.get(); - - ir->accept(this); - } - } -} - -bool -ir_to_mesa_visitor::try_emit_mad(ir_expression *ir, int mul_operand) -{ - int nonmul_operand = 1 - mul_operand; - src_reg a, b, c; - - ir_expression *expr = ir->operands[mul_operand]->as_expression(); - if (!expr || expr->operation != ir_binop_mul) - return false; - - expr->operands[0]->accept(this); - a = this->result; - expr->operands[1]->accept(this); - b = this->result; - ir->operands[nonmul_operand]->accept(this); - c = this->result; - - this->result = get_temp(ir->type); - emit(ir, OPCODE_MAD, dst_reg(this->result), a, b, c); - - return true; -} - -/** - * Emit OPCODE_MAD(a, -b, a) instead of AND(a, NOT(b)) - * - * The logic values are 1.0 for true and 0.0 for false. Logical-and is - * implemented using multiplication, and logical-or is implemented using - * addition. Logical-not can be implemented as (true - x), or (1.0 - x). - * As result, the logical expression (a & !b) can be rewritten as: - * - * - a * !b - * - a * (1 - b) - * - (a * 1) - (a * b) - * - a + -(a * b) - * - a + (a * -b) - * - * This final expression can be implemented as a single MAD(a, -b, a) - * instruction. - */ -bool -ir_to_mesa_visitor::try_emit_mad_for_and_not(ir_expression *ir, int try_operand) -{ - const int other_operand = 1 - try_operand; - src_reg a, b; - - ir_expression *expr = ir->operands[try_operand]->as_expression(); - if (!expr || expr->operation != ir_unop_logic_not) - return false; - - ir->operands[other_operand]->accept(this); - a = this->result; - expr->operands[0]->accept(this); - b = this->result; - - b.negate = ~b.negate; - - this->result = get_temp(ir->type); - emit(ir, OPCODE_MAD, dst_reg(this->result), a, b, a); - - return true; -} - -bool -ir_to_mesa_visitor::try_emit_sat(ir_expression *ir) -{ - /* Saturates were only introduced to vertex programs in - * NV_vertex_program3, so don't give them to drivers in the VP. - */ - if (this->prog->Target == GL_VERTEX_PROGRAM_ARB) - return false; - - ir_rvalue *sat_src = ir->as_rvalue_to_saturate(); - if (!sat_src) - return false; - - sat_src->accept(this); - src_reg src = this->result; - - /* If we generated an expression instruction into a temporary in - * processing the saturate's operand, apply the saturate to that - * instruction. Otherwise, generate a MOV to do the saturate. - * - * Note that we have to be careful to only do this optimization if - * the instruction in question was what generated src->result. For - * example, ir_dereference_array might generate a MUL instruction - * to create the reladdr, and return us a src reg using that - * reladdr. That MUL result is not the value we're trying to - * saturate. - */ - ir_expression *sat_src_expr = sat_src->as_expression(); - ir_to_mesa_instruction *new_inst; - new_inst = (ir_to_mesa_instruction *)this->instructions.get_tail(); - if (sat_src_expr && (sat_src_expr->operation == ir_binop_mul || - sat_src_expr->operation == ir_binop_add || - sat_src_expr->operation == ir_binop_dot)) { - new_inst->saturate = true; - } else { - this->result = get_temp(ir->type); - ir_to_mesa_instruction *inst; - inst = emit(ir, OPCODE_MOV, dst_reg(this->result), src); - inst->saturate = true; - } - - return true; -} - -void -ir_to_mesa_visitor::reladdr_to_temp(ir_instruction *ir, - src_reg *reg, int *num_reladdr) -{ - if (!reg->reladdr) - return; - - emit(ir, OPCODE_ARL, address_reg, *reg->reladdr); - - if (*num_reladdr != 1) { - src_reg temp = get_temp(glsl_type::vec4_type); - - emit(ir, OPCODE_MOV, dst_reg(temp), *reg); - *reg = temp; - } - - (*num_reladdr)--; -} - -void -ir_to_mesa_visitor::emit_swz(ir_expression *ir) -{ - /* Assume that the vector operator is in a form compatible with OPCODE_SWZ. - * This means that each of the operands is either an immediate value of -1, - * 0, or 1, or is a component from one source register (possibly with - * negation). - */ - uint8_t components[4] = { 0 }; - bool negate[4] = { false }; - ir_variable *var = NULL; - - for (unsigned i = 0; i < ir->type->vector_elements; i++) { - ir_rvalue *op = ir->operands[i]; - - assert(op->type->is_scalar()); - - while (op != NULL) { - switch (op->ir_type) { - case ir_type_constant: { - - assert(op->type->is_scalar()); - - const ir_constant *const c = op->as_constant(); - if (c->is_one()) { - components[i] = SWIZZLE_ONE; - } else if (c->is_zero()) { - components[i] = SWIZZLE_ZERO; - } else if (c->is_negative_one()) { - components[i] = SWIZZLE_ONE; - negate[i] = true; - } else { - assert(!"SWZ constant must be 0.0 or 1.0."); - } - - op = NULL; - break; - } - - case ir_type_dereference_variable: { - ir_dereference_variable *const deref = - (ir_dereference_variable *) op; - - assert((var == NULL) || (deref->var == var)); - components[i] = SWIZZLE_X; - var = deref->var; - op = NULL; - break; - } - - case ir_type_expression: { - ir_expression *const expr = (ir_expression *) op; - - assert(expr->operation == ir_unop_neg); - negate[i] = true; - - op = expr->operands[0]; - break; - } - - case ir_type_swizzle: { - ir_swizzle *const swiz = (ir_swizzle *) op; - - components[i] = swiz->mask.x; - op = swiz->val; - break; - } - - default: - assert(!"Should not get here."); - return; - } - } - } - - assert(var != NULL); - - ir_dereference_variable *const deref = - new(mem_ctx) ir_dereference_variable(var); - - this->result.file = PROGRAM_UNDEFINED; - deref->accept(this); - if (this->result.file == PROGRAM_UNDEFINED) { - ir_print_visitor v; - printf("Failed to get tree for expression operand:\n"); - deref->accept(&v); - exit(1); - } - - src_reg src; - - src = this->result; - src.swizzle = MAKE_SWIZZLE4(components[0], - components[1], - components[2], - components[3]); - src.negate = ((unsigned(negate[0]) << 0) - | (unsigned(negate[1]) << 1) - | (unsigned(negate[2]) << 2) - | (unsigned(negate[3]) << 3)); - - /* Storage for our result. Ideally for an assignment we'd be using the - * actual storage for the result here, instead. - */ - const src_reg result_src = get_temp(ir->type); - dst_reg result_dst = dst_reg(result_src); - - /* Limit writes to the channels that will be used by result_src later. - * This does limit this temp's use as a temporary for multi-instruction - * sequences. - */ - result_dst.writemask = (1 << ir->type->vector_elements) - 1; - - emit(ir, OPCODE_SWZ, result_dst, src); - this->result = result_src; -} - -void -ir_to_mesa_visitor::visit(ir_expression *ir) -{ - unsigned int operand; - src_reg op[Elements(ir->operands)]; - src_reg result_src; - dst_reg result_dst; - - /* Quick peephole: Emit OPCODE_MAD(a, b, c) instead of ADD(MUL(a, b), c) - */ - if (ir->operation == ir_binop_add) { - if (try_emit_mad(ir, 1)) - return; - if (try_emit_mad(ir, 0)) - return; - } - - /* Quick peephole: Emit OPCODE_MAD(-a, -b, a) instead of AND(a, NOT(b)) - */ - if (ir->operation == ir_binop_logic_and) { - if (try_emit_mad_for_and_not(ir, 1)) - return; - if (try_emit_mad_for_and_not(ir, 0)) - return; - } - - if (try_emit_sat(ir)) - return; - - if (ir->operation == ir_quadop_vector) { - this->emit_swz(ir); - return; - } - - for (operand = 0; operand < ir->get_num_operands(); operand++) { - this->result.file = PROGRAM_UNDEFINED; - ir->operands[operand]->accept(this); - if (this->result.file == PROGRAM_UNDEFINED) { - ir_print_visitor v; - printf("Failed to get tree for expression operand:\n"); - ir->operands[operand]->accept(&v); - exit(1); - } - op[operand] = this->result; - - /* Matrix expression operands should have been broken down to vector - * operations already. - */ - assert(!ir->operands[operand]->type->is_matrix()); - } - - int vector_elements = ir->operands[0]->type->vector_elements; - if (ir->operands[1]) { - vector_elements = MAX2(vector_elements, - ir->operands[1]->type->vector_elements); - } - - this->result.file = PROGRAM_UNDEFINED; - - /* Storage for our result. Ideally for an assignment we'd be using - * the actual storage for the result here, instead. - */ - result_src = get_temp(ir->type); - /* convenience for the emit functions below. */ - result_dst = dst_reg(result_src); - /* Limit writes to the channels that will be used by result_src later. - * This does limit this temp's use as a temporary for multi-instruction - * sequences. - */ - result_dst.writemask = (1 << ir->type->vector_elements) - 1; - - switch (ir->operation) { - case ir_unop_logic_not: - /* Previously 'SEQ dst, src, 0.0' was used for this. However, many - * older GPUs implement SEQ using multiple instructions (i915 uses two - * SGE instructions and a MUL instruction). Since our logic values are - * 0.0 and 1.0, 1-x also implements !x. - */ - op[0].negate = ~op[0].negate; - emit(ir, OPCODE_ADD, result_dst, op[0], src_reg_for_float(1.0)); - break; - case ir_unop_neg: - op[0].negate = ~op[0].negate; - result_src = op[0]; - break; - case ir_unop_abs: - emit(ir, OPCODE_ABS, result_dst, op[0]); - break; - case ir_unop_sign: - emit(ir, OPCODE_SSG, result_dst, op[0]); - break; - case ir_unop_rcp: - emit_scalar(ir, OPCODE_RCP, result_dst, op[0]); - break; - - case ir_unop_exp2: - emit_scalar(ir, OPCODE_EX2, result_dst, op[0]); - break; - case ir_unop_exp: - case ir_unop_log: - assert(!"not reached: should be handled by ir_explog_to_explog2"); - break; - case ir_unop_log2: - emit_scalar(ir, OPCODE_LG2, result_dst, op[0]); - break; - case ir_unop_sin: - emit_scalar(ir, OPCODE_SIN, result_dst, op[0]); - break; - case ir_unop_cos: - emit_scalar(ir, OPCODE_COS, result_dst, op[0]); - break; - case ir_unop_sin_reduced: - emit_scs(ir, OPCODE_SIN, result_dst, op[0]); - break; - case ir_unop_cos_reduced: - emit_scs(ir, OPCODE_COS, result_dst, op[0]); - break; - - case ir_unop_dFdx: - emit(ir, OPCODE_DDX, result_dst, op[0]); - break; - case ir_unop_dFdy: - emit(ir, OPCODE_DDY, result_dst, op[0]); - break; - - case ir_unop_noise: { - const enum prog_opcode opcode = - prog_opcode(OPCODE_NOISE1 - + (ir->operands[0]->type->vector_elements) - 1); - assert((opcode >= OPCODE_NOISE1) && (opcode <= OPCODE_NOISE4)); - - emit(ir, opcode, result_dst, op[0]); - break; - } - - case ir_binop_add: - emit(ir, OPCODE_ADD, result_dst, op[0], op[1]); - break; - case ir_binop_sub: - emit(ir, OPCODE_SUB, result_dst, op[0], op[1]); - break; - - case ir_binop_mul: - emit(ir, OPCODE_MUL, result_dst, op[0], op[1]); - break; - case ir_binop_div: - assert(!"not reached: should be handled by ir_div_to_mul_rcp"); - case ir_binop_mod: - assert(!"ir_binop_mod should have been converted to b * fract(a/b)"); - break; - - case ir_binop_less: - emit(ir, OPCODE_SLT, result_dst, op[0], op[1]); - break; - case ir_binop_greater: - emit(ir, OPCODE_SGT, result_dst, op[0], op[1]); - break; - case ir_binop_lequal: - emit(ir, OPCODE_SLE, result_dst, op[0], op[1]); - break; - case ir_binop_gequal: - emit(ir, OPCODE_SGE, result_dst, op[0], op[1]); - break; - case ir_binop_equal: - emit(ir, OPCODE_SEQ, result_dst, op[0], op[1]); - break; - case ir_binop_nequal: - emit(ir, OPCODE_SNE, result_dst, op[0], op[1]); - break; - case ir_binop_all_equal: - /* "==" operator producing a scalar boolean. */ - if (ir->operands[0]->type->is_vector() || - ir->operands[1]->type->is_vector()) { - src_reg temp = get_temp(glsl_type::vec4_type); - emit(ir, OPCODE_SNE, dst_reg(temp), op[0], op[1]); - - /* After the dot-product, the value will be an integer on the - * range [0,4]. Zero becomes 1.0, and positive values become zero. - */ - emit_dp(ir, result_dst, temp, temp, vector_elements); - - /* Negating the result of the dot-product gives values on the range - * [-4, 0]. Zero becomes 1.0, and negative values become zero. This - * achieved using SGE. - */ - src_reg sge_src = result_src; - sge_src.negate = ~sge_src.negate; - emit(ir, OPCODE_SGE, result_dst, sge_src, src_reg_for_float(0.0)); - } else { - emit(ir, OPCODE_SEQ, result_dst, op[0], op[1]); - } - break; - case ir_binop_any_nequal: - /* "!=" operator producing a scalar boolean. */ - if (ir->operands[0]->type->is_vector() || - ir->operands[1]->type->is_vector()) { - src_reg temp = get_temp(glsl_type::vec4_type); - emit(ir, OPCODE_SNE, dst_reg(temp), op[0], op[1]); - - /* After the dot-product, the value will be an integer on the - * range [0,4]. Zero stays zero, and positive values become 1.0. - */ - ir_to_mesa_instruction *const dp = - emit_dp(ir, result_dst, temp, temp, vector_elements); - if (this->prog->Target == GL_FRAGMENT_PROGRAM_ARB) { - /* The clamping to [0,1] can be done for free in the fragment - * shader with a saturate. - */ - dp->saturate = true; - } else { - /* Negating the result of the dot-product gives values on the range - * [-4, 0]. Zero stays zero, and negative values become 1.0. This - * achieved using SLT. - */ - src_reg slt_src = result_src; - slt_src.negate = ~slt_src.negate; - emit(ir, OPCODE_SLT, result_dst, slt_src, src_reg_for_float(0.0)); - } - } else { - emit(ir, OPCODE_SNE, result_dst, op[0], op[1]); - } - break; - - case ir_unop_any: { - assert(ir->operands[0]->type->is_vector()); - - /* After the dot-product, the value will be an integer on the - * range [0,4]. Zero stays zero, and positive values become 1.0. - */ - ir_to_mesa_instruction *const dp = - emit_dp(ir, result_dst, op[0], op[0], - ir->operands[0]->type->vector_elements); - if (this->prog->Target == GL_FRAGMENT_PROGRAM_ARB) { - /* The clamping to [0,1] can be done for free in the fragment - * shader with a saturate. - */ - dp->saturate = true; - } else { - /* Negating the result of the dot-product gives values on the range - * [-4, 0]. Zero stays zero, and negative values become 1.0. This - * is achieved using SLT. - */ - src_reg slt_src = result_src; - slt_src.negate = ~slt_src.negate; - emit(ir, OPCODE_SLT, result_dst, slt_src, src_reg_for_float(0.0)); - } - break; - } - - case ir_binop_logic_xor: - emit(ir, OPCODE_SNE, result_dst, op[0], op[1]); - break; - - case ir_binop_logic_or: { - /* After the addition, the value will be an integer on the - * range [0,2]. Zero stays zero, and positive values become 1.0. - */ - ir_to_mesa_instruction *add = - emit(ir, OPCODE_ADD, result_dst, op[0], op[1]); - if (this->prog->Target == GL_FRAGMENT_PROGRAM_ARB) { - /* The clamping to [0,1] can be done for free in the fragment - * shader with a saturate. - */ - add->saturate = true; - } else { - /* Negating the result of the addition gives values on the range - * [-2, 0]. Zero stays zero, and negative values become 1.0. This - * is achieved using SLT. - */ - src_reg slt_src = result_src; - slt_src.negate = ~slt_src.negate; - emit(ir, OPCODE_SLT, result_dst, slt_src, src_reg_for_float(0.0)); - } - break; - } - - case ir_binop_logic_and: - /* the bool args are stored as float 0.0 or 1.0, so "mul" gives us "and". */ - emit(ir, OPCODE_MUL, result_dst, op[0], op[1]); - break; - - case ir_binop_dot: - assert(ir->operands[0]->type->is_vector()); - assert(ir->operands[0]->type == ir->operands[1]->type); - emit_dp(ir, result_dst, op[0], op[1], - ir->operands[0]->type->vector_elements); - break; - - case ir_unop_sqrt: - /* sqrt(x) = x * rsq(x). */ - emit_scalar(ir, OPCODE_RSQ, result_dst, op[0]); - emit(ir, OPCODE_MUL, result_dst, result_src, op[0]); - /* For incoming channels <= 0, set the result to 0. */ - op[0].negate = ~op[0].negate; - emit(ir, OPCODE_CMP, result_dst, - op[0], result_src, src_reg_for_float(0.0)); - break; - case ir_unop_rsq: - emit_scalar(ir, OPCODE_RSQ, result_dst, op[0]); - break; - case ir_unop_i2f: - case ir_unop_u2f: - case ir_unop_b2f: - case ir_unop_b2i: - case ir_unop_i2u: - case ir_unop_u2i: - /* Mesa IR lacks types, ints are stored as truncated floats. */ - result_src = op[0]; - break; - case ir_unop_f2i: - emit(ir, OPCODE_TRUNC, result_dst, op[0]); - break; - case ir_unop_f2b: - case ir_unop_i2b: - emit(ir, OPCODE_SNE, result_dst, - op[0], src_reg_for_float(0.0)); - break; - case ir_unop_trunc: - emit(ir, OPCODE_TRUNC, result_dst, op[0]); - break; - case ir_unop_ceil: - op[0].negate = ~op[0].negate; - emit(ir, OPCODE_FLR, result_dst, op[0]); - result_src.negate = ~result_src.negate; - break; - case ir_unop_floor: - emit(ir, OPCODE_FLR, result_dst, op[0]); - break; - case ir_unop_fract: - emit(ir, OPCODE_FRC, result_dst, op[0]); - break; - - case ir_binop_min: - emit(ir, OPCODE_MIN, result_dst, op[0], op[1]); - break; - case ir_binop_max: - emit(ir, OPCODE_MAX, result_dst, op[0], op[1]); - break; - case ir_binop_pow: - emit_scalar(ir, OPCODE_POW, result_dst, op[0], op[1]); - break; - - case ir_unop_bit_not: - case ir_binop_lshift: - case ir_binop_rshift: - case ir_binop_bit_and: - case ir_binop_bit_xor: - case ir_binop_bit_or: - case ir_unop_round_even: - assert(!"GLSL 1.30 features unsupported"); - break; - - case ir_quadop_vector: - /* This operation should have already been handled. - */ - assert(!"Should not get here."); - break; - } - - this->result = result_src; -} - - -void -ir_to_mesa_visitor::visit(ir_swizzle *ir) -{ - src_reg src; - int i; - int swizzle[4]; - - /* Note that this is only swizzles in expressions, not those on the left - * hand side of an assignment, which do write masking. See ir_assignment - * for that. - */ - - ir->val->accept(this); - src = this->result; - assert(src.file != PROGRAM_UNDEFINED); - - for (i = 0; i < 4; i++) { - if (i < ir->type->vector_elements) { - switch (i) { - case 0: - swizzle[i] = GET_SWZ(src.swizzle, ir->mask.x); - break; - case 1: - swizzle[i] = GET_SWZ(src.swizzle, ir->mask.y); - break; - case 2: - swizzle[i] = GET_SWZ(src.swizzle, ir->mask.z); - break; - case 3: - swizzle[i] = GET_SWZ(src.swizzle, ir->mask.w); - break; - } - } else { - /* If the type is smaller than a vec4, replicate the last - * channel out. - */ - swizzle[i] = swizzle[ir->type->vector_elements - 1]; - } - } - - src.swizzle = MAKE_SWIZZLE4(swizzle[0], swizzle[1], swizzle[2], swizzle[3]); - - this->result = src; -} - -void -ir_to_mesa_visitor::visit(ir_dereference_variable *ir) -{ - variable_storage *entry = find_variable_storage(ir->var); - ir_variable *var = ir->var; - - if (!entry) { - switch (var->mode) { - case ir_var_uniform: - entry = new(mem_ctx) variable_storage(var, PROGRAM_UNIFORM, - var->location); - this->variables.push_tail(entry); - break; - case ir_var_in: - case ir_var_inout: - /* The linker assigns locations for varyings and attributes, - * including deprecated builtins (like gl_Color), - * user-assigned generic attributes (glBindVertexLocation), - * and user-defined varyings. - * - * FINISHME: We would hit this path for function arguments. Fix! - */ - assert(var->location != -1); - entry = new(mem_ctx) variable_storage(var, - PROGRAM_INPUT, - var->location); - if (this->prog->Target == GL_VERTEX_PROGRAM_ARB && - var->location >= VERT_ATTRIB_GENERIC0) { - _mesa_add_attribute(this->prog->Attributes, - var->name, - _mesa_sizeof_glsl_type(var->type->gl_type), - var->type->gl_type, - var->location - VERT_ATTRIB_GENERIC0); - } - break; - case ir_var_out: - assert(var->location != -1); - entry = new(mem_ctx) variable_storage(var, - PROGRAM_OUTPUT, - var->location); - break; - case ir_var_system_value: - entry = new(mem_ctx) variable_storage(var, - PROGRAM_SYSTEM_VALUE, - var->location); - break; - case ir_var_auto: - case ir_var_temporary: - entry = new(mem_ctx) variable_storage(var, PROGRAM_TEMPORARY, - this->next_temp); - this->variables.push_tail(entry); - - next_temp += type_size(var->type); - break; - } - - if (!entry) { - printf("Failed to make storage for %s\n", var->name); - exit(1); - } - } - - this->result = src_reg(entry->file, entry->index, var->type); -} - -void -ir_to_mesa_visitor::visit(ir_dereference_array *ir) -{ - ir_constant *index; - src_reg src; - int element_size = type_size(ir->type); - - index = ir->array_index->constant_expression_value(); - - ir->array->accept(this); - src = this->result; - - if (index) { - src.index += index->value.i[0] * element_size; - } else { - /* Variable index array dereference. It eats the "vec4" of the - * base of the array and an index that offsets the Mesa register - * index. - */ - ir->array_index->accept(this); - - src_reg index_reg; - - if (element_size == 1) { - index_reg = this->result; - } else { - index_reg = get_temp(glsl_type::float_type); - - emit(ir, OPCODE_MUL, dst_reg(index_reg), - this->result, src_reg_for_float(element_size)); - } - - /* If there was already a relative address register involved, add the - * new and the old together to get the new offset. - */ - if (src.reladdr != NULL) { - src_reg accum_reg = get_temp(glsl_type::float_type); - - emit(ir, OPCODE_ADD, dst_reg(accum_reg), - index_reg, *src.reladdr); - - index_reg = accum_reg; - } - - src.reladdr = ralloc(mem_ctx, src_reg); - memcpy(src.reladdr, &index_reg, sizeof(index_reg)); - } - - /* If the type is smaller than a vec4, replicate the last channel out. */ - if (ir->type->is_scalar() || ir->type->is_vector()) - src.swizzle = swizzle_for_size(ir->type->vector_elements); - else - src.swizzle = SWIZZLE_NOOP; - - this->result = src; -} - -void -ir_to_mesa_visitor::visit(ir_dereference_record *ir) -{ - unsigned int i; - const glsl_type *struct_type = ir->record->type; - int offset = 0; - - ir->record->accept(this); - - for (i = 0; i < struct_type->length; i++) { - if (strcmp(struct_type->fields.structure[i].name, ir->field) == 0) - break; - offset += type_size(struct_type->fields.structure[i].type); - } - - /* If the type is smaller than a vec4, replicate the last channel out. */ - if (ir->type->is_scalar() || ir->type->is_vector()) - this->result.swizzle = swizzle_for_size(ir->type->vector_elements); - else - this->result.swizzle = SWIZZLE_NOOP; - - this->result.index += offset; -} - -/** - * We want to be careful in assignment setup to hit the actual storage - * instead of potentially using a temporary like we might with the - * ir_dereference handler. - */ -static dst_reg -get_assignment_lhs(ir_dereference *ir, ir_to_mesa_visitor *v) -{ - /* The LHS must be a dereference. If the LHS is a variable indexed array - * access of a vector, it must be separated into a series conditional moves - * before reaching this point (see ir_vec_index_to_cond_assign). - */ - assert(ir->as_dereference()); - ir_dereference_array *deref_array = ir->as_dereference_array(); - if (deref_array) { - assert(!deref_array->array->type->is_vector()); - } - - /* Use the rvalue deref handler for the most part. We'll ignore - * swizzles in it and write swizzles using writemask, though. - */ - ir->accept(v); - return dst_reg(v->result); -} - -/** - * Process the condition of a conditional assignment - * - * Examines the condition of a conditional assignment to generate the optimal - * first operand of a \c CMP instruction. If the condition is a relational - * operator with 0 (e.g., \c ir_binop_less), the value being compared will be - * used as the source for the \c CMP instruction. Otherwise the comparison - * is processed to a boolean result, and the boolean result is used as the - * operand to the CMP instruction. - */ -bool -ir_to_mesa_visitor::process_move_condition(ir_rvalue *ir) -{ - ir_rvalue *src_ir = ir; - bool negate = true; - bool switch_order = false; - - ir_expression *const expr = ir->as_expression(); - if ((expr != NULL) && (expr->get_num_operands() == 2)) { - bool zero_on_left = false; - - if (expr->operands[0]->is_zero()) { - src_ir = expr->operands[1]; - zero_on_left = true; - } else if (expr->operands[1]->is_zero()) { - src_ir = expr->operands[0]; - zero_on_left = false; - } - - /* a is - 0 + - 0 + - * (a < 0) T F F ( a < 0) T F F - * (0 < a) F F T (-a < 0) F F T - * (a <= 0) T T F (-a < 0) F F T (swap order of other operands) - * (0 <= a) F T T ( a < 0) T F F (swap order of other operands) - * (a > 0) F F T (-a < 0) F F T - * (0 > a) T F F ( a < 0) T F F - * (a >= 0) F T T ( a < 0) T F F (swap order of other operands) - * (0 >= a) T T F (-a < 0) F F T (swap order of other operands) - * - * Note that exchanging the order of 0 and 'a' in the comparison simply - * means that the value of 'a' should be negated. - */ - if (src_ir != ir) { - switch (expr->operation) { - case ir_binop_less: - switch_order = false; - negate = zero_on_left; - break; - - case ir_binop_greater: - switch_order = false; - negate = !zero_on_left; - break; - - case ir_binop_lequal: - switch_order = true; - negate = !zero_on_left; - break; - - case ir_binop_gequal: - switch_order = true; - negate = zero_on_left; - break; - - default: - /* This isn't the right kind of comparison afterall, so make sure - * the whole condition is visited. - */ - src_ir = ir; - break; - } - } - } - - src_ir->accept(this); - - /* We use the OPCODE_CMP (a < 0 ? b : c) for conditional moves, and the - * condition we produced is 0.0 or 1.0. By flipping the sign, we can - * choose which value OPCODE_CMP produces without an extra instruction - * computing the condition. - */ - if (negate) - this->result.negate = ~this->result.negate; - - return switch_order; -} - -void -ir_to_mesa_visitor::visit(ir_assignment *ir) -{ - dst_reg l; - src_reg r; - int i; - - ir->rhs->accept(this); - r = this->result; - - l = get_assignment_lhs(ir->lhs, this); - - /* FINISHME: This should really set to the correct maximal writemask for each - * FINISHME: component written (in the loops below). This case can only - * FINISHME: occur for matrices, arrays, and structures. - */ - if (ir->write_mask == 0) { - assert(!ir->lhs->type->is_scalar() && !ir->lhs->type->is_vector()); - l.writemask = WRITEMASK_XYZW; - } else if (ir->lhs->type->is_scalar()) { - /* FINISHME: This hack makes writing to gl_FragDepth, which lives in the - * FINISHME: W component of fragment shader output zero, work correctly. - */ - l.writemask = WRITEMASK_XYZW; - } else { - int swizzles[4]; - int first_enabled_chan = 0; - int rhs_chan = 0; - - assert(ir->lhs->type->is_vector()); - l.writemask = ir->write_mask; - - for (int i = 0; i < 4; i++) { - if (l.writemask & (1 << i)) { - first_enabled_chan = GET_SWZ(r.swizzle, i); - break; - } - } - - /* Swizzle a small RHS vector into the channels being written. - * - * glsl ir treats write_mask as dictating how many channels are - * present on the RHS while Mesa IR treats write_mask as just - * showing which channels of the vec4 RHS get written. - */ - for (int i = 0; i < 4; i++) { - if (l.writemask & (1 << i)) - swizzles[i] = GET_SWZ(r.swizzle, rhs_chan++); - else - swizzles[i] = first_enabled_chan; - } - r.swizzle = MAKE_SWIZZLE4(swizzles[0], swizzles[1], - swizzles[2], swizzles[3]); - } - - assert(l.file != PROGRAM_UNDEFINED); - assert(r.file != PROGRAM_UNDEFINED); - - if (ir->condition) { - const bool switch_order = this->process_move_condition(ir->condition); - src_reg condition = this->result; - - for (i = 0; i < type_size(ir->lhs->type); i++) { - if (switch_order) { - emit(ir, OPCODE_CMP, l, condition, src_reg(l), r); - } else { - emit(ir, OPCODE_CMP, l, condition, r, src_reg(l)); - } - - l.index++; - r.index++; - } - } else { - for (i = 0; i < type_size(ir->lhs->type); i++) { - emit(ir, OPCODE_MOV, l, r); - l.index++; - r.index++; - } - } -} - - -void -ir_to_mesa_visitor::visit(ir_constant *ir) -{ - src_reg src; - GLfloat stack_vals[4] = { 0 }; - GLfloat *values = stack_vals; - unsigned int i; - - /* Unfortunately, 4 floats is all we can get into - * _mesa_add_unnamed_constant. So, make a temp to store an - * aggregate constant and move each constant value into it. If we - * get lucky, copy propagation will eliminate the extra moves. - */ - - if (ir->type->base_type == GLSL_TYPE_STRUCT) { - src_reg temp_base = get_temp(ir->type); - dst_reg temp = dst_reg(temp_base); - - foreach_iter(exec_list_iterator, iter, ir->components) { - ir_constant *field_value = (ir_constant *)iter.get(); - int size = type_size(field_value->type); - - assert(size > 0); - - field_value->accept(this); - src = this->result; - - for (i = 0; i < (unsigned int)size; i++) { - emit(ir, OPCODE_MOV, temp, src); - - src.index++; - temp.index++; - } - } - this->result = temp_base; - return; - } - - if (ir->type->is_array()) { - src_reg temp_base = get_temp(ir->type); - dst_reg temp = dst_reg(temp_base); - int size = type_size(ir->type->fields.array); - - assert(size > 0); - - for (i = 0; i < ir->type->length; i++) { - ir->array_elements[i]->accept(this); - src = this->result; - for (int j = 0; j < size; j++) { - emit(ir, OPCODE_MOV, temp, src); - - src.index++; - temp.index++; - } - } - this->result = temp_base; - return; - } - - if (ir->type->is_matrix()) { - src_reg mat = get_temp(ir->type); - dst_reg mat_column = dst_reg(mat); - - for (i = 0; i < ir->type->matrix_columns; i++) { - assert(ir->type->base_type == GLSL_TYPE_FLOAT); - values = &ir->value.f[i * ir->type->vector_elements]; - - src = src_reg(PROGRAM_CONSTANT, -1, NULL); - src.index = _mesa_add_unnamed_constant(this->prog->Parameters, - (gl_constant_value *) values, - ir->type->vector_elements, - &src.swizzle); - emit(ir, OPCODE_MOV, mat_column, src); - - mat_column.index++; - } - - this->result = mat; - return; - } - - src.file = PROGRAM_CONSTANT; - switch (ir->type->base_type) { - case GLSL_TYPE_FLOAT: - values = &ir->value.f[0]; - break; - case GLSL_TYPE_UINT: - for (i = 0; i < ir->type->vector_elements; i++) { - values[i] = ir->value.u[i]; - } - break; - case GLSL_TYPE_INT: - for (i = 0; i < ir->type->vector_elements; i++) { - values[i] = ir->value.i[i]; - } - break; - case GLSL_TYPE_BOOL: - for (i = 0; i < ir->type->vector_elements; i++) { - values[i] = ir->value.b[i]; - } - break; - default: - assert(!"Non-float/uint/int/bool constant"); - } - - this->result = src_reg(PROGRAM_CONSTANT, -1, ir->type); - this->result.index = _mesa_add_unnamed_constant(this->prog->Parameters, - (gl_constant_value *) values, - ir->type->vector_elements, - &this->result.swizzle); -} - -function_entry * -ir_to_mesa_visitor::get_function_signature(ir_function_signature *sig) -{ - function_entry *entry; - - foreach_iter(exec_list_iterator, iter, this->function_signatures) { - entry = (function_entry *)iter.get(); - - if (entry->sig == sig) - return entry; - } - - entry = ralloc(mem_ctx, function_entry); - entry->sig = sig; - entry->sig_id = this->next_signature_id++; - entry->bgn_inst = NULL; - - /* Allocate storage for all the parameters. */ - foreach_iter(exec_list_iterator, iter, sig->parameters) { - ir_variable *param = (ir_variable *)iter.get(); - variable_storage *storage; - - storage = find_variable_storage(param); - assert(!storage); - - storage = new(mem_ctx) variable_storage(param, PROGRAM_TEMPORARY, - this->next_temp); - this->variables.push_tail(storage); - - this->next_temp += type_size(param->type); - } - - if (!sig->return_type->is_void()) { - entry->return_reg = get_temp(sig->return_type); - } else { - entry->return_reg = undef_src; - } - - this->function_signatures.push_tail(entry); - return entry; -} - -void -ir_to_mesa_visitor::visit(ir_call *ir) -{ - ir_to_mesa_instruction *call_inst; - ir_function_signature *sig = ir->get_callee(); - function_entry *entry = get_function_signature(sig); - int i; - - /* Process in parameters. */ - exec_list_iterator sig_iter = sig->parameters.iterator(); - foreach_iter(exec_list_iterator, iter, *ir) { - ir_rvalue *param_rval = (ir_rvalue *)iter.get(); - ir_variable *param = (ir_variable *)sig_iter.get(); - - if (param->mode == ir_var_in || - param->mode == ir_var_inout) { - variable_storage *storage = find_variable_storage(param); - assert(storage); - - param_rval->accept(this); - src_reg r = this->result; - - dst_reg l; - l.file = storage->file; - l.index = storage->index; - l.reladdr = NULL; - l.writemask = WRITEMASK_XYZW; - l.cond_mask = COND_TR; - - for (i = 0; i < type_size(param->type); i++) { - emit(ir, OPCODE_MOV, l, r); - l.index++; - r.index++; - } - } - - sig_iter.next(); - } - assert(!sig_iter.has_next()); - - /* Emit call instruction */ - call_inst = emit(ir, OPCODE_CAL); - call_inst->function = entry; - - /* Process out parameters. */ - sig_iter = sig->parameters.iterator(); - foreach_iter(exec_list_iterator, iter, *ir) { - ir_rvalue *param_rval = (ir_rvalue *)iter.get(); - ir_variable *param = (ir_variable *)sig_iter.get(); - - if (param->mode == ir_var_out || - param->mode == ir_var_inout) { - variable_storage *storage = find_variable_storage(param); - assert(storage); - - src_reg r; - r.file = storage->file; - r.index = storage->index; - r.reladdr = NULL; - r.swizzle = SWIZZLE_NOOP; - r.negate = 0; - - param_rval->accept(this); - dst_reg l = dst_reg(this->result); - - for (i = 0; i < type_size(param->type); i++) { - emit(ir, OPCODE_MOV, l, r); - l.index++; - r.index++; - } - } - - sig_iter.next(); - } - assert(!sig_iter.has_next()); - - /* Process return value. */ - this->result = entry->return_reg; -} - -void -ir_to_mesa_visitor::visit(ir_texture *ir) -{ - src_reg result_src, coord, lod_info, projector, dx, dy; - dst_reg result_dst, coord_dst; - ir_to_mesa_instruction *inst = NULL; - prog_opcode opcode = OPCODE_NOP; - - if (ir->op == ir_txs) - this->result = src_reg_for_float(0.0); - else - ir->coordinate->accept(this); - - /* Put our coords in a temp. We'll need to modify them for shadow, - * projection, or LOD, so the only case we'd use it as is is if - * we're doing plain old texturing. Mesa IR optimization should - * handle cleaning up our mess in that case. - */ - coord = get_temp(glsl_type::vec4_type); - coord_dst = dst_reg(coord); - emit(ir, OPCODE_MOV, coord_dst, this->result); - - if (ir->projector) { - ir->projector->accept(this); - projector = this->result; - } - - /* Storage for our result. Ideally for an assignment we'd be using - * the actual storage for the result here, instead. - */ - result_src = get_temp(glsl_type::vec4_type); - result_dst = dst_reg(result_src); - - switch (ir->op) { - case ir_tex: - case ir_txs: - opcode = OPCODE_TEX; - break; - case ir_txb: - opcode = OPCODE_TXB; - ir->lod_info.bias->accept(this); - lod_info = this->result; - break; - case ir_txl: - opcode = OPCODE_TXL; - ir->lod_info.lod->accept(this); - lod_info = this->result; - break; - case ir_txd: - opcode = OPCODE_TXD; - ir->lod_info.grad.dPdx->accept(this); - dx = this->result; - ir->lod_info.grad.dPdy->accept(this); - dy = this->result; - break; - case ir_txf: - assert(!"GLSL 1.30 features unsupported"); - break; - } - - const glsl_type *sampler_type = ir->sampler->type; - - if (ir->projector) { - if (opcode == OPCODE_TEX) { - /* Slot the projector in as the last component of the coord. */ - coord_dst.writemask = WRITEMASK_W; - emit(ir, OPCODE_MOV, coord_dst, projector); - coord_dst.writemask = WRITEMASK_XYZW; - opcode = OPCODE_TXP; - } else { - src_reg coord_w = coord; - coord_w.swizzle = SWIZZLE_WWWW; - - /* For the other TEX opcodes there's no projective version - * since the last slot is taken up by lod info. Do the - * projective divide now. - */ - coord_dst.writemask = WRITEMASK_W; - emit(ir, OPCODE_RCP, coord_dst, projector); - - /* In the case where we have to project the coordinates "by hand," - * the shadow comparitor value must also be projected. - */ - src_reg tmp_src = coord; - if (ir->shadow_comparitor) { - /* Slot the shadow value in as the second to last component of the - * coord. - */ - ir->shadow_comparitor->accept(this); - - tmp_src = get_temp(glsl_type::vec4_type); - dst_reg tmp_dst = dst_reg(tmp_src); - - /* Projective division not allowed for array samplers. */ - assert(!sampler_type->sampler_array); - - tmp_dst.writemask = WRITEMASK_Z; - emit(ir, OPCODE_MOV, tmp_dst, this->result); - - tmp_dst.writemask = WRITEMASK_XY; - emit(ir, OPCODE_MOV, tmp_dst, coord); - } - - coord_dst.writemask = WRITEMASK_XYZ; - emit(ir, OPCODE_MUL, coord_dst, tmp_src, coord_w); - - coord_dst.writemask = WRITEMASK_XYZW; - coord.swizzle = SWIZZLE_XYZW; - } - } - - /* If projection is done and the opcode is not OPCODE_TXP, then the shadow - * comparitor was put in the correct place (and projected) by the code, - * above, that handles by-hand projection. - */ - if (ir->shadow_comparitor && (!ir->projector || opcode == OPCODE_TXP)) { - /* Slot the shadow value in as the second to last component of the - * coord. - */ - ir->shadow_comparitor->accept(this); - - /* XXX This will need to be updated for cubemap array samplers. */ - if (sampler_type->sampler_dimensionality == GLSL_SAMPLER_DIM_2D && - sampler_type->sampler_array) { - coord_dst.writemask = WRITEMASK_W; - } else { - coord_dst.writemask = WRITEMASK_Z; - } - - emit(ir, OPCODE_MOV, coord_dst, this->result); - coord_dst.writemask = WRITEMASK_XYZW; - } - - if (opcode == OPCODE_TXL || opcode == OPCODE_TXB) { - /* Mesa IR stores lod or lod bias in the last channel of the coords. */ - coord_dst.writemask = WRITEMASK_W; - emit(ir, OPCODE_MOV, coord_dst, lod_info); - coord_dst.writemask = WRITEMASK_XYZW; - } - - if (opcode == OPCODE_TXD) - inst = emit(ir, opcode, result_dst, coord, dx, dy); - else - inst = emit(ir, opcode, result_dst, coord); - - if (ir->shadow_comparitor) - inst->tex_shadow = GL_TRUE; - - inst->sampler = _mesa_get_sampler_uniform_value(ir->sampler, - this->shader_program, - this->prog); - - switch (sampler_type->sampler_dimensionality) { - case GLSL_SAMPLER_DIM_1D: - inst->tex_target = (sampler_type->sampler_array) - ? TEXTURE_1D_ARRAY_INDEX : TEXTURE_1D_INDEX; - break; - case GLSL_SAMPLER_DIM_2D: - inst->tex_target = (sampler_type->sampler_array) - ? TEXTURE_2D_ARRAY_INDEX : TEXTURE_2D_INDEX; - break; - case GLSL_SAMPLER_DIM_3D: - inst->tex_target = TEXTURE_3D_INDEX; - break; - case GLSL_SAMPLER_DIM_CUBE: - inst->tex_target = TEXTURE_CUBE_INDEX; - break; - case GLSL_SAMPLER_DIM_RECT: - inst->tex_target = TEXTURE_RECT_INDEX; - break; - case GLSL_SAMPLER_DIM_BUF: - assert(!"FINISHME: Implement ARB_texture_buffer_object"); - break; - default: - assert(!"Should not get here."); - } - - this->result = result_src; -} - -void -ir_to_mesa_visitor::visit(ir_return *ir) -{ - if (ir->get_value()) { - dst_reg l; - int i; - - assert(current_function); - - ir->get_value()->accept(this); - src_reg r = this->result; - - l = dst_reg(current_function->return_reg); - - for (i = 0; i < type_size(current_function->sig->return_type); i++) { - emit(ir, OPCODE_MOV, l, r); - l.index++; - r.index++; - } - } - - emit(ir, OPCODE_RET); -} - -void -ir_to_mesa_visitor::visit(ir_discard *ir) -{ - struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog; - - if (ir->condition) { - ir->condition->accept(this); - this->result.negate = ~this->result.negate; - emit(ir, OPCODE_KIL, undef_dst, this->result); - } else { - emit(ir, OPCODE_KIL_NV); - } - - fp->UsesKill = GL_TRUE; -} - -void -ir_to_mesa_visitor::visit(ir_if *ir) -{ - ir_to_mesa_instruction *cond_inst, *if_inst; - ir_to_mesa_instruction *prev_inst; - - prev_inst = (ir_to_mesa_instruction *)this->instructions.get_tail(); - - ir->condition->accept(this); - assert(this->result.file != PROGRAM_UNDEFINED); - - if (this->options->EmitCondCodes) { - cond_inst = (ir_to_mesa_instruction *)this->instructions.get_tail(); - - /* See if we actually generated any instruction for generating - * the condition. If not, then cook up a move to a temp so we - * have something to set cond_update on. - */ - if (cond_inst == prev_inst) { - src_reg temp = get_temp(glsl_type::bool_type); - cond_inst = emit(ir->condition, OPCODE_MOV, dst_reg(temp), result); - } - cond_inst->cond_update = GL_TRUE; - - if_inst = emit(ir->condition, OPCODE_IF); - if_inst->dst.cond_mask = COND_NE; - } else { - if_inst = emit(ir->condition, OPCODE_IF, undef_dst, this->result); - } - - this->instructions.push_tail(if_inst); - - visit_exec_list(&ir->then_instructions, this); - - if (!ir->else_instructions.is_empty()) { - emit(ir->condition, OPCODE_ELSE); - visit_exec_list(&ir->else_instructions, this); - } - - if_inst = emit(ir->condition, OPCODE_ENDIF); -} - -ir_to_mesa_visitor::ir_to_mesa_visitor() -{ - result.file = PROGRAM_UNDEFINED; - next_temp = 1; - next_signature_id = 1; - current_function = NULL; - mem_ctx = ralloc_context(NULL); -} - -ir_to_mesa_visitor::~ir_to_mesa_visitor() -{ - ralloc_free(mem_ctx); -} - -static struct prog_src_register -mesa_src_reg_from_ir_src_reg(src_reg reg) -{ - struct prog_src_register mesa_reg; - - mesa_reg.File = reg.file; - assert(reg.index < (1 << INST_INDEX_BITS)); - mesa_reg.Index = reg.index; - mesa_reg.Swizzle = reg.swizzle; - mesa_reg.RelAddr = reg.reladdr != NULL; - mesa_reg.Negate = reg.negate; - mesa_reg.Abs = 0; - mesa_reg.HasIndex2 = GL_FALSE; - mesa_reg.RelAddr2 = 0; - mesa_reg.Index2 = 0; - - return mesa_reg; -} - -static void -set_branchtargets(ir_to_mesa_visitor *v, - struct prog_instruction *mesa_instructions, - int num_instructions) -{ - int if_count = 0, loop_count = 0; - int *if_stack, *loop_stack; - int if_stack_pos = 0, loop_stack_pos = 0; - int i, j; - - for (i = 0; i < num_instructions; i++) { - switch (mesa_instructions[i].Opcode) { - case OPCODE_IF: - if_count++; - break; - case OPCODE_BGNLOOP: - loop_count++; - break; - case OPCODE_BRK: - case OPCODE_CONT: - mesa_instructions[i].BranchTarget = -1; - break; - default: - break; - } - } - - if_stack = rzalloc_array(v->mem_ctx, int, if_count); - loop_stack = rzalloc_array(v->mem_ctx, int, loop_count); - - for (i = 0; i < num_instructions; i++) { - switch (mesa_instructions[i].Opcode) { - case OPCODE_IF: - if_stack[if_stack_pos] = i; - if_stack_pos++; - break; - case OPCODE_ELSE: - mesa_instructions[if_stack[if_stack_pos - 1]].BranchTarget = i; - if_stack[if_stack_pos - 1] = i; - break; - case OPCODE_ENDIF: - mesa_instructions[if_stack[if_stack_pos - 1]].BranchTarget = i; - if_stack_pos--; - break; - case OPCODE_BGNLOOP: - loop_stack[loop_stack_pos] = i; - loop_stack_pos++; - break; - case OPCODE_ENDLOOP: - loop_stack_pos--; - /* Rewrite any breaks/conts at this nesting level (haven't - * already had a BranchTarget assigned) to point to the end - * of the loop. - */ - for (j = loop_stack[loop_stack_pos]; j < i; j++) { - if (mesa_instructions[j].Opcode == OPCODE_BRK || - mesa_instructions[j].Opcode == OPCODE_CONT) { - if (mesa_instructions[j].BranchTarget == -1) { - mesa_instructions[j].BranchTarget = i; - } - } - } - /* The loop ends point at each other. */ - mesa_instructions[i].BranchTarget = loop_stack[loop_stack_pos]; - mesa_instructions[loop_stack[loop_stack_pos]].BranchTarget = i; - break; - case OPCODE_CAL: - foreach_iter(exec_list_iterator, iter, v->function_signatures) { - function_entry *entry = (function_entry *)iter.get(); - - if (entry->sig_id == mesa_instructions[i].BranchTarget) { - mesa_instructions[i].BranchTarget = entry->inst; - break; - } - } - break; - default: - break; - } - } -} - -static void -print_program(struct prog_instruction *mesa_instructions, - ir_instruction **mesa_instruction_annotation, - int num_instructions) -{ - ir_instruction *last_ir = NULL; - int i; - int indent = 0; - - for (i = 0; i < num_instructions; i++) { - struct prog_instruction *mesa_inst = mesa_instructions + i; - ir_instruction *ir = mesa_instruction_annotation[i]; - - fprintf(stdout, "%3d: ", i); - - if (last_ir != ir && ir) { - int j; - - for (j = 0; j < indent; j++) { - fprintf(stdout, " "); - } - ir->print(); - printf("\n"); - last_ir = ir; - - fprintf(stdout, " "); /* line number spacing. */ - } - - indent = _mesa_fprint_instruction_opt(stdout, mesa_inst, indent, - PROG_PRINT_DEBUG, NULL); - } -} - - -/** - * Count resources used by the given gpu program (number of texture - * samplers, etc). - */ -static void -count_resources(struct gl_program *prog) -{ - unsigned int i; - - prog->SamplersUsed = 0; - - for (i = 0; i < prog->NumInstructions; i++) { - struct prog_instruction *inst = &prog->Instructions[i]; - - if (_mesa_is_tex_instruction(inst->Opcode)) { - prog->SamplerTargets[inst->TexSrcUnit] = - (gl_texture_index)inst->TexSrcTarget; - prog->SamplersUsed |= 1 << inst->TexSrcUnit; - if (inst->TexShadow) { - prog->ShadowSamplers |= 1 << inst->TexSrcUnit; - } - } - } - - _mesa_update_shader_textures_used(prog); -} - - -/** - * Check if the given vertex/fragment/shader program is within the - * resource limits of the context (number of texture units, etc). - * If any of those checks fail, record a linker error. - * - * XXX more checks are needed... - */ -static void -check_resources(const struct gl_context *ctx, - struct gl_shader_program *shader_program, - struct gl_program *prog) -{ - switch (prog->Target) { - case GL_VERTEX_PROGRAM_ARB: - if (_mesa_bitcount(prog->SamplersUsed) > - ctx->Const.MaxVertexTextureImageUnits) { - linker_error(shader_program, - "Too many vertex shader texture samplers"); - } - if (prog->Parameters->NumParameters > MAX_UNIFORMS) { - linker_error(shader_program, "Too many vertex shader constants"); - } - break; - case MESA_GEOMETRY_PROGRAM: - if (_mesa_bitcount(prog->SamplersUsed) > - ctx->Const.MaxGeometryTextureImageUnits) { - linker_error(shader_program, - "Too many geometry shader texture samplers"); - } - if (prog->Parameters->NumParameters > - MAX_GEOMETRY_UNIFORM_COMPONENTS / 4) { - linker_error(shader_program, "Too many geometry shader constants"); - } - break; - case GL_FRAGMENT_PROGRAM_ARB: - if (_mesa_bitcount(prog->SamplersUsed) > - ctx->Const.MaxTextureImageUnits) { - linker_error(shader_program, - "Too many fragment shader texture samplers"); - } - if (prog->Parameters->NumParameters > MAX_UNIFORMS) { - linker_error(shader_program, "Too many fragment shader constants"); - } - break; - default: - _mesa_problem(ctx, "unexpected program type in check_resources()"); - } -} - - - -struct uniform_sort { - struct gl_uniform *u; - int pos; -}; - -/* The shader_program->Uniforms list is almost sorted in increasing - * uniform->{Frag,Vert}Pos locations, but not quite when there are - * uniforms shared between targets. We need to add parameters in - * increasing order for the targets. - */ -static int -sort_uniforms(const void *a, const void *b) -{ - struct uniform_sort *u1 = (struct uniform_sort *)a; - struct uniform_sort *u2 = (struct uniform_sort *)b; - - return u1->pos - u2->pos; -} - -/* Add the uniforms to the parameters. The linker chose locations - * in our parameters lists (which weren't created yet), which the - * uniforms code will use to poke values into our parameters list - * when uniforms are updated. - */ -static void -add_uniforms_to_parameters_list(struct gl_shader_program *shader_program, - struct gl_shader *shader, - struct gl_program *prog) -{ - unsigned int i; - unsigned int next_sampler = 0, num_uniforms = 0; - struct uniform_sort *sorted_uniforms; - - sorted_uniforms = ralloc_array(NULL, struct uniform_sort, - shader_program->Uniforms->NumUniforms); - - for (i = 0; i < shader_program->Uniforms->NumUniforms; i++) { - struct gl_uniform *uniform = shader_program->Uniforms->Uniforms + i; - int parameter_index = -1; - - switch (shader->Type) { - case GL_VERTEX_SHADER: - parameter_index = uniform->VertPos; - break; - case GL_FRAGMENT_SHADER: - parameter_index = uniform->FragPos; - break; - case GL_GEOMETRY_SHADER: - parameter_index = uniform->GeomPos; - break; - } - - /* Only add uniforms used in our target. */ - if (parameter_index != -1) { - sorted_uniforms[num_uniforms].pos = parameter_index; - sorted_uniforms[num_uniforms].u = uniform; - num_uniforms++; - } - } - - qsort(sorted_uniforms, num_uniforms, sizeof(struct uniform_sort), - sort_uniforms); - - for (i = 0; i < num_uniforms; i++) { - struct gl_uniform *uniform = sorted_uniforms[i].u; - int parameter_index = sorted_uniforms[i].pos; - const glsl_type *type = uniform->Type; - unsigned int size; - - if (type->is_vector() || - type->is_scalar()) { - size = type->vector_elements; - } else { - size = type_size(type) * 4; - } - - gl_register_file file; - if (type->is_sampler() || - (type->is_array() && type->fields.array->is_sampler())) { - file = PROGRAM_SAMPLER; - } else { - file = PROGRAM_UNIFORM; - } - - GLint index = _mesa_lookup_parameter_index(prog->Parameters, -1, - uniform->Name); - - if (index < 0) { - index = _mesa_add_parameter(prog->Parameters, file, - uniform->Name, size, type->gl_type, - NULL, NULL, 0x0); - - /* Sampler uniform values are stored in prog->SamplerUnits, - * and the entry in that array is selected by this index we - * store in ParameterValues[]. - */ - if (file == PROGRAM_SAMPLER) { - for (unsigned int j = 0; j < size / 4; j++) - prog->Parameters->ParameterValues[index + j][0].f = next_sampler++; - } - - /* The location chosen in the Parameters list here (returned - * from _mesa_add_uniform) has to match what the linker chose. - */ - if (index != parameter_index) { - linker_error(shader_program, - "Allocation of uniform `%s' to target failed " - "(%d vs %d)\n", - uniform->Name, index, parameter_index); - } - } - } - - ralloc_free(sorted_uniforms); -} - -static void -set_uniform_initializer(struct gl_context *ctx, void *mem_ctx, - struct gl_shader_program *shader_program, - const char *name, const glsl_type *type, - ir_constant *val) -{ - if (type->is_record()) { - ir_constant *field_constant; - - field_constant = (ir_constant *)val->components.get_head(); - - for (unsigned int i = 0; i < type->length; i++) { - const glsl_type *field_type = type->fields.structure[i].type; - const char *field_name = ralloc_asprintf(mem_ctx, "%s.%s", name, - type->fields.structure[i].name); - set_uniform_initializer(ctx, mem_ctx, shader_program, field_name, - field_type, field_constant); - field_constant = (ir_constant *)field_constant->next; - } - return; - } - - int loc = _mesa_get_uniform_location(ctx, shader_program, name); - - if (loc == -1) { - linker_error(shader_program, - "Couldn't find uniform for initializer %s\n", name); - return; - } - - for (unsigned int i = 0; i < (type->is_array() ? type->length : 1); i++) { - ir_constant *element; - const glsl_type *element_type; - if (type->is_array()) { - element = val->array_elements[i]; - element_type = type->fields.array; - } else { - element = val; - element_type = type; - } - - void *values; - - if (element_type->base_type == GLSL_TYPE_BOOL) { - int *conv = ralloc_array(mem_ctx, int, element_type->components()); - for (unsigned int j = 0; j < element_type->components(); j++) { - conv[j] = element->value.b[j]; - } - values = (void *)conv; - element_type = glsl_type::get_instance(GLSL_TYPE_INT, - element_type->vector_elements, - 1); - } else { - values = &element->value; - } - - if (element_type->is_matrix()) { - _mesa_uniform_matrix(ctx, shader_program, - element_type->matrix_columns, - element_type->vector_elements, - loc, 1, GL_FALSE, (GLfloat *)values); - loc += element_type->matrix_columns; - } else { - _mesa_uniform(ctx, shader_program, loc, element_type->matrix_columns, - values, element_type->gl_type); - loc += type_size(element_type); - } - } -} - -static void -set_uniform_initializers(struct gl_context *ctx, - struct gl_shader_program *shader_program) -{ - void *mem_ctx = NULL; - - for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) { - struct gl_shader *shader = shader_program->_LinkedShaders[i]; - - if (shader == NULL) - continue; - - foreach_iter(exec_list_iterator, iter, *shader->ir) { - ir_instruction *ir = (ir_instruction *)iter.get(); - ir_variable *var = ir->as_variable(); - - if (!var || var->mode != ir_var_uniform || !var->constant_value) - continue; - - if (!mem_ctx) - mem_ctx = ralloc_context(NULL); - - set_uniform_initializer(ctx, mem_ctx, shader_program, var->name, - var->type, var->constant_value); - } - } - - ralloc_free(mem_ctx); -} - -/* - * On a basic block basis, tracks available PROGRAM_TEMPORARY register - * channels for copy propagation and updates following instructions to - * use the original versions. - * - * The ir_to_mesa_visitor lazily produces code assuming that this pass - * will occur. As an example, a TXP production before this pass: - * - * 0: MOV TEMP[1], INPUT[4].xyyy; - * 1: MOV TEMP[1].w, INPUT[4].wwww; - * 2: TXP TEMP[2], TEMP[1], texture[0], 2D; - * - * and after: - * - * 0: MOV TEMP[1], INPUT[4].xyyy; - * 1: MOV TEMP[1].w, INPUT[4].wwww; - * 2: TXP TEMP[2], INPUT[4].xyyw, texture[0], 2D; - * - * which allows for dead code elimination on TEMP[1]'s writes. - */ -void -ir_to_mesa_visitor::copy_propagate(void) -{ - ir_to_mesa_instruction **acp = rzalloc_array(mem_ctx, - ir_to_mesa_instruction *, - this->next_temp * 4); - int *acp_level = rzalloc_array(mem_ctx, int, this->next_temp * 4); - int level = 0; - - foreach_iter(exec_list_iterator, iter, this->instructions) { - ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get(); - - assert(inst->dst.file != PROGRAM_TEMPORARY - || inst->dst.index < this->next_temp); - - /* First, do any copy propagation possible into the src regs. */ - for (int r = 0; r < 3; r++) { - ir_to_mesa_instruction *first = NULL; - bool good = true; - int acp_base = inst->src[r].index * 4; - - if (inst->src[r].file != PROGRAM_TEMPORARY || - inst->src[r].reladdr) - continue; - - /* See if we can find entries in the ACP consisting of MOVs - * from the same src register for all the swizzled channels - * of this src register reference. - */ - for (int i = 0; i < 4; i++) { - int src_chan = GET_SWZ(inst->src[r].swizzle, i); - ir_to_mesa_instruction *copy_chan = acp[acp_base + src_chan]; - - if (!copy_chan) { - good = false; - break; - } - - assert(acp_level[acp_base + src_chan] <= level); - - if (!first) { - first = copy_chan; - } else { - if (first->src[0].file != copy_chan->src[0].file || - first->src[0].index != copy_chan->src[0].index) { - good = false; - break; - } - } - } - - if (good) { - /* We've now validated that we can copy-propagate to - * replace this src register reference. Do it. - */ - inst->src[r].file = first->src[0].file; - inst->src[r].index = first->src[0].index; - - int swizzle = 0; - for (int i = 0; i < 4; i++) { - int src_chan = GET_SWZ(inst->src[r].swizzle, i); - ir_to_mesa_instruction *copy_inst = acp[acp_base + src_chan]; - swizzle |= (GET_SWZ(copy_inst->src[0].swizzle, src_chan) << - (3 * i)); - } - inst->src[r].swizzle = swizzle; - } - } - - switch (inst->op) { - case OPCODE_BGNLOOP: - case OPCODE_ENDLOOP: - /* End of a basic block, clear the ACP entirely. */ - memset(acp, 0, sizeof(*acp) * this->next_temp * 4); - break; - - case OPCODE_IF: - ++level; - break; - - case OPCODE_ENDIF: - case OPCODE_ELSE: - /* Clear all channels written inside the block from the ACP, but - * leaving those that were not touched. - */ - for (int r = 0; r < this->next_temp; r++) { - for (int c = 0; c < 4; c++) { - if (!acp[4 * r + c]) - continue; - - if (acp_level[4 * r + c] >= level) - acp[4 * r + c] = NULL; - } - } - if (inst->op == OPCODE_ENDIF) - --level; - break; - - default: - /* Continuing the block, clear any written channels from - * the ACP. - */ - if (inst->dst.file == PROGRAM_TEMPORARY && inst->dst.reladdr) { - /* Any temporary might be written, so no copy propagation - * across this instruction. - */ - memset(acp, 0, sizeof(*acp) * this->next_temp * 4); - } else if (inst->dst.file == PROGRAM_OUTPUT && - inst->dst.reladdr) { - /* Any output might be written, so no copy propagation - * from outputs across this instruction. - */ - for (int r = 0; r < this->next_temp; r++) { - for (int c = 0; c < 4; c++) { - if (!acp[4 * r + c]) - continue; - - if (acp[4 * r + c]->src[0].file == PROGRAM_OUTPUT) - acp[4 * r + c] = NULL; - } - } - } else if (inst->dst.file == PROGRAM_TEMPORARY || - inst->dst.file == PROGRAM_OUTPUT) { - /* Clear where it's used as dst. */ - if (inst->dst.file == PROGRAM_TEMPORARY) { - for (int c = 0; c < 4; c++) { - if (inst->dst.writemask & (1 << c)) { - acp[4 * inst->dst.index + c] = NULL; - } - } - } - - /* Clear where it's used as src. */ - for (int r = 0; r < this->next_temp; r++) { - for (int c = 0; c < 4; c++) { - if (!acp[4 * r + c]) - continue; - - int src_chan = GET_SWZ(acp[4 * r + c]->src[0].swizzle, c); - - if (acp[4 * r + c]->src[0].file == inst->dst.file && - acp[4 * r + c]->src[0].index == inst->dst.index && - inst->dst.writemask & (1 << src_chan)) - { - acp[4 * r + c] = NULL; - } - } - } - } - break; - } - - /* If this is a copy, add it to the ACP. */ - if (inst->op == OPCODE_MOV && - inst->dst.file == PROGRAM_TEMPORARY && - !inst->dst.reladdr && - !inst->saturate && - !inst->src[0].reladdr && - !inst->src[0].negate) { - for (int i = 0; i < 4; i++) { - if (inst->dst.writemask & (1 << i)) { - acp[4 * inst->dst.index + i] = inst; - acp_level[4 * inst->dst.index + i] = level; - } - } - } - } - - ralloc_free(acp_level); - ralloc_free(acp); -} - - -/** - * Convert a shader's GLSL IR into a Mesa gl_program. - */ -static struct gl_program * -get_mesa_program(struct gl_context *ctx, - struct gl_shader_program *shader_program, - struct gl_shader *shader) -{ - ir_to_mesa_visitor v; - struct prog_instruction *mesa_instructions, *mesa_inst; - ir_instruction **mesa_instruction_annotation; - int i; - struct gl_program *prog; - GLenum target; - const char *target_string; - GLboolean progress; - struct gl_shader_compiler_options *options = - &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(shader->Type)]; - - switch (shader->Type) { - case GL_VERTEX_SHADER: - target = GL_VERTEX_PROGRAM_ARB; - target_string = "vertex"; - break; - case GL_FRAGMENT_SHADER: - target = GL_FRAGMENT_PROGRAM_ARB; - target_string = "fragment"; - break; - case GL_GEOMETRY_SHADER: - target = GL_GEOMETRY_PROGRAM_NV; - target_string = "geometry"; - break; - default: - assert(!"should not be reached"); - return NULL; - } - - validate_ir_tree(shader->ir); - - prog = ctx->Driver.NewProgram(ctx, target, shader_program->Name); - if (!prog) - return NULL; - prog->Parameters = _mesa_new_parameter_list(); - prog->Varying = _mesa_new_parameter_list(); - prog->Attributes = _mesa_new_parameter_list(); - v.ctx = ctx; - v.prog = prog; - v.shader_program = shader_program; - v.options = options; - - add_uniforms_to_parameters_list(shader_program, shader, prog); - - /* Emit Mesa IR for main(). */ - visit_exec_list(shader->ir, &v); - v.emit(NULL, OPCODE_END); - - /* Now emit bodies for any functions that were used. */ - do { - progress = GL_FALSE; - - foreach_iter(exec_list_iterator, iter, v.function_signatures) { - function_entry *entry = (function_entry *)iter.get(); - - if (!entry->bgn_inst) { - v.current_function = entry; - - entry->bgn_inst = v.emit(NULL, OPCODE_BGNSUB); - entry->bgn_inst->function = entry; - - visit_exec_list(&entry->sig->body, &v); - - ir_to_mesa_instruction *last; - last = (ir_to_mesa_instruction *)v.instructions.get_tail(); - if (last->op != OPCODE_RET) - v.emit(NULL, OPCODE_RET); - - ir_to_mesa_instruction *end; - end = v.emit(NULL, OPCODE_ENDSUB); - end->function = entry; - - progress = GL_TRUE; - } - } - } while (progress); - - prog->NumTemporaries = v.next_temp; - - int num_instructions = 0; - foreach_iter(exec_list_iterator, iter, v.instructions) { - num_instructions++; - } - - mesa_instructions = - (struct prog_instruction *)calloc(num_instructions, - sizeof(*mesa_instructions)); - mesa_instruction_annotation = ralloc_array(v.mem_ctx, ir_instruction *, - num_instructions); - - v.copy_propagate(); - - /* Convert ir_mesa_instructions into prog_instructions. - */ - mesa_inst = mesa_instructions; - i = 0; - foreach_iter(exec_list_iterator, iter, v.instructions) { - const ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get(); - - mesa_inst->Opcode = inst->op; - mesa_inst->CondUpdate = inst->cond_update; - if (inst->saturate) - mesa_inst->SaturateMode = SATURATE_ZERO_ONE; - mesa_inst->DstReg.File = inst->dst.file; - mesa_inst->DstReg.Index = inst->dst.index; - mesa_inst->DstReg.CondMask = inst->dst.cond_mask; - mesa_inst->DstReg.WriteMask = inst->dst.writemask; - mesa_inst->DstReg.RelAddr = inst->dst.reladdr != NULL; - mesa_inst->SrcReg[0] = mesa_src_reg_from_ir_src_reg(inst->src[0]); - mesa_inst->SrcReg[1] = mesa_src_reg_from_ir_src_reg(inst->src[1]); - mesa_inst->SrcReg[2] = mesa_src_reg_from_ir_src_reg(inst->src[2]); - mesa_inst->TexSrcUnit = inst->sampler; - mesa_inst->TexSrcTarget = inst->tex_target; - mesa_inst->TexShadow = inst->tex_shadow; - mesa_instruction_annotation[i] = inst->ir; - - /* Set IndirectRegisterFiles. */ - if (mesa_inst->DstReg.RelAddr) - prog->IndirectRegisterFiles |= 1 << mesa_inst->DstReg.File; - - /* Update program's bitmask of indirectly accessed register files */ - for (unsigned src = 0; src < 3; src++) - if (mesa_inst->SrcReg[src].RelAddr) - prog->IndirectRegisterFiles |= 1 << mesa_inst->SrcReg[src].File; - - switch (mesa_inst->Opcode) { - case OPCODE_IF: - if (options->MaxIfDepth == 0) { - linker_warning(shader_program, - "Couldn't flatten if-statement. " - "This will likely result in software " - "rasterization.\n"); - } - break; - case OPCODE_BGNLOOP: - if (options->EmitNoLoops) { - linker_warning(shader_program, - "Couldn't unroll loop. " - "This will likely result in software " - "rasterization.\n"); - } - break; - case OPCODE_CONT: - if (options->EmitNoCont) { - linker_warning(shader_program, - "Couldn't lower continue-statement. " - "This will likely result in software " - "rasterization.\n"); - } - break; - case OPCODE_BGNSUB: - inst->function->inst = i; - mesa_inst->Comment = strdup(inst->function->sig->function_name()); - break; - case OPCODE_ENDSUB: - mesa_inst->Comment = strdup(inst->function->sig->function_name()); - break; - case OPCODE_CAL: - mesa_inst->BranchTarget = inst->function->sig_id; /* rewritten later */ - break; - case OPCODE_ARL: - prog->NumAddressRegs = 1; - break; - default: - break; - } - - mesa_inst++; - i++; - - if (!shader_program->LinkStatus) - break; - } - - if (!shader_program->LinkStatus) { - free(mesa_instructions); - _mesa_reference_program(ctx, &shader->Program, NULL); - return NULL; - } - - set_branchtargets(&v, mesa_instructions, num_instructions); - - if (ctx->Shader.Flags & GLSL_DUMP) { - printf("\n"); - printf("GLSL IR for linked %s program %d:\n", target_string, - shader_program->Name); - _mesa_print_ir(shader->ir, NULL); - printf("\n"); - printf("\n"); - printf("Mesa IR for linked %s program %d:\n", target_string, - shader_program->Name); - print_program(mesa_instructions, mesa_instruction_annotation, - num_instructions); - } - - prog->Instructions = mesa_instructions; - prog->NumInstructions = num_instructions; - - do_set_program_inouts(shader->ir, prog); - count_resources(prog); - - check_resources(ctx, shader_program, prog); - - _mesa_reference_program(ctx, &shader->Program, prog); - - if ((ctx->Shader.Flags & GLSL_NO_OPT) == 0) { - _mesa_optimize_program(ctx, prog); - } - - return prog; -} - -extern "C" { - -/** - * Link a shader. - * Called via ctx->Driver.LinkShader() - * This actually involves converting GLSL IR into Mesa gl_programs with - * code lowering and other optimizations. - */ -GLboolean -_mesa_ir_link_shader(struct gl_context *ctx, struct gl_shader_program *prog) -{ - assert(prog->LinkStatus); - - for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - bool progress; - exec_list *ir = prog->_LinkedShaders[i]->ir; - const struct gl_shader_compiler_options *options = - &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(prog->_LinkedShaders[i]->Type)]; - - do { - progress = false; - - /* Lowering */ - do_mat_op_to_vec(ir); - lower_instructions(ir, (MOD_TO_FRACT | DIV_TO_MUL_RCP | EXP_TO_EXP2 - | LOG_TO_LOG2 | INT_DIV_TO_MUL_RCP - | ((options->EmitNoPow) ? POW_TO_EXP2 : 0))); - - progress = do_lower_jumps(ir, true, true, options->EmitNoMainReturn, options->EmitNoCont, options->EmitNoLoops) || progress; - - progress = do_common_optimization(ir, true, options->MaxUnrollIterations) || progress; - - progress = lower_quadop_vector(ir, true) || progress; - - if (options->MaxIfDepth == 0) - progress = lower_discard(ir) || progress; - - progress = lower_if_to_cond_assign(ir, options->MaxIfDepth) || progress; - - if (options->EmitNoNoise) - progress = lower_noise(ir) || progress; - - /* If there are forms of indirect addressing that the driver - * cannot handle, perform the lowering pass. - */ - if (options->EmitNoIndirectInput || options->EmitNoIndirectOutput - || options->EmitNoIndirectTemp || options->EmitNoIndirectUniform) - progress = - lower_variable_index_to_cond_assign(ir, - options->EmitNoIndirectInput, - options->EmitNoIndirectOutput, - options->EmitNoIndirectTemp, - options->EmitNoIndirectUniform) - || progress; - - progress = do_vec_index_to_cond_assign(ir) || progress; - } while (progress); - - validate_ir_tree(ir); - } - - for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { - struct gl_program *linked_prog; - - if (prog->_LinkedShaders[i] == NULL) - continue; - - linked_prog = get_mesa_program(ctx, prog, prog->_LinkedShaders[i]); - - if (linked_prog) { - bool ok = true; - - switch (prog->_LinkedShaders[i]->Type) { - case GL_VERTEX_SHADER: - _mesa_reference_vertprog(ctx, &prog->VertexProgram, - (struct gl_vertex_program *)linked_prog); - ok = ctx->Driver.ProgramStringNotify(ctx, GL_VERTEX_PROGRAM_ARB, - linked_prog); - break; - case GL_FRAGMENT_SHADER: - _mesa_reference_fragprog(ctx, &prog->FragmentProgram, - (struct gl_fragment_program *)linked_prog); - ok = ctx->Driver.ProgramStringNotify(ctx, GL_FRAGMENT_PROGRAM_ARB, - linked_prog); - break; - case GL_GEOMETRY_SHADER: - _mesa_reference_geomprog(ctx, &prog->GeometryProgram, - (struct gl_geometry_program *)linked_prog); - ok = ctx->Driver.ProgramStringNotify(ctx, GL_GEOMETRY_PROGRAM_NV, - linked_prog); - break; - } - if (!ok) { - return GL_FALSE; - } - } - - _mesa_reference_program(ctx, &linked_prog, NULL); - } - - return GL_TRUE; -} - - -/** - * Compile a GLSL shader. Called via glCompileShader(). - */ -void -_mesa_glsl_compile_shader(struct gl_context *ctx, struct gl_shader *shader) -{ - struct _mesa_glsl_parse_state *state = - new(shader) _mesa_glsl_parse_state(ctx, shader->Type, shader); - - const char *source = shader->Source; - /* Check if the user called glCompileShader without first calling - * glShaderSource. This should fail to compile, but not raise a GL_ERROR. - */ - if (source == NULL) { - shader->CompileStatus = GL_FALSE; - return; - } - - state->error = preprocess(state, &source, &state->info_log, - &ctx->Extensions, ctx->API); - - if (ctx->Shader.Flags & GLSL_DUMP) { - printf("GLSL source for %s shader %d:\n", - _mesa_glsl_shader_target_name(state->target), shader->Name); - printf("%s\n", shader->Source); - } - - if (!state->error) { - _mesa_glsl_lexer_ctor(state, source); - _mesa_glsl_parse(state); - _mesa_glsl_lexer_dtor(state); - } - - ralloc_free(shader->ir); - shader->ir = new(shader) exec_list; - if (!state->error && !state->translation_unit.is_empty()) - _mesa_ast_to_hir(shader->ir, state); - - if (!state->error && !shader->ir->is_empty()) { - validate_ir_tree(shader->ir); - - /* Do some optimization at compile time to reduce shader IR size - * and reduce later work if the same shader is linked multiple times - */ - while (do_common_optimization(shader->ir, false, 32)) - ; - - validate_ir_tree(shader->ir); - } - - shader->symbols = state->symbols; - - shader->CompileStatus = !state->error; - shader->InfoLog = state->info_log; - shader->Version = state->language_version; - memcpy(shader->builtins_to_link, state->builtins_to_link, - sizeof(shader->builtins_to_link[0]) * state->num_builtins_to_link); - shader->num_builtins_to_link = state->num_builtins_to_link; - - if (ctx->Shader.Flags & GLSL_LOG) { - _mesa_write_shader_to_file(shader); - } - - if (ctx->Shader.Flags & GLSL_DUMP) { - if (shader->CompileStatus) { - printf("GLSL IR for shader %d:\n", shader->Name); - _mesa_print_ir(shader->ir, NULL); - printf("\n\n"); - } else { - printf("GLSL shader %d failed to compile.\n", shader->Name); - } - if (shader->InfoLog && shader->InfoLog[0] != 0) { - printf("GLSL shader %d info log:\n", shader->Name); - printf("%s\n", shader->InfoLog); - } - } - - /* Retain any live IR, but trash the rest. */ - reparent_ir(shader->ir, shader->ir); - - ralloc_free(state); -} - - -/** - * Link a GLSL shader program. Called via glLinkProgram(). - */ -void -_mesa_glsl_link_shader(struct gl_context *ctx, struct gl_shader_program *prog) -{ - unsigned int i; - - _mesa_clear_shader_program_data(ctx, prog); - - prog->LinkStatus = GL_TRUE; - - for (i = 0; i < prog->NumShaders; i++) { - if (!prog->Shaders[i]->CompileStatus) { - linker_error(prog, "linking with uncompiled shader"); - prog->LinkStatus = GL_FALSE; - } - } - - prog->Varying = _mesa_new_parameter_list(); - _mesa_reference_vertprog(ctx, &prog->VertexProgram, NULL); - _mesa_reference_fragprog(ctx, &prog->FragmentProgram, NULL); - _mesa_reference_geomprog(ctx, &prog->GeometryProgram, NULL); - - if (prog->LinkStatus) { - link_shaders(ctx, prog); - } - - if (prog->LinkStatus) { - if (!ctx->Driver.LinkShader(ctx, prog)) { - prog->LinkStatus = GL_FALSE; - } - } - - set_uniform_initializers(ctx, prog); - - if (ctx->Shader.Flags & GLSL_DUMP) { - if (!prog->LinkStatus) { - printf("GLSL shader program %d failed to link\n", prog->Name); - } - - if (prog->InfoLog && prog->InfoLog[0] != 0) { - printf("GLSL shader program %d info log:\n", prog->Name); - printf("%s\n", prog->InfoLog); - } - } -} - -} /* extern "C" */ +/* + * Copyright (C) 2005-2007 Brian Paul All Rights Reserved. + * Copyright (C) 2008 VMware, Inc. All Rights Reserved. + * Copyright © 2010 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +/** + * \file ir_to_mesa.cpp + * + * Translate GLSL IR to Mesa's gl_program representation. + */ + +#include +#include "main/compiler.h" +#include "ir.h" +#include "ir_visitor.h" +#include "ir_print_visitor.h" +#include "ir_expression_flattening.h" +#include "glsl_types.h" +#include "glsl_parser_extras.h" +#include "../glsl/program.h" +#include "ir_optimization.h" +#include "ast.h" + +extern "C" { +#include "main/mtypes.h" +#include "main/shaderapi.h" +#include "main/shaderobj.h" +#include "main/uniforms.h" +#include "program/hash_table.h" +#include "program/prog_instruction.h" +#include "program/prog_optimize.h" +#include "program/prog_print.h" +#include "program/program.h" +#include "program/prog_uniform.h" +#include "program/prog_parameter.h" +#include "program/sampler.h" +} + +class src_reg; +class dst_reg; + +static int swizzle_for_size(int size); + +/** + * This struct is a corresponding struct to Mesa prog_src_register, with + * wider fields. + */ +class src_reg { +public: + src_reg(gl_register_file file, int index, const glsl_type *type) + { + this->file = file; + this->index = index; + if (type && (type->is_scalar() || type->is_vector() || type->is_matrix())) + this->swizzle = swizzle_for_size(type->vector_elements); + else + this->swizzle = SWIZZLE_XYZW; + this->negate = 0; + this->reladdr = NULL; + } + + src_reg() + { + this->file = PROGRAM_UNDEFINED; + this->index = 0; + this->swizzle = 0; + this->negate = 0; + this->reladdr = NULL; + } + + explicit src_reg(dst_reg reg); + + gl_register_file file; /**< PROGRAM_* from Mesa */ + int index; /**< temporary index, VERT_ATTRIB_*, FRAG_ATTRIB_*, etc. */ + GLuint swizzle; /**< SWIZZLE_XYZWONEZERO swizzles from Mesa. */ + int negate; /**< NEGATE_XYZW mask from mesa */ + /** Register index should be offset by the integer in this reg. */ + src_reg *reladdr; +}; + +class dst_reg { +public: + dst_reg(gl_register_file file, int writemask) + { + this->file = file; + this->index = 0; + this->writemask = writemask; + this->cond_mask = COND_TR; + this->reladdr = NULL; + } + + dst_reg() + { + this->file = PROGRAM_UNDEFINED; + this->index = 0; + this->writemask = 0; + this->cond_mask = COND_TR; + this->reladdr = NULL; + } + + explicit dst_reg(src_reg reg); + + gl_register_file file; /**< PROGRAM_* from Mesa */ + int index; /**< temporary index, VERT_ATTRIB_*, FRAG_ATTRIB_*, etc. */ + int writemask; /**< Bitfield of WRITEMASK_[XYZW] */ + GLuint cond_mask:4; + /** Register index should be offset by the integer in this reg. */ + src_reg *reladdr; +}; + +src_reg::src_reg(dst_reg reg) +{ + this->file = reg.file; + this->index = reg.index; + this->swizzle = SWIZZLE_XYZW; + this->negate = 0; + this->reladdr = reg.reladdr; +} + +dst_reg::dst_reg(src_reg reg) +{ + this->file = reg.file; + this->index = reg.index; + this->writemask = WRITEMASK_XYZW; + this->cond_mask = COND_TR; + this->reladdr = reg.reladdr; +} + +class ir_to_mesa_instruction : public exec_node { +public: + /* Callers of this ralloc-based new need not call delete. It's + * easier to just ralloc_free 'ctx' (or any of its ancestors). */ + static void* operator new(size_t size, void *ctx) + { + void *node; + + node = rzalloc_size(ctx, size); + assert(node != NULL); + + return node; + } + + enum prog_opcode op; + dst_reg dst; + src_reg src[3]; + /** Pointer to the ir source this tree came from for debugging */ + ir_instruction *ir; + GLboolean cond_update; + bool saturate; + int sampler; /**< sampler index */ + int tex_target; /**< One of TEXTURE_*_INDEX */ + GLboolean tex_shadow; + + class function_entry *function; /* Set on OPCODE_CAL or OPCODE_BGNSUB */ +}; + +class variable_storage : public exec_node { +public: + variable_storage(ir_variable *var, gl_register_file file, int index) + : file(file), index(index), var(var) + { + /* empty */ + } + + gl_register_file file; + int index; + ir_variable *var; /* variable that maps to this, if any */ +}; + +class function_entry : public exec_node { +public: + ir_function_signature *sig; + + /** + * identifier of this function signature used by the program. + * + * At the point that Mesa instructions for function calls are + * generated, we don't know the address of the first instruction of + * the function body. So we make the BranchTarget that is called a + * small integer and rewrite them during set_branchtargets(). + */ + int sig_id; + + /** + * Pointer to first instruction of the function body. + * + * Set during function body emits after main() is processed. + */ + ir_to_mesa_instruction *bgn_inst; + + /** + * Index of the first instruction of the function body in actual + * Mesa IR. + * + * Set after convertion from ir_to_mesa_instruction to prog_instruction. + */ + int inst; + + /** Storage for the return value. */ + src_reg return_reg; +}; + +class ir_to_mesa_visitor : public ir_visitor { +public: + ir_to_mesa_visitor(); + ~ir_to_mesa_visitor(); + + function_entry *current_function; + + struct gl_context *ctx; + struct gl_program *prog; + struct gl_shader_program *shader_program; + struct gl_shader_compiler_options *options; + + int next_temp; + + variable_storage *find_variable_storage(ir_variable *var); + + function_entry *get_function_signature(ir_function_signature *sig); + + src_reg get_temp(const glsl_type *type); + void reladdr_to_temp(ir_instruction *ir, src_reg *reg, int *num_reladdr); + + src_reg src_reg_for_float(float val); + + /** + * \name Visit methods + * + * As typical for the visitor pattern, there must be one \c visit method for + * each concrete subclass of \c ir_instruction. Virtual base classes within + * the hierarchy should not have \c visit methods. + */ + /*@{*/ + virtual void visit(ir_variable *); + virtual void visit(ir_loop *); + virtual void visit(ir_loop_jump *); + virtual void visit(ir_function_signature *); + virtual void visit(ir_function *); + virtual void visit(ir_expression *); + virtual void visit(ir_swizzle *); + virtual void visit(ir_dereference_variable *); + virtual void visit(ir_dereference_array *); + virtual void visit(ir_dereference_record *); + virtual void visit(ir_assignment *); + virtual void visit(ir_constant *); + virtual void visit(ir_call *); + virtual void visit(ir_return *); + virtual void visit(ir_discard *); + virtual void visit(ir_texture *); + virtual void visit(ir_if *); + /*@}*/ + + src_reg result; + + /** List of variable_storage */ + exec_list variables; + + /** List of function_entry */ + exec_list function_signatures; + int next_signature_id; + + /** List of ir_to_mesa_instruction */ + exec_list instructions; + + ir_to_mesa_instruction *emit(ir_instruction *ir, enum prog_opcode op); + + ir_to_mesa_instruction *emit(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, src_reg src0); + + ir_to_mesa_instruction *emit(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, src_reg src0, src_reg src1); + + ir_to_mesa_instruction *emit(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, + src_reg src0, src_reg src1, src_reg src2); + + /** + * Emit the correct dot-product instruction for the type of arguments + */ + ir_to_mesa_instruction * emit_dp(ir_instruction *ir, + dst_reg dst, + src_reg src0, + src_reg src1, + unsigned elements); + + void emit_scalar(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, src_reg src0); + + void emit_scalar(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, src_reg src0, src_reg src1); + + void emit_scs(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, const src_reg &src); + + bool try_emit_mad(ir_expression *ir, + int mul_operand); + bool try_emit_mad_for_and_not(ir_expression *ir, + int mul_operand); + bool try_emit_sat(ir_expression *ir); + + void emit_swz(ir_expression *ir); + + bool process_move_condition(ir_rvalue *ir); + + void copy_propagate(void); + + void *mem_ctx; +}; + +src_reg undef_src = src_reg(PROGRAM_UNDEFINED, 0, NULL); + +dst_reg undef_dst = dst_reg(PROGRAM_UNDEFINED, SWIZZLE_NOOP); + +dst_reg address_reg = dst_reg(PROGRAM_ADDRESS, WRITEMASK_X); + +static int +swizzle_for_size(int size) +{ + int size_swizzles[4] = { + MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), + MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y), + MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z), + MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W), + }; + + assert((size >= 1) && (size <= 4)); + return size_swizzles[size - 1]; +} + +ir_to_mesa_instruction * +ir_to_mesa_visitor::emit(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, + src_reg src0, src_reg src1, src_reg src2) +{ + ir_to_mesa_instruction *inst = new(mem_ctx) ir_to_mesa_instruction(); + int num_reladdr = 0; + + /* If we have to do relative addressing, we want to load the ARL + * reg directly for one of the regs, and preload the other reladdr + * sources into temps. + */ + num_reladdr += dst.reladdr != NULL; + num_reladdr += src0.reladdr != NULL; + num_reladdr += src1.reladdr != NULL; + num_reladdr += src2.reladdr != NULL; + + reladdr_to_temp(ir, &src2, &num_reladdr); + reladdr_to_temp(ir, &src1, &num_reladdr); + reladdr_to_temp(ir, &src0, &num_reladdr); + + if (dst.reladdr) { + emit(ir, OPCODE_ARL, address_reg, *dst.reladdr); + num_reladdr--; + } + assert(num_reladdr == 0); + + inst->op = op; + inst->dst = dst; + inst->src[0] = src0; + inst->src[1] = src1; + inst->src[2] = src2; + inst->ir = ir; + + inst->function = NULL; + + this->instructions.push_tail(inst); + + return inst; +} + + +ir_to_mesa_instruction * +ir_to_mesa_visitor::emit(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, src_reg src0, src_reg src1) +{ + return emit(ir, op, dst, src0, src1, undef_src); +} + +ir_to_mesa_instruction * +ir_to_mesa_visitor::emit(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, src_reg src0) +{ + assert(dst.writemask != 0); + return emit(ir, op, dst, src0, undef_src, undef_src); +} + +ir_to_mesa_instruction * +ir_to_mesa_visitor::emit(ir_instruction *ir, enum prog_opcode op) +{ + return emit(ir, op, undef_dst, undef_src, undef_src, undef_src); +} + +ir_to_mesa_instruction * +ir_to_mesa_visitor::emit_dp(ir_instruction *ir, + dst_reg dst, src_reg src0, src_reg src1, + unsigned elements) +{ + static const gl_inst_opcode dot_opcodes[] = { + OPCODE_DP2, OPCODE_DP3, OPCODE_DP4 + }; + + return emit(ir, dot_opcodes[elements - 2], dst, src0, src1); +} + +/** + * Emits Mesa scalar opcodes to produce unique answers across channels. + * + * Some Mesa opcodes are scalar-only, like ARB_fp/vp. The src X + * channel determines the result across all channels. So to do a vec4 + * of this operation, we want to emit a scalar per source channel used + * to produce dest channels. + */ +void +ir_to_mesa_visitor::emit_scalar(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, + src_reg orig_src0, src_reg orig_src1) +{ + int i, j; + int done_mask = ~dst.writemask; + + /* Mesa RCP is a scalar operation splatting results to all channels, + * like ARB_fp/vp. So emit as many RCPs as necessary to cover our + * dst channels. + */ + for (i = 0; i < 4; i++) { + GLuint this_mask = (1 << i); + ir_to_mesa_instruction *inst; + src_reg src0 = orig_src0; + src_reg src1 = orig_src1; + + if (done_mask & this_mask) + continue; + + GLuint src0_swiz = GET_SWZ(src0.swizzle, i); + GLuint src1_swiz = GET_SWZ(src1.swizzle, i); + for (j = i + 1; j < 4; j++) { + /* If there is another enabled component in the destination that is + * derived from the same inputs, generate its value on this pass as + * well. + */ + if (!(done_mask & (1 << j)) && + GET_SWZ(src0.swizzle, j) == src0_swiz && + GET_SWZ(src1.swizzle, j) == src1_swiz) { + this_mask |= (1 << j); + } + } + src0.swizzle = MAKE_SWIZZLE4(src0_swiz, src0_swiz, + src0_swiz, src0_swiz); + src1.swizzle = MAKE_SWIZZLE4(src1_swiz, src1_swiz, + src1_swiz, src1_swiz); + + inst = emit(ir, op, dst, src0, src1); + inst->dst.writemask = this_mask; + done_mask |= this_mask; + } +} + +void +ir_to_mesa_visitor::emit_scalar(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, src_reg src0) +{ + src_reg undef = undef_src; + + undef.swizzle = SWIZZLE_XXXX; + + emit_scalar(ir, op, dst, src0, undef); +} + +/** + * Emit an OPCODE_SCS instruction + * + * The \c SCS opcode functions a bit differently than the other Mesa (or + * ARB_fragment_program) opcodes. Instead of splatting its result across all + * four components of the destination, it writes one value to the \c x + * component and another value to the \c y component. + * + * \param ir IR instruction being processed + * \param op Either \c OPCODE_SIN or \c OPCODE_COS depending on which + * value is desired. + * \param dst Destination register + * \param src Source register + */ +void +ir_to_mesa_visitor::emit_scs(ir_instruction *ir, enum prog_opcode op, + dst_reg dst, + const src_reg &src) +{ + /* Vertex programs cannot use the SCS opcode. + */ + if (this->prog->Target == GL_VERTEX_PROGRAM_ARB) { + emit_scalar(ir, op, dst, src); + return; + } + + const unsigned component = (op == OPCODE_SIN) ? 0 : 1; + const unsigned scs_mask = (1U << component); + int done_mask = ~dst.writemask; + src_reg tmp; + + assert(op == OPCODE_SIN || op == OPCODE_COS); + + /* If there are compnents in the destination that differ from the component + * that will be written by the SCS instrution, we'll need a temporary. + */ + if (scs_mask != unsigned(dst.writemask)) { + tmp = get_temp(glsl_type::vec4_type); + } + + for (unsigned i = 0; i < 4; i++) { + unsigned this_mask = (1U << i); + src_reg src0 = src; + + if ((done_mask & this_mask) != 0) + continue; + + /* The source swizzle specified which component of the source generates + * sine / cosine for the current component in the destination. The SCS + * instruction requires that this value be swizzle to the X component. + * Replace the current swizzle with a swizzle that puts the source in + * the X component. + */ + unsigned src0_swiz = GET_SWZ(src.swizzle, i); + + src0.swizzle = MAKE_SWIZZLE4(src0_swiz, src0_swiz, + src0_swiz, src0_swiz); + for (unsigned j = i + 1; j < 4; j++) { + /* If there is another enabled component in the destination that is + * derived from the same inputs, generate its value on this pass as + * well. + */ + if (!(done_mask & (1 << j)) && + GET_SWZ(src0.swizzle, j) == src0_swiz) { + this_mask |= (1 << j); + } + } + + if (this_mask != scs_mask) { + ir_to_mesa_instruction *inst; + dst_reg tmp_dst = dst_reg(tmp); + + /* Emit the SCS instruction. + */ + inst = emit(ir, OPCODE_SCS, tmp_dst, src0); + inst->dst.writemask = scs_mask; + + /* Move the result of the SCS instruction to the desired location in + * the destination. + */ + tmp.swizzle = MAKE_SWIZZLE4(component, component, + component, component); + inst = emit(ir, OPCODE_SCS, dst, tmp); + inst->dst.writemask = this_mask; + } else { + /* Emit the SCS instruction to write directly to the destination. + */ + ir_to_mesa_instruction *inst = emit(ir, OPCODE_SCS, dst, src0); + inst->dst.writemask = scs_mask; + } + + done_mask |= this_mask; + } +} + +src_reg +ir_to_mesa_visitor::src_reg_for_float(float val) +{ + src_reg src(PROGRAM_CONSTANT, -1, NULL); + + src.index = _mesa_add_unnamed_constant(this->prog->Parameters, + (const gl_constant_value *)&val, 1, &src.swizzle); + + return src; +} + +static int +type_size(const struct glsl_type *type) +{ + unsigned int i; + int size; + + switch (type->base_type) { + case GLSL_TYPE_UINT: + case GLSL_TYPE_INT: + case GLSL_TYPE_FLOAT: + case GLSL_TYPE_BOOL: + if (type->is_matrix()) { + return type->matrix_columns; + } else { + /* Regardless of size of vector, it gets a vec4. This is bad + * packing for things like floats, but otherwise arrays become a + * mess. Hopefully a later pass over the code can pack scalars + * down if appropriate. + */ + return 1; + } + case GLSL_TYPE_ARRAY: + assert(type->length > 0); + return type_size(type->fields.array) * type->length; + case GLSL_TYPE_STRUCT: + size = 0; + for (i = 0; i < type->length; i++) { + size += type_size(type->fields.structure[i].type); + } + return size; + case GLSL_TYPE_SAMPLER: + /* Samplers take up one slot in UNIFORMS[], but they're baked in + * at link time. + */ + return 1; + default: + assert(0); + return 0; + } +} + +/** + * In the initial pass of codegen, we assign temporary numbers to + * intermediate results. (not SSA -- variable assignments will reuse + * storage). Actual register allocation for the Mesa VM occurs in a + * pass over the Mesa IR later. + */ +src_reg +ir_to_mesa_visitor::get_temp(const glsl_type *type) +{ + src_reg src; + + src.file = PROGRAM_TEMPORARY; + src.index = next_temp; + src.reladdr = NULL; + next_temp += type_size(type); + + if (type->is_array() || type->is_record()) { + src.swizzle = SWIZZLE_NOOP; + } else { + src.swizzle = swizzle_for_size(type->vector_elements); + } + src.negate = 0; + + return src; +} + +variable_storage * +ir_to_mesa_visitor::find_variable_storage(ir_variable *var) +{ + + variable_storage *entry; + + foreach_iter(exec_list_iterator, iter, this->variables) { + entry = (variable_storage *)iter.get(); + + if (entry->var == var) + return entry; + } + + return NULL; +} + +void +ir_to_mesa_visitor::visit(ir_variable *ir) +{ + if (strcmp(ir->name, "gl_FragCoord") == 0) { + struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog; + + fp->OriginUpperLeft = ir->origin_upper_left; + fp->PixelCenterInteger = ir->pixel_center_integer; + + } else if (strcmp(ir->name, "gl_FragDepth") == 0) { + struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog; + switch (ir->depth_layout) { + case ir_depth_layout_none: + fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_NONE; + break; + case ir_depth_layout_any: + fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_ANY; + break; + case ir_depth_layout_greater: + fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_GREATER; + break; + case ir_depth_layout_less: + fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_LESS; + break; + case ir_depth_layout_unchanged: + fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_UNCHANGED; + break; + default: + assert(0); + break; + } + } + + if (ir->mode == ir_var_uniform && strncmp(ir->name, "gl_", 3) == 0) { + unsigned int i; + const ir_state_slot *const slots = ir->state_slots; + assert(ir->state_slots != NULL); + + /* Check if this statevar's setup in the STATE file exactly + * matches how we'll want to reference it as a + * struct/array/whatever. If not, then we need to move it into + * temporary storage and hope that it'll get copy-propagated + * out. + */ + for (i = 0; i < ir->num_state_slots; i++) { + if (slots[i].swizzle != SWIZZLE_XYZW) { + break; + } + } + + variable_storage *storage; + dst_reg dst; + if (i == ir->num_state_slots) { + /* We'll set the index later. */ + storage = new(mem_ctx) variable_storage(ir, PROGRAM_STATE_VAR, -1); + this->variables.push_tail(storage); + + dst = undef_dst; + } else { + /* The variable_storage constructor allocates slots based on the size + * of the type. However, this had better match the number of state + * elements that we're going to copy into the new temporary. + */ + assert((int) ir->num_state_slots == type_size(ir->type)); + + storage = new(mem_ctx) variable_storage(ir, PROGRAM_TEMPORARY, + this->next_temp); + this->variables.push_tail(storage); + this->next_temp += type_size(ir->type); + + dst = dst_reg(src_reg(PROGRAM_TEMPORARY, storage->index, NULL)); + } + + + for (unsigned int i = 0; i < ir->num_state_slots; i++) { + int index = _mesa_add_state_reference(this->prog->Parameters, + (gl_state_index *)slots[i].tokens); + + if (storage->file == PROGRAM_STATE_VAR) { + if (storage->index == -1) { + storage->index = index; + } else { + assert(index == storage->index + (int)i); + } + } else { + src_reg src(PROGRAM_STATE_VAR, index, NULL); + src.swizzle = slots[i].swizzle; + emit(ir, OPCODE_MOV, dst, src); + /* even a float takes up a whole vec4 reg in a struct/array. */ + dst.index++; + } + } + + if (storage->file == PROGRAM_TEMPORARY && + dst.index != storage->index + (int) ir->num_state_slots) { + linker_error(this->shader_program, + "failed to load builtin uniform `%s' " + "(%d/%d regs loaded)\n", + ir->name, dst.index - storage->index, + type_size(ir->type)); + } + } +} + +void +ir_to_mesa_visitor::visit(ir_loop *ir) +{ + ir_dereference_variable *counter = NULL; + + if (ir->counter != NULL) + counter = new(mem_ctx) ir_dereference_variable(ir->counter); + + if (ir->from != NULL) { + assert(ir->counter != NULL); + + ir_assignment *a = + new(mem_ctx) ir_assignment(counter, ir->from, NULL); + + a->accept(this); + } + + emit(NULL, OPCODE_BGNLOOP); + + if (ir->to) { + ir_expression *e = + new(mem_ctx) ir_expression(ir->cmp, glsl_type::bool_type, + counter, ir->to); + ir_if *if_stmt = new(mem_ctx) ir_if(e); + + ir_loop_jump *brk = + new(mem_ctx) ir_loop_jump(ir_loop_jump::jump_break); + + if_stmt->then_instructions.push_tail(brk); + + if_stmt->accept(this); + } + + visit_exec_list(&ir->body_instructions, this); + + if (ir->increment) { + ir_expression *e = + new(mem_ctx) ir_expression(ir_binop_add, counter->type, + counter, ir->increment); + + ir_assignment *a = + new(mem_ctx) ir_assignment(counter, e, NULL); + + a->accept(this); + } + + emit(NULL, OPCODE_ENDLOOP); +} + +void +ir_to_mesa_visitor::visit(ir_loop_jump *ir) +{ + switch (ir->mode) { + case ir_loop_jump::jump_break: + emit(NULL, OPCODE_BRK); + break; + case ir_loop_jump::jump_continue: + emit(NULL, OPCODE_CONT); + break; + } +} + + +void +ir_to_mesa_visitor::visit(ir_function_signature *ir) +{ + assert(0); + (void)ir; +} + +void +ir_to_mesa_visitor::visit(ir_function *ir) +{ + /* Ignore function bodies other than main() -- we shouldn't see calls to + * them since they should all be inlined before we get to ir_to_mesa. + */ + if (strcmp(ir->name, "main") == 0) { + const ir_function_signature *sig; + exec_list empty; + + sig = ir->matching_signature(&empty); + + assert(sig); + + foreach_iter(exec_list_iterator, iter, sig->body) { + ir_instruction *ir = (ir_instruction *)iter.get(); + + ir->accept(this); + } + } +} + +bool +ir_to_mesa_visitor::try_emit_mad(ir_expression *ir, int mul_operand) +{ + int nonmul_operand = 1 - mul_operand; + src_reg a, b, c; + + ir_expression *expr = ir->operands[mul_operand]->as_expression(); + if (!expr || expr->operation != ir_binop_mul) + return false; + + expr->operands[0]->accept(this); + a = this->result; + expr->operands[1]->accept(this); + b = this->result; + ir->operands[nonmul_operand]->accept(this); + c = this->result; + + this->result = get_temp(ir->type); + emit(ir, OPCODE_MAD, dst_reg(this->result), a, b, c); + + return true; +} + +/** + * Emit OPCODE_MAD(a, -b, a) instead of AND(a, NOT(b)) + * + * The logic values are 1.0 for true and 0.0 for false. Logical-and is + * implemented using multiplication, and logical-or is implemented using + * addition. Logical-not can be implemented as (true - x), or (1.0 - x). + * As result, the logical expression (a & !b) can be rewritten as: + * + * - a * !b + * - a * (1 - b) + * - (a * 1) - (a * b) + * - a + -(a * b) + * - a + (a * -b) + * + * This final expression can be implemented as a single MAD(a, -b, a) + * instruction. + */ +bool +ir_to_mesa_visitor::try_emit_mad_for_and_not(ir_expression *ir, int try_operand) +{ + const int other_operand = 1 - try_operand; + src_reg a, b; + + ir_expression *expr = ir->operands[try_operand]->as_expression(); + if (!expr || expr->operation != ir_unop_logic_not) + return false; + + ir->operands[other_operand]->accept(this); + a = this->result; + expr->operands[0]->accept(this); + b = this->result; + + b.negate = ~b.negate; + + this->result = get_temp(ir->type); + emit(ir, OPCODE_MAD, dst_reg(this->result), a, b, a); + + return true; +} + +bool +ir_to_mesa_visitor::try_emit_sat(ir_expression *ir) +{ + /* Saturates were only introduced to vertex programs in + * NV_vertex_program3, so don't give them to drivers in the VP. + */ + if (this->prog->Target == GL_VERTEX_PROGRAM_ARB) + return false; + + ir_rvalue *sat_src = ir->as_rvalue_to_saturate(); + if (!sat_src) + return false; + + sat_src->accept(this); + src_reg src = this->result; + + /* If we generated an expression instruction into a temporary in + * processing the saturate's operand, apply the saturate to that + * instruction. Otherwise, generate a MOV to do the saturate. + * + * Note that we have to be careful to only do this optimization if + * the instruction in question was what generated src->result. For + * example, ir_dereference_array might generate a MUL instruction + * to create the reladdr, and return us a src reg using that + * reladdr. That MUL result is not the value we're trying to + * saturate. + */ + ir_expression *sat_src_expr = sat_src->as_expression(); + ir_to_mesa_instruction *new_inst; + new_inst = (ir_to_mesa_instruction *)this->instructions.get_tail(); + if (sat_src_expr && (sat_src_expr->operation == ir_binop_mul || + sat_src_expr->operation == ir_binop_add || + sat_src_expr->operation == ir_binop_dot)) { + new_inst->saturate = true; + } else { + this->result = get_temp(ir->type); + ir_to_mesa_instruction *inst; + inst = emit(ir, OPCODE_MOV, dst_reg(this->result), src); + inst->saturate = true; + } + + return true; +} + +void +ir_to_mesa_visitor::reladdr_to_temp(ir_instruction *ir, + src_reg *reg, int *num_reladdr) +{ + if (!reg->reladdr) + return; + + emit(ir, OPCODE_ARL, address_reg, *reg->reladdr); + + if (*num_reladdr != 1) { + src_reg temp = get_temp(glsl_type::vec4_type); + + emit(ir, OPCODE_MOV, dst_reg(temp), *reg); + *reg = temp; + } + + (*num_reladdr)--; +} + +void +ir_to_mesa_visitor::emit_swz(ir_expression *ir) +{ + /* Assume that the vector operator is in a form compatible with OPCODE_SWZ. + * This means that each of the operands is either an immediate value of -1, + * 0, or 1, or is a component from one source register (possibly with + * negation). + */ + uint8_t components[4] = { 0 }; + bool negate[4] = { false }; + ir_variable *var = NULL; + + for (unsigned i = 0; i < ir->type->vector_elements; i++) { + ir_rvalue *op = ir->operands[i]; + + assert(op->type->is_scalar()); + + while (op != NULL) { + switch (op->ir_type) { + case ir_type_constant: { + + assert(op->type->is_scalar()); + + const ir_constant *const c = op->as_constant(); + if (c->is_one()) { + components[i] = SWIZZLE_ONE; + } else if (c->is_zero()) { + components[i] = SWIZZLE_ZERO; + } else if (c->is_negative_one()) { + components[i] = SWIZZLE_ONE; + negate[i] = true; + } else { + assert(!"SWZ constant must be 0.0 or 1.0."); + } + + op = NULL; + break; + } + + case ir_type_dereference_variable: { + ir_dereference_variable *const deref = + (ir_dereference_variable *) op; + + assert((var == NULL) || (deref->var == var)); + components[i] = SWIZZLE_X; + var = deref->var; + op = NULL; + break; + } + + case ir_type_expression: { + ir_expression *const expr = (ir_expression *) op; + + assert(expr->operation == ir_unop_neg); + negate[i] = true; + + op = expr->operands[0]; + break; + } + + case ir_type_swizzle: { + ir_swizzle *const swiz = (ir_swizzle *) op; + + components[i] = swiz->mask.x; + op = swiz->val; + break; + } + + default: + assert(!"Should not get here."); + return; + } + } + } + + assert(var != NULL); + + ir_dereference_variable *const deref = + new(mem_ctx) ir_dereference_variable(var); + + this->result.file = PROGRAM_UNDEFINED; + deref->accept(this); + if (this->result.file == PROGRAM_UNDEFINED) { + ir_print_visitor v; + printf("Failed to get tree for expression operand:\n"); + deref->accept(&v); + exit(1); + } + + src_reg src; + + src = this->result; + src.swizzle = MAKE_SWIZZLE4(components[0], + components[1], + components[2], + components[3]); + src.negate = ((unsigned(negate[0]) << 0) + | (unsigned(negate[1]) << 1) + | (unsigned(negate[2]) << 2) + | (unsigned(negate[3]) << 3)); + + /* Storage for our result. Ideally for an assignment we'd be using the + * actual storage for the result here, instead. + */ + const src_reg result_src = get_temp(ir->type); + dst_reg result_dst = dst_reg(result_src); + + /* Limit writes to the channels that will be used by result_src later. + * This does limit this temp's use as a temporary for multi-instruction + * sequences. + */ + result_dst.writemask = (1 << ir->type->vector_elements) - 1; + + emit(ir, OPCODE_SWZ, result_dst, src); + this->result = result_src; +} + +void +ir_to_mesa_visitor::visit(ir_expression *ir) +{ + unsigned int operand; + src_reg op[Elements(ir->operands)]; + src_reg result_src; + dst_reg result_dst; + + /* Quick peephole: Emit OPCODE_MAD(a, b, c) instead of ADD(MUL(a, b), c) + */ + if (ir->operation == ir_binop_add) { + if (try_emit_mad(ir, 1)) + return; + if (try_emit_mad(ir, 0)) + return; + } + + /* Quick peephole: Emit OPCODE_MAD(-a, -b, a) instead of AND(a, NOT(b)) + */ + if (ir->operation == ir_binop_logic_and) { + if (try_emit_mad_for_and_not(ir, 1)) + return; + if (try_emit_mad_for_and_not(ir, 0)) + return; + } + + if (try_emit_sat(ir)) + return; + + if (ir->operation == ir_quadop_vector) { + this->emit_swz(ir); + return; + } + + for (operand = 0; operand < ir->get_num_operands(); operand++) { + this->result.file = PROGRAM_UNDEFINED; + ir->operands[operand]->accept(this); + if (this->result.file == PROGRAM_UNDEFINED) { + ir_print_visitor v; + printf("Failed to get tree for expression operand:\n"); + ir->operands[operand]->accept(&v); + exit(1); + } + op[operand] = this->result; + + /* Matrix expression operands should have been broken down to vector + * operations already. + */ + assert(!ir->operands[operand]->type->is_matrix()); + } + + int vector_elements = ir->operands[0]->type->vector_elements; + if (ir->operands[1]) { + vector_elements = MAX2(vector_elements, + ir->operands[1]->type->vector_elements); + } + + this->result.file = PROGRAM_UNDEFINED; + + /* Storage for our result. Ideally for an assignment we'd be using + * the actual storage for the result here, instead. + */ + result_src = get_temp(ir->type); + /* convenience for the emit functions below. */ + result_dst = dst_reg(result_src); + /* Limit writes to the channels that will be used by result_src later. + * This does limit this temp's use as a temporary for multi-instruction + * sequences. + */ + result_dst.writemask = (1 << ir->type->vector_elements) - 1; + + switch (ir->operation) { + case ir_unop_logic_not: + /* Previously 'SEQ dst, src, 0.0' was used for this. However, many + * older GPUs implement SEQ using multiple instructions (i915 uses two + * SGE instructions and a MUL instruction). Since our logic values are + * 0.0 and 1.0, 1-x also implements !x. + */ + op[0].negate = ~op[0].negate; + emit(ir, OPCODE_ADD, result_dst, op[0], src_reg_for_float(1.0)); + break; + case ir_unop_neg: + op[0].negate = ~op[0].negate; + result_src = op[0]; + break; + case ir_unop_abs: + emit(ir, OPCODE_ABS, result_dst, op[0]); + break; + case ir_unop_sign: + emit(ir, OPCODE_SSG, result_dst, op[0]); + break; + case ir_unop_rcp: + emit_scalar(ir, OPCODE_RCP, result_dst, op[0]); + break; + + case ir_unop_exp2: + emit_scalar(ir, OPCODE_EX2, result_dst, op[0]); + break; + case ir_unop_exp: + case ir_unop_log: + assert(!"not reached: should be handled by ir_explog_to_explog2"); + break; + case ir_unop_log2: + emit_scalar(ir, OPCODE_LG2, result_dst, op[0]); + break; + case ir_unop_sin: + emit_scalar(ir, OPCODE_SIN, result_dst, op[0]); + break; + case ir_unop_cos: + emit_scalar(ir, OPCODE_COS, result_dst, op[0]); + break; + case ir_unop_sin_reduced: + emit_scs(ir, OPCODE_SIN, result_dst, op[0]); + break; + case ir_unop_cos_reduced: + emit_scs(ir, OPCODE_COS, result_dst, op[0]); + break; + + case ir_unop_dFdx: + emit(ir, OPCODE_DDX, result_dst, op[0]); + break; + case ir_unop_dFdy: + emit(ir, OPCODE_DDY, result_dst, op[0]); + break; + + case ir_unop_noise: { + const enum prog_opcode opcode = + prog_opcode(OPCODE_NOISE1 + + (ir->operands[0]->type->vector_elements) - 1); + assert((opcode >= OPCODE_NOISE1) && (opcode <= OPCODE_NOISE4)); + + emit(ir, opcode, result_dst, op[0]); + break; + } + + case ir_binop_add: + emit(ir, OPCODE_ADD, result_dst, op[0], op[1]); + break; + case ir_binop_sub: + emit(ir, OPCODE_SUB, result_dst, op[0], op[1]); + break; + + case ir_binop_mul: + emit(ir, OPCODE_MUL, result_dst, op[0], op[1]); + break; + case ir_binop_div: + assert(!"not reached: should be handled by ir_div_to_mul_rcp"); + case ir_binop_mod: + assert(!"ir_binop_mod should have been converted to b * fract(a/b)"); + break; + + case ir_binop_less: + emit(ir, OPCODE_SLT, result_dst, op[0], op[1]); + break; + case ir_binop_greater: + emit(ir, OPCODE_SGT, result_dst, op[0], op[1]); + break; + case ir_binop_lequal: + emit(ir, OPCODE_SLE, result_dst, op[0], op[1]); + break; + case ir_binop_gequal: + emit(ir, OPCODE_SGE, result_dst, op[0], op[1]); + break; + case ir_binop_equal: + emit(ir, OPCODE_SEQ, result_dst, op[0], op[1]); + break; + case ir_binop_nequal: + emit(ir, OPCODE_SNE, result_dst, op[0], op[1]); + break; + case ir_binop_all_equal: + /* "==" operator producing a scalar boolean. */ + if (ir->operands[0]->type->is_vector() || + ir->operands[1]->type->is_vector()) { + src_reg temp = get_temp(glsl_type::vec4_type); + emit(ir, OPCODE_SNE, dst_reg(temp), op[0], op[1]); + + /* After the dot-product, the value will be an integer on the + * range [0,4]. Zero becomes 1.0, and positive values become zero. + */ + emit_dp(ir, result_dst, temp, temp, vector_elements); + + /* Negating the result of the dot-product gives values on the range + * [-4, 0]. Zero becomes 1.0, and negative values become zero. This + * achieved using SGE. + */ + src_reg sge_src = result_src; + sge_src.negate = ~sge_src.negate; + emit(ir, OPCODE_SGE, result_dst, sge_src, src_reg_for_float(0.0)); + } else { + emit(ir, OPCODE_SEQ, result_dst, op[0], op[1]); + } + break; + case ir_binop_any_nequal: + /* "!=" operator producing a scalar boolean. */ + if (ir->operands[0]->type->is_vector() || + ir->operands[1]->type->is_vector()) { + src_reg temp = get_temp(glsl_type::vec4_type); + emit(ir, OPCODE_SNE, dst_reg(temp), op[0], op[1]); + + /* After the dot-product, the value will be an integer on the + * range [0,4]. Zero stays zero, and positive values become 1.0. + */ + ir_to_mesa_instruction *const dp = + emit_dp(ir, result_dst, temp, temp, vector_elements); + if (this->prog->Target == GL_FRAGMENT_PROGRAM_ARB) { + /* The clamping to [0,1] can be done for free in the fragment + * shader with a saturate. + */ + dp->saturate = true; + } else { + /* Negating the result of the dot-product gives values on the range + * [-4, 0]. Zero stays zero, and negative values become 1.0. This + * achieved using SLT. + */ + src_reg slt_src = result_src; + slt_src.negate = ~slt_src.negate; + emit(ir, OPCODE_SLT, result_dst, slt_src, src_reg_for_float(0.0)); + } + } else { + emit(ir, OPCODE_SNE, result_dst, op[0], op[1]); + } + break; + + case ir_unop_any: { + assert(ir->operands[0]->type->is_vector()); + + /* After the dot-product, the value will be an integer on the + * range [0,4]. Zero stays zero, and positive values become 1.0. + */ + ir_to_mesa_instruction *const dp = + emit_dp(ir, result_dst, op[0], op[0], + ir->operands[0]->type->vector_elements); + if (this->prog->Target == GL_FRAGMENT_PROGRAM_ARB) { + /* The clamping to [0,1] can be done for free in the fragment + * shader with a saturate. + */ + dp->saturate = true; + } else { + /* Negating the result of the dot-product gives values on the range + * [-4, 0]. Zero stays zero, and negative values become 1.0. This + * is achieved using SLT. + */ + src_reg slt_src = result_src; + slt_src.negate = ~slt_src.negate; + emit(ir, OPCODE_SLT, result_dst, slt_src, src_reg_for_float(0.0)); + } + break; + } + + case ir_binop_logic_xor: + emit(ir, OPCODE_SNE, result_dst, op[0], op[1]); + break; + + case ir_binop_logic_or: { + /* After the addition, the value will be an integer on the + * range [0,2]. Zero stays zero, and positive values become 1.0. + */ + ir_to_mesa_instruction *add = + emit(ir, OPCODE_ADD, result_dst, op[0], op[1]); + if (this->prog->Target == GL_FRAGMENT_PROGRAM_ARB) { + /* The clamping to [0,1] can be done for free in the fragment + * shader with a saturate. + */ + add->saturate = true; + } else { + /* Negating the result of the addition gives values on the range + * [-2, 0]. Zero stays zero, and negative values become 1.0. This + * is achieved using SLT. + */ + src_reg slt_src = result_src; + slt_src.negate = ~slt_src.negate; + emit(ir, OPCODE_SLT, result_dst, slt_src, src_reg_for_float(0.0)); + } + break; + } + + case ir_binop_logic_and: + /* the bool args are stored as float 0.0 or 1.0, so "mul" gives us "and". */ + emit(ir, OPCODE_MUL, result_dst, op[0], op[1]); + break; + + case ir_binop_dot: + assert(ir->operands[0]->type->is_vector()); + assert(ir->operands[0]->type == ir->operands[1]->type); + emit_dp(ir, result_dst, op[0], op[1], + ir->operands[0]->type->vector_elements); + break; + + case ir_unop_sqrt: + /* sqrt(x) = x * rsq(x). */ + emit_scalar(ir, OPCODE_RSQ, result_dst, op[0]); + emit(ir, OPCODE_MUL, result_dst, result_src, op[0]); + /* For incoming channels <= 0, set the result to 0. */ + op[0].negate = ~op[0].negate; + emit(ir, OPCODE_CMP, result_dst, + op[0], result_src, src_reg_for_float(0.0)); + break; + case ir_unop_rsq: + emit_scalar(ir, OPCODE_RSQ, result_dst, op[0]); + break; + case ir_unop_i2f: + case ir_unop_u2f: + case ir_unop_b2f: + case ir_unop_b2i: + case ir_unop_i2u: + case ir_unop_u2i: + /* Mesa IR lacks types, ints are stored as truncated floats. */ + result_src = op[0]; + break; + case ir_unop_f2i: + emit(ir, OPCODE_TRUNC, result_dst, op[0]); + break; + case ir_unop_f2b: + case ir_unop_i2b: + emit(ir, OPCODE_SNE, result_dst, + op[0], src_reg_for_float(0.0)); + break; + case ir_unop_trunc: + emit(ir, OPCODE_TRUNC, result_dst, op[0]); + break; + case ir_unop_ceil: + op[0].negate = ~op[0].negate; + emit(ir, OPCODE_FLR, result_dst, op[0]); + result_src.negate = ~result_src.negate; + break; + case ir_unop_floor: + emit(ir, OPCODE_FLR, result_dst, op[0]); + break; + case ir_unop_fract: + emit(ir, OPCODE_FRC, result_dst, op[0]); + break; + + case ir_binop_min: + emit(ir, OPCODE_MIN, result_dst, op[0], op[1]); + break; + case ir_binop_max: + emit(ir, OPCODE_MAX, result_dst, op[0], op[1]); + break; + case ir_binop_pow: + emit_scalar(ir, OPCODE_POW, result_dst, op[0], op[1]); + break; + + case ir_unop_bit_not: + case ir_binop_lshift: + case ir_binop_rshift: + case ir_binop_bit_and: + case ir_binop_bit_xor: + case ir_binop_bit_or: + case ir_unop_round_even: + assert(!"GLSL 1.30 features unsupported"); + break; + + case ir_quadop_vector: + /* This operation should have already been handled. + */ + assert(!"Should not get here."); + break; + } + + this->result = result_src; +} + + +void +ir_to_mesa_visitor::visit(ir_swizzle *ir) +{ + src_reg src; + int i; + int swizzle[4]; + + /* Note that this is only swizzles in expressions, not those on the left + * hand side of an assignment, which do write masking. See ir_assignment + * for that. + */ + + ir->val->accept(this); + src = this->result; + assert(src.file != PROGRAM_UNDEFINED); + + for (i = 0; i < 4; i++) { + if (i < ir->type->vector_elements) { + switch (i) { + case 0: + swizzle[i] = GET_SWZ(src.swizzle, ir->mask.x); + break; + case 1: + swizzle[i] = GET_SWZ(src.swizzle, ir->mask.y); + break; + case 2: + swizzle[i] = GET_SWZ(src.swizzle, ir->mask.z); + break; + case 3: + swizzle[i] = GET_SWZ(src.swizzle, ir->mask.w); + break; + } + } else { + /* If the type is smaller than a vec4, replicate the last + * channel out. + */ + swizzle[i] = swizzle[ir->type->vector_elements - 1]; + } + } + + src.swizzle = MAKE_SWIZZLE4(swizzle[0], swizzle[1], swizzle[2], swizzle[3]); + + this->result = src; +} + +void +ir_to_mesa_visitor::visit(ir_dereference_variable *ir) +{ + variable_storage *entry = find_variable_storage(ir->var); + ir_variable *var = ir->var; + + if (!entry) { + switch (var->mode) { + case ir_var_uniform: + entry = new(mem_ctx) variable_storage(var, PROGRAM_UNIFORM, + var->location); + this->variables.push_tail(entry); + break; + case ir_var_in: + case ir_var_inout: + /* The linker assigns locations for varyings and attributes, + * including deprecated builtins (like gl_Color), + * user-assigned generic attributes (glBindVertexLocation), + * and user-defined varyings. + * + * FINISHME: We would hit this path for function arguments. Fix! + */ + assert(var->location != -1); + entry = new(mem_ctx) variable_storage(var, + PROGRAM_INPUT, + var->location); + if (this->prog->Target == GL_VERTEX_PROGRAM_ARB && + var->location >= VERT_ATTRIB_GENERIC0) { + _mesa_add_attribute(this->prog->Attributes, + var->name, + _mesa_sizeof_glsl_type(var->type->gl_type), + var->type->gl_type, + var->location - VERT_ATTRIB_GENERIC0); + } + break; + case ir_var_out: + assert(var->location != -1); + entry = new(mem_ctx) variable_storage(var, + PROGRAM_OUTPUT, + var->location); + break; + case ir_var_system_value: + entry = new(mem_ctx) variable_storage(var, + PROGRAM_SYSTEM_VALUE, + var->location); + break; + case ir_var_auto: + case ir_var_temporary: + entry = new(mem_ctx) variable_storage(var, PROGRAM_TEMPORARY, + this->next_temp); + this->variables.push_tail(entry); + + next_temp += type_size(var->type); + break; + } + + if (!entry) { + printf("Failed to make storage for %s\n", var->name); + exit(1); + } + } + + this->result = src_reg(entry->file, entry->index, var->type); +} + +void +ir_to_mesa_visitor::visit(ir_dereference_array *ir) +{ + ir_constant *index; + src_reg src; + int element_size = type_size(ir->type); + + index = ir->array_index->constant_expression_value(); + + ir->array->accept(this); + src = this->result; + + if (index) { + src.index += index->value.i[0] * element_size; + } else { + /* Variable index array dereference. It eats the "vec4" of the + * base of the array and an index that offsets the Mesa register + * index. + */ + ir->array_index->accept(this); + + src_reg index_reg; + + if (element_size == 1) { + index_reg = this->result; + } else { + index_reg = get_temp(glsl_type::float_type); + + emit(ir, OPCODE_MUL, dst_reg(index_reg), + this->result, src_reg_for_float(element_size)); + } + + /* If there was already a relative address register involved, add the + * new and the old together to get the new offset. + */ + if (src.reladdr != NULL) { + src_reg accum_reg = get_temp(glsl_type::float_type); + + emit(ir, OPCODE_ADD, dst_reg(accum_reg), + index_reg, *src.reladdr); + + index_reg = accum_reg; + } + + src.reladdr = ralloc(mem_ctx, src_reg); + memcpy(src.reladdr, &index_reg, sizeof(index_reg)); + } + + /* If the type is smaller than a vec4, replicate the last channel out. */ + if (ir->type->is_scalar() || ir->type->is_vector()) + src.swizzle = swizzle_for_size(ir->type->vector_elements); + else + src.swizzle = SWIZZLE_NOOP; + + this->result = src; +} + +void +ir_to_mesa_visitor::visit(ir_dereference_record *ir) +{ + unsigned int i; + const glsl_type *struct_type = ir->record->type; + int offset = 0; + + ir->record->accept(this); + + for (i = 0; i < struct_type->length; i++) { + if (strcmp(struct_type->fields.structure[i].name, ir->field) == 0) + break; + offset += type_size(struct_type->fields.structure[i].type); + } + + /* If the type is smaller than a vec4, replicate the last channel out. */ + if (ir->type->is_scalar() || ir->type->is_vector()) + this->result.swizzle = swizzle_for_size(ir->type->vector_elements); + else + this->result.swizzle = SWIZZLE_NOOP; + + this->result.index += offset; +} + +/** + * We want to be careful in assignment setup to hit the actual storage + * instead of potentially using a temporary like we might with the + * ir_dereference handler. + */ +static dst_reg +get_assignment_lhs(ir_dereference *ir, ir_to_mesa_visitor *v) +{ + /* The LHS must be a dereference. If the LHS is a variable indexed array + * access of a vector, it must be separated into a series conditional moves + * before reaching this point (see ir_vec_index_to_cond_assign). + */ + assert(ir->as_dereference()); + ir_dereference_array *deref_array = ir->as_dereference_array(); + if (deref_array) { + assert(!deref_array->array->type->is_vector()); + } + + /* Use the rvalue deref handler for the most part. We'll ignore + * swizzles in it and write swizzles using writemask, though. + */ + ir->accept(v); + return dst_reg(v->result); +} + +/** + * Process the condition of a conditional assignment + * + * Examines the condition of a conditional assignment to generate the optimal + * first operand of a \c CMP instruction. If the condition is a relational + * operator with 0 (e.g., \c ir_binop_less), the value being compared will be + * used as the source for the \c CMP instruction. Otherwise the comparison + * is processed to a boolean result, and the boolean result is used as the + * operand to the CMP instruction. + */ +bool +ir_to_mesa_visitor::process_move_condition(ir_rvalue *ir) +{ + ir_rvalue *src_ir = ir; + bool negate = true; + bool switch_order = false; + + ir_expression *const expr = ir->as_expression(); + if ((expr != NULL) && (expr->get_num_operands() == 2)) { + bool zero_on_left = false; + + if (expr->operands[0]->is_zero()) { + src_ir = expr->operands[1]; + zero_on_left = true; + } else if (expr->operands[1]->is_zero()) { + src_ir = expr->operands[0]; + zero_on_left = false; + } + + /* a is - 0 + - 0 + + * (a < 0) T F F ( a < 0) T F F + * (0 < a) F F T (-a < 0) F F T + * (a <= 0) T T F (-a < 0) F F T (swap order of other operands) + * (0 <= a) F T T ( a < 0) T F F (swap order of other operands) + * (a > 0) F F T (-a < 0) F F T + * (0 > a) T F F ( a < 0) T F F + * (a >= 0) F T T ( a < 0) T F F (swap order of other operands) + * (0 >= a) T T F (-a < 0) F F T (swap order of other operands) + * + * Note that exchanging the order of 0 and 'a' in the comparison simply + * means that the value of 'a' should be negated. + */ + if (src_ir != ir) { + switch (expr->operation) { + case ir_binop_less: + switch_order = false; + negate = zero_on_left; + break; + + case ir_binop_greater: + switch_order = false; + negate = !zero_on_left; + break; + + case ir_binop_lequal: + switch_order = true; + negate = !zero_on_left; + break; + + case ir_binop_gequal: + switch_order = true; + negate = zero_on_left; + break; + + default: + /* This isn't the right kind of comparison afterall, so make sure + * the whole condition is visited. + */ + src_ir = ir; + break; + } + } + } + + src_ir->accept(this); + + /* We use the OPCODE_CMP (a < 0 ? b : c) for conditional moves, and the + * condition we produced is 0.0 or 1.0. By flipping the sign, we can + * choose which value OPCODE_CMP produces without an extra instruction + * computing the condition. + */ + if (negate) + this->result.negate = ~this->result.negate; + + return switch_order; +} + +void +ir_to_mesa_visitor::visit(ir_assignment *ir) +{ + dst_reg l; + src_reg r; + int i; + + ir->rhs->accept(this); + r = this->result; + + l = get_assignment_lhs(ir->lhs, this); + + /* FINISHME: This should really set to the correct maximal writemask for each + * FINISHME: component written (in the loops below). This case can only + * FINISHME: occur for matrices, arrays, and structures. + */ + if (ir->write_mask == 0) { + assert(!ir->lhs->type->is_scalar() && !ir->lhs->type->is_vector()); + l.writemask = WRITEMASK_XYZW; + } else if (ir->lhs->type->is_scalar()) { + /* FINISHME: This hack makes writing to gl_FragDepth, which lives in the + * FINISHME: W component of fragment shader output zero, work correctly. + */ + l.writemask = WRITEMASK_XYZW; + } else { + int swizzles[4]; + int first_enabled_chan = 0; + int rhs_chan = 0; + + assert(ir->lhs->type->is_vector()); + l.writemask = ir->write_mask; + + for (int i = 0; i < 4; i++) { + if (l.writemask & (1 << i)) { + first_enabled_chan = GET_SWZ(r.swizzle, i); + break; + } + } + + /* Swizzle a small RHS vector into the channels being written. + * + * glsl ir treats write_mask as dictating how many channels are + * present on the RHS while Mesa IR treats write_mask as just + * showing which channels of the vec4 RHS get written. + */ + for (int i = 0; i < 4; i++) { + if (l.writemask & (1 << i)) + swizzles[i] = GET_SWZ(r.swizzle, rhs_chan++); + else + swizzles[i] = first_enabled_chan; + } + r.swizzle = MAKE_SWIZZLE4(swizzles[0], swizzles[1], + swizzles[2], swizzles[3]); + } + + assert(l.file != PROGRAM_UNDEFINED); + assert(r.file != PROGRAM_UNDEFINED); + + if (ir->condition) { + const bool switch_order = this->process_move_condition(ir->condition); + src_reg condition = this->result; + + for (i = 0; i < type_size(ir->lhs->type); i++) { + if (switch_order) { + emit(ir, OPCODE_CMP, l, condition, src_reg(l), r); + } else { + emit(ir, OPCODE_CMP, l, condition, r, src_reg(l)); + } + + l.index++; + r.index++; + } + } else { + for (i = 0; i < type_size(ir->lhs->type); i++) { + emit(ir, OPCODE_MOV, l, r); + l.index++; + r.index++; + } + } +} + + +void +ir_to_mesa_visitor::visit(ir_constant *ir) +{ + src_reg src; + GLfloat stack_vals[4] = { 0 }; + GLfloat *values = stack_vals; + unsigned int i; + + /* Unfortunately, 4 floats is all we can get into + * _mesa_add_unnamed_constant. So, make a temp to store an + * aggregate constant and move each constant value into it. If we + * get lucky, copy propagation will eliminate the extra moves. + */ + + if (ir->type->base_type == GLSL_TYPE_STRUCT) { + src_reg temp_base = get_temp(ir->type); + dst_reg temp = dst_reg(temp_base); + + foreach_iter(exec_list_iterator, iter, ir->components) { + ir_constant *field_value = (ir_constant *)iter.get(); + int size = type_size(field_value->type); + + assert(size > 0); + + field_value->accept(this); + src = this->result; + + for (i = 0; i < (unsigned int)size; i++) { + emit(ir, OPCODE_MOV, temp, src); + + src.index++; + temp.index++; + } + } + this->result = temp_base; + return; + } + + if (ir->type->is_array()) { + src_reg temp_base = get_temp(ir->type); + dst_reg temp = dst_reg(temp_base); + int size = type_size(ir->type->fields.array); + + assert(size > 0); + + for (i = 0; i < ir->type->length; i++) { + ir->array_elements[i]->accept(this); + src = this->result; + for (int j = 0; j < size; j++) { + emit(ir, OPCODE_MOV, temp, src); + + src.index++; + temp.index++; + } + } + this->result = temp_base; + return; + } + + if (ir->type->is_matrix()) { + src_reg mat = get_temp(ir->type); + dst_reg mat_column = dst_reg(mat); + + for (i = 0; i < ir->type->matrix_columns; i++) { + assert(ir->type->base_type == GLSL_TYPE_FLOAT); + values = &ir->value.f[i * ir->type->vector_elements]; + + src = src_reg(PROGRAM_CONSTANT, -1, NULL); + src.index = _mesa_add_unnamed_constant(this->prog->Parameters, + (gl_constant_value *) values, + ir->type->vector_elements, + &src.swizzle); + emit(ir, OPCODE_MOV, mat_column, src); + + mat_column.index++; + } + + this->result = mat; + return; + } + + src.file = PROGRAM_CONSTANT; + switch (ir->type->base_type) { + case GLSL_TYPE_FLOAT: + values = &ir->value.f[0]; + break; + case GLSL_TYPE_UINT: + for (i = 0; i < ir->type->vector_elements; i++) { + values[i] = ir->value.u[i]; + } + break; + case GLSL_TYPE_INT: + for (i = 0; i < ir->type->vector_elements; i++) { + values[i] = ir->value.i[i]; + } + break; + case GLSL_TYPE_BOOL: + for (i = 0; i < ir->type->vector_elements; i++) { + values[i] = ir->value.b[i]; + } + break; + default: + assert(!"Non-float/uint/int/bool constant"); + } + + this->result = src_reg(PROGRAM_CONSTANT, -1, ir->type); + this->result.index = _mesa_add_unnamed_constant(this->prog->Parameters, + (gl_constant_value *) values, + ir->type->vector_elements, + &this->result.swizzle); +} + +function_entry * +ir_to_mesa_visitor::get_function_signature(ir_function_signature *sig) +{ + function_entry *entry; + + foreach_iter(exec_list_iterator, iter, this->function_signatures) { + entry = (function_entry *)iter.get(); + + if (entry->sig == sig) + return entry; + } + + entry = ralloc(mem_ctx, function_entry); + entry->sig = sig; + entry->sig_id = this->next_signature_id++; + entry->bgn_inst = NULL; + + /* Allocate storage for all the parameters. */ + foreach_iter(exec_list_iterator, iter, sig->parameters) { + ir_variable *param = (ir_variable *)iter.get(); + variable_storage *storage; + + storage = find_variable_storage(param); + assert(!storage); + + storage = new(mem_ctx) variable_storage(param, PROGRAM_TEMPORARY, + this->next_temp); + this->variables.push_tail(storage); + + this->next_temp += type_size(param->type); + } + + if (!sig->return_type->is_void()) { + entry->return_reg = get_temp(sig->return_type); + } else { + entry->return_reg = undef_src; + } + + this->function_signatures.push_tail(entry); + return entry; +} + +void +ir_to_mesa_visitor::visit(ir_call *ir) +{ + ir_to_mesa_instruction *call_inst; + ir_function_signature *sig = ir->get_callee(); + function_entry *entry = get_function_signature(sig); + int i; + + /* Process in parameters. */ + exec_list_iterator sig_iter = sig->parameters.iterator(); + foreach_iter(exec_list_iterator, iter, *ir) { + ir_rvalue *param_rval = (ir_rvalue *)iter.get(); + ir_variable *param = (ir_variable *)sig_iter.get(); + + if (param->mode == ir_var_in || + param->mode == ir_var_inout) { + variable_storage *storage = find_variable_storage(param); + assert(storage); + + param_rval->accept(this); + src_reg r = this->result; + + dst_reg l; + l.file = storage->file; + l.index = storage->index; + l.reladdr = NULL; + l.writemask = WRITEMASK_XYZW; + l.cond_mask = COND_TR; + + for (i = 0; i < type_size(param->type); i++) { + emit(ir, OPCODE_MOV, l, r); + l.index++; + r.index++; + } + } + + sig_iter.next(); + } + assert(!sig_iter.has_next()); + + /* Emit call instruction */ + call_inst = emit(ir, OPCODE_CAL); + call_inst->function = entry; + + /* Process out parameters. */ + sig_iter = sig->parameters.iterator(); + foreach_iter(exec_list_iterator, iter, *ir) { + ir_rvalue *param_rval = (ir_rvalue *)iter.get(); + ir_variable *param = (ir_variable *)sig_iter.get(); + + if (param->mode == ir_var_out || + param->mode == ir_var_inout) { + variable_storage *storage = find_variable_storage(param); + assert(storage); + + src_reg r; + r.file = storage->file; + r.index = storage->index; + r.reladdr = NULL; + r.swizzle = SWIZZLE_NOOP; + r.negate = 0; + + param_rval->accept(this); + dst_reg l = dst_reg(this->result); + + for (i = 0; i < type_size(param->type); i++) { + emit(ir, OPCODE_MOV, l, r); + l.index++; + r.index++; + } + } + + sig_iter.next(); + } + assert(!sig_iter.has_next()); + + /* Process return value. */ + this->result = entry->return_reg; +} + +void +ir_to_mesa_visitor::visit(ir_texture *ir) +{ + src_reg result_src, coord, lod_info, projector, dx, dy; + dst_reg result_dst, coord_dst; + ir_to_mesa_instruction *inst = NULL; + prog_opcode opcode = OPCODE_NOP; + + if (ir->op == ir_txs) + this->result = src_reg_for_float(0.0); + else + ir->coordinate->accept(this); + + /* Put our coords in a temp. We'll need to modify them for shadow, + * projection, or LOD, so the only case we'd use it as is is if + * we're doing plain old texturing. Mesa IR optimization should + * handle cleaning up our mess in that case. + */ + coord = get_temp(glsl_type::vec4_type); + coord_dst = dst_reg(coord); + emit(ir, OPCODE_MOV, coord_dst, this->result); + + if (ir->projector) { + ir->projector->accept(this); + projector = this->result; + } + + /* Storage for our result. Ideally for an assignment we'd be using + * the actual storage for the result here, instead. + */ + result_src = get_temp(glsl_type::vec4_type); + result_dst = dst_reg(result_src); + + switch (ir->op) { + case ir_tex: + case ir_txs: + opcode = OPCODE_TEX; + break; + case ir_txb: + opcode = OPCODE_TXB; + ir->lod_info.bias->accept(this); + lod_info = this->result; + break; + case ir_txl: + opcode = OPCODE_TXL; + ir->lod_info.lod->accept(this); + lod_info = this->result; + break; + case ir_txd: + opcode = OPCODE_TXD; + ir->lod_info.grad.dPdx->accept(this); + dx = this->result; + ir->lod_info.grad.dPdy->accept(this); + dy = this->result; + break; + case ir_txf: + assert(!"GLSL 1.30 features unsupported"); + break; + } + + const glsl_type *sampler_type = ir->sampler->type; + + if (ir->projector) { + if (opcode == OPCODE_TEX) { + /* Slot the projector in as the last component of the coord. */ + coord_dst.writemask = WRITEMASK_W; + emit(ir, OPCODE_MOV, coord_dst, projector); + coord_dst.writemask = WRITEMASK_XYZW; + opcode = OPCODE_TXP; + } else { + src_reg coord_w = coord; + coord_w.swizzle = SWIZZLE_WWWW; + + /* For the other TEX opcodes there's no projective version + * since the last slot is taken up by lod info. Do the + * projective divide now. + */ + coord_dst.writemask = WRITEMASK_W; + emit(ir, OPCODE_RCP, coord_dst, projector); + + /* In the case where we have to project the coordinates "by hand," + * the shadow comparitor value must also be projected. + */ + src_reg tmp_src = coord; + if (ir->shadow_comparitor) { + /* Slot the shadow value in as the second to last component of the + * coord. + */ + ir->shadow_comparitor->accept(this); + + tmp_src = get_temp(glsl_type::vec4_type); + dst_reg tmp_dst = dst_reg(tmp_src); + + /* Projective division not allowed for array samplers. */ + assert(!sampler_type->sampler_array); + + tmp_dst.writemask = WRITEMASK_Z; + emit(ir, OPCODE_MOV, tmp_dst, this->result); + + tmp_dst.writemask = WRITEMASK_XY; + emit(ir, OPCODE_MOV, tmp_dst, coord); + } + + coord_dst.writemask = WRITEMASK_XYZ; + emit(ir, OPCODE_MUL, coord_dst, tmp_src, coord_w); + + coord_dst.writemask = WRITEMASK_XYZW; + coord.swizzle = SWIZZLE_XYZW; + } + } + + /* If projection is done and the opcode is not OPCODE_TXP, then the shadow + * comparitor was put in the correct place (and projected) by the code, + * above, that handles by-hand projection. + */ + if (ir->shadow_comparitor && (!ir->projector || opcode == OPCODE_TXP)) { + /* Slot the shadow value in as the second to last component of the + * coord. + */ + ir->shadow_comparitor->accept(this); + + /* XXX This will need to be updated for cubemap array samplers. */ + if (sampler_type->sampler_dimensionality == GLSL_SAMPLER_DIM_2D && + sampler_type->sampler_array) { + coord_dst.writemask = WRITEMASK_W; + } else { + coord_dst.writemask = WRITEMASK_Z; + } + + emit(ir, OPCODE_MOV, coord_dst, this->result); + coord_dst.writemask = WRITEMASK_XYZW; + } + + if (opcode == OPCODE_TXL || opcode == OPCODE_TXB) { + /* Mesa IR stores lod or lod bias in the last channel of the coords. */ + coord_dst.writemask = WRITEMASK_W; + emit(ir, OPCODE_MOV, coord_dst, lod_info); + coord_dst.writemask = WRITEMASK_XYZW; + } + + if (opcode == OPCODE_TXD) + inst = emit(ir, opcode, result_dst, coord, dx, dy); + else + inst = emit(ir, opcode, result_dst, coord); + + if (ir->shadow_comparitor) + inst->tex_shadow = GL_TRUE; + + inst->sampler = _mesa_get_sampler_uniform_value(ir->sampler, + this->shader_program, + this->prog); + + switch (sampler_type->sampler_dimensionality) { + case GLSL_SAMPLER_DIM_1D: + inst->tex_target = (sampler_type->sampler_array) + ? TEXTURE_1D_ARRAY_INDEX : TEXTURE_1D_INDEX; + break; + case GLSL_SAMPLER_DIM_2D: + inst->tex_target = (sampler_type->sampler_array) + ? TEXTURE_2D_ARRAY_INDEX : TEXTURE_2D_INDEX; + break; + case GLSL_SAMPLER_DIM_3D: + inst->tex_target = TEXTURE_3D_INDEX; + break; + case GLSL_SAMPLER_DIM_CUBE: + inst->tex_target = TEXTURE_CUBE_INDEX; + break; + case GLSL_SAMPLER_DIM_RECT: + inst->tex_target = TEXTURE_RECT_INDEX; + break; + case GLSL_SAMPLER_DIM_BUF: + assert(!"FINISHME: Implement ARB_texture_buffer_object"); + break; + default: + assert(!"Should not get here."); + } + + this->result = result_src; +} + +void +ir_to_mesa_visitor::visit(ir_return *ir) +{ + if (ir->get_value()) { + dst_reg l; + int i; + + assert(current_function); + + ir->get_value()->accept(this); + src_reg r = this->result; + + l = dst_reg(current_function->return_reg); + + for (i = 0; i < type_size(current_function->sig->return_type); i++) { + emit(ir, OPCODE_MOV, l, r); + l.index++; + r.index++; + } + } + + emit(ir, OPCODE_RET); +} + +void +ir_to_mesa_visitor::visit(ir_discard *ir) +{ + struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog; + + if (ir->condition) { + ir->condition->accept(this); + this->result.negate = ~this->result.negate; + emit(ir, OPCODE_KIL, undef_dst, this->result); + } else { + emit(ir, OPCODE_KIL_NV); + } + + fp->UsesKill = GL_TRUE; +} + +void +ir_to_mesa_visitor::visit(ir_if *ir) +{ + ir_to_mesa_instruction *cond_inst, *if_inst; + ir_to_mesa_instruction *prev_inst; + + prev_inst = (ir_to_mesa_instruction *)this->instructions.get_tail(); + + ir->condition->accept(this); + assert(this->result.file != PROGRAM_UNDEFINED); + + if (this->options->EmitCondCodes) { + cond_inst = (ir_to_mesa_instruction *)this->instructions.get_tail(); + + /* See if we actually generated any instruction for generating + * the condition. If not, then cook up a move to a temp so we + * have something to set cond_update on. + */ + if (cond_inst == prev_inst) { + src_reg temp = get_temp(glsl_type::bool_type); + cond_inst = emit(ir->condition, OPCODE_MOV, dst_reg(temp), result); + } + cond_inst->cond_update = GL_TRUE; + + if_inst = emit(ir->condition, OPCODE_IF); + if_inst->dst.cond_mask = COND_NE; + } else { + if_inst = emit(ir->condition, OPCODE_IF, undef_dst, this->result); + } + + this->instructions.push_tail(if_inst); + + visit_exec_list(&ir->then_instructions, this); + + if (!ir->else_instructions.is_empty()) { + emit(ir->condition, OPCODE_ELSE); + visit_exec_list(&ir->else_instructions, this); + } + + if_inst = emit(ir->condition, OPCODE_ENDIF); +} + +ir_to_mesa_visitor::ir_to_mesa_visitor() +{ + result.file = PROGRAM_UNDEFINED; + next_temp = 1; + next_signature_id = 1; + current_function = NULL; + mem_ctx = ralloc_context(NULL); +} + +ir_to_mesa_visitor::~ir_to_mesa_visitor() +{ + ralloc_free(mem_ctx); +} + +static struct prog_src_register +mesa_src_reg_from_ir_src_reg(src_reg reg) +{ + struct prog_src_register mesa_reg; + + mesa_reg.File = reg.file; + assert(reg.index < (1 << INST_INDEX_BITS)); + mesa_reg.Index = reg.index; + mesa_reg.Swizzle = reg.swizzle; + mesa_reg.RelAddr = reg.reladdr != NULL; + mesa_reg.Negate = reg.negate; + mesa_reg.Abs = 0; + mesa_reg.HasIndex2 = GL_FALSE; + mesa_reg.RelAddr2 = 0; + mesa_reg.Index2 = 0; + + return mesa_reg; +} + +static void +set_branchtargets(ir_to_mesa_visitor *v, + struct prog_instruction *mesa_instructions, + int num_instructions) +{ + int if_count = 0, loop_count = 0; + int *if_stack, *loop_stack; + int if_stack_pos = 0, loop_stack_pos = 0; + int i, j; + + for (i = 0; i < num_instructions; i++) { + switch (mesa_instructions[i].Opcode) { + case OPCODE_IF: + if_count++; + break; + case OPCODE_BGNLOOP: + loop_count++; + break; + case OPCODE_BRK: + case OPCODE_CONT: + mesa_instructions[i].BranchTarget = -1; + break; + default: + break; + } + } + + if_stack = rzalloc_array(v->mem_ctx, int, if_count); + loop_stack = rzalloc_array(v->mem_ctx, int, loop_count); + + for (i = 0; i < num_instructions; i++) { + switch (mesa_instructions[i].Opcode) { + case OPCODE_IF: + if_stack[if_stack_pos] = i; + if_stack_pos++; + break; + case OPCODE_ELSE: + mesa_instructions[if_stack[if_stack_pos - 1]].BranchTarget = i; + if_stack[if_stack_pos - 1] = i; + break; + case OPCODE_ENDIF: + mesa_instructions[if_stack[if_stack_pos - 1]].BranchTarget = i; + if_stack_pos--; + break; + case OPCODE_BGNLOOP: + loop_stack[loop_stack_pos] = i; + loop_stack_pos++; + break; + case OPCODE_ENDLOOP: + loop_stack_pos--; + /* Rewrite any breaks/conts at this nesting level (haven't + * already had a BranchTarget assigned) to point to the end + * of the loop. + */ + for (j = loop_stack[loop_stack_pos]; j < i; j++) { + if (mesa_instructions[j].Opcode == OPCODE_BRK || + mesa_instructions[j].Opcode == OPCODE_CONT) { + if (mesa_instructions[j].BranchTarget == -1) { + mesa_instructions[j].BranchTarget = i; + } + } + } + /* The loop ends point at each other. */ + mesa_instructions[i].BranchTarget = loop_stack[loop_stack_pos]; + mesa_instructions[loop_stack[loop_stack_pos]].BranchTarget = i; + break; + case OPCODE_CAL: + foreach_iter(exec_list_iterator, iter, v->function_signatures) { + function_entry *entry = (function_entry *)iter.get(); + + if (entry->sig_id == mesa_instructions[i].BranchTarget) { + mesa_instructions[i].BranchTarget = entry->inst; + break; + } + } + break; + default: + break; + } + } +} + +static void +print_program(struct prog_instruction *mesa_instructions, + ir_instruction **mesa_instruction_annotation, + int num_instructions) +{ + ir_instruction *last_ir = NULL; + int i; + int indent = 0; + + for (i = 0; i < num_instructions; i++) { + struct prog_instruction *mesa_inst = mesa_instructions + i; + ir_instruction *ir = mesa_instruction_annotation[i]; + + fprintf(stdout, "%3d: ", i); + + if (last_ir != ir && ir) { + int j; + + for (j = 0; j < indent; j++) { + fprintf(stdout, " "); + } + ir->print(); + printf("\n"); + last_ir = ir; + + fprintf(stdout, " "); /* line number spacing. */ + } + + indent = _mesa_fprint_instruction_opt(stdout, mesa_inst, indent, + PROG_PRINT_DEBUG, NULL); + } +} + + +/** + * Count resources used by the given gpu program (number of texture + * samplers, etc). + */ +static void +count_resources(struct gl_program *prog) +{ + unsigned int i; + + prog->SamplersUsed = 0; + + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = &prog->Instructions[i]; + + if (_mesa_is_tex_instruction(inst->Opcode)) { + prog->SamplerTargets[inst->TexSrcUnit] = + (gl_texture_index)inst->TexSrcTarget; + prog->SamplersUsed |= 1 << inst->TexSrcUnit; + if (inst->TexShadow) { + prog->ShadowSamplers |= 1 << inst->TexSrcUnit; + } + } + } + + _mesa_update_shader_textures_used(prog); +} + + +/** + * Check if the given vertex/fragment/shader program is within the + * resource limits of the context (number of texture units, etc). + * If any of those checks fail, record a linker error. + * + * XXX more checks are needed... + */ +static void +check_resources(const struct gl_context *ctx, + struct gl_shader_program *shader_program, + struct gl_program *prog) +{ + switch (prog->Target) { + case GL_VERTEX_PROGRAM_ARB: + if (_mesa_bitcount(prog->SamplersUsed) > + ctx->Const.MaxVertexTextureImageUnits) { + linker_error(shader_program, + "Too many vertex shader texture samplers"); + } + if (prog->Parameters->NumParameters > MAX_UNIFORMS) { + linker_error(shader_program, "Too many vertex shader constants"); + } + break; + case MESA_GEOMETRY_PROGRAM: + if (_mesa_bitcount(prog->SamplersUsed) > + ctx->Const.MaxGeometryTextureImageUnits) { + linker_error(shader_program, + "Too many geometry shader texture samplers"); + } + if (prog->Parameters->NumParameters > + MAX_GEOMETRY_UNIFORM_COMPONENTS / 4) { + linker_error(shader_program, "Too many geometry shader constants"); + } + break; + case GL_FRAGMENT_PROGRAM_ARB: + if (_mesa_bitcount(prog->SamplersUsed) > + ctx->Const.MaxTextureImageUnits) { + linker_error(shader_program, + "Too many fragment shader texture samplers"); + } + if (prog->Parameters->NumParameters > MAX_UNIFORMS) { + linker_error(shader_program, "Too many fragment shader constants"); + } + break; + default: + _mesa_problem(ctx, "unexpected program type in check_resources()"); + } +} + + + +struct uniform_sort { + struct gl_uniform *u; + int pos; +}; + +/* The shader_program->Uniforms list is almost sorted in increasing + * uniform->{Frag,Vert}Pos locations, but not quite when there are + * uniforms shared between targets. We need to add parameters in + * increasing order for the targets. + */ +static int +sort_uniforms(const void *a, const void *b) +{ + struct uniform_sort *u1 = (struct uniform_sort *)a; + struct uniform_sort *u2 = (struct uniform_sort *)b; + + return u1->pos - u2->pos; +} + +/* Add the uniforms to the parameters. The linker chose locations + * in our parameters lists (which weren't created yet), which the + * uniforms code will use to poke values into our parameters list + * when uniforms are updated. + */ +static void +add_uniforms_to_parameters_list(struct gl_shader_program *shader_program, + struct gl_shader *shader, + struct gl_program *prog) +{ + unsigned int i; + unsigned int next_sampler = 0, num_uniforms = 0; + struct uniform_sort *sorted_uniforms; + + sorted_uniforms = ralloc_array(NULL, struct uniform_sort, + shader_program->Uniforms->NumUniforms); + + for (i = 0; i < shader_program->Uniforms->NumUniforms; i++) { + struct gl_uniform *uniform = shader_program->Uniforms->Uniforms + i; + int parameter_index = -1; + + switch (shader->Type) { + case GL_VERTEX_SHADER: + parameter_index = uniform->VertPos; + break; + case GL_FRAGMENT_SHADER: + parameter_index = uniform->FragPos; + break; + case GL_GEOMETRY_SHADER: + parameter_index = uniform->GeomPos; + break; + } + + /* Only add uniforms used in our target. */ + if (parameter_index != -1) { + sorted_uniforms[num_uniforms].pos = parameter_index; + sorted_uniforms[num_uniforms].u = uniform; + num_uniforms++; + } + } + + qsort(sorted_uniforms, num_uniforms, sizeof(struct uniform_sort), + sort_uniforms); + + for (i = 0; i < num_uniforms; i++) { + struct gl_uniform *uniform = sorted_uniforms[i].u; + int parameter_index = sorted_uniforms[i].pos; + const glsl_type *type = uniform->Type; + unsigned int size; + + if (type->is_vector() || + type->is_scalar()) { + size = type->vector_elements; + } else { + size = type_size(type) * 4; + } + + gl_register_file file; + if (type->is_sampler() || + (type->is_array() && type->fields.array->is_sampler())) { + file = PROGRAM_SAMPLER; + } else { + file = PROGRAM_UNIFORM; + } + + GLint index = _mesa_lookup_parameter_index(prog->Parameters, -1, + uniform->Name); + + if (index < 0) { + index = _mesa_add_parameter(prog->Parameters, file, + uniform->Name, size, type->gl_type, + NULL, NULL, 0x0); + + /* Sampler uniform values are stored in prog->SamplerUnits, + * and the entry in that array is selected by this index we + * store in ParameterValues[]. + */ + if (file == PROGRAM_SAMPLER) { + for (unsigned int j = 0; j < size / 4; j++) + prog->Parameters->ParameterValues[index + j][0].f = next_sampler++; + } + + /* The location chosen in the Parameters list here (returned + * from _mesa_add_uniform) has to match what the linker chose. + */ + if (index != parameter_index) { + linker_error(shader_program, + "Allocation of uniform `%s' to target failed " + "(%d vs %d)\n", + uniform->Name, index, parameter_index); + } + } + } + + ralloc_free(sorted_uniforms); +} + +static void +set_uniform_initializer(struct gl_context *ctx, void *mem_ctx, + struct gl_shader_program *shader_program, + const char *name, const glsl_type *type, + ir_constant *val) +{ + if (type->is_record()) { + ir_constant *field_constant; + + field_constant = (ir_constant *)val->components.get_head(); + + for (unsigned int i = 0; i < type->length; i++) { + const glsl_type *field_type = type->fields.structure[i].type; + const char *field_name = ralloc_asprintf(mem_ctx, "%s.%s", name, + type->fields.structure[i].name); + set_uniform_initializer(ctx, mem_ctx, shader_program, field_name, + field_type, field_constant); + field_constant = (ir_constant *)field_constant->next; + } + return; + } + + int loc = _mesa_get_uniform_location(ctx, shader_program, name); + + if (loc == -1) { + linker_error(shader_program, + "Couldn't find uniform for initializer %s\n", name); + return; + } + + for (unsigned int i = 0; i < (type->is_array() ? type->length : 1); i++) { + ir_constant *element; + const glsl_type *element_type; + if (type->is_array()) { + element = val->array_elements[i]; + element_type = type->fields.array; + } else { + element = val; + element_type = type; + } + + void *values; + + if (element_type->base_type == GLSL_TYPE_BOOL) { + int *conv = ralloc_array(mem_ctx, int, element_type->components()); + for (unsigned int j = 0; j < element_type->components(); j++) { + conv[j] = element->value.b[j]; + } + values = (void *)conv; + element_type = glsl_type::get_instance(GLSL_TYPE_INT, + element_type->vector_elements, + 1); + } else { + values = &element->value; + } + + if (element_type->is_matrix()) { + _mesa_uniform_matrix(ctx, shader_program, + element_type->matrix_columns, + element_type->vector_elements, + loc, 1, GL_FALSE, (GLfloat *)values); + loc += element_type->matrix_columns; + } else { + _mesa_uniform(ctx, shader_program, loc, element_type->matrix_columns, + values, element_type->gl_type); + loc += type_size(element_type); + } + } +} + +static void +set_uniform_initializers(struct gl_context *ctx, + struct gl_shader_program *shader_program) +{ + void *mem_ctx = NULL; + + for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) { + struct gl_shader *shader = shader_program->_LinkedShaders[i]; + + if (shader == NULL) + continue; + + foreach_iter(exec_list_iterator, iter, *shader->ir) { + ir_instruction *ir = (ir_instruction *)iter.get(); + ir_variable *var = ir->as_variable(); + + if (!var || var->mode != ir_var_uniform || !var->constant_value) + continue; + + if (!mem_ctx) + mem_ctx = ralloc_context(NULL); + + set_uniform_initializer(ctx, mem_ctx, shader_program, var->name, + var->type, var->constant_value); + } + } + + ralloc_free(mem_ctx); +} + +/* + * On a basic block basis, tracks available PROGRAM_TEMPORARY register + * channels for copy propagation and updates following instructions to + * use the original versions. + * + * The ir_to_mesa_visitor lazily produces code assuming that this pass + * will occur. As an example, a TXP production before this pass: + * + * 0: MOV TEMP[1], INPUT[4].xyyy; + * 1: MOV TEMP[1].w, INPUT[4].wwww; + * 2: TXP TEMP[2], TEMP[1], texture[0], 2D; + * + * and after: + * + * 0: MOV TEMP[1], INPUT[4].xyyy; + * 1: MOV TEMP[1].w, INPUT[4].wwww; + * 2: TXP TEMP[2], INPUT[4].xyyw, texture[0], 2D; + * + * which allows for dead code elimination on TEMP[1]'s writes. + */ +void +ir_to_mesa_visitor::copy_propagate(void) +{ + ir_to_mesa_instruction **acp = rzalloc_array(mem_ctx, + ir_to_mesa_instruction *, + this->next_temp * 4); + int *acp_level = rzalloc_array(mem_ctx, int, this->next_temp * 4); + int level = 0; + + foreach_iter(exec_list_iterator, iter, this->instructions) { + ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get(); + + assert(inst->dst.file != PROGRAM_TEMPORARY + || inst->dst.index < this->next_temp); + + /* First, do any copy propagation possible into the src regs. */ + for (int r = 0; r < 3; r++) { + ir_to_mesa_instruction *first = NULL; + bool good = true; + int acp_base = inst->src[r].index * 4; + + if (inst->src[r].file != PROGRAM_TEMPORARY || + inst->src[r].reladdr) + continue; + + /* See if we can find entries in the ACP consisting of MOVs + * from the same src register for all the swizzled channels + * of this src register reference. + */ + for (int i = 0; i < 4; i++) { + int src_chan = GET_SWZ(inst->src[r].swizzle, i); + ir_to_mesa_instruction *copy_chan = acp[acp_base + src_chan]; + + if (!copy_chan) { + good = false; + break; + } + + assert(acp_level[acp_base + src_chan] <= level); + + if (!first) { + first = copy_chan; + } else { + if (first->src[0].file != copy_chan->src[0].file || + first->src[0].index != copy_chan->src[0].index) { + good = false; + break; + } + } + } + + if (good) { + /* We've now validated that we can copy-propagate to + * replace this src register reference. Do it. + */ + inst->src[r].file = first->src[0].file; + inst->src[r].index = first->src[0].index; + + int swizzle = 0; + for (int i = 0; i < 4; i++) { + int src_chan = GET_SWZ(inst->src[r].swizzle, i); + ir_to_mesa_instruction *copy_inst = acp[acp_base + src_chan]; + swizzle |= (GET_SWZ(copy_inst->src[0].swizzle, src_chan) << + (3 * i)); + } + inst->src[r].swizzle = swizzle; + } + } + + switch (inst->op) { + case OPCODE_BGNLOOP: + case OPCODE_ENDLOOP: + /* End of a basic block, clear the ACP entirely. */ + memset(acp, 0, sizeof(*acp) * this->next_temp * 4); + break; + + case OPCODE_IF: + ++level; + break; + + case OPCODE_ENDIF: + case OPCODE_ELSE: + /* Clear all channels written inside the block from the ACP, but + * leaving those that were not touched. + */ + for (int r = 0; r < this->next_temp; r++) { + for (int c = 0; c < 4; c++) { + if (!acp[4 * r + c]) + continue; + + if (acp_level[4 * r + c] >= level) + acp[4 * r + c] = NULL; + } + } + if (inst->op == OPCODE_ENDIF) + --level; + break; + + default: + /* Continuing the block, clear any written channels from + * the ACP. + */ + if (inst->dst.file == PROGRAM_TEMPORARY && inst->dst.reladdr) { + /* Any temporary might be written, so no copy propagation + * across this instruction. + */ + memset(acp, 0, sizeof(*acp) * this->next_temp * 4); + } else if (inst->dst.file == PROGRAM_OUTPUT && + inst->dst.reladdr) { + /* Any output might be written, so no copy propagation + * from outputs across this instruction. + */ + for (int r = 0; r < this->next_temp; r++) { + for (int c = 0; c < 4; c++) { + if (!acp[4 * r + c]) + continue; + + if (acp[4 * r + c]->src[0].file == PROGRAM_OUTPUT) + acp[4 * r + c] = NULL; + } + } + } else if (inst->dst.file == PROGRAM_TEMPORARY || + inst->dst.file == PROGRAM_OUTPUT) { + /* Clear where it's used as dst. */ + if (inst->dst.file == PROGRAM_TEMPORARY) { + for (int c = 0; c < 4; c++) { + if (inst->dst.writemask & (1 << c)) { + acp[4 * inst->dst.index + c] = NULL; + } + } + } + + /* Clear where it's used as src. */ + for (int r = 0; r < this->next_temp; r++) { + for (int c = 0; c < 4; c++) { + if (!acp[4 * r + c]) + continue; + + int src_chan = GET_SWZ(acp[4 * r + c]->src[0].swizzle, c); + + if (acp[4 * r + c]->src[0].file == inst->dst.file && + acp[4 * r + c]->src[0].index == inst->dst.index && + inst->dst.writemask & (1 << src_chan)) + { + acp[4 * r + c] = NULL; + } + } + } + } + break; + } + + /* If this is a copy, add it to the ACP. */ + if (inst->op == OPCODE_MOV && + inst->dst.file == PROGRAM_TEMPORARY && + !inst->dst.reladdr && + !inst->saturate && + !inst->src[0].reladdr && + !inst->src[0].negate) { + for (int i = 0; i < 4; i++) { + if (inst->dst.writemask & (1 << i)) { + acp[4 * inst->dst.index + i] = inst; + acp_level[4 * inst->dst.index + i] = level; + } + } + } + } + + ralloc_free(acp_level); + ralloc_free(acp); +} + + +/** + * Convert a shader's GLSL IR into a Mesa gl_program. + */ +static struct gl_program * +get_mesa_program(struct gl_context *ctx, + struct gl_shader_program *shader_program, + struct gl_shader *shader) +{ + ir_to_mesa_visitor v; + struct prog_instruction *mesa_instructions, *mesa_inst; + ir_instruction **mesa_instruction_annotation; + int i; + struct gl_program *prog; + GLenum target; + const char *target_string; + GLboolean progress; + struct gl_shader_compiler_options *options = + &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(shader->Type)]; + + switch (shader->Type) { + case GL_VERTEX_SHADER: + target = GL_VERTEX_PROGRAM_ARB; + target_string = "vertex"; + break; + case GL_FRAGMENT_SHADER: + target = GL_FRAGMENT_PROGRAM_ARB; + target_string = "fragment"; + break; + case GL_GEOMETRY_SHADER: + target = GL_GEOMETRY_PROGRAM_NV; + target_string = "geometry"; + break; + default: + assert(!"should not be reached"); + return NULL; + } + + validate_ir_tree(shader->ir); + + prog = ctx->Driver.NewProgram(ctx, target, shader_program->Name); + if (!prog) + return NULL; + prog->Parameters = _mesa_new_parameter_list(); + prog->Varying = _mesa_new_parameter_list(); + prog->Attributes = _mesa_new_parameter_list(); + v.ctx = ctx; + v.prog = prog; + v.shader_program = shader_program; + v.options = options; + + add_uniforms_to_parameters_list(shader_program, shader, prog); + + /* Emit Mesa IR for main(). */ + visit_exec_list(shader->ir, &v); + v.emit(NULL, OPCODE_END); + + /* Now emit bodies for any functions that were used. */ + do { + progress = GL_FALSE; + + foreach_iter(exec_list_iterator, iter, v.function_signatures) { + function_entry *entry = (function_entry *)iter.get(); + + if (!entry->bgn_inst) { + v.current_function = entry; + + entry->bgn_inst = v.emit(NULL, OPCODE_BGNSUB); + entry->bgn_inst->function = entry; + + visit_exec_list(&entry->sig->body, &v); + + ir_to_mesa_instruction *last; + last = (ir_to_mesa_instruction *)v.instructions.get_tail(); + if (last->op != OPCODE_RET) + v.emit(NULL, OPCODE_RET); + + ir_to_mesa_instruction *end; + end = v.emit(NULL, OPCODE_ENDSUB); + end->function = entry; + + progress = GL_TRUE; + } + } + } while (progress); + + prog->NumTemporaries = v.next_temp; + + int num_instructions = 0; + foreach_iter(exec_list_iterator, iter, v.instructions) { + num_instructions++; + } + + mesa_instructions = + (struct prog_instruction *)calloc(num_instructions, + sizeof(*mesa_instructions)); + mesa_instruction_annotation = ralloc_array(v.mem_ctx, ir_instruction *, + num_instructions); + + v.copy_propagate(); + + /* Convert ir_mesa_instructions into prog_instructions. + */ + mesa_inst = mesa_instructions; + i = 0; + foreach_iter(exec_list_iterator, iter, v.instructions) { + const ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get(); + + mesa_inst->Opcode = inst->op; + mesa_inst->CondUpdate = inst->cond_update; + if (inst->saturate) + mesa_inst->SaturateMode = SATURATE_ZERO_ONE; + mesa_inst->DstReg.File = inst->dst.file; + mesa_inst->DstReg.Index = inst->dst.index; + mesa_inst->DstReg.CondMask = inst->dst.cond_mask; + mesa_inst->DstReg.WriteMask = inst->dst.writemask; + mesa_inst->DstReg.RelAddr = inst->dst.reladdr != NULL; + mesa_inst->SrcReg[0] = mesa_src_reg_from_ir_src_reg(inst->src[0]); + mesa_inst->SrcReg[1] = mesa_src_reg_from_ir_src_reg(inst->src[1]); + mesa_inst->SrcReg[2] = mesa_src_reg_from_ir_src_reg(inst->src[2]); + mesa_inst->TexSrcUnit = inst->sampler; + mesa_inst->TexSrcTarget = inst->tex_target; + mesa_inst->TexShadow = inst->tex_shadow; + mesa_instruction_annotation[i] = inst->ir; + + /* Set IndirectRegisterFiles. */ + if (mesa_inst->DstReg.RelAddr) + prog->IndirectRegisterFiles |= 1 << mesa_inst->DstReg.File; + + /* Update program's bitmask of indirectly accessed register files */ + for (unsigned src = 0; src < 3; src++) + if (mesa_inst->SrcReg[src].RelAddr) + prog->IndirectRegisterFiles |= 1 << mesa_inst->SrcReg[src].File; + + switch (mesa_inst->Opcode) { + case OPCODE_IF: + if (options->MaxIfDepth == 0) { + linker_warning(shader_program, + "Couldn't flatten if-statement. " + "This will likely result in software " + "rasterization.\n"); + } + break; + case OPCODE_BGNLOOP: + if (options->EmitNoLoops) { + linker_warning(shader_program, + "Couldn't unroll loop. " + "This will likely result in software " + "rasterization.\n"); + } + break; + case OPCODE_CONT: + if (options->EmitNoCont) { + linker_warning(shader_program, + "Couldn't lower continue-statement. " + "This will likely result in software " + "rasterization.\n"); + } + break; + case OPCODE_BGNSUB: + inst->function->inst = i; + mesa_inst->Comment = strdup(inst->function->sig->function_name()); + break; + case OPCODE_ENDSUB: + mesa_inst->Comment = strdup(inst->function->sig->function_name()); + break; + case OPCODE_CAL: + mesa_inst->BranchTarget = inst->function->sig_id; /* rewritten later */ + break; + case OPCODE_ARL: + prog->NumAddressRegs = 1; + break; + default: + break; + } + + mesa_inst++; + i++; + + if (!shader_program->LinkStatus) + break; + } + + if (!shader_program->LinkStatus) { + free(mesa_instructions); + _mesa_reference_program(ctx, &shader->Program, NULL); + return NULL; + } + + set_branchtargets(&v, mesa_instructions, num_instructions); + + if (ctx->Shader.Flags & GLSL_DUMP) { + printf("\n"); + printf("GLSL IR for linked %s program %d:\n", target_string, + shader_program->Name); + _mesa_print_ir(shader->ir, NULL); + printf("\n"); + printf("\n"); + printf("Mesa IR for linked %s program %d:\n", target_string, + shader_program->Name); + print_program(mesa_instructions, mesa_instruction_annotation, + num_instructions); + } + + prog->Instructions = mesa_instructions; + prog->NumInstructions = num_instructions; + + do_set_program_inouts(shader->ir, prog); + count_resources(prog); + + check_resources(ctx, shader_program, prog); + + _mesa_reference_program(ctx, &shader->Program, prog); + + if ((ctx->Shader.Flags & GLSL_NO_OPT) == 0) { + _mesa_optimize_program(ctx, prog); + } + + return prog; +} + +extern "C" { + +/** + * Link a shader. + * Called via ctx->Driver.LinkShader() + * This actually involves converting GLSL IR into Mesa gl_programs with + * code lowering and other optimizations. + */ +GLboolean +_mesa_ir_link_shader(struct gl_context *ctx, struct gl_shader_program *prog) +{ + assert(prog->LinkStatus); + + for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + bool progress; + exec_list *ir = prog->_LinkedShaders[i]->ir; + const struct gl_shader_compiler_options *options = + &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(prog->_LinkedShaders[i]->Type)]; + + do { + progress = false; + + /* Lowering */ + do_mat_op_to_vec(ir); + lower_instructions(ir, (MOD_TO_FRACT | DIV_TO_MUL_RCP | EXP_TO_EXP2 + | LOG_TO_LOG2 | INT_DIV_TO_MUL_RCP + | ((options->EmitNoPow) ? POW_TO_EXP2 : 0))); + + progress = do_lower_jumps(ir, true, true, options->EmitNoMainReturn, options->EmitNoCont, options->EmitNoLoops) || progress; + + progress = do_common_optimization(ir, true, options->MaxUnrollIterations) || progress; + + progress = lower_quadop_vector(ir, true) || progress; + + if (options->MaxIfDepth == 0) + progress = lower_discard(ir) || progress; + + progress = lower_if_to_cond_assign(ir, options->MaxIfDepth) || progress; + + if (options->EmitNoNoise) + progress = lower_noise(ir) || progress; + + /* If there are forms of indirect addressing that the driver + * cannot handle, perform the lowering pass. + */ + if (options->EmitNoIndirectInput || options->EmitNoIndirectOutput + || options->EmitNoIndirectTemp || options->EmitNoIndirectUniform) + progress = + lower_variable_index_to_cond_assign(ir, + options->EmitNoIndirectInput, + options->EmitNoIndirectOutput, + options->EmitNoIndirectTemp, + options->EmitNoIndirectUniform) + || progress; + + progress = do_vec_index_to_cond_assign(ir) || progress; + } while (progress); + + validate_ir_tree(ir); + } + + for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { + struct gl_program *linked_prog; + + if (prog->_LinkedShaders[i] == NULL) + continue; + + linked_prog = get_mesa_program(ctx, prog, prog->_LinkedShaders[i]); + + if (linked_prog) { + bool ok = true; + + switch (prog->_LinkedShaders[i]->Type) { + case GL_VERTEX_SHADER: + _mesa_reference_vertprog(ctx, &prog->VertexProgram, + (struct gl_vertex_program *)linked_prog); + ok = ctx->Driver.ProgramStringNotify(ctx, GL_VERTEX_PROGRAM_ARB, + linked_prog); + break; + case GL_FRAGMENT_SHADER: + _mesa_reference_fragprog(ctx, &prog->FragmentProgram, + (struct gl_fragment_program *)linked_prog); + ok = ctx->Driver.ProgramStringNotify(ctx, GL_FRAGMENT_PROGRAM_ARB, + linked_prog); + break; + case GL_GEOMETRY_SHADER: + _mesa_reference_geomprog(ctx, &prog->GeometryProgram, + (struct gl_geometry_program *)linked_prog); + ok = ctx->Driver.ProgramStringNotify(ctx, GL_GEOMETRY_PROGRAM_NV, + linked_prog); + break; + } + if (!ok) { + return GL_FALSE; + } + } + + _mesa_reference_program(ctx, &linked_prog, NULL); + } + + return GL_TRUE; +} + + +/** + * Compile a GLSL shader. Called via glCompileShader(). + */ +void +_mesa_glsl_compile_shader(struct gl_context *ctx, struct gl_shader *shader) +{ + struct _mesa_glsl_parse_state *state = + new(shader) _mesa_glsl_parse_state(ctx, shader->Type, shader); + + const char *source = shader->Source; + /* Check if the user called glCompileShader without first calling + * glShaderSource. This should fail to compile, but not raise a GL_ERROR. + */ + if (source == NULL) { + shader->CompileStatus = GL_FALSE; + return; + } + + state->error = preprocess(state, &source, &state->info_log, + &ctx->Extensions, ctx->API); + + if (ctx->Shader.Flags & GLSL_DUMP) { + printf("GLSL source for %s shader %d:\n", + _mesa_glsl_shader_target_name(state->target), shader->Name); + printf("%s\n", shader->Source); + } + + if (!state->error) { + _mesa_glsl_lexer_ctor(state, source); + _mesa_glsl_parse(state); + _mesa_glsl_lexer_dtor(state); + } + + ralloc_free(shader->ir); + shader->ir = new(shader) exec_list; + if (!state->error && !state->translation_unit.is_empty()) + _mesa_ast_to_hir(shader->ir, state); + + if (!state->error && !shader->ir->is_empty()) { + validate_ir_tree(shader->ir); + + /* Do some optimization at compile time to reduce shader IR size + * and reduce later work if the same shader is linked multiple times + */ + while (do_common_optimization(shader->ir, false, 32)) + ; + + validate_ir_tree(shader->ir); + } + + shader->symbols = state->symbols; + + shader->CompileStatus = !state->error; + shader->InfoLog = state->info_log; + shader->Version = state->language_version; + memcpy(shader->builtins_to_link, state->builtins_to_link, + sizeof(shader->builtins_to_link[0]) * state->num_builtins_to_link); + shader->num_builtins_to_link = state->num_builtins_to_link; + + if (ctx->Shader.Flags & GLSL_LOG) { + _mesa_write_shader_to_file(shader); + } + + if (ctx->Shader.Flags & GLSL_DUMP) { + if (shader->CompileStatus) { + printf("GLSL IR for shader %d:\n", shader->Name); + _mesa_print_ir(shader->ir, NULL); + printf("\n\n"); + } else { + printf("GLSL shader %d failed to compile.\n", shader->Name); + } + if (shader->InfoLog && shader->InfoLog[0] != 0) { + printf("GLSL shader %d info log:\n", shader->Name); + printf("%s\n", shader->InfoLog); + } + } + + /* Retain any live IR, but trash the rest. */ + reparent_ir(shader->ir, shader->ir); + + ralloc_free(state); +} + + +/** + * Link a GLSL shader program. Called via glLinkProgram(). + */ +void +_mesa_glsl_link_shader(struct gl_context *ctx, struct gl_shader_program *prog) +{ + unsigned int i; + + _mesa_clear_shader_program_data(ctx, prog); + + prog->LinkStatus = GL_TRUE; + + for (i = 0; i < prog->NumShaders; i++) { + if (!prog->Shaders[i]->CompileStatus) { + linker_error(prog, "linking with uncompiled shader"); + prog->LinkStatus = GL_FALSE; + } + } + + prog->Varying = _mesa_new_parameter_list(); + _mesa_reference_vertprog(ctx, &prog->VertexProgram, NULL); + _mesa_reference_fragprog(ctx, &prog->FragmentProgram, NULL); + _mesa_reference_geomprog(ctx, &prog->GeometryProgram, NULL); + + if (prog->LinkStatus) { + link_shaders(ctx, prog); + } + + if (prog->LinkStatus) { + if (!ctx->Driver.LinkShader(ctx, prog)) { + prog->LinkStatus = GL_FALSE; + } + } + + set_uniform_initializers(ctx, prog); + + if (ctx->Shader.Flags & GLSL_DUMP) { + if (!prog->LinkStatus) { + printf("GLSL shader program %d failed to link\n", prog->Name); + } + + if (prog->InfoLog && prog->InfoLog[0] != 0) { + printf("GLSL shader program %d info log:\n", prog->Name); + printf("%s\n", prog->InfoLog); + } + } +} + +} /* extern "C" */ diff --git a/mesalib/src/mesa/program/prog_instruction.h b/mesalib/src/mesa/program/prog_instruction.h index 19e9b95a3..db2b594e7 100644 --- a/mesalib/src/mesa/program/prog_instruction.h +++ b/mesalib/src/mesa/program/prog_instruction.h @@ -1,454 +1,454 @@ -/* - * Mesa 3-D graphics library - * Version: 7.3 - * - * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - - -/** - * \file prog_instruction.h - * - * Vertex/fragment program instruction datatypes and constants. - * - * \author Brian Paul - * \author Keith Whitwell - * \author Ian Romanick - */ - - -#ifndef PROG_INSTRUCTION_H -#define PROG_INSTRUCTION_H - - -#include "main/glheader.h" - - -/** - * Swizzle indexes. - * Do not change! - */ -/*@{*/ -#define SWIZZLE_X 0 -#define SWIZZLE_Y 1 -#define SWIZZLE_Z 2 -#define SWIZZLE_W 3 -#define SWIZZLE_ZERO 4 /**< For SWZ instruction only */ -#define SWIZZLE_ONE 5 /**< For SWZ instruction only */ -#define SWIZZLE_NIL 7 /**< used during shader code gen (undefined value) */ -/*@}*/ - -#define MAKE_SWIZZLE4(a,b,c,d) (((a)<<0) | ((b)<<3) | ((c)<<6) | ((d)<<9)) -#define SWIZZLE_NOOP MAKE_SWIZZLE4(0,1,2,3) -#define GET_SWZ(swz, idx) (((swz) >> ((idx)*3)) & 0x7) -#define GET_BIT(msk, idx) (((msk) >> (idx)) & 0x1) - -#define SWIZZLE_XYZW MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W) -#define SWIZZLE_XXXX MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X) -#define SWIZZLE_YYYY MAKE_SWIZZLE4(SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y) -#define SWIZZLE_ZZZZ MAKE_SWIZZLE4(SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z) -#define SWIZZLE_WWWW MAKE_SWIZZLE4(SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W) - - -/** - * Writemask values, 1 bit per component. - */ -/*@{*/ -#define WRITEMASK_X 0x1 -#define WRITEMASK_Y 0x2 -#define WRITEMASK_XY 0x3 -#define WRITEMASK_Z 0x4 -#define WRITEMASK_XZ 0x5 -#define WRITEMASK_YZ 0x6 -#define WRITEMASK_XYZ 0x7 -#define WRITEMASK_W 0x8 -#define WRITEMASK_XW 0x9 -#define WRITEMASK_YW 0xa -#define WRITEMASK_XYW 0xb -#define WRITEMASK_ZW 0xc -#define WRITEMASK_XZW 0xd -#define WRITEMASK_YZW 0xe -#define WRITEMASK_XYZW 0xf -/*@}*/ - - -/** - * Condition codes - */ -/*@{*/ -#define COND_GT 1 /**< greater than zero */ -#define COND_EQ 2 /**< equal to zero */ -#define COND_LT 3 /**< less than zero */ -#define COND_UN 4 /**< unordered (NaN) */ -#define COND_GE 5 /**< greater than or equal to zero */ -#define COND_LE 6 /**< less than or equal to zero */ -#define COND_NE 7 /**< not equal to zero */ -#define COND_TR 8 /**< always true */ -#define COND_FL 9 /**< always false */ -/*@}*/ - - -/** - * Instruction precision for GL_NV_fragment_program - */ -/*@{*/ -#define FLOAT32 0x1 -#define FLOAT16 0x2 -#define FIXED12 0x4 -/*@}*/ - - -/** - * Saturation modes when storing values. - */ -/*@{*/ -#define SATURATE_OFF 0 -#define SATURATE_ZERO_ONE 1 -/*@}*/ - - -/** - * Per-component negation masks - */ -/*@{*/ -#define NEGATE_X 0x1 -#define NEGATE_Y 0x2 -#define NEGATE_Z 0x4 -#define NEGATE_W 0x8 -#define NEGATE_XYZ 0x7 -#define NEGATE_XYZW 0xf -#define NEGATE_NONE 0x0 -/*@}*/ - - -/** - * Program instruction opcodes for vertex, fragment and geometry programs. - */ -typedef enum prog_opcode { - /* ARB_vp ARB_fp NV_vp NV_fp GLSL */ - /*------------------------------------------*/ - OPCODE_NOP = 0, /* X */ - OPCODE_ABS, /* X X 1.1 X */ - OPCODE_ADD, /* X X X X X */ - OPCODE_AND, /* */ - OPCODE_ARA, /* 2 */ - OPCODE_ARL, /* X X X */ - OPCODE_ARL_NV, /* 2 */ - OPCODE_ARR, /* 2 */ - OPCODE_BGNLOOP, /* opt */ - OPCODE_BGNSUB, /* opt */ - OPCODE_BRA, /* 2 */ - OPCODE_BRK, /* 2 opt */ - OPCODE_CAL, /* 2 2 opt */ - OPCODE_CMP, /* X X */ - OPCODE_CONT, /* opt */ - OPCODE_COS, /* X 2 X X */ - OPCODE_DDX, /* X X */ - OPCODE_DDY, /* X X */ - OPCODE_DP2, /* 2 X */ - OPCODE_DP2A, /* 2 */ - OPCODE_DP3, /* X X X X X */ - OPCODE_DP4, /* X X X X X */ - OPCODE_DPH, /* X X 1.1 */ - OPCODE_DST, /* X X X X */ - OPCODE_ELSE, /* opt */ - OPCODE_EMIT_VERTEX,/* X */ - OPCODE_END, /* X X X X opt */ - OPCODE_END_PRIMITIVE,/* X */ - OPCODE_ENDIF, /* opt */ - OPCODE_ENDLOOP, /* opt */ - OPCODE_ENDSUB, /* opt */ - OPCODE_EX2, /* X X 2 X X */ - OPCODE_EXP, /* X X */ - OPCODE_FLR, /* X X 2 X X */ - OPCODE_FRC, /* X X 2 X X */ - OPCODE_IF, /* opt */ - OPCODE_KIL, /* X X */ - OPCODE_KIL_NV, /* X X */ - OPCODE_LG2, /* X X 2 X X */ - OPCODE_LIT, /* X X X X */ - OPCODE_LOG, /* X X */ - OPCODE_LRP, /* X X */ - OPCODE_MAD, /* X X X X X */ - OPCODE_MAX, /* X X X X X */ - OPCODE_MIN, /* X X X X X */ - OPCODE_MOV, /* X X X X X */ - OPCODE_MUL, /* X X X X X */ - OPCODE_NOISE1, /* X */ - OPCODE_NOISE2, /* X */ - OPCODE_NOISE3, /* X */ - OPCODE_NOISE4, /* X */ - OPCODE_NOT, /* */ - OPCODE_NRM3, /* */ - OPCODE_NRM4, /* */ - OPCODE_OR, /* */ - OPCODE_PK2H, /* X */ - OPCODE_PK2US, /* X */ - OPCODE_PK4B, /* X */ - OPCODE_PK4UB, /* X */ - OPCODE_POW, /* X X X X */ - OPCODE_POPA, /* 3 */ - OPCODE_PRINT, /* X X */ - OPCODE_PUSHA, /* 3 */ - OPCODE_RCC, /* 1.1 */ - OPCODE_RCP, /* X X X X X */ - OPCODE_RET, /* 2 2 opt */ - OPCODE_RFL, /* X X */ - OPCODE_RSQ, /* X X X X X */ - OPCODE_SCS, /* X X */ - OPCODE_SEQ, /* 2 X X */ - OPCODE_SFL, /* 2 X */ - OPCODE_SGE, /* X X X X X */ - OPCODE_SGT, /* 2 X X */ - OPCODE_SIN, /* X 2 X X */ - OPCODE_SLE, /* 2 X X */ - OPCODE_SLT, /* X X X X X */ - OPCODE_SNE, /* 2 X X */ - OPCODE_SSG, /* 2 X */ - OPCODE_STR, /* 2 X */ - OPCODE_SUB, /* X X 1.1 X X */ - OPCODE_SWZ, /* X X X */ - OPCODE_TEX, /* X 3 X X */ - OPCODE_TXB, /* X 3 X */ - OPCODE_TXD, /* X X */ - OPCODE_TXL, /* 3 2 X */ - OPCODE_TXP, /* X X */ - OPCODE_TXP_NV, /* 3 X */ - OPCODE_TRUNC, /* X */ - OPCODE_UP2H, /* X */ - OPCODE_UP2US, /* X */ - OPCODE_UP4B, /* X */ - OPCODE_UP4UB, /* X */ - OPCODE_X2D, /* X */ - OPCODE_XOR, /* */ - OPCODE_XPD, /* X X */ - MAX_OPCODE -} gl_inst_opcode; - - -/** - * Number of bits for the src/dst register Index field. - * This limits the size of temp/uniform register files. - */ -#define INST_INDEX_BITS 12 - - -/** - * Instruction source register. - */ -struct prog_src_register -{ - GLuint File:4; /**< One of the PROGRAM_* register file values. */ - GLint Index:(INST_INDEX_BITS+1); /**< Extra bit here for sign bit. - * May be negative for relative addressing. - */ - GLuint Swizzle:12; - GLuint RelAddr:1; - - /** Take the component-wise absolute value */ - GLuint Abs:1; - - /** - * Post-Abs negation. - * This will either be NEGATE_NONE or NEGATE_XYZW, except for the SWZ - * instruction which allows per-component negation. - */ - GLuint Negate:4; - - /** - * Is the register two-dimensional. - * Two dimensional registers are of the - * REGISTER[index][index2] format. - * They are used by the geometry shaders where - * the first index is the index within an array - * and the second index is the semantic of the - * array, e.g. gl_PositionIn[index] would become - * INPUT[index][gl_PositionIn] - */ - GLuint HasIndex2:1; - GLuint RelAddr2:1; - GLint Index2:(INST_INDEX_BITS+1); /**< Extra bit here for sign bit. - * May be negative for relative - * addressing. */ -}; - - -/** - * Instruction destination register. - */ -struct prog_dst_register -{ - GLuint File:4; /**< One of the PROGRAM_* register file values */ - GLuint Index:INST_INDEX_BITS; /**< Unsigned, never negative */ - GLuint WriteMask:4; - GLuint RelAddr:1; - - /** - * \name Conditional destination update control. - * - * \since - * NV_fragment_program, NV_fragment_program_option, NV_vertex_program2, - * NV_vertex_program2_option. - */ - /*@{*/ - /** - * Takes one of the 9 possible condition values (EQ, FL, GT, GE, LE, LT, - * NE, TR, or UN). Dest reg is only written to if the matching - * (swizzled) condition code value passes. When a conditional update mask - * is not specified, this will be \c COND_TR. - */ - GLuint CondMask:4; - - /** - * Condition code swizzle value. - */ - GLuint CondSwizzle:12; - - /** - * Selects the condition code register to use for conditional destination - * update masking. In NV_fragmnet_program or NV_vertex_program2 mode, only - * condition code register 0 is available. In NV_vertex_program3 mode, - * condition code registers 0 and 1 are available. - */ - GLuint CondSrc:1; - /*@}*/ -}; - - -/** - * Vertex/fragment program instruction. - */ -struct prog_instruction -{ - gl_inst_opcode Opcode; - struct prog_src_register SrcReg[3]; - struct prog_dst_register DstReg; - - /** - * Indicates that the instruction should update the condition code - * register. - * - * \since - * NV_fragment_program, NV_fragment_program_option, NV_vertex_program2, - * NV_vertex_program2_option. - */ - GLuint CondUpdate:1; - - /** - * If prog_instruction::CondUpdate is \c GL_TRUE, this value selects the - * condition code register that is to be updated. - * - * In GL_NV_fragment_program or GL_NV_vertex_program2 mode, only condition - * code register 0 is available. In GL_NV_vertex_program3 mode, condition - * code registers 0 and 1 are available. - * - * \since - * NV_fragment_program, NV_fragment_program_option, NV_vertex_program2, - * NV_vertex_program2_option. - */ - GLuint CondDst:1; - - /** - * Saturate each value of the vectored result to the range [0,1] or the - * range [-1,1]. \c SSAT mode (i.e., saturation to the range [-1,1]) is - * only available in NV_fragment_program2 mode. - * Value is one of the SATURATE_* tokens. - * - * \since - * NV_fragment_program, NV_fragment_program_option, NV_vertex_program3. - */ - GLuint SaturateMode:2; - - /** - * Per-instruction selectable precision: FLOAT32, FLOAT16, FIXED12. - * - * \since - * NV_fragment_program, NV_fragment_program_option. - */ - GLuint Precision:3; - - /** - * \name Extra fields for TEX, TXB, TXD, TXL, TXP instructions. - */ - /*@{*/ - /** Source texture unit. */ - GLuint TexSrcUnit:5; - - /** Source texture target, one of TEXTURE_{1D,2D,3D,CUBE,RECT}_INDEX */ - GLuint TexSrcTarget:3; - - /** True if tex instruction should do shadow comparison */ - GLuint TexShadow:1; - /*@}*/ - - /** - * For BRA and CAL instructions, the location to jump to. - * For BGNLOOP, points to ENDLOOP (and vice-versa). - * For BRK, points to ENDLOOP - * For IF, points to ELSE or ENDIF. - * For ELSE, points to ENDIF. - */ - GLint BranchTarget; - - /** for debugging purposes */ - const char *Comment; - - /** Arbitrary data. Used for OPCODE_PRINT and some drivers */ - void *Data; - - /** for driver use (try to remove someday) */ - GLint Aux; -}; - - -extern void -_mesa_init_instructions(struct prog_instruction *inst, GLuint count); - -extern struct prog_instruction * -_mesa_alloc_instructions(GLuint numInst); - -extern struct prog_instruction * -_mesa_realloc_instructions(struct prog_instruction *oldInst, - GLuint numOldInst, GLuint numNewInst); - -extern struct prog_instruction * -_mesa_copy_instructions(struct prog_instruction *dest, - const struct prog_instruction *src, GLuint n); - -extern void -_mesa_free_instructions(struct prog_instruction *inst, GLuint count); - -extern GLuint -_mesa_num_inst_src_regs(gl_inst_opcode opcode); - -extern GLuint -_mesa_num_inst_dst_regs(gl_inst_opcode opcode); - -extern GLboolean -_mesa_is_tex_instruction(gl_inst_opcode opcode); - -extern GLboolean -_mesa_check_soa_dependencies(const struct prog_instruction *inst); - -extern const char * -_mesa_opcode_string(gl_inst_opcode opcode); - - -#endif /* PROG_INSTRUCTION_H */ +/* + * Mesa 3-D graphics library + * Version: 7.3 + * + * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + + +/** + * \file prog_instruction.h + * + * Vertex/fragment program instruction datatypes and constants. + * + * \author Brian Paul + * \author Keith Whitwell + * \author Ian Romanick + */ + + +#ifndef PROG_INSTRUCTION_H +#define PROG_INSTRUCTION_H + + +#include "main/glheader.h" + + +/** + * Swizzle indexes. + * Do not change! + */ +/*@{*/ +#define SWIZZLE_X 0 +#define SWIZZLE_Y 1 +#define SWIZZLE_Z 2 +#define SWIZZLE_W 3 +#define SWIZZLE_ZERO 4 /**< For SWZ instruction only */ +#define SWIZZLE_ONE 5 /**< For SWZ instruction only */ +#define SWIZZLE_NIL 7 /**< used during shader code gen (undefined value) */ +/*@}*/ + +#define MAKE_SWIZZLE4(a,b,c,d) (((a)<<0) | ((b)<<3) | ((c)<<6) | ((d)<<9)) +#define SWIZZLE_NOOP MAKE_SWIZZLE4(0,1,2,3) +#define GET_SWZ(swz, idx) (((swz) >> ((idx)*3)) & 0x7) +#define GET_BIT(msk, idx) (((msk) >> (idx)) & 0x1) + +#define SWIZZLE_XYZW MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W) +#define SWIZZLE_XXXX MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X) +#define SWIZZLE_YYYY MAKE_SWIZZLE4(SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y) +#define SWIZZLE_ZZZZ MAKE_SWIZZLE4(SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z) +#define SWIZZLE_WWWW MAKE_SWIZZLE4(SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W) + + +/** + * Writemask values, 1 bit per component. + */ +/*@{*/ +#define WRITEMASK_X 0x1 +#define WRITEMASK_Y 0x2 +#define WRITEMASK_XY 0x3 +#define WRITEMASK_Z 0x4 +#define WRITEMASK_XZ 0x5 +#define WRITEMASK_YZ 0x6 +#define WRITEMASK_XYZ 0x7 +#define WRITEMASK_W 0x8 +#define WRITEMASK_XW 0x9 +#define WRITEMASK_YW 0xa +#define WRITEMASK_XYW 0xb +#define WRITEMASK_ZW 0xc +#define WRITEMASK_XZW 0xd +#define WRITEMASK_YZW 0xe +#define WRITEMASK_XYZW 0xf +/*@}*/ + + +/** + * Condition codes + */ +/*@{*/ +#define COND_GT 1 /**< greater than zero */ +#define COND_EQ 2 /**< equal to zero */ +#define COND_LT 3 /**< less than zero */ +#define COND_UN 4 /**< unordered (NaN) */ +#define COND_GE 5 /**< greater than or equal to zero */ +#define COND_LE 6 /**< less than or equal to zero */ +#define COND_NE 7 /**< not equal to zero */ +#define COND_TR 8 /**< always true */ +#define COND_FL 9 /**< always false */ +/*@}*/ + + +/** + * Instruction precision for GL_NV_fragment_program + */ +/*@{*/ +#define FLOAT32 0x1 +#define FLOAT16 0x2 +#define FIXED12 0x4 +/*@}*/ + + +/** + * Saturation modes when storing values. + */ +/*@{*/ +#define SATURATE_OFF 0 +#define SATURATE_ZERO_ONE 1 +/*@}*/ + + +/** + * Per-component negation masks + */ +/*@{*/ +#define NEGATE_X 0x1 +#define NEGATE_Y 0x2 +#define NEGATE_Z 0x4 +#define NEGATE_W 0x8 +#define NEGATE_XYZ 0x7 +#define NEGATE_XYZW 0xf +#define NEGATE_NONE 0x0 +/*@}*/ + + +/** + * Program instruction opcodes for vertex, fragment and geometry programs. + */ +typedef enum prog_opcode { + /* ARB_vp ARB_fp NV_vp NV_fp GLSL */ + /*------------------------------------------*/ + OPCODE_NOP = 0, /* X */ + OPCODE_ABS, /* X X 1.1 X */ + OPCODE_ADD, /* X X X X X */ + OPCODE_AND, /* */ + OPCODE_ARA, /* 2 */ + OPCODE_ARL, /* X X X */ + OPCODE_ARL_NV, /* 2 */ + OPCODE_ARR, /* 2 */ + OPCODE_BGNLOOP, /* opt */ + OPCODE_BGNSUB, /* opt */ + OPCODE_BRA, /* 2 */ + OPCODE_BRK, /* 2 opt */ + OPCODE_CAL, /* 2 2 opt */ + OPCODE_CMP, /* X X */ + OPCODE_CONT, /* opt */ + OPCODE_COS, /* X 2 X X */ + OPCODE_DDX, /* X X */ + OPCODE_DDY, /* X X */ + OPCODE_DP2, /* 2 X */ + OPCODE_DP2A, /* 2 */ + OPCODE_DP3, /* X X X X X */ + OPCODE_DP4, /* X X X X X */ + OPCODE_DPH, /* X X 1.1 */ + OPCODE_DST, /* X X X X */ + OPCODE_ELSE, /* opt */ + OPCODE_EMIT_VERTEX,/* X */ + OPCODE_END, /* X X X X opt */ + OPCODE_END_PRIMITIVE,/* X */ + OPCODE_ENDIF, /* opt */ + OPCODE_ENDLOOP, /* opt */ + OPCODE_ENDSUB, /* opt */ + OPCODE_EX2, /* X X 2 X X */ + OPCODE_EXP, /* X X */ + OPCODE_FLR, /* X X 2 X X */ + OPCODE_FRC, /* X X 2 X X */ + OPCODE_IF, /* opt */ + OPCODE_KIL, /* X X */ + OPCODE_KIL_NV, /* X X */ + OPCODE_LG2, /* X X 2 X X */ + OPCODE_LIT, /* X X X X */ + OPCODE_LOG, /* X X */ + OPCODE_LRP, /* X X */ + OPCODE_MAD, /* X X X X X */ + OPCODE_MAX, /* X X X X X */ + OPCODE_MIN, /* X X X X X */ + OPCODE_MOV, /* X X X X X */ + OPCODE_MUL, /* X X X X X */ + OPCODE_NOISE1, /* X */ + OPCODE_NOISE2, /* X */ + OPCODE_NOISE3, /* X */ + OPCODE_NOISE4, /* X */ + OPCODE_NOT, /* */ + OPCODE_NRM3, /* */ + OPCODE_NRM4, /* */ + OPCODE_OR, /* */ + OPCODE_PK2H, /* X */ + OPCODE_PK2US, /* X */ + OPCODE_PK4B, /* X */ + OPCODE_PK4UB, /* X */ + OPCODE_POW, /* X X X X */ + OPCODE_POPA, /* 3 */ + OPCODE_PRINT, /* X X */ + OPCODE_PUSHA, /* 3 */ + OPCODE_RCC, /* 1.1 */ + OPCODE_RCP, /* X X X X X */ + OPCODE_RET, /* 2 2 opt */ + OPCODE_RFL, /* X X */ + OPCODE_RSQ, /* X X X X X */ + OPCODE_SCS, /* X X */ + OPCODE_SEQ, /* 2 X X */ + OPCODE_SFL, /* 2 X */ + OPCODE_SGE, /* X X X X X */ + OPCODE_SGT, /* 2 X X */ + OPCODE_SIN, /* X 2 X X */ + OPCODE_SLE, /* 2 X X */ + OPCODE_SLT, /* X X X X X */ + OPCODE_SNE, /* 2 X X */ + OPCODE_SSG, /* 2 X */ + OPCODE_STR, /* 2 X */ + OPCODE_SUB, /* X X 1.1 X X */ + OPCODE_SWZ, /* X X X */ + OPCODE_TEX, /* X 3 X X */ + OPCODE_TXB, /* X 3 X */ + OPCODE_TXD, /* X X */ + OPCODE_TXL, /* 3 2 X */ + OPCODE_TXP, /* X X */ + OPCODE_TXP_NV, /* 3 X */ + OPCODE_TRUNC, /* X */ + OPCODE_UP2H, /* X */ + OPCODE_UP2US, /* X */ + OPCODE_UP4B, /* X */ + OPCODE_UP4UB, /* X */ + OPCODE_X2D, /* X */ + OPCODE_XOR, /* */ + OPCODE_XPD, /* X X */ + MAX_OPCODE +} gl_inst_opcode; + + +/** + * Number of bits for the src/dst register Index field. + * This limits the size of temp/uniform register files. + */ +#define INST_INDEX_BITS 12 + + +/** + * Instruction source register. + */ +struct prog_src_register +{ + GLuint File:4; /**< One of the PROGRAM_* register file values. */ + GLint Index:(INST_INDEX_BITS+1); /**< Extra bit here for sign bit. + * May be negative for relative addressing. + */ + GLuint Swizzle:12; + GLuint RelAddr:1; + + /** Take the component-wise absolute value */ + GLuint Abs:1; + + /** + * Post-Abs negation. + * This will either be NEGATE_NONE or NEGATE_XYZW, except for the SWZ + * instruction which allows per-component negation. + */ + GLuint Negate:4; + + /** + * Is the register two-dimensional. + * Two dimensional registers are of the + * REGISTER[index][index2] format. + * They are used by the geometry shaders where + * the first index is the index within an array + * and the second index is the semantic of the + * array, e.g. gl_PositionIn[index] would become + * INPUT[index][gl_PositionIn] + */ + GLuint HasIndex2:1; + GLuint RelAddr2:1; + GLint Index2:(INST_INDEX_BITS+1); /**< Extra bit here for sign bit. + * May be negative for relative + * addressing. */ +}; + + +/** + * Instruction destination register. + */ +struct prog_dst_register +{ + GLuint File:4; /**< One of the PROGRAM_* register file values */ + GLuint Index:INST_INDEX_BITS; /**< Unsigned, never negative */ + GLuint WriteMask:4; + GLuint RelAddr:1; + + /** + * \name Conditional destination update control. + * + * \since + * NV_fragment_program, NV_fragment_program_option, NV_vertex_program2, + * NV_vertex_program2_option. + */ + /*@{*/ + /** + * Takes one of the 9 possible condition values (EQ, FL, GT, GE, LE, LT, + * NE, TR, or UN). Dest reg is only written to if the matching + * (swizzled) condition code value passes. When a conditional update mask + * is not specified, this will be \c COND_TR. + */ + GLuint CondMask:4; + + /** + * Condition code swizzle value. + */ + GLuint CondSwizzle:12; + + /** + * Selects the condition code register to use for conditional destination + * update masking. In NV_fragmnet_program or NV_vertex_program2 mode, only + * condition code register 0 is available. In NV_vertex_program3 mode, + * condition code registers 0 and 1 are available. + */ + GLuint CondSrc:1; + /*@}*/ +}; + + +/** + * Vertex/fragment program instruction. + */ +struct prog_instruction +{ + gl_inst_opcode Opcode; + struct prog_src_register SrcReg[3]; + struct prog_dst_register DstReg; + + /** + * Indicates that the instruction should update the condition code + * register. + * + * \since + * NV_fragment_program, NV_fragment_program_option, NV_vertex_program2, + * NV_vertex_program2_option. + */ + GLuint CondUpdate:1; + + /** + * If prog_instruction::CondUpdate is \c GL_TRUE, this value selects the + * condition code register that is to be updated. + * + * In GL_NV_fragment_program or GL_NV_vertex_program2 mode, only condition + * code register 0 is available. In GL_NV_vertex_program3 mode, condition + * code registers 0 and 1 are available. + * + * \since + * NV_fragment_program, NV_fragment_program_option, NV_vertex_program2, + * NV_vertex_program2_option. + */ + GLuint CondDst:1; + + /** + * Saturate each value of the vectored result to the range [0,1] or the + * range [-1,1]. \c SSAT mode (i.e., saturation to the range [-1,1]) is + * only available in NV_fragment_program2 mode. + * Value is one of the SATURATE_* tokens. + * + * \since + * NV_fragment_program, NV_fragment_program_option, NV_vertex_program3. + */ + GLuint SaturateMode:2; + + /** + * Per-instruction selectable precision: FLOAT32, FLOAT16, FIXED12. + * + * \since + * NV_fragment_program, NV_fragment_program_option. + */ + GLuint Precision:3; + + /** + * \name Extra fields for TEX, TXB, TXD, TXL, TXP instructions. + */ + /*@{*/ + /** Source texture unit. */ + GLuint TexSrcUnit:5; + + /** Source texture target, one of TEXTURE_{1D,2D,3D,CUBE,RECT}_INDEX */ + GLuint TexSrcTarget:3; + + /** True if tex instruction should do shadow comparison */ + GLuint TexShadow:1; + /*@}*/ + + /** + * For BRA and CAL instructions, the location to jump to. + * For BGNLOOP, points to ENDLOOP (and vice-versa). + * For BRK, points to ENDLOOP + * For IF, points to ELSE or ENDIF. + * For ELSE, points to ENDIF. + */ + GLint BranchTarget; + + /** for debugging purposes */ + const char *Comment; + + /** Arbitrary data. Used for OPCODE_PRINT and some drivers */ + void *Data; + + /** for driver use (try to remove someday) */ + GLint Aux; +}; + + +extern void +_mesa_init_instructions(struct prog_instruction *inst, GLuint count); + +extern struct prog_instruction * +_mesa_alloc_instructions(GLuint numInst); + +extern struct prog_instruction * +_mesa_realloc_instructions(struct prog_instruction *oldInst, + GLuint numOldInst, GLuint numNewInst); + +extern struct prog_instruction * +_mesa_copy_instructions(struct prog_instruction *dest, + const struct prog_instruction *src, GLuint n); + +extern void +_mesa_free_instructions(struct prog_instruction *inst, GLuint count); + +extern GLuint +_mesa_num_inst_src_regs(gl_inst_opcode opcode); + +extern GLuint +_mesa_num_inst_dst_regs(gl_inst_opcode opcode); + +extern GLboolean +_mesa_is_tex_instruction(gl_inst_opcode opcode); + +extern GLboolean +_mesa_check_soa_dependencies(const struct prog_instruction *inst); + +extern const char * +_mesa_opcode_string(gl_inst_opcode opcode); + + +#endif /* PROG_INSTRUCTION_H */ diff --git a/mesalib/src/mesa/program/prog_optimize.c b/mesalib/src/mesa/program/prog_optimize.c index 1fe1c9add..25d9684b1 100644 --- a/mesalib/src/mesa/program/prog_optimize.c +++ b/mesalib/src/mesa/program/prog_optimize.c @@ -1,1366 +1,1366 @@ -/* - * Mesa 3-D graphics library - * Version: 7.5 - * - * Copyright (C) 2009 VMware, Inc. All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * VMWARE BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - - - -#include "main/glheader.h" -#include "main/context.h" -#include "main/macros.h" -#include "program.h" -#include "prog_instruction.h" -#include "prog_optimize.h" -#include "prog_print.h" - - -#define MAX_LOOP_NESTING 50 -/* MAX_PROGRAM_TEMPS is a low number (256), and we want to be able to - * register allocate many temporary values into that small number of - * temps. So allow large temporary indices coming into the register - * allocator. - */ -#define REG_ALLOCATE_MAX_PROGRAM_TEMPS ((1 << INST_INDEX_BITS) - 1) - -static GLboolean dbg = GL_FALSE; - -#define NO_MASK 0xf - -/** - * Returns the mask of channels (bitmask of WRITEMASK_X,Y,Z,W) which - * are read from the given src in this instruction, We also provide - * one optional masks which may mask other components in the dst - * register - */ -static GLuint -get_src_arg_mask(const struct prog_instruction *inst, - GLuint arg, GLuint dst_mask) -{ - GLuint read_mask, channel_mask; - GLuint comp; - - ASSERT(arg < _mesa_num_inst_src_regs(inst->Opcode)); - - /* Form the dst register, find the written channels */ - if (inst->CondUpdate) { - channel_mask = WRITEMASK_XYZW; - } - else { - switch (inst->Opcode) { - case OPCODE_MOV: - case OPCODE_MIN: - case OPCODE_MAX: - case OPCODE_ABS: - case OPCODE_ADD: - case OPCODE_MAD: - case OPCODE_MUL: - case OPCODE_SUB: - case OPCODE_CMP: - case OPCODE_FLR: - case OPCODE_FRC: - case OPCODE_LRP: - case OPCODE_SEQ: - case OPCODE_SGE: - case OPCODE_SGT: - case OPCODE_SLE: - case OPCODE_SLT: - case OPCODE_SNE: - case OPCODE_SSG: - channel_mask = inst->DstReg.WriteMask & dst_mask; - break; - case OPCODE_RCP: - case OPCODE_SIN: - case OPCODE_COS: - case OPCODE_RSQ: - case OPCODE_POW: - case OPCODE_EX2: - case OPCODE_LOG: - channel_mask = WRITEMASK_X; - break; - case OPCODE_DP2: - channel_mask = WRITEMASK_XY; - break; - case OPCODE_DP3: - case OPCODE_XPD: - channel_mask = WRITEMASK_XYZ; - break; - default: - channel_mask = WRITEMASK_XYZW; - break; - } - } - - /* Now, given the src swizzle and the written channels, find which - * components are actually read - */ - read_mask = 0x0; - for (comp = 0; comp < 4; ++comp) { - const GLuint coord = GET_SWZ(inst->SrcReg[arg].Swizzle, comp); - ASSERT(coord < 4); - if (channel_mask & (1 << comp) && coord <= SWIZZLE_W) - read_mask |= 1 << coord; - } - - return read_mask; -} - - -/** - * For a MOV instruction, compute a write mask when src register also has - * a mask - */ -static GLuint -get_dst_mask_for_mov(const struct prog_instruction *mov, GLuint src_mask) -{ - const GLuint mask = mov->DstReg.WriteMask; - GLuint comp; - GLuint updated_mask = 0x0; - - ASSERT(mov->Opcode == OPCODE_MOV); - - for (comp = 0; comp < 4; ++comp) { - GLuint src_comp; - if ((mask & (1 << comp)) == 0) - continue; - src_comp = GET_SWZ(mov->SrcReg[0].Swizzle, comp); - if ((src_mask & (1 << src_comp)) == 0) - continue; - updated_mask |= 1 << comp; - } - - return updated_mask; -} - - -/** - * Ensure that the swizzle is regular. That is, all of the swizzle - * terms are SWIZZLE_X,Y,Z,W and not SWIZZLE_ZERO or SWIZZLE_ONE. - */ -static GLboolean -is_swizzle_regular(GLuint swz) -{ - return GET_SWZ(swz,0) <= SWIZZLE_W && - GET_SWZ(swz,1) <= SWIZZLE_W && - GET_SWZ(swz,2) <= SWIZZLE_W && - GET_SWZ(swz,3) <= SWIZZLE_W; -} - - -/** - * In 'prog' remove instruction[i] if removeFlags[i] == TRUE. - * \return number of instructions removed - */ -static GLuint -remove_instructions(struct gl_program *prog, const GLboolean *removeFlags) -{ - GLint i, removeEnd = 0, removeCount = 0; - GLuint totalRemoved = 0; - - /* go backward */ - for (i = prog->NumInstructions - 1; i >= 0; i--) { - if (removeFlags[i]) { - totalRemoved++; - if (removeCount == 0) { - /* begin a run of instructions to remove */ - removeEnd = i; - removeCount = 1; - } - else { - /* extend the run of instructions to remove */ - removeCount++; - } - } - else { - /* don't remove this instruction, but check if the preceeding - * instructions are to be removed. - */ - if (removeCount > 0) { - GLint removeStart = removeEnd - removeCount + 1; - _mesa_delete_instructions(prog, removeStart, removeCount); - removeStart = removeCount = 0; /* reset removal info */ - } - } - } - /* Finish removing if the first instruction was to be removed. */ - if (removeCount > 0) { - GLint removeStart = removeEnd - removeCount + 1; - _mesa_delete_instructions(prog, removeStart, removeCount); - } - return totalRemoved; -} - - -/** - * Remap register indexes according to map. - * \param prog the program to search/replace - * \param file the type of register file to search/replace - * \param map maps old register indexes to new indexes - */ -static void -replace_regs(struct gl_program *prog, gl_register_file file, const GLint map[]) -{ - GLuint i; - - for (i = 0; i < prog->NumInstructions; i++) { - struct prog_instruction *inst = prog->Instructions + i; - const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode); - GLuint j; - for (j = 0; j < numSrc; j++) { - if (inst->SrcReg[j].File == file) { - GLuint index = inst->SrcReg[j].Index; - ASSERT(map[index] >= 0); - inst->SrcReg[j].Index = map[index]; - } - } - if (inst->DstReg.File == file) { - const GLuint index = inst->DstReg.Index; - ASSERT(map[index] >= 0); - inst->DstReg.Index = map[index]; - } - } -} - - -/** - * Remove dead instructions from the given program. - * This is very primitive for now. Basically look for temp registers - * that are written to but never read. Remove any instructions that - * write to such registers. Be careful with condition code setters. - */ -static GLboolean -_mesa_remove_dead_code_global(struct gl_program *prog) -{ - GLboolean tempRead[REG_ALLOCATE_MAX_PROGRAM_TEMPS][4]; - GLboolean *removeInst; /* per-instruction removal flag */ - GLuint i, rem = 0, comp; - - memset(tempRead, 0, sizeof(tempRead)); - - if (dbg) { - printf("Optimize: Begin dead code removal\n"); - /*_mesa_print_program(prog);*/ - } - - removeInst = (GLboolean *) - calloc(1, prog->NumInstructions * sizeof(GLboolean)); - - /* Determine which temps are read and written */ - for (i = 0; i < prog->NumInstructions; i++) { - const struct prog_instruction *inst = prog->Instructions + i; - const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode); - GLuint j; - - /* check src regs */ - for (j = 0; j < numSrc; j++) { - if (inst->SrcReg[j].File == PROGRAM_TEMPORARY) { - const GLuint index = inst->SrcReg[j].Index; - GLuint read_mask; - ASSERT(index < REG_ALLOCATE_MAX_PROGRAM_TEMPS); - read_mask = get_src_arg_mask(inst, j, NO_MASK); - - if (inst->SrcReg[j].RelAddr) { - if (dbg) - printf("abort remove dead code (indirect temp)\n"); - goto done; - } - - for (comp = 0; comp < 4; comp++) { - const GLuint swz = GET_SWZ(inst->SrcReg[j].Swizzle, comp); - ASSERT(swz < 4); - if ((read_mask & (1 << swz)) == 0) - continue; - if (swz <= SWIZZLE_W) - tempRead[index][swz] = GL_TRUE; - } - } - } - - /* check dst reg */ - if (inst->DstReg.File == PROGRAM_TEMPORARY) { - const GLuint index = inst->DstReg.Index; - ASSERT(index < REG_ALLOCATE_MAX_PROGRAM_TEMPS); - - if (inst->DstReg.RelAddr) { - if (dbg) - printf("abort remove dead code (indirect temp)\n"); - goto done; - } - - if (inst->CondUpdate) { - /* If we're writing to this register and setting condition - * codes we cannot remove the instruction. Prevent removal - * by setting the 'read' flag. - */ - tempRead[index][0] = GL_TRUE; - tempRead[index][1] = GL_TRUE; - tempRead[index][2] = GL_TRUE; - tempRead[index][3] = GL_TRUE; - } - } - } - - /* find instructions that write to dead registers, flag for removal */ - for (i = 0; i < prog->NumInstructions; i++) { - struct prog_instruction *inst = prog->Instructions + i; - const GLuint numDst = _mesa_num_inst_dst_regs(inst->Opcode); - - if (numDst != 0 && inst->DstReg.File == PROGRAM_TEMPORARY) { - GLint chan, index = inst->DstReg.Index; - - for (chan = 0; chan < 4; chan++) { - if (!tempRead[index][chan] && - inst->DstReg.WriteMask & (1 << chan)) { - if (dbg) { - printf("Remove writemask on %u.%c\n", i, - chan == 3 ? 'w' : 'x' + chan); - } - inst->DstReg.WriteMask &= ~(1 << chan); - rem++; - } - } - - if (inst->DstReg.WriteMask == 0) { - /* If we cleared all writes, the instruction can be removed. */ - if (dbg) - printf("Remove instruction %u: \n", i); - removeInst[i] = GL_TRUE; - } - } - } - - /* now remove the instructions which aren't needed */ - rem = remove_instructions(prog, removeInst); - - if (dbg) { - printf("Optimize: End dead code removal.\n"); - printf(" %u channel writes removed\n", rem); - printf(" %u instructions removed\n", rem); - /*_mesa_print_program(prog);*/ - } - -done: - free(removeInst); - return rem != 0; -} - - -enum inst_use -{ - READ, - WRITE, - FLOW, - END -}; - - -/** - * Scan forward in program from 'start' for the next occurances of TEMP[index]. - * We look if an instruction reads the component given by the masks and if they - * are overwritten. - * Return READ, WRITE, FLOW or END to indicate the next usage or an indicator - * that we can't look further. - */ -static enum inst_use -find_next_use(const struct gl_program *prog, - GLuint start, - GLuint index, - GLuint mask) -{ - GLuint i; - - for (i = start; i < prog->NumInstructions; i++) { - const struct prog_instruction *inst = prog->Instructions + i; - switch (inst->Opcode) { - case OPCODE_BGNLOOP: - case OPCODE_BGNSUB: - case OPCODE_BRA: - case OPCODE_CAL: - case OPCODE_CONT: - case OPCODE_IF: - case OPCODE_ELSE: - case OPCODE_ENDIF: - case OPCODE_ENDLOOP: - case OPCODE_ENDSUB: - case OPCODE_RET: - return FLOW; - case OPCODE_END: - return END; - default: - { - const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode); - GLuint j; - for (j = 0; j < numSrc; j++) { - if (inst->SrcReg[j].RelAddr || - (inst->SrcReg[j].File == PROGRAM_TEMPORARY && - inst->SrcReg[j].Index == index && - (get_src_arg_mask(inst,j,NO_MASK) & mask))) - return READ; - } - if (_mesa_num_inst_dst_regs(inst->Opcode) == 1 && - inst->DstReg.File == PROGRAM_TEMPORARY && - inst->DstReg.Index == index) { - mask &= ~inst->DstReg.WriteMask; - if (mask == 0) - return WRITE; - } - } - } - } - return END; -} - - -/** - * Is the given instruction opcode a flow-control opcode? - * XXX maybe move this into prog_instruction.[ch] - */ -static GLboolean -_mesa_is_flow_control_opcode(enum prog_opcode opcode) -{ - switch (opcode) { - case OPCODE_BGNLOOP: - case OPCODE_BGNSUB: - case OPCODE_BRA: - case OPCODE_CAL: - case OPCODE_CONT: - case OPCODE_IF: - case OPCODE_ELSE: - case OPCODE_END: - case OPCODE_ENDIF: - case OPCODE_ENDLOOP: - case OPCODE_ENDSUB: - case OPCODE_RET: - return GL_TRUE; - default: - return GL_FALSE; - } -} - - -/** - * Test if the given instruction is a simple MOV (no conditional updating, - * not relative addressing, no negation/abs, etc). - */ -static GLboolean -can_downward_mov_be_modifed(const struct prog_instruction *mov) -{ - return - mov->Opcode == OPCODE_MOV && - mov->CondUpdate == GL_FALSE && - mov->SrcReg[0].RelAddr == 0 && - mov->SrcReg[0].Negate == 0 && - mov->SrcReg[0].Abs == 0 && - mov->SrcReg[0].HasIndex2 == 0 && - mov->SrcReg[0].RelAddr2 == 0 && - mov->DstReg.RelAddr == 0 && - mov->DstReg.CondMask == COND_TR; -} - - -static GLboolean -can_upward_mov_be_modifed(const struct prog_instruction *mov) -{ - return - can_downward_mov_be_modifed(mov) && - mov->DstReg.File == PROGRAM_TEMPORARY && - mov->SaturateMode == SATURATE_OFF; -} - - -/** - * Try to remove use of extraneous MOV instructions, to free them up for dead - * code removal. - */ -static void -_mesa_remove_extra_move_use(struct gl_program *prog) -{ - GLuint i, j; - - if (dbg) { - printf("Optimize: Begin remove extra move use\n"); - _mesa_print_program(prog); - } - - /* - * Look for sequences such as this: - * MOV tmpX, arg0; - * ... - * FOO tmpY, tmpX, arg1; - * and convert into: - * MOV tmpX, arg0; - * ... - * FOO tmpY, arg0, arg1; - */ - - for (i = 0; i + 1 < prog->NumInstructions; i++) { - const struct prog_instruction *mov = prog->Instructions + i; - GLuint dst_mask, src_mask; - if (can_upward_mov_be_modifed(mov) == GL_FALSE) - continue; - - /* Scanning the code, we maintain the components which are still active in - * these two masks - */ - dst_mask = mov->DstReg.WriteMask; - src_mask = get_src_arg_mask(mov, 0, NO_MASK); - - /* Walk through remaining instructions until the or src reg gets - * rewritten or we get into some flow-control, eliminating the use of - * this MOV. - */ - for (j = i + 1; j < prog->NumInstructions; j++) { - struct prog_instruction *inst2 = prog->Instructions + j; - GLuint arg; - - if (_mesa_is_flow_control_opcode(inst2->Opcode)) - break; - - /* First rewrite this instruction's args if appropriate. */ - for (arg = 0; arg < _mesa_num_inst_src_regs(inst2->Opcode); arg++) { - GLuint comp, read_mask; - - if (inst2->SrcReg[arg].File != mov->DstReg.File || - inst2->SrcReg[arg].Index != mov->DstReg.Index || - inst2->SrcReg[arg].RelAddr || - inst2->SrcReg[arg].Abs) - continue; - read_mask = get_src_arg_mask(inst2, arg, NO_MASK); - - /* Adjust the swizzles of inst2 to point at MOV's source if ALL the - * components read still come from the mov instructions - */ - if (is_swizzle_regular(inst2->SrcReg[arg].Swizzle) && - (read_mask & dst_mask) == read_mask) { - for (comp = 0; comp < 4; comp++) { - const GLuint inst2_swz = - GET_SWZ(inst2->SrcReg[arg].Swizzle, comp); - const GLuint s = GET_SWZ(mov->SrcReg[0].Swizzle, inst2_swz); - inst2->SrcReg[arg].Swizzle &= ~(7 << (3 * comp)); - inst2->SrcReg[arg].Swizzle |= s << (3 * comp); - inst2->SrcReg[arg].Negate ^= (((mov->SrcReg[0].Negate >> - inst2_swz) & 0x1) << comp); - } - inst2->SrcReg[arg].File = mov->SrcReg[0].File; - inst2->SrcReg[arg].Index = mov->SrcReg[0].Index; - } - } - - /* The source of MOV is written. This potentially deactivates some - * components from the src and dst of the MOV instruction - */ - if (inst2->DstReg.File == mov->DstReg.File && - (inst2->DstReg.RelAddr || - inst2->DstReg.Index == mov->DstReg.Index)) { - dst_mask &= ~inst2->DstReg.WriteMask; - src_mask = get_src_arg_mask(mov, 0, dst_mask); - } - - /* Idem when the destination of mov is written */ - if (inst2->DstReg.File == mov->SrcReg[0].File && - (inst2->DstReg.RelAddr || - inst2->DstReg.Index == mov->SrcReg[0].Index)) { - src_mask &= ~inst2->DstReg.WriteMask; - dst_mask &= get_dst_mask_for_mov(mov, src_mask); - } - if (dst_mask == 0) - break; - } - } - - if (dbg) { - printf("Optimize: End remove extra move use.\n"); - /*_mesa_print_program(prog);*/ - } -} - - -/** - * Complements dead_code_global. Try to remove code in block of code by - * carefully monitoring the swizzles. Both functions should be merged into one - * with a proper control flow graph - */ -static GLboolean -_mesa_remove_dead_code_local(struct gl_program *prog) -{ - GLboolean *removeInst; - GLuint i, arg, rem = 0; - - removeInst = (GLboolean *) - calloc(1, prog->NumInstructions * sizeof(GLboolean)); - - for (i = 0; i < prog->NumInstructions; i++) { - const struct prog_instruction *inst = prog->Instructions + i; - const GLuint index = inst->DstReg.Index; - const GLuint mask = inst->DstReg.WriteMask; - enum inst_use use; - - /* We must deactivate the pass as soon as some indirection is used */ - if (inst->DstReg.RelAddr) - goto done; - for (arg = 0; arg < _mesa_num_inst_src_regs(inst->Opcode); arg++) - if (inst->SrcReg[arg].RelAddr) - goto done; - - if (_mesa_is_flow_control_opcode(inst->Opcode) || - _mesa_num_inst_dst_regs(inst->Opcode) == 0 || - inst->DstReg.File != PROGRAM_TEMPORARY || - inst->DstReg.RelAddr) - continue; - - use = find_next_use(prog, i+1, index, mask); - if (use == WRITE || use == END) - removeInst[i] = GL_TRUE; - } - - rem = remove_instructions(prog, removeInst); - -done: - free(removeInst); - return rem != 0; -} - - -/** - * Try to inject the destination of mov as the destination of inst and recompute - * the swizzles operators for the sources of inst if required. Return GL_TRUE - * of the substitution was possible, GL_FALSE otherwise - */ -static GLboolean -_mesa_merge_mov_into_inst(struct prog_instruction *inst, - const struct prog_instruction *mov) -{ - /* Indirection table which associates destination and source components for - * the mov instruction - */ - const GLuint mask = get_src_arg_mask(mov, 0, NO_MASK); - - /* Some components are not written by inst. We cannot remove the mov */ - if (mask != (inst->DstReg.WriteMask & mask)) - return GL_FALSE; - - inst->SaturateMode |= mov->SaturateMode; - - /* Depending on the instruction, we may need to recompute the swizzles. - * Also, some other instructions (like TEX) are not linear. We will only - * consider completely active sources and destinations - */ - switch (inst->Opcode) { - - /* Carstesian instructions: we compute the swizzle */ - case OPCODE_MOV: - case OPCODE_MIN: - case OPCODE_MAX: - case OPCODE_ABS: - case OPCODE_ADD: - case OPCODE_MAD: - case OPCODE_MUL: - case OPCODE_SUB: - { - GLuint dst_to_src_comp[4] = {0,0,0,0}; - GLuint dst_comp, arg; - for (dst_comp = 0; dst_comp < 4; ++dst_comp) { - if (mov->DstReg.WriteMask & (1 << dst_comp)) { - const GLuint src_comp = GET_SWZ(mov->SrcReg[0].Swizzle, dst_comp); - ASSERT(src_comp < 4); - dst_to_src_comp[dst_comp] = src_comp; - } - } - - /* Patch each source of the instruction */ - for (arg = 0; arg < _mesa_num_inst_src_regs(inst->Opcode); arg++) { - const GLuint arg_swz = inst->SrcReg[arg].Swizzle; - inst->SrcReg[arg].Swizzle = 0; - - /* Reset each active component of the swizzle */ - for (dst_comp = 0; dst_comp < 4; ++dst_comp) { - GLuint src_comp, arg_comp; - if ((mov->DstReg.WriteMask & (1 << dst_comp)) == 0) - continue; - src_comp = dst_to_src_comp[dst_comp]; - ASSERT(src_comp < 4); - arg_comp = GET_SWZ(arg_swz, src_comp); - ASSERT(arg_comp < 4); - inst->SrcReg[arg].Swizzle |= arg_comp << (3*dst_comp); - } - } - inst->DstReg = mov->DstReg; - return GL_TRUE; - } - - /* Dot products and scalar instructions: we only change the destination */ - case OPCODE_RCP: - case OPCODE_SIN: - case OPCODE_COS: - case OPCODE_RSQ: - case OPCODE_POW: - case OPCODE_EX2: - case OPCODE_LOG: - case OPCODE_DP2: - case OPCODE_DP3: - case OPCODE_DP4: - inst->DstReg = mov->DstReg; - return GL_TRUE; - - /* All other instructions require fully active components with no swizzle */ - default: - if (mov->SrcReg[0].Swizzle != SWIZZLE_XYZW || - inst->DstReg.WriteMask != WRITEMASK_XYZW) - return GL_FALSE; - inst->DstReg = mov->DstReg; - return GL_TRUE; - } -} - - -/** - * Try to remove extraneous MOV instructions from the given program. - */ -static GLboolean -_mesa_remove_extra_moves(struct gl_program *prog) -{ - GLboolean *removeInst; /* per-instruction removal flag */ - GLuint i, rem = 0, nesting = 0; - - if (dbg) { - printf("Optimize: Begin remove extra moves\n"); - _mesa_print_program(prog); - } - - removeInst = (GLboolean *) - calloc(1, prog->NumInstructions * sizeof(GLboolean)); - - /* - * Look for sequences such as this: - * FOO tmpX, arg0, arg1; - * MOV tmpY, tmpX; - * and convert into: - * FOO tmpY, arg0, arg1; - */ - - for (i = 0; i < prog->NumInstructions; i++) { - const struct prog_instruction *mov = prog->Instructions + i; - - switch (mov->Opcode) { - case OPCODE_BGNLOOP: - case OPCODE_BGNSUB: - case OPCODE_IF: - nesting++; - break; - case OPCODE_ENDLOOP: - case OPCODE_ENDSUB: - case OPCODE_ENDIF: - nesting--; - break; - case OPCODE_MOV: - if (i > 0 && - can_downward_mov_be_modifed(mov) && - mov->SrcReg[0].File == PROGRAM_TEMPORARY && - nesting == 0) - { - - /* see if this MOV can be removed */ - const GLuint id = mov->SrcReg[0].Index; - struct prog_instruction *prevInst; - GLuint prevI; - - /* get pointer to previous instruction */ - prevI = i - 1; - while (prevI > 0 && removeInst[prevI]) - prevI--; - prevInst = prog->Instructions + prevI; - - if (prevInst->DstReg.File == PROGRAM_TEMPORARY && - prevInst->DstReg.Index == id && - prevInst->DstReg.RelAddr == 0 && - prevInst->DstReg.CondSrc == 0 && - prevInst->DstReg.CondMask == COND_TR) { - - const GLuint dst_mask = prevInst->DstReg.WriteMask; - enum inst_use next_use = find_next_use(prog, i+1, id, dst_mask); - - if (next_use == WRITE || next_use == END) { - /* OK, we can safely remove this MOV instruction. - * Transform: - * prevI: FOO tempIndex, x, y; - * i: MOV z, tempIndex; - * Into: - * prevI: FOO z, x, y; - */ - if (_mesa_merge_mov_into_inst(prevInst, mov)) { - removeInst[i] = GL_TRUE; - if (dbg) { - printf("Remove MOV at %u\n", i); - printf("new prev inst %u: ", prevI); - _mesa_print_instruction(prevInst); - } - } - } - } - } - break; - default: - ; /* nothing */ - } - } - - /* now remove the instructions which aren't needed */ - rem = remove_instructions(prog, removeInst); - - free(removeInst); - - if (dbg) { - printf("Optimize: End remove extra moves. %u instructions removed\n", rem); - /*_mesa_print_program(prog);*/ - } - - return rem != 0; -} - - -/** A live register interval */ -struct interval -{ - GLuint Reg; /** The temporary register index */ - GLuint Start, End; /** Start/end instruction numbers */ -}; - - -/** A list of register intervals */ -struct interval_list -{ - GLuint Num; - struct interval Intervals[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; -}; - - -static void -append_interval(struct interval_list *list, const struct interval *inv) -{ - list->Intervals[list->Num++] = *inv; -} - - -/** Insert interval inv into list, sorted by interval end */ -static void -insert_interval_by_end(struct interval_list *list, const struct interval *inv) -{ - /* XXX we could do a binary search insertion here since list is sorted */ - GLint i = list->Num - 1; - while (i >= 0 && list->Intervals[i].End > inv->End) { - list->Intervals[i + 1] = list->Intervals[i]; - i--; - } - list->Intervals[i + 1] = *inv; - list->Num++; - -#ifdef DEBUG - { - GLuint i; - for (i = 0; i + 1 < list->Num; i++) { - ASSERT(list->Intervals[i].End <= list->Intervals[i + 1].End); - } - } -#endif -} - - -/** Remove the given interval from the interval list */ -static void -remove_interval(struct interval_list *list, const struct interval *inv) -{ - /* XXX we could binary search since list is sorted */ - GLuint k; - for (k = 0; k < list->Num; k++) { - if (list->Intervals[k].Reg == inv->Reg) { - /* found, remove it */ - ASSERT(list->Intervals[k].Start == inv->Start); - ASSERT(list->Intervals[k].End == inv->End); - while (k < list->Num - 1) { - list->Intervals[k] = list->Intervals[k + 1]; - k++; - } - list->Num--; - return; - } - } -} - - -/** called by qsort() */ -static int -compare_start(const void *a, const void *b) -{ - const struct interval *ia = (const struct interval *) a; - const struct interval *ib = (const struct interval *) b; - if (ia->Start < ib->Start) - return -1; - else if (ia->Start > ib->Start) - return +1; - else - return 0; -} - - -/** sort the interval list according to interval starts */ -static void -sort_interval_list_by_start(struct interval_list *list) -{ - qsort(list->Intervals, list->Num, sizeof(struct interval), compare_start); -#ifdef DEBUG - { - GLuint i; - for (i = 0; i + 1 < list->Num; i++) { - ASSERT(list->Intervals[i].Start <= list->Intervals[i + 1].Start); - } - } -#endif -} - -struct loop_info -{ - GLuint Start, End; /**< Start, end instructions of loop */ -}; - -/** - * Update the intermediate interval info for register 'index' and - * instruction 'ic'. - */ -static void -update_interval(GLint intBegin[], GLint intEnd[], - struct loop_info *loopStack, GLuint loopStackDepth, - GLuint index, GLuint ic) -{ - int i; - GLuint begin = ic; - GLuint end = ic; - - /* If the register is used in a loop, extend its lifetime through the end - * of the outermost loop that doesn't contain its definition. - */ - for (i = 0; i < loopStackDepth; i++) { - if (intBegin[index] < loopStack[i].Start) { - end = loopStack[i].End; - break; - } - } - - /* Variables that are live at the end of a loop will also be live at the - * beginning, so an instruction inside of a loop should have its live - * interval begin at the start of the outermost loop. - */ - if (loopStackDepth > 0 && ic > loopStack[0].Start && ic < loopStack[0].End) { - begin = loopStack[0].Start; - } - - ASSERT(index < REG_ALLOCATE_MAX_PROGRAM_TEMPS); - if (intBegin[index] == -1) { - ASSERT(intEnd[index] == -1); - intBegin[index] = begin; - intEnd[index] = end; - } - else { - intEnd[index] = end; - } -} - - -/** - * Find first/last instruction that references each temporary register. - */ -GLboolean -_mesa_find_temp_intervals(const struct prog_instruction *instructions, - GLuint numInstructions, - GLint intBegin[REG_ALLOCATE_MAX_PROGRAM_TEMPS], - GLint intEnd[REG_ALLOCATE_MAX_PROGRAM_TEMPS]) -{ - struct loop_info loopStack[MAX_LOOP_NESTING]; - GLuint loopStackDepth = 0; - GLuint i; - - for (i = 0; i < REG_ALLOCATE_MAX_PROGRAM_TEMPS; i++){ - intBegin[i] = intEnd[i] = -1; - } - - /* Scan instructions looking for temporary registers */ - for (i = 0; i < numInstructions; i++) { - const struct prog_instruction *inst = instructions + i; - if (inst->Opcode == OPCODE_BGNLOOP) { - loopStack[loopStackDepth].Start = i; - loopStack[loopStackDepth].End = inst->BranchTarget; - loopStackDepth++; - } - else if (inst->Opcode == OPCODE_ENDLOOP) { - loopStackDepth--; - } - else if (inst->Opcode == OPCODE_CAL) { - return GL_FALSE; - } - else { - const GLuint numSrc = 3;/*_mesa_num_inst_src_regs(inst->Opcode);*/ - GLuint j; - for (j = 0; j < numSrc; j++) { - if (inst->SrcReg[j].File == PROGRAM_TEMPORARY) { - const GLuint index = inst->SrcReg[j].Index; - if (inst->SrcReg[j].RelAddr) - return GL_FALSE; - update_interval(intBegin, intEnd, loopStack, loopStackDepth, - index, i); - } - } - if (inst->DstReg.File == PROGRAM_TEMPORARY) { - const GLuint index = inst->DstReg.Index; - if (inst->DstReg.RelAddr) - return GL_FALSE; - update_interval(intBegin, intEnd, loopStack, loopStackDepth, - index, i); - } - } - } - - return GL_TRUE; -} - - -/** - * Find the live intervals for each temporary register in the program. - * For register R, the interval [A,B] indicates that R is referenced - * from instruction A through instruction B. - * Special consideration is needed for loops and subroutines. - * \return GL_TRUE if success, GL_FALSE if we cannot proceed for some reason - */ -static GLboolean -find_live_intervals(struct gl_program *prog, - struct interval_list *liveIntervals) -{ - GLint intBegin[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; - GLint intEnd[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; - GLuint i; - - /* - * Note: we'll return GL_FALSE below if we find relative indexing - * into the TEMP register file. We can't handle that yet. - * We also give up on subroutines for now. - */ - - if (dbg) { - printf("Optimize: Begin find intervals\n"); - } - - /* build intermediate arrays */ - if (!_mesa_find_temp_intervals(prog->Instructions, prog->NumInstructions, - intBegin, intEnd)) - return GL_FALSE; - - /* Build live intervals list from intermediate arrays */ - liveIntervals->Num = 0; - for (i = 0; i < REG_ALLOCATE_MAX_PROGRAM_TEMPS; i++) { - if (intBegin[i] >= 0) { - struct interval inv; - inv.Reg = i; - inv.Start = intBegin[i]; - inv.End = intEnd[i]; - append_interval(liveIntervals, &inv); - } - } - - /* Sort the list according to interval starts */ - sort_interval_list_by_start(liveIntervals); - - if (dbg) { - /* print interval info */ - for (i = 0; i < liveIntervals->Num; i++) { - const struct interval *inv = liveIntervals->Intervals + i; - printf("Reg[%d] live [%d, %d]:", - inv->Reg, inv->Start, inv->End); - if (1) { - GLuint j; - for (j = 0; j < inv->Start; j++) - printf(" "); - for (j = inv->Start; j <= inv->End; j++) - printf("x"); - } - printf("\n"); - } - } - - return GL_TRUE; -} - - -/** Scan the array of used register flags to find free entry */ -static GLint -alloc_register(GLboolean usedRegs[REG_ALLOCATE_MAX_PROGRAM_TEMPS]) -{ - GLuint k; - for (k = 0; k < REG_ALLOCATE_MAX_PROGRAM_TEMPS; k++) { - if (!usedRegs[k]) { - usedRegs[k] = GL_TRUE; - return k; - } - } - return -1; -} - - -/** - * This function implements "Linear Scan Register Allocation" to reduce - * the number of temporary registers used by the program. - * - * We compute the "live interval" for all temporary registers then - * examine the overlap of the intervals to allocate new registers. - * Basically, if two intervals do not overlap, they can use the same register. - */ -static void -_mesa_reallocate_registers(struct gl_program *prog) -{ - struct interval_list liveIntervals; - GLint registerMap[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; - GLboolean usedRegs[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; - GLuint i; - GLint maxTemp = -1; - - if (dbg) { - printf("Optimize: Begin live-interval register reallocation\n"); - _mesa_print_program(prog); - } - - for (i = 0; i < REG_ALLOCATE_MAX_PROGRAM_TEMPS; i++){ - registerMap[i] = -1; - usedRegs[i] = GL_FALSE; - } - - if (!find_live_intervals(prog, &liveIntervals)) { - if (dbg) - printf("Aborting register reallocation\n"); - return; - } - - { - struct interval_list activeIntervals; - activeIntervals.Num = 0; - - /* loop over live intervals, allocating a new register for each */ - for (i = 0; i < liveIntervals.Num; i++) { - const struct interval *live = liveIntervals.Intervals + i; - - if (dbg) - printf("Consider register %u\n", live->Reg); - - /* Expire old intervals. Intervals which have ended with respect - * to the live interval can have their remapped registers freed. - */ - { - GLint j; - for (j = 0; j < (GLint) activeIntervals.Num; j++) { - const struct interval *inv = activeIntervals.Intervals + j; - if (inv->End >= live->Start) { - /* Stop now. Since the activeInterval list is sorted - * we know we don't have to go further. - */ - break; - } - else { - /* Interval 'inv' has expired */ - const GLint regNew = registerMap[inv->Reg]; - ASSERT(regNew >= 0); - - if (dbg) - printf(" expire interval for reg %u\n", inv->Reg); - - /* remove interval j from active list */ - remove_interval(&activeIntervals, inv); - j--; /* counter-act j++ in for-loop above */ - - /* return register regNew to the free pool */ - if (dbg) - printf(" free reg %d\n", regNew); - ASSERT(usedRegs[regNew] == GL_TRUE); - usedRegs[regNew] = GL_FALSE; - } - } - } - - /* find a free register for this live interval */ - { - const GLint k = alloc_register(usedRegs); - if (k < 0) { - /* out of registers, give up */ - return; - } - registerMap[live->Reg] = k; - maxTemp = MAX2(maxTemp, k); - if (dbg) - printf(" remap register %u -> %d\n", live->Reg, k); - } - - /* Insert this live interval into the active list which is sorted - * by increasing end points. - */ - insert_interval_by_end(&activeIntervals, live); - } - } - - if (maxTemp + 1 < (GLint) liveIntervals.Num) { - /* OK, we've reduced the number of registers needed. - * Scan the program and replace all the old temporary register - * indexes with the new indexes. - */ - replace_regs(prog, PROGRAM_TEMPORARY, registerMap); - - prog->NumTemporaries = maxTemp + 1; - } - - if (dbg) { - printf("Optimize: End live-interval register reallocation\n"); - printf("Num temp regs before: %u after: %u\n", - liveIntervals.Num, maxTemp + 1); - _mesa_print_program(prog); - } -} - - -#if 0 -static void -print_it(struct gl_context *ctx, struct gl_program *program, const char *txt) { - fprintf(stderr, "%s (%u inst):\n", txt, program->NumInstructions); - _mesa_print_program(program); - _mesa_print_program_parameters(ctx, program); - fprintf(stderr, "\n\n"); -} -#endif - -/** - * This pass replaces CMP T0, T1 T2 T0 with MOV T0, T2 when the CMP - * instruction is the first instruction to write to register T0. The are - * several lowering passes done in GLSL IR (e.g. branches and - * relative addressing) that create a large number of conditional assignments - * that ir_to_mesa converts to CMP instructions like the one mentioned above. - * - * Here is why this conversion is safe: - * CMP T0, T1 T2 T0 can be expanded to: - * if (T1 < 0.0) - * MOV T0, T2; - * else - * MOV T0, T0; - * - * If (T1 < 0.0) evaluates to true then our replacement MOV T0, T2 is the same - * as the original program. If (T1 < 0.0) evaluates to false, executing - * MOV T0, T0 will store a garbage value in T0 since T0 is uninitialized. - * Therefore, it doesn't matter that we are replacing MOV T0, T0 with MOV T0, T2 - * because any instruction that was going to read from T0 after this was going - * to read a garbage value anyway. - */ -static void -_mesa_simplify_cmp(struct gl_program * program) -{ - GLuint tempWrites[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; - GLuint outputWrites[MAX_PROGRAM_OUTPUTS]; - GLuint i; - - if (dbg) { - printf("Optimize: Begin reads without writes\n"); - _mesa_print_program(program); - } - - for (i = 0; i < REG_ALLOCATE_MAX_PROGRAM_TEMPS; i++) { - tempWrites[i] = 0; - } - - for (i = 0; i < MAX_PROGRAM_OUTPUTS; i++) { - outputWrites[i] = 0; - } - - for (i = 0; i < program->NumInstructions; i++) { - struct prog_instruction *inst = program->Instructions + i; - GLuint prevWriteMask; - - /* Give up if we encounter relative addressing or flow control. */ - if (_mesa_is_flow_control_opcode(inst->Opcode) || inst->DstReg.RelAddr) { - return; - } - - if (inst->DstReg.File == PROGRAM_OUTPUT) { - assert(inst->DstReg.Index < MAX_PROGRAM_OUTPUTS); - prevWriteMask = outputWrites[inst->DstReg.Index]; - outputWrites[inst->DstReg.Index] |= inst->DstReg.WriteMask; - } else if (inst->DstReg.File == PROGRAM_TEMPORARY) { - assert(inst->DstReg.Index < REG_ALLOCATE_MAX_PROGRAM_TEMPS); - prevWriteMask = tempWrites[inst->DstReg.Index]; - tempWrites[inst->DstReg.Index] |= inst->DstReg.WriteMask; - } else { - /* No other register type can be a destination register. */ - continue; - } - - /* For a CMP to be considered a conditional write, the destination - * register and source register two must be the same. */ - if (inst->Opcode == OPCODE_CMP - && !(inst->DstReg.WriteMask & prevWriteMask) - && inst->SrcReg[2].File == inst->DstReg.File - && inst->SrcReg[2].Index == inst->DstReg.Index - && inst->DstReg.WriteMask == get_src_arg_mask(inst, 2, NO_MASK)) { - - inst->Opcode = OPCODE_MOV; - inst->SrcReg[0] = inst->SrcReg[1]; - - /* Unused operands are expected to have the file set to - * PROGRAM_UNDEFINED. This is how _mesa_init_instructions initializes - * all of the sources. - */ - inst->SrcReg[1].File = PROGRAM_UNDEFINED; - inst->SrcReg[1].Swizzle = SWIZZLE_NOOP; - inst->SrcReg[2].File = PROGRAM_UNDEFINED; - inst->SrcReg[2].Swizzle = SWIZZLE_NOOP; - } - } - if (dbg) { - printf("Optimize: End reads without writes\n"); - _mesa_print_program(program); - } -} - -/** - * Apply optimizations to the given program to eliminate unnecessary - * instructions, temp regs, etc. - */ -void -_mesa_optimize_program(struct gl_context *ctx, struct gl_program *program) -{ - GLboolean any_change; - - _mesa_simplify_cmp(program); - /* Stop when no modifications were output */ - do { - any_change = GL_FALSE; - _mesa_remove_extra_move_use(program); - if (_mesa_remove_dead_code_global(program)) - any_change = GL_TRUE; - if (_mesa_remove_extra_moves(program)) - any_change = GL_TRUE; - if (_mesa_remove_dead_code_local(program)) - any_change = GL_TRUE; - - any_change = _mesa_constant_fold(program) || any_change; - _mesa_reallocate_registers(program); - } while (any_change); -} - +/* + * Mesa 3-D graphics library + * Version: 7.5 + * + * Copyright (C) 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * VMWARE BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + + + +#include "main/glheader.h" +#include "main/context.h" +#include "main/macros.h" +#include "program.h" +#include "prog_instruction.h" +#include "prog_optimize.h" +#include "prog_print.h" + + +#define MAX_LOOP_NESTING 50 +/* MAX_PROGRAM_TEMPS is a low number (256), and we want to be able to + * register allocate many temporary values into that small number of + * temps. So allow large temporary indices coming into the register + * allocator. + */ +#define REG_ALLOCATE_MAX_PROGRAM_TEMPS ((1 << INST_INDEX_BITS) - 1) + +static GLboolean dbg = GL_FALSE; + +#define NO_MASK 0xf + +/** + * Returns the mask of channels (bitmask of WRITEMASK_X,Y,Z,W) which + * are read from the given src in this instruction, We also provide + * one optional masks which may mask other components in the dst + * register + */ +static GLuint +get_src_arg_mask(const struct prog_instruction *inst, + GLuint arg, GLuint dst_mask) +{ + GLuint read_mask, channel_mask; + GLuint comp; + + ASSERT(arg < _mesa_num_inst_src_regs(inst->Opcode)); + + /* Form the dst register, find the written channels */ + if (inst->CondUpdate) { + channel_mask = WRITEMASK_XYZW; + } + else { + switch (inst->Opcode) { + case OPCODE_MOV: + case OPCODE_MIN: + case OPCODE_MAX: + case OPCODE_ABS: + case OPCODE_ADD: + case OPCODE_MAD: + case OPCODE_MUL: + case OPCODE_SUB: + case OPCODE_CMP: + case OPCODE_FLR: + case OPCODE_FRC: + case OPCODE_LRP: + case OPCODE_SEQ: + case OPCODE_SGE: + case OPCODE_SGT: + case OPCODE_SLE: + case OPCODE_SLT: + case OPCODE_SNE: + case OPCODE_SSG: + channel_mask = inst->DstReg.WriteMask & dst_mask; + break; + case OPCODE_RCP: + case OPCODE_SIN: + case OPCODE_COS: + case OPCODE_RSQ: + case OPCODE_POW: + case OPCODE_EX2: + case OPCODE_LOG: + channel_mask = WRITEMASK_X; + break; + case OPCODE_DP2: + channel_mask = WRITEMASK_XY; + break; + case OPCODE_DP3: + case OPCODE_XPD: + channel_mask = WRITEMASK_XYZ; + break; + default: + channel_mask = WRITEMASK_XYZW; + break; + } + } + + /* Now, given the src swizzle and the written channels, find which + * components are actually read + */ + read_mask = 0x0; + for (comp = 0; comp < 4; ++comp) { + const GLuint coord = GET_SWZ(inst->SrcReg[arg].Swizzle, comp); + ASSERT(coord < 4); + if (channel_mask & (1 << comp) && coord <= SWIZZLE_W) + read_mask |= 1 << coord; + } + + return read_mask; +} + + +/** + * For a MOV instruction, compute a write mask when src register also has + * a mask + */ +static GLuint +get_dst_mask_for_mov(const struct prog_instruction *mov, GLuint src_mask) +{ + const GLuint mask = mov->DstReg.WriteMask; + GLuint comp; + GLuint updated_mask = 0x0; + + ASSERT(mov->Opcode == OPCODE_MOV); + + for (comp = 0; comp < 4; ++comp) { + GLuint src_comp; + if ((mask & (1 << comp)) == 0) + continue; + src_comp = GET_SWZ(mov->SrcReg[0].Swizzle, comp); + if ((src_mask & (1 << src_comp)) == 0) + continue; + updated_mask |= 1 << comp; + } + + return updated_mask; +} + + +/** + * Ensure that the swizzle is regular. That is, all of the swizzle + * terms are SWIZZLE_X,Y,Z,W and not SWIZZLE_ZERO or SWIZZLE_ONE. + */ +static GLboolean +is_swizzle_regular(GLuint swz) +{ + return GET_SWZ(swz,0) <= SWIZZLE_W && + GET_SWZ(swz,1) <= SWIZZLE_W && + GET_SWZ(swz,2) <= SWIZZLE_W && + GET_SWZ(swz,3) <= SWIZZLE_W; +} + + +/** + * In 'prog' remove instruction[i] if removeFlags[i] == TRUE. + * \return number of instructions removed + */ +static GLuint +remove_instructions(struct gl_program *prog, const GLboolean *removeFlags) +{ + GLint i, removeEnd = 0, removeCount = 0; + GLuint totalRemoved = 0; + + /* go backward */ + for (i = prog->NumInstructions - 1; i >= 0; i--) { + if (removeFlags[i]) { + totalRemoved++; + if (removeCount == 0) { + /* begin a run of instructions to remove */ + removeEnd = i; + removeCount = 1; + } + else { + /* extend the run of instructions to remove */ + removeCount++; + } + } + else { + /* don't remove this instruction, but check if the preceeding + * instructions are to be removed. + */ + if (removeCount > 0) { + GLint removeStart = removeEnd - removeCount + 1; + _mesa_delete_instructions(prog, removeStart, removeCount); + removeStart = removeCount = 0; /* reset removal info */ + } + } + } + /* Finish removing if the first instruction was to be removed. */ + if (removeCount > 0) { + GLint removeStart = removeEnd - removeCount + 1; + _mesa_delete_instructions(prog, removeStart, removeCount); + } + return totalRemoved; +} + + +/** + * Remap register indexes according to map. + * \param prog the program to search/replace + * \param file the type of register file to search/replace + * \param map maps old register indexes to new indexes + */ +static void +replace_regs(struct gl_program *prog, gl_register_file file, const GLint map[]) +{ + GLuint i; + + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = prog->Instructions + i; + const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode); + GLuint j; + for (j = 0; j < numSrc; j++) { + if (inst->SrcReg[j].File == file) { + GLuint index = inst->SrcReg[j].Index; + ASSERT(map[index] >= 0); + inst->SrcReg[j].Index = map[index]; + } + } + if (inst->DstReg.File == file) { + const GLuint index = inst->DstReg.Index; + ASSERT(map[index] >= 0); + inst->DstReg.Index = map[index]; + } + } +} + + +/** + * Remove dead instructions from the given program. + * This is very primitive for now. Basically look for temp registers + * that are written to but never read. Remove any instructions that + * write to such registers. Be careful with condition code setters. + */ +static GLboolean +_mesa_remove_dead_code_global(struct gl_program *prog) +{ + GLboolean tempRead[REG_ALLOCATE_MAX_PROGRAM_TEMPS][4]; + GLboolean *removeInst; /* per-instruction removal flag */ + GLuint i, rem = 0, comp; + + memset(tempRead, 0, sizeof(tempRead)); + + if (dbg) { + printf("Optimize: Begin dead code removal\n"); + /*_mesa_print_program(prog);*/ + } + + removeInst = (GLboolean *) + calloc(1, prog->NumInstructions * sizeof(GLboolean)); + + /* Determine which temps are read and written */ + for (i = 0; i < prog->NumInstructions; i++) { + const struct prog_instruction *inst = prog->Instructions + i; + const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode); + GLuint j; + + /* check src regs */ + for (j = 0; j < numSrc; j++) { + if (inst->SrcReg[j].File == PROGRAM_TEMPORARY) { + const GLuint index = inst->SrcReg[j].Index; + GLuint read_mask; + ASSERT(index < REG_ALLOCATE_MAX_PROGRAM_TEMPS); + read_mask = get_src_arg_mask(inst, j, NO_MASK); + + if (inst->SrcReg[j].RelAddr) { + if (dbg) + printf("abort remove dead code (indirect temp)\n"); + goto done; + } + + for (comp = 0; comp < 4; comp++) { + const GLuint swz = GET_SWZ(inst->SrcReg[j].Swizzle, comp); + ASSERT(swz < 4); + if ((read_mask & (1 << swz)) == 0) + continue; + if (swz <= SWIZZLE_W) + tempRead[index][swz] = GL_TRUE; + } + } + } + + /* check dst reg */ + if (inst->DstReg.File == PROGRAM_TEMPORARY) { + const GLuint index = inst->DstReg.Index; + ASSERT(index < REG_ALLOCATE_MAX_PROGRAM_TEMPS); + + if (inst->DstReg.RelAddr) { + if (dbg) + printf("abort remove dead code (indirect temp)\n"); + goto done; + } + + if (inst->CondUpdate) { + /* If we're writing to this register and setting condition + * codes we cannot remove the instruction. Prevent removal + * by setting the 'read' flag. + */ + tempRead[index][0] = GL_TRUE; + tempRead[index][1] = GL_TRUE; + tempRead[index][2] = GL_TRUE; + tempRead[index][3] = GL_TRUE; + } + } + } + + /* find instructions that write to dead registers, flag for removal */ + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = prog->Instructions + i; + const GLuint numDst = _mesa_num_inst_dst_regs(inst->Opcode); + + if (numDst != 0 && inst->DstReg.File == PROGRAM_TEMPORARY) { + GLint chan, index = inst->DstReg.Index; + + for (chan = 0; chan < 4; chan++) { + if (!tempRead[index][chan] && + inst->DstReg.WriteMask & (1 << chan)) { + if (dbg) { + printf("Remove writemask on %u.%c\n", i, + chan == 3 ? 'w' : 'x' + chan); + } + inst->DstReg.WriteMask &= ~(1 << chan); + rem++; + } + } + + if (inst->DstReg.WriteMask == 0) { + /* If we cleared all writes, the instruction can be removed. */ + if (dbg) + printf("Remove instruction %u: \n", i); + removeInst[i] = GL_TRUE; + } + } + } + + /* now remove the instructions which aren't needed */ + rem = remove_instructions(prog, removeInst); + + if (dbg) { + printf("Optimize: End dead code removal.\n"); + printf(" %u channel writes removed\n", rem); + printf(" %u instructions removed\n", rem); + /*_mesa_print_program(prog);*/ + } + +done: + free(removeInst); + return rem != 0; +} + + +enum inst_use +{ + READ, + WRITE, + FLOW, + END +}; + + +/** + * Scan forward in program from 'start' for the next occurances of TEMP[index]. + * We look if an instruction reads the component given by the masks and if they + * are overwritten. + * Return READ, WRITE, FLOW or END to indicate the next usage or an indicator + * that we can't look further. + */ +static enum inst_use +find_next_use(const struct gl_program *prog, + GLuint start, + GLuint index, + GLuint mask) +{ + GLuint i; + + for (i = start; i < prog->NumInstructions; i++) { + const struct prog_instruction *inst = prog->Instructions + i; + switch (inst->Opcode) { + case OPCODE_BGNLOOP: + case OPCODE_BGNSUB: + case OPCODE_BRA: + case OPCODE_CAL: + case OPCODE_CONT: + case OPCODE_IF: + case OPCODE_ELSE: + case OPCODE_ENDIF: + case OPCODE_ENDLOOP: + case OPCODE_ENDSUB: + case OPCODE_RET: + return FLOW; + case OPCODE_END: + return END; + default: + { + const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode); + GLuint j; + for (j = 0; j < numSrc; j++) { + if (inst->SrcReg[j].RelAddr || + (inst->SrcReg[j].File == PROGRAM_TEMPORARY && + inst->SrcReg[j].Index == index && + (get_src_arg_mask(inst,j,NO_MASK) & mask))) + return READ; + } + if (_mesa_num_inst_dst_regs(inst->Opcode) == 1 && + inst->DstReg.File == PROGRAM_TEMPORARY && + inst->DstReg.Index == index) { + mask &= ~inst->DstReg.WriteMask; + if (mask == 0) + return WRITE; + } + } + } + } + return END; +} + + +/** + * Is the given instruction opcode a flow-control opcode? + * XXX maybe move this into prog_instruction.[ch] + */ +static GLboolean +_mesa_is_flow_control_opcode(enum prog_opcode opcode) +{ + switch (opcode) { + case OPCODE_BGNLOOP: + case OPCODE_BGNSUB: + case OPCODE_BRA: + case OPCODE_CAL: + case OPCODE_CONT: + case OPCODE_IF: + case OPCODE_ELSE: + case OPCODE_END: + case OPCODE_ENDIF: + case OPCODE_ENDLOOP: + case OPCODE_ENDSUB: + case OPCODE_RET: + return GL_TRUE; + default: + return GL_FALSE; + } +} + + +/** + * Test if the given instruction is a simple MOV (no conditional updating, + * not relative addressing, no negation/abs, etc). + */ +static GLboolean +can_downward_mov_be_modifed(const struct prog_instruction *mov) +{ + return + mov->Opcode == OPCODE_MOV && + mov->CondUpdate == GL_FALSE && + mov->SrcReg[0].RelAddr == 0 && + mov->SrcReg[0].Negate == 0 && + mov->SrcReg[0].Abs == 0 && + mov->SrcReg[0].HasIndex2 == 0 && + mov->SrcReg[0].RelAddr2 == 0 && + mov->DstReg.RelAddr == 0 && + mov->DstReg.CondMask == COND_TR; +} + + +static GLboolean +can_upward_mov_be_modifed(const struct prog_instruction *mov) +{ + return + can_downward_mov_be_modifed(mov) && + mov->DstReg.File == PROGRAM_TEMPORARY && + mov->SaturateMode == SATURATE_OFF; +} + + +/** + * Try to remove use of extraneous MOV instructions, to free them up for dead + * code removal. + */ +static void +_mesa_remove_extra_move_use(struct gl_program *prog) +{ + GLuint i, j; + + if (dbg) { + printf("Optimize: Begin remove extra move use\n"); + _mesa_print_program(prog); + } + + /* + * Look for sequences such as this: + * MOV tmpX, arg0; + * ... + * FOO tmpY, tmpX, arg1; + * and convert into: + * MOV tmpX, arg0; + * ... + * FOO tmpY, arg0, arg1; + */ + + for (i = 0; i + 1 < prog->NumInstructions; i++) { + const struct prog_instruction *mov = prog->Instructions + i; + GLuint dst_mask, src_mask; + if (can_upward_mov_be_modifed(mov) == GL_FALSE) + continue; + + /* Scanning the code, we maintain the components which are still active in + * these two masks + */ + dst_mask = mov->DstReg.WriteMask; + src_mask = get_src_arg_mask(mov, 0, NO_MASK); + + /* Walk through remaining instructions until the or src reg gets + * rewritten or we get into some flow-control, eliminating the use of + * this MOV. + */ + for (j = i + 1; j < prog->NumInstructions; j++) { + struct prog_instruction *inst2 = prog->Instructions + j; + GLuint arg; + + if (_mesa_is_flow_control_opcode(inst2->Opcode)) + break; + + /* First rewrite this instruction's args if appropriate. */ + for (arg = 0; arg < _mesa_num_inst_src_regs(inst2->Opcode); arg++) { + GLuint comp, read_mask; + + if (inst2->SrcReg[arg].File != mov->DstReg.File || + inst2->SrcReg[arg].Index != mov->DstReg.Index || + inst2->SrcReg[arg].RelAddr || + inst2->SrcReg[arg].Abs) + continue; + read_mask = get_src_arg_mask(inst2, arg, NO_MASK); + + /* Adjust the swizzles of inst2 to point at MOV's source if ALL the + * components read still come from the mov instructions + */ + if (is_swizzle_regular(inst2->SrcReg[arg].Swizzle) && + (read_mask & dst_mask) == read_mask) { + for (comp = 0; comp < 4; comp++) { + const GLuint inst2_swz = + GET_SWZ(inst2->SrcReg[arg].Swizzle, comp); + const GLuint s = GET_SWZ(mov->SrcReg[0].Swizzle, inst2_swz); + inst2->SrcReg[arg].Swizzle &= ~(7 << (3 * comp)); + inst2->SrcReg[arg].Swizzle |= s << (3 * comp); + inst2->SrcReg[arg].Negate ^= (((mov->SrcReg[0].Negate >> + inst2_swz) & 0x1) << comp); + } + inst2->SrcReg[arg].File = mov->SrcReg[0].File; + inst2->SrcReg[arg].Index = mov->SrcReg[0].Index; + } + } + + /* The source of MOV is written. This potentially deactivates some + * components from the src and dst of the MOV instruction + */ + if (inst2->DstReg.File == mov->DstReg.File && + (inst2->DstReg.RelAddr || + inst2->DstReg.Index == mov->DstReg.Index)) { + dst_mask &= ~inst2->DstReg.WriteMask; + src_mask = get_src_arg_mask(mov, 0, dst_mask); + } + + /* Idem when the destination of mov is written */ + if (inst2->DstReg.File == mov->SrcReg[0].File && + (inst2->DstReg.RelAddr || + inst2->DstReg.Index == mov->SrcReg[0].Index)) { + src_mask &= ~inst2->DstReg.WriteMask; + dst_mask &= get_dst_mask_for_mov(mov, src_mask); + } + if (dst_mask == 0) + break; + } + } + + if (dbg) { + printf("Optimize: End remove extra move use.\n"); + /*_mesa_print_program(prog);*/ + } +} + + +/** + * Complements dead_code_global. Try to remove code in block of code by + * carefully monitoring the swizzles. Both functions should be merged into one + * with a proper control flow graph + */ +static GLboolean +_mesa_remove_dead_code_local(struct gl_program *prog) +{ + GLboolean *removeInst; + GLuint i, arg, rem = 0; + + removeInst = (GLboolean *) + calloc(1, prog->NumInstructions * sizeof(GLboolean)); + + for (i = 0; i < prog->NumInstructions; i++) { + const struct prog_instruction *inst = prog->Instructions + i; + const GLuint index = inst->DstReg.Index; + const GLuint mask = inst->DstReg.WriteMask; + enum inst_use use; + + /* We must deactivate the pass as soon as some indirection is used */ + if (inst->DstReg.RelAddr) + goto done; + for (arg = 0; arg < _mesa_num_inst_src_regs(inst->Opcode); arg++) + if (inst->SrcReg[arg].RelAddr) + goto done; + + if (_mesa_is_flow_control_opcode(inst->Opcode) || + _mesa_num_inst_dst_regs(inst->Opcode) == 0 || + inst->DstReg.File != PROGRAM_TEMPORARY || + inst->DstReg.RelAddr) + continue; + + use = find_next_use(prog, i+1, index, mask); + if (use == WRITE || use == END) + removeInst[i] = GL_TRUE; + } + + rem = remove_instructions(prog, removeInst); + +done: + free(removeInst); + return rem != 0; +} + + +/** + * Try to inject the destination of mov as the destination of inst and recompute + * the swizzles operators for the sources of inst if required. Return GL_TRUE + * of the substitution was possible, GL_FALSE otherwise + */ +static GLboolean +_mesa_merge_mov_into_inst(struct prog_instruction *inst, + const struct prog_instruction *mov) +{ + /* Indirection table which associates destination and source components for + * the mov instruction + */ + const GLuint mask = get_src_arg_mask(mov, 0, NO_MASK); + + /* Some components are not written by inst. We cannot remove the mov */ + if (mask != (inst->DstReg.WriteMask & mask)) + return GL_FALSE; + + inst->SaturateMode |= mov->SaturateMode; + + /* Depending on the instruction, we may need to recompute the swizzles. + * Also, some other instructions (like TEX) are not linear. We will only + * consider completely active sources and destinations + */ + switch (inst->Opcode) { + + /* Carstesian instructions: we compute the swizzle */ + case OPCODE_MOV: + case OPCODE_MIN: + case OPCODE_MAX: + case OPCODE_ABS: + case OPCODE_ADD: + case OPCODE_MAD: + case OPCODE_MUL: + case OPCODE_SUB: + { + GLuint dst_to_src_comp[4] = {0,0,0,0}; + GLuint dst_comp, arg; + for (dst_comp = 0; dst_comp < 4; ++dst_comp) { + if (mov->DstReg.WriteMask & (1 << dst_comp)) { + const GLuint src_comp = GET_SWZ(mov->SrcReg[0].Swizzle, dst_comp); + ASSERT(src_comp < 4); + dst_to_src_comp[dst_comp] = src_comp; + } + } + + /* Patch each source of the instruction */ + for (arg = 0; arg < _mesa_num_inst_src_regs(inst->Opcode); arg++) { + const GLuint arg_swz = inst->SrcReg[arg].Swizzle; + inst->SrcReg[arg].Swizzle = 0; + + /* Reset each active component of the swizzle */ + for (dst_comp = 0; dst_comp < 4; ++dst_comp) { + GLuint src_comp, arg_comp; + if ((mov->DstReg.WriteMask & (1 << dst_comp)) == 0) + continue; + src_comp = dst_to_src_comp[dst_comp]; + ASSERT(src_comp < 4); + arg_comp = GET_SWZ(arg_swz, src_comp); + ASSERT(arg_comp < 4); + inst->SrcReg[arg].Swizzle |= arg_comp << (3*dst_comp); + } + } + inst->DstReg = mov->DstReg; + return GL_TRUE; + } + + /* Dot products and scalar instructions: we only change the destination */ + case OPCODE_RCP: + case OPCODE_SIN: + case OPCODE_COS: + case OPCODE_RSQ: + case OPCODE_POW: + case OPCODE_EX2: + case OPCODE_LOG: + case OPCODE_DP2: + case OPCODE_DP3: + case OPCODE_DP4: + inst->DstReg = mov->DstReg; + return GL_TRUE; + + /* All other instructions require fully active components with no swizzle */ + default: + if (mov->SrcReg[0].Swizzle != SWIZZLE_XYZW || + inst->DstReg.WriteMask != WRITEMASK_XYZW) + return GL_FALSE; + inst->DstReg = mov->DstReg; + return GL_TRUE; + } +} + + +/** + * Try to remove extraneous MOV instructions from the given program. + */ +static GLboolean +_mesa_remove_extra_moves(struct gl_program *prog) +{ + GLboolean *removeInst; /* per-instruction removal flag */ + GLuint i, rem = 0, nesting = 0; + + if (dbg) { + printf("Optimize: Begin remove extra moves\n"); + _mesa_print_program(prog); + } + + removeInst = (GLboolean *) + calloc(1, prog->NumInstructions * sizeof(GLboolean)); + + /* + * Look for sequences such as this: + * FOO tmpX, arg0, arg1; + * MOV tmpY, tmpX; + * and convert into: + * FOO tmpY, arg0, arg1; + */ + + for (i = 0; i < prog->NumInstructions; i++) { + const struct prog_instruction *mov = prog->Instructions + i; + + switch (mov->Opcode) { + case OPCODE_BGNLOOP: + case OPCODE_BGNSUB: + case OPCODE_IF: + nesting++; + break; + case OPCODE_ENDLOOP: + case OPCODE_ENDSUB: + case OPCODE_ENDIF: + nesting--; + break; + case OPCODE_MOV: + if (i > 0 && + can_downward_mov_be_modifed(mov) && + mov->SrcReg[0].File == PROGRAM_TEMPORARY && + nesting == 0) + { + + /* see if this MOV can be removed */ + const GLuint id = mov->SrcReg[0].Index; + struct prog_instruction *prevInst; + GLuint prevI; + + /* get pointer to previous instruction */ + prevI = i - 1; + while (prevI > 0 && removeInst[prevI]) + prevI--; + prevInst = prog->Instructions + prevI; + + if (prevInst->DstReg.File == PROGRAM_TEMPORARY && + prevInst->DstReg.Index == id && + prevInst->DstReg.RelAddr == 0 && + prevInst->DstReg.CondSrc == 0 && + prevInst->DstReg.CondMask == COND_TR) { + + const GLuint dst_mask = prevInst->DstReg.WriteMask; + enum inst_use next_use = find_next_use(prog, i+1, id, dst_mask); + + if (next_use == WRITE || next_use == END) { + /* OK, we can safely remove this MOV instruction. + * Transform: + * prevI: FOO tempIndex, x, y; + * i: MOV z, tempIndex; + * Into: + * prevI: FOO z, x, y; + */ + if (_mesa_merge_mov_into_inst(prevInst, mov)) { + removeInst[i] = GL_TRUE; + if (dbg) { + printf("Remove MOV at %u\n", i); + printf("new prev inst %u: ", prevI); + _mesa_print_instruction(prevInst); + } + } + } + } + } + break; + default: + ; /* nothing */ + } + } + + /* now remove the instructions which aren't needed */ + rem = remove_instructions(prog, removeInst); + + free(removeInst); + + if (dbg) { + printf("Optimize: End remove extra moves. %u instructions removed\n", rem); + /*_mesa_print_program(prog);*/ + } + + return rem != 0; +} + + +/** A live register interval */ +struct interval +{ + GLuint Reg; /** The temporary register index */ + GLuint Start, End; /** Start/end instruction numbers */ +}; + + +/** A list of register intervals */ +struct interval_list +{ + GLuint Num; + struct interval Intervals[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; +}; + + +static void +append_interval(struct interval_list *list, const struct interval *inv) +{ + list->Intervals[list->Num++] = *inv; +} + + +/** Insert interval inv into list, sorted by interval end */ +static void +insert_interval_by_end(struct interval_list *list, const struct interval *inv) +{ + /* XXX we could do a binary search insertion here since list is sorted */ + GLint i = list->Num - 1; + while (i >= 0 && list->Intervals[i].End > inv->End) { + list->Intervals[i + 1] = list->Intervals[i]; + i--; + } + list->Intervals[i + 1] = *inv; + list->Num++; + +#ifdef DEBUG + { + GLuint i; + for (i = 0; i + 1 < list->Num; i++) { + ASSERT(list->Intervals[i].End <= list->Intervals[i + 1].End); + } + } +#endif +} + + +/** Remove the given interval from the interval list */ +static void +remove_interval(struct interval_list *list, const struct interval *inv) +{ + /* XXX we could binary search since list is sorted */ + GLuint k; + for (k = 0; k < list->Num; k++) { + if (list->Intervals[k].Reg == inv->Reg) { + /* found, remove it */ + ASSERT(list->Intervals[k].Start == inv->Start); + ASSERT(list->Intervals[k].End == inv->End); + while (k < list->Num - 1) { + list->Intervals[k] = list->Intervals[k + 1]; + k++; + } + list->Num--; + return; + } + } +} + + +/** called by qsort() */ +static int +compare_start(const void *a, const void *b) +{ + const struct interval *ia = (const struct interval *) a; + const struct interval *ib = (const struct interval *) b; + if (ia->Start < ib->Start) + return -1; + else if (ia->Start > ib->Start) + return +1; + else + return 0; +} + + +/** sort the interval list according to interval starts */ +static void +sort_interval_list_by_start(struct interval_list *list) +{ + qsort(list->Intervals, list->Num, sizeof(struct interval), compare_start); +#ifdef DEBUG + { + GLuint i; + for (i = 0; i + 1 < list->Num; i++) { + ASSERT(list->Intervals[i].Start <= list->Intervals[i + 1].Start); + } + } +#endif +} + +struct loop_info +{ + GLuint Start, End; /**< Start, end instructions of loop */ +}; + +/** + * Update the intermediate interval info for register 'index' and + * instruction 'ic'. + */ +static void +update_interval(GLint intBegin[], GLint intEnd[], + struct loop_info *loopStack, GLuint loopStackDepth, + GLuint index, GLuint ic) +{ + int i; + GLuint begin = ic; + GLuint end = ic; + + /* If the register is used in a loop, extend its lifetime through the end + * of the outermost loop that doesn't contain its definition. + */ + for (i = 0; i < loopStackDepth; i++) { + if (intBegin[index] < loopStack[i].Start) { + end = loopStack[i].End; + break; + } + } + + /* Variables that are live at the end of a loop will also be live at the + * beginning, so an instruction inside of a loop should have its live + * interval begin at the start of the outermost loop. + */ + if (loopStackDepth > 0 && ic > loopStack[0].Start && ic < loopStack[0].End) { + begin = loopStack[0].Start; + } + + ASSERT(index < REG_ALLOCATE_MAX_PROGRAM_TEMPS); + if (intBegin[index] == -1) { + ASSERT(intEnd[index] == -1); + intBegin[index] = begin; + intEnd[index] = end; + } + else { + intEnd[index] = end; + } +} + + +/** + * Find first/last instruction that references each temporary register. + */ +GLboolean +_mesa_find_temp_intervals(const struct prog_instruction *instructions, + GLuint numInstructions, + GLint intBegin[REG_ALLOCATE_MAX_PROGRAM_TEMPS], + GLint intEnd[REG_ALLOCATE_MAX_PROGRAM_TEMPS]) +{ + struct loop_info loopStack[MAX_LOOP_NESTING]; + GLuint loopStackDepth = 0; + GLuint i; + + for (i = 0; i < REG_ALLOCATE_MAX_PROGRAM_TEMPS; i++){ + intBegin[i] = intEnd[i] = -1; + } + + /* Scan instructions looking for temporary registers */ + for (i = 0; i < numInstructions; i++) { + const struct prog_instruction *inst = instructions + i; + if (inst->Opcode == OPCODE_BGNLOOP) { + loopStack[loopStackDepth].Start = i; + loopStack[loopStackDepth].End = inst->BranchTarget; + loopStackDepth++; + } + else if (inst->Opcode == OPCODE_ENDLOOP) { + loopStackDepth--; + } + else if (inst->Opcode == OPCODE_CAL) { + return GL_FALSE; + } + else { + const GLuint numSrc = 3;/*_mesa_num_inst_src_regs(inst->Opcode);*/ + GLuint j; + for (j = 0; j < numSrc; j++) { + if (inst->SrcReg[j].File == PROGRAM_TEMPORARY) { + const GLuint index = inst->SrcReg[j].Index; + if (inst->SrcReg[j].RelAddr) + return GL_FALSE; + update_interval(intBegin, intEnd, loopStack, loopStackDepth, + index, i); + } + } + if (inst->DstReg.File == PROGRAM_TEMPORARY) { + const GLuint index = inst->DstReg.Index; + if (inst->DstReg.RelAddr) + return GL_FALSE; + update_interval(intBegin, intEnd, loopStack, loopStackDepth, + index, i); + } + } + } + + return GL_TRUE; +} + + +/** + * Find the live intervals for each temporary register in the program. + * For register R, the interval [A,B] indicates that R is referenced + * from instruction A through instruction B. + * Special consideration is needed for loops and subroutines. + * \return GL_TRUE if success, GL_FALSE if we cannot proceed for some reason + */ +static GLboolean +find_live_intervals(struct gl_program *prog, + struct interval_list *liveIntervals) +{ + GLint intBegin[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; + GLint intEnd[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; + GLuint i; + + /* + * Note: we'll return GL_FALSE below if we find relative indexing + * into the TEMP register file. We can't handle that yet. + * We also give up on subroutines for now. + */ + + if (dbg) { + printf("Optimize: Begin find intervals\n"); + } + + /* build intermediate arrays */ + if (!_mesa_find_temp_intervals(prog->Instructions, prog->NumInstructions, + intBegin, intEnd)) + return GL_FALSE; + + /* Build live intervals list from intermediate arrays */ + liveIntervals->Num = 0; + for (i = 0; i < REG_ALLOCATE_MAX_PROGRAM_TEMPS; i++) { + if (intBegin[i] >= 0) { + struct interval inv; + inv.Reg = i; + inv.Start = intBegin[i]; + inv.End = intEnd[i]; + append_interval(liveIntervals, &inv); + } + } + + /* Sort the list according to interval starts */ + sort_interval_list_by_start(liveIntervals); + + if (dbg) { + /* print interval info */ + for (i = 0; i < liveIntervals->Num; i++) { + const struct interval *inv = liveIntervals->Intervals + i; + printf("Reg[%d] live [%d, %d]:", + inv->Reg, inv->Start, inv->End); + if (1) { + GLuint j; + for (j = 0; j < inv->Start; j++) + printf(" "); + for (j = inv->Start; j <= inv->End; j++) + printf("x"); + } + printf("\n"); + } + } + + return GL_TRUE; +} + + +/** Scan the array of used register flags to find free entry */ +static GLint +alloc_register(GLboolean usedRegs[REG_ALLOCATE_MAX_PROGRAM_TEMPS]) +{ + GLuint k; + for (k = 0; k < REG_ALLOCATE_MAX_PROGRAM_TEMPS; k++) { + if (!usedRegs[k]) { + usedRegs[k] = GL_TRUE; + return k; + } + } + return -1; +} + + +/** + * This function implements "Linear Scan Register Allocation" to reduce + * the number of temporary registers used by the program. + * + * We compute the "live interval" for all temporary registers then + * examine the overlap of the intervals to allocate new registers. + * Basically, if two intervals do not overlap, they can use the same register. + */ +static void +_mesa_reallocate_registers(struct gl_program *prog) +{ + struct interval_list liveIntervals; + GLint registerMap[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; + GLboolean usedRegs[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; + GLuint i; + GLint maxTemp = -1; + + if (dbg) { + printf("Optimize: Begin live-interval register reallocation\n"); + _mesa_print_program(prog); + } + + for (i = 0; i < REG_ALLOCATE_MAX_PROGRAM_TEMPS; i++){ + registerMap[i] = -1; + usedRegs[i] = GL_FALSE; + } + + if (!find_live_intervals(prog, &liveIntervals)) { + if (dbg) + printf("Aborting register reallocation\n"); + return; + } + + { + struct interval_list activeIntervals; + activeIntervals.Num = 0; + + /* loop over live intervals, allocating a new register for each */ + for (i = 0; i < liveIntervals.Num; i++) { + const struct interval *live = liveIntervals.Intervals + i; + + if (dbg) + printf("Consider register %u\n", live->Reg); + + /* Expire old intervals. Intervals which have ended with respect + * to the live interval can have their remapped registers freed. + */ + { + GLint j; + for (j = 0; j < (GLint) activeIntervals.Num; j++) { + const struct interval *inv = activeIntervals.Intervals + j; + if (inv->End >= live->Start) { + /* Stop now. Since the activeInterval list is sorted + * we know we don't have to go further. + */ + break; + } + else { + /* Interval 'inv' has expired */ + const GLint regNew = registerMap[inv->Reg]; + ASSERT(regNew >= 0); + + if (dbg) + printf(" expire interval for reg %u\n", inv->Reg); + + /* remove interval j from active list */ + remove_interval(&activeIntervals, inv); + j--; /* counter-act j++ in for-loop above */ + + /* return register regNew to the free pool */ + if (dbg) + printf(" free reg %d\n", regNew); + ASSERT(usedRegs[regNew] == GL_TRUE); + usedRegs[regNew] = GL_FALSE; + } + } + } + + /* find a free register for this live interval */ + { + const GLint k = alloc_register(usedRegs); + if (k < 0) { + /* out of registers, give up */ + return; + } + registerMap[live->Reg] = k; + maxTemp = MAX2(maxTemp, k); + if (dbg) + printf(" remap register %u -> %d\n", live->Reg, k); + } + + /* Insert this live interval into the active list which is sorted + * by increasing end points. + */ + insert_interval_by_end(&activeIntervals, live); + } + } + + if (maxTemp + 1 < (GLint) liveIntervals.Num) { + /* OK, we've reduced the number of registers needed. + * Scan the program and replace all the old temporary register + * indexes with the new indexes. + */ + replace_regs(prog, PROGRAM_TEMPORARY, registerMap); + + prog->NumTemporaries = maxTemp + 1; + } + + if (dbg) { + printf("Optimize: End live-interval register reallocation\n"); + printf("Num temp regs before: %u after: %u\n", + liveIntervals.Num, maxTemp + 1); + _mesa_print_program(prog); + } +} + + +#if 0 +static void +print_it(struct gl_context *ctx, struct gl_program *program, const char *txt) { + fprintf(stderr, "%s (%u inst):\n", txt, program->NumInstructions); + _mesa_print_program(program); + _mesa_print_program_parameters(ctx, program); + fprintf(stderr, "\n\n"); +} +#endif + +/** + * This pass replaces CMP T0, T1 T2 T0 with MOV T0, T2 when the CMP + * instruction is the first instruction to write to register T0. The are + * several lowering passes done in GLSL IR (e.g. branches and + * relative addressing) that create a large number of conditional assignments + * that ir_to_mesa converts to CMP instructions like the one mentioned above. + * + * Here is why this conversion is safe: + * CMP T0, T1 T2 T0 can be expanded to: + * if (T1 < 0.0) + * MOV T0, T2; + * else + * MOV T0, T0; + * + * If (T1 < 0.0) evaluates to true then our replacement MOV T0, T2 is the same + * as the original program. If (T1 < 0.0) evaluates to false, executing + * MOV T0, T0 will store a garbage value in T0 since T0 is uninitialized. + * Therefore, it doesn't matter that we are replacing MOV T0, T0 with MOV T0, T2 + * because any instruction that was going to read from T0 after this was going + * to read a garbage value anyway. + */ +static void +_mesa_simplify_cmp(struct gl_program * program) +{ + GLuint tempWrites[REG_ALLOCATE_MAX_PROGRAM_TEMPS]; + GLuint outputWrites[MAX_PROGRAM_OUTPUTS]; + GLuint i; + + if (dbg) { + printf("Optimize: Begin reads without writes\n"); + _mesa_print_program(program); + } + + for (i = 0; i < REG_ALLOCATE_MAX_PROGRAM_TEMPS; i++) { + tempWrites[i] = 0; + } + + for (i = 0; i < MAX_PROGRAM_OUTPUTS; i++) { + outputWrites[i] = 0; + } + + for (i = 0; i < program->NumInstructions; i++) { + struct prog_instruction *inst = program->Instructions + i; + GLuint prevWriteMask; + + /* Give up if we encounter relative addressing or flow control. */ + if (_mesa_is_flow_control_opcode(inst->Opcode) || inst->DstReg.RelAddr) { + return; + } + + if (inst->DstReg.File == PROGRAM_OUTPUT) { + assert(inst->DstReg.Index < MAX_PROGRAM_OUTPUTS); + prevWriteMask = outputWrites[inst->DstReg.Index]; + outputWrites[inst->DstReg.Index] |= inst->DstReg.WriteMask; + } else if (inst->DstReg.File == PROGRAM_TEMPORARY) { + assert(inst->DstReg.Index < REG_ALLOCATE_MAX_PROGRAM_TEMPS); + prevWriteMask = tempWrites[inst->DstReg.Index]; + tempWrites[inst->DstReg.Index] |= inst->DstReg.WriteMask; + } else { + /* No other register type can be a destination register. */ + continue; + } + + /* For a CMP to be considered a conditional write, the destination + * register and source register two must be the same. */ + if (inst->Opcode == OPCODE_CMP + && !(inst->DstReg.WriteMask & prevWriteMask) + && inst->SrcReg[2].File == inst->DstReg.File + && inst->SrcReg[2].Index == inst->DstReg.Index + && inst->DstReg.WriteMask == get_src_arg_mask(inst, 2, NO_MASK)) { + + inst->Opcode = OPCODE_MOV; + inst->SrcReg[0] = inst->SrcReg[1]; + + /* Unused operands are expected to have the file set to + * PROGRAM_UNDEFINED. This is how _mesa_init_instructions initializes + * all of the sources. + */ + inst->SrcReg[1].File = PROGRAM_UNDEFINED; + inst->SrcReg[1].Swizzle = SWIZZLE_NOOP; + inst->SrcReg[2].File = PROGRAM_UNDEFINED; + inst->SrcReg[2].Swizzle = SWIZZLE_NOOP; + } + } + if (dbg) { + printf("Optimize: End reads without writes\n"); + _mesa_print_program(program); + } +} + +/** + * Apply optimizations to the given program to eliminate unnecessary + * instructions, temp regs, etc. + */ +void +_mesa_optimize_program(struct gl_context *ctx, struct gl_program *program) +{ + GLboolean any_change; + + _mesa_simplify_cmp(program); + /* Stop when no modifications were output */ + do { + any_change = GL_FALSE; + _mesa_remove_extra_move_use(program); + if (_mesa_remove_dead_code_global(program)) + any_change = GL_TRUE; + if (_mesa_remove_extra_moves(program)) + any_change = GL_TRUE; + if (_mesa_remove_dead_code_local(program)) + any_change = GL_TRUE; + + any_change = _mesa_constant_fold(program) || any_change; + _mesa_reallocate_registers(program); + } while (any_change); +} + diff --git a/mesalib/src/mesa/program/prog_print.c b/mesalib/src/mesa/program/prog_print.c index 2cfec13ec..70412b1fa 100644 --- a/mesalib/src/mesa/program/prog_print.c +++ b/mesalib/src/mesa/program/prog_print.c @@ -1,1096 +1,1096 @@ -/* - * Mesa 3-D graphics library - * Version: 7.3 - * - * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. - * Copyright (C) 2009 VMware, Inc. All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - -/** - * \file prog_print.c - * Print vertex/fragment programs - for debugging. - * \author Brian Paul - */ - -#include "main/glheader.h" -#include "main/context.h" -#include "main/imports.h" -#include "prog_instruction.h" -#include "prog_parameter.h" -#include "prog_print.h" -#include "prog_statevars.h" - - - -/** - * Return string name for given program/register file. - */ -const char * -_mesa_register_file_name(gl_register_file f) -{ - switch (f) { - case PROGRAM_TEMPORARY: - return "TEMP"; - case PROGRAM_LOCAL_PARAM: - return "LOCAL"; - case PROGRAM_ENV_PARAM: - return "ENV"; - case PROGRAM_STATE_VAR: - return "STATE"; - case PROGRAM_INPUT: - return "INPUT"; - case PROGRAM_OUTPUT: - return "OUTPUT"; - case PROGRAM_NAMED_PARAM: - return "NAMED"; - case PROGRAM_CONSTANT: - return "CONST"; - case PROGRAM_UNIFORM: - return "UNIFORM"; - case PROGRAM_VARYING: - return "VARYING"; - case PROGRAM_WRITE_ONLY: - return "WRITE_ONLY"; - case PROGRAM_ADDRESS: - return "ADDR"; - case PROGRAM_SAMPLER: - return "SAMPLER"; - case PROGRAM_SYSTEM_VALUE: - return "SYSVAL"; - case PROGRAM_UNDEFINED: - return "UNDEFINED"; - default: - { - static char s[20]; - _mesa_snprintf(s, sizeof(s), "FILE%u", f); - return s; - } - } -} - - -/** - * Return ARB_v/f_prog-style input attrib string. - */ -static const char * -arb_input_attrib_string(GLint index, GLenum progType) -{ - /* - * These strings should match the VERT_ATTRIB_x and FRAG_ATTRIB_x tokens. - */ - const char *vertAttribs[] = { - "vertex.position", - "vertex.weight", - "vertex.normal", - "vertex.color.primary", - "vertex.color.secondary", - "vertex.fogcoord", - "vertex.(six)", - "vertex.(seven)", - "vertex.texcoord[0]", - "vertex.texcoord[1]", - "vertex.texcoord[2]", - "vertex.texcoord[3]", - "vertex.texcoord[4]", - "vertex.texcoord[5]", - "vertex.texcoord[6]", - "vertex.texcoord[7]", - "vertex.attrib[0]", - "vertex.attrib[1]", - "vertex.attrib[2]", - "vertex.attrib[3]", - "vertex.attrib[4]", - "vertex.attrib[5]", - "vertex.attrib[6]", - "vertex.attrib[7]", - "vertex.attrib[8]", - "vertex.attrib[9]", - "vertex.attrib[10]", - "vertex.attrib[11]", - "vertex.attrib[12]", - "vertex.attrib[13]", - "vertex.attrib[14]", - "vertex.attrib[15]" - }; - const char *fragAttribs[] = { - "fragment.position", - "fragment.color.primary", - "fragment.color.secondary", - "fragment.fogcoord", - "fragment.texcoord[0]", - "fragment.texcoord[1]", - "fragment.texcoord[2]", - "fragment.texcoord[3]", - "fragment.texcoord[4]", - "fragment.texcoord[5]", - "fragment.texcoord[6]", - "fragment.texcoord[7]", - "fragment.varying[0]", - "fragment.varying[1]", - "fragment.varying[2]", - "fragment.varying[3]", - "fragment.varying[4]", - "fragment.varying[5]", - "fragment.varying[6]", - "fragment.varying[7]" - }; - - /* sanity checks */ - assert(strcmp(vertAttribs[VERT_ATTRIB_TEX0], "vertex.texcoord[0]") == 0); - assert(strcmp(vertAttribs[VERT_ATTRIB_GENERIC15], "vertex.attrib[15]") == 0); - - if (progType == GL_VERTEX_PROGRAM_ARB) { - assert(index < sizeof(vertAttribs) / sizeof(vertAttribs[0])); - return vertAttribs[index]; - } - else { - assert(index < sizeof(fragAttribs) / sizeof(fragAttribs[0])); - return fragAttribs[index]; - } -} - - -/** - * Print a vertex program's InputsRead field in human-readable format. - * For debugging. - */ -void -_mesa_print_vp_inputs(GLbitfield inputs) -{ - printf("VP Inputs 0x%x: \n", inputs); - while (inputs) { - GLint attr = _mesa_ffs(inputs) - 1; - const char *name = arb_input_attrib_string(attr, - GL_VERTEX_PROGRAM_ARB); - printf(" %d: %s\n", attr, name); - inputs &= ~(1 << attr); - } -} - - -/** - * Print a fragment program's InputsRead field in human-readable format. - * For debugging. - */ -void -_mesa_print_fp_inputs(GLbitfield inputs) -{ - printf("FP Inputs 0x%x: \n", inputs); - while (inputs) { - GLint attr = _mesa_ffs(inputs) - 1; - const char *name = arb_input_attrib_string(attr, - GL_FRAGMENT_PROGRAM_ARB); - printf(" %d: %s\n", attr, name); - inputs &= ~(1 << attr); - } -} - - - -/** - * Return ARB_v/f_prog-style output attrib string. - */ -static const char * -arb_output_attrib_string(GLint index, GLenum progType) -{ - /* - * These strings should match the VERT_RESULT_x and FRAG_RESULT_x tokens. - */ - const char *vertResults[] = { - "result.position", - "result.color.primary", - "result.color.secondary", - "result.fogcoord", - "result.texcoord[0]", - "result.texcoord[1]", - "result.texcoord[2]", - "result.texcoord[3]", - "result.texcoord[4]", - "result.texcoord[5]", - "result.texcoord[6]", - "result.texcoord[7]", - "result.varying[0]", - "result.varying[1]", - "result.varying[2]", - "result.varying[3]", - "result.varying[4]", - "result.varying[5]", - "result.varying[6]", - "result.varying[7]" - }; - const char *fragResults[] = { - "result.color", - "result.color(half)", - "result.depth", - "result.color[0]", - "result.color[1]", - "result.color[2]", - "result.color[3]" - }; - - if (progType == GL_VERTEX_PROGRAM_ARB) { - assert(index < sizeof(vertResults) / sizeof(vertResults[0])); - return vertResults[index]; - } - else { - assert(index < sizeof(fragResults) / sizeof(fragResults[0])); - return fragResults[index]; - } -} - - -/** - * Return string representation of the given register. - * Note that some types of registers (like PROGRAM_UNIFORM) aren't defined - * by the ARB/NV program languages so we've taken some liberties here. - * \param f the register file (PROGRAM_INPUT, PROGRAM_TEMPORARY, etc) - * \param index number of the register in the register file - * \param mode the output format/mode/style - * \param prog pointer to containing program - */ -static const char * -reg_string(gl_register_file f, GLint index, gl_prog_print_mode mode, - GLboolean relAddr, const struct gl_program *prog, - GLboolean hasIndex2, GLboolean relAddr2, GLint index2) -{ - static char str[100]; - const char *addr = relAddr ? "ADDR+" : ""; - - str[0] = 0; - - switch (mode) { - case PROG_PRINT_DEBUG: - sprintf(str, "%s[%s%d]", - _mesa_register_file_name(f), addr, index); - if (hasIndex2) { - int offset = strlen(str); - const char *addr2 = relAddr2 ? "ADDR+" : ""; - sprintf(str+offset, "[%s%d]", addr2, index2); - } - break; - - case PROG_PRINT_ARB: - switch (f) { - case PROGRAM_INPUT: - sprintf(str, "%s", arb_input_attrib_string(index, prog->Target)); - break; - case PROGRAM_OUTPUT: - sprintf(str, "%s", arb_output_attrib_string(index, prog->Target)); - break; - case PROGRAM_TEMPORARY: - sprintf(str, "temp%d", index); - break; - case PROGRAM_ENV_PARAM: - sprintf(str, "program.env[%s%d]", addr, index); - break; - case PROGRAM_LOCAL_PARAM: - sprintf(str, "program.local[%s%d]", addr, index); - break; - case PROGRAM_VARYING: /* extension */ - sprintf(str, "varying[%s%d]", addr, index); - break; - case PROGRAM_CONSTANT: /* extension */ - sprintf(str, "constant[%s%d]", addr, index); - break; - case PROGRAM_UNIFORM: /* extension */ - sprintf(str, "uniform[%s%d]", addr, index); - break; - case PROGRAM_SYSTEM_VALUE: - sprintf(str, "sysvalue[%s%d]", addr, index); - break; - case PROGRAM_STATE_VAR: - { - struct gl_program_parameter *param - = prog->Parameters->Parameters + index; - char *state = _mesa_program_state_string(param->StateIndexes); - sprintf(str, "%s", state); - free(state); - } - break; - case PROGRAM_ADDRESS: - sprintf(str, "A%d", index); - break; - default: - _mesa_problem(NULL, "bad file in reg_string()"); - } - break; - - case PROG_PRINT_NV: - switch (f) { - case PROGRAM_INPUT: - if (prog->Target == GL_VERTEX_PROGRAM_ARB) - sprintf(str, "v[%d]", index); - else - sprintf(str, "f[%d]", index); - break; - case PROGRAM_OUTPUT: - sprintf(str, "o[%d]", index); - break; - case PROGRAM_TEMPORARY: - sprintf(str, "R%d", index); - break; - case PROGRAM_ENV_PARAM: - sprintf(str, "c[%d]", index); - break; - case PROGRAM_VARYING: /* extension */ - sprintf(str, "varying[%s%d]", addr, index); - break; - case PROGRAM_UNIFORM: /* extension */ - sprintf(str, "uniform[%s%d]", addr, index); - break; - case PROGRAM_CONSTANT: /* extension */ - sprintf(str, "constant[%s%d]", addr, index); - break; - case PROGRAM_STATE_VAR: /* extension */ - sprintf(str, "state[%s%d]", addr, index); - break; - default: - _mesa_problem(NULL, "bad file in reg_string()"); - } - break; - - default: - _mesa_problem(NULL, "bad mode in reg_string()"); - } - - return str; -} - - -/** - * Return a string representation of the given swizzle word. - * If extended is true, use extended (comma-separated) format. - * \param swizzle the swizzle field - * \param negateBase 4-bit negation vector - * \param extended if true, also allow 0, 1 values - */ -const char * -_mesa_swizzle_string(GLuint swizzle, GLuint negateMask, GLboolean extended) -{ - static const char swz[] = "xyzw01!?"; /* See SWIZZLE_x definitions */ - static char s[20]; - GLuint i = 0; - - if (!extended && swizzle == SWIZZLE_NOOP && negateMask == 0) - return ""; /* no swizzle/negation */ - - if (!extended) - s[i++] = '.'; - - if (negateMask & NEGATE_X) - s[i++] = '-'; - s[i++] = swz[GET_SWZ(swizzle, 0)]; - - if (extended) { - s[i++] = ','; - } - - if (negateMask & NEGATE_Y) - s[i++] = '-'; - s[i++] = swz[GET_SWZ(swizzle, 1)]; - - if (extended) { - s[i++] = ','; - } - - if (negateMask & NEGATE_Z) - s[i++] = '-'; - s[i++] = swz[GET_SWZ(swizzle, 2)]; - - if (extended) { - s[i++] = ','; - } - - if (negateMask & NEGATE_W) - s[i++] = '-'; - s[i++] = swz[GET_SWZ(swizzle, 3)]; - - s[i] = 0; - return s; -} - - -void -_mesa_print_swizzle(GLuint swizzle) -{ - if (swizzle == SWIZZLE_XYZW) { - printf(".xyzw\n"); - } - else { - const char *s = _mesa_swizzle_string(swizzle, 0, 0); - printf("%s\n", s); - } -} - - -const char * -_mesa_writemask_string(GLuint writeMask) -{ - static char s[10]; - GLuint i = 0; - - if (writeMask == WRITEMASK_XYZW) - return ""; - - s[i++] = '.'; - if (writeMask & WRITEMASK_X) - s[i++] = 'x'; - if (writeMask & WRITEMASK_Y) - s[i++] = 'y'; - if (writeMask & WRITEMASK_Z) - s[i++] = 'z'; - if (writeMask & WRITEMASK_W) - s[i++] = 'w'; - - s[i] = 0; - return s; -} - - -const char * -_mesa_condcode_string(GLuint condcode) -{ - switch (condcode) { - case COND_GT: return "GT"; - case COND_EQ: return "EQ"; - case COND_LT: return "LT"; - case COND_UN: return "UN"; - case COND_GE: return "GE"; - case COND_LE: return "LE"; - case COND_NE: return "NE"; - case COND_TR: return "TR"; - case COND_FL: return "FL"; - default: return "cond???"; - } -} - - -static void -fprint_dst_reg(FILE * f, - const struct prog_dst_register *dstReg, - gl_prog_print_mode mode, - const struct gl_program *prog) -{ - fprintf(f, "%s%s", - reg_string((gl_register_file) dstReg->File, - dstReg->Index, mode, dstReg->RelAddr, prog, - GL_FALSE, GL_FALSE, 0), - _mesa_writemask_string(dstReg->WriteMask)); - - if (dstReg->CondMask != COND_TR) { - fprintf(f, " (%s.%s)", - _mesa_condcode_string(dstReg->CondMask), - _mesa_swizzle_string(dstReg->CondSwizzle, - GL_FALSE, GL_FALSE)); - } - -#if 0 - fprintf(f, "%s[%d]%s", - _mesa_register_file_name((gl_register_file) dstReg->File), - dstReg->Index, - _mesa_writemask_string(dstReg->WriteMask)); -#endif -} - - -static void -fprint_src_reg(FILE *f, - const struct prog_src_register *srcReg, - gl_prog_print_mode mode, - const struct gl_program *prog) -{ - const char *abs = srcReg->Abs ? "|" : ""; - - fprintf(f, "%s%s%s%s", - abs, - reg_string((gl_register_file) srcReg->File, - srcReg->Index, mode, srcReg->RelAddr, prog, - srcReg->HasIndex2, srcReg->RelAddr2, srcReg->Index2), - _mesa_swizzle_string(srcReg->Swizzle, - srcReg->Negate, GL_FALSE), - abs); -#if 0 - fprintf(f, "%s[%d]%s", - _mesa_register_file_name((gl_register_file) srcReg->File), - srcReg->Index, - _mesa_swizzle_string(srcReg->Swizzle, - srcReg->Negate, GL_FALSE)); -#endif -} - - -static void -fprint_comment(FILE *f, const struct prog_instruction *inst) -{ - if (inst->Comment) - fprintf(f, "; # %s\n", inst->Comment); - else - fprintf(f, ";\n"); -} - - -void -_mesa_fprint_alu_instruction(FILE *f, - const struct prog_instruction *inst, - const char *opcode_string, GLuint numRegs, - gl_prog_print_mode mode, - const struct gl_program *prog) -{ - GLuint j; - - fprintf(f, "%s", opcode_string); - if (inst->CondUpdate) - fprintf(f, ".C"); - - /* frag prog only */ - if (inst->SaturateMode == SATURATE_ZERO_ONE) - fprintf(f, "_SAT"); - - fprintf(f, " "); - if (inst->DstReg.File != PROGRAM_UNDEFINED) { - fprint_dst_reg(f, &inst->DstReg, mode, prog); - } - else { - fprintf(f, " ???"); - } - - if (numRegs > 0) - fprintf(f, ", "); - - for (j = 0; j < numRegs; j++) { - fprint_src_reg(f, inst->SrcReg + j, mode, prog); - if (j + 1 < numRegs) - fprintf(f, ", "); - } - - fprint_comment(f, inst); -} - - -void -_mesa_print_alu_instruction(const struct prog_instruction *inst, - const char *opcode_string, GLuint numRegs) -{ - _mesa_fprint_alu_instruction(stderr, inst, opcode_string, - numRegs, PROG_PRINT_DEBUG, NULL); -} - - -/** - * Print a single vertex/fragment program instruction. - */ -GLint -_mesa_fprint_instruction_opt(FILE *f, - const struct prog_instruction *inst, - GLint indent, - gl_prog_print_mode mode, - const struct gl_program *prog) -{ - GLint i; - - if (inst->Opcode == OPCODE_ELSE || - inst->Opcode == OPCODE_ENDIF || - inst->Opcode == OPCODE_ENDLOOP || - inst->Opcode == OPCODE_ENDSUB) { - indent -= 3; - } - for (i = 0; i < indent; i++) { - fprintf(f, " "); - } - - switch (inst->Opcode) { - case OPCODE_PRINT: - fprintf(f, "PRINT '%s'", (char *) inst->Data); - if (inst->SrcReg[0].File != PROGRAM_UNDEFINED) { - fprintf(f, ", "); - fprintf(f, "%s[%d]%s", - _mesa_register_file_name((gl_register_file) inst->SrcReg[0].File), - inst->SrcReg[0].Index, - _mesa_swizzle_string(inst->SrcReg[0].Swizzle, - inst->SrcReg[0].Negate, GL_FALSE)); - } - if (inst->Comment) - fprintf(f, " # %s", inst->Comment); - fprint_comment(f, inst); - break; - case OPCODE_SWZ: - fprintf(f, "SWZ"); - if (inst->SaturateMode == SATURATE_ZERO_ONE) - fprintf(f, "_SAT"); - fprintf(f, " "); - fprint_dst_reg(f, &inst->DstReg, mode, prog); - fprintf(f, ", %s[%d], %s", - _mesa_register_file_name((gl_register_file) inst->SrcReg[0].File), - inst->SrcReg[0].Index, - _mesa_swizzle_string(inst->SrcReg[0].Swizzle, - inst->SrcReg[0].Negate, GL_TRUE)); - fprint_comment(f, inst); - break; - case OPCODE_TEX: - case OPCODE_TXP: - case OPCODE_TXL: - case OPCODE_TXB: - case OPCODE_TXD: - fprintf(f, "%s", _mesa_opcode_string(inst->Opcode)); - if (inst->SaturateMode == SATURATE_ZERO_ONE) - fprintf(f, "_SAT"); - fprintf(f, " "); - fprint_dst_reg(f, &inst->DstReg, mode, prog); - fprintf(f, ", "); - fprint_src_reg(f, &inst->SrcReg[0], mode, prog); - if (inst->Opcode == OPCODE_TXD) { - fprintf(f, ", "); - fprint_src_reg(f, &inst->SrcReg[1], mode, prog); - fprintf(f, ", "); - fprint_src_reg(f, &inst->SrcReg[2], mode, prog); - } - fprintf(f, ", texture[%d], ", inst->TexSrcUnit); - switch (inst->TexSrcTarget) { - case TEXTURE_1D_INDEX: fprintf(f, "1D"); break; - case TEXTURE_2D_INDEX: fprintf(f, "2D"); break; - case TEXTURE_3D_INDEX: fprintf(f, "3D"); break; - case TEXTURE_CUBE_INDEX: fprintf(f, "CUBE"); break; - case TEXTURE_RECT_INDEX: fprintf(f, "RECT"); break; - case TEXTURE_1D_ARRAY_INDEX: fprintf(f, "1D_ARRAY"); break; - case TEXTURE_2D_ARRAY_INDEX: fprintf(f, "2D_ARRAY"); break; - default: - ; - } - if (inst->TexShadow) - fprintf(f, " SHADOW"); - fprint_comment(f, inst); - break; - - case OPCODE_KIL: - fprintf(f, "%s", _mesa_opcode_string(inst->Opcode)); - fprintf(f, " "); - fprint_src_reg(f, &inst->SrcReg[0], mode, prog); - fprint_comment(f, inst); - break; - case OPCODE_KIL_NV: - fprintf(f, "%s", _mesa_opcode_string(inst->Opcode)); - fprintf(f, " "); - fprintf(f, "%s.%s", - _mesa_condcode_string(inst->DstReg.CondMask), - _mesa_swizzle_string(inst->DstReg.CondSwizzle, - GL_FALSE, GL_FALSE)); - fprint_comment(f, inst); - break; - - case OPCODE_ARL: - fprintf(f, "ARL "); - fprint_dst_reg(f, &inst->DstReg, mode, prog); - fprintf(f, ", "); - fprint_src_reg(f, &inst->SrcReg[0], mode, prog); - fprint_comment(f, inst); - break; - case OPCODE_BRA: - fprintf(f, "BRA %d (%s%s)", - inst->BranchTarget, - _mesa_condcode_string(inst->DstReg.CondMask), - _mesa_swizzle_string(inst->DstReg.CondSwizzle, 0, GL_FALSE)); - fprint_comment(f, inst); - break; - case OPCODE_IF: - if (inst->SrcReg[0].File != PROGRAM_UNDEFINED) { - /* Use ordinary register */ - fprintf(f, "IF "); - fprint_src_reg(f, &inst->SrcReg[0], mode, prog); - fprintf(f, "; "); - } - else { - /* Use cond codes */ - fprintf(f, "IF (%s%s);", - _mesa_condcode_string(inst->DstReg.CondMask), - _mesa_swizzle_string(inst->DstReg.CondSwizzle, - 0, GL_FALSE)); - } - fprintf(f, " # (if false, goto %d)", inst->BranchTarget); - fprint_comment(f, inst); - return indent + 3; - case OPCODE_ELSE: - fprintf(f, "ELSE; # (goto %d)\n", inst->BranchTarget); - return indent + 3; - case OPCODE_ENDIF: - fprintf(f, "ENDIF;\n"); - break; - case OPCODE_BGNLOOP: - fprintf(f, "BGNLOOP; # (end at %d)\n", inst->BranchTarget); - return indent + 3; - case OPCODE_ENDLOOP: - fprintf(f, "ENDLOOP; # (goto %d)\n", inst->BranchTarget); - break; - case OPCODE_BRK: - case OPCODE_CONT: - fprintf(f, "%s (%s%s); # (goto %d)", - _mesa_opcode_string(inst->Opcode), - _mesa_condcode_string(inst->DstReg.CondMask), - _mesa_swizzle_string(inst->DstReg.CondSwizzle, 0, GL_FALSE), - inst->BranchTarget); - fprint_comment(f, inst); - break; - - case OPCODE_BGNSUB: - if (mode == PROG_PRINT_NV) { - fprintf(f, "%s:\n", inst->Comment); /* comment is label */ - return indent; - } - else { - fprintf(f, "BGNSUB"); - fprint_comment(f, inst); - return indent + 3; - } - case OPCODE_ENDSUB: - if (mode == PROG_PRINT_DEBUG) { - fprintf(f, "ENDSUB"); - fprint_comment(f, inst); - } - break; - case OPCODE_CAL: - if (mode == PROG_PRINT_NV) { - fprintf(f, "CAL %s; # (goto %d)\n", inst->Comment, inst->BranchTarget); - } - else { - fprintf(f, "CAL %u", inst->BranchTarget); - fprint_comment(f, inst); - } - break; - case OPCODE_RET: - fprintf(f, "RET (%s%s)", - _mesa_condcode_string(inst->DstReg.CondMask), - _mesa_swizzle_string(inst->DstReg.CondSwizzle, 0, GL_FALSE)); - fprint_comment(f, inst); - break; - - case OPCODE_END: - fprintf(f, "END\n"); - break; - case OPCODE_NOP: - if (mode == PROG_PRINT_DEBUG) { - fprintf(f, "NOP"); - fprint_comment(f, inst); - } - else if (inst->Comment) { - /* ARB/NV extensions don't have NOP instruction */ - fprintf(f, "# %s\n", inst->Comment); - } - break; - case OPCODE_EMIT_VERTEX: - fprintf(f, "EMIT_VERTEX\n"); - break; - case OPCODE_END_PRIMITIVE: - fprintf(f, "END_PRIMITIVE\n"); - break; - /* XXX may need other special-case instructions */ - default: - if (inst->Opcode < MAX_OPCODE) { - /* typical alu instruction */ - _mesa_fprint_alu_instruction(f, inst, - _mesa_opcode_string(inst->Opcode), - _mesa_num_inst_src_regs(inst->Opcode), - mode, prog); - } - else { - _mesa_fprint_alu_instruction(f, inst, - _mesa_opcode_string(inst->Opcode), - 3/*_mesa_num_inst_src_regs(inst->Opcode)*/, - mode, prog); - } - break; - } - return indent; -} - - -GLint -_mesa_print_instruction_opt(const struct prog_instruction *inst, - GLint indent, - gl_prog_print_mode mode, - const struct gl_program *prog) -{ - return _mesa_fprint_instruction_opt(stderr, inst, indent, mode, prog); -} - - -void -_mesa_print_instruction(const struct prog_instruction *inst) -{ - /* note: 4th param should be ignored for PROG_PRINT_DEBUG */ - _mesa_fprint_instruction_opt(stderr, inst, 0, PROG_PRINT_DEBUG, NULL); -} - - - -/** - * Print program, with options. - */ -void -_mesa_fprint_program_opt(FILE *f, - const struct gl_program *prog, - gl_prog_print_mode mode, - GLboolean lineNumbers) -{ - GLuint i, indent = 0; - - switch (prog->Target) { - case GL_VERTEX_PROGRAM_ARB: - if (mode == PROG_PRINT_ARB) - fprintf(f, "!!ARBvp1.0\n"); - else if (mode == PROG_PRINT_NV) - fprintf(f, "!!VP1.0\n"); - else - fprintf(f, "# Vertex Program/Shader %u\n", prog->Id); - break; - case GL_FRAGMENT_PROGRAM_ARB: - case GL_FRAGMENT_PROGRAM_NV: - if (mode == PROG_PRINT_ARB) - fprintf(f, "!!ARBfp1.0\n"); - else if (mode == PROG_PRINT_NV) - fprintf(f, "!!FP1.0\n"); - else - fprintf(f, "# Fragment Program/Shader %u\n", prog->Id); - break; - case MESA_GEOMETRY_PROGRAM: - fprintf(f, "# Geometry Shader\n"); - } - - for (i = 0; i < prog->NumInstructions; i++) { - if (lineNumbers) - fprintf(f, "%3d: ", i); - indent = _mesa_fprint_instruction_opt(f, prog->Instructions + i, - indent, mode, prog); - } -} - - -/** - * Print program to stderr, default options. - */ -void -_mesa_print_program(const struct gl_program *prog) -{ - _mesa_fprint_program_opt(stderr, prog, PROG_PRINT_DEBUG, GL_TRUE); -} - - -/** - * Return binary representation of 64-bit value (as a string). - * Insert a comma to separate each group of 8 bits. - * Note we return a pointer to local static storage so this is not - * re-entrant, etc. - * XXX move to imports.[ch] if useful elsewhere. - */ -static const char * -binary(GLbitfield64 val) -{ - static char buf[80]; - GLint i, len = 0; - for (i = 63; i >= 0; --i) { - if (val & (BITFIELD64_BIT(i))) - buf[len++] = '1'; - else if (len > 0 || i == 0) - buf[len++] = '0'; - if (len > 0 && ((i-1) % 8) == 7) - buf[len++] = ','; - } - buf[len] = '\0'; - return buf; -} - - -/** - * Print all of a program's parameters/fields to given file. - */ -static void -_mesa_fprint_program_parameters(FILE *f, - struct gl_context *ctx, - const struct gl_program *prog) -{ - GLuint i; - - fprintf(f, "InputsRead: 0x%x (0b%s)\n", - prog->InputsRead, binary(prog->InputsRead)); - fprintf(f, "OutputsWritten: 0x%llx (0b%s)\n", - (unsigned long long)prog->OutputsWritten, - binary(prog->OutputsWritten)); - fprintf(f, "NumInstructions=%d\n", prog->NumInstructions); - fprintf(f, "NumTemporaries=%d\n", prog->NumTemporaries); - fprintf(f, "NumParameters=%d\n", prog->NumParameters); - fprintf(f, "NumAttributes=%d\n", prog->NumAttributes); - fprintf(f, "NumAddressRegs=%d\n", prog->NumAddressRegs); - fprintf(f, "IndirectRegisterFiles: 0x%x (0b%s)\n", - prog->IndirectRegisterFiles, binary(prog->IndirectRegisterFiles)); - fprintf(f, "SamplersUsed: 0x%x (0b%s)\n", - prog->SamplersUsed, binary(prog->SamplersUsed)); - fprintf(f, "Samplers=[ "); - for (i = 0; i < MAX_SAMPLERS; i++) { - fprintf(f, "%d ", prog->SamplerUnits[i]); - } - fprintf(f, "]\n"); - - _mesa_load_state_parameters(ctx, prog->Parameters); - -#if 0 - fprintf(f, "Local Params:\n"); - for (i = 0; i < MAX_PROGRAM_LOCAL_PARAMS; i++){ - const GLfloat *p = prog->LocalParams[i]; - fprintf(f, "%2d: %f, %f, %f, %f\n", i, p[0], p[1], p[2], p[3]); - } -#endif - _mesa_print_parameter_list(prog->Parameters); -} - - -/** - * Print all of a program's parameters/fields to stderr. - */ -void -_mesa_print_program_parameters(struct gl_context *ctx, const struct gl_program *prog) -{ - _mesa_fprint_program_parameters(stderr, ctx, prog); -} - - -/** - * Print a program parameter list to given file. - */ -static void -_mesa_fprint_parameter_list(FILE *f, - const struct gl_program_parameter_list *list) -{ - GLuint i; - - if (!list) - return; - - if (0) - fprintf(f, "param list %p\n", (void *) list); - fprintf(f, "dirty state flags: 0x%x\n", list->StateFlags); - for (i = 0; i < list->NumParameters; i++){ - struct gl_program_parameter *param = list->Parameters + i; - const GLfloat *v = (GLfloat *) list->ParameterValues[i]; - fprintf(f, "param[%d] sz=%d %s %s = {%.3g, %.3g, %.3g, %.3g}", - i, param->Size, - _mesa_register_file_name(list->Parameters[i].Type), - param->Name, v[0], v[1], v[2], v[3]); - if (param->Flags & PROG_PARAM_BIT_CENTROID) - fprintf(f, " Centroid"); - if (param->Flags & PROG_PARAM_BIT_INVARIANT) - fprintf(f, " Invariant"); - if (param->Flags & PROG_PARAM_BIT_FLAT) - fprintf(f, " Flat"); - if (param->Flags & PROG_PARAM_BIT_LINEAR) - fprintf(f, " Linear"); - fprintf(f, "\n"); - } -} - - -/** - * Print a program parameter list to stderr. - */ -void -_mesa_print_parameter_list(const struct gl_program_parameter_list *list) -{ - _mesa_fprint_parameter_list(stderr, list); -} - - -/** - * Write shader and associated info to a file. - */ -void -_mesa_write_shader_to_file(const struct gl_shader *shader) -{ - const char *type; - char filename[100]; - FILE *f; - - if (shader->Type == GL_FRAGMENT_SHADER) - type = "frag"; - else if (shader->Type == GL_VERTEX_SHADER) - type = "vert"; - else - type = "geom"; - - _mesa_snprintf(filename, sizeof(filename), "shader_%u.%s", shader->Name, type); - f = fopen(filename, "w"); - if (!f) { - fprintf(stderr, "Unable to open %s for writing\n", filename); - return; - } - - fprintf(f, "/* Shader %u source, checksum %u */\n", shader->Name, shader->SourceChecksum); - fputs(shader->Source, f); - fprintf(f, "\n"); - - fprintf(f, "/* Compile status: %s */\n", - shader->CompileStatus ? "ok" : "fail"); - fprintf(f, "/* Log Info: */\n"); - if (shader->InfoLog) { - fputs(shader->InfoLog, f); - } - if (shader->CompileStatus && shader->Program) { - fprintf(f, "/* GPU code */\n"); - fprintf(f, "/*\n"); - _mesa_fprint_program_opt(f, shader->Program, PROG_PRINT_DEBUG, GL_TRUE); - fprintf(f, "*/\n"); - fprintf(f, "/* Parameters / constants */\n"); - fprintf(f, "/*\n"); - _mesa_fprint_parameter_list(f, shader->Program->Parameters); - fprintf(f, "*/\n"); - } - - fclose(f); -} - - -/** - * Append the shader's uniform info/values to the shader log file. - * The log file will typically have been created by the - * _mesa_write_shader_to_file function. - */ -void -_mesa_append_uniforms_to_file(const struct gl_shader *shader, - const struct gl_program *prog) -{ - const char *type; - char filename[100]; - FILE *f; - - if (shader->Type == GL_FRAGMENT_SHADER) - type = "frag"; - else - type = "vert"; - - _mesa_snprintf(filename, sizeof(filename), "shader_%u.%s", shader->Name, type); - f = fopen(filename, "a"); /* append */ - if (!f) { - fprintf(stderr, "Unable to open %s for appending\n", filename); - return; - } - - fprintf(f, "/* First-draw parameters / constants */\n"); - fprintf(f, "/*\n"); - _mesa_fprint_parameter_list(f, prog->Parameters); - fprintf(f, "*/\n"); - - fclose(f); -} +/* + * Mesa 3-D graphics library + * Version: 7.3 + * + * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. + * Copyright (C) 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/** + * \file prog_print.c + * Print vertex/fragment programs - for debugging. + * \author Brian Paul + */ + +#include "main/glheader.h" +#include "main/context.h" +#include "main/imports.h" +#include "prog_instruction.h" +#include "prog_parameter.h" +#include "prog_print.h" +#include "prog_statevars.h" + + + +/** + * Return string name for given program/register file. + */ +const char * +_mesa_register_file_name(gl_register_file f) +{ + switch (f) { + case PROGRAM_TEMPORARY: + return "TEMP"; + case PROGRAM_LOCAL_PARAM: + return "LOCAL"; + case PROGRAM_ENV_PARAM: + return "ENV"; + case PROGRAM_STATE_VAR: + return "STATE"; + case PROGRAM_INPUT: + return "INPUT"; + case PROGRAM_OUTPUT: + return "OUTPUT"; + case PROGRAM_NAMED_PARAM: + return "NAMED"; + case PROGRAM_CONSTANT: + return "CONST"; + case PROGRAM_UNIFORM: + return "UNIFORM"; + case PROGRAM_VARYING: + return "VARYING"; + case PROGRAM_WRITE_ONLY: + return "WRITE_ONLY"; + case PROGRAM_ADDRESS: + return "ADDR"; + case PROGRAM_SAMPLER: + return "SAMPLER"; + case PROGRAM_SYSTEM_VALUE: + return "SYSVAL"; + case PROGRAM_UNDEFINED: + return "UNDEFINED"; + default: + { + static char s[20]; + _mesa_snprintf(s, sizeof(s), "FILE%u", f); + return s; + } + } +} + + +/** + * Return ARB_v/f_prog-style input attrib string. + */ +static const char * +arb_input_attrib_string(GLint index, GLenum progType) +{ + /* + * These strings should match the VERT_ATTRIB_x and FRAG_ATTRIB_x tokens. + */ + const char *vertAttribs[] = { + "vertex.position", + "vertex.weight", + "vertex.normal", + "vertex.color.primary", + "vertex.color.secondary", + "vertex.fogcoord", + "vertex.(six)", + "vertex.(seven)", + "vertex.texcoord[0]", + "vertex.texcoord[1]", + "vertex.texcoord[2]", + "vertex.texcoord[3]", + "vertex.texcoord[4]", + "vertex.texcoord[5]", + "vertex.texcoord[6]", + "vertex.texcoord[7]", + "vertex.attrib[0]", + "vertex.attrib[1]", + "vertex.attrib[2]", + "vertex.attrib[3]", + "vertex.attrib[4]", + "vertex.attrib[5]", + "vertex.attrib[6]", + "vertex.attrib[7]", + "vertex.attrib[8]", + "vertex.attrib[9]", + "vertex.attrib[10]", + "vertex.attrib[11]", + "vertex.attrib[12]", + "vertex.attrib[13]", + "vertex.attrib[14]", + "vertex.attrib[15]" + }; + const char *fragAttribs[] = { + "fragment.position", + "fragment.color.primary", + "fragment.color.secondary", + "fragment.fogcoord", + "fragment.texcoord[0]", + "fragment.texcoord[1]", + "fragment.texcoord[2]", + "fragment.texcoord[3]", + "fragment.texcoord[4]", + "fragment.texcoord[5]", + "fragment.texcoord[6]", + "fragment.texcoord[7]", + "fragment.varying[0]", + "fragment.varying[1]", + "fragment.varying[2]", + "fragment.varying[3]", + "fragment.varying[4]", + "fragment.varying[5]", + "fragment.varying[6]", + "fragment.varying[7]" + }; + + /* sanity checks */ + assert(strcmp(vertAttribs[VERT_ATTRIB_TEX0], "vertex.texcoord[0]") == 0); + assert(strcmp(vertAttribs[VERT_ATTRIB_GENERIC15], "vertex.attrib[15]") == 0); + + if (progType == GL_VERTEX_PROGRAM_ARB) { + assert(index < sizeof(vertAttribs) / sizeof(vertAttribs[0])); + return vertAttribs[index]; + } + else { + assert(index < sizeof(fragAttribs) / sizeof(fragAttribs[0])); + return fragAttribs[index]; + } +} + + +/** + * Print a vertex program's InputsRead field in human-readable format. + * For debugging. + */ +void +_mesa_print_vp_inputs(GLbitfield inputs) +{ + printf("VP Inputs 0x%x: \n", inputs); + while (inputs) { + GLint attr = _mesa_ffs(inputs) - 1; + const char *name = arb_input_attrib_string(attr, + GL_VERTEX_PROGRAM_ARB); + printf(" %d: %s\n", attr, name); + inputs &= ~(1 << attr); + } +} + + +/** + * Print a fragment program's InputsRead field in human-readable format. + * For debugging. + */ +void +_mesa_print_fp_inputs(GLbitfield inputs) +{ + printf("FP Inputs 0x%x: \n", inputs); + while (inputs) { + GLint attr = _mesa_ffs(inputs) - 1; + const char *name = arb_input_attrib_string(attr, + GL_FRAGMENT_PROGRAM_ARB); + printf(" %d: %s\n", attr, name); + inputs &= ~(1 << attr); + } +} + + + +/** + * Return ARB_v/f_prog-style output attrib string. + */ +static const char * +arb_output_attrib_string(GLint index, GLenum progType) +{ + /* + * These strings should match the VERT_RESULT_x and FRAG_RESULT_x tokens. + */ + const char *vertResults[] = { + "result.position", + "result.color.primary", + "result.color.secondary", + "result.fogcoord", + "result.texcoord[0]", + "result.texcoord[1]", + "result.texcoord[2]", + "result.texcoord[3]", + "result.texcoord[4]", + "result.texcoord[5]", + "result.texcoord[6]", + "result.texcoord[7]", + "result.varying[0]", + "result.varying[1]", + "result.varying[2]", + "result.varying[3]", + "result.varying[4]", + "result.varying[5]", + "result.varying[6]", + "result.varying[7]" + }; + const char *fragResults[] = { + "result.color", + "result.color(half)", + "result.depth", + "result.color[0]", + "result.color[1]", + "result.color[2]", + "result.color[3]" + }; + + if (progType == GL_VERTEX_PROGRAM_ARB) { + assert(index < sizeof(vertResults) / sizeof(vertResults[0])); + return vertResults[index]; + } + else { + assert(index < sizeof(fragResults) / sizeof(fragResults[0])); + return fragResults[index]; + } +} + + +/** + * Return string representation of the given register. + * Note that some types of registers (like PROGRAM_UNIFORM) aren't defined + * by the ARB/NV program languages so we've taken some liberties here. + * \param f the register file (PROGRAM_INPUT, PROGRAM_TEMPORARY, etc) + * \param index number of the register in the register file + * \param mode the output format/mode/style + * \param prog pointer to containing program + */ +static const char * +reg_string(gl_register_file f, GLint index, gl_prog_print_mode mode, + GLboolean relAddr, const struct gl_program *prog, + GLboolean hasIndex2, GLboolean relAddr2, GLint index2) +{ + static char str[100]; + const char *addr = relAddr ? "ADDR+" : ""; + + str[0] = 0; + + switch (mode) { + case PROG_PRINT_DEBUG: + sprintf(str, "%s[%s%d]", + _mesa_register_file_name(f), addr, index); + if (hasIndex2) { + int offset = strlen(str); + const char *addr2 = relAddr2 ? "ADDR+" : ""; + sprintf(str+offset, "[%s%d]", addr2, index2); + } + break; + + case PROG_PRINT_ARB: + switch (f) { + case PROGRAM_INPUT: + sprintf(str, "%s", arb_input_attrib_string(index, prog->Target)); + break; + case PROGRAM_OUTPUT: + sprintf(str, "%s", arb_output_attrib_string(index, prog->Target)); + break; + case PROGRAM_TEMPORARY: + sprintf(str, "temp%d", index); + break; + case PROGRAM_ENV_PARAM: + sprintf(str, "program.env[%s%d]", addr, index); + break; + case PROGRAM_LOCAL_PARAM: + sprintf(str, "program.local[%s%d]", addr, index); + break; + case PROGRAM_VARYING: /* extension */ + sprintf(str, "varying[%s%d]", addr, index); + break; + case PROGRAM_CONSTANT: /* extension */ + sprintf(str, "constant[%s%d]", addr, index); + break; + case PROGRAM_UNIFORM: /* extension */ + sprintf(str, "uniform[%s%d]", addr, index); + break; + case PROGRAM_SYSTEM_VALUE: + sprintf(str, "sysvalue[%s%d]", addr, index); + break; + case PROGRAM_STATE_VAR: + { + struct gl_program_parameter *param + = prog->Parameters->Parameters + index; + char *state = _mesa_program_state_string(param->StateIndexes); + sprintf(str, "%s", state); + free(state); + } + break; + case PROGRAM_ADDRESS: + sprintf(str, "A%d", index); + break; + default: + _mesa_problem(NULL, "bad file in reg_string()"); + } + break; + + case PROG_PRINT_NV: + switch (f) { + case PROGRAM_INPUT: + if (prog->Target == GL_VERTEX_PROGRAM_ARB) + sprintf(str, "v[%d]", index); + else + sprintf(str, "f[%d]", index); + break; + case PROGRAM_OUTPUT: + sprintf(str, "o[%d]", index); + break; + case PROGRAM_TEMPORARY: + sprintf(str, "R%d", index); + break; + case PROGRAM_ENV_PARAM: + sprintf(str, "c[%d]", index); + break; + case PROGRAM_VARYING: /* extension */ + sprintf(str, "varying[%s%d]", addr, index); + break; + case PROGRAM_UNIFORM: /* extension */ + sprintf(str, "uniform[%s%d]", addr, index); + break; + case PROGRAM_CONSTANT: /* extension */ + sprintf(str, "constant[%s%d]", addr, index); + break; + case PROGRAM_STATE_VAR: /* extension */ + sprintf(str, "state[%s%d]", addr, index); + break; + default: + _mesa_problem(NULL, "bad file in reg_string()"); + } + break; + + default: + _mesa_problem(NULL, "bad mode in reg_string()"); + } + + return str; +} + + +/** + * Return a string representation of the given swizzle word. + * If extended is true, use extended (comma-separated) format. + * \param swizzle the swizzle field + * \param negateBase 4-bit negation vector + * \param extended if true, also allow 0, 1 values + */ +const char * +_mesa_swizzle_string(GLuint swizzle, GLuint negateMask, GLboolean extended) +{ + static const char swz[] = "xyzw01!?"; /* See SWIZZLE_x definitions */ + static char s[20]; + GLuint i = 0; + + if (!extended && swizzle == SWIZZLE_NOOP && negateMask == 0) + return ""; /* no swizzle/negation */ + + if (!extended) + s[i++] = '.'; + + if (negateMask & NEGATE_X) + s[i++] = '-'; + s[i++] = swz[GET_SWZ(swizzle, 0)]; + + if (extended) { + s[i++] = ','; + } + + if (negateMask & NEGATE_Y) + s[i++] = '-'; + s[i++] = swz[GET_SWZ(swizzle, 1)]; + + if (extended) { + s[i++] = ','; + } + + if (negateMask & NEGATE_Z) + s[i++] = '-'; + s[i++] = swz[GET_SWZ(swizzle, 2)]; + + if (extended) { + s[i++] = ','; + } + + if (negateMask & NEGATE_W) + s[i++] = '-'; + s[i++] = swz[GET_SWZ(swizzle, 3)]; + + s[i] = 0; + return s; +} + + +void +_mesa_print_swizzle(GLuint swizzle) +{ + if (swizzle == SWIZZLE_XYZW) { + printf(".xyzw\n"); + } + else { + const char *s = _mesa_swizzle_string(swizzle, 0, 0); + printf("%s\n", s); + } +} + + +const char * +_mesa_writemask_string(GLuint writeMask) +{ + static char s[10]; + GLuint i = 0; + + if (writeMask == WRITEMASK_XYZW) + return ""; + + s[i++] = '.'; + if (writeMask & WRITEMASK_X) + s[i++] = 'x'; + if (writeMask & WRITEMASK_Y) + s[i++] = 'y'; + if (writeMask & WRITEMASK_Z) + s[i++] = 'z'; + if (writeMask & WRITEMASK_W) + s[i++] = 'w'; + + s[i] = 0; + return s; +} + + +const char * +_mesa_condcode_string(GLuint condcode) +{ + switch (condcode) { + case COND_GT: return "GT"; + case COND_EQ: return "EQ"; + case COND_LT: return "LT"; + case COND_UN: return "UN"; + case COND_GE: return "GE"; + case COND_LE: return "LE"; + case COND_NE: return "NE"; + case COND_TR: return "TR"; + case COND_FL: return "FL"; + default: return "cond???"; + } +} + + +static void +fprint_dst_reg(FILE * f, + const struct prog_dst_register *dstReg, + gl_prog_print_mode mode, + const struct gl_program *prog) +{ + fprintf(f, "%s%s", + reg_string((gl_register_file) dstReg->File, + dstReg->Index, mode, dstReg->RelAddr, prog, + GL_FALSE, GL_FALSE, 0), + _mesa_writemask_string(dstReg->WriteMask)); + + if (dstReg->CondMask != COND_TR) { + fprintf(f, " (%s.%s)", + _mesa_condcode_string(dstReg->CondMask), + _mesa_swizzle_string(dstReg->CondSwizzle, + GL_FALSE, GL_FALSE)); + } + +#if 0 + fprintf(f, "%s[%d]%s", + _mesa_register_file_name((gl_register_file) dstReg->File), + dstReg->Index, + _mesa_writemask_string(dstReg->WriteMask)); +#endif +} + + +static void +fprint_src_reg(FILE *f, + const struct prog_src_register *srcReg, + gl_prog_print_mode mode, + const struct gl_program *prog) +{ + const char *abs = srcReg->Abs ? "|" : ""; + + fprintf(f, "%s%s%s%s", + abs, + reg_string((gl_register_file) srcReg->File, + srcReg->Index, mode, srcReg->RelAddr, prog, + srcReg->HasIndex2, srcReg->RelAddr2, srcReg->Index2), + _mesa_swizzle_string(srcReg->Swizzle, + srcReg->Negate, GL_FALSE), + abs); +#if 0 + fprintf(f, "%s[%d]%s", + _mesa_register_file_name((gl_register_file) srcReg->File), + srcReg->Index, + _mesa_swizzle_string(srcReg->Swizzle, + srcReg->Negate, GL_FALSE)); +#endif +} + + +static void +fprint_comment(FILE *f, const struct prog_instruction *inst) +{ + if (inst->Comment) + fprintf(f, "; # %s\n", inst->Comment); + else + fprintf(f, ";\n"); +} + + +void +_mesa_fprint_alu_instruction(FILE *f, + const struct prog_instruction *inst, + const char *opcode_string, GLuint numRegs, + gl_prog_print_mode mode, + const struct gl_program *prog) +{ + GLuint j; + + fprintf(f, "%s", opcode_string); + if (inst->CondUpdate) + fprintf(f, ".C"); + + /* frag prog only */ + if (inst->SaturateMode == SATURATE_ZERO_ONE) + fprintf(f, "_SAT"); + + fprintf(f, " "); + if (inst->DstReg.File != PROGRAM_UNDEFINED) { + fprint_dst_reg(f, &inst->DstReg, mode, prog); + } + else { + fprintf(f, " ???"); + } + + if (numRegs > 0) + fprintf(f, ", "); + + for (j = 0; j < numRegs; j++) { + fprint_src_reg(f, inst->SrcReg + j, mode, prog); + if (j + 1 < numRegs) + fprintf(f, ", "); + } + + fprint_comment(f, inst); +} + + +void +_mesa_print_alu_instruction(const struct prog_instruction *inst, + const char *opcode_string, GLuint numRegs) +{ + _mesa_fprint_alu_instruction(stderr, inst, opcode_string, + numRegs, PROG_PRINT_DEBUG, NULL); +} + + +/** + * Print a single vertex/fragment program instruction. + */ +GLint +_mesa_fprint_instruction_opt(FILE *f, + const struct prog_instruction *inst, + GLint indent, + gl_prog_print_mode mode, + const struct gl_program *prog) +{ + GLint i; + + if (inst->Opcode == OPCODE_ELSE || + inst->Opcode == OPCODE_ENDIF || + inst->Opcode == OPCODE_ENDLOOP || + inst->Opcode == OPCODE_ENDSUB) { + indent -= 3; + } + for (i = 0; i < indent; i++) { + fprintf(f, " "); + } + + switch (inst->Opcode) { + case OPCODE_PRINT: + fprintf(f, "PRINT '%s'", (char *) inst->Data); + if (inst->SrcReg[0].File != PROGRAM_UNDEFINED) { + fprintf(f, ", "); + fprintf(f, "%s[%d]%s", + _mesa_register_file_name((gl_register_file) inst->SrcReg[0].File), + inst->SrcReg[0].Index, + _mesa_swizzle_string(inst->SrcReg[0].Swizzle, + inst->SrcReg[0].Negate, GL_FALSE)); + } + if (inst->Comment) + fprintf(f, " # %s", inst->Comment); + fprint_comment(f, inst); + break; + case OPCODE_SWZ: + fprintf(f, "SWZ"); + if (inst->SaturateMode == SATURATE_ZERO_ONE) + fprintf(f, "_SAT"); + fprintf(f, " "); + fprint_dst_reg(f, &inst->DstReg, mode, prog); + fprintf(f, ", %s[%d], %s", + _mesa_register_file_name((gl_register_file) inst->SrcReg[0].File), + inst->SrcReg[0].Index, + _mesa_swizzle_string(inst->SrcReg[0].Swizzle, + inst->SrcReg[0].Negate, GL_TRUE)); + fprint_comment(f, inst); + break; + case OPCODE_TEX: + case OPCODE_TXP: + case OPCODE_TXL: + case OPCODE_TXB: + case OPCODE_TXD: + fprintf(f, "%s", _mesa_opcode_string(inst->Opcode)); + if (inst->SaturateMode == SATURATE_ZERO_ONE) + fprintf(f, "_SAT"); + fprintf(f, " "); + fprint_dst_reg(f, &inst->DstReg, mode, prog); + fprintf(f, ", "); + fprint_src_reg(f, &inst->SrcReg[0], mode, prog); + if (inst->Opcode == OPCODE_TXD) { + fprintf(f, ", "); + fprint_src_reg(f, &inst->SrcReg[1], mode, prog); + fprintf(f, ", "); + fprint_src_reg(f, &inst->SrcReg[2], mode, prog); + } + fprintf(f, ", texture[%d], ", inst->TexSrcUnit); + switch (inst->TexSrcTarget) { + case TEXTURE_1D_INDEX: fprintf(f, "1D"); break; + case TEXTURE_2D_INDEX: fprintf(f, "2D"); break; + case TEXTURE_3D_INDEX: fprintf(f, "3D"); break; + case TEXTURE_CUBE_INDEX: fprintf(f, "CUBE"); break; + case TEXTURE_RECT_INDEX: fprintf(f, "RECT"); break; + case TEXTURE_1D_ARRAY_INDEX: fprintf(f, "1D_ARRAY"); break; + case TEXTURE_2D_ARRAY_INDEX: fprintf(f, "2D_ARRAY"); break; + default: + ; + } + if (inst->TexShadow) + fprintf(f, " SHADOW"); + fprint_comment(f, inst); + break; + + case OPCODE_KIL: + fprintf(f, "%s", _mesa_opcode_string(inst->Opcode)); + fprintf(f, " "); + fprint_src_reg(f, &inst->SrcReg[0], mode, prog); + fprint_comment(f, inst); + break; + case OPCODE_KIL_NV: + fprintf(f, "%s", _mesa_opcode_string(inst->Opcode)); + fprintf(f, " "); + fprintf(f, "%s.%s", + _mesa_condcode_string(inst->DstReg.CondMask), + _mesa_swizzle_string(inst->DstReg.CondSwizzle, + GL_FALSE, GL_FALSE)); + fprint_comment(f, inst); + break; + + case OPCODE_ARL: + fprintf(f, "ARL "); + fprint_dst_reg(f, &inst->DstReg, mode, prog); + fprintf(f, ", "); + fprint_src_reg(f, &inst->SrcReg[0], mode, prog); + fprint_comment(f, inst); + break; + case OPCODE_BRA: + fprintf(f, "BRA %d (%s%s)", + inst->BranchTarget, + _mesa_condcode_string(inst->DstReg.CondMask), + _mesa_swizzle_string(inst->DstReg.CondSwizzle, 0, GL_FALSE)); + fprint_comment(f, inst); + break; + case OPCODE_IF: + if (inst->SrcReg[0].File != PROGRAM_UNDEFINED) { + /* Use ordinary register */ + fprintf(f, "IF "); + fprint_src_reg(f, &inst->SrcReg[0], mode, prog); + fprintf(f, "; "); + } + else { + /* Use cond codes */ + fprintf(f, "IF (%s%s);", + _mesa_condcode_string(inst->DstReg.CondMask), + _mesa_swizzle_string(inst->DstReg.CondSwizzle, + 0, GL_FALSE)); + } + fprintf(f, " # (if false, goto %d)", inst->BranchTarget); + fprint_comment(f, inst); + return indent + 3; + case OPCODE_ELSE: + fprintf(f, "ELSE; # (goto %d)\n", inst->BranchTarget); + return indent + 3; + case OPCODE_ENDIF: + fprintf(f, "ENDIF;\n"); + break; + case OPCODE_BGNLOOP: + fprintf(f, "BGNLOOP; # (end at %d)\n", inst->BranchTarget); + return indent + 3; + case OPCODE_ENDLOOP: + fprintf(f, "ENDLOOP; # (goto %d)\n", inst->BranchTarget); + break; + case OPCODE_BRK: + case OPCODE_CONT: + fprintf(f, "%s (%s%s); # (goto %d)", + _mesa_opcode_string(inst->Opcode), + _mesa_condcode_string(inst->DstReg.CondMask), + _mesa_swizzle_string(inst->DstReg.CondSwizzle, 0, GL_FALSE), + inst->BranchTarget); + fprint_comment(f, inst); + break; + + case OPCODE_BGNSUB: + if (mode == PROG_PRINT_NV) { + fprintf(f, "%s:\n", inst->Comment); /* comment is label */ + return indent; + } + else { + fprintf(f, "BGNSUB"); + fprint_comment(f, inst); + return indent + 3; + } + case OPCODE_ENDSUB: + if (mode == PROG_PRINT_DEBUG) { + fprintf(f, "ENDSUB"); + fprint_comment(f, inst); + } + break; + case OPCODE_CAL: + if (mode == PROG_PRINT_NV) { + fprintf(f, "CAL %s; # (goto %d)\n", inst->Comment, inst->BranchTarget); + } + else { + fprintf(f, "CAL %u", inst->BranchTarget); + fprint_comment(f, inst); + } + break; + case OPCODE_RET: + fprintf(f, "RET (%s%s)", + _mesa_condcode_string(inst->DstReg.CondMask), + _mesa_swizzle_string(inst->DstReg.CondSwizzle, 0, GL_FALSE)); + fprint_comment(f, inst); + break; + + case OPCODE_END: + fprintf(f, "END\n"); + break; + case OPCODE_NOP: + if (mode == PROG_PRINT_DEBUG) { + fprintf(f, "NOP"); + fprint_comment(f, inst); + } + else if (inst->Comment) { + /* ARB/NV extensions don't have NOP instruction */ + fprintf(f, "# %s\n", inst->Comment); + } + break; + case OPCODE_EMIT_VERTEX: + fprintf(f, "EMIT_VERTEX\n"); + break; + case OPCODE_END_PRIMITIVE: + fprintf(f, "END_PRIMITIVE\n"); + break; + /* XXX may need other special-case instructions */ + default: + if (inst->Opcode < MAX_OPCODE) { + /* typical alu instruction */ + _mesa_fprint_alu_instruction(f, inst, + _mesa_opcode_string(inst->Opcode), + _mesa_num_inst_src_regs(inst->Opcode), + mode, prog); + } + else { + _mesa_fprint_alu_instruction(f, inst, + _mesa_opcode_string(inst->Opcode), + 3/*_mesa_num_inst_src_regs(inst->Opcode)*/, + mode, prog); + } + break; + } + return indent; +} + + +GLint +_mesa_print_instruction_opt(const struct prog_instruction *inst, + GLint indent, + gl_prog_print_mode mode, + const struct gl_program *prog) +{ + return _mesa_fprint_instruction_opt(stderr, inst, indent, mode, prog); +} + + +void +_mesa_print_instruction(const struct prog_instruction *inst) +{ + /* note: 4th param should be ignored for PROG_PRINT_DEBUG */ + _mesa_fprint_instruction_opt(stderr, inst, 0, PROG_PRINT_DEBUG, NULL); +} + + + +/** + * Print program, with options. + */ +void +_mesa_fprint_program_opt(FILE *f, + const struct gl_program *prog, + gl_prog_print_mode mode, + GLboolean lineNumbers) +{ + GLuint i, indent = 0; + + switch (prog->Target) { + case GL_VERTEX_PROGRAM_ARB: + if (mode == PROG_PRINT_ARB) + fprintf(f, "!!ARBvp1.0\n"); + else if (mode == PROG_PRINT_NV) + fprintf(f, "!!VP1.0\n"); + else + fprintf(f, "# Vertex Program/Shader %u\n", prog->Id); + break; + case GL_FRAGMENT_PROGRAM_ARB: + case GL_FRAGMENT_PROGRAM_NV: + if (mode == PROG_PRINT_ARB) + fprintf(f, "!!ARBfp1.0\n"); + else if (mode == PROG_PRINT_NV) + fprintf(f, "!!FP1.0\n"); + else + fprintf(f, "# Fragment Program/Shader %u\n", prog->Id); + break; + case MESA_GEOMETRY_PROGRAM: + fprintf(f, "# Geometry Shader\n"); + } + + for (i = 0; i < prog->NumInstructions; i++) { + if (lineNumbers) + fprintf(f, "%3d: ", i); + indent = _mesa_fprint_instruction_opt(f, prog->Instructions + i, + indent, mode, prog); + } +} + + +/** + * Print program to stderr, default options. + */ +void +_mesa_print_program(const struct gl_program *prog) +{ + _mesa_fprint_program_opt(stderr, prog, PROG_PRINT_DEBUG, GL_TRUE); +} + + +/** + * Return binary representation of 64-bit value (as a string). + * Insert a comma to separate each group of 8 bits. + * Note we return a pointer to local static storage so this is not + * re-entrant, etc. + * XXX move to imports.[ch] if useful elsewhere. + */ +static const char * +binary(GLbitfield64 val) +{ + static char buf[80]; + GLint i, len = 0; + for (i = 63; i >= 0; --i) { + if (val & (BITFIELD64_BIT(i))) + buf[len++] = '1'; + else if (len > 0 || i == 0) + buf[len++] = '0'; + if (len > 0 && ((i-1) % 8) == 7) + buf[len++] = ','; + } + buf[len] = '\0'; + return buf; +} + + +/** + * Print all of a program's parameters/fields to given file. + */ +static void +_mesa_fprint_program_parameters(FILE *f, + struct gl_context *ctx, + const struct gl_program *prog) +{ + GLuint i; + + fprintf(f, "InputsRead: 0x%x (0b%s)\n", + prog->InputsRead, binary(prog->InputsRead)); + fprintf(f, "OutputsWritten: 0x%llx (0b%s)\n", + (unsigned long long)prog->OutputsWritten, + binary(prog->OutputsWritten)); + fprintf(f, "NumInstructions=%d\n", prog->NumInstructions); + fprintf(f, "NumTemporaries=%d\n", prog->NumTemporaries); + fprintf(f, "NumParameters=%d\n", prog->NumParameters); + fprintf(f, "NumAttributes=%d\n", prog->NumAttributes); + fprintf(f, "NumAddressRegs=%d\n", prog->NumAddressRegs); + fprintf(f, "IndirectRegisterFiles: 0x%x (0b%s)\n", + prog->IndirectRegisterFiles, binary(prog->IndirectRegisterFiles)); + fprintf(f, "SamplersUsed: 0x%x (0b%s)\n", + prog->SamplersUsed, binary(prog->SamplersUsed)); + fprintf(f, "Samplers=[ "); + for (i = 0; i < MAX_SAMPLERS; i++) { + fprintf(f, "%d ", prog->SamplerUnits[i]); + } + fprintf(f, "]\n"); + + _mesa_load_state_parameters(ctx, prog->Parameters); + +#if 0 + fprintf(f, "Local Params:\n"); + for (i = 0; i < MAX_PROGRAM_LOCAL_PARAMS; i++){ + const GLfloat *p = prog->LocalParams[i]; + fprintf(f, "%2d: %f, %f, %f, %f\n", i, p[0], p[1], p[2], p[3]); + } +#endif + _mesa_print_parameter_list(prog->Parameters); +} + + +/** + * Print all of a program's parameters/fields to stderr. + */ +void +_mesa_print_program_parameters(struct gl_context *ctx, const struct gl_program *prog) +{ + _mesa_fprint_program_parameters(stderr, ctx, prog); +} + + +/** + * Print a program parameter list to given file. + */ +static void +_mesa_fprint_parameter_list(FILE *f, + const struct gl_program_parameter_list *list) +{ + GLuint i; + + if (!list) + return; + + if (0) + fprintf(f, "param list %p\n", (void *) list); + fprintf(f, "dirty state flags: 0x%x\n", list->StateFlags); + for (i = 0; i < list->NumParameters; i++){ + struct gl_program_parameter *param = list->Parameters + i; + const GLfloat *v = (GLfloat *) list->ParameterValues[i]; + fprintf(f, "param[%d] sz=%d %s %s = {%.3g, %.3g, %.3g, %.3g}", + i, param->Size, + _mesa_register_file_name(list->Parameters[i].Type), + param->Name, v[0], v[1], v[2], v[3]); + if (param->Flags & PROG_PARAM_BIT_CENTROID) + fprintf(f, " Centroid"); + if (param->Flags & PROG_PARAM_BIT_INVARIANT) + fprintf(f, " Invariant"); + if (param->Flags & PROG_PARAM_BIT_FLAT) + fprintf(f, " Flat"); + if (param->Flags & PROG_PARAM_BIT_LINEAR) + fprintf(f, " Linear"); + fprintf(f, "\n"); + } +} + + +/** + * Print a program parameter list to stderr. + */ +void +_mesa_print_parameter_list(const struct gl_program_parameter_list *list) +{ + _mesa_fprint_parameter_list(stderr, list); +} + + +/** + * Write shader and associated info to a file. + */ +void +_mesa_write_shader_to_file(const struct gl_shader *shader) +{ + const char *type; + char filename[100]; + FILE *f; + + if (shader->Type == GL_FRAGMENT_SHADER) + type = "frag"; + else if (shader->Type == GL_VERTEX_SHADER) + type = "vert"; + else + type = "geom"; + + _mesa_snprintf(filename, sizeof(filename), "shader_%u.%s", shader->Name, type); + f = fopen(filename, "w"); + if (!f) { + fprintf(stderr, "Unable to open %s for writing\n", filename); + return; + } + + fprintf(f, "/* Shader %u source, checksum %u */\n", shader->Name, shader->SourceChecksum); + fputs(shader->Source, f); + fprintf(f, "\n"); + + fprintf(f, "/* Compile status: %s */\n", + shader->CompileStatus ? "ok" : "fail"); + fprintf(f, "/* Log Info: */\n"); + if (shader->InfoLog) { + fputs(shader->InfoLog, f); + } + if (shader->CompileStatus && shader->Program) { + fprintf(f, "/* GPU code */\n"); + fprintf(f, "/*\n"); + _mesa_fprint_program_opt(f, shader->Program, PROG_PRINT_DEBUG, GL_TRUE); + fprintf(f, "*/\n"); + fprintf(f, "/* Parameters / constants */\n"); + fprintf(f, "/*\n"); + _mesa_fprint_parameter_list(f, shader->Program->Parameters); + fprintf(f, "*/\n"); + } + + fclose(f); +} + + +/** + * Append the shader's uniform info/values to the shader log file. + * The log file will typically have been created by the + * _mesa_write_shader_to_file function. + */ +void +_mesa_append_uniforms_to_file(const struct gl_shader *shader, + const struct gl_program *prog) +{ + const char *type; + char filename[100]; + FILE *f; + + if (shader->Type == GL_FRAGMENT_SHADER) + type = "frag"; + else + type = "vert"; + + _mesa_snprintf(filename, sizeof(filename), "shader_%u.%s", shader->Name, type); + f = fopen(filename, "a"); /* append */ + if (!f) { + fprintf(stderr, "Unable to open %s for appending\n", filename); + return; + } + + fprintf(f, "/* First-draw parameters / constants */\n"); + fprintf(f, "/*\n"); + _mesa_fprint_parameter_list(f, prog->Parameters); + fprintf(f, "*/\n"); + + fclose(f); +} diff --git a/mesalib/src/mesa/program/prog_statevars.c b/mesalib/src/mesa/program/prog_statevars.c index c68e1643a..6aa2409e8 100644 --- a/mesalib/src/mesa/program/prog_statevars.c +++ b/mesalib/src/mesa/program/prog_statevars.c @@ -1,1208 +1,1208 @@ -/* - * Mesa 3-D graphics library - * Version: 7.1 - * - * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - -/** - * \file prog_statevars.c - * Program state variable management. - * \author Brian Paul - */ - - -#include "main/glheader.h" -#include "main/context.h" -#include "main/imports.h" -#include "main/macros.h" -#include "main/mtypes.h" -#include "prog_statevars.h" -#include "prog_parameter.h" - - -/** - * Use the list of tokens in the state[] array to find global GL state - * and return it in . Usually, four values are returned in - * but matrix queries may return as many as 16 values. - * This function is used for ARB vertex/fragment programs. - * The program parser will produce the state[] values. - */ -static void -_mesa_fetch_state(struct gl_context *ctx, const gl_state_index state[], - GLfloat *value) -{ - switch (state[0]) { - case STATE_MATERIAL: - { - /* state[1] is either 0=front or 1=back side */ - const GLuint face = (GLuint) state[1]; - const struct gl_material *mat = &ctx->Light.Material; - ASSERT(face == 0 || face == 1); - /* we rely on tokens numbered so that _BACK_ == _FRONT_+ 1 */ - ASSERT(MAT_ATTRIB_FRONT_AMBIENT + 1 == MAT_ATTRIB_BACK_AMBIENT); - /* XXX we could get rid of this switch entirely with a little - * work in arbprogparse.c's parse_state_single_item(). - */ - /* state[2] is the material attribute */ - switch (state[2]) { - case STATE_AMBIENT: - COPY_4V(value, mat->Attrib[MAT_ATTRIB_FRONT_AMBIENT + face]); - return; - case STATE_DIFFUSE: - COPY_4V(value, mat->Attrib[MAT_ATTRIB_FRONT_DIFFUSE + face]); - return; - case STATE_SPECULAR: - COPY_4V(value, mat->Attrib[MAT_ATTRIB_FRONT_SPECULAR + face]); - return; - case STATE_EMISSION: - COPY_4V(value, mat->Attrib[MAT_ATTRIB_FRONT_EMISSION + face]); - return; - case STATE_SHININESS: - value[0] = mat->Attrib[MAT_ATTRIB_FRONT_SHININESS + face][0]; - value[1] = 0.0F; - value[2] = 0.0F; - value[3] = 1.0F; - return; - default: - _mesa_problem(ctx, "Invalid material state in fetch_state"); - return; - } - } - case STATE_LIGHT: - { - /* state[1] is the light number */ - const GLuint ln = (GLuint) state[1]; - /* state[2] is the light attribute */ - switch (state[2]) { - case STATE_AMBIENT: - COPY_4V(value, ctx->Light.Light[ln].Ambient); - return; - case STATE_DIFFUSE: - COPY_4V(value, ctx->Light.Light[ln].Diffuse); - return; - case STATE_SPECULAR: - COPY_4V(value, ctx->Light.Light[ln].Specular); - return; - case STATE_POSITION: - COPY_4V(value, ctx->Light.Light[ln].EyePosition); - return; - case STATE_ATTENUATION: - value[0] = ctx->Light.Light[ln].ConstantAttenuation; - value[1] = ctx->Light.Light[ln].LinearAttenuation; - value[2] = ctx->Light.Light[ln].QuadraticAttenuation; - value[3] = ctx->Light.Light[ln].SpotExponent; - return; - case STATE_SPOT_DIRECTION: - COPY_3V(value, ctx->Light.Light[ln].SpotDirection); - value[3] = ctx->Light.Light[ln]._CosCutoff; - return; - case STATE_SPOT_CUTOFF: - value[0] = ctx->Light.Light[ln].SpotCutoff; - return; - case STATE_HALF_VECTOR: - { - static const GLfloat eye_z[] = {0, 0, 1}; - GLfloat p[3]; - /* Compute infinite half angle vector: - * halfVector = normalize(normalize(lightPos) + (0, 0, 1)) - * light.EyePosition.w should be 0 for infinite lights. - */ - COPY_3V(p, ctx->Light.Light[ln].EyePosition); - NORMALIZE_3FV(p); - ADD_3V(value, p, eye_z); - NORMALIZE_3FV(value); - value[3] = 1.0; - } - return; - default: - _mesa_problem(ctx, "Invalid light state in fetch_state"); - return; - } - } - case STATE_LIGHTMODEL_AMBIENT: - COPY_4V(value, ctx->Light.Model.Ambient); - return; - case STATE_LIGHTMODEL_SCENECOLOR: - if (state[1] == 0) { - /* front */ - GLint i; - for (i = 0; i < 3; i++) { - value[i] = ctx->Light.Model.Ambient[i] - * ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_AMBIENT][i] - + ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_EMISSION][i]; - } - value[3] = ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_DIFFUSE][3]; - } - else { - /* back */ - GLint i; - for (i = 0; i < 3; i++) { - value[i] = ctx->Light.Model.Ambient[i] - * ctx->Light.Material.Attrib[MAT_ATTRIB_BACK_AMBIENT][i] - + ctx->Light.Material.Attrib[MAT_ATTRIB_BACK_EMISSION][i]; - } - value[3] = ctx->Light.Material.Attrib[MAT_ATTRIB_BACK_DIFFUSE][3]; - } - return; - case STATE_LIGHTPROD: - { - const GLuint ln = (GLuint) state[1]; - const GLuint face = (GLuint) state[2]; - GLint i; - ASSERT(face == 0 || face == 1); - switch (state[3]) { - case STATE_AMBIENT: - for (i = 0; i < 3; i++) { - value[i] = ctx->Light.Light[ln].Ambient[i] * - ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_AMBIENT+face][i]; - } - /* [3] = material alpha */ - value[3] = ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_AMBIENT+face][3]; - return; - case STATE_DIFFUSE: - for (i = 0; i < 3; i++) { - value[i] = ctx->Light.Light[ln].Diffuse[i] * - ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_DIFFUSE+face][i]; - } - /* [3] = material alpha */ - value[3] = ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_DIFFUSE+face][3]; - return; - case STATE_SPECULAR: - for (i = 0; i < 3; i++) { - value[i] = ctx->Light.Light[ln].Specular[i] * - ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_SPECULAR+face][i]; - } - /* [3] = material alpha */ - value[3] = ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_SPECULAR+face][3]; - return; - default: - _mesa_problem(ctx, "Invalid lightprod state in fetch_state"); - return; - } - } - case STATE_TEXGEN: - { - /* state[1] is the texture unit */ - const GLuint unit = (GLuint) state[1]; - /* state[2] is the texgen attribute */ - switch (state[2]) { - case STATE_TEXGEN_EYE_S: - COPY_4V(value, ctx->Texture.Unit[unit].GenS.EyePlane); - return; - case STATE_TEXGEN_EYE_T: - COPY_4V(value, ctx->Texture.Unit[unit].GenT.EyePlane); - return; - case STATE_TEXGEN_EYE_R: - COPY_4V(value, ctx->Texture.Unit[unit].GenR.EyePlane); - return; - case STATE_TEXGEN_EYE_Q: - COPY_4V(value, ctx->Texture.Unit[unit].GenQ.EyePlane); - return; - case STATE_TEXGEN_OBJECT_S: - COPY_4V(value, ctx->Texture.Unit[unit].GenS.ObjectPlane); - return; - case STATE_TEXGEN_OBJECT_T: - COPY_4V(value, ctx->Texture.Unit[unit].GenT.ObjectPlane); - return; - case STATE_TEXGEN_OBJECT_R: - COPY_4V(value, ctx->Texture.Unit[unit].GenR.ObjectPlane); - return; - case STATE_TEXGEN_OBJECT_Q: - COPY_4V(value, ctx->Texture.Unit[unit].GenQ.ObjectPlane); - return; - default: - _mesa_problem(ctx, "Invalid texgen state in fetch_state"); - return; - } - } - case STATE_TEXENV_COLOR: - { - /* state[1] is the texture unit */ - const GLuint unit = (GLuint) state[1]; - if(ctx->Color._ClampFragmentColor) - COPY_4V(value, ctx->Texture.Unit[unit].EnvColor); - else - COPY_4V(value, ctx->Texture.Unit[unit].EnvColorUnclamped); - } - return; - case STATE_FOG_COLOR: - if(ctx->Color._ClampFragmentColor) - COPY_4V(value, ctx->Fog.Color); - else - COPY_4V(value, ctx->Fog.ColorUnclamped); - return; - case STATE_FOG_PARAMS: - value[0] = ctx->Fog.Density; - value[1] = ctx->Fog.Start; - value[2] = ctx->Fog.End; - value[3] = (ctx->Fog.End == ctx->Fog.Start) - ? 1.0f : (GLfloat)(1.0 / (ctx->Fog.End - ctx->Fog.Start)); - return; - case STATE_CLIPPLANE: - { - const GLuint plane = (GLuint) state[1]; - COPY_4V(value, ctx->Transform.EyeUserPlane[plane]); - } - return; - case STATE_POINT_SIZE: - value[0] = ctx->Point.Size; - value[1] = ctx->Point.MinSize; - value[2] = ctx->Point.MaxSize; - value[3] = ctx->Point.Threshold; - return; - case STATE_POINT_ATTENUATION: - value[0] = ctx->Point.Params[0]; - value[1] = ctx->Point.Params[1]; - value[2] = ctx->Point.Params[2]; - value[3] = 1.0F; - return; - case STATE_MODELVIEW_MATRIX: - case STATE_PROJECTION_MATRIX: - case STATE_MVP_MATRIX: - case STATE_TEXTURE_MATRIX: - case STATE_PROGRAM_MATRIX: - { - /* state[0] = modelview, projection, texture, etc. */ - /* state[1] = which texture matrix or program matrix */ - /* state[2] = first row to fetch */ - /* state[3] = last row to fetch */ - /* state[4] = transpose, inverse or invtrans */ - const GLmatrix *matrix; - const gl_state_index mat = state[0]; - const GLuint index = (GLuint) state[1]; - const GLuint firstRow = (GLuint) state[2]; - const GLuint lastRow = (GLuint) state[3]; - const gl_state_index modifier = state[4]; - const GLfloat *m; - GLuint row, i; - ASSERT(firstRow >= 0); - ASSERT(firstRow < 4); - ASSERT(lastRow >= 0); - ASSERT(lastRow < 4); - if (mat == STATE_MODELVIEW_MATRIX) { - matrix = ctx->ModelviewMatrixStack.Top; - } - else if (mat == STATE_PROJECTION_MATRIX) { - matrix = ctx->ProjectionMatrixStack.Top; - } - else if (mat == STATE_MVP_MATRIX) { - matrix = &ctx->_ModelProjectMatrix; - } - else if (mat == STATE_TEXTURE_MATRIX) { - ASSERT(index < Elements(ctx->TextureMatrixStack)); - matrix = ctx->TextureMatrixStack[index].Top; - } - else if (mat == STATE_PROGRAM_MATRIX) { - ASSERT(index < Elements(ctx->ProgramMatrixStack)); - matrix = ctx->ProgramMatrixStack[index].Top; - } - else { - _mesa_problem(ctx, "Bad matrix name in _mesa_fetch_state()"); - return; - } - if (modifier == STATE_MATRIX_INVERSE || - modifier == STATE_MATRIX_INVTRANS) { - /* Be sure inverse is up to date: - */ - _math_matrix_alloc_inv( (GLmatrix *) matrix ); - _math_matrix_analyse( (GLmatrix*) matrix ); - m = matrix->inv; - } - else { - m = matrix->m; - } - if (modifier == STATE_MATRIX_TRANSPOSE || - modifier == STATE_MATRIX_INVTRANS) { - for (i = 0, row = firstRow; row <= lastRow; row++) { - value[i++] = m[row * 4 + 0]; - value[i++] = m[row * 4 + 1]; - value[i++] = m[row * 4 + 2]; - value[i++] = m[row * 4 + 3]; - } - } - else { - for (i = 0, row = firstRow; row <= lastRow; row++) { - value[i++] = m[row + 0]; - value[i++] = m[row + 4]; - value[i++] = m[row + 8]; - value[i++] = m[row + 12]; - } - } - } - return; - case STATE_DEPTH_RANGE: - value[0] = ctx->Viewport.Near; /* near */ - value[1] = ctx->Viewport.Far; /* far */ - value[2] = ctx->Viewport.Far - ctx->Viewport.Near; /* far - near */ - value[3] = 1.0; - return; - case STATE_FRAGMENT_PROGRAM: - { - /* state[1] = {STATE_ENV, STATE_LOCAL} */ - /* state[2] = parameter index */ - const int idx = (int) state[2]; - switch (state[1]) { - case STATE_ENV: - COPY_4V(value, ctx->FragmentProgram.Parameters[idx]); - return; - case STATE_LOCAL: - COPY_4V(value, ctx->FragmentProgram.Current->Base.LocalParams[idx]); - return; - default: - _mesa_problem(ctx, "Bad state switch in _mesa_fetch_state()"); - return; - } - } - return; - - case STATE_VERTEX_PROGRAM: - { - /* state[1] = {STATE_ENV, STATE_LOCAL} */ - /* state[2] = parameter index */ - const int idx = (int) state[2]; - switch (state[1]) { - case STATE_ENV: - COPY_4V(value, ctx->VertexProgram.Parameters[idx]); - return; - case STATE_LOCAL: - COPY_4V(value, ctx->VertexProgram.Current->Base.LocalParams[idx]); - return; - default: - _mesa_problem(ctx, "Bad state switch in _mesa_fetch_state()"); - return; - } - } - return; - - case STATE_NORMAL_SCALE: - ASSIGN_4V(value, ctx->_ModelViewInvScale, 0, 0, 1); - return; - - case STATE_INTERNAL: - switch (state[1]) { - case STATE_CURRENT_ATTRIB: - { - const GLuint idx = (GLuint) state[2]; - COPY_4V(value, ctx->Current.Attrib[idx]); - } - return; - - case STATE_CURRENT_ATTRIB_MAYBE_VP_CLAMPED: - { - const GLuint idx = (GLuint) state[2]; - if(ctx->Light._ClampVertexColor && - (idx == VERT_ATTRIB_COLOR0 || - idx == VERT_ATTRIB_COLOR1)) { - value[0] = CLAMP(ctx->Current.Attrib[idx][0], 0.0f, 1.0f); - value[1] = CLAMP(ctx->Current.Attrib[idx][1], 0.0f, 1.0f); - value[2] = CLAMP(ctx->Current.Attrib[idx][2], 0.0f, 1.0f); - value[3] = CLAMP(ctx->Current.Attrib[idx][3], 0.0f, 1.0f); - } - else - COPY_4V(value, ctx->Current.Attrib[idx]); - } - return; - - case STATE_NORMAL_SCALE: - ASSIGN_4V(value, - ctx->_ModelViewInvScale, - ctx->_ModelViewInvScale, - ctx->_ModelViewInvScale, - 1); - return; - - case STATE_TEXRECT_SCALE: - /* Value = { 1/texWidth, 1/texHeight, 0, 1 }. - * Used to convert unnormalized texcoords to normalized texcoords. - */ - { - const int unit = (int) state[2]; - const struct gl_texture_object *texObj - = ctx->Texture.Unit[unit]._Current; - if (texObj) { - struct gl_texture_image *texImage = texObj->Image[0][0]; - ASSIGN_4V(value, - (GLfloat) (1.0 / texImage->Width), - (GLfloat) (1.0 / texImage->Height), - 0.0f, 1.0f); - } - } - return; - - case STATE_FOG_PARAMS_OPTIMIZED: - /* for simpler per-vertex/pixel fog calcs. POW (for EXP/EXP2 fog) - * might be more expensive than EX2 on some hw, plus it needs - * another constant (e) anyway. Linear fog can now be done with a - * single MAD. - * linear: fogcoord * -1/(end-start) + end/(end-start) - * exp: 2^-(density/ln(2) * fogcoord) - * exp2: 2^-((density/(ln(2)^2) * fogcoord)^2) - */ - value[0] = (ctx->Fog.End == ctx->Fog.Start) - ? 1.0f : (GLfloat)(-1.0F / (ctx->Fog.End - ctx->Fog.Start)); - value[1] = ctx->Fog.End * -value[0]; - value[2] = (GLfloat)(ctx->Fog.Density * M_LOG2E); /* M_LOG2E == 1/ln(2) */ - value[3] = (GLfloat)(ctx->Fog.Density * ONE_DIV_SQRT_LN2); - return; - - case STATE_POINT_SIZE_CLAMPED: - { - /* this includes implementation dependent limits, to avoid - * another potentially necessary clamp. - * Note: for sprites, point smooth (point AA) is ignored - * and we'll clamp to MinPointSizeAA and MaxPointSize, because we - * expect drivers will want to say their minimum for AA size is 0.0 - * but for non-AA it's 1.0 (because normal points with size below 1.0 - * need to get rounded up to 1.0, hence never disappear). GL does - * not specify max clamp size for sprites, other than it needs to be - * at least as large as max AA size, hence use non-AA size there. - */ - GLfloat minImplSize; - GLfloat maxImplSize; - if (ctx->Point.PointSprite) { - minImplSize = ctx->Const.MinPointSizeAA; - maxImplSize = ctx->Const.MaxPointSize; - } - else if (ctx->Point.SmoothFlag || ctx->Multisample._Enabled) { - minImplSize = ctx->Const.MinPointSizeAA; - maxImplSize = ctx->Const.MaxPointSizeAA; - } - else { - minImplSize = ctx->Const.MinPointSize; - maxImplSize = ctx->Const.MaxPointSize; - } - value[0] = ctx->Point.Size; - value[1] = ctx->Point.MinSize >= minImplSize ? ctx->Point.MinSize : minImplSize; - value[2] = ctx->Point.MaxSize <= maxImplSize ? ctx->Point.MaxSize : maxImplSize; - value[3] = ctx->Point.Threshold; - } - return; - case STATE_POINT_SIZE_IMPL_CLAMP: - { - /* for implementation clamp only in vs */ - GLfloat minImplSize; - GLfloat maxImplSize; - if (ctx->Point.PointSprite) { - minImplSize = ctx->Const.MinPointSizeAA; - maxImplSize = ctx->Const.MaxPointSize; - } - else if (ctx->Point.SmoothFlag || ctx->Multisample._Enabled) { - minImplSize = ctx->Const.MinPointSizeAA; - maxImplSize = ctx->Const.MaxPointSizeAA; - } - else { - minImplSize = ctx->Const.MinPointSize; - maxImplSize = ctx->Const.MaxPointSize; - } - value[0] = ctx->Point.Size; - value[1] = minImplSize; - value[2] = maxImplSize; - value[3] = ctx->Point.Threshold; - } - return; - case STATE_LIGHT_SPOT_DIR_NORMALIZED: - { - /* here, state[2] is the light number */ - /* pre-normalize spot dir */ - const GLuint ln = (GLuint) state[2]; - COPY_3V(value, ctx->Light.Light[ln]._NormSpotDirection); - value[3] = ctx->Light.Light[ln]._CosCutoff; - } - return; - - case STATE_LIGHT_POSITION: - { - const GLuint ln = (GLuint) state[2]; - COPY_4V(value, ctx->Light.Light[ln]._Position); - } - return; - - case STATE_LIGHT_POSITION_NORMALIZED: - { - const GLuint ln = (GLuint) state[2]; - COPY_4V(value, ctx->Light.Light[ln]._Position); - NORMALIZE_3FV( value ); - } - return; - - case STATE_LIGHT_HALF_VECTOR: - { - const GLuint ln = (GLuint) state[2]; - GLfloat p[3]; - /* Compute infinite half angle vector: - * halfVector = normalize(normalize(lightPos) + (0, 0, 1)) - * light.EyePosition.w should be 0 for infinite lights. - */ - COPY_3V(p, ctx->Light.Light[ln]._Position); - NORMALIZE_3FV(p); - ADD_3V(value, p, ctx->_EyeZDir); - NORMALIZE_3FV(value); - value[3] = 1.0; - } - return; - - case STATE_PT_SCALE: - value[0] = ctx->Pixel.RedScale; - value[1] = ctx->Pixel.GreenScale; - value[2] = ctx->Pixel.BlueScale; - value[3] = ctx->Pixel.AlphaScale; - return; - - case STATE_PT_BIAS: - value[0] = ctx->Pixel.RedBias; - value[1] = ctx->Pixel.GreenBias; - value[2] = ctx->Pixel.BlueBias; - value[3] = ctx->Pixel.AlphaBias; - return; - - case STATE_SHADOW_AMBIENT: - { - const int unit = (int) state[2]; - const struct gl_texture_object *texObj - = ctx->Texture.Unit[unit]._Current; - if (texObj) { - value[0] = - value[1] = - value[2] = - value[3] = texObj->Sampler.CompareFailValue; - } - } - return; - - case STATE_FB_SIZE: - value[0] = (GLfloat) (ctx->DrawBuffer->Width - 1); - value[1] = (GLfloat) (ctx->DrawBuffer->Height - 1); - value[2] = 0.0F; - value[3] = 0.0F; - return; - - case STATE_FB_WPOS_Y_TRANSFORM: - /* A driver may negate this conditional by using ZW swizzle - * instead of XY (based on e.g. some other state). */ - if (ctx->DrawBuffer->Name != 0) { - /* Identity (XY) followed by flipping Y upside down (ZW). */ - value[0] = 1.0F; - value[1] = 0.0F; - value[2] = -1.0F; - value[3] = (GLfloat) ctx->DrawBuffer->Height; - } else { - /* Flipping Y upside down (XY) followed by identity (ZW). */ - value[0] = -1.0F; - value[1] = (GLfloat) ctx->DrawBuffer->Height; - value[2] = 1.0F; - value[3] = 0.0F; - } - return; - - case STATE_ROT_MATRIX_0: - { - const int unit = (int) state[2]; - GLfloat *rotMat22 = ctx->Texture.Unit[unit].RotMatrix; - value[0] = rotMat22[0]; - value[1] = rotMat22[2]; - value[2] = 0.0; - value[3] = 0.0; - } - return; - - case STATE_ROT_MATRIX_1: - { - const int unit = (int) state[2]; - GLfloat *rotMat22 = ctx->Texture.Unit[unit].RotMatrix; - value[0] = rotMat22[1]; - value[1] = rotMat22[3]; - value[2] = 0.0; - value[3] = 0.0; - } - return; - - /* XXX: make sure new tokens added here are also handled in the - * _mesa_program_state_flags() switch, below. - */ - default: - /* Unknown state indexes are silently ignored here. - * Drivers may do something special. - */ - return; - } - return; - - default: - _mesa_problem(ctx, "Invalid state in _mesa_fetch_state"); - return; - } -} - - -/** - * Return a bitmask of the Mesa state flags (_NEW_* values) which would - * indicate that the given context state may have changed. - * The bitmask is used during validation to determine if we need to update - * vertex/fragment program parameters (like "state.material.color") when - * some GL state has changed. - */ -GLbitfield -_mesa_program_state_flags(const gl_state_index state[STATE_LENGTH]) -{ - switch (state[0]) { - case STATE_MATERIAL: - case STATE_LIGHT: - case STATE_LIGHTMODEL_AMBIENT: - case STATE_LIGHTMODEL_SCENECOLOR: - case STATE_LIGHTPROD: - return _NEW_LIGHT; - - case STATE_TEXGEN: - return _NEW_TEXTURE; - case STATE_TEXENV_COLOR: - return _NEW_TEXTURE | _NEW_BUFFERS | _NEW_FRAG_CLAMP; - - case STATE_FOG_COLOR: - return _NEW_FOG | _NEW_BUFFERS | _NEW_FRAG_CLAMP; - case STATE_FOG_PARAMS: - return _NEW_FOG; - - case STATE_CLIPPLANE: - return _NEW_TRANSFORM; - - case STATE_POINT_SIZE: - case STATE_POINT_ATTENUATION: - return _NEW_POINT; - - case STATE_MODELVIEW_MATRIX: - return _NEW_MODELVIEW; - case STATE_PROJECTION_MATRIX: - return _NEW_PROJECTION; - case STATE_MVP_MATRIX: - return _NEW_MODELVIEW | _NEW_PROJECTION; - case STATE_TEXTURE_MATRIX: - return _NEW_TEXTURE_MATRIX; - case STATE_PROGRAM_MATRIX: - return _NEW_TRACK_MATRIX; - - case STATE_DEPTH_RANGE: - return _NEW_VIEWPORT; - - case STATE_FRAGMENT_PROGRAM: - case STATE_VERTEX_PROGRAM: - return _NEW_PROGRAM; - - case STATE_NORMAL_SCALE: - return _NEW_MODELVIEW; - - case STATE_INTERNAL: - switch (state[1]) { - case STATE_CURRENT_ATTRIB: - return _NEW_CURRENT_ATTRIB; - case STATE_CURRENT_ATTRIB_MAYBE_VP_CLAMPED: - return _NEW_CURRENT_ATTRIB | _NEW_LIGHT | _NEW_BUFFERS; - - case STATE_NORMAL_SCALE: - return _NEW_MODELVIEW; - - case STATE_TEXRECT_SCALE: - case STATE_SHADOW_AMBIENT: - case STATE_ROT_MATRIX_0: - case STATE_ROT_MATRIX_1: - return _NEW_TEXTURE; - case STATE_FOG_PARAMS_OPTIMIZED: - return _NEW_FOG; - case STATE_POINT_SIZE_CLAMPED: - case STATE_POINT_SIZE_IMPL_CLAMP: - return _NEW_POINT | _NEW_MULTISAMPLE; - case STATE_LIGHT_SPOT_DIR_NORMALIZED: - case STATE_LIGHT_POSITION: - case STATE_LIGHT_POSITION_NORMALIZED: - case STATE_LIGHT_HALF_VECTOR: - return _NEW_LIGHT; - - case STATE_PT_SCALE: - case STATE_PT_BIAS: - return _NEW_PIXEL; - - case STATE_FB_SIZE: - case STATE_FB_WPOS_Y_TRANSFORM: - return _NEW_BUFFERS; - - default: - /* unknown state indexes are silently ignored and - * no flag set, since it is handled by the driver. - */ - return 0; - } - - default: - _mesa_problem(NULL, "unexpected state[0] in make_state_flags()"); - return 0; - } -} - - -static void -append(char *dst, const char *src) -{ - while (*dst) - dst++; - while (*src) - *dst++ = *src++; - *dst = 0; -} - - -/** - * Convert token 'k' to a string, append it onto 'dst' string. - */ -static void -append_token(char *dst, gl_state_index k) -{ - switch (k) { - case STATE_MATERIAL: - append(dst, "material"); - break; - case STATE_LIGHT: - append(dst, "light"); - break; - case STATE_LIGHTMODEL_AMBIENT: - append(dst, "lightmodel.ambient"); - break; - case STATE_LIGHTMODEL_SCENECOLOR: - break; - case STATE_LIGHTPROD: - append(dst, "lightprod"); - break; - case STATE_TEXGEN: - append(dst, "texgen"); - break; - case STATE_FOG_COLOR: - append(dst, "fog.color"); - break; - case STATE_FOG_PARAMS: - append(dst, "fog.params"); - break; - case STATE_CLIPPLANE: - append(dst, "clip"); - break; - case STATE_POINT_SIZE: - append(dst, "point.size"); - break; - case STATE_POINT_ATTENUATION: - append(dst, "point.attenuation"); - break; - case STATE_MODELVIEW_MATRIX: - append(dst, "matrix.modelview"); - break; - case STATE_PROJECTION_MATRIX: - append(dst, "matrix.projection"); - break; - case STATE_MVP_MATRIX: - append(dst, "matrix.mvp"); - break; - case STATE_TEXTURE_MATRIX: - append(dst, "matrix.texture"); - break; - case STATE_PROGRAM_MATRIX: - append(dst, "matrix.program"); - break; - case STATE_MATRIX_INVERSE: - append(dst, ".inverse"); - break; - case STATE_MATRIX_TRANSPOSE: - append(dst, ".transpose"); - break; - case STATE_MATRIX_INVTRANS: - append(dst, ".invtrans"); - break; - case STATE_AMBIENT: - append(dst, ".ambient"); - break; - case STATE_DIFFUSE: - append(dst, ".diffuse"); - break; - case STATE_SPECULAR: - append(dst, ".specular"); - break; - case STATE_EMISSION: - append(dst, ".emission"); - break; - case STATE_SHININESS: - append(dst, "lshininess"); - break; - case STATE_HALF_VECTOR: - append(dst, ".half"); - break; - case STATE_POSITION: - append(dst, ".position"); - break; - case STATE_ATTENUATION: - append(dst, ".attenuation"); - break; - case STATE_SPOT_DIRECTION: - append(dst, ".spot.direction"); - break; - case STATE_SPOT_CUTOFF: - append(dst, ".spot.cutoff"); - break; - case STATE_TEXGEN_EYE_S: - append(dst, ".eye.s"); - break; - case STATE_TEXGEN_EYE_T: - append(dst, ".eye.t"); - break; - case STATE_TEXGEN_EYE_R: - append(dst, ".eye.r"); - break; - case STATE_TEXGEN_EYE_Q: - append(dst, ".eye.q"); - break; - case STATE_TEXGEN_OBJECT_S: - append(dst, ".object.s"); - break; - case STATE_TEXGEN_OBJECT_T: - append(dst, ".object.t"); - break; - case STATE_TEXGEN_OBJECT_R: - append(dst, ".object.r"); - break; - case STATE_TEXGEN_OBJECT_Q: - append(dst, ".object.q"); - break; - case STATE_TEXENV_COLOR: - append(dst, "texenv"); - break; - case STATE_DEPTH_RANGE: - append(dst, "depth.range"); - break; - case STATE_VERTEX_PROGRAM: - case STATE_FRAGMENT_PROGRAM: - break; - case STATE_ENV: - append(dst, "env"); - break; - case STATE_LOCAL: - append(dst, "local"); - break; - /* BEGIN internal state vars */ - case STATE_INTERNAL: - append(dst, ".internal."); - break; - case STATE_CURRENT_ATTRIB: - append(dst, "current"); - break; - case STATE_NORMAL_SCALE: - append(dst, "normalScale"); - break; - case STATE_TEXRECT_SCALE: - append(dst, "texrectScale"); - break; - case STATE_FOG_PARAMS_OPTIMIZED: - append(dst, "fogParamsOptimized"); - break; - case STATE_POINT_SIZE_CLAMPED: - append(dst, "pointSizeClamped"); - break; - case STATE_POINT_SIZE_IMPL_CLAMP: - append(dst, "pointSizeImplClamp"); - break; - case STATE_LIGHT_SPOT_DIR_NORMALIZED: - append(dst, "lightSpotDirNormalized"); - break; - case STATE_LIGHT_POSITION: - append(dst, "lightPosition"); - break; - case STATE_LIGHT_POSITION_NORMALIZED: - append(dst, "light.position.normalized"); - break; - case STATE_LIGHT_HALF_VECTOR: - append(dst, "lightHalfVector"); - break; - case STATE_PT_SCALE: - append(dst, "PTscale"); - break; - case STATE_PT_BIAS: - append(dst, "PTbias"); - break; - case STATE_SHADOW_AMBIENT: - append(dst, "CompareFailValue"); - break; - case STATE_FB_SIZE: - append(dst, "FbSize"); - break; - case STATE_FB_WPOS_Y_TRANSFORM: - append(dst, "FbWposYTransform"); - break; - case STATE_ROT_MATRIX_0: - append(dst, "rotMatrixRow0"); - break; - case STATE_ROT_MATRIX_1: - append(dst, "rotMatrixRow1"); - break; - default: - /* probably STATE_INTERNAL_DRIVER+i (driver private state) */ - append(dst, "driverState"); - } -} - -static void -append_face(char *dst, GLint face) -{ - if (face == 0) - append(dst, "front."); - else - append(dst, "back."); -} - -static void -append_index(char *dst, GLint index) -{ - char s[20]; - sprintf(s, "[%d]", index); - append(dst, s); -} - -/** - * Make a string from the given state vector. - * For example, return "state.matrix.texture[2].inverse". - * Use free() to deallocate the string. - */ -char * -_mesa_program_state_string(const gl_state_index state[STATE_LENGTH]) -{ - char str[1000] = ""; - char tmp[30]; - - append(str, "state."); - append_token(str, state[0]); - - switch (state[0]) { - case STATE_MATERIAL: - append_face(str, state[1]); - append_token(str, state[2]); - break; - case STATE_LIGHT: - append_index(str, state[1]); /* light number [i]. */ - append_token(str, state[2]); /* coefficients */ - break; - case STATE_LIGHTMODEL_AMBIENT: - append(str, "lightmodel.ambient"); - break; - case STATE_LIGHTMODEL_SCENECOLOR: - if (state[1] == 0) { - append(str, "lightmodel.front.scenecolor"); - } - else { - append(str, "lightmodel.back.scenecolor"); - } - break; - case STATE_LIGHTPROD: - append_index(str, state[1]); /* light number [i]. */ - append_face(str, state[2]); - append_token(str, state[3]); - break; - case STATE_TEXGEN: - append_index(str, state[1]); /* tex unit [i] */ - append_token(str, state[2]); /* plane coef */ - break; - case STATE_TEXENV_COLOR: - append_index(str, state[1]); /* tex unit [i] */ - append(str, "color"); - break; - case STATE_CLIPPLANE: - append_index(str, state[1]); /* plane [i] */ - append(str, ".plane"); - break; - case STATE_MODELVIEW_MATRIX: - case STATE_PROJECTION_MATRIX: - case STATE_MVP_MATRIX: - case STATE_TEXTURE_MATRIX: - case STATE_PROGRAM_MATRIX: - { - /* state[0] = modelview, projection, texture, etc. */ - /* state[1] = which texture matrix or program matrix */ - /* state[2] = first row to fetch */ - /* state[3] = last row to fetch */ - /* state[4] = transpose, inverse or invtrans */ - const gl_state_index mat = state[0]; - const GLuint index = (GLuint) state[1]; - const GLuint firstRow = (GLuint) state[2]; - const GLuint lastRow = (GLuint) state[3]; - const gl_state_index modifier = state[4]; - if (index || - mat == STATE_TEXTURE_MATRIX || - mat == STATE_PROGRAM_MATRIX) - append_index(str, index); - if (modifier) - append_token(str, modifier); - if (firstRow == lastRow) - sprintf(tmp, ".row[%d]", firstRow); - else - sprintf(tmp, ".row[%d..%d]", firstRow, lastRow); - append(str, tmp); - } - break; - case STATE_POINT_SIZE: - break; - case STATE_POINT_ATTENUATION: - break; - case STATE_FOG_PARAMS: - break; - case STATE_FOG_COLOR: - break; - case STATE_DEPTH_RANGE: - break; - case STATE_FRAGMENT_PROGRAM: - case STATE_VERTEX_PROGRAM: - /* state[1] = {STATE_ENV, STATE_LOCAL} */ - /* state[2] = parameter index */ - append_token(str, state[1]); - append_index(str, state[2]); - break; - case STATE_NORMAL_SCALE: - break; - case STATE_INTERNAL: - append_token(str, state[1]); - if (state[1] == STATE_CURRENT_ATTRIB) - append_index(str, state[2]); - break; - default: - _mesa_problem(NULL, "Invalid state in _mesa_program_state_string"); - break; - } - - return _mesa_strdup(str); -} - - -/** - * Loop over all the parameters in a parameter list. If the parameter - * is a GL state reference, look up the current value of that state - * variable and put it into the parameter's Value[4] array. - * Other parameter types never change or are explicitly set by the user - * with glUniform() or glProgramParameter(), etc. - * This would be called at glBegin time. - */ -void -_mesa_load_state_parameters(struct gl_context *ctx, - struct gl_program_parameter_list *paramList) -{ - GLuint i; - - if (!paramList) - return; - - for (i = 0; i < paramList->NumParameters; i++) { - if (paramList->Parameters[i].Type == PROGRAM_STATE_VAR) { - _mesa_fetch_state(ctx, - paramList->Parameters[i].StateIndexes, - ¶mList->ParameterValues[i][0].f); - } - } -} - - -/** - * Copy the 16 elements of a matrix into four consecutive program - * registers starting at 'pos'. - */ -static void -load_matrix(GLfloat registers[][4], GLuint pos, const GLfloat mat[16]) -{ - GLuint i; - for (i = 0; i < 4; i++) { - registers[pos + i][0] = mat[0 + i]; - registers[pos + i][1] = mat[4 + i]; - registers[pos + i][2] = mat[8 + i]; - registers[pos + i][3] = mat[12 + i]; - } -} - - -/** - * As above, but transpose the matrix. - */ -static void -load_transpose_matrix(GLfloat registers[][4], GLuint pos, - const GLfloat mat[16]) -{ - memcpy(registers[pos], mat, 16 * sizeof(GLfloat)); -} - - -/** - * Load current vertex program's parameter registers with tracked - * matrices (if NV program). This only needs to be done per - * glBegin/glEnd, not per-vertex. - */ -void -_mesa_load_tracked_matrices(struct gl_context *ctx) -{ - GLuint i; - - for (i = 0; i < MAX_NV_VERTEX_PROGRAM_PARAMS / 4; i++) { - /* point 'mat' at source matrix */ - GLmatrix *mat; - if (ctx->VertexProgram.TrackMatrix[i] == GL_MODELVIEW) { - mat = ctx->ModelviewMatrixStack.Top; - } - else if (ctx->VertexProgram.TrackMatrix[i] == GL_PROJECTION) { - mat = ctx->ProjectionMatrixStack.Top; - } - else if (ctx->VertexProgram.TrackMatrix[i] == GL_TEXTURE) { - GLuint unit = MIN2(ctx->Texture.CurrentUnit, - Elements(ctx->TextureMatrixStack) - 1); - mat = ctx->TextureMatrixStack[unit].Top; - } - else if (ctx->VertexProgram.TrackMatrix[i]==GL_MODELVIEW_PROJECTION_NV) { - /* XXX verify the combined matrix is up to date */ - mat = &ctx->_ModelProjectMatrix; - } - else if (ctx->VertexProgram.TrackMatrix[i] >= GL_MATRIX0_NV && - ctx->VertexProgram.TrackMatrix[i] <= GL_MATRIX7_NV) { - GLuint n = ctx->VertexProgram.TrackMatrix[i] - GL_MATRIX0_NV; - ASSERT(n < Elements(ctx->ProgramMatrixStack)); - mat = ctx->ProgramMatrixStack[n].Top; - } - else { - /* no matrix is tracked, but we leave the register values as-is */ - assert(ctx->VertexProgram.TrackMatrix[i] == GL_NONE); - continue; - } - - /* load the matrix values into sequential registers */ - if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_IDENTITY_NV) { - load_matrix(ctx->VertexProgram.Parameters, i*4, mat->m); - } - else if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_INVERSE_NV) { - _math_matrix_analyse(mat); /* update the inverse */ - ASSERT(!_math_matrix_is_dirty(mat)); - load_matrix(ctx->VertexProgram.Parameters, i*4, mat->inv); - } - else if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_TRANSPOSE_NV) { - load_transpose_matrix(ctx->VertexProgram.Parameters, i*4, mat->m); - } - else { - assert(ctx->VertexProgram.TrackMatrixTransform[i] - == GL_INVERSE_TRANSPOSE_NV); - _math_matrix_analyse(mat); /* update the inverse */ - ASSERT(!_math_matrix_is_dirty(mat)); - load_transpose_matrix(ctx->VertexProgram.Parameters, i*4, mat->inv); - } - } -} +/* + * Mesa 3-D graphics library + * Version: 7.1 + * + * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/** + * \file prog_statevars.c + * Program state variable management. + * \author Brian Paul + */ + + +#include "main/glheader.h" +#include "main/context.h" +#include "main/imports.h" +#include "main/macros.h" +#include "main/mtypes.h" +#include "prog_statevars.h" +#include "prog_parameter.h" + + +/** + * Use the list of tokens in the state[] array to find global GL state + * and return it in . Usually, four values are returned in + * but matrix queries may return as many as 16 values. + * This function is used for ARB vertex/fragment programs. + * The program parser will produce the state[] values. + */ +static void +_mesa_fetch_state(struct gl_context *ctx, const gl_state_index state[], + GLfloat *value) +{ + switch (state[0]) { + case STATE_MATERIAL: + { + /* state[1] is either 0=front or 1=back side */ + const GLuint face = (GLuint) state[1]; + const struct gl_material *mat = &ctx->Light.Material; + ASSERT(face == 0 || face == 1); + /* we rely on tokens numbered so that _BACK_ == _FRONT_+ 1 */ + ASSERT(MAT_ATTRIB_FRONT_AMBIENT + 1 == MAT_ATTRIB_BACK_AMBIENT); + /* XXX we could get rid of this switch entirely with a little + * work in arbprogparse.c's parse_state_single_item(). + */ + /* state[2] is the material attribute */ + switch (state[2]) { + case STATE_AMBIENT: + COPY_4V(value, mat->Attrib[MAT_ATTRIB_FRONT_AMBIENT + face]); + return; + case STATE_DIFFUSE: + COPY_4V(value, mat->Attrib[MAT_ATTRIB_FRONT_DIFFUSE + face]); + return; + case STATE_SPECULAR: + COPY_4V(value, mat->Attrib[MAT_ATTRIB_FRONT_SPECULAR + face]); + return; + case STATE_EMISSION: + COPY_4V(value, mat->Attrib[MAT_ATTRIB_FRONT_EMISSION + face]); + return; + case STATE_SHININESS: + value[0] = mat->Attrib[MAT_ATTRIB_FRONT_SHININESS + face][0]; + value[1] = 0.0F; + value[2] = 0.0F; + value[3] = 1.0F; + return; + default: + _mesa_problem(ctx, "Invalid material state in fetch_state"); + return; + } + } + case STATE_LIGHT: + { + /* state[1] is the light number */ + const GLuint ln = (GLuint) state[1]; + /* state[2] is the light attribute */ + switch (state[2]) { + case STATE_AMBIENT: + COPY_4V(value, ctx->Light.Light[ln].Ambient); + return; + case STATE_DIFFUSE: + COPY_4V(value, ctx->Light.Light[ln].Diffuse); + return; + case STATE_SPECULAR: + COPY_4V(value, ctx->Light.Light[ln].Specular); + return; + case STATE_POSITION: + COPY_4V(value, ctx->Light.Light[ln].EyePosition); + return; + case STATE_ATTENUATION: + value[0] = ctx->Light.Light[ln].ConstantAttenuation; + value[1] = ctx->Light.Light[ln].LinearAttenuation; + value[2] = ctx->Light.Light[ln].QuadraticAttenuation; + value[3] = ctx->Light.Light[ln].SpotExponent; + return; + case STATE_SPOT_DIRECTION: + COPY_3V(value, ctx->Light.Light[ln].SpotDirection); + value[3] = ctx->Light.Light[ln]._CosCutoff; + return; + case STATE_SPOT_CUTOFF: + value[0] = ctx->Light.Light[ln].SpotCutoff; + return; + case STATE_HALF_VECTOR: + { + static const GLfloat eye_z[] = {0, 0, 1}; + GLfloat p[3]; + /* Compute infinite half angle vector: + * halfVector = normalize(normalize(lightPos) + (0, 0, 1)) + * light.EyePosition.w should be 0 for infinite lights. + */ + COPY_3V(p, ctx->Light.Light[ln].EyePosition); + NORMALIZE_3FV(p); + ADD_3V(value, p, eye_z); + NORMALIZE_3FV(value); + value[3] = 1.0; + } + return; + default: + _mesa_problem(ctx, "Invalid light state in fetch_state"); + return; + } + } + case STATE_LIGHTMODEL_AMBIENT: + COPY_4V(value, ctx->Light.Model.Ambient); + return; + case STATE_LIGHTMODEL_SCENECOLOR: + if (state[1] == 0) { + /* front */ + GLint i; + for (i = 0; i < 3; i++) { + value[i] = ctx->Light.Model.Ambient[i] + * ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_AMBIENT][i] + + ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_EMISSION][i]; + } + value[3] = ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_DIFFUSE][3]; + } + else { + /* back */ + GLint i; + for (i = 0; i < 3; i++) { + value[i] = ctx->Light.Model.Ambient[i] + * ctx->Light.Material.Attrib[MAT_ATTRIB_BACK_AMBIENT][i] + + ctx->Light.Material.Attrib[MAT_ATTRIB_BACK_EMISSION][i]; + } + value[3] = ctx->Light.Material.Attrib[MAT_ATTRIB_BACK_DIFFUSE][3]; + } + return; + case STATE_LIGHTPROD: + { + const GLuint ln = (GLuint) state[1]; + const GLuint face = (GLuint) state[2]; + GLint i; + ASSERT(face == 0 || face == 1); + switch (state[3]) { + case STATE_AMBIENT: + for (i = 0; i < 3; i++) { + value[i] = ctx->Light.Light[ln].Ambient[i] * + ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_AMBIENT+face][i]; + } + /* [3] = material alpha */ + value[3] = ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_AMBIENT+face][3]; + return; + case STATE_DIFFUSE: + for (i = 0; i < 3; i++) { + value[i] = ctx->Light.Light[ln].Diffuse[i] * + ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_DIFFUSE+face][i]; + } + /* [3] = material alpha */ + value[3] = ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_DIFFUSE+face][3]; + return; + case STATE_SPECULAR: + for (i = 0; i < 3; i++) { + value[i] = ctx->Light.Light[ln].Specular[i] * + ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_SPECULAR+face][i]; + } + /* [3] = material alpha */ + value[3] = ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_SPECULAR+face][3]; + return; + default: + _mesa_problem(ctx, "Invalid lightprod state in fetch_state"); + return; + } + } + case STATE_TEXGEN: + { + /* state[1] is the texture unit */ + const GLuint unit = (GLuint) state[1]; + /* state[2] is the texgen attribute */ + switch (state[2]) { + case STATE_TEXGEN_EYE_S: + COPY_4V(value, ctx->Texture.Unit[unit].GenS.EyePlane); + return; + case STATE_TEXGEN_EYE_T: + COPY_4V(value, ctx->Texture.Unit[unit].GenT.EyePlane); + return; + case STATE_TEXGEN_EYE_R: + COPY_4V(value, ctx->Texture.Unit[unit].GenR.EyePlane); + return; + case STATE_TEXGEN_EYE_Q: + COPY_4V(value, ctx->Texture.Unit[unit].GenQ.EyePlane); + return; + case STATE_TEXGEN_OBJECT_S: + COPY_4V(value, ctx->Texture.Unit[unit].GenS.ObjectPlane); + return; + case STATE_TEXGEN_OBJECT_T: + COPY_4V(value, ctx->Texture.Unit[unit].GenT.ObjectPlane); + return; + case STATE_TEXGEN_OBJECT_R: + COPY_4V(value, ctx->Texture.Unit[unit].GenR.ObjectPlane); + return; + case STATE_TEXGEN_OBJECT_Q: + COPY_4V(value, ctx->Texture.Unit[unit].GenQ.ObjectPlane); + return; + default: + _mesa_problem(ctx, "Invalid texgen state in fetch_state"); + return; + } + } + case STATE_TEXENV_COLOR: + { + /* state[1] is the texture unit */ + const GLuint unit = (GLuint) state[1]; + if(ctx->Color._ClampFragmentColor) + COPY_4V(value, ctx->Texture.Unit[unit].EnvColor); + else + COPY_4V(value, ctx->Texture.Unit[unit].EnvColorUnclamped); + } + return; + case STATE_FOG_COLOR: + if(ctx->Color._ClampFragmentColor) + COPY_4V(value, ctx->Fog.Color); + else + COPY_4V(value, ctx->Fog.ColorUnclamped); + return; + case STATE_FOG_PARAMS: + value[0] = ctx->Fog.Density; + value[1] = ctx->Fog.Start; + value[2] = ctx->Fog.End; + value[3] = (ctx->Fog.End == ctx->Fog.Start) + ? 1.0f : (GLfloat)(1.0 / (ctx->Fog.End - ctx->Fog.Start)); + return; + case STATE_CLIPPLANE: + { + const GLuint plane = (GLuint) state[1]; + COPY_4V(value, ctx->Transform.EyeUserPlane[plane]); + } + return; + case STATE_POINT_SIZE: + value[0] = ctx->Point.Size; + value[1] = ctx->Point.MinSize; + value[2] = ctx->Point.MaxSize; + value[3] = ctx->Point.Threshold; + return; + case STATE_POINT_ATTENUATION: + value[0] = ctx->Point.Params[0]; + value[1] = ctx->Point.Params[1]; + value[2] = ctx->Point.Params[2]; + value[3] = 1.0F; + return; + case STATE_MODELVIEW_MATRIX: + case STATE_PROJECTION_MATRIX: + case STATE_MVP_MATRIX: + case STATE_TEXTURE_MATRIX: + case STATE_PROGRAM_MATRIX: + { + /* state[0] = modelview, projection, texture, etc. */ + /* state[1] = which texture matrix or program matrix */ + /* state[2] = first row to fetch */ + /* state[3] = last row to fetch */ + /* state[4] = transpose, inverse or invtrans */ + const GLmatrix *matrix; + const gl_state_index mat = state[0]; + const GLuint index = (GLuint) state[1]; + const GLuint firstRow = (GLuint) state[2]; + const GLuint lastRow = (GLuint) state[3]; + const gl_state_index modifier = state[4]; + const GLfloat *m; + GLuint row, i; + ASSERT(firstRow >= 0); + ASSERT(firstRow < 4); + ASSERT(lastRow >= 0); + ASSERT(lastRow < 4); + if (mat == STATE_MODELVIEW_MATRIX) { + matrix = ctx->ModelviewMatrixStack.Top; + } + else if (mat == STATE_PROJECTION_MATRIX) { + matrix = ctx->ProjectionMatrixStack.Top; + } + else if (mat == STATE_MVP_MATRIX) { + matrix = &ctx->_ModelProjectMatrix; + } + else if (mat == STATE_TEXTURE_MATRIX) { + ASSERT(index < Elements(ctx->TextureMatrixStack)); + matrix = ctx->TextureMatrixStack[index].Top; + } + else if (mat == STATE_PROGRAM_MATRIX) { + ASSERT(index < Elements(ctx->ProgramMatrixStack)); + matrix = ctx->ProgramMatrixStack[index].Top; + } + else { + _mesa_problem(ctx, "Bad matrix name in _mesa_fetch_state()"); + return; + } + if (modifier == STATE_MATRIX_INVERSE || + modifier == STATE_MATRIX_INVTRANS) { + /* Be sure inverse is up to date: + */ + _math_matrix_alloc_inv( (GLmatrix *) matrix ); + _math_matrix_analyse( (GLmatrix*) matrix ); + m = matrix->inv; + } + else { + m = matrix->m; + } + if (modifier == STATE_MATRIX_TRANSPOSE || + modifier == STATE_MATRIX_INVTRANS) { + for (i = 0, row = firstRow; row <= lastRow; row++) { + value[i++] = m[row * 4 + 0]; + value[i++] = m[row * 4 + 1]; + value[i++] = m[row * 4 + 2]; + value[i++] = m[row * 4 + 3]; + } + } + else { + for (i = 0, row = firstRow; row <= lastRow; row++) { + value[i++] = m[row + 0]; + value[i++] = m[row + 4]; + value[i++] = m[row + 8]; + value[i++] = m[row + 12]; + } + } + } + return; + case STATE_DEPTH_RANGE: + value[0] = ctx->Viewport.Near; /* near */ + value[1] = ctx->Viewport.Far; /* far */ + value[2] = ctx->Viewport.Far - ctx->Viewport.Near; /* far - near */ + value[3] = 1.0; + return; + case STATE_FRAGMENT_PROGRAM: + { + /* state[1] = {STATE_ENV, STATE_LOCAL} */ + /* state[2] = parameter index */ + const int idx = (int) state[2]; + switch (state[1]) { + case STATE_ENV: + COPY_4V(value, ctx->FragmentProgram.Parameters[idx]); + return; + case STATE_LOCAL: + COPY_4V(value, ctx->FragmentProgram.Current->Base.LocalParams[idx]); + return; + default: + _mesa_problem(ctx, "Bad state switch in _mesa_fetch_state()"); + return; + } + } + return; + + case STATE_VERTEX_PROGRAM: + { + /* state[1] = {STATE_ENV, STATE_LOCAL} */ + /* state[2] = parameter index */ + const int idx = (int) state[2]; + switch (state[1]) { + case STATE_ENV: + COPY_4V(value, ctx->VertexProgram.Parameters[idx]); + return; + case STATE_LOCAL: + COPY_4V(value, ctx->VertexProgram.Current->Base.LocalParams[idx]); + return; + default: + _mesa_problem(ctx, "Bad state switch in _mesa_fetch_state()"); + return; + } + } + return; + + case STATE_NORMAL_SCALE: + ASSIGN_4V(value, ctx->_ModelViewInvScale, 0, 0, 1); + return; + + case STATE_INTERNAL: + switch (state[1]) { + case STATE_CURRENT_ATTRIB: + { + const GLuint idx = (GLuint) state[2]; + COPY_4V(value, ctx->Current.Attrib[idx]); + } + return; + + case STATE_CURRENT_ATTRIB_MAYBE_VP_CLAMPED: + { + const GLuint idx = (GLuint) state[2]; + if(ctx->Light._ClampVertexColor && + (idx == VERT_ATTRIB_COLOR0 || + idx == VERT_ATTRIB_COLOR1)) { + value[0] = CLAMP(ctx->Current.Attrib[idx][0], 0.0f, 1.0f); + value[1] = CLAMP(ctx->Current.Attrib[idx][1], 0.0f, 1.0f); + value[2] = CLAMP(ctx->Current.Attrib[idx][2], 0.0f, 1.0f); + value[3] = CLAMP(ctx->Current.Attrib[idx][3], 0.0f, 1.0f); + } + else + COPY_4V(value, ctx->Current.Attrib[idx]); + } + return; + + case STATE_NORMAL_SCALE: + ASSIGN_4V(value, + ctx->_ModelViewInvScale, + ctx->_ModelViewInvScale, + ctx->_ModelViewInvScale, + 1); + return; + + case STATE_TEXRECT_SCALE: + /* Value = { 1/texWidth, 1/texHeight, 0, 1 }. + * Used to convert unnormalized texcoords to normalized texcoords. + */ + { + const int unit = (int) state[2]; + const struct gl_texture_object *texObj + = ctx->Texture.Unit[unit]._Current; + if (texObj) { + struct gl_texture_image *texImage = texObj->Image[0][0]; + ASSIGN_4V(value, + (GLfloat) (1.0 / texImage->Width), + (GLfloat) (1.0 / texImage->Height), + 0.0f, 1.0f); + } + } + return; + + case STATE_FOG_PARAMS_OPTIMIZED: + /* for simpler per-vertex/pixel fog calcs. POW (for EXP/EXP2 fog) + * might be more expensive than EX2 on some hw, plus it needs + * another constant (e) anyway. Linear fog can now be done with a + * single MAD. + * linear: fogcoord * -1/(end-start) + end/(end-start) + * exp: 2^-(density/ln(2) * fogcoord) + * exp2: 2^-((density/(ln(2)^2) * fogcoord)^2) + */ + value[0] = (ctx->Fog.End == ctx->Fog.Start) + ? 1.0f : (GLfloat)(-1.0F / (ctx->Fog.End - ctx->Fog.Start)); + value[1] = ctx->Fog.End * -value[0]; + value[2] = (GLfloat)(ctx->Fog.Density * M_LOG2E); /* M_LOG2E == 1/ln(2) */ + value[3] = (GLfloat)(ctx->Fog.Density * ONE_DIV_SQRT_LN2); + return; + + case STATE_POINT_SIZE_CLAMPED: + { + /* this includes implementation dependent limits, to avoid + * another potentially necessary clamp. + * Note: for sprites, point smooth (point AA) is ignored + * and we'll clamp to MinPointSizeAA and MaxPointSize, because we + * expect drivers will want to say their minimum for AA size is 0.0 + * but for non-AA it's 1.0 (because normal points with size below 1.0 + * need to get rounded up to 1.0, hence never disappear). GL does + * not specify max clamp size for sprites, other than it needs to be + * at least as large as max AA size, hence use non-AA size there. + */ + GLfloat minImplSize; + GLfloat maxImplSize; + if (ctx->Point.PointSprite) { + minImplSize = ctx->Const.MinPointSizeAA; + maxImplSize = ctx->Const.MaxPointSize; + } + else if (ctx->Point.SmoothFlag || ctx->Multisample._Enabled) { + minImplSize = ctx->Const.MinPointSizeAA; + maxImplSize = ctx->Const.MaxPointSizeAA; + } + else { + minImplSize = ctx->Const.MinPointSize; + maxImplSize = ctx->Const.MaxPointSize; + } + value[0] = ctx->Point.Size; + value[1] = ctx->Point.MinSize >= minImplSize ? ctx->Point.MinSize : minImplSize; + value[2] = ctx->Point.MaxSize <= maxImplSize ? ctx->Point.MaxSize : maxImplSize; + value[3] = ctx->Point.Threshold; + } + return; + case STATE_POINT_SIZE_IMPL_CLAMP: + { + /* for implementation clamp only in vs */ + GLfloat minImplSize; + GLfloat maxImplSize; + if (ctx->Point.PointSprite) { + minImplSize = ctx->Const.MinPointSizeAA; + maxImplSize = ctx->Const.MaxPointSize; + } + else if (ctx->Point.SmoothFlag || ctx->Multisample._Enabled) { + minImplSize = ctx->Const.MinPointSizeAA; + maxImplSize = ctx->Const.MaxPointSizeAA; + } + else { + minImplSize = ctx->Const.MinPointSize; + maxImplSize = ctx->Const.MaxPointSize; + } + value[0] = ctx->Point.Size; + value[1] = minImplSize; + value[2] = maxImplSize; + value[3] = ctx->Point.Threshold; + } + return; + case STATE_LIGHT_SPOT_DIR_NORMALIZED: + { + /* here, state[2] is the light number */ + /* pre-normalize spot dir */ + const GLuint ln = (GLuint) state[2]; + COPY_3V(value, ctx->Light.Light[ln]._NormSpotDirection); + value[3] = ctx->Light.Light[ln]._CosCutoff; + } + return; + + case STATE_LIGHT_POSITION: + { + const GLuint ln = (GLuint) state[2]; + COPY_4V(value, ctx->Light.Light[ln]._Position); + } + return; + + case STATE_LIGHT_POSITION_NORMALIZED: + { + const GLuint ln = (GLuint) state[2]; + COPY_4V(value, ctx->Light.Light[ln]._Position); + NORMALIZE_3FV( value ); + } + return; + + case STATE_LIGHT_HALF_VECTOR: + { + const GLuint ln = (GLuint) state[2]; + GLfloat p[3]; + /* Compute infinite half angle vector: + * halfVector = normalize(normalize(lightPos) + (0, 0, 1)) + * light.EyePosition.w should be 0 for infinite lights. + */ + COPY_3V(p, ctx->Light.Light[ln]._Position); + NORMALIZE_3FV(p); + ADD_3V(value, p, ctx->_EyeZDir); + NORMALIZE_3FV(value); + value[3] = 1.0; + } + return; + + case STATE_PT_SCALE: + value[0] = ctx->Pixel.RedScale; + value[1] = ctx->Pixel.GreenScale; + value[2] = ctx->Pixel.BlueScale; + value[3] = ctx->Pixel.AlphaScale; + return; + + case STATE_PT_BIAS: + value[0] = ctx->Pixel.RedBias; + value[1] = ctx->Pixel.GreenBias; + value[2] = ctx->Pixel.BlueBias; + value[3] = ctx->Pixel.AlphaBias; + return; + + case STATE_SHADOW_AMBIENT: + { + const int unit = (int) state[2]; + const struct gl_texture_object *texObj + = ctx->Texture.Unit[unit]._Current; + if (texObj) { + value[0] = + value[1] = + value[2] = + value[3] = texObj->Sampler.CompareFailValue; + } + } + return; + + case STATE_FB_SIZE: + value[0] = (GLfloat) (ctx->DrawBuffer->Width - 1); + value[1] = (GLfloat) (ctx->DrawBuffer->Height - 1); + value[2] = 0.0F; + value[3] = 0.0F; + return; + + case STATE_FB_WPOS_Y_TRANSFORM: + /* A driver may negate this conditional by using ZW swizzle + * instead of XY (based on e.g. some other state). */ + if (ctx->DrawBuffer->Name != 0) { + /* Identity (XY) followed by flipping Y upside down (ZW). */ + value[0] = 1.0F; + value[1] = 0.0F; + value[2] = -1.0F; + value[3] = (GLfloat) ctx->DrawBuffer->Height; + } else { + /* Flipping Y upside down (XY) followed by identity (ZW). */ + value[0] = -1.0F; + value[1] = (GLfloat) ctx->DrawBuffer->Height; + value[2] = 1.0F; + value[3] = 0.0F; + } + return; + + case STATE_ROT_MATRIX_0: + { + const int unit = (int) state[2]; + GLfloat *rotMat22 = ctx->Texture.Unit[unit].RotMatrix; + value[0] = rotMat22[0]; + value[1] = rotMat22[2]; + value[2] = 0.0; + value[3] = 0.0; + } + return; + + case STATE_ROT_MATRIX_1: + { + const int unit = (int) state[2]; + GLfloat *rotMat22 = ctx->Texture.Unit[unit].RotMatrix; + value[0] = rotMat22[1]; + value[1] = rotMat22[3]; + value[2] = 0.0; + value[3] = 0.0; + } + return; + + /* XXX: make sure new tokens added here are also handled in the + * _mesa_program_state_flags() switch, below. + */ + default: + /* Unknown state indexes are silently ignored here. + * Drivers may do something special. + */ + return; + } + return; + + default: + _mesa_problem(ctx, "Invalid state in _mesa_fetch_state"); + return; + } +} + + +/** + * Return a bitmask of the Mesa state flags (_NEW_* values) which would + * indicate that the given context state may have changed. + * The bitmask is used during validation to determine if we need to update + * vertex/fragment program parameters (like "state.material.color") when + * some GL state has changed. + */ +GLbitfield +_mesa_program_state_flags(const gl_state_index state[STATE_LENGTH]) +{ + switch (state[0]) { + case STATE_MATERIAL: + case STATE_LIGHT: + case STATE_LIGHTMODEL_AMBIENT: + case STATE_LIGHTMODEL_SCENECOLOR: + case STATE_LIGHTPROD: + return _NEW_LIGHT; + + case STATE_TEXGEN: + return _NEW_TEXTURE; + case STATE_TEXENV_COLOR: + return _NEW_TEXTURE | _NEW_BUFFERS | _NEW_FRAG_CLAMP; + + case STATE_FOG_COLOR: + return _NEW_FOG | _NEW_BUFFERS | _NEW_FRAG_CLAMP; + case STATE_FOG_PARAMS: + return _NEW_FOG; + + case STATE_CLIPPLANE: + return _NEW_TRANSFORM; + + case STATE_POINT_SIZE: + case STATE_POINT_ATTENUATION: + return _NEW_POINT; + + case STATE_MODELVIEW_MATRIX: + return _NEW_MODELVIEW; + case STATE_PROJECTION_MATRIX: + return _NEW_PROJECTION; + case STATE_MVP_MATRIX: + return _NEW_MODELVIEW | _NEW_PROJECTION; + case STATE_TEXTURE_MATRIX: + return _NEW_TEXTURE_MATRIX; + case STATE_PROGRAM_MATRIX: + return _NEW_TRACK_MATRIX; + + case STATE_DEPTH_RANGE: + return _NEW_VIEWPORT; + + case STATE_FRAGMENT_PROGRAM: + case STATE_VERTEX_PROGRAM: + return _NEW_PROGRAM; + + case STATE_NORMAL_SCALE: + return _NEW_MODELVIEW; + + case STATE_INTERNAL: + switch (state[1]) { + case STATE_CURRENT_ATTRIB: + return _NEW_CURRENT_ATTRIB; + case STATE_CURRENT_ATTRIB_MAYBE_VP_CLAMPED: + return _NEW_CURRENT_ATTRIB | _NEW_LIGHT | _NEW_BUFFERS; + + case STATE_NORMAL_SCALE: + return _NEW_MODELVIEW; + + case STATE_TEXRECT_SCALE: + case STATE_SHADOW_AMBIENT: + case STATE_ROT_MATRIX_0: + case STATE_ROT_MATRIX_1: + return _NEW_TEXTURE; + case STATE_FOG_PARAMS_OPTIMIZED: + return _NEW_FOG; + case STATE_POINT_SIZE_CLAMPED: + case STATE_POINT_SIZE_IMPL_CLAMP: + return _NEW_POINT | _NEW_MULTISAMPLE; + case STATE_LIGHT_SPOT_DIR_NORMALIZED: + case STATE_LIGHT_POSITION: + case STATE_LIGHT_POSITION_NORMALIZED: + case STATE_LIGHT_HALF_VECTOR: + return _NEW_LIGHT; + + case STATE_PT_SCALE: + case STATE_PT_BIAS: + return _NEW_PIXEL; + + case STATE_FB_SIZE: + case STATE_FB_WPOS_Y_TRANSFORM: + return _NEW_BUFFERS; + + default: + /* unknown state indexes are silently ignored and + * no flag set, since it is handled by the driver. + */ + return 0; + } + + default: + _mesa_problem(NULL, "unexpected state[0] in make_state_flags()"); + return 0; + } +} + + +static void +append(char *dst, const char *src) +{ + while (*dst) + dst++; + while (*src) + *dst++ = *src++; + *dst = 0; +} + + +/** + * Convert token 'k' to a string, append it onto 'dst' string. + */ +static void +append_token(char *dst, gl_state_index k) +{ + switch (k) { + case STATE_MATERIAL: + append(dst, "material"); + break; + case STATE_LIGHT: + append(dst, "light"); + break; + case STATE_LIGHTMODEL_AMBIENT: + append(dst, "lightmodel.ambient"); + break; + case STATE_LIGHTMODEL_SCENECOLOR: + break; + case STATE_LIGHTPROD: + append(dst, "lightprod"); + break; + case STATE_TEXGEN: + append(dst, "texgen"); + break; + case STATE_FOG_COLOR: + append(dst, "fog.color"); + break; + case STATE_FOG_PARAMS: + append(dst, "fog.params"); + break; + case STATE_CLIPPLANE: + append(dst, "clip"); + break; + case STATE_POINT_SIZE: + append(dst, "point.size"); + break; + case STATE_POINT_ATTENUATION: + append(dst, "point.attenuation"); + break; + case STATE_MODELVIEW_MATRIX: + append(dst, "matrix.modelview"); + break; + case STATE_PROJECTION_MATRIX: + append(dst, "matrix.projection"); + break; + case STATE_MVP_MATRIX: + append(dst, "matrix.mvp"); + break; + case STATE_TEXTURE_MATRIX: + append(dst, "matrix.texture"); + break; + case STATE_PROGRAM_MATRIX: + append(dst, "matrix.program"); + break; + case STATE_MATRIX_INVERSE: + append(dst, ".inverse"); + break; + case STATE_MATRIX_TRANSPOSE: + append(dst, ".transpose"); + break; + case STATE_MATRIX_INVTRANS: + append(dst, ".invtrans"); + break; + case STATE_AMBIENT: + append(dst, ".ambient"); + break; + case STATE_DIFFUSE: + append(dst, ".diffuse"); + break; + case STATE_SPECULAR: + append(dst, ".specular"); + break; + case STATE_EMISSION: + append(dst, ".emission"); + break; + case STATE_SHININESS: + append(dst, "lshininess"); + break; + case STATE_HALF_VECTOR: + append(dst, ".half"); + break; + case STATE_POSITION: + append(dst, ".position"); + break; + case STATE_ATTENUATION: + append(dst, ".attenuation"); + break; + case STATE_SPOT_DIRECTION: + append(dst, ".spot.direction"); + break; + case STATE_SPOT_CUTOFF: + append(dst, ".spot.cutoff"); + break; + case STATE_TEXGEN_EYE_S: + append(dst, ".eye.s"); + break; + case STATE_TEXGEN_EYE_T: + append(dst, ".eye.t"); + break; + case STATE_TEXGEN_EYE_R: + append(dst, ".eye.r"); + break; + case STATE_TEXGEN_EYE_Q: + append(dst, ".eye.q"); + break; + case STATE_TEXGEN_OBJECT_S: + append(dst, ".object.s"); + break; + case STATE_TEXGEN_OBJECT_T: + append(dst, ".object.t"); + break; + case STATE_TEXGEN_OBJECT_R: + append(dst, ".object.r"); + break; + case STATE_TEXGEN_OBJECT_Q: + append(dst, ".object.q"); + break; + case STATE_TEXENV_COLOR: + append(dst, "texenv"); + break; + case STATE_DEPTH_RANGE: + append(dst, "depth.range"); + break; + case STATE_VERTEX_PROGRAM: + case STATE_FRAGMENT_PROGRAM: + break; + case STATE_ENV: + append(dst, "env"); + break; + case STATE_LOCAL: + append(dst, "local"); + break; + /* BEGIN internal state vars */ + case STATE_INTERNAL: + append(dst, ".internal."); + break; + case STATE_CURRENT_ATTRIB: + append(dst, "current"); + break; + case STATE_NORMAL_SCALE: + append(dst, "normalScale"); + break; + case STATE_TEXRECT_SCALE: + append(dst, "texrectScale"); + break; + case STATE_FOG_PARAMS_OPTIMIZED: + append(dst, "fogParamsOptimized"); + break; + case STATE_POINT_SIZE_CLAMPED: + append(dst, "pointSizeClamped"); + break; + case STATE_POINT_SIZE_IMPL_CLAMP: + append(dst, "pointSizeImplClamp"); + break; + case STATE_LIGHT_SPOT_DIR_NORMALIZED: + append(dst, "lightSpotDirNormalized"); + break; + case STATE_LIGHT_POSITION: + append(dst, "lightPosition"); + break; + case STATE_LIGHT_POSITION_NORMALIZED: + append(dst, "light.position.normalized"); + break; + case STATE_LIGHT_HALF_VECTOR: + append(dst, "lightHalfVector"); + break; + case STATE_PT_SCALE: + append(dst, "PTscale"); + break; + case STATE_PT_BIAS: + append(dst, "PTbias"); + break; + case STATE_SHADOW_AMBIENT: + append(dst, "CompareFailValue"); + break; + case STATE_FB_SIZE: + append(dst, "FbSize"); + break; + case STATE_FB_WPOS_Y_TRANSFORM: + append(dst, "FbWposYTransform"); + break; + case STATE_ROT_MATRIX_0: + append(dst, "rotMatrixRow0"); + break; + case STATE_ROT_MATRIX_1: + append(dst, "rotMatrixRow1"); + break; + default: + /* probably STATE_INTERNAL_DRIVER+i (driver private state) */ + append(dst, "driverState"); + } +} + +static void +append_face(char *dst, GLint face) +{ + if (face == 0) + append(dst, "front."); + else + append(dst, "back."); +} + +static void +append_index(char *dst, GLint index) +{ + char s[20]; + sprintf(s, "[%d]", index); + append(dst, s); +} + +/** + * Make a string from the given state vector. + * For example, return "state.matrix.texture[2].inverse". + * Use free() to deallocate the string. + */ +char * +_mesa_program_state_string(const gl_state_index state[STATE_LENGTH]) +{ + char str[1000] = ""; + char tmp[30]; + + append(str, "state."); + append_token(str, state[0]); + + switch (state[0]) { + case STATE_MATERIAL: + append_face(str, state[1]); + append_token(str, state[2]); + break; + case STATE_LIGHT: + append_index(str, state[1]); /* light number [i]. */ + append_token(str, state[2]); /* coefficients */ + break; + case STATE_LIGHTMODEL_AMBIENT: + append(str, "lightmodel.ambient"); + break; + case STATE_LIGHTMODEL_SCENECOLOR: + if (state[1] == 0) { + append(str, "lightmodel.front.scenecolor"); + } + else { + append(str, "lightmodel.back.scenecolor"); + } + break; + case STATE_LIGHTPROD: + append_index(str, state[1]); /* light number [i]. */ + append_face(str, state[2]); + append_token(str, state[3]); + break; + case STATE_TEXGEN: + append_index(str, state[1]); /* tex unit [i] */ + append_token(str, state[2]); /* plane coef */ + break; + case STATE_TEXENV_COLOR: + append_index(str, state[1]); /* tex unit [i] */ + append(str, "color"); + break; + case STATE_CLIPPLANE: + append_index(str, state[1]); /* plane [i] */ + append(str, ".plane"); + break; + case STATE_MODELVIEW_MATRIX: + case STATE_PROJECTION_MATRIX: + case STATE_MVP_MATRIX: + case STATE_TEXTURE_MATRIX: + case STATE_PROGRAM_MATRIX: + { + /* state[0] = modelview, projection, texture, etc. */ + /* state[1] = which texture matrix or program matrix */ + /* state[2] = first row to fetch */ + /* state[3] = last row to fetch */ + /* state[4] = transpose, inverse or invtrans */ + const gl_state_index mat = state[0]; + const GLuint index = (GLuint) state[1]; + const GLuint firstRow = (GLuint) state[2]; + const GLuint lastRow = (GLuint) state[3]; + const gl_state_index modifier = state[4]; + if (index || + mat == STATE_TEXTURE_MATRIX || + mat == STATE_PROGRAM_MATRIX) + append_index(str, index); + if (modifier) + append_token(str, modifier); + if (firstRow == lastRow) + sprintf(tmp, ".row[%d]", firstRow); + else + sprintf(tmp, ".row[%d..%d]", firstRow, lastRow); + append(str, tmp); + } + break; + case STATE_POINT_SIZE: + break; + case STATE_POINT_ATTENUATION: + break; + case STATE_FOG_PARAMS: + break; + case STATE_FOG_COLOR: + break; + case STATE_DEPTH_RANGE: + break; + case STATE_FRAGMENT_PROGRAM: + case STATE_VERTEX_PROGRAM: + /* state[1] = {STATE_ENV, STATE_LOCAL} */ + /* state[2] = parameter index */ + append_token(str, state[1]); + append_index(str, state[2]); + break; + case STATE_NORMAL_SCALE: + break; + case STATE_INTERNAL: + append_token(str, state[1]); + if (state[1] == STATE_CURRENT_ATTRIB) + append_index(str, state[2]); + break; + default: + _mesa_problem(NULL, "Invalid state in _mesa_program_state_string"); + break; + } + + return _mesa_strdup(str); +} + + +/** + * Loop over all the parameters in a parameter list. If the parameter + * is a GL state reference, look up the current value of that state + * variable and put it into the parameter's Value[4] array. + * Other parameter types never change or are explicitly set by the user + * with glUniform() or glProgramParameter(), etc. + * This would be called at glBegin time. + */ +void +_mesa_load_state_parameters(struct gl_context *ctx, + struct gl_program_parameter_list *paramList) +{ + GLuint i; + + if (!paramList) + return; + + for (i = 0; i < paramList->NumParameters; i++) { + if (paramList->Parameters[i].Type == PROGRAM_STATE_VAR) { + _mesa_fetch_state(ctx, + paramList->Parameters[i].StateIndexes, + ¶mList->ParameterValues[i][0].f); + } + } +} + + +/** + * Copy the 16 elements of a matrix into four consecutive program + * registers starting at 'pos'. + */ +static void +load_matrix(GLfloat registers[][4], GLuint pos, const GLfloat mat[16]) +{ + GLuint i; + for (i = 0; i < 4; i++) { + registers[pos + i][0] = mat[0 + i]; + registers[pos + i][1] = mat[4 + i]; + registers[pos + i][2] = mat[8 + i]; + registers[pos + i][3] = mat[12 + i]; + } +} + + +/** + * As above, but transpose the matrix. + */ +static void +load_transpose_matrix(GLfloat registers[][4], GLuint pos, + const GLfloat mat[16]) +{ + memcpy(registers[pos], mat, 16 * sizeof(GLfloat)); +} + + +/** + * Load current vertex program's parameter registers with tracked + * matrices (if NV program). This only needs to be done per + * glBegin/glEnd, not per-vertex. + */ +void +_mesa_load_tracked_matrices(struct gl_context *ctx) +{ + GLuint i; + + for (i = 0; i < MAX_NV_VERTEX_PROGRAM_PARAMS / 4; i++) { + /* point 'mat' at source matrix */ + GLmatrix *mat; + if (ctx->VertexProgram.TrackMatrix[i] == GL_MODELVIEW) { + mat = ctx->ModelviewMatrixStack.Top; + } + else if (ctx->VertexProgram.TrackMatrix[i] == GL_PROJECTION) { + mat = ctx->ProjectionMatrixStack.Top; + } + else if (ctx->VertexProgram.TrackMatrix[i] == GL_TEXTURE) { + GLuint unit = MIN2(ctx->Texture.CurrentUnit, + Elements(ctx->TextureMatrixStack) - 1); + mat = ctx->TextureMatrixStack[unit].Top; + } + else if (ctx->VertexProgram.TrackMatrix[i]==GL_MODELVIEW_PROJECTION_NV) { + /* XXX verify the combined matrix is up to date */ + mat = &ctx->_ModelProjectMatrix; + } + else if (ctx->VertexProgram.TrackMatrix[i] >= GL_MATRIX0_NV && + ctx->VertexProgram.TrackMatrix[i] <= GL_MATRIX7_NV) { + GLuint n = ctx->VertexProgram.TrackMatrix[i] - GL_MATRIX0_NV; + ASSERT(n < Elements(ctx->ProgramMatrixStack)); + mat = ctx->ProgramMatrixStack[n].Top; + } + else { + /* no matrix is tracked, but we leave the register values as-is */ + assert(ctx->VertexProgram.TrackMatrix[i] == GL_NONE); + continue; + } + + /* load the matrix values into sequential registers */ + if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_IDENTITY_NV) { + load_matrix(ctx->VertexProgram.Parameters, i*4, mat->m); + } + else if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_INVERSE_NV) { + _math_matrix_analyse(mat); /* update the inverse */ + ASSERT(!_math_matrix_is_dirty(mat)); + load_matrix(ctx->VertexProgram.Parameters, i*4, mat->inv); + } + else if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_TRANSPOSE_NV) { + load_transpose_matrix(ctx->VertexProgram.Parameters, i*4, mat->m); + } + else { + assert(ctx->VertexProgram.TrackMatrixTransform[i] + == GL_INVERSE_TRANSPOSE_NV); + _math_matrix_analyse(mat); /* update the inverse */ + ASSERT(!_math_matrix_is_dirty(mat)); + load_transpose_matrix(ctx->VertexProgram.Parameters, i*4, mat->inv); + } + } +} diff --git a/mesalib/src/mesa/program/prog_statevars.h b/mesalib/src/mesa/program/prog_statevars.h index 07e0a2798..04af3f4cf 100644 --- a/mesalib/src/mesa/program/prog_statevars.h +++ b/mesalib/src/mesa/program/prog_statevars.h @@ -1,148 +1,148 @@ -/* - * Mesa 3-D graphics library - * Version: 7.1 - * - * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - -#ifndef PROG_STATEVARS_H -#define PROG_STATEVARS_H - -#include "main/glheader.h" - -struct gl_context; -struct gl_program_parameter_list; - -/** - * Number of STATE_* values we need to address any GL state. - * Used to dimension arrays. - */ -#define STATE_LENGTH 5 - - -/** - * Used for describing GL state referenced from inside ARB vertex and - * fragment programs. - * A string such as "state.light[0].ambient" gets translated into a - * sequence of tokens such as [ STATE_LIGHT, 0, STATE_AMBIENT ]. - * - * For state that's an array, like STATE_CLIPPLANE, the 2nd token [1] should - * always be the array index. - */ -typedef enum gl_state_index_ { - STATE_MATERIAL = 100, /* start at 100 so small ints are seen as ints */ - - STATE_LIGHT, - STATE_LIGHTMODEL_AMBIENT, - STATE_LIGHTMODEL_SCENECOLOR, - STATE_LIGHTPROD, - - STATE_TEXGEN, - - STATE_FOG_COLOR, - STATE_FOG_PARAMS, - - STATE_CLIPPLANE, - - STATE_POINT_SIZE, - STATE_POINT_ATTENUATION, - - STATE_MODELVIEW_MATRIX, - STATE_PROJECTION_MATRIX, - STATE_MVP_MATRIX, - STATE_TEXTURE_MATRIX, - STATE_PROGRAM_MATRIX, - STATE_MATRIX_INVERSE, - STATE_MATRIX_TRANSPOSE, - STATE_MATRIX_INVTRANS, - - STATE_AMBIENT, - STATE_DIFFUSE, - STATE_SPECULAR, - STATE_EMISSION, - STATE_SHININESS, - STATE_HALF_VECTOR, - - STATE_POSITION, /**< xyzw = position */ - STATE_ATTENUATION, /**< xyz = attenuation, w = spot exponent */ - STATE_SPOT_DIRECTION, /**< xyz = direction, w = cos(cutoff) */ - STATE_SPOT_CUTOFF, /**< x = cutoff, yzw = undefined */ - - STATE_TEXGEN_EYE_S, - STATE_TEXGEN_EYE_T, - STATE_TEXGEN_EYE_R, - STATE_TEXGEN_EYE_Q, - STATE_TEXGEN_OBJECT_S, - STATE_TEXGEN_OBJECT_T, - STATE_TEXGEN_OBJECT_R, - STATE_TEXGEN_OBJECT_Q, - - STATE_TEXENV_COLOR, - - STATE_DEPTH_RANGE, - - STATE_VERTEX_PROGRAM, - STATE_FRAGMENT_PROGRAM, - - STATE_ENV, - STATE_LOCAL, - - STATE_INTERNAL, /* Mesa additions */ - STATE_CURRENT_ATTRIB, /* ctx->Current vertex attrib value */ - STATE_CURRENT_ATTRIB_MAYBE_VP_CLAMPED, /* ctx->Current vertex attrib value after passthrough vertex processing */ - STATE_NORMAL_SCALE, - STATE_TEXRECT_SCALE, - STATE_FOG_PARAMS_OPTIMIZED, /* for faster fog calc */ - STATE_POINT_SIZE_CLAMPED, /* includes implementation dependent size clamp */ - STATE_POINT_SIZE_IMPL_CLAMP, /* for implementation clamp only in vs */ - STATE_LIGHT_SPOT_DIR_NORMALIZED, /* pre-normalized spot dir */ - STATE_LIGHT_POSITION, /* object vs eye space */ - STATE_LIGHT_POSITION_NORMALIZED, /* object vs eye space */ - STATE_LIGHT_HALF_VECTOR, /* object vs eye space */ - STATE_PT_SCALE, /**< Pixel transfer RGBA scale */ - STATE_PT_BIAS, /**< Pixel transfer RGBA bias */ - STATE_SHADOW_AMBIENT, /**< ARB_shadow_ambient fail value; token[2] is texture unit index */ - STATE_FB_SIZE, /**< (width-1, height-1, 0, 0) */ - STATE_FB_WPOS_Y_TRANSFORM, /**< (1, 0, -1, height) if a FBO is bound, (-1, height, 1, 0) otherwise */ - STATE_ROT_MATRIX_0, /**< ATI_envmap_bumpmap, rot matrix row 0 */ - STATE_ROT_MATRIX_1, /**< ATI_envmap_bumpmap, rot matrix row 1 */ - STATE_INTERNAL_DRIVER /* first available state index for drivers (must be last) */ -} gl_state_index; - - - -extern void -_mesa_load_state_parameters(struct gl_context *ctx, - struct gl_program_parameter_list *paramList); - - -extern GLbitfield -_mesa_program_state_flags(const gl_state_index state[STATE_LENGTH]); - - -extern char * -_mesa_program_state_string(const gl_state_index state[STATE_LENGTH]); - - -extern void -_mesa_load_tracked_matrices(struct gl_context *ctx); - - -#endif /* PROG_STATEVARS_H */ +/* + * Mesa 3-D graphics library + * Version: 7.1 + * + * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +#ifndef PROG_STATEVARS_H +#define PROG_STATEVARS_H + +#include "main/glheader.h" + +struct gl_context; +struct gl_program_parameter_list; + +/** + * Number of STATE_* values we need to address any GL state. + * Used to dimension arrays. + */ +#define STATE_LENGTH 5 + + +/** + * Used for describing GL state referenced from inside ARB vertex and + * fragment programs. + * A string such as "state.light[0].ambient" gets translated into a + * sequence of tokens such as [ STATE_LIGHT, 0, STATE_AMBIENT ]. + * + * For state that's an array, like STATE_CLIPPLANE, the 2nd token [1] should + * always be the array index. + */ +typedef enum gl_state_index_ { + STATE_MATERIAL = 100, /* start at 100 so small ints are seen as ints */ + + STATE_LIGHT, + STATE_LIGHTMODEL_AMBIENT, + STATE_LIGHTMODEL_SCENECOLOR, + STATE_LIGHTPROD, + + STATE_TEXGEN, + + STATE_FOG_COLOR, + STATE_FOG_PARAMS, + + STATE_CLIPPLANE, + + STATE_POINT_SIZE, + STATE_POINT_ATTENUATION, + + STATE_MODELVIEW_MATRIX, + STATE_PROJECTION_MATRIX, + STATE_MVP_MATRIX, + STATE_TEXTURE_MATRIX, + STATE_PROGRAM_MATRIX, + STATE_MATRIX_INVERSE, + STATE_MATRIX_TRANSPOSE, + STATE_MATRIX_INVTRANS, + + STATE_AMBIENT, + STATE_DIFFUSE, + STATE_SPECULAR, + STATE_EMISSION, + STATE_SHININESS, + STATE_HALF_VECTOR, + + STATE_POSITION, /**< xyzw = position */ + STATE_ATTENUATION, /**< xyz = attenuation, w = spot exponent */ + STATE_SPOT_DIRECTION, /**< xyz = direction, w = cos(cutoff) */ + STATE_SPOT_CUTOFF, /**< x = cutoff, yzw = undefined */ + + STATE_TEXGEN_EYE_S, + STATE_TEXGEN_EYE_T, + STATE_TEXGEN_EYE_R, + STATE_TEXGEN_EYE_Q, + STATE_TEXGEN_OBJECT_S, + STATE_TEXGEN_OBJECT_T, + STATE_TEXGEN_OBJECT_R, + STATE_TEXGEN_OBJECT_Q, + + STATE_TEXENV_COLOR, + + STATE_DEPTH_RANGE, + + STATE_VERTEX_PROGRAM, + STATE_FRAGMENT_PROGRAM, + + STATE_ENV, + STATE_LOCAL, + + STATE_INTERNAL, /* Mesa additions */ + STATE_CURRENT_ATTRIB, /* ctx->Current vertex attrib value */ + STATE_CURRENT_ATTRIB_MAYBE_VP_CLAMPED, /* ctx->Current vertex attrib value after passthrough vertex processing */ + STATE_NORMAL_SCALE, + STATE_TEXRECT_SCALE, + STATE_FOG_PARAMS_OPTIMIZED, /* for faster fog calc */ + STATE_POINT_SIZE_CLAMPED, /* includes implementation dependent size clamp */ + STATE_POINT_SIZE_IMPL_CLAMP, /* for implementation clamp only in vs */ + STATE_LIGHT_SPOT_DIR_NORMALIZED, /* pre-normalized spot dir */ + STATE_LIGHT_POSITION, /* object vs eye space */ + STATE_LIGHT_POSITION_NORMALIZED, /* object vs eye space */ + STATE_LIGHT_HALF_VECTOR, /* object vs eye space */ + STATE_PT_SCALE, /**< Pixel transfer RGBA scale */ + STATE_PT_BIAS, /**< Pixel transfer RGBA bias */ + STATE_SHADOW_AMBIENT, /**< ARB_shadow_ambient fail value; token[2] is texture unit index */ + STATE_FB_SIZE, /**< (width-1, height-1, 0, 0) */ + STATE_FB_WPOS_Y_TRANSFORM, /**< (1, 0, -1, height) if a FBO is bound, (-1, height, 1, 0) otherwise */ + STATE_ROT_MATRIX_0, /**< ATI_envmap_bumpmap, rot matrix row 0 */ + STATE_ROT_MATRIX_1, /**< ATI_envmap_bumpmap, rot matrix row 1 */ + STATE_INTERNAL_DRIVER /* first available state index for drivers (must be last) */ +} gl_state_index; + + + +extern void +_mesa_load_state_parameters(struct gl_context *ctx, + struct gl_program_parameter_list *paramList); + + +extern GLbitfield +_mesa_program_state_flags(const gl_state_index state[STATE_LENGTH]); + + +extern char * +_mesa_program_state_string(const gl_state_index state[STATE_LENGTH]); + + +extern void +_mesa_load_tracked_matrices(struct gl_context *ctx); + + +#endif /* PROG_STATEVARS_H */ diff --git a/mesalib/src/mesa/program/program.c b/mesalib/src/mesa/program/program.c index 5a9bc0fae..ecff2344a 100644 --- a/mesalib/src/mesa/program/program.c +++ b/mesalib/src/mesa/program/program.c @@ -1,1080 +1,1080 @@ -/* - * Mesa 3-D graphics library - * Version: 6.5.3 - * - * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - -/** - * \file program.c - * Vertex and fragment program support functions. - * \author Brian Paul - */ - - -#include "main/glheader.h" -#include "main/context.h" -#include "main/hash.h" -#include "main/mfeatures.h" -#include "program.h" -#include "prog_cache.h" -#include "prog_parameter.h" -#include "prog_instruction.h" - - -/** - * A pointer to this dummy program is put into the hash table when - * glGenPrograms is called. - */ -struct gl_program _mesa_DummyProgram; - - -/** - * Init context's vertex/fragment program state - */ -void -_mesa_init_program(struct gl_context *ctx) -{ - GLuint i; - - /* - * If this assertion fails, we need to increase the field - * size for register indexes (see INST_INDEX_BITS). - */ - ASSERT(ctx->Const.VertexProgram.MaxUniformComponents / 4 - <= (1 << INST_INDEX_BITS)); - ASSERT(ctx->Const.FragmentProgram.MaxUniformComponents / 4 - <= (1 << INST_INDEX_BITS)); - - ASSERT(ctx->Const.VertexProgram.MaxTemps <= (1 << INST_INDEX_BITS)); - ASSERT(ctx->Const.VertexProgram.MaxLocalParams <= (1 << INST_INDEX_BITS)); - ASSERT(ctx->Const.FragmentProgram.MaxTemps <= (1 << INST_INDEX_BITS)); - ASSERT(ctx->Const.FragmentProgram.MaxLocalParams <= (1 << INST_INDEX_BITS)); - - ASSERT(ctx->Const.VertexProgram.MaxUniformComponents <= 4 * MAX_UNIFORMS); - ASSERT(ctx->Const.FragmentProgram.MaxUniformComponents <= 4 * MAX_UNIFORMS); - - ASSERT(ctx->Const.VertexProgram.MaxAddressOffset <= (1 << INST_INDEX_BITS)); - ASSERT(ctx->Const.FragmentProgram.MaxAddressOffset <= (1 << INST_INDEX_BITS)); - - /* If this fails, increase prog_instruction::TexSrcUnit size */ - ASSERT(MAX_TEXTURE_UNITS <= (1 << 5)); - - /* If this fails, increase prog_instruction::TexSrcTarget size */ - ASSERT(NUM_TEXTURE_TARGETS <= (1 << 3)); - - ctx->Program.ErrorPos = -1; - ctx->Program.ErrorString = _mesa_strdup(""); - -#if FEATURE_NV_vertex_program || FEATURE_ARB_vertex_program - ctx->VertexProgram.Enabled = GL_FALSE; -#if FEATURE_es2_glsl - ctx->VertexProgram.PointSizeEnabled = - (ctx->API == API_OPENGLES2) ? GL_TRUE : GL_FALSE; -#else - ctx->VertexProgram.PointSizeEnabled = GL_FALSE; -#endif - ctx->VertexProgram.TwoSideEnabled = GL_FALSE; - _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current, - ctx->Shared->DefaultVertexProgram); - assert(ctx->VertexProgram.Current); - for (i = 0; i < MAX_NV_VERTEX_PROGRAM_PARAMS / 4; i++) { - ctx->VertexProgram.TrackMatrix[i] = GL_NONE; - ctx->VertexProgram.TrackMatrixTransform[i] = GL_IDENTITY_NV; - } - ctx->VertexProgram.Cache = _mesa_new_program_cache(); -#endif - -#if FEATURE_NV_fragment_program || FEATURE_ARB_fragment_program - ctx->FragmentProgram.Enabled = GL_FALSE; - _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current, - ctx->Shared->DefaultFragmentProgram); - assert(ctx->FragmentProgram.Current); - ctx->FragmentProgram.Cache = _mesa_new_program_cache(); -#endif - -#if FEATURE_ARB_geometry_shader4 - ctx->GeometryProgram.Enabled = GL_FALSE; - /* right now by default we don't have a geometry program */ - _mesa_reference_geomprog(ctx, &ctx->GeometryProgram.Current, - NULL); - ctx->GeometryProgram.Cache = _mesa_new_program_cache(); -#endif - - /* XXX probably move this stuff */ -#if FEATURE_ATI_fragment_shader - ctx->ATIFragmentShader.Enabled = GL_FALSE; - ctx->ATIFragmentShader.Current = ctx->Shared->DefaultFragmentShader; - assert(ctx->ATIFragmentShader.Current); - ctx->ATIFragmentShader.Current->RefCount++; -#endif -} - - -/** - * Free a context's vertex/fragment program state - */ -void -_mesa_free_program_data(struct gl_context *ctx) -{ -#if FEATURE_NV_vertex_program || FEATURE_ARB_vertex_program - _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current, NULL); - _mesa_delete_program_cache(ctx, ctx->VertexProgram.Cache); -#endif -#if FEATURE_NV_fragment_program || FEATURE_ARB_fragment_program - _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current, NULL); - _mesa_delete_program_cache(ctx, ctx->FragmentProgram.Cache); -#endif -#if FEATURE_ARB_geometry_shader4 - _mesa_reference_geomprog(ctx, &ctx->GeometryProgram.Current, NULL); - _mesa_delete_program_cache(ctx, ctx->GeometryProgram.Cache); -#endif - /* XXX probably move this stuff */ -#if FEATURE_ATI_fragment_shader - if (ctx->ATIFragmentShader.Current) { - ctx->ATIFragmentShader.Current->RefCount--; - if (ctx->ATIFragmentShader.Current->RefCount <= 0) { - free(ctx->ATIFragmentShader.Current); - } - } -#endif - free((void *) ctx->Program.ErrorString); -} - - -/** - * Update the default program objects in the given context to reference those - * specified in the shared state and release those referencing the old - * shared state. - */ -void -_mesa_update_default_objects_program(struct gl_context *ctx) -{ -#if FEATURE_NV_vertex_program || FEATURE_ARB_vertex_program - _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current, - (struct gl_vertex_program *) - ctx->Shared->DefaultVertexProgram); - assert(ctx->VertexProgram.Current); -#endif - -#if FEATURE_NV_fragment_program || FEATURE_ARB_fragment_program - _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current, - (struct gl_fragment_program *) - ctx->Shared->DefaultFragmentProgram); - assert(ctx->FragmentProgram.Current); -#endif - -#if FEATURE_ARB_geometry_shader4 - _mesa_reference_geomprog(ctx, &ctx->GeometryProgram.Current, - (struct gl_geometry_program *) - ctx->Shared->DefaultGeometryProgram); -#endif - - /* XXX probably move this stuff */ -#if FEATURE_ATI_fragment_shader - if (ctx->ATIFragmentShader.Current) { - ctx->ATIFragmentShader.Current->RefCount--; - if (ctx->ATIFragmentShader.Current->RefCount <= 0) { - free(ctx->ATIFragmentShader.Current); - } - } - ctx->ATIFragmentShader.Current = (struct ati_fragment_shader *) ctx->Shared->DefaultFragmentShader; - assert(ctx->ATIFragmentShader.Current); - ctx->ATIFragmentShader.Current->RefCount++; -#endif -} - - -/** - * Set the vertex/fragment program error state (position and error string). - * This is generally called from within the parsers. - */ -void -_mesa_set_program_error(struct gl_context *ctx, GLint pos, const char *string) -{ - ctx->Program.ErrorPos = pos; - free((void *) ctx->Program.ErrorString); - if (!string) - string = ""; - ctx->Program.ErrorString = _mesa_strdup(string); -} - - -/** - * Find the line number and column for 'pos' within 'string'. - * Return a copy of the line which contains 'pos'. Free the line with - * free(). - * \param string the program string - * \param pos the position within the string - * \param line returns the line number corresponding to 'pos'. - * \param col returns the column number corresponding to 'pos'. - * \return copy of the line containing 'pos'. - */ -const GLubyte * -_mesa_find_line_column(const GLubyte *string, const GLubyte *pos, - GLint *line, GLint *col) -{ - const GLubyte *lineStart = string; - const GLubyte *p = string; - GLubyte *s; - int len; - - *line = 1; - - while (p != pos) { - if (*p == (GLubyte) '\n') { - (*line)++; - lineStart = p + 1; - } - p++; - } - - *col = (pos - lineStart) + 1; - - /* return copy of this line */ - while (*p != 0 && *p != '\n') - p++; - len = p - lineStart; - s = (GLubyte *) malloc(len + 1); - memcpy(s, lineStart, len); - s[len] = 0; - - return s; -} - - -/** - * Initialize a new vertex/fragment program object. - */ -static struct gl_program * -_mesa_init_program_struct( struct gl_context *ctx, struct gl_program *prog, - GLenum target, GLuint id) -{ - (void) ctx; - if (prog) { - GLuint i; - memset(prog, 0, sizeof(*prog)); - prog->Id = id; - prog->Target = target; - prog->Resident = GL_TRUE; - prog->RefCount = 1; - prog->Format = GL_PROGRAM_FORMAT_ASCII_ARB; - - /* default mapping from samplers to texture units */ - for (i = 0; i < MAX_SAMPLERS; i++) - prog->SamplerUnits[i] = i; - } - - return prog; -} - - -/** - * Initialize a new fragment program object. - */ -struct gl_program * -_mesa_init_fragment_program( struct gl_context *ctx, struct gl_fragment_program *prog, - GLenum target, GLuint id) -{ - if (prog) - return _mesa_init_program_struct( ctx, &prog->Base, target, id ); - else - return NULL; -} - - -/** - * Initialize a new vertex program object. - */ -struct gl_program * -_mesa_init_vertex_program( struct gl_context *ctx, struct gl_vertex_program *prog, - GLenum target, GLuint id) -{ - if (prog) - return _mesa_init_program_struct( ctx, &prog->Base, target, id ); - else - return NULL; -} - - -/** - * Initialize a new geometry program object. - */ -struct gl_program * -_mesa_init_geometry_program( struct gl_context *ctx, struct gl_geometry_program *prog, - GLenum target, GLuint id) -{ - if (prog) - return _mesa_init_program_struct( ctx, &prog->Base, target, id ); - else - return NULL; -} - - -/** - * Allocate and initialize a new fragment/vertex program object but - * don't put it into the program hash table. Called via - * ctx->Driver.NewProgram. May be overridden (ie. replaced) by a - * device driver function to implement OO deriviation with additional - * types not understood by this function. - * - * \param ctx context - * \param id program id/number - * \param target program target/type - * \return pointer to new program object - */ -struct gl_program * -_mesa_new_program(struct gl_context *ctx, GLenum target, GLuint id) -{ - struct gl_program *prog; - switch (target) { - case GL_VERTEX_PROGRAM_ARB: /* == GL_VERTEX_PROGRAM_NV */ - case GL_VERTEX_STATE_PROGRAM_NV: - prog = _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program), - target, id ); - break; - case GL_FRAGMENT_PROGRAM_NV: - case GL_FRAGMENT_PROGRAM_ARB: - prog =_mesa_init_fragment_program(ctx, - CALLOC_STRUCT(gl_fragment_program), - target, id ); - break; - case MESA_GEOMETRY_PROGRAM: - prog = _mesa_init_geometry_program(ctx, - CALLOC_STRUCT(gl_geometry_program), - target, id); - break; - default: - _mesa_problem(ctx, "bad target in _mesa_new_program"); - prog = NULL; - } - return prog; -} - - -/** - * Delete a program and remove it from the hash table, ignoring the - * reference count. - * Called via ctx->Driver.DeleteProgram. May be wrapped (OO deriviation) - * by a device driver function. - */ -void -_mesa_delete_program(struct gl_context *ctx, struct gl_program *prog) -{ - (void) ctx; - ASSERT(prog); - ASSERT(prog->RefCount==0); - - if (prog == &_mesa_DummyProgram) - return; - - if (prog->String) - free(prog->String); - - if (prog->Instructions) { - _mesa_free_instructions(prog->Instructions, prog->NumInstructions); - } - if (prog->Parameters) { - _mesa_free_parameter_list(prog->Parameters); - } - if (prog->Varying) { - _mesa_free_parameter_list(prog->Varying); - } - if (prog->Attributes) { - _mesa_free_parameter_list(prog->Attributes); - } - - free(prog); -} - - -/** - * Return the gl_program object for a given ID. - * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of - * casts elsewhere. - */ -struct gl_program * -_mesa_lookup_program(struct gl_context *ctx, GLuint id) -{ - if (id) - return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id); - else - return NULL; -} - - -/** - * Reference counting for vertex/fragment programs - * This is normally only called from the _mesa_reference_program() macro - * when there's a real pointer change. - */ -void -_mesa_reference_program_(struct gl_context *ctx, - struct gl_program **ptr, - struct gl_program *prog) -{ -#ifndef NDEBUG - assert(ptr); - if (*ptr && prog) { - /* sanity check */ - if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB) - ASSERT(prog->Target == GL_VERTEX_PROGRAM_ARB); - else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB) - ASSERT(prog->Target == GL_FRAGMENT_PROGRAM_ARB || - prog->Target == GL_FRAGMENT_PROGRAM_NV); - else if ((*ptr)->Target == MESA_GEOMETRY_PROGRAM) - ASSERT(prog->Target == MESA_GEOMETRY_PROGRAM); - } -#endif - - if (*ptr) { - GLboolean deleteFlag; - - /*_glthread_LOCK_MUTEX((*ptr)->Mutex);*/ -#if 0 - printf("Program %p ID=%u Target=%s Refcount-- to %d\n", - *ptr, (*ptr)->Id, - ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB ? "VP" : - ((*ptr)->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")), - (*ptr)->RefCount - 1); -#endif - ASSERT((*ptr)->RefCount > 0); - (*ptr)->RefCount--; - - deleteFlag = ((*ptr)->RefCount == 0); - /*_glthread_UNLOCK_MUTEX((*ptr)->Mutex);*/ - - if (deleteFlag) { - ASSERT(ctx); - ctx->Driver.DeleteProgram(ctx, *ptr); - } - - *ptr = NULL; - } - - assert(!*ptr); - if (prog) { - /*_glthread_LOCK_MUTEX(prog->Mutex);*/ - prog->RefCount++; -#if 0 - printf("Program %p ID=%u Target=%s Refcount++ to %d\n", - prog, prog->Id, - (prog->Target == GL_VERTEX_PROGRAM_ARB ? "VP" : - (prog->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")), - prog->RefCount); -#endif - /*_glthread_UNLOCK_MUTEX(prog->Mutex);*/ - } - - *ptr = prog; -} - - -/** - * Return a copy of a program. - * XXX Problem here if the program object is actually OO-derivation - * made by a device driver. - */ -struct gl_program * -_mesa_clone_program(struct gl_context *ctx, const struct gl_program *prog) -{ - struct gl_program *clone; - - clone = ctx->Driver.NewProgram(ctx, prog->Target, prog->Id); - if (!clone) - return NULL; - - assert(clone->Target == prog->Target); - assert(clone->RefCount == 1); - - clone->String = (GLubyte *) _mesa_strdup((char *) prog->String); - clone->Format = prog->Format; - clone->Instructions = _mesa_alloc_instructions(prog->NumInstructions); - if (!clone->Instructions) { - _mesa_reference_program(ctx, &clone, NULL); - return NULL; - } - _mesa_copy_instructions(clone->Instructions, prog->Instructions, - prog->NumInstructions); - clone->InputsRead = prog->InputsRead; - clone->OutputsWritten = prog->OutputsWritten; - clone->SamplersUsed = prog->SamplersUsed; - clone->ShadowSamplers = prog->ShadowSamplers; - memcpy(clone->TexturesUsed, prog->TexturesUsed, sizeof(prog->TexturesUsed)); - - if (prog->Parameters) - clone->Parameters = _mesa_clone_parameter_list(prog->Parameters); - memcpy(clone->LocalParams, prog->LocalParams, sizeof(clone->LocalParams)); - if (prog->Varying) - clone->Varying = _mesa_clone_parameter_list(prog->Varying); - if (prog->Attributes) - clone->Attributes = _mesa_clone_parameter_list(prog->Attributes); - memcpy(clone->LocalParams, prog->LocalParams, sizeof(clone->LocalParams)); - clone->IndirectRegisterFiles = prog->IndirectRegisterFiles; - clone->NumInstructions = prog->NumInstructions; - clone->NumTemporaries = prog->NumTemporaries; - clone->NumParameters = prog->NumParameters; - clone->NumAttributes = prog->NumAttributes; - clone->NumAddressRegs = prog->NumAddressRegs; - clone->NumNativeInstructions = prog->NumNativeInstructions; - clone->NumNativeTemporaries = prog->NumNativeTemporaries; - clone->NumNativeParameters = prog->NumNativeParameters; - clone->NumNativeAttributes = prog->NumNativeAttributes; - clone->NumNativeAddressRegs = prog->NumNativeAddressRegs; - clone->NumAluInstructions = prog->NumAluInstructions; - clone->NumTexInstructions = prog->NumTexInstructions; - clone->NumTexIndirections = prog->NumTexIndirections; - clone->NumNativeAluInstructions = prog->NumNativeAluInstructions; - clone->NumNativeTexInstructions = prog->NumNativeTexInstructions; - clone->NumNativeTexIndirections = prog->NumNativeTexIndirections; - - switch (prog->Target) { - case GL_VERTEX_PROGRAM_ARB: - { - const struct gl_vertex_program *vp - = (const struct gl_vertex_program *) prog; - struct gl_vertex_program *vpc = (struct gl_vertex_program *) clone; - vpc->IsPositionInvariant = vp->IsPositionInvariant; - vpc->IsNVProgram = vp->IsNVProgram; - } - break; - case GL_FRAGMENT_PROGRAM_ARB: - { - const struct gl_fragment_program *fp - = (const struct gl_fragment_program *) prog; - struct gl_fragment_program *fpc = (struct gl_fragment_program *) clone; - fpc->UsesKill = fp->UsesKill; - fpc->OriginUpperLeft = fp->OriginUpperLeft; - fpc->PixelCenterInteger = fp->PixelCenterInteger; - } - break; - case MESA_GEOMETRY_PROGRAM: - { - const struct gl_geometry_program *gp - = (const struct gl_geometry_program *) prog; - struct gl_geometry_program *gpc = (struct gl_geometry_program *) clone; - gpc->VerticesOut = gp->VerticesOut; - gpc->InputType = gp->InputType; - gpc->OutputType = gp->OutputType; - } - break; - default: - _mesa_problem(NULL, "Unexpected target in _mesa_clone_program"); - } - - return clone; -} - - -/** - * Insert 'count' NOP instructions at 'start' in the given program. - * Adjust branch targets accordingly. - */ -GLboolean -_mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count) -{ - const GLuint origLen = prog->NumInstructions; - const GLuint newLen = origLen + count; - struct prog_instruction *newInst; - GLuint i; - - /* adjust branches */ - for (i = 0; i < prog->NumInstructions; i++) { - struct prog_instruction *inst = prog->Instructions + i; - if (inst->BranchTarget > 0) { - if ((GLuint)inst->BranchTarget >= start) { - inst->BranchTarget += count; - } - } - } - - /* Alloc storage for new instructions */ - newInst = _mesa_alloc_instructions(newLen); - if (!newInst) { - return GL_FALSE; - } - - /* Copy 'start' instructions into new instruction buffer */ - _mesa_copy_instructions(newInst, prog->Instructions, start); - - /* init the new instructions */ - _mesa_init_instructions(newInst + start, count); - - /* Copy the remaining/tail instructions to new inst buffer */ - _mesa_copy_instructions(newInst + start + count, - prog->Instructions + start, - origLen - start); - - /* free old instructions */ - _mesa_free_instructions(prog->Instructions, origLen); - - /* install new instructions */ - prog->Instructions = newInst; - prog->NumInstructions = newLen; - - return GL_TRUE; -} - -/** - * Delete 'count' instructions at 'start' in the given program. - * Adjust branch targets accordingly. - */ -GLboolean -_mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count) -{ - const GLuint origLen = prog->NumInstructions; - const GLuint newLen = origLen - count; - struct prog_instruction *newInst; - GLuint i; - - /* adjust branches */ - for (i = 0; i < prog->NumInstructions; i++) { - struct prog_instruction *inst = prog->Instructions + i; - if (inst->BranchTarget > 0) { - if (inst->BranchTarget > (GLint) start) { - inst->BranchTarget -= count; - } - } - } - - /* Alloc storage for new instructions */ - newInst = _mesa_alloc_instructions(newLen); - if (!newInst) { - return GL_FALSE; - } - - /* Copy 'start' instructions into new instruction buffer */ - _mesa_copy_instructions(newInst, prog->Instructions, start); - - /* Copy the remaining/tail instructions to new inst buffer */ - _mesa_copy_instructions(newInst + start, - prog->Instructions + start + count, - newLen - start); - - /* free old instructions */ - _mesa_free_instructions(prog->Instructions, origLen); - - /* install new instructions */ - prog->Instructions = newInst; - prog->NumInstructions = newLen; - - return GL_TRUE; -} - - -/** - * Search instructions for registers that match (oldFile, oldIndex), - * replacing them with (newFile, newIndex). - */ -static void -replace_registers(struct prog_instruction *inst, GLuint numInst, - GLuint oldFile, GLuint oldIndex, - GLuint newFile, GLuint newIndex) -{ - GLuint i, j; - for (i = 0; i < numInst; i++) { - /* src regs */ - for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) { - if (inst[i].SrcReg[j].File == oldFile && - inst[i].SrcReg[j].Index == oldIndex) { - inst[i].SrcReg[j].File = newFile; - inst[i].SrcReg[j].Index = newIndex; - } - } - /* dst reg */ - if (inst[i].DstReg.File == oldFile && inst[i].DstReg.Index == oldIndex) { - inst[i].DstReg.File = newFile; - inst[i].DstReg.Index = newIndex; - } - } -} - - -/** - * Search instructions for references to program parameters. When found, - * increment the parameter index by 'offset'. - * Used when combining programs. - */ -static void -adjust_param_indexes(struct prog_instruction *inst, GLuint numInst, - GLuint offset) -{ - GLuint i, j; - for (i = 0; i < numInst; i++) { - for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) { - GLuint f = inst[i].SrcReg[j].File; - if (f == PROGRAM_CONSTANT || - f == PROGRAM_UNIFORM || - f == PROGRAM_STATE_VAR) { - inst[i].SrcReg[j].Index += offset; - } - } - } -} - - -/** - * Combine two programs into one. Fix instructions so the outputs of - * the first program go to the inputs of the second program. - */ -struct gl_program * -_mesa_combine_programs(struct gl_context *ctx, - const struct gl_program *progA, - const struct gl_program *progB) -{ - struct prog_instruction *newInst; - struct gl_program *newProg; - const GLuint lenA = progA->NumInstructions - 1; /* omit END instr */ - const GLuint lenB = progB->NumInstructions; - const GLuint numParamsA = _mesa_num_parameters(progA->Parameters); - const GLuint newLength = lenA + lenB; - GLboolean usedTemps[MAX_PROGRAM_TEMPS]; - GLuint firstTemp = 0; - GLbitfield inputsB; - GLuint i; - - ASSERT(progA->Target == progB->Target); - - newInst = _mesa_alloc_instructions(newLength); - if (!newInst) - return GL_FALSE; - - _mesa_copy_instructions(newInst, progA->Instructions, lenA); - _mesa_copy_instructions(newInst + lenA, progB->Instructions, lenB); - - /* adjust branch / instruction addresses for B's instructions */ - for (i = 0; i < lenB; i++) { - newInst[lenA + i].BranchTarget += lenA; - } - - newProg = ctx->Driver.NewProgram(ctx, progA->Target, 0); - newProg->Instructions = newInst; - newProg->NumInstructions = newLength; - - /* find used temp regs (we may need new temps below) */ - _mesa_find_used_registers(newProg, PROGRAM_TEMPORARY, - usedTemps, MAX_PROGRAM_TEMPS); - - if (newProg->Target == GL_FRAGMENT_PROGRAM_ARB) { - struct gl_fragment_program *fprogA, *fprogB, *newFprog; - GLbitfield progB_inputsRead = progB->InputsRead; - GLint progB_colorFile, progB_colorIndex; - - fprogA = (struct gl_fragment_program *) progA; - fprogB = (struct gl_fragment_program *) progB; - newFprog = (struct gl_fragment_program *) newProg; - - newFprog->UsesKill = fprogA->UsesKill || fprogB->UsesKill; - - /* We'll do a search and replace for instances - * of progB_colorFile/progB_colorIndex below... - */ - progB_colorFile = PROGRAM_INPUT; - progB_colorIndex = FRAG_ATTRIB_COL0; - - /* - * The fragment program may get color from a state var rather than - * a fragment input (vertex output) if it's constant. - * See the texenvprogram.c code. - * So, search the program's parameter list now to see if the program - * gets color from a state var instead of a conventional fragment - * input register. - */ - for (i = 0; i < progB->Parameters->NumParameters; i++) { - struct gl_program_parameter *p = &progB->Parameters->Parameters[i]; - if (p->Type == PROGRAM_STATE_VAR && - p->StateIndexes[0] == STATE_INTERNAL && - p->StateIndexes[1] == STATE_CURRENT_ATTRIB && - (int) p->StateIndexes[2] == (int) VERT_ATTRIB_COLOR0) { - progB_inputsRead |= FRAG_BIT_COL0; - progB_colorFile = PROGRAM_STATE_VAR; - progB_colorIndex = i; - break; - } - } - - /* Connect color outputs of fprogA to color inputs of fprogB, via a - * new temporary register. - */ - if ((progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) && - (progB_inputsRead & FRAG_BIT_COL0)) { - GLint tempReg = _mesa_find_free_register(usedTemps, MAX_PROGRAM_TEMPS, - firstTemp); - if (tempReg < 0) { - _mesa_problem(ctx, "No free temp regs found in " - "_mesa_combine_programs(), using 31"); - tempReg = 31; - } - firstTemp = tempReg + 1; - - /* replace writes to result.color[0] with tempReg */ - replace_registers(newInst, lenA, - PROGRAM_OUTPUT, FRAG_RESULT_COLOR, - PROGRAM_TEMPORARY, tempReg); - /* replace reads from the input color with tempReg */ - replace_registers(newInst + lenA, lenB, - progB_colorFile, progB_colorIndex, /* search for */ - PROGRAM_TEMPORARY, tempReg /* replace with */ ); - } - - /* compute combined program's InputsRead */ - inputsB = progB_inputsRead; - if (progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) { - inputsB &= ~(1 << FRAG_ATTRIB_COL0); - } - newProg->InputsRead = progA->InputsRead | inputsB; - newProg->OutputsWritten = progB->OutputsWritten; - newProg->SamplersUsed = progA->SamplersUsed | progB->SamplersUsed; - } - else { - /* vertex program */ - assert(0); /* XXX todo */ - } - - /* - * Merge parameters (uniforms, constants, etc) - */ - newProg->Parameters = _mesa_combine_parameter_lists(progA->Parameters, - progB->Parameters); - - adjust_param_indexes(newInst + lenA, lenB, numParamsA); - - - return newProg; -} - - -/** - * Populate the 'used' array with flags indicating which registers (TEMPs, - * INPUTs, OUTPUTs, etc, are used by the given program. - * \param file type of register to scan for - * \param used returns true/false flags for in use / free - * \param usedSize size of the 'used' array - */ -void -_mesa_find_used_registers(const struct gl_program *prog, - gl_register_file file, - GLboolean used[], GLuint usedSize) -{ - GLuint i, j; - - memset(used, 0, usedSize); - - for (i = 0; i < prog->NumInstructions; i++) { - const struct prog_instruction *inst = prog->Instructions + i; - const GLuint n = _mesa_num_inst_src_regs(inst->Opcode); - - if (inst->DstReg.File == file) { - ASSERT(inst->DstReg.Index < usedSize); - if(inst->DstReg.Index < usedSize) - used[inst->DstReg.Index] = GL_TRUE; - } - - for (j = 0; j < n; j++) { - if (inst->SrcReg[j].File == file) { - ASSERT(inst->SrcReg[j].Index < usedSize); - if(inst->SrcReg[j].Index < usedSize) - used[inst->SrcReg[j].Index] = GL_TRUE; - } - } - } -} - - -/** - * Scan the given 'used' register flag array for the first entry - * that's >= firstReg. - * \param used vector of flags indicating registers in use (as returned - * by _mesa_find_used_registers()) - * \param usedSize size of the 'used' array - * \param firstReg first register to start searching at - * \return index of unused register, or -1 if none. - */ -GLint -_mesa_find_free_register(const GLboolean used[], - GLuint usedSize, GLuint firstReg) -{ - GLuint i; - - assert(firstReg < usedSize); - - for (i = firstReg; i < usedSize; i++) - if (!used[i]) - return i; - - return -1; -} - - - -/** - * Check if the given register index is valid (doesn't exceed implementation- - * dependent limits). - * \return GL_TRUE if OK, GL_FALSE if bad index - */ -GLboolean -_mesa_valid_register_index(const struct gl_context *ctx, - gl_shader_type shaderType, - gl_register_file file, GLint index) -{ - const struct gl_program_constants *c; - - switch (shaderType) { - case MESA_SHADER_VERTEX: - c = &ctx->Const.VertexProgram; - break; - case MESA_SHADER_FRAGMENT: - c = &ctx->Const.FragmentProgram; - break; - case MESA_SHADER_GEOMETRY: - c = &ctx->Const.GeometryProgram; - break; - default: - _mesa_problem(ctx, - "unexpected shader type in _mesa_valid_register_index()"); - return GL_FALSE; - } - - switch (file) { - case PROGRAM_UNDEFINED: - return GL_TRUE; /* XXX or maybe false? */ - - case PROGRAM_TEMPORARY: - return index >= 0 && index < c->MaxTemps; - - case PROGRAM_ENV_PARAM: - return index >= 0 && index < c->MaxEnvParams; - - case PROGRAM_LOCAL_PARAM: - return index >= 0 && index < c->MaxLocalParams; - - case PROGRAM_NAMED_PARAM: - return index >= 0 && index < c->MaxParameters; - - case PROGRAM_UNIFORM: - case PROGRAM_STATE_VAR: - /* aka constant buffer */ - return index >= 0 && index < c->MaxUniformComponents / 4; - - case PROGRAM_CONSTANT: - /* constant buffer w/ possible relative negative addressing */ - return (index > (int) c->MaxUniformComponents / -4 && - index < c->MaxUniformComponents / 4); - - case PROGRAM_INPUT: - if (index < 0) - return GL_FALSE; - - switch (shaderType) { - case MESA_SHADER_VERTEX: - return index < VERT_ATTRIB_GENERIC0 + c->MaxAttribs; - case MESA_SHADER_FRAGMENT: - return index < FRAG_ATTRIB_VAR0 + ctx->Const.MaxVarying; - case MESA_SHADER_GEOMETRY: - return index < GEOM_ATTRIB_VAR0 + ctx->Const.MaxVarying; - default: - return GL_FALSE; - } - - case PROGRAM_OUTPUT: - if (index < 0) - return GL_FALSE; - - switch (shaderType) { - case MESA_SHADER_VERTEX: - return index < VERT_RESULT_VAR0 + ctx->Const.MaxVarying; - case MESA_SHADER_FRAGMENT: - return index < FRAG_RESULT_DATA0 + ctx->Const.MaxDrawBuffers; - case MESA_SHADER_GEOMETRY: - return index < GEOM_RESULT_VAR0 + ctx->Const.MaxVarying; - default: - return GL_FALSE; - } - - case PROGRAM_ADDRESS: - return index >= 0 && index < c->MaxAddressRegs; - - default: - _mesa_problem(ctx, - "unexpected register file in _mesa_valid_register_index()"); - return GL_FALSE; - } -} - - - -/** - * "Post-process" a GPU program. This is intended to be used for debugging. - * Example actions include no-op'ing instructions or changing instruction - * behaviour. - */ -void -_mesa_postprocess_program(struct gl_context *ctx, struct gl_program *prog) -{ - static const GLfloat white[4] = { 0.5, 0.5, 0.5, 0.5 }; - GLuint i; - GLuint whiteSwizzle; - GLint whiteIndex = _mesa_add_unnamed_constant(prog->Parameters, - (gl_constant_value *) white, - 4, &whiteSwizzle); - - (void) whiteIndex; - - for (i = 0; i < prog->NumInstructions; i++) { - struct prog_instruction *inst = prog->Instructions + i; - const GLuint n = _mesa_num_inst_src_regs(inst->Opcode); - - (void) n; - - if (_mesa_is_tex_instruction(inst->Opcode)) { -#if 0 - /* replace TEX/TXP/TXB with MOV */ - inst->Opcode = OPCODE_MOV; - inst->DstReg.WriteMask = WRITEMASK_XYZW; - inst->SrcReg[0].Swizzle = SWIZZLE_XYZW; - inst->SrcReg[0].Negate = NEGATE_NONE; -#endif - -#if 0 - /* disable shadow texture mode */ - inst->TexShadow = 0; -#endif - } - - if (inst->Opcode == OPCODE_TXP) { -#if 0 - inst->Opcode = OPCODE_MOV; - inst->DstReg.WriteMask = WRITEMASK_XYZW; - inst->SrcReg[0].File = PROGRAM_CONSTANT; - inst->SrcReg[0].Index = whiteIndex; - inst->SrcReg[0].Swizzle = SWIZZLE_XYZW; - inst->SrcReg[0].Negate = NEGATE_NONE; -#endif -#if 0 - inst->TexShadow = 0; -#endif -#if 0 - inst->Opcode = OPCODE_TEX; - inst->TexShadow = 0; -#endif - } - - } -} +/* + * Mesa 3-D graphics library + * Version: 6.5.3 + * + * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/** + * \file program.c + * Vertex and fragment program support functions. + * \author Brian Paul + */ + + +#include "main/glheader.h" +#include "main/context.h" +#include "main/hash.h" +#include "main/mfeatures.h" +#include "program.h" +#include "prog_cache.h" +#include "prog_parameter.h" +#include "prog_instruction.h" + + +/** + * A pointer to this dummy program is put into the hash table when + * glGenPrograms is called. + */ +struct gl_program _mesa_DummyProgram; + + +/** + * Init context's vertex/fragment program state + */ +void +_mesa_init_program(struct gl_context *ctx) +{ + GLuint i; + + /* + * If this assertion fails, we need to increase the field + * size for register indexes (see INST_INDEX_BITS). + */ + ASSERT(ctx->Const.VertexProgram.MaxUniformComponents / 4 + <= (1 << INST_INDEX_BITS)); + ASSERT(ctx->Const.FragmentProgram.MaxUniformComponents / 4 + <= (1 << INST_INDEX_BITS)); + + ASSERT(ctx->Const.VertexProgram.MaxTemps <= (1 << INST_INDEX_BITS)); + ASSERT(ctx->Const.VertexProgram.MaxLocalParams <= (1 << INST_INDEX_BITS)); + ASSERT(ctx->Const.FragmentProgram.MaxTemps <= (1 << INST_INDEX_BITS)); + ASSERT(ctx->Const.FragmentProgram.MaxLocalParams <= (1 << INST_INDEX_BITS)); + + ASSERT(ctx->Const.VertexProgram.MaxUniformComponents <= 4 * MAX_UNIFORMS); + ASSERT(ctx->Const.FragmentProgram.MaxUniformComponents <= 4 * MAX_UNIFORMS); + + ASSERT(ctx->Const.VertexProgram.MaxAddressOffset <= (1 << INST_INDEX_BITS)); + ASSERT(ctx->Const.FragmentProgram.MaxAddressOffset <= (1 << INST_INDEX_BITS)); + + /* If this fails, increase prog_instruction::TexSrcUnit size */ + ASSERT(MAX_TEXTURE_UNITS <= (1 << 5)); + + /* If this fails, increase prog_instruction::TexSrcTarget size */ + ASSERT(NUM_TEXTURE_TARGETS <= (1 << 3)); + + ctx->Program.ErrorPos = -1; + ctx->Program.ErrorString = _mesa_strdup(""); + +#if FEATURE_NV_vertex_program || FEATURE_ARB_vertex_program + ctx->VertexProgram.Enabled = GL_FALSE; +#if FEATURE_es2_glsl + ctx->VertexProgram.PointSizeEnabled = + (ctx->API == API_OPENGLES2) ? GL_TRUE : GL_FALSE; +#else + ctx->VertexProgram.PointSizeEnabled = GL_FALSE; +#endif + ctx->VertexProgram.TwoSideEnabled = GL_FALSE; + _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current, + ctx->Shared->DefaultVertexProgram); + assert(ctx->VertexProgram.Current); + for (i = 0; i < MAX_NV_VERTEX_PROGRAM_PARAMS / 4; i++) { + ctx->VertexProgram.TrackMatrix[i] = GL_NONE; + ctx->VertexProgram.TrackMatrixTransform[i] = GL_IDENTITY_NV; + } + ctx->VertexProgram.Cache = _mesa_new_program_cache(); +#endif + +#if FEATURE_NV_fragment_program || FEATURE_ARB_fragment_program + ctx->FragmentProgram.Enabled = GL_FALSE; + _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current, + ctx->Shared->DefaultFragmentProgram); + assert(ctx->FragmentProgram.Current); + ctx->FragmentProgram.Cache = _mesa_new_program_cache(); +#endif + +#if FEATURE_ARB_geometry_shader4 + ctx->GeometryProgram.Enabled = GL_FALSE; + /* right now by default we don't have a geometry program */ + _mesa_reference_geomprog(ctx, &ctx->GeometryProgram.Current, + NULL); + ctx->GeometryProgram.Cache = _mesa_new_program_cache(); +#endif + + /* XXX probably move this stuff */ +#if FEATURE_ATI_fragment_shader + ctx->ATIFragmentShader.Enabled = GL_FALSE; + ctx->ATIFragmentShader.Current = ctx->Shared->DefaultFragmentShader; + assert(ctx->ATIFragmentShader.Current); + ctx->ATIFragmentShader.Current->RefCount++; +#endif +} + + +/** + * Free a context's vertex/fragment program state + */ +void +_mesa_free_program_data(struct gl_context *ctx) +{ +#if FEATURE_NV_vertex_program || FEATURE_ARB_vertex_program + _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current, NULL); + _mesa_delete_program_cache(ctx, ctx->VertexProgram.Cache); +#endif +#if FEATURE_NV_fragment_program || FEATURE_ARB_fragment_program + _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current, NULL); + _mesa_delete_program_cache(ctx, ctx->FragmentProgram.Cache); +#endif +#if FEATURE_ARB_geometry_shader4 + _mesa_reference_geomprog(ctx, &ctx->GeometryProgram.Current, NULL); + _mesa_delete_program_cache(ctx, ctx->GeometryProgram.Cache); +#endif + /* XXX probably move this stuff */ +#if FEATURE_ATI_fragment_shader + if (ctx->ATIFragmentShader.Current) { + ctx->ATIFragmentShader.Current->RefCount--; + if (ctx->ATIFragmentShader.Current->RefCount <= 0) { + free(ctx->ATIFragmentShader.Current); + } + } +#endif + free((void *) ctx->Program.ErrorString); +} + + +/** + * Update the default program objects in the given context to reference those + * specified in the shared state and release those referencing the old + * shared state. + */ +void +_mesa_update_default_objects_program(struct gl_context *ctx) +{ +#if FEATURE_NV_vertex_program || FEATURE_ARB_vertex_program + _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current, + (struct gl_vertex_program *) + ctx->Shared->DefaultVertexProgram); + assert(ctx->VertexProgram.Current); +#endif + +#if FEATURE_NV_fragment_program || FEATURE_ARB_fragment_program + _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current, + (struct gl_fragment_program *) + ctx->Shared->DefaultFragmentProgram); + assert(ctx->FragmentProgram.Current); +#endif + +#if FEATURE_ARB_geometry_shader4 + _mesa_reference_geomprog(ctx, &ctx->GeometryProgram.Current, + (struct gl_geometry_program *) + ctx->Shared->DefaultGeometryProgram); +#endif + + /* XXX probably move this stuff */ +#if FEATURE_ATI_fragment_shader + if (ctx->ATIFragmentShader.Current) { + ctx->ATIFragmentShader.Current->RefCount--; + if (ctx->ATIFragmentShader.Current->RefCount <= 0) { + free(ctx->ATIFragmentShader.Current); + } + } + ctx->ATIFragmentShader.Current = (struct ati_fragment_shader *) ctx->Shared->DefaultFragmentShader; + assert(ctx->ATIFragmentShader.Current); + ctx->ATIFragmentShader.Current->RefCount++; +#endif +} + + +/** + * Set the vertex/fragment program error state (position and error string). + * This is generally called from within the parsers. + */ +void +_mesa_set_program_error(struct gl_context *ctx, GLint pos, const char *string) +{ + ctx->Program.ErrorPos = pos; + free((void *) ctx->Program.ErrorString); + if (!string) + string = ""; + ctx->Program.ErrorString = _mesa_strdup(string); +} + + +/** + * Find the line number and column for 'pos' within 'string'. + * Return a copy of the line which contains 'pos'. Free the line with + * free(). + * \param string the program string + * \param pos the position within the string + * \param line returns the line number corresponding to 'pos'. + * \param col returns the column number corresponding to 'pos'. + * \return copy of the line containing 'pos'. + */ +const GLubyte * +_mesa_find_line_column(const GLubyte *string, const GLubyte *pos, + GLint *line, GLint *col) +{ + const GLubyte *lineStart = string; + const GLubyte *p = string; + GLubyte *s; + int len; + + *line = 1; + + while (p != pos) { + if (*p == (GLubyte) '\n') { + (*line)++; + lineStart = p + 1; + } + p++; + } + + *col = (pos - lineStart) + 1; + + /* return copy of this line */ + while (*p != 0 && *p != '\n') + p++; + len = p - lineStart; + s = (GLubyte *) malloc(len + 1); + memcpy(s, lineStart, len); + s[len] = 0; + + return s; +} + + +/** + * Initialize a new vertex/fragment program object. + */ +static struct gl_program * +_mesa_init_program_struct( struct gl_context *ctx, struct gl_program *prog, + GLenum target, GLuint id) +{ + (void) ctx; + if (prog) { + GLuint i; + memset(prog, 0, sizeof(*prog)); + prog->Id = id; + prog->Target = target; + prog->Resident = GL_TRUE; + prog->RefCount = 1; + prog->Format = GL_PROGRAM_FORMAT_ASCII_ARB; + + /* default mapping from samplers to texture units */ + for (i = 0; i < MAX_SAMPLERS; i++) + prog->SamplerUnits[i] = i; + } + + return prog; +} + + +/** + * Initialize a new fragment program object. + */ +struct gl_program * +_mesa_init_fragment_program( struct gl_context *ctx, struct gl_fragment_program *prog, + GLenum target, GLuint id) +{ + if (prog) + return _mesa_init_program_struct( ctx, &prog->Base, target, id ); + else + return NULL; +} + + +/** + * Initialize a new vertex program object. + */ +struct gl_program * +_mesa_init_vertex_program( struct gl_context *ctx, struct gl_vertex_program *prog, + GLenum target, GLuint id) +{ + if (prog) + return _mesa_init_program_struct( ctx, &prog->Base, target, id ); + else + return NULL; +} + + +/** + * Initialize a new geometry program object. + */ +struct gl_program * +_mesa_init_geometry_program( struct gl_context *ctx, struct gl_geometry_program *prog, + GLenum target, GLuint id) +{ + if (prog) + return _mesa_init_program_struct( ctx, &prog->Base, target, id ); + else + return NULL; +} + + +/** + * Allocate and initialize a new fragment/vertex program object but + * don't put it into the program hash table. Called via + * ctx->Driver.NewProgram. May be overridden (ie. replaced) by a + * device driver function to implement OO deriviation with additional + * types not understood by this function. + * + * \param ctx context + * \param id program id/number + * \param target program target/type + * \return pointer to new program object + */ +struct gl_program * +_mesa_new_program(struct gl_context *ctx, GLenum target, GLuint id) +{ + struct gl_program *prog; + switch (target) { + case GL_VERTEX_PROGRAM_ARB: /* == GL_VERTEX_PROGRAM_NV */ + case GL_VERTEX_STATE_PROGRAM_NV: + prog = _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program), + target, id ); + break; + case GL_FRAGMENT_PROGRAM_NV: + case GL_FRAGMENT_PROGRAM_ARB: + prog =_mesa_init_fragment_program(ctx, + CALLOC_STRUCT(gl_fragment_program), + target, id ); + break; + case MESA_GEOMETRY_PROGRAM: + prog = _mesa_init_geometry_program(ctx, + CALLOC_STRUCT(gl_geometry_program), + target, id); + break; + default: + _mesa_problem(ctx, "bad target in _mesa_new_program"); + prog = NULL; + } + return prog; +} + + +/** + * Delete a program and remove it from the hash table, ignoring the + * reference count. + * Called via ctx->Driver.DeleteProgram. May be wrapped (OO deriviation) + * by a device driver function. + */ +void +_mesa_delete_program(struct gl_context *ctx, struct gl_program *prog) +{ + (void) ctx; + ASSERT(prog); + ASSERT(prog->RefCount==0); + + if (prog == &_mesa_DummyProgram) + return; + + if (prog->String) + free(prog->String); + + if (prog->Instructions) { + _mesa_free_instructions(prog->Instructions, prog->NumInstructions); + } + if (prog->Parameters) { + _mesa_free_parameter_list(prog->Parameters); + } + if (prog->Varying) { + _mesa_free_parameter_list(prog->Varying); + } + if (prog->Attributes) { + _mesa_free_parameter_list(prog->Attributes); + } + + free(prog); +} + + +/** + * Return the gl_program object for a given ID. + * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of + * casts elsewhere. + */ +struct gl_program * +_mesa_lookup_program(struct gl_context *ctx, GLuint id) +{ + if (id) + return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id); + else + return NULL; +} + + +/** + * Reference counting for vertex/fragment programs + * This is normally only called from the _mesa_reference_program() macro + * when there's a real pointer change. + */ +void +_mesa_reference_program_(struct gl_context *ctx, + struct gl_program **ptr, + struct gl_program *prog) +{ +#ifndef NDEBUG + assert(ptr); + if (*ptr && prog) { + /* sanity check */ + if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB) + ASSERT(prog->Target == GL_VERTEX_PROGRAM_ARB); + else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB) + ASSERT(prog->Target == GL_FRAGMENT_PROGRAM_ARB || + prog->Target == GL_FRAGMENT_PROGRAM_NV); + else if ((*ptr)->Target == MESA_GEOMETRY_PROGRAM) + ASSERT(prog->Target == MESA_GEOMETRY_PROGRAM); + } +#endif + + if (*ptr) { + GLboolean deleteFlag; + + /*_glthread_LOCK_MUTEX((*ptr)->Mutex);*/ +#if 0 + printf("Program %p ID=%u Target=%s Refcount-- to %d\n", + *ptr, (*ptr)->Id, + ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB ? "VP" : + ((*ptr)->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")), + (*ptr)->RefCount - 1); +#endif + ASSERT((*ptr)->RefCount > 0); + (*ptr)->RefCount--; + + deleteFlag = ((*ptr)->RefCount == 0); + /*_glthread_UNLOCK_MUTEX((*ptr)->Mutex);*/ + + if (deleteFlag) { + ASSERT(ctx); + ctx->Driver.DeleteProgram(ctx, *ptr); + } + + *ptr = NULL; + } + + assert(!*ptr); + if (prog) { + /*_glthread_LOCK_MUTEX(prog->Mutex);*/ + prog->RefCount++; +#if 0 + printf("Program %p ID=%u Target=%s Refcount++ to %d\n", + prog, prog->Id, + (prog->Target == GL_VERTEX_PROGRAM_ARB ? "VP" : + (prog->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")), + prog->RefCount); +#endif + /*_glthread_UNLOCK_MUTEX(prog->Mutex);*/ + } + + *ptr = prog; +} + + +/** + * Return a copy of a program. + * XXX Problem here if the program object is actually OO-derivation + * made by a device driver. + */ +struct gl_program * +_mesa_clone_program(struct gl_context *ctx, const struct gl_program *prog) +{ + struct gl_program *clone; + + clone = ctx->Driver.NewProgram(ctx, prog->Target, prog->Id); + if (!clone) + return NULL; + + assert(clone->Target == prog->Target); + assert(clone->RefCount == 1); + + clone->String = (GLubyte *) _mesa_strdup((char *) prog->String); + clone->Format = prog->Format; + clone->Instructions = _mesa_alloc_instructions(prog->NumInstructions); + if (!clone->Instructions) { + _mesa_reference_program(ctx, &clone, NULL); + return NULL; + } + _mesa_copy_instructions(clone->Instructions, prog->Instructions, + prog->NumInstructions); + clone->InputsRead = prog->InputsRead; + clone->OutputsWritten = prog->OutputsWritten; + clone->SamplersUsed = prog->SamplersUsed; + clone->ShadowSamplers = prog->ShadowSamplers; + memcpy(clone->TexturesUsed, prog->TexturesUsed, sizeof(prog->TexturesUsed)); + + if (prog->Parameters) + clone->Parameters = _mesa_clone_parameter_list(prog->Parameters); + memcpy(clone->LocalParams, prog->LocalParams, sizeof(clone->LocalParams)); + if (prog->Varying) + clone->Varying = _mesa_clone_parameter_list(prog->Varying); + if (prog->Attributes) + clone->Attributes = _mesa_clone_parameter_list(prog->Attributes); + memcpy(clone->LocalParams, prog->LocalParams, sizeof(clone->LocalParams)); + clone->IndirectRegisterFiles = prog->IndirectRegisterFiles; + clone->NumInstructions = prog->NumInstructions; + clone->NumTemporaries = prog->NumTemporaries; + clone->NumParameters = prog->NumParameters; + clone->NumAttributes = prog->NumAttributes; + clone->NumAddressRegs = prog->NumAddressRegs; + clone->NumNativeInstructions = prog->NumNativeInstructions; + clone->NumNativeTemporaries = prog->NumNativeTemporaries; + clone->NumNativeParameters = prog->NumNativeParameters; + clone->NumNativeAttributes = prog->NumNativeAttributes; + clone->NumNativeAddressRegs = prog->NumNativeAddressRegs; + clone->NumAluInstructions = prog->NumAluInstructions; + clone->NumTexInstructions = prog->NumTexInstructions; + clone->NumTexIndirections = prog->NumTexIndirections; + clone->NumNativeAluInstructions = prog->NumNativeAluInstructions; + clone->NumNativeTexInstructions = prog->NumNativeTexInstructions; + clone->NumNativeTexIndirections = prog->NumNativeTexIndirections; + + switch (prog->Target) { + case GL_VERTEX_PROGRAM_ARB: + { + const struct gl_vertex_program *vp + = (const struct gl_vertex_program *) prog; + struct gl_vertex_program *vpc = (struct gl_vertex_program *) clone; + vpc->IsPositionInvariant = vp->IsPositionInvariant; + vpc->IsNVProgram = vp->IsNVProgram; + } + break; + case GL_FRAGMENT_PROGRAM_ARB: + { + const struct gl_fragment_program *fp + = (const struct gl_fragment_program *) prog; + struct gl_fragment_program *fpc = (struct gl_fragment_program *) clone; + fpc->UsesKill = fp->UsesKill; + fpc->OriginUpperLeft = fp->OriginUpperLeft; + fpc->PixelCenterInteger = fp->PixelCenterInteger; + } + break; + case MESA_GEOMETRY_PROGRAM: + { + const struct gl_geometry_program *gp + = (const struct gl_geometry_program *) prog; + struct gl_geometry_program *gpc = (struct gl_geometry_program *) clone; + gpc->VerticesOut = gp->VerticesOut; + gpc->InputType = gp->InputType; + gpc->OutputType = gp->OutputType; + } + break; + default: + _mesa_problem(NULL, "Unexpected target in _mesa_clone_program"); + } + + return clone; +} + + +/** + * Insert 'count' NOP instructions at 'start' in the given program. + * Adjust branch targets accordingly. + */ +GLboolean +_mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count) +{ + const GLuint origLen = prog->NumInstructions; + const GLuint newLen = origLen + count; + struct prog_instruction *newInst; + GLuint i; + + /* adjust branches */ + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = prog->Instructions + i; + if (inst->BranchTarget > 0) { + if ((GLuint)inst->BranchTarget >= start) { + inst->BranchTarget += count; + } + } + } + + /* Alloc storage for new instructions */ + newInst = _mesa_alloc_instructions(newLen); + if (!newInst) { + return GL_FALSE; + } + + /* Copy 'start' instructions into new instruction buffer */ + _mesa_copy_instructions(newInst, prog->Instructions, start); + + /* init the new instructions */ + _mesa_init_instructions(newInst + start, count); + + /* Copy the remaining/tail instructions to new inst buffer */ + _mesa_copy_instructions(newInst + start + count, + prog->Instructions + start, + origLen - start); + + /* free old instructions */ + _mesa_free_instructions(prog->Instructions, origLen); + + /* install new instructions */ + prog->Instructions = newInst; + prog->NumInstructions = newLen; + + return GL_TRUE; +} + +/** + * Delete 'count' instructions at 'start' in the given program. + * Adjust branch targets accordingly. + */ +GLboolean +_mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count) +{ + const GLuint origLen = prog->NumInstructions; + const GLuint newLen = origLen - count; + struct prog_instruction *newInst; + GLuint i; + + /* adjust branches */ + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = prog->Instructions + i; + if (inst->BranchTarget > 0) { + if (inst->BranchTarget > (GLint) start) { + inst->BranchTarget -= count; + } + } + } + + /* Alloc storage for new instructions */ + newInst = _mesa_alloc_instructions(newLen); + if (!newInst) { + return GL_FALSE; + } + + /* Copy 'start' instructions into new instruction buffer */ + _mesa_copy_instructions(newInst, prog->Instructions, start); + + /* Copy the remaining/tail instructions to new inst buffer */ + _mesa_copy_instructions(newInst + start, + prog->Instructions + start + count, + newLen - start); + + /* free old instructions */ + _mesa_free_instructions(prog->Instructions, origLen); + + /* install new instructions */ + prog->Instructions = newInst; + prog->NumInstructions = newLen; + + return GL_TRUE; +} + + +/** + * Search instructions for registers that match (oldFile, oldIndex), + * replacing them with (newFile, newIndex). + */ +static void +replace_registers(struct prog_instruction *inst, GLuint numInst, + GLuint oldFile, GLuint oldIndex, + GLuint newFile, GLuint newIndex) +{ + GLuint i, j; + for (i = 0; i < numInst; i++) { + /* src regs */ + for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) { + if (inst[i].SrcReg[j].File == oldFile && + inst[i].SrcReg[j].Index == oldIndex) { + inst[i].SrcReg[j].File = newFile; + inst[i].SrcReg[j].Index = newIndex; + } + } + /* dst reg */ + if (inst[i].DstReg.File == oldFile && inst[i].DstReg.Index == oldIndex) { + inst[i].DstReg.File = newFile; + inst[i].DstReg.Index = newIndex; + } + } +} + + +/** + * Search instructions for references to program parameters. When found, + * increment the parameter index by 'offset'. + * Used when combining programs. + */ +static void +adjust_param_indexes(struct prog_instruction *inst, GLuint numInst, + GLuint offset) +{ + GLuint i, j; + for (i = 0; i < numInst; i++) { + for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) { + GLuint f = inst[i].SrcReg[j].File; + if (f == PROGRAM_CONSTANT || + f == PROGRAM_UNIFORM || + f == PROGRAM_STATE_VAR) { + inst[i].SrcReg[j].Index += offset; + } + } + } +} + + +/** + * Combine two programs into one. Fix instructions so the outputs of + * the first program go to the inputs of the second program. + */ +struct gl_program * +_mesa_combine_programs(struct gl_context *ctx, + const struct gl_program *progA, + const struct gl_program *progB) +{ + struct prog_instruction *newInst; + struct gl_program *newProg; + const GLuint lenA = progA->NumInstructions - 1; /* omit END instr */ + const GLuint lenB = progB->NumInstructions; + const GLuint numParamsA = _mesa_num_parameters(progA->Parameters); + const GLuint newLength = lenA + lenB; + GLboolean usedTemps[MAX_PROGRAM_TEMPS]; + GLuint firstTemp = 0; + GLbitfield inputsB; + GLuint i; + + ASSERT(progA->Target == progB->Target); + + newInst = _mesa_alloc_instructions(newLength); + if (!newInst) + return GL_FALSE; + + _mesa_copy_instructions(newInst, progA->Instructions, lenA); + _mesa_copy_instructions(newInst + lenA, progB->Instructions, lenB); + + /* adjust branch / instruction addresses for B's instructions */ + for (i = 0; i < lenB; i++) { + newInst[lenA + i].BranchTarget += lenA; + } + + newProg = ctx->Driver.NewProgram(ctx, progA->Target, 0); + newProg->Instructions = newInst; + newProg->NumInstructions = newLength; + + /* find used temp regs (we may need new temps below) */ + _mesa_find_used_registers(newProg, PROGRAM_TEMPORARY, + usedTemps, MAX_PROGRAM_TEMPS); + + if (newProg->Target == GL_FRAGMENT_PROGRAM_ARB) { + struct gl_fragment_program *fprogA, *fprogB, *newFprog; + GLbitfield progB_inputsRead = progB->InputsRead; + GLint progB_colorFile, progB_colorIndex; + + fprogA = (struct gl_fragment_program *) progA; + fprogB = (struct gl_fragment_program *) progB; + newFprog = (struct gl_fragment_program *) newProg; + + newFprog->UsesKill = fprogA->UsesKill || fprogB->UsesKill; + + /* We'll do a search and replace for instances + * of progB_colorFile/progB_colorIndex below... + */ + progB_colorFile = PROGRAM_INPUT; + progB_colorIndex = FRAG_ATTRIB_COL0; + + /* + * The fragment program may get color from a state var rather than + * a fragment input (vertex output) if it's constant. + * See the texenvprogram.c code. + * So, search the program's parameter list now to see if the program + * gets color from a state var instead of a conventional fragment + * input register. + */ + for (i = 0; i < progB->Parameters->NumParameters; i++) { + struct gl_program_parameter *p = &progB->Parameters->Parameters[i]; + if (p->Type == PROGRAM_STATE_VAR && + p->StateIndexes[0] == STATE_INTERNAL && + p->StateIndexes[1] == STATE_CURRENT_ATTRIB && + (int) p->StateIndexes[2] == (int) VERT_ATTRIB_COLOR0) { + progB_inputsRead |= FRAG_BIT_COL0; + progB_colorFile = PROGRAM_STATE_VAR; + progB_colorIndex = i; + break; + } + } + + /* Connect color outputs of fprogA to color inputs of fprogB, via a + * new temporary register. + */ + if ((progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) && + (progB_inputsRead & FRAG_BIT_COL0)) { + GLint tempReg = _mesa_find_free_register(usedTemps, MAX_PROGRAM_TEMPS, + firstTemp); + if (tempReg < 0) { + _mesa_problem(ctx, "No free temp regs found in " + "_mesa_combine_programs(), using 31"); + tempReg = 31; + } + firstTemp = tempReg + 1; + + /* replace writes to result.color[0] with tempReg */ + replace_registers(newInst, lenA, + PROGRAM_OUTPUT, FRAG_RESULT_COLOR, + PROGRAM_TEMPORARY, tempReg); + /* replace reads from the input color with tempReg */ + replace_registers(newInst + lenA, lenB, + progB_colorFile, progB_colorIndex, /* search for */ + PROGRAM_TEMPORARY, tempReg /* replace with */ ); + } + + /* compute combined program's InputsRead */ + inputsB = progB_inputsRead; + if (progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) { + inputsB &= ~(1 << FRAG_ATTRIB_COL0); + } + newProg->InputsRead = progA->InputsRead | inputsB; + newProg->OutputsWritten = progB->OutputsWritten; + newProg->SamplersUsed = progA->SamplersUsed | progB->SamplersUsed; + } + else { + /* vertex program */ + assert(0); /* XXX todo */ + } + + /* + * Merge parameters (uniforms, constants, etc) + */ + newProg->Parameters = _mesa_combine_parameter_lists(progA->Parameters, + progB->Parameters); + + adjust_param_indexes(newInst + lenA, lenB, numParamsA); + + + return newProg; +} + + +/** + * Populate the 'used' array with flags indicating which registers (TEMPs, + * INPUTs, OUTPUTs, etc, are used by the given program. + * \param file type of register to scan for + * \param used returns true/false flags for in use / free + * \param usedSize size of the 'used' array + */ +void +_mesa_find_used_registers(const struct gl_program *prog, + gl_register_file file, + GLboolean used[], GLuint usedSize) +{ + GLuint i, j; + + memset(used, 0, usedSize); + + for (i = 0; i < prog->NumInstructions; i++) { + const struct prog_instruction *inst = prog->Instructions + i; + const GLuint n = _mesa_num_inst_src_regs(inst->Opcode); + + if (inst->DstReg.File == file) { + ASSERT(inst->DstReg.Index < usedSize); + if(inst->DstReg.Index < usedSize) + used[inst->DstReg.Index] = GL_TRUE; + } + + for (j = 0; j < n; j++) { + if (inst->SrcReg[j].File == file) { + ASSERT(inst->SrcReg[j].Index < usedSize); + if(inst->SrcReg[j].Index < usedSize) + used[inst->SrcReg[j].Index] = GL_TRUE; + } + } + } +} + + +/** + * Scan the given 'used' register flag array for the first entry + * that's >= firstReg. + * \param used vector of flags indicating registers in use (as returned + * by _mesa_find_used_registers()) + * \param usedSize size of the 'used' array + * \param firstReg first register to start searching at + * \return index of unused register, or -1 if none. + */ +GLint +_mesa_find_free_register(const GLboolean used[], + GLuint usedSize, GLuint firstReg) +{ + GLuint i; + + assert(firstReg < usedSize); + + for (i = firstReg; i < usedSize; i++) + if (!used[i]) + return i; + + return -1; +} + + + +/** + * Check if the given register index is valid (doesn't exceed implementation- + * dependent limits). + * \return GL_TRUE if OK, GL_FALSE if bad index + */ +GLboolean +_mesa_valid_register_index(const struct gl_context *ctx, + gl_shader_type shaderType, + gl_register_file file, GLint index) +{ + const struct gl_program_constants *c; + + switch (shaderType) { + case MESA_SHADER_VERTEX: + c = &ctx->Const.VertexProgram; + break; + case MESA_SHADER_FRAGMENT: + c = &ctx->Const.FragmentProgram; + break; + case MESA_SHADER_GEOMETRY: + c = &ctx->Const.GeometryProgram; + break; + default: + _mesa_problem(ctx, + "unexpected shader type in _mesa_valid_register_index()"); + return GL_FALSE; + } + + switch (file) { + case PROGRAM_UNDEFINED: + return GL_TRUE; /* XXX or maybe false? */ + + case PROGRAM_TEMPORARY: + return index >= 0 && index < c->MaxTemps; + + case PROGRAM_ENV_PARAM: + return index >= 0 && index < c->MaxEnvParams; + + case PROGRAM_LOCAL_PARAM: + return index >= 0 && index < c->MaxLocalParams; + + case PROGRAM_NAMED_PARAM: + return index >= 0 && index < c->MaxParameters; + + case PROGRAM_UNIFORM: + case PROGRAM_STATE_VAR: + /* aka constant buffer */ + return index >= 0 && index < c->MaxUniformComponents / 4; + + case PROGRAM_CONSTANT: + /* constant buffer w/ possible relative negative addressing */ + return (index > (int) c->MaxUniformComponents / -4 && + index < c->MaxUniformComponents / 4); + + case PROGRAM_INPUT: + if (index < 0) + return GL_FALSE; + + switch (shaderType) { + case MESA_SHADER_VERTEX: + return index < VERT_ATTRIB_GENERIC0 + c->MaxAttribs; + case MESA_SHADER_FRAGMENT: + return index < FRAG_ATTRIB_VAR0 + ctx->Const.MaxVarying; + case MESA_SHADER_GEOMETRY: + return index < GEOM_ATTRIB_VAR0 + ctx->Const.MaxVarying; + default: + return GL_FALSE; + } + + case PROGRAM_OUTPUT: + if (index < 0) + return GL_FALSE; + + switch (shaderType) { + case MESA_SHADER_VERTEX: + return index < VERT_RESULT_VAR0 + ctx->Const.MaxVarying; + case MESA_SHADER_FRAGMENT: + return index < FRAG_RESULT_DATA0 + ctx->Const.MaxDrawBuffers; + case MESA_SHADER_GEOMETRY: + return index < GEOM_RESULT_VAR0 + ctx->Const.MaxVarying; + default: + return GL_FALSE; + } + + case PROGRAM_ADDRESS: + return index >= 0 && index < c->MaxAddressRegs; + + default: + _mesa_problem(ctx, + "unexpected register file in _mesa_valid_register_index()"); + return GL_FALSE; + } +} + + + +/** + * "Post-process" a GPU program. This is intended to be used for debugging. + * Example actions include no-op'ing instructions or changing instruction + * behaviour. + */ +void +_mesa_postprocess_program(struct gl_context *ctx, struct gl_program *prog) +{ + static const GLfloat white[4] = { 0.5, 0.5, 0.5, 0.5 }; + GLuint i; + GLuint whiteSwizzle; + GLint whiteIndex = _mesa_add_unnamed_constant(prog->Parameters, + (gl_constant_value *) white, + 4, &whiteSwizzle); + + (void) whiteIndex; + + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = prog->Instructions + i; + const GLuint n = _mesa_num_inst_src_regs(inst->Opcode); + + (void) n; + + if (_mesa_is_tex_instruction(inst->Opcode)) { +#if 0 + /* replace TEX/TXP/TXB with MOV */ + inst->Opcode = OPCODE_MOV; + inst->DstReg.WriteMask = WRITEMASK_XYZW; + inst->SrcReg[0].Swizzle = SWIZZLE_XYZW; + inst->SrcReg[0].Negate = NEGATE_NONE; +#endif + +#if 0 + /* disable shadow texture mode */ + inst->TexShadow = 0; +#endif + } + + if (inst->Opcode == OPCODE_TXP) { +#if 0 + inst->Opcode = OPCODE_MOV; + inst->DstReg.WriteMask = WRITEMASK_XYZW; + inst->SrcReg[0].File = PROGRAM_CONSTANT; + inst->SrcReg[0].Index = whiteIndex; + inst->SrcReg[0].Swizzle = SWIZZLE_XYZW; + inst->SrcReg[0].Negate = NEGATE_NONE; +#endif +#if 0 + inst->TexShadow = 0; +#endif +#if 0 + inst->Opcode = OPCODE_TEX; + inst->TexShadow = 0; +#endif + } + + } +} diff --git a/mesalib/src/mesa/program/program_parse.y b/mesalib/src/mesa/program/program_parse.y index 4efdc0184..dec35038b 100644 --- a/mesalib/src/mesa/program/program_parse.y +++ b/mesalib/src/mesa/program/program_parse.y @@ -1,2806 +1,2806 @@ -%{ -/* - * Copyright © 2009 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER - * DEALINGS IN THE SOFTWARE. - */ -#include -#include -#include - -#include "main/mtypes.h" -#include "main/imports.h" -#include "program/program.h" -#include "program/prog_parameter.h" -#include "program/prog_parameter_layout.h" -#include "program/prog_statevars.h" -#include "program/prog_instruction.h" - -#include "program/symbol_table.h" -#include "program/program_parser.h" - -extern void *yy_scan_string(char *); -extern void yy_delete_buffer(void *); - -static struct asm_symbol *declare_variable(struct asm_parser_state *state, - char *name, enum asm_type t, struct YYLTYPE *locp); - -static int add_state_reference(struct gl_program_parameter_list *param_list, - const gl_state_index tokens[STATE_LENGTH]); - -static int initialize_symbol_from_state(struct gl_program *prog, - struct asm_symbol *param_var, const gl_state_index tokens[STATE_LENGTH]); - -static int initialize_symbol_from_param(struct gl_program *prog, - struct asm_symbol *param_var, const gl_state_index tokens[STATE_LENGTH]); - -static int initialize_symbol_from_const(struct gl_program *prog, - struct asm_symbol *param_var, const struct asm_vector *vec, - GLboolean allowSwizzle); - -static int yyparse(struct asm_parser_state *state); - -static char *make_error_string(const char *fmt, ...); - -static void yyerror(struct YYLTYPE *locp, struct asm_parser_state *state, - const char *s); - -static int validate_inputs(struct YYLTYPE *locp, - struct asm_parser_state *state); - -static void init_dst_reg(struct prog_dst_register *r); - -static void set_dst_reg(struct prog_dst_register *r, - gl_register_file file, GLint index); - -static void init_src_reg(struct asm_src_register *r); - -static void set_src_reg(struct asm_src_register *r, - gl_register_file file, GLint index); - -static void set_src_reg_swz(struct asm_src_register *r, - gl_register_file file, GLint index, GLuint swizzle); - -static void asm_instruction_set_operands(struct asm_instruction *inst, - const struct prog_dst_register *dst, const struct asm_src_register *src0, - const struct asm_src_register *src1, const struct asm_src_register *src2); - -static struct asm_instruction *asm_instruction_ctor(gl_inst_opcode op, - const struct prog_dst_register *dst, const struct asm_src_register *src0, - const struct asm_src_register *src1, const struct asm_src_register *src2); - -static struct asm_instruction *asm_instruction_copy_ctor( - const struct prog_instruction *base, const struct prog_dst_register *dst, - const struct asm_src_register *src0, const struct asm_src_register *src1, - const struct asm_src_register *src2); - -#ifndef FALSE -#define FALSE 0 -#define TRUE (!FALSE) -#endif - -#define YYLLOC_DEFAULT(Current, Rhs, N) \ - do { \ - if (YYID(N)) { \ - (Current).first_line = YYRHSLOC(Rhs, 1).first_line; \ - (Current).first_column = YYRHSLOC(Rhs, 1).first_column; \ - (Current).position = YYRHSLOC(Rhs, 1).position; \ - (Current).last_line = YYRHSLOC(Rhs, N).last_line; \ - (Current).last_column = YYRHSLOC(Rhs, N).last_column; \ - } else { \ - (Current).first_line = YYRHSLOC(Rhs, 0).last_line; \ - (Current).last_line = (Current).first_line; \ - (Current).first_column = YYRHSLOC(Rhs, 0).last_column; \ - (Current).last_column = (Current).first_column; \ - (Current).position = YYRHSLOC(Rhs, 0).position \ - + (Current).first_column; \ - } \ - } while(YYID(0)) - -#define YYLEX_PARAM state->scanner -%} - -%pure-parser -%locations -%parse-param { struct asm_parser_state *state } -%error-verbose -%lex-param { void *scanner } - -%union { - struct asm_instruction *inst; - struct asm_symbol *sym; - struct asm_symbol temp_sym; - struct asm_swizzle_mask swiz_mask; - struct asm_src_register src_reg; - struct prog_dst_register dst_reg; - struct prog_instruction temp_inst; - char *string; - unsigned result; - unsigned attrib; - int integer; - float real; - gl_state_index state[STATE_LENGTH]; - int negate; - struct asm_vector vector; - gl_inst_opcode opcode; - - struct { - unsigned swz; - unsigned rgba_valid:1; - unsigned xyzw_valid:1; - unsigned negate:1; - } ext_swizzle; -} - -%token ARBvp_10 ARBfp_10 - -/* Tokens for assembler pseudo-ops */ -%token ADDRESS -%token ALIAS ATTRIB -%token OPTION OUTPUT -%token PARAM -%token TEMP -%token END - - /* Tokens for instructions */ -%token BIN_OP BINSC_OP SAMPLE_OP SCALAR_OP TRI_OP VECTOR_OP -%token ARL KIL SWZ TXD_OP - -%token INTEGER -%token REAL - -%token AMBIENT ATTENUATION -%token BACK -%token CLIP COLOR -%token DEPTH DIFFUSE DIRECTION -%token EMISSION ENV EYE -%token FOG FOGCOORD FRAGMENT FRONT -%token HALF -%token INVERSE INVTRANS -%token LIGHT LIGHTMODEL LIGHTPROD LOCAL -%token MATERIAL MAT_PROGRAM MATRIX MATRIXINDEX MODELVIEW MVP -%token NORMAL -%token OBJECT -%token PALETTE PARAMS PLANE POINT_TOK POINTSIZE POSITION PRIMARY PROGRAM PROJECTION -%token RANGE RESULT ROW -%token SCENECOLOR SECONDARY SHININESS SIZE_TOK SPECULAR SPOT STATE -%token TEXCOORD TEXENV TEXGEN TEXGEN_Q TEXGEN_R TEXGEN_S TEXGEN_T TEXTURE TRANSPOSE -%token TEXTURE_UNIT TEX_1D TEX_2D TEX_3D TEX_CUBE TEX_RECT -%token TEX_SHADOW1D TEX_SHADOW2D TEX_SHADOWRECT -%token TEX_ARRAY1D TEX_ARRAY2D TEX_ARRAYSHADOW1D TEX_ARRAYSHADOW2D -%token VERTEX VTXATTRIB -%token WEIGHT - -%token IDENTIFIER USED_IDENTIFIER -%type string -%token MASK4 MASK3 MASK2 MASK1 SWIZZLE -%token DOT_DOT -%token DOT - -%type instruction ALU_instruction TexInstruction -%type ARL_instruction VECTORop_instruction -%type SCALARop_instruction BINSCop_instruction BINop_instruction -%type TRIop_instruction TXD_instruction SWZ_instruction SAMPLE_instruction -%type KIL_instruction - -%type dstReg maskedDstReg maskedAddrReg -%type srcReg scalarUse scalarSrcReg swizzleSrcReg -%type scalarSuffix swizzleSuffix extendedSwizzle -%type extSwizComp extSwizSel -%type optionalMask - -%type progParamArray -%type addrRegRelOffset addrRegPosOffset addrRegNegOffset -%type progParamArrayMem progParamArrayAbs progParamArrayRel -%type addrReg -%type addrComponent addrWriteMask - -%type ccMaskRule ccTest ccMaskRule2 ccTest2 optionalCcMask - -%type resultBinding resultColBinding -%type optFaceType optColorType -%type optResultFaceType optResultColorType - -%type optTexImageUnitNum texImageUnitNum -%type optTexCoordUnitNum texCoordUnitNum -%type optLegacyTexUnitNum legacyTexUnitNum -%type texImageUnit texTarget -%type vtxAttribNum - -%type attribBinding vtxAttribItem fragAttribItem - -%type paramSingleInit paramSingleItemDecl -%type optArraySize - -%type stateSingleItem stateMultipleItem -%type stateMaterialItem -%type stateLightItem stateLightModelItem stateLightProdItem -%type stateTexGenItem stateFogItem stateClipPlaneItem statePointItem -%type stateMatrixItem stateMatrixRow stateMatrixRows -%type stateTexEnvItem stateDepthItem - -%type stateLModProperty -%type stateMatrixName optMatrixRows - -%type stateMatProperty -%type stateLightProperty stateSpotProperty -%type stateLightNumber stateLProdProperty -%type stateTexGenType stateTexGenCoord -%type stateTexEnvProperty -%type stateFogProperty -%type stateClipPlaneNum -%type statePointProperty - -%type stateOptMatModifier stateMatModifier stateMatrixRowNum -%type stateOptModMatNum stateModMatNum statePaletteMatNum -%type stateProgramMatNum - -%type ambDiffSpecProperty - -%type programSingleItem progEnvParam progLocalParam -%type programMultipleItem progEnvParams progLocalParams - -%type paramMultipleInit paramMultInitList paramMultipleItem -%type paramSingleItemUse - -%type progEnvParamNum progLocalParamNum -%type progEnvParamNums progLocalParamNums - -%type paramConstDecl paramConstUse -%type paramConstScalarDecl paramConstScalarUse paramConstVector -%type signedFloatConstant -%type optionalSign - -%{ -extern int yylex(YYSTYPE *yylval_param, YYLTYPE *yylloc_param, - void *yyscanner); -%} - -%% - -program: language optionSequence statementSequence END - ; - -language: ARBvp_10 - { - if (state->prog->Target != GL_VERTEX_PROGRAM_ARB) { - yyerror(& @1, state, "invalid fragment program header"); - - } - state->mode = ARB_vertex; - } - | ARBfp_10 - { - if (state->prog->Target != GL_FRAGMENT_PROGRAM_ARB) { - yyerror(& @1, state, "invalid vertex program header"); - } - state->mode = ARB_fragment; - - state->option.TexRect = - (state->ctx->Extensions.NV_texture_rectangle != GL_FALSE); - } - ; - -optionSequence: optionSequence option - | - ; - -option: OPTION string ';' - { - int valid = 0; - - if (state->mode == ARB_vertex) { - valid = _mesa_ARBvp_parse_option(state, $2); - } else if (state->mode == ARB_fragment) { - valid = _mesa_ARBfp_parse_option(state, $2); - } - - - free($2); - - if (!valid) { - const char *const err_str = (state->mode == ARB_vertex) - ? "invalid ARB vertex program option" - : "invalid ARB fragment program option"; - - yyerror(& @2, state, err_str); - YYERROR; - } - } - ; - -statementSequence: statementSequence statement - | - ; - -statement: instruction ';' - { - if ($1 != NULL) { - if (state->inst_tail == NULL) { - state->inst_head = $1; - } else { - state->inst_tail->next = $1; - } - - state->inst_tail = $1; - $1->next = NULL; - - state->prog->NumInstructions++; - } - } - | namingStatement ';' - ; - -instruction: ALU_instruction - { - $$ = $1; - state->prog->NumAluInstructions++; - } - | TexInstruction - { - $$ = $1; - state->prog->NumTexInstructions++; - } - ; - -ALU_instruction: ARL_instruction - | VECTORop_instruction - | SCALARop_instruction - | BINSCop_instruction - | BINop_instruction - | TRIop_instruction - | SWZ_instruction - ; - -TexInstruction: SAMPLE_instruction - | KIL_instruction - | TXD_instruction - ; - -ARL_instruction: ARL maskedAddrReg ',' scalarSrcReg - { - $$ = asm_instruction_ctor(OPCODE_ARL, & $2, & $4, NULL, NULL); - } - ; - -VECTORop_instruction: VECTOR_OP maskedDstReg ',' swizzleSrcReg - { - $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, NULL, NULL); - } - ; - -SCALARop_instruction: SCALAR_OP maskedDstReg ',' scalarSrcReg - { - $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, NULL, NULL); - } - ; - -BINSCop_instruction: BINSC_OP maskedDstReg ',' scalarSrcReg ',' scalarSrcReg - { - $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, & $6, NULL); - } - ; - - -BINop_instruction: BIN_OP maskedDstReg ',' swizzleSrcReg ',' swizzleSrcReg - { - $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, & $6, NULL); - } - ; - -TRIop_instruction: TRI_OP maskedDstReg ',' - swizzleSrcReg ',' swizzleSrcReg ',' swizzleSrcReg - { - $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, & $6, & $8); - } - ; - -SAMPLE_instruction: SAMPLE_OP maskedDstReg ',' swizzleSrcReg ',' texImageUnit ',' texTarget - { - $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, NULL, NULL); - if ($$ != NULL) { - const GLbitfield tex_mask = (1U << $6); - GLbitfield shadow_tex = 0; - GLbitfield target_mask = 0; - - - $$->Base.TexSrcUnit = $6; - - if ($8 < 0) { - shadow_tex = tex_mask; - - $$->Base.TexSrcTarget = -$8; - $$->Base.TexShadow = 1; - } else { - $$->Base.TexSrcTarget = $8; - } - - target_mask = (1U << $$->Base.TexSrcTarget); - - /* If this texture unit was previously accessed and that access - * had a different texture target, generate an error. - * - * If this texture unit was previously accessed and that access - * had a different shadow mode, generate an error. - */ - if ((state->prog->TexturesUsed[$6] != 0) - && ((state->prog->TexturesUsed[$6] != target_mask) - || ((state->prog->ShadowSamplers & tex_mask) - != shadow_tex))) { - yyerror(& @8, state, - "multiple targets used on one texture image unit"); - YYERROR; - } - - - state->prog->TexturesUsed[$6] |= target_mask; - state->prog->ShadowSamplers |= shadow_tex; - } - } - ; - -KIL_instruction: KIL swizzleSrcReg - { - $$ = asm_instruction_ctor(OPCODE_KIL, NULL, & $2, NULL, NULL); - state->fragment.UsesKill = 1; - } - | KIL ccTest - { - $$ = asm_instruction_ctor(OPCODE_KIL_NV, NULL, NULL, NULL, NULL); - $$->Base.DstReg.CondMask = $2.CondMask; - $$->Base.DstReg.CondSwizzle = $2.CondSwizzle; - $$->Base.DstReg.CondSrc = $2.CondSrc; - state->fragment.UsesKill = 1; - } - ; - -TXD_instruction: TXD_OP maskedDstReg ',' swizzleSrcReg ',' swizzleSrcReg ',' swizzleSrcReg ',' texImageUnit ',' texTarget - { - $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, & $6, & $8); - if ($$ != NULL) { - const GLbitfield tex_mask = (1U << $10); - GLbitfield shadow_tex = 0; - GLbitfield target_mask = 0; - - - $$->Base.TexSrcUnit = $10; - - if ($12 < 0) { - shadow_tex = tex_mask; - - $$->Base.TexSrcTarget = -$12; - $$->Base.TexShadow = 1; - } else { - $$->Base.TexSrcTarget = $12; - } - - target_mask = (1U << $$->Base.TexSrcTarget); - - /* If this texture unit was previously accessed and that access - * had a different texture target, generate an error. - * - * If this texture unit was previously accessed and that access - * had a different shadow mode, generate an error. - */ - if ((state->prog->TexturesUsed[$10] != 0) - && ((state->prog->TexturesUsed[$10] != target_mask) - || ((state->prog->ShadowSamplers & tex_mask) - != shadow_tex))) { - yyerror(& @12, state, - "multiple targets used on one texture image unit"); - YYERROR; - } - - - state->prog->TexturesUsed[$10] |= target_mask; - state->prog->ShadowSamplers |= shadow_tex; - } - } - ; - -texImageUnit: TEXTURE_UNIT optTexImageUnitNum - { - $$ = $2; - } - ; - -texTarget: TEX_1D { $$ = TEXTURE_1D_INDEX; } - | TEX_2D { $$ = TEXTURE_2D_INDEX; } - | TEX_3D { $$ = TEXTURE_3D_INDEX; } - | TEX_CUBE { $$ = TEXTURE_CUBE_INDEX; } - | TEX_RECT { $$ = TEXTURE_RECT_INDEX; } - | TEX_SHADOW1D { $$ = -TEXTURE_1D_INDEX; } - | TEX_SHADOW2D { $$ = -TEXTURE_2D_INDEX; } - | TEX_SHADOWRECT { $$ = -TEXTURE_RECT_INDEX; } - | TEX_ARRAY1D { $$ = TEXTURE_1D_ARRAY_INDEX; } - | TEX_ARRAY2D { $$ = TEXTURE_2D_ARRAY_INDEX; } - | TEX_ARRAYSHADOW1D { $$ = -TEXTURE_1D_ARRAY_INDEX; } - | TEX_ARRAYSHADOW2D { $$ = -TEXTURE_2D_ARRAY_INDEX; } - ; - -SWZ_instruction: SWZ maskedDstReg ',' srcReg ',' extendedSwizzle - { - /* FIXME: Is this correct? Should the extenedSwizzle be applied - * FIXME: to the existing swizzle? - */ - $4.Base.Swizzle = $6.swizzle; - $4.Base.Negate = $6.mask; - - $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, NULL, NULL); - } - ; - -scalarSrcReg: optionalSign scalarUse - { - $$ = $2; - - if ($1) { - $$.Base.Negate = ~$$.Base.Negate; - } - } - | optionalSign '|' scalarUse '|' - { - $$ = $3; - - if (!state->option.NV_fragment) { - yyerror(& @2, state, "unexpected character '|'"); - YYERROR; - } - - if ($1) { - $$.Base.Negate = ~$$.Base.Negate; - } - - $$.Base.Abs = 1; - } - ; - -scalarUse: srcReg scalarSuffix - { - $$ = $1; - - $$.Base.Swizzle = _mesa_combine_swizzles($$.Base.Swizzle, - $2.swizzle); - } - | paramConstScalarUse - { - struct asm_symbol temp_sym; - - if (!state->option.NV_fragment) { - yyerror(& @1, state, "expected scalar suffix"); - YYERROR; - } - - memset(& temp_sym, 0, sizeof(temp_sym)); - temp_sym.param_binding_begin = ~0; - initialize_symbol_from_const(state->prog, & temp_sym, & $1, GL_TRUE); - - set_src_reg_swz(& $$, PROGRAM_CONSTANT, - temp_sym.param_binding_begin, - temp_sym.param_binding_swizzle); - } - ; - -swizzleSrcReg: optionalSign srcReg swizzleSuffix - { - $$ = $2; - - if ($1) { - $$.Base.Negate = ~$$.Base.Negate; - } - - $$.Base.Swizzle = _mesa_combine_swizzles($$.Base.Swizzle, - $3.swizzle); - } - | optionalSign '|' srcReg swizzleSuffix '|' - { - $$ = $3; - - if (!state->option.NV_fragment) { - yyerror(& @2, state, "unexpected character '|'"); - YYERROR; - } - - if ($1) { - $$.Base.Negate = ~$$.Base.Negate; - } - - $$.Base.Abs = 1; - $$.Base.Swizzle = _mesa_combine_swizzles($$.Base.Swizzle, - $4.swizzle); - } - - ; - -maskedDstReg: dstReg optionalMask optionalCcMask - { - $$ = $1; - $$.WriteMask = $2.mask; - $$.CondMask = $3.CondMask; - $$.CondSwizzle = $3.CondSwizzle; - $$.CondSrc = $3.CondSrc; - - if ($$.File == PROGRAM_OUTPUT) { - /* Technically speaking, this should check that it is in - * vertex program mode. However, PositionInvariant can never be - * set in fragment program mode, so it is somewhat irrelevant. - */ - if (state->option.PositionInvariant - && ($$.Index == VERT_RESULT_HPOS)) { - yyerror(& @1, state, "position-invariant programs cannot " - "write position"); - YYERROR; - } - - state->prog->OutputsWritten |= BITFIELD64_BIT($$.Index); - } - } - ; - -maskedAddrReg: addrReg addrWriteMask - { - set_dst_reg(& $$, PROGRAM_ADDRESS, 0); - $$.WriteMask = $2.mask; - } - ; - -extendedSwizzle: extSwizComp ',' extSwizComp ',' extSwizComp ',' extSwizComp - { - const unsigned xyzw_valid = - ($1.xyzw_valid << 0) - | ($3.xyzw_valid << 1) - | ($5.xyzw_valid << 2) - | ($7.xyzw_valid << 3); - const unsigned rgba_valid = - ($1.rgba_valid << 0) - | ($3.rgba_valid << 1) - | ($5.rgba_valid << 2) - | ($7.rgba_valid << 3); - - /* All of the swizzle components have to be valid in either RGBA - * or XYZW. Note that 0 and 1 are valid in both, so both masks - * can have some bits set. - * - * We somewhat deviate from the spec here. It would be really hard - * to figure out which component is the error, and there probably - * isn't a lot of benefit. - */ - if ((rgba_valid != 0x0f) && (xyzw_valid != 0x0f)) { - yyerror(& @1, state, "cannot combine RGBA and XYZW swizzle " - "components"); - YYERROR; - } - - $$.swizzle = MAKE_SWIZZLE4($1.swz, $3.swz, $5.swz, $7.swz); - $$.mask = ($1.negate) | ($3.negate << 1) | ($5.negate << 2) - | ($7.negate << 3); - } - ; - -extSwizComp: optionalSign extSwizSel - { - $$ = $2; - $$.negate = ($1) ? 1 : 0; - } - ; - -extSwizSel: INTEGER - { - if (($1 != 0) && ($1 != 1)) { - yyerror(& @1, state, "invalid extended swizzle selector"); - YYERROR; - } - - $$.swz = ($1 == 0) ? SWIZZLE_ZERO : SWIZZLE_ONE; - - /* 0 and 1 are valid for both RGBA swizzle names and XYZW - * swizzle names. - */ - $$.xyzw_valid = 1; - $$.rgba_valid = 1; - } - | string - { - char s; - - if (strlen($1) > 1) { - yyerror(& @1, state, "invalid extended swizzle selector"); - YYERROR; - } - - s = $1[0]; - free($1); - - switch (s) { - case 'x': - $$.swz = SWIZZLE_X; - $$.xyzw_valid = 1; - break; - case 'y': - $$.swz = SWIZZLE_Y; - $$.xyzw_valid = 1; - break; - case 'z': - $$.swz = SWIZZLE_Z; - $$.xyzw_valid = 1; - break; - case 'w': - $$.swz = SWIZZLE_W; - $$.xyzw_valid = 1; - break; - - case 'r': - $$.swz = SWIZZLE_X; - $$.rgba_valid = 1; - break; - case 'g': - $$.swz = SWIZZLE_Y; - $$.rgba_valid = 1; - break; - case 'b': - $$.swz = SWIZZLE_Z; - $$.rgba_valid = 1; - break; - case 'a': - $$.swz = SWIZZLE_W; - $$.rgba_valid = 1; - break; - - default: - yyerror(& @1, state, "invalid extended swizzle selector"); - YYERROR; - break; - } - } - ; - -srcReg: USED_IDENTIFIER /* temporaryReg | progParamSingle */ - { - struct asm_symbol *const s = (struct asm_symbol *) - _mesa_symbol_table_find_symbol(state->st, 0, $1); - - free($1); - - if (s == NULL) { - yyerror(& @1, state, "invalid operand variable"); - YYERROR; - } else if ((s->type != at_param) && (s->type != at_temp) - && (s->type != at_attrib)) { - yyerror(& @1, state, "invalid operand variable"); - YYERROR; - } else if ((s->type == at_param) && s->param_is_array) { - yyerror(& @1, state, "non-array access to array PARAM"); - YYERROR; - } - - init_src_reg(& $$); - switch (s->type) { - case at_temp: - set_src_reg(& $$, PROGRAM_TEMPORARY, s->temp_binding); - break; - case at_param: - set_src_reg_swz(& $$, s->param_binding_type, - s->param_binding_begin, - s->param_binding_swizzle); - break; - case at_attrib: - set_src_reg(& $$, PROGRAM_INPUT, s->attrib_binding); - state->prog->InputsRead |= (1U << $$.Base.Index); - - if (!validate_inputs(& @1, state)) { - YYERROR; - } - break; - - default: - YYERROR; - break; - } - } - | attribBinding - { - set_src_reg(& $$, PROGRAM_INPUT, $1); - state->prog->InputsRead |= (1U << $$.Base.Index); - - if (!validate_inputs(& @1, state)) { - YYERROR; - } - } - | progParamArray '[' progParamArrayMem ']' - { - if (! $3.Base.RelAddr - && ((unsigned) $3.Base.Index >= $1->param_binding_length)) { - yyerror(& @3, state, "out of bounds array access"); - YYERROR; - } - - init_src_reg(& $$); - $$.Base.File = $1->param_binding_type; - - if ($3.Base.RelAddr) { - state->prog->IndirectRegisterFiles |= (1 << $$.Base.File); - $1->param_accessed_indirectly = 1; - - $$.Base.RelAddr = 1; - $$.Base.Index = $3.Base.Index; - $$.Symbol = $1; - } else { - $$.Base.Index = $1->param_binding_begin + $3.Base.Index; - } - } - | paramSingleItemUse - { - gl_register_file file = ($1.name != NULL) - ? $1.param_binding_type - : PROGRAM_CONSTANT; - set_src_reg_swz(& $$, file, $1.param_binding_begin, - $1.param_binding_swizzle); - } - ; - -dstReg: resultBinding - { - set_dst_reg(& $$, PROGRAM_OUTPUT, $1); - } - | USED_IDENTIFIER /* temporaryReg | vertexResultReg */ - { - struct asm_symbol *const s = (struct asm_symbol *) - _mesa_symbol_table_find_symbol(state->st, 0, $1); - - free($1); - - if (s == NULL) { - yyerror(& @1, state, "invalid operand variable"); - YYERROR; - } else if ((s->type != at_output) && (s->type != at_temp)) { - yyerror(& @1, state, "invalid operand variable"); - YYERROR; - } - - switch (s->type) { - case at_temp: - set_dst_reg(& $$, PROGRAM_TEMPORARY, s->temp_binding); - break; - case at_output: - set_dst_reg(& $$, PROGRAM_OUTPUT, s->output_binding); - break; - default: - set_dst_reg(& $$, s->param_binding_type, s->param_binding_begin); - break; - } - } - ; - -progParamArray: USED_IDENTIFIER - { - struct asm_symbol *const s = (struct asm_symbol *) - _mesa_symbol_table_find_symbol(state->st, 0, $1); - - free($1); - - if (s == NULL) { - yyerror(& @1, state, "invalid operand variable"); - YYERROR; - } else if ((s->type != at_param) || !s->param_is_array) { - yyerror(& @1, state, "array access to non-PARAM variable"); - YYERROR; - } else { - $$ = s; - } - } - ; - -progParamArrayMem: progParamArrayAbs | progParamArrayRel; - -progParamArrayAbs: INTEGER - { - init_src_reg(& $$); - $$.Base.Index = $1; - } - ; - -progParamArrayRel: addrReg addrComponent addrRegRelOffset - { - /* FINISHME: Add support for multiple address registers. - */ - /* FINISHME: Add support for 4-component address registers. - */ - init_src_reg(& $$); - $$.Base.RelAddr = 1; - $$.Base.Index = $3; - } - ; - -addrRegRelOffset: { $$ = 0; } - | '+' addrRegPosOffset { $$ = $2; } - | '-' addrRegNegOffset { $$ = -$2; } - ; - -addrRegPosOffset: INTEGER - { - if (($1 < 0) || ($1 > (state->limits->MaxAddressOffset - 1))) { - char s[100]; - _mesa_snprintf(s, sizeof(s), - "relative address offset too large (%d)", $1); - yyerror(& @1, state, s); - YYERROR; - } else { - $$ = $1; - } - } - ; - -addrRegNegOffset: INTEGER - { - if (($1 < 0) || ($1 > state->limits->MaxAddressOffset)) { - char s[100]; - _mesa_snprintf(s, sizeof(s), - "relative address offset too large (%d)", $1); - yyerror(& @1, state, s); - YYERROR; - } else { - $$ = $1; - } - } - ; - -addrReg: USED_IDENTIFIER - { - struct asm_symbol *const s = (struct asm_symbol *) - _mesa_symbol_table_find_symbol(state->st, 0, $1); - - free($1); - - if (s == NULL) { - yyerror(& @1, state, "invalid array member"); - YYERROR; - } else if (s->type != at_address) { - yyerror(& @1, state, - "invalid variable for indexed array access"); - YYERROR; - } else { - $$ = s; - } - } - ; - -addrComponent: MASK1 - { - if ($1.mask != WRITEMASK_X) { - yyerror(& @1, state, "invalid address component selector"); - YYERROR; - } else { - $$ = $1; - } - } - ; - -addrWriteMask: MASK1 - { - if ($1.mask != WRITEMASK_X) { - yyerror(& @1, state, - "address register write mask must be \".x\""); - YYERROR; - } else { - $$ = $1; - } - } - ; - -scalarSuffix: MASK1; - -swizzleSuffix: MASK1 - | MASK4 - | SWIZZLE - | { $$.swizzle = SWIZZLE_NOOP; $$.mask = WRITEMASK_XYZW; } - ; - -optionalMask: MASK4 | MASK3 | MASK2 | MASK1 - | { $$.swizzle = SWIZZLE_NOOP; $$.mask = WRITEMASK_XYZW; } - ; - -optionalCcMask: '(' ccTest ')' - { - $$ = $2; - } - | '(' ccTest2 ')' - { - $$ = $2; - } - | - { - $$.CondMask = COND_TR; - $$.CondSwizzle = SWIZZLE_NOOP; - $$.CondSrc = 0; - } - ; - -ccTest: ccMaskRule swizzleSuffix - { - $$ = $1; - $$.CondSwizzle = $2.swizzle; - } - ; - -ccTest2: ccMaskRule2 swizzleSuffix - { - $$ = $1; - $$.CondSwizzle = $2.swizzle; - } - ; - -ccMaskRule: IDENTIFIER - { - const int cond = _mesa_parse_cc($1); - if ((cond == 0) || ($1[2] != '\0')) { - char *const err_str = - make_error_string("invalid condition code \"%s\"", $1); - - yyerror(& @1, state, (err_str != NULL) - ? err_str : "invalid condition code"); - - if (err_str != NULL) { - free(err_str); - } - - YYERROR; - } - - $$.CondMask = cond; - $$.CondSwizzle = SWIZZLE_NOOP; - $$.CondSrc = 0; - } - ; - -ccMaskRule2: USED_IDENTIFIER - { - const int cond = _mesa_parse_cc($1); - if ((cond == 0) || ($1[2] != '\0')) { - char *const err_str = - make_error_string("invalid condition code \"%s\"", $1); - - yyerror(& @1, state, (err_str != NULL) - ? err_str : "invalid condition code"); - - if (err_str != NULL) { - free(err_str); - } - - YYERROR; - } - - $$.CondMask = cond; - $$.CondSwizzle = SWIZZLE_NOOP; - $$.CondSrc = 0; - } - ; - -namingStatement: ATTRIB_statement - | PARAM_statement - | TEMP_statement - | ADDRESS_statement - | OUTPUT_statement - | ALIAS_statement - ; - -ATTRIB_statement: ATTRIB IDENTIFIER '=' attribBinding - { - struct asm_symbol *const s = - declare_variable(state, $2, at_attrib, & @2); - - if (s == NULL) { - free($2); - YYERROR; - } else { - s->attrib_binding = $4; - state->InputsBound |= (1U << s->attrib_binding); - - if (!validate_inputs(& @4, state)) { - YYERROR; - } - } - } - ; - -attribBinding: VERTEX vtxAttribItem - { - $$ = $2; - } - | FRAGMENT fragAttribItem - { - $$ = $2; - } - ; - -vtxAttribItem: POSITION - { - $$ = VERT_ATTRIB_POS; - } - | WEIGHT vtxOptWeightNum - { - $$ = VERT_ATTRIB_WEIGHT; - } - | NORMAL - { - $$ = VERT_ATTRIB_NORMAL; - } - | COLOR optColorType - { - if (!state->ctx->Extensions.EXT_secondary_color) { - yyerror(& @2, state, "GL_EXT_secondary_color not supported"); - YYERROR; - } - - $$ = VERT_ATTRIB_COLOR0 + $2; - } - | FOGCOORD - { - if (!state->ctx->Extensions.EXT_fog_coord) { - yyerror(& @1, state, "GL_EXT_fog_coord not supported"); - YYERROR; - } - - $$ = VERT_ATTRIB_FOG; - } - | TEXCOORD optTexCoordUnitNum - { - $$ = VERT_ATTRIB_TEX0 + $2; - } - | MATRIXINDEX '[' vtxWeightNum ']' - { - yyerror(& @1, state, "GL_ARB_matrix_palette not supported"); - YYERROR; - } - | VTXATTRIB '[' vtxAttribNum ']' - { - $$ = VERT_ATTRIB_GENERIC0 + $3; - } - ; - -vtxAttribNum: INTEGER - { - if ((unsigned) $1 >= state->limits->MaxAttribs) { - yyerror(& @1, state, "invalid vertex attribute reference"); - YYERROR; - } - - $$ = $1; - } - ; - -vtxOptWeightNum: | '[' vtxWeightNum ']'; -vtxWeightNum: INTEGER; - -fragAttribItem: POSITION - { - $$ = FRAG_ATTRIB_WPOS; - } - | COLOR optColorType - { - $$ = FRAG_ATTRIB_COL0 + $2; - } - | FOGCOORD - { - $$ = FRAG_ATTRIB_FOGC; - } - | TEXCOORD optTexCoordUnitNum - { - $$ = FRAG_ATTRIB_TEX0 + $2; - } - ; - -PARAM_statement: PARAM_singleStmt | PARAM_multipleStmt; - -PARAM_singleStmt: PARAM IDENTIFIER paramSingleInit - { - struct asm_symbol *const s = - declare_variable(state, $2, at_param, & @2); - - if (s == NULL) { - free($2); - YYERROR; - } else { - s->param_binding_type = $3.param_binding_type; - s->param_binding_begin = $3.param_binding_begin; - s->param_binding_length = $3.param_binding_length; - s->param_binding_swizzle = $3.param_binding_swizzle; - s->param_is_array = 0; - } - } - ; - -PARAM_multipleStmt: PARAM IDENTIFIER '[' optArraySize ']' paramMultipleInit - { - if (($4 != 0) && ((unsigned) $4 != $6.param_binding_length)) { - free($2); - yyerror(& @4, state, - "parameter array size and number of bindings must match"); - YYERROR; - } else { - struct asm_symbol *const s = - declare_variable(state, $2, $6.type, & @2); - - if (s == NULL) { - free($2); - YYERROR; - } else { - s->param_binding_type = $6.param_binding_type; - s->param_binding_begin = $6.param_binding_begin; - s->param_binding_length = $6.param_binding_length; - s->param_binding_swizzle = SWIZZLE_XYZW; - s->param_is_array = 1; - } - } - } - ; - -optArraySize: - { - $$ = 0; - } - | INTEGER - { - if (($1 < 1) || ((unsigned) $1 > state->limits->MaxParameters)) { - char msg[100]; - _mesa_snprintf(msg, sizeof(msg), - "invalid parameter array size (size=%d max=%u)", - $1, state->limits->MaxParameters); - yyerror(& @1, state, msg); - YYERROR; - } else { - $$ = $1; - } - } - ; - -paramSingleInit: '=' paramSingleItemDecl - { - $$ = $2; - } - ; - -paramMultipleInit: '=' '{' paramMultInitList '}' - { - $$ = $3; - } - ; - -paramMultInitList: paramMultipleItem - | paramMultInitList ',' paramMultipleItem - { - $1.param_binding_length += $3.param_binding_length; - $$ = $1; - } - ; - -paramSingleItemDecl: stateSingleItem - { - memset(& $$, 0, sizeof($$)); - $$.param_binding_begin = ~0; - initialize_symbol_from_state(state->prog, & $$, $1); - } - | programSingleItem - { - memset(& $$, 0, sizeof($$)); - $$.param_binding_begin = ~0; - initialize_symbol_from_param(state->prog, & $$, $1); - } - | paramConstDecl - { - memset(& $$, 0, sizeof($$)); - $$.param_binding_begin = ~0; - initialize_symbol_from_const(state->prog, & $$, & $1, GL_TRUE); - } - ; - -paramSingleItemUse: stateSingleItem - { - memset(& $$, 0, sizeof($$)); - $$.param_binding_begin = ~0; - initialize_symbol_from_state(state->prog, & $$, $1); - } - | programSingleItem - { - memset(& $$, 0, sizeof($$)); - $$.param_binding_begin = ~0; - initialize_symbol_from_param(state->prog, & $$, $1); - } - | paramConstUse - { - memset(& $$, 0, sizeof($$)); - $$.param_binding_begin = ~0; - initialize_symbol_from_const(state->prog, & $$, & $1, GL_TRUE); - } - ; - -paramMultipleItem: stateMultipleItem - { - memset(& $$, 0, sizeof($$)); - $$.param_binding_begin = ~0; - initialize_symbol_from_state(state->prog, & $$, $1); - } - | programMultipleItem - { - memset(& $$, 0, sizeof($$)); - $$.param_binding_begin = ~0; - initialize_symbol_from_param(state->prog, & $$, $1); - } - | paramConstDecl - { - memset(& $$, 0, sizeof($$)); - $$.param_binding_begin = ~0; - initialize_symbol_from_const(state->prog, & $$, & $1, GL_FALSE); - } - ; - -stateMultipleItem: stateSingleItem { memcpy($$, $1, sizeof($$)); } - | STATE stateMatrixRows { memcpy($$, $2, sizeof($$)); } - ; - -stateSingleItem: STATE stateMaterialItem { memcpy($$, $2, sizeof($$)); } - | STATE stateLightItem { memcpy($$, $2, sizeof($$)); } - | STATE stateLightModelItem { memcpy($$, $2, sizeof($$)); } - | STATE stateLightProdItem { memcpy($$, $2, sizeof($$)); } - | STATE stateTexGenItem { memcpy($$, $2, sizeof($$)); } - | STATE stateTexEnvItem { memcpy($$, $2, sizeof($$)); } - | STATE stateFogItem { memcpy($$, $2, sizeof($$)); } - | STATE stateClipPlaneItem { memcpy($$, $2, sizeof($$)); } - | STATE statePointItem { memcpy($$, $2, sizeof($$)); } - | STATE stateMatrixRow { memcpy($$, $2, sizeof($$)); } - | STATE stateDepthItem { memcpy($$, $2, sizeof($$)); } - ; - -stateMaterialItem: MATERIAL optFaceType stateMatProperty - { - memset($$, 0, sizeof($$)); - $$[0] = STATE_MATERIAL; - $$[1] = $2; - $$[2] = $3; - } - ; - -stateMatProperty: ambDiffSpecProperty - { - $$ = $1; - } - | EMISSION - { - $$ = STATE_EMISSION; - } - | SHININESS - { - $$ = STATE_SHININESS; - } - ; - -stateLightItem: LIGHT '[' stateLightNumber ']' stateLightProperty - { - memset($$, 0, sizeof($$)); - $$[0] = STATE_LIGHT; - $$[1] = $3; - $$[2] = $5; - } - ; - -stateLightProperty: ambDiffSpecProperty - { - $$ = $1; - } - | POSITION - { - $$ = STATE_POSITION; - } - | ATTENUATION - { - if (!state->ctx->Extensions.EXT_point_parameters) { - yyerror(& @1, state, "GL_ARB_point_parameters not supported"); - YYERROR; - } - - $$ = STATE_ATTENUATION; - } - | SPOT stateSpotProperty - { - $$ = $2; - } - | HALF - { - $$ = STATE_HALF_VECTOR; - } - ; - -stateSpotProperty: DIRECTION - { - $$ = STATE_SPOT_DIRECTION; - } - ; - -stateLightModelItem: LIGHTMODEL stateLModProperty - { - $$[0] = $2[0]; - $$[1] = $2[1]; - } - ; - -stateLModProperty: AMBIENT - { - memset($$, 0, sizeof($$)); - $$[0] = STATE_LIGHTMODEL_AMBIENT; - } - | optFaceType SCENECOLOR - { - memset($$, 0, sizeof($$)); - $$[0] = STATE_LIGHTMODEL_SCENECOLOR; - $$[1] = $1; - } - ; - -stateLightProdItem: LIGHTPROD '[' stateLightNumber ']' optFaceType stateLProdProperty - { - memset($$, 0, sizeof($$)); - $$[0] = STATE_LIGHTPROD; - $$[1] = $3; - $$[2] = $5; - $$[3] = $6; - } - ; - -stateLProdProperty: ambDiffSpecProperty; - -stateTexEnvItem: TEXENV optLegacyTexUnitNum stateTexEnvProperty - { - memset($$, 0, sizeof($$)); - $$[0] = $3; - $$[1] = $2; - } - ; - -stateTexEnvProperty: COLOR - { - $$ = STATE_TEXENV_COLOR; - } - ; - -ambDiffSpecProperty: AMBIENT - { - $$ = STATE_AMBIENT; - } - | DIFFUSE - { - $$ = STATE_DIFFUSE; - } - | SPECULAR - { - $$ = STATE_SPECULAR; - } - ; - -stateLightNumber: INTEGER - { - if ((unsigned) $1 >= state->MaxLights) { - yyerror(& @1, state, "invalid light selector"); - YYERROR; - } - - $$ = $1; - } - ; - -stateTexGenItem: TEXGEN optTexCoordUnitNum stateTexGenType stateTexGenCoord - { - memset($$, 0, sizeof($$)); - $$[0] = STATE_TEXGEN; - $$[1] = $2; - $$[2] = $3 + $4; - } - ; - -stateTexGenType: EYE - { - $$ = STATE_TEXGEN_EYE_S; - } - | OBJECT - { - $$ = STATE_TEXGEN_OBJECT_S; - } - ; -stateTexGenCoord: TEXGEN_S - { - $$ = STATE_TEXGEN_EYE_S - STATE_TEXGEN_EYE_S; - } - | TEXGEN_T - { - $$ = STATE_TEXGEN_EYE_T - STATE_TEXGEN_EYE_S; - } - | TEXGEN_R - { - $$ = STATE_TEXGEN_EYE_R - STATE_TEXGEN_EYE_S; - } - | TEXGEN_Q - { - $$ = STATE_TEXGEN_EYE_Q - STATE_TEXGEN_EYE_S; - } - ; - -stateFogItem: FOG stateFogProperty - { - memset($$, 0, sizeof($$)); - $$[0] = $2; - } - ; - -stateFogProperty: COLOR - { - $$ = STATE_FOG_COLOR; - } - | PARAMS - { - $$ = STATE_FOG_PARAMS; - } - ; - -stateClipPlaneItem: CLIP '[' stateClipPlaneNum ']' PLANE - { - memset($$, 0, sizeof($$)); - $$[0] = STATE_CLIPPLANE; - $$[1] = $3; - } - ; - -stateClipPlaneNum: INTEGER - { - if ((unsigned) $1 >= state->MaxClipPlanes) { - yyerror(& @1, state, "invalid clip plane selector"); - YYERROR; - } - - $$ = $1; - } - ; - -statePointItem: POINT_TOK statePointProperty - { - memset($$, 0, sizeof($$)); - $$[0] = $2; - } - ; - -statePointProperty: SIZE_TOK - { - $$ = STATE_POINT_SIZE; - } - | ATTENUATION - { - $$ = STATE_POINT_ATTENUATION; - } - ; - -stateMatrixRow: stateMatrixItem ROW '[' stateMatrixRowNum ']' - { - $$[0] = $1[0]; - $$[1] = $1[1]; - $$[2] = $4; - $$[3] = $4; - $$[4] = $1[2]; - } - ; - -stateMatrixRows: stateMatrixItem optMatrixRows - { - $$[0] = $1[0]; - $$[1] = $1[1]; - $$[2] = $2[2]; - $$[3] = $2[3]; - $$[4] = $1[2]; - } - ; - -optMatrixRows: - { - $$[2] = 0; - $$[3] = 3; - } - | ROW '[' stateMatrixRowNum DOT_DOT stateMatrixRowNum ']' - { - /* It seems logical that the matrix row range specifier would have - * to specify a range or more than one row (i.e., $5 > $3). - * However, the ARB_vertex_program spec says "a program will fail - * to load if is greater than ." This means that $3 == $5 - * is valid. - */ - if ($3 > $5) { - yyerror(& @3, state, "invalid matrix row range"); - YYERROR; - } - - $$[2] = $3; - $$[3] = $5; - } - ; - -stateMatrixItem: MATRIX stateMatrixName stateOptMatModifier - { - $$[0] = $2[0]; - $$[1] = $2[1]; - $$[2] = $3; - } - ; - -stateOptMatModifier: - { - $$ = 0; - } - | stateMatModifier - { - $$ = $1; - } - ; - -stateMatModifier: INVERSE - { - $$ = STATE_MATRIX_INVERSE; - } - | TRANSPOSE - { - $$ = STATE_MATRIX_TRANSPOSE; - } - | INVTRANS - { - $$ = STATE_MATRIX_INVTRANS; - } - ; - -stateMatrixRowNum: INTEGER - { - if ($1 > 3) { - yyerror(& @1, state, "invalid matrix row reference"); - YYERROR; - } - - $$ = $1; - } - ; - -stateMatrixName: MODELVIEW stateOptModMatNum - { - $$[0] = STATE_MODELVIEW_MATRIX; - $$[1] = $2; - } - | PROJECTION - { - $$[0] = STATE_PROJECTION_MATRIX; - $$[1] = 0; - } - | MVP - { - $$[0] = STATE_MVP_MATRIX; - $$[1] = 0; - } - | TEXTURE optTexCoordUnitNum - { - $$[0] = STATE_TEXTURE_MATRIX; - $$[1] = $2; - } - | PALETTE '[' statePaletteMatNum ']' - { - yyerror(& @1, state, "GL_ARB_matrix_palette not supported"); - YYERROR; - } - | MAT_PROGRAM '[' stateProgramMatNum ']' - { - $$[0] = STATE_PROGRAM_MATRIX; - $$[1] = $3; - } - ; - -stateOptModMatNum: - { - $$ = 0; - } - | '[' stateModMatNum ']' - { - $$ = $2; - } - ; -stateModMatNum: INTEGER - { - /* Since GL_ARB_vertex_blend isn't supported, only modelview matrix - * zero is valid. - */ - if ($1 != 0) { - yyerror(& @1, state, "invalid modelview matrix index"); - YYERROR; - } - - $$ = $1; - } - ; -statePaletteMatNum: INTEGER - { - /* Since GL_ARB_matrix_palette isn't supported, just let any value - * through here. The error will be generated later. - */ - $$ = $1; - } - ; -stateProgramMatNum: INTEGER - { - if ((unsigned) $1 >= state->MaxProgramMatrices) { - yyerror(& @1, state, "invalid program matrix selector"); - YYERROR; - } - - $$ = $1; - } - ; - -stateDepthItem: DEPTH RANGE - { - memset($$, 0, sizeof($$)); - $$[0] = STATE_DEPTH_RANGE; - } - ; - - -programSingleItem: progEnvParam | progLocalParam; - -programMultipleItem: progEnvParams | progLocalParams; - -progEnvParams: PROGRAM ENV '[' progEnvParamNums ']' - { - memset($$, 0, sizeof($$)); - $$[0] = state->state_param_enum; - $$[1] = STATE_ENV; - $$[2] = $4[0]; - $$[3] = $4[1]; - } - ; - -progEnvParamNums: progEnvParamNum - { - $$[0] = $1; - $$[1] = $1; - } - | progEnvParamNum DOT_DOT progEnvParamNum - { - $$[0] = $1; - $$[1] = $3; - } - ; - -progEnvParam: PROGRAM ENV '[' progEnvParamNum ']' - { - memset($$, 0, sizeof($$)); - $$[0] = state->state_param_enum; - $$[1] = STATE_ENV; - $$[2] = $4; - $$[3] = $4; - } - ; - -progLocalParams: PROGRAM LOCAL '[' progLocalParamNums ']' - { - memset($$, 0, sizeof($$)); - $$[0] = state->state_param_enum; - $$[1] = STATE_LOCAL; - $$[2] = $4[0]; - $$[3] = $4[1]; - } - -progLocalParamNums: progLocalParamNum - { - $$[0] = $1; - $$[1] = $1; - } - | progLocalParamNum DOT_DOT progLocalParamNum - { - $$[0] = $1; - $$[1] = $3; - } - ; - -progLocalParam: PROGRAM LOCAL '[' progLocalParamNum ']' - { - memset($$, 0, sizeof($$)); - $$[0] = state->state_param_enum; - $$[1] = STATE_LOCAL; - $$[2] = $4; - $$[3] = $4; - } - ; - -progEnvParamNum: INTEGER - { - if ((unsigned) $1 >= state->limits->MaxEnvParams) { - yyerror(& @1, state, "invalid environment parameter reference"); - YYERROR; - } - $$ = $1; - } - ; - -progLocalParamNum: INTEGER - { - if ((unsigned) $1 >= state->limits->MaxLocalParams) { - yyerror(& @1, state, "invalid local parameter reference"); - YYERROR; - } - $$ = $1; - } - ; - - - -paramConstDecl: paramConstScalarDecl | paramConstVector; -paramConstUse: paramConstScalarUse | paramConstVector; - -paramConstScalarDecl: signedFloatConstant - { - $$.count = 4; - $$.data[0].f = $1; - $$.data[1].f = $1; - $$.data[2].f = $1; - $$.data[3].f = $1; - } - ; - -paramConstScalarUse: REAL - { - $$.count = 1; - $$.data[0].f = $1; - $$.data[1].f = $1; - $$.data[2].f = $1; - $$.data[3].f = $1; - } - | INTEGER - { - $$.count = 1; - $$.data[0].f = (float) $1; - $$.data[1].f = (float) $1; - $$.data[2].f = (float) $1; - $$.data[3].f = (float) $1; - } - ; - -paramConstVector: '{' signedFloatConstant '}' - { - $$.count = 4; - $$.data[0].f = $2; - $$.data[1].f = 0.0f; - $$.data[2].f = 0.0f; - $$.data[3].f = 1.0f; - } - | '{' signedFloatConstant ',' signedFloatConstant '}' - { - $$.count = 4; - $$.data[0].f = $2; - $$.data[1].f = $4; - $$.data[2].f = 0.0f; - $$.data[3].f = 1.0f; - } - | '{' signedFloatConstant ',' signedFloatConstant ',' - signedFloatConstant '}' - { - $$.count = 4; - $$.data[0].f = $2; - $$.data[1].f = $4; - $$.data[2].f = $6; - $$.data[3].f = 1.0f; - } - | '{' signedFloatConstant ',' signedFloatConstant ',' - signedFloatConstant ',' signedFloatConstant '}' - { - $$.count = 4; - $$.data[0].f = $2; - $$.data[1].f = $4; - $$.data[2].f = $6; - $$.data[3].f = $8; - } - ; - -signedFloatConstant: optionalSign REAL - { - $$ = ($1) ? -$2 : $2; - } - | optionalSign INTEGER - { - $$ = (float)(($1) ? -$2 : $2); - } - ; - -optionalSign: '+' { $$ = FALSE; } - | '-' { $$ = TRUE; } - | { $$ = FALSE; } - ; - -TEMP_statement: optVarSize TEMP { $$ = $2; } varNameList - ; - -optVarSize: string - { - /* NV_fragment_program_option defines the size qualifiers in a - * fairly broken way. "SHORT" or "LONG" can optionally be used - * before TEMP or OUTPUT. However, neither is a reserved word! - * This means that we have to parse it as an identifier, then check - * to make sure it's one of the valid values. *sigh* - * - * In addition, the grammar in the extension spec does *not* allow - * the size specifier to be optional, but all known implementations - * do. - */ - if (!state->option.NV_fragment) { - yyerror(& @1, state, "unexpected IDENTIFIER"); - YYERROR; - } - - if (strcmp("SHORT", $1) == 0) { - } else if (strcmp("LONG", $1) == 0) { - } else { - char *const err_str = - make_error_string("invalid storage size specifier \"%s\"", - $1); - - yyerror(& @1, state, (err_str != NULL) - ? err_str : "invalid storage size specifier"); - - if (err_str != NULL) { - free(err_str); - } - - YYERROR; - } - } - | - { - } - ; - -ADDRESS_statement: ADDRESS { $$ = $1; } varNameList - ; - -varNameList: varNameList ',' IDENTIFIER - { - if (!declare_variable(state, $3, $0, & @3)) { - free($3); - YYERROR; - } - } - | IDENTIFIER - { - if (!declare_variable(state, $1, $0, & @1)) { - free($1); - YYERROR; - } - } - ; - -OUTPUT_statement: optVarSize OUTPUT IDENTIFIER '=' resultBinding - { - struct asm_symbol *const s = - declare_variable(state, $3, at_output, & @3); - - if (s == NULL) { - free($3); - YYERROR; - } else { - s->output_binding = $5; - } - } - ; - -resultBinding: RESULT POSITION - { - if (state->mode == ARB_vertex) { - $$ = VERT_RESULT_HPOS; - } else { - yyerror(& @2, state, "invalid program result name"); - YYERROR; - } - } - | RESULT FOGCOORD - { - if (state->mode == ARB_vertex) { - $$ = VERT_RESULT_FOGC; - } else { - yyerror(& @2, state, "invalid program result name"); - YYERROR; - } - } - | RESULT resultColBinding - { - $$ = $2; - } - | RESULT POINTSIZE - { - if (state->mode == ARB_vertex) { - $$ = VERT_RESULT_PSIZ; - } else { - yyerror(& @2, state, "invalid program result name"); - YYERROR; - } - } - | RESULT TEXCOORD optTexCoordUnitNum - { - if (state->mode == ARB_vertex) { - $$ = VERT_RESULT_TEX0 + $3; - } else { - yyerror(& @2, state, "invalid program result name"); - YYERROR; - } - } - | RESULT DEPTH - { - if (state->mode == ARB_fragment) { - $$ = FRAG_RESULT_DEPTH; - } else { - yyerror(& @2, state, "invalid program result name"); - YYERROR; - } - } - ; - -resultColBinding: COLOR optResultFaceType optResultColorType - { - $$ = $2 + $3; - } - ; - -optResultFaceType: - { - if (state->mode == ARB_vertex) { - $$ = VERT_RESULT_COL0; - } else { - if (state->option.DrawBuffers) - $$ = FRAG_RESULT_DATA0; - else - $$ = FRAG_RESULT_COLOR; - } - } - | '[' INTEGER ']' - { - if (state->mode == ARB_vertex) { - yyerror(& @1, state, "invalid program result name"); - YYERROR; - } else { - if (!state->option.DrawBuffers) { - /* From the ARB_draw_buffers spec (same text exists - * for ATI_draw_buffers): - * - * If this option is not specified, a fragment - * program that attempts to bind - * "result.color[n]" will fail to load, and only - * "result.color" will be allowed. - */ - yyerror(& @1, state, - "result.color[] used without " - "`OPTION ARB_draw_buffers' or " - "`OPTION ATI_draw_buffers'"); - YYERROR; - } else if ($2 >= state->MaxDrawBuffers) { - yyerror(& @1, state, - "result.color[] exceeds MAX_DRAW_BUFFERS_ARB"); - YYERROR; - } - $$ = FRAG_RESULT_DATA0 + $2; - } - } - | FRONT - { - if (state->mode == ARB_vertex) { - $$ = VERT_RESULT_COL0; - } else { - yyerror(& @1, state, "invalid program result name"); - YYERROR; - } - } - | BACK - { - if (state->mode == ARB_vertex) { - $$ = VERT_RESULT_BFC0; - } else { - yyerror(& @1, state, "invalid program result name"); - YYERROR; - } - } - ; - -optResultColorType: - { - $$ = 0; - } - | PRIMARY - { - if (state->mode == ARB_vertex) { - $$ = 0; - } else { - yyerror(& @1, state, "invalid program result name"); - YYERROR; - } - } - | SECONDARY - { - if (state->mode == ARB_vertex) { - $$ = 1; - } else { - yyerror(& @1, state, "invalid program result name"); - YYERROR; - } - } - ; - -optFaceType: { $$ = 0; } - | FRONT { $$ = 0; } - | BACK { $$ = 1; } - ; - -optColorType: { $$ = 0; } - | PRIMARY { $$ = 0; } - | SECONDARY { $$ = 1; } - ; - -optTexCoordUnitNum: { $$ = 0; } - | '[' texCoordUnitNum ']' { $$ = $2; } - ; - -optTexImageUnitNum: { $$ = 0; } - | '[' texImageUnitNum ']' { $$ = $2; } - ; - -optLegacyTexUnitNum: { $$ = 0; } - | '[' legacyTexUnitNum ']' { $$ = $2; } - ; - -texCoordUnitNum: INTEGER - { - if ((unsigned) $1 >= state->MaxTextureCoordUnits) { - yyerror(& @1, state, "invalid texture coordinate unit selector"); - YYERROR; - } - - $$ = $1; - } - ; - -texImageUnitNum: INTEGER - { - if ((unsigned) $1 >= state->MaxTextureImageUnits) { - yyerror(& @1, state, "invalid texture image unit selector"); - YYERROR; - } - - $$ = $1; - } - ; - -legacyTexUnitNum: INTEGER - { - if ((unsigned) $1 >= state->MaxTextureUnits) { - yyerror(& @1, state, "invalid texture unit selector"); - YYERROR; - } - - $$ = $1; - } - ; - -ALIAS_statement: ALIAS IDENTIFIER '=' USED_IDENTIFIER - { - struct asm_symbol *exist = (struct asm_symbol *) - _mesa_symbol_table_find_symbol(state->st, 0, $2); - struct asm_symbol *target = (struct asm_symbol *) - _mesa_symbol_table_find_symbol(state->st, 0, $4); - - free($4); - - if (exist != NULL) { - char m[1000]; - _mesa_snprintf(m, sizeof(m), "redeclared identifier: %s", $2); - free($2); - yyerror(& @2, state, m); - YYERROR; - } else if (target == NULL) { - free($2); - yyerror(& @4, state, - "undefined variable binding in ALIAS statement"); - YYERROR; - } else { - _mesa_symbol_table_add_symbol(state->st, 0, $2, target); - } - } - ; - -string: IDENTIFIER - | USED_IDENTIFIER - ; - -%% - -void -asm_instruction_set_operands(struct asm_instruction *inst, - const struct prog_dst_register *dst, - const struct asm_src_register *src0, - const struct asm_src_register *src1, - const struct asm_src_register *src2) -{ - /* In the core ARB extensions only the KIL instruction doesn't have a - * destination register. - */ - if (dst == NULL) { - init_dst_reg(& inst->Base.DstReg); - } else { - inst->Base.DstReg = *dst; - } - - /* The only instruction that doesn't have any source registers is the - * condition-code based KIL instruction added by NV_fragment_program_option. - */ - if (src0 != NULL) { - inst->Base.SrcReg[0] = src0->Base; - inst->SrcReg[0] = *src0; - } else { - init_src_reg(& inst->SrcReg[0]); - } - - if (src1 != NULL) { - inst->Base.SrcReg[1] = src1->Base; - inst->SrcReg[1] = *src1; - } else { - init_src_reg(& inst->SrcReg[1]); - } - - if (src2 != NULL) { - inst->Base.SrcReg[2] = src2->Base; - inst->SrcReg[2] = *src2; - } else { - init_src_reg(& inst->SrcReg[2]); - } -} - - -struct asm_instruction * -asm_instruction_ctor(gl_inst_opcode op, - const struct prog_dst_register *dst, - const struct asm_src_register *src0, - const struct asm_src_register *src1, - const struct asm_src_register *src2) -{ - struct asm_instruction *inst = CALLOC_STRUCT(asm_instruction); - - if (inst) { - _mesa_init_instructions(& inst->Base, 1); - inst->Base.Opcode = op; - - asm_instruction_set_operands(inst, dst, src0, src1, src2); - } - - return inst; -} - - -struct asm_instruction * -asm_instruction_copy_ctor(const struct prog_instruction *base, - const struct prog_dst_register *dst, - const struct asm_src_register *src0, - const struct asm_src_register *src1, - const struct asm_src_register *src2) -{ - struct asm_instruction *inst = CALLOC_STRUCT(asm_instruction); - - if (inst) { - _mesa_init_instructions(& inst->Base, 1); - inst->Base.Opcode = base->Opcode; - inst->Base.CondUpdate = base->CondUpdate; - inst->Base.CondDst = base->CondDst; - inst->Base.SaturateMode = base->SaturateMode; - inst->Base.Precision = base->Precision; - - asm_instruction_set_operands(inst, dst, src0, src1, src2); - } - - return inst; -} - - -void -init_dst_reg(struct prog_dst_register *r) -{ - memset(r, 0, sizeof(*r)); - r->File = PROGRAM_UNDEFINED; - r->WriteMask = WRITEMASK_XYZW; - r->CondMask = COND_TR; - r->CondSwizzle = SWIZZLE_NOOP; -} - - -/** Like init_dst_reg() but set the File and Index fields. */ -void -set_dst_reg(struct prog_dst_register *r, gl_register_file file, GLint index) -{ - const GLint maxIndex = 1 << INST_INDEX_BITS; - const GLint minIndex = 0; - ASSERT(index >= minIndex); - (void) minIndex; - ASSERT(index <= maxIndex); - (void) maxIndex; - ASSERT(file == PROGRAM_TEMPORARY || - file == PROGRAM_ADDRESS || - file == PROGRAM_OUTPUT); - memset(r, 0, sizeof(*r)); - r->File = file; - r->Index = index; - r->WriteMask = WRITEMASK_XYZW; - r->CondMask = COND_TR; - r->CondSwizzle = SWIZZLE_NOOP; -} - - -void -init_src_reg(struct asm_src_register *r) -{ - memset(r, 0, sizeof(*r)); - r->Base.File = PROGRAM_UNDEFINED; - r->Base.Swizzle = SWIZZLE_NOOP; - r->Symbol = NULL; -} - - -/** Like init_src_reg() but set the File and Index fields. - * \return GL_TRUE if a valid src register, GL_FALSE otherwise - */ -void -set_src_reg(struct asm_src_register *r, gl_register_file file, GLint index) -{ - set_src_reg_swz(r, file, index, SWIZZLE_XYZW); -} - - -void -set_src_reg_swz(struct asm_src_register *r, gl_register_file file, GLint index, - GLuint swizzle) -{ - const GLint maxIndex = (1 << INST_INDEX_BITS) - 1; - const GLint minIndex = -(1 << INST_INDEX_BITS); - ASSERT(file < PROGRAM_FILE_MAX); - ASSERT(index >= minIndex); - (void) minIndex; - ASSERT(index <= maxIndex); - (void) maxIndex; - memset(r, 0, sizeof(*r)); - r->Base.File = file; - r->Base.Index = index; - r->Base.Swizzle = swizzle; - r->Symbol = NULL; -} - - -/** - * Validate the set of inputs used by a program - * - * Validates that legal sets of inputs are used by the program. In this case - * "used" included both reading the input or binding the input to a name using - * the \c ATTRIB command. - * - * \return - * \c TRUE if the combination of inputs used is valid, \c FALSE otherwise. - */ -int -validate_inputs(struct YYLTYPE *locp, struct asm_parser_state *state) -{ - const int inputs = state->prog->InputsRead | state->InputsBound; - - if (((inputs & 0x0ffff) & (inputs >> 16)) != 0) { - yyerror(locp, state, "illegal use of generic attribute and name attribute"); - return 0; - } - - return 1; -} - - -struct asm_symbol * -declare_variable(struct asm_parser_state *state, char *name, enum asm_type t, - struct YYLTYPE *locp) -{ - struct asm_symbol *s = NULL; - struct asm_symbol *exist = (struct asm_symbol *) - _mesa_symbol_table_find_symbol(state->st, 0, name); - - - if (exist != NULL) { - yyerror(locp, state, "redeclared identifier"); - } else { - s = calloc(1, sizeof(struct asm_symbol)); - s->name = name; - s->type = t; - - switch (t) { - case at_temp: - if (state->prog->NumTemporaries >= state->limits->MaxTemps) { - yyerror(locp, state, "too many temporaries declared"); - free(s); - return NULL; - } - - s->temp_binding = state->prog->NumTemporaries; - state->prog->NumTemporaries++; - break; - - case at_address: - if (state->prog->NumAddressRegs >= state->limits->MaxAddressRegs) { - yyerror(locp, state, "too many address registers declared"); - free(s); - return NULL; - } - - /* FINISHME: Add support for multiple address registers. - */ - state->prog->NumAddressRegs++; - break; - - default: - break; - } - - _mesa_symbol_table_add_symbol(state->st, 0, s->name, s); - s->next = state->sym; - state->sym = s; - } - - return s; -} - - -int add_state_reference(struct gl_program_parameter_list *param_list, - const gl_state_index tokens[STATE_LENGTH]) -{ - const GLuint size = 4; /* XXX fix */ - char *name; - GLint index; - - name = _mesa_program_state_string(tokens); - index = _mesa_add_parameter(param_list, PROGRAM_STATE_VAR, name, - size, GL_NONE, NULL, tokens, 0x0); - param_list->StateFlags |= _mesa_program_state_flags(tokens); - - /* free name string here since we duplicated it in add_parameter() */ - free(name); - - return index; -} - - -int -initialize_symbol_from_state(struct gl_program *prog, - struct asm_symbol *param_var, - const gl_state_index tokens[STATE_LENGTH]) -{ - int idx = -1; - gl_state_index state_tokens[STATE_LENGTH]; - - - memcpy(state_tokens, tokens, sizeof(state_tokens)); - - param_var->type = at_param; - param_var->param_binding_type = PROGRAM_STATE_VAR; - - /* If we are adding a STATE_MATRIX that has multiple rows, we need to - * unroll it and call add_state_reference() for each row - */ - if ((state_tokens[0] == STATE_MODELVIEW_MATRIX || - state_tokens[0] == STATE_PROJECTION_MATRIX || - state_tokens[0] == STATE_MVP_MATRIX || - state_tokens[0] == STATE_TEXTURE_MATRIX || - state_tokens[0] == STATE_PROGRAM_MATRIX) - && (state_tokens[2] != state_tokens[3])) { - int row; - const int first_row = state_tokens[2]; - const int last_row = state_tokens[3]; - - for (row = first_row; row <= last_row; row++) { - state_tokens[2] = state_tokens[3] = row; - - idx = add_state_reference(prog->Parameters, state_tokens); - if (param_var->param_binding_begin == ~0U) { - param_var->param_binding_begin = idx; - param_var->param_binding_swizzle = SWIZZLE_XYZW; - } - - param_var->param_binding_length++; - } - } - else { - idx = add_state_reference(prog->Parameters, state_tokens); - if (param_var->param_binding_begin == ~0U) { - param_var->param_binding_begin = idx; - param_var->param_binding_swizzle = SWIZZLE_XYZW; - } - param_var->param_binding_length++; - } - - return idx; -} - - -int -initialize_symbol_from_param(struct gl_program *prog, - struct asm_symbol *param_var, - const gl_state_index tokens[STATE_LENGTH]) -{ - int idx = -1; - gl_state_index state_tokens[STATE_LENGTH]; - - - memcpy(state_tokens, tokens, sizeof(state_tokens)); - - assert((state_tokens[0] == STATE_VERTEX_PROGRAM) - || (state_tokens[0] == STATE_FRAGMENT_PROGRAM)); - assert((state_tokens[1] == STATE_ENV) - || (state_tokens[1] == STATE_LOCAL)); - - /* - * The param type is STATE_VAR. The program parameter entry will - * effectively be a pointer into the LOCAL or ENV parameter array. - */ - param_var->type = at_param; - param_var->param_binding_type = PROGRAM_STATE_VAR; - - /* If we are adding a STATE_ENV or STATE_LOCAL that has multiple elements, - * we need to unroll it and call add_state_reference() for each row - */ - if (state_tokens[2] != state_tokens[3]) { - int row; - const int first_row = state_tokens[2]; - const int last_row = state_tokens[3]; - - for (row = first_row; row <= last_row; row++) { - state_tokens[2] = state_tokens[3] = row; - - idx = add_state_reference(prog->Parameters, state_tokens); - if (param_var->param_binding_begin == ~0U) { - param_var->param_binding_begin = idx; - param_var->param_binding_swizzle = SWIZZLE_XYZW; - } - param_var->param_binding_length++; - } - } - else { - idx = add_state_reference(prog->Parameters, state_tokens); - if (param_var->param_binding_begin == ~0U) { - param_var->param_binding_begin = idx; - param_var->param_binding_swizzle = SWIZZLE_XYZW; - } - param_var->param_binding_length++; - } - - return idx; -} - - -/** - * Put a float/vector constant/literal into the parameter list. - * \param param_var returns info about the parameter/constant's location, - * binding, type, etc. - * \param vec the vector/constant to add - * \param allowSwizzle if true, try to consolidate constants which only differ - * by a swizzle. We don't want to do this when building - * arrays of constants that may be indexed indirectly. - * \return index of the constant in the parameter list. - */ -int -initialize_symbol_from_const(struct gl_program *prog, - struct asm_symbol *param_var, - const struct asm_vector *vec, - GLboolean allowSwizzle) -{ - unsigned swizzle; - const int idx = _mesa_add_unnamed_constant(prog->Parameters, - vec->data, vec->count, - allowSwizzle ? &swizzle : NULL); - - param_var->type = at_param; - param_var->param_binding_type = PROGRAM_CONSTANT; - - if (param_var->param_binding_begin == ~0U) { - param_var->param_binding_begin = idx; - param_var->param_binding_swizzle = allowSwizzle ? swizzle : SWIZZLE_XYZW; - } - param_var->param_binding_length++; - - return idx; -} - - -char * -make_error_string(const char *fmt, ...) -{ - int length; - char *str; - va_list args; - - - /* Call vsnprintf once to determine how large the final string is. Call it - * again to do the actual formatting. from the vsnprintf manual page: - * - * Upon successful return, these functions return the number of - * characters printed (not including the trailing '\0' used to end - * output to strings). - */ - va_start(args, fmt); - length = 1 + vsnprintf(NULL, 0, fmt, args); - va_end(args); - - str = malloc(length); - if (str) { - va_start(args, fmt); - vsnprintf(str, length, fmt, args); - va_end(args); - } - - return str; -} - - -void -yyerror(YYLTYPE *locp, struct asm_parser_state *state, const char *s) -{ - char *err_str; - - - err_str = make_error_string("glProgramStringARB(%s)\n", s); - if (err_str) { - _mesa_error(state->ctx, GL_INVALID_OPERATION, "%s", err_str); - free(err_str); - } - - err_str = make_error_string("line %u, char %u: error: %s\n", - locp->first_line, locp->first_column, s); - _mesa_set_program_error(state->ctx, locp->position, err_str); - - if (err_str) { - free(err_str); - } -} - - -GLboolean -_mesa_parse_arb_program(struct gl_context *ctx, GLenum target, const GLubyte *str, - GLsizei len, struct asm_parser_state *state) -{ - struct asm_instruction *inst; - unsigned i; - GLubyte *strz; - GLboolean result = GL_FALSE; - void *temp; - struct asm_symbol *sym; - - state->ctx = ctx; - state->prog->Target = target; - state->prog->Parameters = _mesa_new_parameter_list(); - - /* Make a copy of the program string and force it to be NUL-terminated. - */ - strz = (GLubyte *) malloc(len + 1); - if (strz == NULL) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glProgramStringARB"); - return GL_FALSE; - } - memcpy (strz, str, len); - strz[len] = '\0'; - - state->prog->String = strz; - - state->st = _mesa_symbol_table_ctor(); - - state->limits = (target == GL_VERTEX_PROGRAM_ARB) - ? & ctx->Const.VertexProgram - : & ctx->Const.FragmentProgram; - - state->MaxTextureImageUnits = ctx->Const.MaxTextureImageUnits; - state->MaxTextureCoordUnits = ctx->Const.MaxTextureCoordUnits; - state->MaxTextureUnits = ctx->Const.MaxTextureUnits; - state->MaxClipPlanes = ctx->Const.MaxClipPlanes; - state->MaxLights = ctx->Const.MaxLights; - state->MaxProgramMatrices = ctx->Const.MaxProgramMatrices; - state->MaxDrawBuffers = ctx->Const.MaxDrawBuffers; - - state->state_param_enum = (target == GL_VERTEX_PROGRAM_ARB) - ? STATE_VERTEX_PROGRAM : STATE_FRAGMENT_PROGRAM; - - _mesa_set_program_error(ctx, -1, NULL); - - _mesa_program_lexer_ctor(& state->scanner, state, (const char *) str, len); - yyparse(state); - _mesa_program_lexer_dtor(state->scanner); - - - if (ctx->Program.ErrorPos != -1) { - goto error; - } - - if (! _mesa_layout_parameters(state)) { - struct YYLTYPE loc; - - loc.first_line = 0; - loc.first_column = 0; - loc.position = len; - - yyerror(& loc, state, "invalid PARAM usage"); - goto error; - } - - - - /* Add one instruction to store the "END" instruction. - */ - state->prog->Instructions = - _mesa_alloc_instructions(state->prog->NumInstructions + 1); - inst = state->inst_head; - for (i = 0; i < state->prog->NumInstructions; i++) { - struct asm_instruction *const temp = inst->next; - - state->prog->Instructions[i] = inst->Base; - inst = temp; - } - - /* Finally, tag on an OPCODE_END instruction */ - { - const GLuint numInst = state->prog->NumInstructions; - _mesa_init_instructions(state->prog->Instructions + numInst, 1); - state->prog->Instructions[numInst].Opcode = OPCODE_END; - } - state->prog->NumInstructions++; - - state->prog->NumParameters = state->prog->Parameters->NumParameters; - state->prog->NumAttributes = _mesa_bitcount(state->prog->InputsRead); - - /* - * Initialize native counts to logical counts. The device driver may - * change them if program is translated into a hardware program. - */ - state->prog->NumNativeInstructions = state->prog->NumInstructions; - state->prog->NumNativeTemporaries = state->prog->NumTemporaries; - state->prog->NumNativeParameters = state->prog->NumParameters; - state->prog->NumNativeAttributes = state->prog->NumAttributes; - state->prog->NumNativeAddressRegs = state->prog->NumAddressRegs; - - result = GL_TRUE; - -error: - for (inst = state->inst_head; inst != NULL; inst = temp) { - temp = inst->next; - free(inst); - } - - state->inst_head = NULL; - state->inst_tail = NULL; - - for (sym = state->sym; sym != NULL; sym = temp) { - temp = sym->next; - - free((void *) sym->name); - free(sym); - } - state->sym = NULL; - - _mesa_symbol_table_dtor(state->st); - state->st = NULL; - - return result; -} +%{ +/* + * Copyright © 2009 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ +#include +#include +#include + +#include "main/mtypes.h" +#include "main/imports.h" +#include "program/program.h" +#include "program/prog_parameter.h" +#include "program/prog_parameter_layout.h" +#include "program/prog_statevars.h" +#include "program/prog_instruction.h" + +#include "program/symbol_table.h" +#include "program/program_parser.h" + +extern void *yy_scan_string(char *); +extern void yy_delete_buffer(void *); + +static struct asm_symbol *declare_variable(struct asm_parser_state *state, + char *name, enum asm_type t, struct YYLTYPE *locp); + +static int add_state_reference(struct gl_program_parameter_list *param_list, + const gl_state_index tokens[STATE_LENGTH]); + +static int initialize_symbol_from_state(struct gl_program *prog, + struct asm_symbol *param_var, const gl_state_index tokens[STATE_LENGTH]); + +static int initialize_symbol_from_param(struct gl_program *prog, + struct asm_symbol *param_var, const gl_state_index tokens[STATE_LENGTH]); + +static int initialize_symbol_from_const(struct gl_program *prog, + struct asm_symbol *param_var, const struct asm_vector *vec, + GLboolean allowSwizzle); + +static int yyparse(struct asm_parser_state *state); + +static char *make_error_string(const char *fmt, ...); + +static void yyerror(struct YYLTYPE *locp, struct asm_parser_state *state, + const char *s); + +static int validate_inputs(struct YYLTYPE *locp, + struct asm_parser_state *state); + +static void init_dst_reg(struct prog_dst_register *r); + +static void set_dst_reg(struct prog_dst_register *r, + gl_register_file file, GLint index); + +static void init_src_reg(struct asm_src_register *r); + +static void set_src_reg(struct asm_src_register *r, + gl_register_file file, GLint index); + +static void set_src_reg_swz(struct asm_src_register *r, + gl_register_file file, GLint index, GLuint swizzle); + +static void asm_instruction_set_operands(struct asm_instruction *inst, + const struct prog_dst_register *dst, const struct asm_src_register *src0, + const struct asm_src_register *src1, const struct asm_src_register *src2); + +static struct asm_instruction *asm_instruction_ctor(gl_inst_opcode op, + const struct prog_dst_register *dst, const struct asm_src_register *src0, + const struct asm_src_register *src1, const struct asm_src_register *src2); + +static struct asm_instruction *asm_instruction_copy_ctor( + const struct prog_instruction *base, const struct prog_dst_register *dst, + const struct asm_src_register *src0, const struct asm_src_register *src1, + const struct asm_src_register *src2); + +#ifndef FALSE +#define FALSE 0 +#define TRUE (!FALSE) +#endif + +#define YYLLOC_DEFAULT(Current, Rhs, N) \ + do { \ + if (YYID(N)) { \ + (Current).first_line = YYRHSLOC(Rhs, 1).first_line; \ + (Current).first_column = YYRHSLOC(Rhs, 1).first_column; \ + (Current).position = YYRHSLOC(Rhs, 1).position; \ + (Current).last_line = YYRHSLOC(Rhs, N).last_line; \ + (Current).last_column = YYRHSLOC(Rhs, N).last_column; \ + } else { \ + (Current).first_line = YYRHSLOC(Rhs, 0).last_line; \ + (Current).last_line = (Current).first_line; \ + (Current).first_column = YYRHSLOC(Rhs, 0).last_column; \ + (Current).last_column = (Current).first_column; \ + (Current).position = YYRHSLOC(Rhs, 0).position \ + + (Current).first_column; \ + } \ + } while(YYID(0)) + +#define YYLEX_PARAM state->scanner +%} + +%pure-parser +%locations +%parse-param { struct asm_parser_state *state } +%error-verbose +%lex-param { void *scanner } + +%union { + struct asm_instruction *inst; + struct asm_symbol *sym; + struct asm_symbol temp_sym; + struct asm_swizzle_mask swiz_mask; + struct asm_src_register src_reg; + struct prog_dst_register dst_reg; + struct prog_instruction temp_inst; + char *string; + unsigned result; + unsigned attrib; + int integer; + float real; + gl_state_index state[STATE_LENGTH]; + int negate; + struct asm_vector vector; + gl_inst_opcode opcode; + + struct { + unsigned swz; + unsigned rgba_valid:1; + unsigned xyzw_valid:1; + unsigned negate:1; + } ext_swizzle; +} + +%token ARBvp_10 ARBfp_10 + +/* Tokens for assembler pseudo-ops */ +%token ADDRESS +%token ALIAS ATTRIB +%token OPTION OUTPUT +%token PARAM +%token TEMP +%token END + + /* Tokens for instructions */ +%token BIN_OP BINSC_OP SAMPLE_OP SCALAR_OP TRI_OP VECTOR_OP +%token ARL KIL SWZ TXD_OP + +%token INTEGER +%token REAL + +%token AMBIENT ATTENUATION +%token BACK +%token CLIP COLOR +%token DEPTH DIFFUSE DIRECTION +%token EMISSION ENV EYE +%token FOG FOGCOORD FRAGMENT FRONT +%token HALF +%token INVERSE INVTRANS +%token LIGHT LIGHTMODEL LIGHTPROD LOCAL +%token MATERIAL MAT_PROGRAM MATRIX MATRIXINDEX MODELVIEW MVP +%token NORMAL +%token OBJECT +%token PALETTE PARAMS PLANE POINT_TOK POINTSIZE POSITION PRIMARY PROGRAM PROJECTION +%token RANGE RESULT ROW +%token SCENECOLOR SECONDARY SHININESS SIZE_TOK SPECULAR SPOT STATE +%token TEXCOORD TEXENV TEXGEN TEXGEN_Q TEXGEN_R TEXGEN_S TEXGEN_T TEXTURE TRANSPOSE +%token TEXTURE_UNIT TEX_1D TEX_2D TEX_3D TEX_CUBE TEX_RECT +%token TEX_SHADOW1D TEX_SHADOW2D TEX_SHADOWRECT +%token TEX_ARRAY1D TEX_ARRAY2D TEX_ARRAYSHADOW1D TEX_ARRAYSHADOW2D +%token VERTEX VTXATTRIB +%token WEIGHT + +%token IDENTIFIER USED_IDENTIFIER +%type string +%token MASK4 MASK3 MASK2 MASK1 SWIZZLE +%token DOT_DOT +%token DOT + +%type instruction ALU_instruction TexInstruction +%type ARL_instruction VECTORop_instruction +%type SCALARop_instruction BINSCop_instruction BINop_instruction +%type TRIop_instruction TXD_instruction SWZ_instruction SAMPLE_instruction +%type KIL_instruction + +%type dstReg maskedDstReg maskedAddrReg +%type srcReg scalarUse scalarSrcReg swizzleSrcReg +%type scalarSuffix swizzleSuffix extendedSwizzle +%type extSwizComp extSwizSel +%type optionalMask + +%type progParamArray +%type addrRegRelOffset addrRegPosOffset addrRegNegOffset +%type progParamArrayMem progParamArrayAbs progParamArrayRel +%type addrReg +%type addrComponent addrWriteMask + +%type ccMaskRule ccTest ccMaskRule2 ccTest2 optionalCcMask + +%type resultBinding resultColBinding +%type optFaceType optColorType +%type optResultFaceType optResultColorType + +%type optTexImageUnitNum texImageUnitNum +%type optTexCoordUnitNum texCoordUnitNum +%type optLegacyTexUnitNum legacyTexUnitNum +%type texImageUnit texTarget +%type vtxAttribNum + +%type attribBinding vtxAttribItem fragAttribItem + +%type paramSingleInit paramSingleItemDecl +%type optArraySize + +%type stateSingleItem stateMultipleItem +%type stateMaterialItem +%type stateLightItem stateLightModelItem stateLightProdItem +%type stateTexGenItem stateFogItem stateClipPlaneItem statePointItem +%type stateMatrixItem stateMatrixRow stateMatrixRows +%type stateTexEnvItem stateDepthItem + +%type stateLModProperty +%type stateMatrixName optMatrixRows + +%type stateMatProperty +%type stateLightProperty stateSpotProperty +%type stateLightNumber stateLProdProperty +%type stateTexGenType stateTexGenCoord +%type stateTexEnvProperty +%type stateFogProperty +%type stateClipPlaneNum +%type statePointProperty + +%type stateOptMatModifier stateMatModifier stateMatrixRowNum +%type stateOptModMatNum stateModMatNum statePaletteMatNum +%type stateProgramMatNum + +%type ambDiffSpecProperty + +%type programSingleItem progEnvParam progLocalParam +%type programMultipleItem progEnvParams progLocalParams + +%type paramMultipleInit paramMultInitList paramMultipleItem +%type paramSingleItemUse + +%type progEnvParamNum progLocalParamNum +%type progEnvParamNums progLocalParamNums + +%type paramConstDecl paramConstUse +%type paramConstScalarDecl paramConstScalarUse paramConstVector +%type signedFloatConstant +%type optionalSign + +%{ +extern int yylex(YYSTYPE *yylval_param, YYLTYPE *yylloc_param, + void *yyscanner); +%} + +%% + +program: language optionSequence statementSequence END + ; + +language: ARBvp_10 + { + if (state->prog->Target != GL_VERTEX_PROGRAM_ARB) { + yyerror(& @1, state, "invalid fragment program header"); + + } + state->mode = ARB_vertex; + } + | ARBfp_10 + { + if (state->prog->Target != GL_FRAGMENT_PROGRAM_ARB) { + yyerror(& @1, state, "invalid vertex program header"); + } + state->mode = ARB_fragment; + + state->option.TexRect = + (state->ctx->Extensions.NV_texture_rectangle != GL_FALSE); + } + ; + +optionSequence: optionSequence option + | + ; + +option: OPTION string ';' + { + int valid = 0; + + if (state->mode == ARB_vertex) { + valid = _mesa_ARBvp_parse_option(state, $2); + } else if (state->mode == ARB_fragment) { + valid = _mesa_ARBfp_parse_option(state, $2); + } + + + free($2); + + if (!valid) { + const char *const err_str = (state->mode == ARB_vertex) + ? "invalid ARB vertex program option" + : "invalid ARB fragment program option"; + + yyerror(& @2, state, err_str); + YYERROR; + } + } + ; + +statementSequence: statementSequence statement + | + ; + +statement: instruction ';' + { + if ($1 != NULL) { + if (state->inst_tail == NULL) { + state->inst_head = $1; + } else { + state->inst_tail->next = $1; + } + + state->inst_tail = $1; + $1->next = NULL; + + state->prog->NumInstructions++; + } + } + | namingStatement ';' + ; + +instruction: ALU_instruction + { + $$ = $1; + state->prog->NumAluInstructions++; + } + | TexInstruction + { + $$ = $1; + state->prog->NumTexInstructions++; + } + ; + +ALU_instruction: ARL_instruction + | VECTORop_instruction + | SCALARop_instruction + | BINSCop_instruction + | BINop_instruction + | TRIop_instruction + | SWZ_instruction + ; + +TexInstruction: SAMPLE_instruction + | KIL_instruction + | TXD_instruction + ; + +ARL_instruction: ARL maskedAddrReg ',' scalarSrcReg + { + $$ = asm_instruction_ctor(OPCODE_ARL, & $2, & $4, NULL, NULL); + } + ; + +VECTORop_instruction: VECTOR_OP maskedDstReg ',' swizzleSrcReg + { + $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, NULL, NULL); + } + ; + +SCALARop_instruction: SCALAR_OP maskedDstReg ',' scalarSrcReg + { + $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, NULL, NULL); + } + ; + +BINSCop_instruction: BINSC_OP maskedDstReg ',' scalarSrcReg ',' scalarSrcReg + { + $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, & $6, NULL); + } + ; + + +BINop_instruction: BIN_OP maskedDstReg ',' swizzleSrcReg ',' swizzleSrcReg + { + $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, & $6, NULL); + } + ; + +TRIop_instruction: TRI_OP maskedDstReg ',' + swizzleSrcReg ',' swizzleSrcReg ',' swizzleSrcReg + { + $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, & $6, & $8); + } + ; + +SAMPLE_instruction: SAMPLE_OP maskedDstReg ',' swizzleSrcReg ',' texImageUnit ',' texTarget + { + $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, NULL, NULL); + if ($$ != NULL) { + const GLbitfield tex_mask = (1U << $6); + GLbitfield shadow_tex = 0; + GLbitfield target_mask = 0; + + + $$->Base.TexSrcUnit = $6; + + if ($8 < 0) { + shadow_tex = tex_mask; + + $$->Base.TexSrcTarget = -$8; + $$->Base.TexShadow = 1; + } else { + $$->Base.TexSrcTarget = $8; + } + + target_mask = (1U << $$->Base.TexSrcTarget); + + /* If this texture unit was previously accessed and that access + * had a different texture target, generate an error. + * + * If this texture unit was previously accessed and that access + * had a different shadow mode, generate an error. + */ + if ((state->prog->TexturesUsed[$6] != 0) + && ((state->prog->TexturesUsed[$6] != target_mask) + || ((state->prog->ShadowSamplers & tex_mask) + != shadow_tex))) { + yyerror(& @8, state, + "multiple targets used on one texture image unit"); + YYERROR; + } + + + state->prog->TexturesUsed[$6] |= target_mask; + state->prog->ShadowSamplers |= shadow_tex; + } + } + ; + +KIL_instruction: KIL swizzleSrcReg + { + $$ = asm_instruction_ctor(OPCODE_KIL, NULL, & $2, NULL, NULL); + state->fragment.UsesKill = 1; + } + | KIL ccTest + { + $$ = asm_instruction_ctor(OPCODE_KIL_NV, NULL, NULL, NULL, NULL); + $$->Base.DstReg.CondMask = $2.CondMask; + $$->Base.DstReg.CondSwizzle = $2.CondSwizzle; + $$->Base.DstReg.CondSrc = $2.CondSrc; + state->fragment.UsesKill = 1; + } + ; + +TXD_instruction: TXD_OP maskedDstReg ',' swizzleSrcReg ',' swizzleSrcReg ',' swizzleSrcReg ',' texImageUnit ',' texTarget + { + $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, & $6, & $8); + if ($$ != NULL) { + const GLbitfield tex_mask = (1U << $10); + GLbitfield shadow_tex = 0; + GLbitfield target_mask = 0; + + + $$->Base.TexSrcUnit = $10; + + if ($12 < 0) { + shadow_tex = tex_mask; + + $$->Base.TexSrcTarget = -$12; + $$->Base.TexShadow = 1; + } else { + $$->Base.TexSrcTarget = $12; + } + + target_mask = (1U << $$->Base.TexSrcTarget); + + /* If this texture unit was previously accessed and that access + * had a different texture target, generate an error. + * + * If this texture unit was previously accessed and that access + * had a different shadow mode, generate an error. + */ + if ((state->prog->TexturesUsed[$10] != 0) + && ((state->prog->TexturesUsed[$10] != target_mask) + || ((state->prog->ShadowSamplers & tex_mask) + != shadow_tex))) { + yyerror(& @12, state, + "multiple targets used on one texture image unit"); + YYERROR; + } + + + state->prog->TexturesUsed[$10] |= target_mask; + state->prog->ShadowSamplers |= shadow_tex; + } + } + ; + +texImageUnit: TEXTURE_UNIT optTexImageUnitNum + { + $$ = $2; + } + ; + +texTarget: TEX_1D { $$ = TEXTURE_1D_INDEX; } + | TEX_2D { $$ = TEXTURE_2D_INDEX; } + | TEX_3D { $$ = TEXTURE_3D_INDEX; } + | TEX_CUBE { $$ = TEXTURE_CUBE_INDEX; } + | TEX_RECT { $$ = TEXTURE_RECT_INDEX; } + | TEX_SHADOW1D { $$ = -TEXTURE_1D_INDEX; } + | TEX_SHADOW2D { $$ = -TEXTURE_2D_INDEX; } + | TEX_SHADOWRECT { $$ = -TEXTURE_RECT_INDEX; } + | TEX_ARRAY1D { $$ = TEXTURE_1D_ARRAY_INDEX; } + | TEX_ARRAY2D { $$ = TEXTURE_2D_ARRAY_INDEX; } + | TEX_ARRAYSHADOW1D { $$ = -TEXTURE_1D_ARRAY_INDEX; } + | TEX_ARRAYSHADOW2D { $$ = -TEXTURE_2D_ARRAY_INDEX; } + ; + +SWZ_instruction: SWZ maskedDstReg ',' srcReg ',' extendedSwizzle + { + /* FIXME: Is this correct? Should the extenedSwizzle be applied + * FIXME: to the existing swizzle? + */ + $4.Base.Swizzle = $6.swizzle; + $4.Base.Negate = $6.mask; + + $$ = asm_instruction_copy_ctor(& $1, & $2, & $4, NULL, NULL); + } + ; + +scalarSrcReg: optionalSign scalarUse + { + $$ = $2; + + if ($1) { + $$.Base.Negate = ~$$.Base.Negate; + } + } + | optionalSign '|' scalarUse '|' + { + $$ = $3; + + if (!state->option.NV_fragment) { + yyerror(& @2, state, "unexpected character '|'"); + YYERROR; + } + + if ($1) { + $$.Base.Negate = ~$$.Base.Negate; + } + + $$.Base.Abs = 1; + } + ; + +scalarUse: srcReg scalarSuffix + { + $$ = $1; + + $$.Base.Swizzle = _mesa_combine_swizzles($$.Base.Swizzle, + $2.swizzle); + } + | paramConstScalarUse + { + struct asm_symbol temp_sym; + + if (!state->option.NV_fragment) { + yyerror(& @1, state, "expected scalar suffix"); + YYERROR; + } + + memset(& temp_sym, 0, sizeof(temp_sym)); + temp_sym.param_binding_begin = ~0; + initialize_symbol_from_const(state->prog, & temp_sym, & $1, GL_TRUE); + + set_src_reg_swz(& $$, PROGRAM_CONSTANT, + temp_sym.param_binding_begin, + temp_sym.param_binding_swizzle); + } + ; + +swizzleSrcReg: optionalSign srcReg swizzleSuffix + { + $$ = $2; + + if ($1) { + $$.Base.Negate = ~$$.Base.Negate; + } + + $$.Base.Swizzle = _mesa_combine_swizzles($$.Base.Swizzle, + $3.swizzle); + } + | optionalSign '|' srcReg swizzleSuffix '|' + { + $$ = $3; + + if (!state->option.NV_fragment) { + yyerror(& @2, state, "unexpected character '|'"); + YYERROR; + } + + if ($1) { + $$.Base.Negate = ~$$.Base.Negate; + } + + $$.Base.Abs = 1; + $$.Base.Swizzle = _mesa_combine_swizzles($$.Base.Swizzle, + $4.swizzle); + } + + ; + +maskedDstReg: dstReg optionalMask optionalCcMask + { + $$ = $1; + $$.WriteMask = $2.mask; + $$.CondMask = $3.CondMask; + $$.CondSwizzle = $3.CondSwizzle; + $$.CondSrc = $3.CondSrc; + + if ($$.File == PROGRAM_OUTPUT) { + /* Technically speaking, this should check that it is in + * vertex program mode. However, PositionInvariant can never be + * set in fragment program mode, so it is somewhat irrelevant. + */ + if (state->option.PositionInvariant + && ($$.Index == VERT_RESULT_HPOS)) { + yyerror(& @1, state, "position-invariant programs cannot " + "write position"); + YYERROR; + } + + state->prog->OutputsWritten |= BITFIELD64_BIT($$.Index); + } + } + ; + +maskedAddrReg: addrReg addrWriteMask + { + set_dst_reg(& $$, PROGRAM_ADDRESS, 0); + $$.WriteMask = $2.mask; + } + ; + +extendedSwizzle: extSwizComp ',' extSwizComp ',' extSwizComp ',' extSwizComp + { + const unsigned xyzw_valid = + ($1.xyzw_valid << 0) + | ($3.xyzw_valid << 1) + | ($5.xyzw_valid << 2) + | ($7.xyzw_valid << 3); + const unsigned rgba_valid = + ($1.rgba_valid << 0) + | ($3.rgba_valid << 1) + | ($5.rgba_valid << 2) + | ($7.rgba_valid << 3); + + /* All of the swizzle components have to be valid in either RGBA + * or XYZW. Note that 0 and 1 are valid in both, so both masks + * can have some bits set. + * + * We somewhat deviate from the spec here. It would be really hard + * to figure out which component is the error, and there probably + * isn't a lot of benefit. + */ + if ((rgba_valid != 0x0f) && (xyzw_valid != 0x0f)) { + yyerror(& @1, state, "cannot combine RGBA and XYZW swizzle " + "components"); + YYERROR; + } + + $$.swizzle = MAKE_SWIZZLE4($1.swz, $3.swz, $5.swz, $7.swz); + $$.mask = ($1.negate) | ($3.negate << 1) | ($5.negate << 2) + | ($7.negate << 3); + } + ; + +extSwizComp: optionalSign extSwizSel + { + $$ = $2; + $$.negate = ($1) ? 1 : 0; + } + ; + +extSwizSel: INTEGER + { + if (($1 != 0) && ($1 != 1)) { + yyerror(& @1, state, "invalid extended swizzle selector"); + YYERROR; + } + + $$.swz = ($1 == 0) ? SWIZZLE_ZERO : SWIZZLE_ONE; + + /* 0 and 1 are valid for both RGBA swizzle names and XYZW + * swizzle names. + */ + $$.xyzw_valid = 1; + $$.rgba_valid = 1; + } + | string + { + char s; + + if (strlen($1) > 1) { + yyerror(& @1, state, "invalid extended swizzle selector"); + YYERROR; + } + + s = $1[0]; + free($1); + + switch (s) { + case 'x': + $$.swz = SWIZZLE_X; + $$.xyzw_valid = 1; + break; + case 'y': + $$.swz = SWIZZLE_Y; + $$.xyzw_valid = 1; + break; + case 'z': + $$.swz = SWIZZLE_Z; + $$.xyzw_valid = 1; + break; + case 'w': + $$.swz = SWIZZLE_W; + $$.xyzw_valid = 1; + break; + + case 'r': + $$.swz = SWIZZLE_X; + $$.rgba_valid = 1; + break; + case 'g': + $$.swz = SWIZZLE_Y; + $$.rgba_valid = 1; + break; + case 'b': + $$.swz = SWIZZLE_Z; + $$.rgba_valid = 1; + break; + case 'a': + $$.swz = SWIZZLE_W; + $$.rgba_valid = 1; + break; + + default: + yyerror(& @1, state, "invalid extended swizzle selector"); + YYERROR; + break; + } + } + ; + +srcReg: USED_IDENTIFIER /* temporaryReg | progParamSingle */ + { + struct asm_symbol *const s = (struct asm_symbol *) + _mesa_symbol_table_find_symbol(state->st, 0, $1); + + free($1); + + if (s == NULL) { + yyerror(& @1, state, "invalid operand variable"); + YYERROR; + } else if ((s->type != at_param) && (s->type != at_temp) + && (s->type != at_attrib)) { + yyerror(& @1, state, "invalid operand variable"); + YYERROR; + } else if ((s->type == at_param) && s->param_is_array) { + yyerror(& @1, state, "non-array access to array PARAM"); + YYERROR; + } + + init_src_reg(& $$); + switch (s->type) { + case at_temp: + set_src_reg(& $$, PROGRAM_TEMPORARY, s->temp_binding); + break; + case at_param: + set_src_reg_swz(& $$, s->param_binding_type, + s->param_binding_begin, + s->param_binding_swizzle); + break; + case at_attrib: + set_src_reg(& $$, PROGRAM_INPUT, s->attrib_binding); + state->prog->InputsRead |= (1U << $$.Base.Index); + + if (!validate_inputs(& @1, state)) { + YYERROR; + } + break; + + default: + YYERROR; + break; + } + } + | attribBinding + { + set_src_reg(& $$, PROGRAM_INPUT, $1); + state->prog->InputsRead |= (1U << $$.Base.Index); + + if (!validate_inputs(& @1, state)) { + YYERROR; + } + } + | progParamArray '[' progParamArrayMem ']' + { + if (! $3.Base.RelAddr + && ((unsigned) $3.Base.Index >= $1->param_binding_length)) { + yyerror(& @3, state, "out of bounds array access"); + YYERROR; + } + + init_src_reg(& $$); + $$.Base.File = $1->param_binding_type; + + if ($3.Base.RelAddr) { + state->prog->IndirectRegisterFiles |= (1 << $$.Base.File); + $1->param_accessed_indirectly = 1; + + $$.Base.RelAddr = 1; + $$.Base.Index = $3.Base.Index; + $$.Symbol = $1; + } else { + $$.Base.Index = $1->param_binding_begin + $3.Base.Index; + } + } + | paramSingleItemUse + { + gl_register_file file = ($1.name != NULL) + ? $1.param_binding_type + : PROGRAM_CONSTANT; + set_src_reg_swz(& $$, file, $1.param_binding_begin, + $1.param_binding_swizzle); + } + ; + +dstReg: resultBinding + { + set_dst_reg(& $$, PROGRAM_OUTPUT, $1); + } + | USED_IDENTIFIER /* temporaryReg | vertexResultReg */ + { + struct asm_symbol *const s = (struct asm_symbol *) + _mesa_symbol_table_find_symbol(state->st, 0, $1); + + free($1); + + if (s == NULL) { + yyerror(& @1, state, "invalid operand variable"); + YYERROR; + } else if ((s->type != at_output) && (s->type != at_temp)) { + yyerror(& @1, state, "invalid operand variable"); + YYERROR; + } + + switch (s->type) { + case at_temp: + set_dst_reg(& $$, PROGRAM_TEMPORARY, s->temp_binding); + break; + case at_output: + set_dst_reg(& $$, PROGRAM_OUTPUT, s->output_binding); + break; + default: + set_dst_reg(& $$, s->param_binding_type, s->param_binding_begin); + break; + } + } + ; + +progParamArray: USED_IDENTIFIER + { + struct asm_symbol *const s = (struct asm_symbol *) + _mesa_symbol_table_find_symbol(state->st, 0, $1); + + free($1); + + if (s == NULL) { + yyerror(& @1, state, "invalid operand variable"); + YYERROR; + } else if ((s->type != at_param) || !s->param_is_array) { + yyerror(& @1, state, "array access to non-PARAM variable"); + YYERROR; + } else { + $$ = s; + } + } + ; + +progParamArrayMem: progParamArrayAbs | progParamArrayRel; + +progParamArrayAbs: INTEGER + { + init_src_reg(& $$); + $$.Base.Index = $1; + } + ; + +progParamArrayRel: addrReg addrComponent addrRegRelOffset + { + /* FINISHME: Add support for multiple address registers. + */ + /* FINISHME: Add support for 4-component address registers. + */ + init_src_reg(& $$); + $$.Base.RelAddr = 1; + $$.Base.Index = $3; + } + ; + +addrRegRelOffset: { $$ = 0; } + | '+' addrRegPosOffset { $$ = $2; } + | '-' addrRegNegOffset { $$ = -$2; } + ; + +addrRegPosOffset: INTEGER + { + if (($1 < 0) || ($1 > (state->limits->MaxAddressOffset - 1))) { + char s[100]; + _mesa_snprintf(s, sizeof(s), + "relative address offset too large (%d)", $1); + yyerror(& @1, state, s); + YYERROR; + } else { + $$ = $1; + } + } + ; + +addrRegNegOffset: INTEGER + { + if (($1 < 0) || ($1 > state->limits->MaxAddressOffset)) { + char s[100]; + _mesa_snprintf(s, sizeof(s), + "relative address offset too large (%d)", $1); + yyerror(& @1, state, s); + YYERROR; + } else { + $$ = $1; + } + } + ; + +addrReg: USED_IDENTIFIER + { + struct asm_symbol *const s = (struct asm_symbol *) + _mesa_symbol_table_find_symbol(state->st, 0, $1); + + free($1); + + if (s == NULL) { + yyerror(& @1, state, "invalid array member"); + YYERROR; + } else if (s->type != at_address) { + yyerror(& @1, state, + "invalid variable for indexed array access"); + YYERROR; + } else { + $$ = s; + } + } + ; + +addrComponent: MASK1 + { + if ($1.mask != WRITEMASK_X) { + yyerror(& @1, state, "invalid address component selector"); + YYERROR; + } else { + $$ = $1; + } + } + ; + +addrWriteMask: MASK1 + { + if ($1.mask != WRITEMASK_X) { + yyerror(& @1, state, + "address register write mask must be \".x\""); + YYERROR; + } else { + $$ = $1; + } + } + ; + +scalarSuffix: MASK1; + +swizzleSuffix: MASK1 + | MASK4 + | SWIZZLE + | { $$.swizzle = SWIZZLE_NOOP; $$.mask = WRITEMASK_XYZW; } + ; + +optionalMask: MASK4 | MASK3 | MASK2 | MASK1 + | { $$.swizzle = SWIZZLE_NOOP; $$.mask = WRITEMASK_XYZW; } + ; + +optionalCcMask: '(' ccTest ')' + { + $$ = $2; + } + | '(' ccTest2 ')' + { + $$ = $2; + } + | + { + $$.CondMask = COND_TR; + $$.CondSwizzle = SWIZZLE_NOOP; + $$.CondSrc = 0; + } + ; + +ccTest: ccMaskRule swizzleSuffix + { + $$ = $1; + $$.CondSwizzle = $2.swizzle; + } + ; + +ccTest2: ccMaskRule2 swizzleSuffix + { + $$ = $1; + $$.CondSwizzle = $2.swizzle; + } + ; + +ccMaskRule: IDENTIFIER + { + const int cond = _mesa_parse_cc($1); + if ((cond == 0) || ($1[2] != '\0')) { + char *const err_str = + make_error_string("invalid condition code \"%s\"", $1); + + yyerror(& @1, state, (err_str != NULL) + ? err_str : "invalid condition code"); + + if (err_str != NULL) { + free(err_str); + } + + YYERROR; + } + + $$.CondMask = cond; + $$.CondSwizzle = SWIZZLE_NOOP; + $$.CondSrc = 0; + } + ; + +ccMaskRule2: USED_IDENTIFIER + { + const int cond = _mesa_parse_cc($1); + if ((cond == 0) || ($1[2] != '\0')) { + char *const err_str = + make_error_string("invalid condition code \"%s\"", $1); + + yyerror(& @1, state, (err_str != NULL) + ? err_str : "invalid condition code"); + + if (err_str != NULL) { + free(err_str); + } + + YYERROR; + } + + $$.CondMask = cond; + $$.CondSwizzle = SWIZZLE_NOOP; + $$.CondSrc = 0; + } + ; + +namingStatement: ATTRIB_statement + | PARAM_statement + | TEMP_statement + | ADDRESS_statement + | OUTPUT_statement + | ALIAS_statement + ; + +ATTRIB_statement: ATTRIB IDENTIFIER '=' attribBinding + { + struct asm_symbol *const s = + declare_variable(state, $2, at_attrib, & @2); + + if (s == NULL) { + free($2); + YYERROR; + } else { + s->attrib_binding = $4; + state->InputsBound |= (1U << s->attrib_binding); + + if (!validate_inputs(& @4, state)) { + YYERROR; + } + } + } + ; + +attribBinding: VERTEX vtxAttribItem + { + $$ = $2; + } + | FRAGMENT fragAttribItem + { + $$ = $2; + } + ; + +vtxAttribItem: POSITION + { + $$ = VERT_ATTRIB_POS; + } + | WEIGHT vtxOptWeightNum + { + $$ = VERT_ATTRIB_WEIGHT; + } + | NORMAL + { + $$ = VERT_ATTRIB_NORMAL; + } + | COLOR optColorType + { + if (!state->ctx->Extensions.EXT_secondary_color) { + yyerror(& @2, state, "GL_EXT_secondary_color not supported"); + YYERROR; + } + + $$ = VERT_ATTRIB_COLOR0 + $2; + } + | FOGCOORD + { + if (!state->ctx->Extensions.EXT_fog_coord) { + yyerror(& @1, state, "GL_EXT_fog_coord not supported"); + YYERROR; + } + + $$ = VERT_ATTRIB_FOG; + } + | TEXCOORD optTexCoordUnitNum + { + $$ = VERT_ATTRIB_TEX0 + $2; + } + | MATRIXINDEX '[' vtxWeightNum ']' + { + yyerror(& @1, state, "GL_ARB_matrix_palette not supported"); + YYERROR; + } + | VTXATTRIB '[' vtxAttribNum ']' + { + $$ = VERT_ATTRIB_GENERIC0 + $3; + } + ; + +vtxAttribNum: INTEGER + { + if ((unsigned) $1 >= state->limits->MaxAttribs) { + yyerror(& @1, state, "invalid vertex attribute reference"); + YYERROR; + } + + $$ = $1; + } + ; + +vtxOptWeightNum: | '[' vtxWeightNum ']'; +vtxWeightNum: INTEGER; + +fragAttribItem: POSITION + { + $$ = FRAG_ATTRIB_WPOS; + } + | COLOR optColorType + { + $$ = FRAG_ATTRIB_COL0 + $2; + } + | FOGCOORD + { + $$ = FRAG_ATTRIB_FOGC; + } + | TEXCOORD optTexCoordUnitNum + { + $$ = FRAG_ATTRIB_TEX0 + $2; + } + ; + +PARAM_statement: PARAM_singleStmt | PARAM_multipleStmt; + +PARAM_singleStmt: PARAM IDENTIFIER paramSingleInit + { + struct asm_symbol *const s = + declare_variable(state, $2, at_param, & @2); + + if (s == NULL) { + free($2); + YYERROR; + } else { + s->param_binding_type = $3.param_binding_type; + s->param_binding_begin = $3.param_binding_begin; + s->param_binding_length = $3.param_binding_length; + s->param_binding_swizzle = $3.param_binding_swizzle; + s->param_is_array = 0; + } + } + ; + +PARAM_multipleStmt: PARAM IDENTIFIER '[' optArraySize ']' paramMultipleInit + { + if (($4 != 0) && ((unsigned) $4 != $6.param_binding_length)) { + free($2); + yyerror(& @4, state, + "parameter array size and number of bindings must match"); + YYERROR; + } else { + struct asm_symbol *const s = + declare_variable(state, $2, $6.type, & @2); + + if (s == NULL) { + free($2); + YYERROR; + } else { + s->param_binding_type = $6.param_binding_type; + s->param_binding_begin = $6.param_binding_begin; + s->param_binding_length = $6.param_binding_length; + s->param_binding_swizzle = SWIZZLE_XYZW; + s->param_is_array = 1; + } + } + } + ; + +optArraySize: + { + $$ = 0; + } + | INTEGER + { + if (($1 < 1) || ((unsigned) $1 > state->limits->MaxParameters)) { + char msg[100]; + _mesa_snprintf(msg, sizeof(msg), + "invalid parameter array size (size=%d max=%u)", + $1, state->limits->MaxParameters); + yyerror(& @1, state, msg); + YYERROR; + } else { + $$ = $1; + } + } + ; + +paramSingleInit: '=' paramSingleItemDecl + { + $$ = $2; + } + ; + +paramMultipleInit: '=' '{' paramMultInitList '}' + { + $$ = $3; + } + ; + +paramMultInitList: paramMultipleItem + | paramMultInitList ',' paramMultipleItem + { + $1.param_binding_length += $3.param_binding_length; + $$ = $1; + } + ; + +paramSingleItemDecl: stateSingleItem + { + memset(& $$, 0, sizeof($$)); + $$.param_binding_begin = ~0; + initialize_symbol_from_state(state->prog, & $$, $1); + } + | programSingleItem + { + memset(& $$, 0, sizeof($$)); + $$.param_binding_begin = ~0; + initialize_symbol_from_param(state->prog, & $$, $1); + } + | paramConstDecl + { + memset(& $$, 0, sizeof($$)); + $$.param_binding_begin = ~0; + initialize_symbol_from_const(state->prog, & $$, & $1, GL_TRUE); + } + ; + +paramSingleItemUse: stateSingleItem + { + memset(& $$, 0, sizeof($$)); + $$.param_binding_begin = ~0; + initialize_symbol_from_state(state->prog, & $$, $1); + } + | programSingleItem + { + memset(& $$, 0, sizeof($$)); + $$.param_binding_begin = ~0; + initialize_symbol_from_param(state->prog, & $$, $1); + } + | paramConstUse + { + memset(& $$, 0, sizeof($$)); + $$.param_binding_begin = ~0; + initialize_symbol_from_const(state->prog, & $$, & $1, GL_TRUE); + } + ; + +paramMultipleItem: stateMultipleItem + { + memset(& $$, 0, sizeof($$)); + $$.param_binding_begin = ~0; + initialize_symbol_from_state(state->prog, & $$, $1); + } + | programMultipleItem + { + memset(& $$, 0, sizeof($$)); + $$.param_binding_begin = ~0; + initialize_symbol_from_param(state->prog, & $$, $1); + } + | paramConstDecl + { + memset(& $$, 0, sizeof($$)); + $$.param_binding_begin = ~0; + initialize_symbol_from_const(state->prog, & $$, & $1, GL_FALSE); + } + ; + +stateMultipleItem: stateSingleItem { memcpy($$, $1, sizeof($$)); } + | STATE stateMatrixRows { memcpy($$, $2, sizeof($$)); } + ; + +stateSingleItem: STATE stateMaterialItem { memcpy($$, $2, sizeof($$)); } + | STATE stateLightItem { memcpy($$, $2, sizeof($$)); } + | STATE stateLightModelItem { memcpy($$, $2, sizeof($$)); } + | STATE stateLightProdItem { memcpy($$, $2, sizeof($$)); } + | STATE stateTexGenItem { memcpy($$, $2, sizeof($$)); } + | STATE stateTexEnvItem { memcpy($$, $2, sizeof($$)); } + | STATE stateFogItem { memcpy($$, $2, sizeof($$)); } + | STATE stateClipPlaneItem { memcpy($$, $2, sizeof($$)); } + | STATE statePointItem { memcpy($$, $2, sizeof($$)); } + | STATE stateMatrixRow { memcpy($$, $2, sizeof($$)); } + | STATE stateDepthItem { memcpy($$, $2, sizeof($$)); } + ; + +stateMaterialItem: MATERIAL optFaceType stateMatProperty + { + memset($$, 0, sizeof($$)); + $$[0] = STATE_MATERIAL; + $$[1] = $2; + $$[2] = $3; + } + ; + +stateMatProperty: ambDiffSpecProperty + { + $$ = $1; + } + | EMISSION + { + $$ = STATE_EMISSION; + } + | SHININESS + { + $$ = STATE_SHININESS; + } + ; + +stateLightItem: LIGHT '[' stateLightNumber ']' stateLightProperty + { + memset($$, 0, sizeof($$)); + $$[0] = STATE_LIGHT; + $$[1] = $3; + $$[2] = $5; + } + ; + +stateLightProperty: ambDiffSpecProperty + { + $$ = $1; + } + | POSITION + { + $$ = STATE_POSITION; + } + | ATTENUATION + { + if (!state->ctx->Extensions.EXT_point_parameters) { + yyerror(& @1, state, "GL_ARB_point_parameters not supported"); + YYERROR; + } + + $$ = STATE_ATTENUATION; + } + | SPOT stateSpotProperty + { + $$ = $2; + } + | HALF + { + $$ = STATE_HALF_VECTOR; + } + ; + +stateSpotProperty: DIRECTION + { + $$ = STATE_SPOT_DIRECTION; + } + ; + +stateLightModelItem: LIGHTMODEL stateLModProperty + { + $$[0] = $2[0]; + $$[1] = $2[1]; + } + ; + +stateLModProperty: AMBIENT + { + memset($$, 0, sizeof($$)); + $$[0] = STATE_LIGHTMODEL_AMBIENT; + } + | optFaceType SCENECOLOR + { + memset($$, 0, sizeof($$)); + $$[0] = STATE_LIGHTMODEL_SCENECOLOR; + $$[1] = $1; + } + ; + +stateLightProdItem: LIGHTPROD '[' stateLightNumber ']' optFaceType stateLProdProperty + { + memset($$, 0, sizeof($$)); + $$[0] = STATE_LIGHTPROD; + $$[1] = $3; + $$[2] = $5; + $$[3] = $6; + } + ; + +stateLProdProperty: ambDiffSpecProperty; + +stateTexEnvItem: TEXENV optLegacyTexUnitNum stateTexEnvProperty + { + memset($$, 0, sizeof($$)); + $$[0] = $3; + $$[1] = $2; + } + ; + +stateTexEnvProperty: COLOR + { + $$ = STATE_TEXENV_COLOR; + } + ; + +ambDiffSpecProperty: AMBIENT + { + $$ = STATE_AMBIENT; + } + | DIFFUSE + { + $$ = STATE_DIFFUSE; + } + | SPECULAR + { + $$ = STATE_SPECULAR; + } + ; + +stateLightNumber: INTEGER + { + if ((unsigned) $1 >= state->MaxLights) { + yyerror(& @1, state, "invalid light selector"); + YYERROR; + } + + $$ = $1; + } + ; + +stateTexGenItem: TEXGEN optTexCoordUnitNum stateTexGenType stateTexGenCoord + { + memset($$, 0, sizeof($$)); + $$[0] = STATE_TEXGEN; + $$[1] = $2; + $$[2] = $3 + $4; + } + ; + +stateTexGenType: EYE + { + $$ = STATE_TEXGEN_EYE_S; + } + | OBJECT + { + $$ = STATE_TEXGEN_OBJECT_S; + } + ; +stateTexGenCoord: TEXGEN_S + { + $$ = STATE_TEXGEN_EYE_S - STATE_TEXGEN_EYE_S; + } + | TEXGEN_T + { + $$ = STATE_TEXGEN_EYE_T - STATE_TEXGEN_EYE_S; + } + | TEXGEN_R + { + $$ = STATE_TEXGEN_EYE_R - STATE_TEXGEN_EYE_S; + } + | TEXGEN_Q + { + $$ = STATE_TEXGEN_EYE_Q - STATE_TEXGEN_EYE_S; + } + ; + +stateFogItem: FOG stateFogProperty + { + memset($$, 0, sizeof($$)); + $$[0] = $2; + } + ; + +stateFogProperty: COLOR + { + $$ = STATE_FOG_COLOR; + } + | PARAMS + { + $$ = STATE_FOG_PARAMS; + } + ; + +stateClipPlaneItem: CLIP '[' stateClipPlaneNum ']' PLANE + { + memset($$, 0, sizeof($$)); + $$[0] = STATE_CLIPPLANE; + $$[1] = $3; + } + ; + +stateClipPlaneNum: INTEGER + { + if ((unsigned) $1 >= state->MaxClipPlanes) { + yyerror(& @1, state, "invalid clip plane selector"); + YYERROR; + } + + $$ = $1; + } + ; + +statePointItem: POINT_TOK statePointProperty + { + memset($$, 0, sizeof($$)); + $$[0] = $2; + } + ; + +statePointProperty: SIZE_TOK + { + $$ = STATE_POINT_SIZE; + } + | ATTENUATION + { + $$ = STATE_POINT_ATTENUATION; + } + ; + +stateMatrixRow: stateMatrixItem ROW '[' stateMatrixRowNum ']' + { + $$[0] = $1[0]; + $$[1] = $1[1]; + $$[2] = $4; + $$[3] = $4; + $$[4] = $1[2]; + } + ; + +stateMatrixRows: stateMatrixItem optMatrixRows + { + $$[0] = $1[0]; + $$[1] = $1[1]; + $$[2] = $2[2]; + $$[3] = $2[3]; + $$[4] = $1[2]; + } + ; + +optMatrixRows: + { + $$[2] = 0; + $$[3] = 3; + } + | ROW '[' stateMatrixRowNum DOT_DOT stateMatrixRowNum ']' + { + /* It seems logical that the matrix row range specifier would have + * to specify a range or more than one row (i.e., $5 > $3). + * However, the ARB_vertex_program spec says "a program will fail + * to load if is greater than ." This means that $3 == $5 + * is valid. + */ + if ($3 > $5) { + yyerror(& @3, state, "invalid matrix row range"); + YYERROR; + } + + $$[2] = $3; + $$[3] = $5; + } + ; + +stateMatrixItem: MATRIX stateMatrixName stateOptMatModifier + { + $$[0] = $2[0]; + $$[1] = $2[1]; + $$[2] = $3; + } + ; + +stateOptMatModifier: + { + $$ = 0; + } + | stateMatModifier + { + $$ = $1; + } + ; + +stateMatModifier: INVERSE + { + $$ = STATE_MATRIX_INVERSE; + } + | TRANSPOSE + { + $$ = STATE_MATRIX_TRANSPOSE; + } + | INVTRANS + { + $$ = STATE_MATRIX_INVTRANS; + } + ; + +stateMatrixRowNum: INTEGER + { + if ($1 > 3) { + yyerror(& @1, state, "invalid matrix row reference"); + YYERROR; + } + + $$ = $1; + } + ; + +stateMatrixName: MODELVIEW stateOptModMatNum + { + $$[0] = STATE_MODELVIEW_MATRIX; + $$[1] = $2; + } + | PROJECTION + { + $$[0] = STATE_PROJECTION_MATRIX; + $$[1] = 0; + } + | MVP + { + $$[0] = STATE_MVP_MATRIX; + $$[1] = 0; + } + | TEXTURE optTexCoordUnitNum + { + $$[0] = STATE_TEXTURE_MATRIX; + $$[1] = $2; + } + | PALETTE '[' statePaletteMatNum ']' + { + yyerror(& @1, state, "GL_ARB_matrix_palette not supported"); + YYERROR; + } + | MAT_PROGRAM '[' stateProgramMatNum ']' + { + $$[0] = STATE_PROGRAM_MATRIX; + $$[1] = $3; + } + ; + +stateOptModMatNum: + { + $$ = 0; + } + | '[' stateModMatNum ']' + { + $$ = $2; + } + ; +stateModMatNum: INTEGER + { + /* Since GL_ARB_vertex_blend isn't supported, only modelview matrix + * zero is valid. + */ + if ($1 != 0) { + yyerror(& @1, state, "invalid modelview matrix index"); + YYERROR; + } + + $$ = $1; + } + ; +statePaletteMatNum: INTEGER + { + /* Since GL_ARB_matrix_palette isn't supported, just let any value + * through here. The error will be generated later. + */ + $$ = $1; + } + ; +stateProgramMatNum: INTEGER + { + if ((unsigned) $1 >= state->MaxProgramMatrices) { + yyerror(& @1, state, "invalid program matrix selector"); + YYERROR; + } + + $$ = $1; + } + ; + +stateDepthItem: DEPTH RANGE + { + memset($$, 0, sizeof($$)); + $$[0] = STATE_DEPTH_RANGE; + } + ; + + +programSingleItem: progEnvParam | progLocalParam; + +programMultipleItem: progEnvParams | progLocalParams; + +progEnvParams: PROGRAM ENV '[' progEnvParamNums ']' + { + memset($$, 0, sizeof($$)); + $$[0] = state->state_param_enum; + $$[1] = STATE_ENV; + $$[2] = $4[0]; + $$[3] = $4[1]; + } + ; + +progEnvParamNums: progEnvParamNum + { + $$[0] = $1; + $$[1] = $1; + } + | progEnvParamNum DOT_DOT progEnvParamNum + { + $$[0] = $1; + $$[1] = $3; + } + ; + +progEnvParam: PROGRAM ENV '[' progEnvParamNum ']' + { + memset($$, 0, sizeof($$)); + $$[0] = state->state_param_enum; + $$[1] = STATE_ENV; + $$[2] = $4; + $$[3] = $4; + } + ; + +progLocalParams: PROGRAM LOCAL '[' progLocalParamNums ']' + { + memset($$, 0, sizeof($$)); + $$[0] = state->state_param_enum; + $$[1] = STATE_LOCAL; + $$[2] = $4[0]; + $$[3] = $4[1]; + } + +progLocalParamNums: progLocalParamNum + { + $$[0] = $1; + $$[1] = $1; + } + | progLocalParamNum DOT_DOT progLocalParamNum + { + $$[0] = $1; + $$[1] = $3; + } + ; + +progLocalParam: PROGRAM LOCAL '[' progLocalParamNum ']' + { + memset($$, 0, sizeof($$)); + $$[0] = state->state_param_enum; + $$[1] = STATE_LOCAL; + $$[2] = $4; + $$[3] = $4; + } + ; + +progEnvParamNum: INTEGER + { + if ((unsigned) $1 >= state->limits->MaxEnvParams) { + yyerror(& @1, state, "invalid environment parameter reference"); + YYERROR; + } + $$ = $1; + } + ; + +progLocalParamNum: INTEGER + { + if ((unsigned) $1 >= state->limits->MaxLocalParams) { + yyerror(& @1, state, "invalid local parameter reference"); + YYERROR; + } + $$ = $1; + } + ; + + + +paramConstDecl: paramConstScalarDecl | paramConstVector; +paramConstUse: paramConstScalarUse | paramConstVector; + +paramConstScalarDecl: signedFloatConstant + { + $$.count = 4; + $$.data[0].f = $1; + $$.data[1].f = $1; + $$.data[2].f = $1; + $$.data[3].f = $1; + } + ; + +paramConstScalarUse: REAL + { + $$.count = 1; + $$.data[0].f = $1; + $$.data[1].f = $1; + $$.data[2].f = $1; + $$.data[3].f = $1; + } + | INTEGER + { + $$.count = 1; + $$.data[0].f = (float) $1; + $$.data[1].f = (float) $1; + $$.data[2].f = (float) $1; + $$.data[3].f = (float) $1; + } + ; + +paramConstVector: '{' signedFloatConstant '}' + { + $$.count = 4; + $$.data[0].f = $2; + $$.data[1].f = 0.0f; + $$.data[2].f = 0.0f; + $$.data[3].f = 1.0f; + } + | '{' signedFloatConstant ',' signedFloatConstant '}' + { + $$.count = 4; + $$.data[0].f = $2; + $$.data[1].f = $4; + $$.data[2].f = 0.0f; + $$.data[3].f = 1.0f; + } + | '{' signedFloatConstant ',' signedFloatConstant ',' + signedFloatConstant '}' + { + $$.count = 4; + $$.data[0].f = $2; + $$.data[1].f = $4; + $$.data[2].f = $6; + $$.data[3].f = 1.0f; + } + | '{' signedFloatConstant ',' signedFloatConstant ',' + signedFloatConstant ',' signedFloatConstant '}' + { + $$.count = 4; + $$.data[0].f = $2; + $$.data[1].f = $4; + $$.data[2].f = $6; + $$.data[3].f = $8; + } + ; + +signedFloatConstant: optionalSign REAL + { + $$ = ($1) ? -$2 : $2; + } + | optionalSign INTEGER + { + $$ = (float)(($1) ? -$2 : $2); + } + ; + +optionalSign: '+' { $$ = FALSE; } + | '-' { $$ = TRUE; } + | { $$ = FALSE; } + ; + +TEMP_statement: optVarSize TEMP { $$ = $2; } varNameList + ; + +optVarSize: string + { + /* NV_fragment_program_option defines the size qualifiers in a + * fairly broken way. "SHORT" or "LONG" can optionally be used + * before TEMP or OUTPUT. However, neither is a reserved word! + * This means that we have to parse it as an identifier, then check + * to make sure it's one of the valid values. *sigh* + * + * In addition, the grammar in the extension spec does *not* allow + * the size specifier to be optional, but all known implementations + * do. + */ + if (!state->option.NV_fragment) { + yyerror(& @1, state, "unexpected IDENTIFIER"); + YYERROR; + } + + if (strcmp("SHORT", $1) == 0) { + } else if (strcmp("LONG", $1) == 0) { + } else { + char *const err_str = + make_error_string("invalid storage size specifier \"%s\"", + $1); + + yyerror(& @1, state, (err_str != NULL) + ? err_str : "invalid storage size specifier"); + + if (err_str != NULL) { + free(err_str); + } + + YYERROR; + } + } + | + { + } + ; + +ADDRESS_statement: ADDRESS { $$ = $1; } varNameList + ; + +varNameList: varNameList ',' IDENTIFIER + { + if (!declare_variable(state, $3, $0, & @3)) { + free($3); + YYERROR; + } + } + | IDENTIFIER + { + if (!declare_variable(state, $1, $0, & @1)) { + free($1); + YYERROR; + } + } + ; + +OUTPUT_statement: optVarSize OUTPUT IDENTIFIER '=' resultBinding + { + struct asm_symbol *const s = + declare_variable(state, $3, at_output, & @3); + + if (s == NULL) { + free($3); + YYERROR; + } else { + s->output_binding = $5; + } + } + ; + +resultBinding: RESULT POSITION + { + if (state->mode == ARB_vertex) { + $$ = VERT_RESULT_HPOS; + } else { + yyerror(& @2, state, "invalid program result name"); + YYERROR; + } + } + | RESULT FOGCOORD + { + if (state->mode == ARB_vertex) { + $$ = VERT_RESULT_FOGC; + } else { + yyerror(& @2, state, "invalid program result name"); + YYERROR; + } + } + | RESULT resultColBinding + { + $$ = $2; + } + | RESULT POINTSIZE + { + if (state->mode == ARB_vertex) { + $$ = VERT_RESULT_PSIZ; + } else { + yyerror(& @2, state, "invalid program result name"); + YYERROR; + } + } + | RESULT TEXCOORD optTexCoordUnitNum + { + if (state->mode == ARB_vertex) { + $$ = VERT_RESULT_TEX0 + $3; + } else { + yyerror(& @2, state, "invalid program result name"); + YYERROR; + } + } + | RESULT DEPTH + { + if (state->mode == ARB_fragment) { + $$ = FRAG_RESULT_DEPTH; + } else { + yyerror(& @2, state, "invalid program result name"); + YYERROR; + } + } + ; + +resultColBinding: COLOR optResultFaceType optResultColorType + { + $$ = $2 + $3; + } + ; + +optResultFaceType: + { + if (state->mode == ARB_vertex) { + $$ = VERT_RESULT_COL0; + } else { + if (state->option.DrawBuffers) + $$ = FRAG_RESULT_DATA0; + else + $$ = FRAG_RESULT_COLOR; + } + } + | '[' INTEGER ']' + { + if (state->mode == ARB_vertex) { + yyerror(& @1, state, "invalid program result name"); + YYERROR; + } else { + if (!state->option.DrawBuffers) { + /* From the ARB_draw_buffers spec (same text exists + * for ATI_draw_buffers): + * + * If this option is not specified, a fragment + * program that attempts to bind + * "result.color[n]" will fail to load, and only + * "result.color" will be allowed. + */ + yyerror(& @1, state, + "result.color[] used without " + "`OPTION ARB_draw_buffers' or " + "`OPTION ATI_draw_buffers'"); + YYERROR; + } else if ($2 >= state->MaxDrawBuffers) { + yyerror(& @1, state, + "result.color[] exceeds MAX_DRAW_BUFFERS_ARB"); + YYERROR; + } + $$ = FRAG_RESULT_DATA0 + $2; + } + } + | FRONT + { + if (state->mode == ARB_vertex) { + $$ = VERT_RESULT_COL0; + } else { + yyerror(& @1, state, "invalid program result name"); + YYERROR; + } + } + | BACK + { + if (state->mode == ARB_vertex) { + $$ = VERT_RESULT_BFC0; + } else { + yyerror(& @1, state, "invalid program result name"); + YYERROR; + } + } + ; + +optResultColorType: + { + $$ = 0; + } + | PRIMARY + { + if (state->mode == ARB_vertex) { + $$ = 0; + } else { + yyerror(& @1, state, "invalid program result name"); + YYERROR; + } + } + | SECONDARY + { + if (state->mode == ARB_vertex) { + $$ = 1; + } else { + yyerror(& @1, state, "invalid program result name"); + YYERROR; + } + } + ; + +optFaceType: { $$ = 0; } + | FRONT { $$ = 0; } + | BACK { $$ = 1; } + ; + +optColorType: { $$ = 0; } + | PRIMARY { $$ = 0; } + | SECONDARY { $$ = 1; } + ; + +optTexCoordUnitNum: { $$ = 0; } + | '[' texCoordUnitNum ']' { $$ = $2; } + ; + +optTexImageUnitNum: { $$ = 0; } + | '[' texImageUnitNum ']' { $$ = $2; } + ; + +optLegacyTexUnitNum: { $$ = 0; } + | '[' legacyTexUnitNum ']' { $$ = $2; } + ; + +texCoordUnitNum: INTEGER + { + if ((unsigned) $1 >= state->MaxTextureCoordUnits) { + yyerror(& @1, state, "invalid texture coordinate unit selector"); + YYERROR; + } + + $$ = $1; + } + ; + +texImageUnitNum: INTEGER + { + if ((unsigned) $1 >= state->MaxTextureImageUnits) { + yyerror(& @1, state, "invalid texture image unit selector"); + YYERROR; + } + + $$ = $1; + } + ; + +legacyTexUnitNum: INTEGER + { + if ((unsigned) $1 >= state->MaxTextureUnits) { + yyerror(& @1, state, "invalid texture unit selector"); + YYERROR; + } + + $$ = $1; + } + ; + +ALIAS_statement: ALIAS IDENTIFIER '=' USED_IDENTIFIER + { + struct asm_symbol *exist = (struct asm_symbol *) + _mesa_symbol_table_find_symbol(state->st, 0, $2); + struct asm_symbol *target = (struct asm_symbol *) + _mesa_symbol_table_find_symbol(state->st, 0, $4); + + free($4); + + if (exist != NULL) { + char m[1000]; + _mesa_snprintf(m, sizeof(m), "redeclared identifier: %s", $2); + free($2); + yyerror(& @2, state, m); + YYERROR; + } else if (target == NULL) { + free($2); + yyerror(& @4, state, + "undefined variable binding in ALIAS statement"); + YYERROR; + } else { + _mesa_symbol_table_add_symbol(state->st, 0, $2, target); + } + } + ; + +string: IDENTIFIER + | USED_IDENTIFIER + ; + +%% + +void +asm_instruction_set_operands(struct asm_instruction *inst, + const struct prog_dst_register *dst, + const struct asm_src_register *src0, + const struct asm_src_register *src1, + const struct asm_src_register *src2) +{ + /* In the core ARB extensions only the KIL instruction doesn't have a + * destination register. + */ + if (dst == NULL) { + init_dst_reg(& inst->Base.DstReg); + } else { + inst->Base.DstReg = *dst; + } + + /* The only instruction that doesn't have any source registers is the + * condition-code based KIL instruction added by NV_fragment_program_option. + */ + if (src0 != NULL) { + inst->Base.SrcReg[0] = src0->Base; + inst->SrcReg[0] = *src0; + } else { + init_src_reg(& inst->SrcReg[0]); + } + + if (src1 != NULL) { + inst->Base.SrcReg[1] = src1->Base; + inst->SrcReg[1] = *src1; + } else { + init_src_reg(& inst->SrcReg[1]); + } + + if (src2 != NULL) { + inst->Base.SrcReg[2] = src2->Base; + inst->SrcReg[2] = *src2; + } else { + init_src_reg(& inst->SrcReg[2]); + } +} + + +struct asm_instruction * +asm_instruction_ctor(gl_inst_opcode op, + const struct prog_dst_register *dst, + const struct asm_src_register *src0, + const struct asm_src_register *src1, + const struct asm_src_register *src2) +{ + struct asm_instruction *inst = CALLOC_STRUCT(asm_instruction); + + if (inst) { + _mesa_init_instructions(& inst->Base, 1); + inst->Base.Opcode = op; + + asm_instruction_set_operands(inst, dst, src0, src1, src2); + } + + return inst; +} + + +struct asm_instruction * +asm_instruction_copy_ctor(const struct prog_instruction *base, + const struct prog_dst_register *dst, + const struct asm_src_register *src0, + const struct asm_src_register *src1, + const struct asm_src_register *src2) +{ + struct asm_instruction *inst = CALLOC_STRUCT(asm_instruction); + + if (inst) { + _mesa_init_instructions(& inst->Base, 1); + inst->Base.Opcode = base->Opcode; + inst->Base.CondUpdate = base->CondUpdate; + inst->Base.CondDst = base->CondDst; + inst->Base.SaturateMode = base->SaturateMode; + inst->Base.Precision = base->Precision; + + asm_instruction_set_operands(inst, dst, src0, src1, src2); + } + + return inst; +} + + +void +init_dst_reg(struct prog_dst_register *r) +{ + memset(r, 0, sizeof(*r)); + r->File = PROGRAM_UNDEFINED; + r->WriteMask = WRITEMASK_XYZW; + r->CondMask = COND_TR; + r->CondSwizzle = SWIZZLE_NOOP; +} + + +/** Like init_dst_reg() but set the File and Index fields. */ +void +set_dst_reg(struct prog_dst_register *r, gl_register_file file, GLint index) +{ + const GLint maxIndex = 1 << INST_INDEX_BITS; + const GLint minIndex = 0; + ASSERT(index >= minIndex); + (void) minIndex; + ASSERT(index <= maxIndex); + (void) maxIndex; + ASSERT(file == PROGRAM_TEMPORARY || + file == PROGRAM_ADDRESS || + file == PROGRAM_OUTPUT); + memset(r, 0, sizeof(*r)); + r->File = file; + r->Index = index; + r->WriteMask = WRITEMASK_XYZW; + r->CondMask = COND_TR; + r->CondSwizzle = SWIZZLE_NOOP; +} + + +void +init_src_reg(struct asm_src_register *r) +{ + memset(r, 0, sizeof(*r)); + r->Base.File = PROGRAM_UNDEFINED; + r->Base.Swizzle = SWIZZLE_NOOP; + r->Symbol = NULL; +} + + +/** Like init_src_reg() but set the File and Index fields. + * \return GL_TRUE if a valid src register, GL_FALSE otherwise + */ +void +set_src_reg(struct asm_src_register *r, gl_register_file file, GLint index) +{ + set_src_reg_swz(r, file, index, SWIZZLE_XYZW); +} + + +void +set_src_reg_swz(struct asm_src_register *r, gl_register_file file, GLint index, + GLuint swizzle) +{ + const GLint maxIndex = (1 << INST_INDEX_BITS) - 1; + const GLint minIndex = -(1 << INST_INDEX_BITS); + ASSERT(file < PROGRAM_FILE_MAX); + ASSERT(index >= minIndex); + (void) minIndex; + ASSERT(index <= maxIndex); + (void) maxIndex; + memset(r, 0, sizeof(*r)); + r->Base.File = file; + r->Base.Index = index; + r->Base.Swizzle = swizzle; + r->Symbol = NULL; +} + + +/** + * Validate the set of inputs used by a program + * + * Validates that legal sets of inputs are used by the program. In this case + * "used" included both reading the input or binding the input to a name using + * the \c ATTRIB command. + * + * \return + * \c TRUE if the combination of inputs used is valid, \c FALSE otherwise. + */ +int +validate_inputs(struct YYLTYPE *locp, struct asm_parser_state *state) +{ + const int inputs = state->prog->InputsRead | state->InputsBound; + + if (((inputs & 0x0ffff) & (inputs >> 16)) != 0) { + yyerror(locp, state, "illegal use of generic attribute and name attribute"); + return 0; + } + + return 1; +} + + +struct asm_symbol * +declare_variable(struct asm_parser_state *state, char *name, enum asm_type t, + struct YYLTYPE *locp) +{ + struct asm_symbol *s = NULL; + struct asm_symbol *exist = (struct asm_symbol *) + _mesa_symbol_table_find_symbol(state->st, 0, name); + + + if (exist != NULL) { + yyerror(locp, state, "redeclared identifier"); + } else { + s = calloc(1, sizeof(struct asm_symbol)); + s->name = name; + s->type = t; + + switch (t) { + case at_temp: + if (state->prog->NumTemporaries >= state->limits->MaxTemps) { + yyerror(locp, state, "too many temporaries declared"); + free(s); + return NULL; + } + + s->temp_binding = state->prog->NumTemporaries; + state->prog->NumTemporaries++; + break; + + case at_address: + if (state->prog->NumAddressRegs >= state->limits->MaxAddressRegs) { + yyerror(locp, state, "too many address registers declared"); + free(s); + return NULL; + } + + /* FINISHME: Add support for multiple address registers. + */ + state->prog->NumAddressRegs++; + break; + + default: + break; + } + + _mesa_symbol_table_add_symbol(state->st, 0, s->name, s); + s->next = state->sym; + state->sym = s; + } + + return s; +} + + +int add_state_reference(struct gl_program_parameter_list *param_list, + const gl_state_index tokens[STATE_LENGTH]) +{ + const GLuint size = 4; /* XXX fix */ + char *name; + GLint index; + + name = _mesa_program_state_string(tokens); + index = _mesa_add_parameter(param_list, PROGRAM_STATE_VAR, name, + size, GL_NONE, NULL, tokens, 0x0); + param_list->StateFlags |= _mesa_program_state_flags(tokens); + + /* free name string here since we duplicated it in add_parameter() */ + free(name); + + return index; +} + + +int +initialize_symbol_from_state(struct gl_program *prog, + struct asm_symbol *param_var, + const gl_state_index tokens[STATE_LENGTH]) +{ + int idx = -1; + gl_state_index state_tokens[STATE_LENGTH]; + + + memcpy(state_tokens, tokens, sizeof(state_tokens)); + + param_var->type = at_param; + param_var->param_binding_type = PROGRAM_STATE_VAR; + + /* If we are adding a STATE_MATRIX that has multiple rows, we need to + * unroll it and call add_state_reference() for each row + */ + if ((state_tokens[0] == STATE_MODELVIEW_MATRIX || + state_tokens[0] == STATE_PROJECTION_MATRIX || + state_tokens[0] == STATE_MVP_MATRIX || + state_tokens[0] == STATE_TEXTURE_MATRIX || + state_tokens[0] == STATE_PROGRAM_MATRIX) + && (state_tokens[2] != state_tokens[3])) { + int row; + const int first_row = state_tokens[2]; + const int last_row = state_tokens[3]; + + for (row = first_row; row <= last_row; row++) { + state_tokens[2] = state_tokens[3] = row; + + idx = add_state_reference(prog->Parameters, state_tokens); + if (param_var->param_binding_begin == ~0U) { + param_var->param_binding_begin = idx; + param_var->param_binding_swizzle = SWIZZLE_XYZW; + } + + param_var->param_binding_length++; + } + } + else { + idx = add_state_reference(prog->Parameters, state_tokens); + if (param_var->param_binding_begin == ~0U) { + param_var->param_binding_begin = idx; + param_var->param_binding_swizzle = SWIZZLE_XYZW; + } + param_var->param_binding_length++; + } + + return idx; +} + + +int +initialize_symbol_from_param(struct gl_program *prog, + struct asm_symbol *param_var, + const gl_state_index tokens[STATE_LENGTH]) +{ + int idx = -1; + gl_state_index state_tokens[STATE_LENGTH]; + + + memcpy(state_tokens, tokens, sizeof(state_tokens)); + + assert((state_tokens[0] == STATE_VERTEX_PROGRAM) + || (state_tokens[0] == STATE_FRAGMENT_PROGRAM)); + assert((state_tokens[1] == STATE_ENV) + || (state_tokens[1] == STATE_LOCAL)); + + /* + * The param type is STATE_VAR. The program parameter entry will + * effectively be a pointer into the LOCAL or ENV parameter array. + */ + param_var->type = at_param; + param_var->param_binding_type = PROGRAM_STATE_VAR; + + /* If we are adding a STATE_ENV or STATE_LOCAL that has multiple elements, + * we need to unroll it and call add_state_reference() for each row + */ + if (state_tokens[2] != state_tokens[3]) { + int row; + const int first_row = state_tokens[2]; + const int last_row = state_tokens[3]; + + for (row = first_row; row <= last_row; row++) { + state_tokens[2] = state_tokens[3] = row; + + idx = add_state_reference(prog->Parameters, state_tokens); + if (param_var->param_binding_begin == ~0U) { + param_var->param_binding_begin = idx; + param_var->param_binding_swizzle = SWIZZLE_XYZW; + } + param_var->param_binding_length++; + } + } + else { + idx = add_state_reference(prog->Parameters, state_tokens); + if (param_var->param_binding_begin == ~0U) { + param_var->param_binding_begin = idx; + param_var->param_binding_swizzle = SWIZZLE_XYZW; + } + param_var->param_binding_length++; + } + + return idx; +} + + +/** + * Put a float/vector constant/literal into the parameter list. + * \param param_var returns info about the parameter/constant's location, + * binding, type, etc. + * \param vec the vector/constant to add + * \param allowSwizzle if true, try to consolidate constants which only differ + * by a swizzle. We don't want to do this when building + * arrays of constants that may be indexed indirectly. + * \return index of the constant in the parameter list. + */ +int +initialize_symbol_from_const(struct gl_program *prog, + struct asm_symbol *param_var, + const struct asm_vector *vec, + GLboolean allowSwizzle) +{ + unsigned swizzle; + const int idx = _mesa_add_unnamed_constant(prog->Parameters, + vec->data, vec->count, + allowSwizzle ? &swizzle : NULL); + + param_var->type = at_param; + param_var->param_binding_type = PROGRAM_CONSTANT; + + if (param_var->param_binding_begin == ~0U) { + param_var->param_binding_begin = idx; + param_var->param_binding_swizzle = allowSwizzle ? swizzle : SWIZZLE_XYZW; + } + param_var->param_binding_length++; + + return idx; +} + + +char * +make_error_string(const char *fmt, ...) +{ + int length; + char *str; + va_list args; + + + /* Call vsnprintf once to determine how large the final string is. Call it + * again to do the actual formatting. from the vsnprintf manual page: + * + * Upon successful return, these functions return the number of + * characters printed (not including the trailing '\0' used to end + * output to strings). + */ + va_start(args, fmt); + length = 1 + vsnprintf(NULL, 0, fmt, args); + va_end(args); + + str = malloc(length); + if (str) { + va_start(args, fmt); + vsnprintf(str, length, fmt, args); + va_end(args); + } + + return str; +} + + +void +yyerror(YYLTYPE *locp, struct asm_parser_state *state, const char *s) +{ + char *err_str; + + + err_str = make_error_string("glProgramStringARB(%s)\n", s); + if (err_str) { + _mesa_error(state->ctx, GL_INVALID_OPERATION, "%s", err_str); + free(err_str); + } + + err_str = make_error_string("line %u, char %u: error: %s\n", + locp->first_line, locp->first_column, s); + _mesa_set_program_error(state->ctx, locp->position, err_str); + + if (err_str) { + free(err_str); + } +} + + +GLboolean +_mesa_parse_arb_program(struct gl_context *ctx, GLenum target, const GLubyte *str, + GLsizei len, struct asm_parser_state *state) +{ + struct asm_instruction *inst; + unsigned i; + GLubyte *strz; + GLboolean result = GL_FALSE; + void *temp; + struct asm_symbol *sym; + + state->ctx = ctx; + state->prog->Target = target; + state->prog->Parameters = _mesa_new_parameter_list(); + + /* Make a copy of the program string and force it to be NUL-terminated. + */ + strz = (GLubyte *) malloc(len + 1); + if (strz == NULL) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glProgramStringARB"); + return GL_FALSE; + } + memcpy (strz, str, len); + strz[len] = '\0'; + + state->prog->String = strz; + + state->st = _mesa_symbol_table_ctor(); + + state->limits = (target == GL_VERTEX_PROGRAM_ARB) + ? & ctx->Const.VertexProgram + : & ctx->Const.FragmentProgram; + + state->MaxTextureImageUnits = ctx->Const.MaxTextureImageUnits; + state->MaxTextureCoordUnits = ctx->Const.MaxTextureCoordUnits; + state->MaxTextureUnits = ctx->Const.MaxTextureUnits; + state->MaxClipPlanes = ctx->Const.MaxClipPlanes; + state->MaxLights = ctx->Const.MaxLights; + state->MaxProgramMatrices = ctx->Const.MaxProgramMatrices; + state->MaxDrawBuffers = ctx->Const.MaxDrawBuffers; + + state->state_param_enum = (target == GL_VERTEX_PROGRAM_ARB) + ? STATE_VERTEX_PROGRAM : STATE_FRAGMENT_PROGRAM; + + _mesa_set_program_error(ctx, -1, NULL); + + _mesa_program_lexer_ctor(& state->scanner, state, (const char *) str, len); + yyparse(state); + _mesa_program_lexer_dtor(state->scanner); + + + if (ctx->Program.ErrorPos != -1) { + goto error; + } + + if (! _mesa_layout_parameters(state)) { + struct YYLTYPE loc; + + loc.first_line = 0; + loc.first_column = 0; + loc.position = len; + + yyerror(& loc, state, "invalid PARAM usage"); + goto error; + } + + + + /* Add one instruction to store the "END" instruction. + */ + state->prog->Instructions = + _mesa_alloc_instructions(state->prog->NumInstructions + 1); + inst = state->inst_head; + for (i = 0; i < state->prog->NumInstructions; i++) { + struct asm_instruction *const temp = inst->next; + + state->prog->Instructions[i] = inst->Base; + inst = temp; + } + + /* Finally, tag on an OPCODE_END instruction */ + { + const GLuint numInst = state->prog->NumInstructions; + _mesa_init_instructions(state->prog->Instructions + numInst, 1); + state->prog->Instructions[numInst].Opcode = OPCODE_END; + } + state->prog->NumInstructions++; + + state->prog->NumParameters = state->prog->Parameters->NumParameters; + state->prog->NumAttributes = _mesa_bitcount(state->prog->InputsRead); + + /* + * Initialize native counts to logical counts. The device driver may + * change them if program is translated into a hardware program. + */ + state->prog->NumNativeInstructions = state->prog->NumInstructions; + state->prog->NumNativeTemporaries = state->prog->NumTemporaries; + state->prog->NumNativeParameters = state->prog->NumParameters; + state->prog->NumNativeAttributes = state->prog->NumAttributes; + state->prog->NumNativeAddressRegs = state->prog->NumAddressRegs; + + result = GL_TRUE; + +error: + for (inst = state->inst_head; inst != NULL; inst = temp) { + temp = inst->next; + free(inst); + } + + state->inst_head = NULL; + state->inst_tail = NULL; + + for (sym = state->sym; sym != NULL; sym = temp) { + temp = sym->next; + + free((void *) sym->name); + free(sym); + } + state->sym = NULL; + + _mesa_symbol_table_dtor(state->st); + state->st = NULL; + + return result; +} diff --git a/mesalib/src/mesa/program/program_parse_extra.c b/mesalib/src/mesa/program/program_parse_extra.c index 67ab9b99d..4d928483e 100644 --- a/mesalib/src/mesa/program/program_parse_extra.c +++ b/mesalib/src/mesa/program/program_parse_extra.c @@ -1,265 +1,265 @@ -/* - * Copyright © 2009 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER - * DEALINGS IN THE SOFTWARE. - */ - -#include -#include "main/mtypes.h" -#include "prog_instruction.h" -#include "program_parser.h" - - -/** - * Extra assembly-level parser routines - * - * \author Ian Romanick - */ - -int -_mesa_parse_instruction_suffix(const struct asm_parser_state *state, - const char *suffix, - struct prog_instruction *inst) -{ - inst->CondUpdate = 0; - inst->CondDst = 0; - inst->SaturateMode = SATURATE_OFF; - inst->Precision = FLOAT32; - - - /* The first possible suffix element is the precision specifier from - * NV_fragment_program_option. - */ - if (state->option.NV_fragment) { - switch (suffix[0]) { - case 'H': - inst->Precision = FLOAT16; - suffix++; - break; - case 'R': - inst->Precision = FLOAT32; - suffix++; - break; - case 'X': - inst->Precision = FIXED12; - suffix++; - break; - default: - break; - } - } - - /* The next possible suffix element is the condition code modifier selection - * from NV_fragment_program_option. - */ - if (state->option.NV_fragment) { - if (suffix[0] == 'C') { - inst->CondUpdate = 1; - suffix++; - } - } - - - /* The final possible suffix element is the saturation selector from - * ARB_fragment_program. - */ - if (state->mode == ARB_fragment) { - if (strcmp(suffix, "_SAT") == 0) { - inst->SaturateMode = SATURATE_ZERO_ONE; - suffix += 4; - } - } - - - /* It is an error for all of the suffix string not to be consumed. - */ - return suffix[0] == '\0'; -} - - -int -_mesa_parse_cc(const char *s) -{ - int cond = 0; - - switch (s[0]) { - case 'E': - if (s[1] == 'Q') { - cond = COND_EQ; - } - break; - - case 'F': - if (s[1] == 'L') { - cond = COND_FL; - } - break; - - case 'G': - if (s[1] == 'E') { - cond = COND_GE; - } else if (s[1] == 'T') { - cond = COND_GT; - } - break; - - case 'L': - if (s[1] == 'E') { - cond = COND_LE; - } else if (s[1] == 'T') { - cond = COND_LT; - } - break; - - case 'N': - if (s[1] == 'E') { - cond = COND_NE; - } - break; - - case 'T': - if (s[1] == 'R') { - cond = COND_TR; - } - break; - - default: - break; - } - - return ((cond == 0) || (s[2] != '\0')) ? 0 : cond; -} - - -int -_mesa_ARBvp_parse_option(struct asm_parser_state *state, const char *option) -{ - if (strcmp(option, "ARB_position_invariant") == 0) { - state->option.PositionInvariant = 1; - return 1; - } - - return 0; -} - - -int -_mesa_ARBfp_parse_option(struct asm_parser_state *state, const char *option) -{ - /* All of the options currently supported start with "ARB_". The code is - * currently structured with nested if-statements because eventually options - * that start with "NV_" will be supported. This structure will result in - * less churn when those options are added. - */ - if (strncmp(option, "ARB_", 4) == 0) { - /* Advance the pointer past the "ARB_" prefix. - */ - option += 4; - - - if (strncmp(option, "fog_", 4) == 0) { - option += 4; - - if (state->option.Fog == OPTION_NONE) { - if (strcmp(option, "exp") == 0) { - state->option.Fog = OPTION_FOG_EXP; - return 1; - } else if (strcmp(option, "exp2") == 0) { - state->option.Fog = OPTION_FOG_EXP2; - return 1; - } else if (strcmp(option, "linear") == 0) { - state->option.Fog = OPTION_FOG_LINEAR; - return 1; - } - } - - return 0; - } else if (strncmp(option, "precision_hint_", 15) == 0) { - option += 15; - - if (state->option.PrecisionHint == OPTION_NONE) { - if (strcmp(option, "nicest") == 0) { - state->option.PrecisionHint = OPTION_NICEST; - return 1; - } else if (strcmp(option, "fastest") == 0) { - state->option.PrecisionHint = OPTION_FASTEST; - return 1; - } - } - - return 0; - } else if (strcmp(option, "draw_buffers") == 0) { - /* Don't need to check extension availability because all Mesa-based - * drivers support GL_ARB_draw_buffers. - */ - state->option.DrawBuffers = 1; - return 1; - } else if (strcmp(option, "fragment_program_shadow") == 0) { - if (state->ctx->Extensions.ARB_fragment_program_shadow) { - state->option.Shadow = 1; - return 1; - } - } else if (strncmp(option, "fragment_coord_", 15) == 0) { - option += 15; - if (state->ctx->Extensions.ARB_fragment_coord_conventions) { - if (strcmp(option, "origin_upper_left") == 0) { - state->option.OriginUpperLeft = 1; - return 1; - } - else if (strcmp(option, "pixel_center_integer") == 0) { - state->option.PixelCenterInteger = 1; - return 1; - } - } - } - } else if (strncmp(option, "ATI_", 4) == 0) { - option += 4; - - if (strcmp(option, "draw_buffers") == 0) { - /* Don't need to check extension availability because all Mesa-based - * drivers support GL_ATI_draw_buffers. - */ - state->option.DrawBuffers = 1; - return 1; - } - } else if (strncmp(option, "NV_fragment_program", 19) == 0) { - option += 19; - - /* Other NV_fragment_program strings may be supported later. - */ - if (option[0] == '\0') { - if (state->ctx->Extensions.NV_fragment_program_option) { - state->option.NV_fragment = 1; - return 1; - } - } - } else if (strncmp(option, "MESA_", 5) == 0) { - option += 5; - - if (strcmp(option, "texture_array") == 0) { - if (state->ctx->Extensions.MESA_texture_array) { - state->option.TexArray = 1; - return 1; - } - } - } - - return 0; -} +/* + * Copyright © 2009 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +#include +#include "main/mtypes.h" +#include "prog_instruction.h" +#include "program_parser.h" + + +/** + * Extra assembly-level parser routines + * + * \author Ian Romanick + */ + +int +_mesa_parse_instruction_suffix(const struct asm_parser_state *state, + const char *suffix, + struct prog_instruction *inst) +{ + inst->CondUpdate = 0; + inst->CondDst = 0; + inst->SaturateMode = SATURATE_OFF; + inst->Precision = FLOAT32; + + + /* The first possible suffix element is the precision specifier from + * NV_fragment_program_option. + */ + if (state->option.NV_fragment) { + switch (suffix[0]) { + case 'H': + inst->Precision = FLOAT16; + suffix++; + break; + case 'R': + inst->Precision = FLOAT32; + suffix++; + break; + case 'X': + inst->Precision = FIXED12; + suffix++; + break; + default: + break; + } + } + + /* The next possible suffix element is the condition code modifier selection + * from NV_fragment_program_option. + */ + if (state->option.NV_fragment) { + if (suffix[0] == 'C') { + inst->CondUpdate = 1; + suffix++; + } + } + + + /* The final possible suffix element is the saturation selector from + * ARB_fragment_program. + */ + if (state->mode == ARB_fragment) { + if (strcmp(suffix, "_SAT") == 0) { + inst->SaturateMode = SATURATE_ZERO_ONE; + suffix += 4; + } + } + + + /* It is an error for all of the suffix string not to be consumed. + */ + return suffix[0] == '\0'; +} + + +int +_mesa_parse_cc(const char *s) +{ + int cond = 0; + + switch (s[0]) { + case 'E': + if (s[1] == 'Q') { + cond = COND_EQ; + } + break; + + case 'F': + if (s[1] == 'L') { + cond = COND_FL; + } + break; + + case 'G': + if (s[1] == 'E') { + cond = COND_GE; + } else if (s[1] == 'T') { + cond = COND_GT; + } + break; + + case 'L': + if (s[1] == 'E') { + cond = COND_LE; + } else if (s[1] == 'T') { + cond = COND_LT; + } + break; + + case 'N': + if (s[1] == 'E') { + cond = COND_NE; + } + break; + + case 'T': + if (s[1] == 'R') { + cond = COND_TR; + } + break; + + default: + break; + } + + return ((cond == 0) || (s[2] != '\0')) ? 0 : cond; +} + + +int +_mesa_ARBvp_parse_option(struct asm_parser_state *state, const char *option) +{ + if (strcmp(option, "ARB_position_invariant") == 0) { + state->option.PositionInvariant = 1; + return 1; + } + + return 0; +} + + +int +_mesa_ARBfp_parse_option(struct asm_parser_state *state, const char *option) +{ + /* All of the options currently supported start with "ARB_". The code is + * currently structured with nested if-statements because eventually options + * that start with "NV_" will be supported. This structure will result in + * less churn when those options are added. + */ + if (strncmp(option, "ARB_", 4) == 0) { + /* Advance the pointer past the "ARB_" prefix. + */ + option += 4; + + + if (strncmp(option, "fog_", 4) == 0) { + option += 4; + + if (state->option.Fog == OPTION_NONE) { + if (strcmp(option, "exp") == 0) { + state->option.Fog = OPTION_FOG_EXP; + return 1; + } else if (strcmp(option, "exp2") == 0) { + state->option.Fog = OPTION_FOG_EXP2; + return 1; + } else if (strcmp(option, "linear") == 0) { + state->option.Fog = OPTION_FOG_LINEAR; + return 1; + } + } + + return 0; + } else if (strncmp(option, "precision_hint_", 15) == 0) { + option += 15; + + if (state->option.PrecisionHint == OPTION_NONE) { + if (strcmp(option, "nicest") == 0) { + state->option.PrecisionHint = OPTION_NICEST; + return 1; + } else if (strcmp(option, "fastest") == 0) { + state->option.PrecisionHint = OPTION_FASTEST; + return 1; + } + } + + return 0; + } else if (strcmp(option, "draw_buffers") == 0) { + /* Don't need to check extension availability because all Mesa-based + * drivers support GL_ARB_draw_buffers. + */ + state->option.DrawBuffers = 1; + return 1; + } else if (strcmp(option, "fragment_program_shadow") == 0) { + if (state->ctx->Extensions.ARB_fragment_program_shadow) { + state->option.Shadow = 1; + return 1; + } + } else if (strncmp(option, "fragment_coord_", 15) == 0) { + option += 15; + if (state->ctx->Extensions.ARB_fragment_coord_conventions) { + if (strcmp(option, "origin_upper_left") == 0) { + state->option.OriginUpperLeft = 1; + return 1; + } + else if (strcmp(option, "pixel_center_integer") == 0) { + state->option.PixelCenterInteger = 1; + return 1; + } + } + } + } else if (strncmp(option, "ATI_", 4) == 0) { + option += 4; + + if (strcmp(option, "draw_buffers") == 0) { + /* Don't need to check extension availability because all Mesa-based + * drivers support GL_ATI_draw_buffers. + */ + state->option.DrawBuffers = 1; + return 1; + } + } else if (strncmp(option, "NV_fragment_program", 19) == 0) { + option += 19; + + /* Other NV_fragment_program strings may be supported later. + */ + if (option[0] == '\0') { + if (state->ctx->Extensions.NV_fragment_program_option) { + state->option.NV_fragment = 1; + return 1; + } + } + } else if (strncmp(option, "MESA_", 5) == 0) { + option += 5; + + if (strcmp(option, "texture_array") == 0) { + if (state->ctx->Extensions.MESA_texture_array) { + state->option.TexArray = 1; + return 1; + } + } + } + + return 0; +} diff --git a/mesalib/src/mesa/program/program_parser.h b/mesalib/src/mesa/program/program_parser.h index b6853a229..5637598f3 100644 --- a/mesalib/src/mesa/program/program_parser.h +++ b/mesalib/src/mesa/program/program_parser.h @@ -1,301 +1,301 @@ -/* - * Copyright © 2009 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER - * DEALINGS IN THE SOFTWARE. - */ -#pragma once - -#include "main/config.h" -#include "program/prog_parameter.h" - -struct gl_context; - -enum asm_type { - at_none, - at_address, - at_attrib, - at_param, - at_temp, - at_output -}; - -struct asm_symbol { - struct asm_symbol *next; /**< List linkage for freeing. */ - const char *name; - enum asm_type type; - unsigned attrib_binding; - unsigned output_binding; /**< Output / result register number. */ - - /** - * One of PROGRAM_STATE_VAR, PROGRAM_LOCAL_PARAM, or PROGRAM_ENV_PARAM. - */ - unsigned param_binding_type; - - /** - * Offset into the program_parameter_list where the tokens representing our - * bound state (or constants) start. - */ - unsigned param_binding_begin; - - /** - * Constants put into the parameter list may be swizzled. This - * field contain's the symbol's swizzle. (SWIZZLE_X/Y/Z/W) - */ - unsigned param_binding_swizzle; - - /* This is how many entries in the program_parameter_list we take up - * with our state tokens or constants. Note that this is _not_ the same as - * the number of param registers we eventually use. - */ - unsigned param_binding_length; - - /** - * Index of the temp register assigned to this variable. - */ - unsigned temp_binding; - - /** - * Flag whether or not a PARAM is an array - */ - unsigned param_is_array:1; - - - /** - * Flag whether or not a PARAM array is accessed indirectly - */ - unsigned param_accessed_indirectly:1; - - - /** - * \brief Is first pass of parameter layout done with this variable? - * - * The parameter layout routine operates in two passes. This flag tracks - * whether or not the first pass has handled this variable. - * - * \sa _mesa_layout_parameters - */ - unsigned pass1_done:1; -}; - - -struct asm_vector { - unsigned count; - gl_constant_value data[4]; -}; - - -struct asm_swizzle_mask { - unsigned swizzle:12; - unsigned mask:4; -}; - - -struct asm_src_register { - struct prog_src_register Base; - - /** - * Symbol associated with indirect access to parameter arrays. - * - * If \c Base::RelAddr is 1, this will point to the symbol for the parameter - * that is being dereferenced. Further, \c Base::Index will be the offset - * from the address register being used. - */ - struct asm_symbol *Symbol; -}; - - -struct asm_instruction { - struct prog_instruction Base; - struct asm_instruction *next; - struct asm_src_register SrcReg[3]; -}; - - -struct asm_parser_state { - struct gl_context *ctx; - struct gl_program *prog; - - /** - * Per-program target limits - */ - struct gl_program_constants *limits; - - struct _mesa_symbol_table *st; - - /** - * Linked list of symbols - * - * This list is \b only used when cleaning up compiler state and freeing - * memory. - */ - struct asm_symbol *sym; - - /** - * State for the lexer. - */ - void *scanner; - - /** - * Linked list of instructions generated during parsing. - */ - /*@{*/ - struct asm_instruction *inst_head; - struct asm_instruction *inst_tail; - /*@}*/ - - - /** - * Selected limits copied from gl_constants - * - * These are limits from the GL context, but various bits in the program - * must be validated against these values. - */ - /*@{*/ - unsigned MaxTextureCoordUnits; - unsigned MaxTextureImageUnits; - unsigned MaxTextureUnits; - unsigned MaxClipPlanes; - unsigned MaxLights; - unsigned MaxProgramMatrices; - unsigned MaxDrawBuffers; - /*@}*/ - - /** - * Value to use in state vector accessors for environment and local - * parameters - */ - unsigned state_param_enum; - - - /** - * Input attributes bound to specific names - * - * This is only needed so that errors can be properly produced when - * multiple ATTRIB statements bind illegal combinations of vertex - * attributes. - */ - unsigned InputsBound; - - enum { - invalid_mode = 0, - ARB_vertex, - ARB_fragment - } mode; - - struct { - unsigned PositionInvariant:1; - unsigned Fog:2; - unsigned PrecisionHint:2; - unsigned DrawBuffers:1; - unsigned Shadow:1; - unsigned TexRect:1; - unsigned TexArray:1; - unsigned NV_fragment:1; - unsigned OriginUpperLeft:1; - unsigned PixelCenterInteger:1; - } option; - - struct { - unsigned UsesKill:1; - } fragment; -}; - -#define OPTION_NONE 0 -#define OPTION_FOG_EXP 1 -#define OPTION_FOG_EXP2 2 -#define OPTION_FOG_LINEAR 3 -#define OPTION_NICEST 1 -#define OPTION_FASTEST 2 - -typedef struct YYLTYPE { - int first_line; - int first_column; - int last_line; - int last_column; - int position; -} YYLTYPE; - -#define YYLTYPE_IS_DECLARED 1 -#define YYLTYPE_IS_TRIVIAL 1 - - -extern GLboolean _mesa_parse_arb_program(struct gl_context *ctx, GLenum target, - const GLubyte *str, GLsizei len, struct asm_parser_state *state); - - - -/* From program_lexer.l. */ -extern void _mesa_program_lexer_dtor(void *scanner); - -extern void _mesa_program_lexer_ctor(void **scanner, - struct asm_parser_state *state, const char *string, size_t len); - - -/** - *\name From program_parse_extra.c - */ -/*@{*/ - -/** - * Parses and processes an option string to an ARB vertex program - * - * \return - * Non-zero on success, zero on failure. - */ -extern int _mesa_ARBvp_parse_option(struct asm_parser_state *state, - const char *option); - -/** - * Parses and processes an option string to an ARB fragment program - * - * \return - * Non-zero on success, zero on failure. - */ -extern int _mesa_ARBfp_parse_option(struct asm_parser_state *state, - const char *option); - -/** - * Parses and processes instruction suffixes - * - * Instruction suffixes, such as \c _SAT, are processed. The relevant bits - * are set in \c inst. If suffixes are encountered that are either not known - * or not supported by the modes and options set in \c state, zero will be - * returned. - * - * \return - * Non-zero on success, zero on failure. - */ -extern int _mesa_parse_instruction_suffix(const struct asm_parser_state *state, - const char *suffix, struct prog_instruction *inst); - -/** - * Parses a condition code name - * - * The condition code names (e.g., \c LT, \c GT, \c NE) were added to assembly - * shaders with the \c GL_NV_fragment_program_option extension. This function - * converts a string representation into one of the \c COND_ macros. - * - * \return - * One of the \c COND_ macros defined in prog_instruction.h on success or zero - * on failure. - */ -extern int _mesa_parse_cc(const char *s); - -/*@}*/ +/* + * Copyright © 2009 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ +#pragma once + +#include "main/config.h" +#include "program/prog_parameter.h" + +struct gl_context; + +enum asm_type { + at_none, + at_address, + at_attrib, + at_param, + at_temp, + at_output +}; + +struct asm_symbol { + struct asm_symbol *next; /**< List linkage for freeing. */ + const char *name; + enum asm_type type; + unsigned attrib_binding; + unsigned output_binding; /**< Output / result register number. */ + + /** + * One of PROGRAM_STATE_VAR, PROGRAM_LOCAL_PARAM, or PROGRAM_ENV_PARAM. + */ + unsigned param_binding_type; + + /** + * Offset into the program_parameter_list where the tokens representing our + * bound state (or constants) start. + */ + unsigned param_binding_begin; + + /** + * Constants put into the parameter list may be swizzled. This + * field contain's the symbol's swizzle. (SWIZZLE_X/Y/Z/W) + */ + unsigned param_binding_swizzle; + + /* This is how many entries in the program_parameter_list we take up + * with our state tokens or constants. Note that this is _not_ the same as + * the number of param registers we eventually use. + */ + unsigned param_binding_length; + + /** + * Index of the temp register assigned to this variable. + */ + unsigned temp_binding; + + /** + * Flag whether or not a PARAM is an array + */ + unsigned param_is_array:1; + + + /** + * Flag whether or not a PARAM array is accessed indirectly + */ + unsigned param_accessed_indirectly:1; + + + /** + * \brief Is first pass of parameter layout done with this variable? + * + * The parameter layout routine operates in two passes. This flag tracks + * whether or not the first pass has handled this variable. + * + * \sa _mesa_layout_parameters + */ + unsigned pass1_done:1; +}; + + +struct asm_vector { + unsigned count; + gl_constant_value data[4]; +}; + + +struct asm_swizzle_mask { + unsigned swizzle:12; + unsigned mask:4; +}; + + +struct asm_src_register { + struct prog_src_register Base; + + /** + * Symbol associated with indirect access to parameter arrays. + * + * If \c Base::RelAddr is 1, this will point to the symbol for the parameter + * that is being dereferenced. Further, \c Base::Index will be the offset + * from the address register being used. + */ + struct asm_symbol *Symbol; +}; + + +struct asm_instruction { + struct prog_instruction Base; + struct asm_instruction *next; + struct asm_src_register SrcReg[3]; +}; + + +struct asm_parser_state { + struct gl_context *ctx; + struct gl_program *prog; + + /** + * Per-program target limits + */ + struct gl_program_constants *limits; + + struct _mesa_symbol_table *st; + + /** + * Linked list of symbols + * + * This list is \b only used when cleaning up compiler state and freeing + * memory. + */ + struct asm_symbol *sym; + + /** + * State for the lexer. + */ + void *scanner; + + /** + * Linked list of instructions generated during parsing. + */ + /*@{*/ + struct asm_instruction *inst_head; + struct asm_instruction *inst_tail; + /*@}*/ + + + /** + * Selected limits copied from gl_constants + * + * These are limits from the GL context, but various bits in the program + * must be validated against these values. + */ + /*@{*/ + unsigned MaxTextureCoordUnits; + unsigned MaxTextureImageUnits; + unsigned MaxTextureUnits; + unsigned MaxClipPlanes; + unsigned MaxLights; + unsigned MaxProgramMatrices; + unsigned MaxDrawBuffers; + /*@}*/ + + /** + * Value to use in state vector accessors for environment and local + * parameters + */ + unsigned state_param_enum; + + + /** + * Input attributes bound to specific names + * + * This is only needed so that errors can be properly produced when + * multiple ATTRIB statements bind illegal combinations of vertex + * attributes. + */ + unsigned InputsBound; + + enum { + invalid_mode = 0, + ARB_vertex, + ARB_fragment + } mode; + + struct { + unsigned PositionInvariant:1; + unsigned Fog:2; + unsigned PrecisionHint:2; + unsigned DrawBuffers:1; + unsigned Shadow:1; + unsigned TexRect:1; + unsigned TexArray:1; + unsigned NV_fragment:1; + unsigned OriginUpperLeft:1; + unsigned PixelCenterInteger:1; + } option; + + struct { + unsigned UsesKill:1; + } fragment; +}; + +#define OPTION_NONE 0 +#define OPTION_FOG_EXP 1 +#define OPTION_FOG_EXP2 2 +#define OPTION_FOG_LINEAR 3 +#define OPTION_NICEST 1 +#define OPTION_FASTEST 2 + +typedef struct YYLTYPE { + int first_line; + int first_column; + int last_line; + int last_column; + int position; +} YYLTYPE; + +#define YYLTYPE_IS_DECLARED 1 +#define YYLTYPE_IS_TRIVIAL 1 + + +extern GLboolean _mesa_parse_arb_program(struct gl_context *ctx, GLenum target, + const GLubyte *str, GLsizei len, struct asm_parser_state *state); + + + +/* From program_lexer.l. */ +extern void _mesa_program_lexer_dtor(void *scanner); + +extern void _mesa_program_lexer_ctor(void **scanner, + struct asm_parser_state *state, const char *string, size_t len); + + +/** + *\name From program_parse_extra.c + */ +/*@{*/ + +/** + * Parses and processes an option string to an ARB vertex program + * + * \return + * Non-zero on success, zero on failure. + */ +extern int _mesa_ARBvp_parse_option(struct asm_parser_state *state, + const char *option); + +/** + * Parses and processes an option string to an ARB fragment program + * + * \return + * Non-zero on success, zero on failure. + */ +extern int _mesa_ARBfp_parse_option(struct asm_parser_state *state, + const char *option); + +/** + * Parses and processes instruction suffixes + * + * Instruction suffixes, such as \c _SAT, are processed. The relevant bits + * are set in \c inst. If suffixes are encountered that are either not known + * or not supported by the modes and options set in \c state, zero will be + * returned. + * + * \return + * Non-zero on success, zero on failure. + */ +extern int _mesa_parse_instruction_suffix(const struct asm_parser_state *state, + const char *suffix, struct prog_instruction *inst); + +/** + * Parses a condition code name + * + * The condition code names (e.g., \c LT, \c GT, \c NE) were added to assembly + * shaders with the \c GL_NV_fragment_program_option extension. This function + * converts a string representation into one of the \c COND_ macros. + * + * \return + * One of the \c COND_ macros defined in prog_instruction.h on success or zero + * on failure. + */ +extern int _mesa_parse_cc(const char *s); + +/*@}*/ diff --git a/mesalib/src/mesa/program/programopt.c b/mesalib/src/mesa/program/programopt.c index 62b406653..c72dfb23b 100644 --- a/mesalib/src/mesa/program/programopt.c +++ b/mesalib/src/mesa/program/programopt.c @@ -1,685 +1,685 @@ -/* - * Mesa 3-D graphics library - * Version: 6.5.3 - * - * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - -/** - * \file programopt.c - * Vertex/Fragment program optimizations and transformations for program - * options, etc. - * - * \author Brian Paul - */ - - -#include "main/glheader.h" -#include "main/context.h" -#include "prog_parameter.h" -#include "prog_statevars.h" -#include "program.h" -#include "programopt.h" -#include "prog_instruction.h" - - -/** - * This function inserts instructions for coordinate modelview * projection - * into a vertex program. - * May be used to implement the position_invariant option. - */ -static void -_mesa_insert_mvp_dp4_code(struct gl_context *ctx, struct gl_vertex_program *vprog) -{ - struct prog_instruction *newInst; - const GLuint origLen = vprog->Base.NumInstructions; - const GLuint newLen = origLen + 4; - GLuint i; - - /* - * Setup state references for the modelview/projection matrix. - * XXX we should check if these state vars are already declared. - */ - static const gl_state_index mvpState[4][STATE_LENGTH] = { - { STATE_MVP_MATRIX, 0, 0, 0, 0 }, /* state.matrix.mvp.row[0] */ - { STATE_MVP_MATRIX, 0, 1, 1, 0 }, /* state.matrix.mvp.row[1] */ - { STATE_MVP_MATRIX, 0, 2, 2, 0 }, /* state.matrix.mvp.row[2] */ - { STATE_MVP_MATRIX, 0, 3, 3, 0 }, /* state.matrix.mvp.row[3] */ - }; - GLint mvpRef[4]; - - for (i = 0; i < 4; i++) { - mvpRef[i] = _mesa_add_state_reference(vprog->Base.Parameters, - mvpState[i]); - } - - /* Alloc storage for new instructions */ - newInst = _mesa_alloc_instructions(newLen); - if (!newInst) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, - "glProgramString(inserting position_invariant code)"); - return; - } - - /* - * Generated instructions: - * newInst[0] = DP4 result.position.x, mvp.row[0], vertex.position; - * newInst[1] = DP4 result.position.y, mvp.row[1], vertex.position; - * newInst[2] = DP4 result.position.z, mvp.row[2], vertex.position; - * newInst[3] = DP4 result.position.w, mvp.row[3], vertex.position; - */ - _mesa_init_instructions(newInst, 4); - for (i = 0; i < 4; i++) { - newInst[i].Opcode = OPCODE_DP4; - newInst[i].DstReg.File = PROGRAM_OUTPUT; - newInst[i].DstReg.Index = VERT_RESULT_HPOS; - newInst[i].DstReg.WriteMask = (WRITEMASK_X << i); - newInst[i].SrcReg[0].File = PROGRAM_STATE_VAR; - newInst[i].SrcReg[0].Index = mvpRef[i]; - newInst[i].SrcReg[0].Swizzle = SWIZZLE_NOOP; - newInst[i].SrcReg[1].File = PROGRAM_INPUT; - newInst[i].SrcReg[1].Index = VERT_ATTRIB_POS; - newInst[i].SrcReg[1].Swizzle = SWIZZLE_NOOP; - } - - /* Append original instructions after new instructions */ - _mesa_copy_instructions (newInst + 4, vprog->Base.Instructions, origLen); - - /* free old instructions */ - _mesa_free_instructions(vprog->Base.Instructions, origLen); - - /* install new instructions */ - vprog->Base.Instructions = newInst; - vprog->Base.NumInstructions = newLen; - vprog->Base.InputsRead |= VERT_BIT_POS; - vprog->Base.OutputsWritten |= BITFIELD64_BIT(VERT_RESULT_HPOS); -} - - -static void -_mesa_insert_mvp_mad_code(struct gl_context *ctx, struct gl_vertex_program *vprog) -{ - struct prog_instruction *newInst; - const GLuint origLen = vprog->Base.NumInstructions; - const GLuint newLen = origLen + 4; - GLuint hposTemp; - GLuint i; - - /* - * Setup state references for the modelview/projection matrix. - * XXX we should check if these state vars are already declared. - */ - static const gl_state_index mvpState[4][STATE_LENGTH] = { - { STATE_MVP_MATRIX, 0, 0, 0, STATE_MATRIX_TRANSPOSE }, - { STATE_MVP_MATRIX, 0, 1, 1, STATE_MATRIX_TRANSPOSE }, - { STATE_MVP_MATRIX, 0, 2, 2, STATE_MATRIX_TRANSPOSE }, - { STATE_MVP_MATRIX, 0, 3, 3, STATE_MATRIX_TRANSPOSE }, - }; - GLint mvpRef[4]; - - for (i = 0; i < 4; i++) { - mvpRef[i] = _mesa_add_state_reference(vprog->Base.Parameters, - mvpState[i]); - } - - /* Alloc storage for new instructions */ - newInst = _mesa_alloc_instructions(newLen); - if (!newInst) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, - "glProgramString(inserting position_invariant code)"); - return; - } - - /* TEMP hposTemp; */ - hposTemp = vprog->Base.NumTemporaries++; - - /* - * Generated instructions: - * emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]); - * emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp); - * emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp); - * emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp); - */ - _mesa_init_instructions(newInst, 4); - - newInst[0].Opcode = OPCODE_MUL; - newInst[0].DstReg.File = PROGRAM_TEMPORARY; - newInst[0].DstReg.Index = hposTemp; - newInst[0].DstReg.WriteMask = WRITEMASK_XYZW; - newInst[0].SrcReg[0].File = PROGRAM_INPUT; - newInst[0].SrcReg[0].Index = VERT_ATTRIB_POS; - newInst[0].SrcReg[0].Swizzle = SWIZZLE_XXXX; - newInst[0].SrcReg[1].File = PROGRAM_STATE_VAR; - newInst[0].SrcReg[1].Index = mvpRef[0]; - newInst[0].SrcReg[1].Swizzle = SWIZZLE_NOOP; - - for (i = 1; i <= 2; i++) { - newInst[i].Opcode = OPCODE_MAD; - newInst[i].DstReg.File = PROGRAM_TEMPORARY; - newInst[i].DstReg.Index = hposTemp; - newInst[i].DstReg.WriteMask = WRITEMASK_XYZW; - newInst[i].SrcReg[0].File = PROGRAM_INPUT; - newInst[i].SrcReg[0].Index = VERT_ATTRIB_POS; - newInst[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(i,i,i,i); - newInst[i].SrcReg[1].File = PROGRAM_STATE_VAR; - newInst[i].SrcReg[1].Index = mvpRef[i]; - newInst[i].SrcReg[1].Swizzle = SWIZZLE_NOOP; - newInst[i].SrcReg[2].File = PROGRAM_TEMPORARY; - newInst[i].SrcReg[2].Index = hposTemp; - newInst[1].SrcReg[2].Swizzle = SWIZZLE_NOOP; - } - - newInst[3].Opcode = OPCODE_MAD; - newInst[3].DstReg.File = PROGRAM_OUTPUT; - newInst[3].DstReg.Index = VERT_RESULT_HPOS; - newInst[3].DstReg.WriteMask = WRITEMASK_XYZW; - newInst[3].SrcReg[0].File = PROGRAM_INPUT; - newInst[3].SrcReg[0].Index = VERT_ATTRIB_POS; - newInst[3].SrcReg[0].Swizzle = SWIZZLE_WWWW; - newInst[3].SrcReg[1].File = PROGRAM_STATE_VAR; - newInst[3].SrcReg[1].Index = mvpRef[3]; - newInst[3].SrcReg[1].Swizzle = SWIZZLE_NOOP; - newInst[3].SrcReg[2].File = PROGRAM_TEMPORARY; - newInst[3].SrcReg[2].Index = hposTemp; - newInst[3].SrcReg[2].Swizzle = SWIZZLE_NOOP; - - - /* Append original instructions after new instructions */ - _mesa_copy_instructions (newInst + 4, vprog->Base.Instructions, origLen); - - /* free old instructions */ - _mesa_free_instructions(vprog->Base.Instructions, origLen); - - /* install new instructions */ - vprog->Base.Instructions = newInst; - vprog->Base.NumInstructions = newLen; - vprog->Base.InputsRead |= VERT_BIT_POS; - vprog->Base.OutputsWritten |= BITFIELD64_BIT(VERT_RESULT_HPOS); -} - - -void -_mesa_insert_mvp_code(struct gl_context *ctx, struct gl_vertex_program *vprog) -{ - if (ctx->mvp_with_dp4) - _mesa_insert_mvp_dp4_code( ctx, vprog ); - else - _mesa_insert_mvp_mad_code( ctx, vprog ); -} - - - - - - -/** - * Append instructions to implement fog - * - * The \c fragment.fogcoord input is used to compute the fog blend factor. - * - * \param ctx The GL context - * \param fprog Fragment program that fog instructions will be appended to. - * \param fog_mode Fog mode. One of \c GL_EXP, \c GL_EXP2, or \c GL_LINEAR. - * \param saturate True if writes to color outputs should be clamped to [0, 1] - * - * \note - * This function sets \c FRAG_BIT_FOGC in \c fprog->Base.InputsRead. - * - * \todo With a little work, this function could be adapted to add fog code - * to vertex programs too. - */ -void -_mesa_append_fog_code(struct gl_context *ctx, - struct gl_fragment_program *fprog, GLenum fog_mode, - GLboolean saturate) -{ - static const gl_state_index fogPStateOpt[STATE_LENGTH] - = { STATE_INTERNAL, STATE_FOG_PARAMS_OPTIMIZED, 0, 0, 0 }; - static const gl_state_index fogColorState[STATE_LENGTH] - = { STATE_FOG_COLOR, 0, 0, 0, 0}; - struct prog_instruction *newInst, *inst; - const GLuint origLen = fprog->Base.NumInstructions; - const GLuint newLen = origLen + 5; - GLuint i; - GLint fogPRefOpt, fogColorRef; /* state references */ - GLuint colorTemp, fogFactorTemp; /* temporary registerss */ - - if (fog_mode == GL_NONE) { - _mesa_problem(ctx, "_mesa_append_fog_code() called for fragment program" - " with fog_mode == GL_NONE"); - return; - } - - if (!(fprog->Base.OutputsWritten & (1 << FRAG_RESULT_COLOR))) { - /* program doesn't output color, so nothing to do */ - return; - } - - /* Alloc storage for new instructions */ - newInst = _mesa_alloc_instructions(newLen); - if (!newInst) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, - "glProgramString(inserting fog_option code)"); - return; - } - - /* Copy orig instructions into new instruction buffer */ - _mesa_copy_instructions(newInst, fprog->Base.Instructions, origLen); - - /* PARAM fogParamsRefOpt = internal optimized fog params; */ - fogPRefOpt - = _mesa_add_state_reference(fprog->Base.Parameters, fogPStateOpt); - /* PARAM fogColorRef = state.fog.color; */ - fogColorRef - = _mesa_add_state_reference(fprog->Base.Parameters, fogColorState); - - /* TEMP colorTemp; */ - colorTemp = fprog->Base.NumTemporaries++; - /* TEMP fogFactorTemp; */ - fogFactorTemp = fprog->Base.NumTemporaries++; - - /* Scan program to find where result.color is written */ - inst = newInst; - for (i = 0; i < fprog->Base.NumInstructions; i++) { - if (inst->Opcode == OPCODE_END) - break; - if (inst->DstReg.File == PROGRAM_OUTPUT && - inst->DstReg.Index == FRAG_RESULT_COLOR) { - /* change the instruction to write to colorTemp w/ clamping */ - inst->DstReg.File = PROGRAM_TEMPORARY; - inst->DstReg.Index = colorTemp; - inst->SaturateMode = saturate; - /* don't break (may be several writes to result.color) */ - } - inst++; - } - assert(inst->Opcode == OPCODE_END); /* we'll overwrite this inst */ - - _mesa_init_instructions(inst, 5); - - /* emit instructions to compute fog blending factor */ - /* this is always clamped to [0, 1] regardless of fragment clamping */ - if (fog_mode == GL_LINEAR) { - /* MAD fogFactorTemp.x, fragment.fogcoord.x, fogPRefOpt.x, fogPRefOpt.y; */ - inst->Opcode = OPCODE_MAD; - inst->DstReg.File = PROGRAM_TEMPORARY; - inst->DstReg.Index = fogFactorTemp; - inst->DstReg.WriteMask = WRITEMASK_X; - inst->SrcReg[0].File = PROGRAM_INPUT; - inst->SrcReg[0].Index = FRAG_ATTRIB_FOGC; - inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; - inst->SrcReg[1].File = PROGRAM_STATE_VAR; - inst->SrcReg[1].Index = fogPRefOpt; - inst->SrcReg[1].Swizzle = SWIZZLE_XXXX; - inst->SrcReg[2].File = PROGRAM_STATE_VAR; - inst->SrcReg[2].Index = fogPRefOpt; - inst->SrcReg[2].Swizzle = SWIZZLE_YYYY; - inst->SaturateMode = SATURATE_ZERO_ONE; - inst++; - } - else { - ASSERT(fog_mode == GL_EXP || fog_mode == GL_EXP2); - /* fogPRefOpt.z = d/ln(2), fogPRefOpt.w = d/sqrt(ln(2) */ - /* EXP: MUL fogFactorTemp.x, fogPRefOpt.z, fragment.fogcoord.x; */ - /* EXP2: MUL fogFactorTemp.x, fogPRefOpt.w, fragment.fogcoord.x; */ - inst->Opcode = OPCODE_MUL; - inst->DstReg.File = PROGRAM_TEMPORARY; - inst->DstReg.Index = fogFactorTemp; - inst->DstReg.WriteMask = WRITEMASK_X; - inst->SrcReg[0].File = PROGRAM_STATE_VAR; - inst->SrcReg[0].Index = fogPRefOpt; - inst->SrcReg[0].Swizzle - = (fog_mode == GL_EXP) ? SWIZZLE_ZZZZ : SWIZZLE_WWWW; - inst->SrcReg[1].File = PROGRAM_INPUT; - inst->SrcReg[1].Index = FRAG_ATTRIB_FOGC; - inst->SrcReg[1].Swizzle = SWIZZLE_XXXX; - inst++; - if (fog_mode == GL_EXP2) { - /* MUL fogFactorTemp.x, fogFactorTemp.x, fogFactorTemp.x; */ - inst->Opcode = OPCODE_MUL; - inst->DstReg.File = PROGRAM_TEMPORARY; - inst->DstReg.Index = fogFactorTemp; - inst->DstReg.WriteMask = WRITEMASK_X; - inst->SrcReg[0].File = PROGRAM_TEMPORARY; - inst->SrcReg[0].Index = fogFactorTemp; - inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; - inst->SrcReg[1].File = PROGRAM_TEMPORARY; - inst->SrcReg[1].Index = fogFactorTemp; - inst->SrcReg[1].Swizzle = SWIZZLE_XXXX; - inst++; - } - /* EX2_SAT fogFactorTemp.x, -fogFactorTemp.x; */ - inst->Opcode = OPCODE_EX2; - inst->DstReg.File = PROGRAM_TEMPORARY; - inst->DstReg.Index = fogFactorTemp; - inst->DstReg.WriteMask = WRITEMASK_X; - inst->SrcReg[0].File = PROGRAM_TEMPORARY; - inst->SrcReg[0].Index = fogFactorTemp; - inst->SrcReg[0].Negate = NEGATE_XYZW; - inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; - inst->SaturateMode = SATURATE_ZERO_ONE; - inst++; - } - /* LRP result.color.xyz, fogFactorTemp.xxxx, colorTemp, fogColorRef; */ - inst->Opcode = OPCODE_LRP; - inst->DstReg.File = PROGRAM_OUTPUT; - inst->DstReg.Index = FRAG_RESULT_COLOR; - inst->DstReg.WriteMask = WRITEMASK_XYZ; - inst->SrcReg[0].File = PROGRAM_TEMPORARY; - inst->SrcReg[0].Index = fogFactorTemp; - inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; - inst->SrcReg[1].File = PROGRAM_TEMPORARY; - inst->SrcReg[1].Index = colorTemp; - inst->SrcReg[1].Swizzle = SWIZZLE_NOOP; - inst->SrcReg[2].File = PROGRAM_STATE_VAR; - inst->SrcReg[2].Index = fogColorRef; - inst->SrcReg[2].Swizzle = SWIZZLE_NOOP; - inst++; - /* MOV result.color.w, colorTemp.x; # copy alpha */ - inst->Opcode = OPCODE_MOV; - inst->DstReg.File = PROGRAM_OUTPUT; - inst->DstReg.Index = FRAG_RESULT_COLOR; - inst->DstReg.WriteMask = WRITEMASK_W; - inst->SrcReg[0].File = PROGRAM_TEMPORARY; - inst->SrcReg[0].Index = colorTemp; - inst->SrcReg[0].Swizzle = SWIZZLE_NOOP; - inst++; - /* END; */ - inst->Opcode = OPCODE_END; - inst++; - - /* free old instructions */ - _mesa_free_instructions(fprog->Base.Instructions, origLen); - - /* install new instructions */ - fprog->Base.Instructions = newInst; - fprog->Base.NumInstructions = inst - newInst; - fprog->Base.InputsRead |= FRAG_BIT_FOGC; - assert(fprog->Base.OutputsWritten & (1 << FRAG_RESULT_COLOR)); -} - - - -static GLboolean -is_texture_instruction(const struct prog_instruction *inst) -{ - switch (inst->Opcode) { - case OPCODE_TEX: - case OPCODE_TXB: - case OPCODE_TXD: - case OPCODE_TXL: - case OPCODE_TXP: - case OPCODE_TXP_NV: - return GL_TRUE; - default: - return GL_FALSE; - } -} - - -/** - * Count the number of texure indirections in the given program. - * The program's NumTexIndirections field will be updated. - * See the GL_ARB_fragment_program spec (issue 24) for details. - * XXX we count texture indirections in texenvprogram.c (maybe use this code - * instead and elsewhere). - */ -void -_mesa_count_texture_indirections(struct gl_program *prog) -{ - GLuint indirections = 1; - GLbitfield tempsOutput = 0x0; - GLbitfield aluTemps = 0x0; - GLuint i; - - for (i = 0; i < prog->NumInstructions; i++) { - const struct prog_instruction *inst = prog->Instructions + i; - - if (is_texture_instruction(inst)) { - if (((inst->SrcReg[0].File == PROGRAM_TEMPORARY) && - (tempsOutput & (1 << inst->SrcReg[0].Index))) || - ((inst->Opcode != OPCODE_KIL) && - (inst->DstReg.File == PROGRAM_TEMPORARY) && - (aluTemps & (1 << inst->DstReg.Index)))) - { - indirections++; - tempsOutput = 0x0; - aluTemps = 0x0; - } - } - else { - GLuint j; - for (j = 0; j < 3; j++) { - if (inst->SrcReg[j].File == PROGRAM_TEMPORARY) - aluTemps |= (1 << inst->SrcReg[j].Index); - } - if (inst->DstReg.File == PROGRAM_TEMPORARY) - aluTemps |= (1 << inst->DstReg.Index); - } - - if ((inst->Opcode != OPCODE_KIL) && (inst->DstReg.File == PROGRAM_TEMPORARY)) - tempsOutput |= (1 << inst->DstReg.Index); - } - - prog->NumTexIndirections = indirections; -} - - -/** - * Count number of texture instructions in given program and update the - * program's NumTexInstructions field. - */ -void -_mesa_count_texture_instructions(struct gl_program *prog) -{ - GLuint i; - prog->NumTexInstructions = 0; - for (i = 0; i < prog->NumInstructions; i++) { - prog->NumTexInstructions += is_texture_instruction(prog->Instructions + i); - } -} - - -/** - * Scan/rewrite program to remove reads of custom (output) registers. - * The passed type has to be either PROGRAM_OUTPUT or PROGRAM_VARYING - * (for vertex shaders). - * In GLSL shaders, varying vars can be read and written. - * On some hardware, trying to read an output register causes trouble. - * So, rewrite the program to use a temporary register in this case. - */ -void -_mesa_remove_output_reads(struct gl_program *prog, gl_register_file type) -{ - GLuint i; - GLint outputMap[VERT_RESULT_MAX]; - GLuint numVaryingReads = 0; - GLboolean usedTemps[MAX_PROGRAM_TEMPS]; - GLuint firstTemp = 0; - - _mesa_find_used_registers(prog, PROGRAM_TEMPORARY, - usedTemps, MAX_PROGRAM_TEMPS); - - assert(type == PROGRAM_VARYING || type == PROGRAM_OUTPUT); - assert(prog->Target == GL_VERTEX_PROGRAM_ARB || type != PROGRAM_VARYING); - - for (i = 0; i < VERT_RESULT_MAX; i++) - outputMap[i] = -1; - - /* look for instructions which read from varying vars */ - for (i = 0; i < prog->NumInstructions; i++) { - struct prog_instruction *inst = prog->Instructions + i; - const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode); - GLuint j; - for (j = 0; j < numSrc; j++) { - if (inst->SrcReg[j].File == type) { - /* replace the read with a temp reg */ - const GLuint var = inst->SrcReg[j].Index; - if (outputMap[var] == -1) { - numVaryingReads++; - outputMap[var] = _mesa_find_free_register(usedTemps, - MAX_PROGRAM_TEMPS, - firstTemp); - firstTemp = outputMap[var] + 1; - } - inst->SrcReg[j].File = PROGRAM_TEMPORARY; - inst->SrcReg[j].Index = outputMap[var]; - } - } - } - - if (numVaryingReads == 0) - return; /* nothing to be done */ - - /* look for instructions which write to the varying vars identified above */ - for (i = 0; i < prog->NumInstructions; i++) { - struct prog_instruction *inst = prog->Instructions + i; - if (inst->DstReg.File == type && - outputMap[inst->DstReg.Index] >= 0) { - /* change inst to write to the temp reg, instead of the varying */ - inst->DstReg.File = PROGRAM_TEMPORARY; - inst->DstReg.Index = outputMap[inst->DstReg.Index]; - } - } - - /* insert new instructions to copy the temp vars to the varying vars */ - { - struct prog_instruction *inst; - GLint endPos, var; - - /* Look for END instruction and insert the new varying writes */ - endPos = -1; - for (i = 0; i < prog->NumInstructions; i++) { - struct prog_instruction *inst = prog->Instructions + i; - if (inst->Opcode == OPCODE_END) { - endPos = i; - _mesa_insert_instructions(prog, i, numVaryingReads); - break; - } - } - - assert(endPos >= 0); - - /* insert new MOV instructions here */ - inst = prog->Instructions + endPos; - for (var = 0; var < VERT_RESULT_MAX; var++) { - if (outputMap[var] >= 0) { - /* MOV VAR[var], TEMP[tmp]; */ - inst->Opcode = OPCODE_MOV; - inst->DstReg.File = type; - inst->DstReg.Index = var; - inst->SrcReg[0].File = PROGRAM_TEMPORARY; - inst->SrcReg[0].Index = outputMap[var]; - inst++; - } - } - } -} - - -/** - * Make the given fragment program into a "no-op" shader. - * Actually, just copy the incoming fragment color (or texcoord) - * to the output color. - * This is for debug/test purposes. - */ -void -_mesa_nop_fragment_program(struct gl_context *ctx, struct gl_fragment_program *prog) -{ - struct prog_instruction *inst; - GLuint inputAttr; - - inst = _mesa_alloc_instructions(2); - if (!inst) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "_mesa_nop_fragment_program"); - return; - } - - _mesa_init_instructions(inst, 2); - - inst[0].Opcode = OPCODE_MOV; - inst[0].DstReg.File = PROGRAM_OUTPUT; - inst[0].DstReg.Index = FRAG_RESULT_COLOR; - inst[0].SrcReg[0].File = PROGRAM_INPUT; - if (prog->Base.InputsRead & FRAG_BIT_COL0) - inputAttr = FRAG_ATTRIB_COL0; - else - inputAttr = FRAG_ATTRIB_TEX0; - inst[0].SrcReg[0].Index = inputAttr; - - inst[1].Opcode = OPCODE_END; - - _mesa_free_instructions(prog->Base.Instructions, - prog->Base.NumInstructions); - - prog->Base.Instructions = inst; - prog->Base.NumInstructions = 2; - prog->Base.InputsRead = 1 << inputAttr; - prog->Base.OutputsWritten = BITFIELD64_BIT(FRAG_RESULT_COLOR); -} - - -/** - * \sa _mesa_nop_fragment_program - * Replace the given vertex program with a "no-op" program that just - * transforms vertex position and emits color. - */ -void -_mesa_nop_vertex_program(struct gl_context *ctx, struct gl_vertex_program *prog) -{ - struct prog_instruction *inst; - GLuint inputAttr; - - /* - * Start with a simple vertex program that emits color. - */ - inst = _mesa_alloc_instructions(2); - if (!inst) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "_mesa_nop_vertex_program"); - return; - } - - _mesa_init_instructions(inst, 2); - - inst[0].Opcode = OPCODE_MOV; - inst[0].DstReg.File = PROGRAM_OUTPUT; - inst[0].DstReg.Index = VERT_RESULT_COL0; - inst[0].SrcReg[0].File = PROGRAM_INPUT; - if (prog->Base.InputsRead & VERT_BIT_COLOR0) - inputAttr = VERT_ATTRIB_COLOR0; - else - inputAttr = VERT_ATTRIB_TEX0; - inst[0].SrcReg[0].Index = inputAttr; - - inst[1].Opcode = OPCODE_END; - - _mesa_free_instructions(prog->Base.Instructions, - prog->Base.NumInstructions); - - prog->Base.Instructions = inst; - prog->Base.NumInstructions = 2; - prog->Base.InputsRead = 1 << inputAttr; - prog->Base.OutputsWritten = BITFIELD64_BIT(VERT_RESULT_COL0); - - /* - * Now insert code to do standard modelview/projection transformation. - */ - _mesa_insert_mvp_code(ctx, prog); -} +/* + * Mesa 3-D graphics library + * Version: 6.5.3 + * + * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/** + * \file programopt.c + * Vertex/Fragment program optimizations and transformations for program + * options, etc. + * + * \author Brian Paul + */ + + +#include "main/glheader.h" +#include "main/context.h" +#include "prog_parameter.h" +#include "prog_statevars.h" +#include "program.h" +#include "programopt.h" +#include "prog_instruction.h" + + +/** + * This function inserts instructions for coordinate modelview * projection + * into a vertex program. + * May be used to implement the position_invariant option. + */ +static void +_mesa_insert_mvp_dp4_code(struct gl_context *ctx, struct gl_vertex_program *vprog) +{ + struct prog_instruction *newInst; + const GLuint origLen = vprog->Base.NumInstructions; + const GLuint newLen = origLen + 4; + GLuint i; + + /* + * Setup state references for the modelview/projection matrix. + * XXX we should check if these state vars are already declared. + */ + static const gl_state_index mvpState[4][STATE_LENGTH] = { + { STATE_MVP_MATRIX, 0, 0, 0, 0 }, /* state.matrix.mvp.row[0] */ + { STATE_MVP_MATRIX, 0, 1, 1, 0 }, /* state.matrix.mvp.row[1] */ + { STATE_MVP_MATRIX, 0, 2, 2, 0 }, /* state.matrix.mvp.row[2] */ + { STATE_MVP_MATRIX, 0, 3, 3, 0 }, /* state.matrix.mvp.row[3] */ + }; + GLint mvpRef[4]; + + for (i = 0; i < 4; i++) { + mvpRef[i] = _mesa_add_state_reference(vprog->Base.Parameters, + mvpState[i]); + } + + /* Alloc storage for new instructions */ + newInst = _mesa_alloc_instructions(newLen); + if (!newInst) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, + "glProgramString(inserting position_invariant code)"); + return; + } + + /* + * Generated instructions: + * newInst[0] = DP4 result.position.x, mvp.row[0], vertex.position; + * newInst[1] = DP4 result.position.y, mvp.row[1], vertex.position; + * newInst[2] = DP4 result.position.z, mvp.row[2], vertex.position; + * newInst[3] = DP4 result.position.w, mvp.row[3], vertex.position; + */ + _mesa_init_instructions(newInst, 4); + for (i = 0; i < 4; i++) { + newInst[i].Opcode = OPCODE_DP4; + newInst[i].DstReg.File = PROGRAM_OUTPUT; + newInst[i].DstReg.Index = VERT_RESULT_HPOS; + newInst[i].DstReg.WriteMask = (WRITEMASK_X << i); + newInst[i].SrcReg[0].File = PROGRAM_STATE_VAR; + newInst[i].SrcReg[0].Index = mvpRef[i]; + newInst[i].SrcReg[0].Swizzle = SWIZZLE_NOOP; + newInst[i].SrcReg[1].File = PROGRAM_INPUT; + newInst[i].SrcReg[1].Index = VERT_ATTRIB_POS; + newInst[i].SrcReg[1].Swizzle = SWIZZLE_NOOP; + } + + /* Append original instructions after new instructions */ + _mesa_copy_instructions (newInst + 4, vprog->Base.Instructions, origLen); + + /* free old instructions */ + _mesa_free_instructions(vprog->Base.Instructions, origLen); + + /* install new instructions */ + vprog->Base.Instructions = newInst; + vprog->Base.NumInstructions = newLen; + vprog->Base.InputsRead |= VERT_BIT_POS; + vprog->Base.OutputsWritten |= BITFIELD64_BIT(VERT_RESULT_HPOS); +} + + +static void +_mesa_insert_mvp_mad_code(struct gl_context *ctx, struct gl_vertex_program *vprog) +{ + struct prog_instruction *newInst; + const GLuint origLen = vprog->Base.NumInstructions; + const GLuint newLen = origLen + 4; + GLuint hposTemp; + GLuint i; + + /* + * Setup state references for the modelview/projection matrix. + * XXX we should check if these state vars are already declared. + */ + static const gl_state_index mvpState[4][STATE_LENGTH] = { + { STATE_MVP_MATRIX, 0, 0, 0, STATE_MATRIX_TRANSPOSE }, + { STATE_MVP_MATRIX, 0, 1, 1, STATE_MATRIX_TRANSPOSE }, + { STATE_MVP_MATRIX, 0, 2, 2, STATE_MATRIX_TRANSPOSE }, + { STATE_MVP_MATRIX, 0, 3, 3, STATE_MATRIX_TRANSPOSE }, + }; + GLint mvpRef[4]; + + for (i = 0; i < 4; i++) { + mvpRef[i] = _mesa_add_state_reference(vprog->Base.Parameters, + mvpState[i]); + } + + /* Alloc storage for new instructions */ + newInst = _mesa_alloc_instructions(newLen); + if (!newInst) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, + "glProgramString(inserting position_invariant code)"); + return; + } + + /* TEMP hposTemp; */ + hposTemp = vprog->Base.NumTemporaries++; + + /* + * Generated instructions: + * emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]); + * emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp); + * emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp); + * emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp); + */ + _mesa_init_instructions(newInst, 4); + + newInst[0].Opcode = OPCODE_MUL; + newInst[0].DstReg.File = PROGRAM_TEMPORARY; + newInst[0].DstReg.Index = hposTemp; + newInst[0].DstReg.WriteMask = WRITEMASK_XYZW; + newInst[0].SrcReg[0].File = PROGRAM_INPUT; + newInst[0].SrcReg[0].Index = VERT_ATTRIB_POS; + newInst[0].SrcReg[0].Swizzle = SWIZZLE_XXXX; + newInst[0].SrcReg[1].File = PROGRAM_STATE_VAR; + newInst[0].SrcReg[1].Index = mvpRef[0]; + newInst[0].SrcReg[1].Swizzle = SWIZZLE_NOOP; + + for (i = 1; i <= 2; i++) { + newInst[i].Opcode = OPCODE_MAD; + newInst[i].DstReg.File = PROGRAM_TEMPORARY; + newInst[i].DstReg.Index = hposTemp; + newInst[i].DstReg.WriteMask = WRITEMASK_XYZW; + newInst[i].SrcReg[0].File = PROGRAM_INPUT; + newInst[i].SrcReg[0].Index = VERT_ATTRIB_POS; + newInst[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(i,i,i,i); + newInst[i].SrcReg[1].File = PROGRAM_STATE_VAR; + newInst[i].SrcReg[1].Index = mvpRef[i]; + newInst[i].SrcReg[1].Swizzle = SWIZZLE_NOOP; + newInst[i].SrcReg[2].File = PROGRAM_TEMPORARY; + newInst[i].SrcReg[2].Index = hposTemp; + newInst[1].SrcReg[2].Swizzle = SWIZZLE_NOOP; + } + + newInst[3].Opcode = OPCODE_MAD; + newInst[3].DstReg.File = PROGRAM_OUTPUT; + newInst[3].DstReg.Index = VERT_RESULT_HPOS; + newInst[3].DstReg.WriteMask = WRITEMASK_XYZW; + newInst[3].SrcReg[0].File = PROGRAM_INPUT; + newInst[3].SrcReg[0].Index = VERT_ATTRIB_POS; + newInst[3].SrcReg[0].Swizzle = SWIZZLE_WWWW; + newInst[3].SrcReg[1].File = PROGRAM_STATE_VAR; + newInst[3].SrcReg[1].Index = mvpRef[3]; + newInst[3].SrcReg[1].Swizzle = SWIZZLE_NOOP; + newInst[3].SrcReg[2].File = PROGRAM_TEMPORARY; + newInst[3].SrcReg[2].Index = hposTemp; + newInst[3].SrcReg[2].Swizzle = SWIZZLE_NOOP; + + + /* Append original instructions after new instructions */ + _mesa_copy_instructions (newInst + 4, vprog->Base.Instructions, origLen); + + /* free old instructions */ + _mesa_free_instructions(vprog->Base.Instructions, origLen); + + /* install new instructions */ + vprog->Base.Instructions = newInst; + vprog->Base.NumInstructions = newLen; + vprog->Base.InputsRead |= VERT_BIT_POS; + vprog->Base.OutputsWritten |= BITFIELD64_BIT(VERT_RESULT_HPOS); +} + + +void +_mesa_insert_mvp_code(struct gl_context *ctx, struct gl_vertex_program *vprog) +{ + if (ctx->mvp_with_dp4) + _mesa_insert_mvp_dp4_code( ctx, vprog ); + else + _mesa_insert_mvp_mad_code( ctx, vprog ); +} + + + + + + +/** + * Append instructions to implement fog + * + * The \c fragment.fogcoord input is used to compute the fog blend factor. + * + * \param ctx The GL context + * \param fprog Fragment program that fog instructions will be appended to. + * \param fog_mode Fog mode. One of \c GL_EXP, \c GL_EXP2, or \c GL_LINEAR. + * \param saturate True if writes to color outputs should be clamped to [0, 1] + * + * \note + * This function sets \c FRAG_BIT_FOGC in \c fprog->Base.InputsRead. + * + * \todo With a little work, this function could be adapted to add fog code + * to vertex programs too. + */ +void +_mesa_append_fog_code(struct gl_context *ctx, + struct gl_fragment_program *fprog, GLenum fog_mode, + GLboolean saturate) +{ + static const gl_state_index fogPStateOpt[STATE_LENGTH] + = { STATE_INTERNAL, STATE_FOG_PARAMS_OPTIMIZED, 0, 0, 0 }; + static const gl_state_index fogColorState[STATE_LENGTH] + = { STATE_FOG_COLOR, 0, 0, 0, 0}; + struct prog_instruction *newInst, *inst; + const GLuint origLen = fprog->Base.NumInstructions; + const GLuint newLen = origLen + 5; + GLuint i; + GLint fogPRefOpt, fogColorRef; /* state references */ + GLuint colorTemp, fogFactorTemp; /* temporary registerss */ + + if (fog_mode == GL_NONE) { + _mesa_problem(ctx, "_mesa_append_fog_code() called for fragment program" + " with fog_mode == GL_NONE"); + return; + } + + if (!(fprog->Base.OutputsWritten & (1 << FRAG_RESULT_COLOR))) { + /* program doesn't output color, so nothing to do */ + return; + } + + /* Alloc storage for new instructions */ + newInst = _mesa_alloc_instructions(newLen); + if (!newInst) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, + "glProgramString(inserting fog_option code)"); + return; + } + + /* Copy orig instructions into new instruction buffer */ + _mesa_copy_instructions(newInst, fprog->Base.Instructions, origLen); + + /* PARAM fogParamsRefOpt = internal optimized fog params; */ + fogPRefOpt + = _mesa_add_state_reference(fprog->Base.Parameters, fogPStateOpt); + /* PARAM fogColorRef = state.fog.color; */ + fogColorRef + = _mesa_add_state_reference(fprog->Base.Parameters, fogColorState); + + /* TEMP colorTemp; */ + colorTemp = fprog->Base.NumTemporaries++; + /* TEMP fogFactorTemp; */ + fogFactorTemp = fprog->Base.NumTemporaries++; + + /* Scan program to find where result.color is written */ + inst = newInst; + for (i = 0; i < fprog->Base.NumInstructions; i++) { + if (inst->Opcode == OPCODE_END) + break; + if (inst->DstReg.File == PROGRAM_OUTPUT && + inst->DstReg.Index == FRAG_RESULT_COLOR) { + /* change the instruction to write to colorTemp w/ clamping */ + inst->DstReg.File = PROGRAM_TEMPORARY; + inst->DstReg.Index = colorTemp; + inst->SaturateMode = saturate; + /* don't break (may be several writes to result.color) */ + } + inst++; + } + assert(inst->Opcode == OPCODE_END); /* we'll overwrite this inst */ + + _mesa_init_instructions(inst, 5); + + /* emit instructions to compute fog blending factor */ + /* this is always clamped to [0, 1] regardless of fragment clamping */ + if (fog_mode == GL_LINEAR) { + /* MAD fogFactorTemp.x, fragment.fogcoord.x, fogPRefOpt.x, fogPRefOpt.y; */ + inst->Opcode = OPCODE_MAD; + inst->DstReg.File = PROGRAM_TEMPORARY; + inst->DstReg.Index = fogFactorTemp; + inst->DstReg.WriteMask = WRITEMASK_X; + inst->SrcReg[0].File = PROGRAM_INPUT; + inst->SrcReg[0].Index = FRAG_ATTRIB_FOGC; + inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; + inst->SrcReg[1].File = PROGRAM_STATE_VAR; + inst->SrcReg[1].Index = fogPRefOpt; + inst->SrcReg[1].Swizzle = SWIZZLE_XXXX; + inst->SrcReg[2].File = PROGRAM_STATE_VAR; + inst->SrcReg[2].Index = fogPRefOpt; + inst->SrcReg[2].Swizzle = SWIZZLE_YYYY; + inst->SaturateMode = SATURATE_ZERO_ONE; + inst++; + } + else { + ASSERT(fog_mode == GL_EXP || fog_mode == GL_EXP2); + /* fogPRefOpt.z = d/ln(2), fogPRefOpt.w = d/sqrt(ln(2) */ + /* EXP: MUL fogFactorTemp.x, fogPRefOpt.z, fragment.fogcoord.x; */ + /* EXP2: MUL fogFactorTemp.x, fogPRefOpt.w, fragment.fogcoord.x; */ + inst->Opcode = OPCODE_MUL; + inst->DstReg.File = PROGRAM_TEMPORARY; + inst->DstReg.Index = fogFactorTemp; + inst->DstReg.WriteMask = WRITEMASK_X; + inst->SrcReg[0].File = PROGRAM_STATE_VAR; + inst->SrcReg[0].Index = fogPRefOpt; + inst->SrcReg[0].Swizzle + = (fog_mode == GL_EXP) ? SWIZZLE_ZZZZ : SWIZZLE_WWWW; + inst->SrcReg[1].File = PROGRAM_INPUT; + inst->SrcReg[1].Index = FRAG_ATTRIB_FOGC; + inst->SrcReg[1].Swizzle = SWIZZLE_XXXX; + inst++; + if (fog_mode == GL_EXP2) { + /* MUL fogFactorTemp.x, fogFactorTemp.x, fogFactorTemp.x; */ + inst->Opcode = OPCODE_MUL; + inst->DstReg.File = PROGRAM_TEMPORARY; + inst->DstReg.Index = fogFactorTemp; + inst->DstReg.WriteMask = WRITEMASK_X; + inst->SrcReg[0].File = PROGRAM_TEMPORARY; + inst->SrcReg[0].Index = fogFactorTemp; + inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; + inst->SrcReg[1].File = PROGRAM_TEMPORARY; + inst->SrcReg[1].Index = fogFactorTemp; + inst->SrcReg[1].Swizzle = SWIZZLE_XXXX; + inst++; + } + /* EX2_SAT fogFactorTemp.x, -fogFactorTemp.x; */ + inst->Opcode = OPCODE_EX2; + inst->DstReg.File = PROGRAM_TEMPORARY; + inst->DstReg.Index = fogFactorTemp; + inst->DstReg.WriteMask = WRITEMASK_X; + inst->SrcReg[0].File = PROGRAM_TEMPORARY; + inst->SrcReg[0].Index = fogFactorTemp; + inst->SrcReg[0].Negate = NEGATE_XYZW; + inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; + inst->SaturateMode = SATURATE_ZERO_ONE; + inst++; + } + /* LRP result.color.xyz, fogFactorTemp.xxxx, colorTemp, fogColorRef; */ + inst->Opcode = OPCODE_LRP; + inst->DstReg.File = PROGRAM_OUTPUT; + inst->DstReg.Index = FRAG_RESULT_COLOR; + inst->DstReg.WriteMask = WRITEMASK_XYZ; + inst->SrcReg[0].File = PROGRAM_TEMPORARY; + inst->SrcReg[0].Index = fogFactorTemp; + inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; + inst->SrcReg[1].File = PROGRAM_TEMPORARY; + inst->SrcReg[1].Index = colorTemp; + inst->SrcReg[1].Swizzle = SWIZZLE_NOOP; + inst->SrcReg[2].File = PROGRAM_STATE_VAR; + inst->SrcReg[2].Index = fogColorRef; + inst->SrcReg[2].Swizzle = SWIZZLE_NOOP; + inst++; + /* MOV result.color.w, colorTemp.x; # copy alpha */ + inst->Opcode = OPCODE_MOV; + inst->DstReg.File = PROGRAM_OUTPUT; + inst->DstReg.Index = FRAG_RESULT_COLOR; + inst->DstReg.WriteMask = WRITEMASK_W; + inst->SrcReg[0].File = PROGRAM_TEMPORARY; + inst->SrcReg[0].Index = colorTemp; + inst->SrcReg[0].Swizzle = SWIZZLE_NOOP; + inst++; + /* END; */ + inst->Opcode = OPCODE_END; + inst++; + + /* free old instructions */ + _mesa_free_instructions(fprog->Base.Instructions, origLen); + + /* install new instructions */ + fprog->Base.Instructions = newInst; + fprog->Base.NumInstructions = inst - newInst; + fprog->Base.InputsRead |= FRAG_BIT_FOGC; + assert(fprog->Base.OutputsWritten & (1 << FRAG_RESULT_COLOR)); +} + + + +static GLboolean +is_texture_instruction(const struct prog_instruction *inst) +{ + switch (inst->Opcode) { + case OPCODE_TEX: + case OPCODE_TXB: + case OPCODE_TXD: + case OPCODE_TXL: + case OPCODE_TXP: + case OPCODE_TXP_NV: + return GL_TRUE; + default: + return GL_FALSE; + } +} + + +/** + * Count the number of texure indirections in the given program. + * The program's NumTexIndirections field will be updated. + * See the GL_ARB_fragment_program spec (issue 24) for details. + * XXX we count texture indirections in texenvprogram.c (maybe use this code + * instead and elsewhere). + */ +void +_mesa_count_texture_indirections(struct gl_program *prog) +{ + GLuint indirections = 1; + GLbitfield tempsOutput = 0x0; + GLbitfield aluTemps = 0x0; + GLuint i; + + for (i = 0; i < prog->NumInstructions; i++) { + const struct prog_instruction *inst = prog->Instructions + i; + + if (is_texture_instruction(inst)) { + if (((inst->SrcReg[0].File == PROGRAM_TEMPORARY) && + (tempsOutput & (1 << inst->SrcReg[0].Index))) || + ((inst->Opcode != OPCODE_KIL) && + (inst->DstReg.File == PROGRAM_TEMPORARY) && + (aluTemps & (1 << inst->DstReg.Index)))) + { + indirections++; + tempsOutput = 0x0; + aluTemps = 0x0; + } + } + else { + GLuint j; + for (j = 0; j < 3; j++) { + if (inst->SrcReg[j].File == PROGRAM_TEMPORARY) + aluTemps |= (1 << inst->SrcReg[j].Index); + } + if (inst->DstReg.File == PROGRAM_TEMPORARY) + aluTemps |= (1 << inst->DstReg.Index); + } + + if ((inst->Opcode != OPCODE_KIL) && (inst->DstReg.File == PROGRAM_TEMPORARY)) + tempsOutput |= (1 << inst->DstReg.Index); + } + + prog->NumTexIndirections = indirections; +} + + +/** + * Count number of texture instructions in given program and update the + * program's NumTexInstructions field. + */ +void +_mesa_count_texture_instructions(struct gl_program *prog) +{ + GLuint i; + prog->NumTexInstructions = 0; + for (i = 0; i < prog->NumInstructions; i++) { + prog->NumTexInstructions += is_texture_instruction(prog->Instructions + i); + } +} + + +/** + * Scan/rewrite program to remove reads of custom (output) registers. + * The passed type has to be either PROGRAM_OUTPUT or PROGRAM_VARYING + * (for vertex shaders). + * In GLSL shaders, varying vars can be read and written. + * On some hardware, trying to read an output register causes trouble. + * So, rewrite the program to use a temporary register in this case. + */ +void +_mesa_remove_output_reads(struct gl_program *prog, gl_register_file type) +{ + GLuint i; + GLint outputMap[VERT_RESULT_MAX]; + GLuint numVaryingReads = 0; + GLboolean usedTemps[MAX_PROGRAM_TEMPS]; + GLuint firstTemp = 0; + + _mesa_find_used_registers(prog, PROGRAM_TEMPORARY, + usedTemps, MAX_PROGRAM_TEMPS); + + assert(type == PROGRAM_VARYING || type == PROGRAM_OUTPUT); + assert(prog->Target == GL_VERTEX_PROGRAM_ARB || type != PROGRAM_VARYING); + + for (i = 0; i < VERT_RESULT_MAX; i++) + outputMap[i] = -1; + + /* look for instructions which read from varying vars */ + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = prog->Instructions + i; + const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode); + GLuint j; + for (j = 0; j < numSrc; j++) { + if (inst->SrcReg[j].File == type) { + /* replace the read with a temp reg */ + const GLuint var = inst->SrcReg[j].Index; + if (outputMap[var] == -1) { + numVaryingReads++; + outputMap[var] = _mesa_find_free_register(usedTemps, + MAX_PROGRAM_TEMPS, + firstTemp); + firstTemp = outputMap[var] + 1; + } + inst->SrcReg[j].File = PROGRAM_TEMPORARY; + inst->SrcReg[j].Index = outputMap[var]; + } + } + } + + if (numVaryingReads == 0) + return; /* nothing to be done */ + + /* look for instructions which write to the varying vars identified above */ + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = prog->Instructions + i; + if (inst->DstReg.File == type && + outputMap[inst->DstReg.Index] >= 0) { + /* change inst to write to the temp reg, instead of the varying */ + inst->DstReg.File = PROGRAM_TEMPORARY; + inst->DstReg.Index = outputMap[inst->DstReg.Index]; + } + } + + /* insert new instructions to copy the temp vars to the varying vars */ + { + struct prog_instruction *inst; + GLint endPos, var; + + /* Look for END instruction and insert the new varying writes */ + endPos = -1; + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = prog->Instructions + i; + if (inst->Opcode == OPCODE_END) { + endPos = i; + _mesa_insert_instructions(prog, i, numVaryingReads); + break; + } + } + + assert(endPos >= 0); + + /* insert new MOV instructions here */ + inst = prog->Instructions + endPos; + for (var = 0; var < VERT_RESULT_MAX; var++) { + if (outputMap[var] >= 0) { + /* MOV VAR[var], TEMP[tmp]; */ + inst->Opcode = OPCODE_MOV; + inst->DstReg.File = type; + inst->DstReg.Index = var; + inst->SrcReg[0].File = PROGRAM_TEMPORARY; + inst->SrcReg[0].Index = outputMap[var]; + inst++; + } + } + } +} + + +/** + * Make the given fragment program into a "no-op" shader. + * Actually, just copy the incoming fragment color (or texcoord) + * to the output color. + * This is for debug/test purposes. + */ +void +_mesa_nop_fragment_program(struct gl_context *ctx, struct gl_fragment_program *prog) +{ + struct prog_instruction *inst; + GLuint inputAttr; + + inst = _mesa_alloc_instructions(2); + if (!inst) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "_mesa_nop_fragment_program"); + return; + } + + _mesa_init_instructions(inst, 2); + + inst[0].Opcode = OPCODE_MOV; + inst[0].DstReg.File = PROGRAM_OUTPUT; + inst[0].DstReg.Index = FRAG_RESULT_COLOR; + inst[0].SrcReg[0].File = PROGRAM_INPUT; + if (prog->Base.InputsRead & FRAG_BIT_COL0) + inputAttr = FRAG_ATTRIB_COL0; + else + inputAttr = FRAG_ATTRIB_TEX0; + inst[0].SrcReg[0].Index = inputAttr; + + inst[1].Opcode = OPCODE_END; + + _mesa_free_instructions(prog->Base.Instructions, + prog->Base.NumInstructions); + + prog->Base.Instructions = inst; + prog->Base.NumInstructions = 2; + prog->Base.InputsRead = 1 << inputAttr; + prog->Base.OutputsWritten = BITFIELD64_BIT(FRAG_RESULT_COLOR); +} + + +/** + * \sa _mesa_nop_fragment_program + * Replace the given vertex program with a "no-op" program that just + * transforms vertex position and emits color. + */ +void +_mesa_nop_vertex_program(struct gl_context *ctx, struct gl_vertex_program *prog) +{ + struct prog_instruction *inst; + GLuint inputAttr; + + /* + * Start with a simple vertex program that emits color. + */ + inst = _mesa_alloc_instructions(2); + if (!inst) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "_mesa_nop_vertex_program"); + return; + } + + _mesa_init_instructions(inst, 2); + + inst[0].Opcode = OPCODE_MOV; + inst[0].DstReg.File = PROGRAM_OUTPUT; + inst[0].DstReg.Index = VERT_RESULT_COL0; + inst[0].SrcReg[0].File = PROGRAM_INPUT; + if (prog->Base.InputsRead & VERT_BIT_COLOR0) + inputAttr = VERT_ATTRIB_COLOR0; + else + inputAttr = VERT_ATTRIB_TEX0; + inst[0].SrcReg[0].Index = inputAttr; + + inst[1].Opcode = OPCODE_END; + + _mesa_free_instructions(prog->Base.Instructions, + prog->Base.NumInstructions); + + prog->Base.Instructions = inst; + prog->Base.NumInstructions = 2; + prog->Base.InputsRead = 1 << inputAttr; + prog->Base.OutputsWritten = BITFIELD64_BIT(VERT_RESULT_COL0); + + /* + * Now insert code to do standard modelview/projection transformation. + */ + _mesa_insert_mvp_code(ctx, prog); +} diff --git a/mesalib/src/mesa/program/programopt.h b/mesalib/src/mesa/program/programopt.h index 99d7c3757..b9205823c 100644 --- a/mesalib/src/mesa/program/programopt.h +++ b/mesalib/src/mesa/program/programopt.h @@ -1,55 +1,55 @@ -/* - * Mesa 3-D graphics library - * Version: 6.5.3 - * - * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN - * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN - * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - - -#ifndef PROGRAMOPT_H -#define PROGRAMOPT_H 1 - -#include "main/mtypes.h" - -extern void -_mesa_insert_mvp_code(struct gl_context *ctx, struct gl_vertex_program *vprog); - -extern void -_mesa_append_fog_code(struct gl_context *ctx, - struct gl_fragment_program *fprog, GLenum fog_mode, - GLboolean saturate); - -extern void -_mesa_count_texture_indirections(struct gl_program *prog); - -extern void -_mesa_count_texture_instructions(struct gl_program *prog); - -extern void -_mesa_remove_output_reads(struct gl_program *prog, gl_register_file type); - -extern void -_mesa_nop_fragment_program(struct gl_context *ctx, struct gl_fragment_program *prog); - -extern void -_mesa_nop_vertex_program(struct gl_context *ctx, struct gl_vertex_program *prog); - - -#endif /* PROGRAMOPT_H */ +/* + * Mesa 3-D graphics library + * Version: 6.5.3 + * + * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + + +#ifndef PROGRAMOPT_H +#define PROGRAMOPT_H 1 + +#include "main/mtypes.h" + +extern void +_mesa_insert_mvp_code(struct gl_context *ctx, struct gl_vertex_program *vprog); + +extern void +_mesa_append_fog_code(struct gl_context *ctx, + struct gl_fragment_program *fprog, GLenum fog_mode, + GLboolean saturate); + +extern void +_mesa_count_texture_indirections(struct gl_program *prog); + +extern void +_mesa_count_texture_instructions(struct gl_program *prog); + +extern void +_mesa_remove_output_reads(struct gl_program *prog, gl_register_file type); + +extern void +_mesa_nop_fragment_program(struct gl_context *ctx, struct gl_fragment_program *prog); + +extern void +_mesa_nop_vertex_program(struct gl_context *ctx, struct gl_vertex_program *prog); + + +#endif /* PROGRAMOPT_H */ diff --git a/mesalib/src/mesa/program/register_allocate.c b/mesalib/src/mesa/program/register_allocate.c index a41a6ad6d..f5b5174fc 100644 --- a/mesalib/src/mesa/program/register_allocate.c +++ b/mesalib/src/mesa/program/register_allocate.c @@ -1,558 +1,558 @@ -/* - * Copyright © 2010 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS - * IN THE SOFTWARE. - * - * Authors: - * Eric Anholt - * - */ - -/** @file register_allocate.c - * - * Graph-coloring register allocator. - * - * The basic idea of graph coloring is to make a node in a graph for - * every thing that needs a register (color) number assigned, and make - * edges in the graph between nodes that interfere (can't be allocated - * to the same register at the same time). - * - * During the "simplify" process, any any node with fewer edges than - * there are registers means that that edge can get assigned a - * register regardless of what its neighbors choose, so that node is - * pushed on a stack and removed (with its edges) from the graph. - * That likely causes other nodes to become trivially colorable as well. - * - * Then during the "select" process, nodes are popped off of that - * stack, their edges restored, and assigned a color different from - * their neighbors. Because they were pushed on the stack only when - * they were trivially colorable, any color chosen won't interfere - * with the registers to be popped later. - * - * The downside to most graph coloring is that real hardware often has - * limitations, like registers that need to be allocated to a node in - * pairs, or aligned on some boundary. This implementation follows - * the paper "Retargetable Graph-Coloring Register Allocation for - * Irregular Architectures" by Johan Runeson and Sven-Olof Nyström. - * - * In this system, there are register classes each containing various - * registers, and registers may interfere with other registers. For - * example, one might have a class of base registers, and a class of - * aligned register pairs that would each interfere with their pair of - * the base registers. Each node has a register class it needs to be - * assigned to. Define p(B) to be the size of register class B, and - * q(B,C) to be the number of registers in B that the worst choice - * register in C could conflict with. Then, this system replaces the - * basic graph coloring test of "fewer edges from this node than there - * are registers" with "For this node of class B, the sum of q(B,C) - * for each neighbor node of class C is less than pB". - * - * A nice feature of the pq test is that q(B,C) can be computed once - * up front and stored in a 2-dimensional array, so that the cost of - * coloring a node is constant with the number of registers. We do - * this during ra_set_finalize(). - */ - -#include - -#include "main/imports.h" -#include "main/macros.h" -#include "main/mtypes.h" -#include "register_allocate.h" - -#define NO_REG ~0 - -struct ra_reg { - GLboolean *conflicts; - unsigned int *conflict_list; - unsigned int conflict_list_size; - unsigned int num_conflicts; -}; - -struct ra_regs { - struct ra_reg *regs; - unsigned int count; - - struct ra_class **classes; - unsigned int class_count; -}; - -struct ra_class { - GLboolean *regs; - - /** - * p(B) in Runeson/Nyström paper. - * - * This is "how many regs are in the set." - */ - unsigned int p; - - /** - * q(B,C) (indexed by C, B is this register class) in - * Runeson/Nyström paper. This is "how many registers of B could - * the worst choice register from C conflict with". - */ - unsigned int *q; -}; - -struct ra_node { - /** @{ - * - * List of which nodes this node interferes with. This should be - * symmetric with the other node. - */ - GLboolean *adjacency; - unsigned int *adjacency_list; - unsigned int adjacency_count; - /** @} */ - - unsigned int class; - - /* Register, if assigned, or NO_REG. */ - unsigned int reg; - - /** - * Set when the node is in the trivially colorable stack. When - * set, the adjacency to this node is ignored, to implement the - * "remove the edge from the graph" in simplification without - * having to actually modify the adjacency_list. - */ - GLboolean in_stack; - - /* For an implementation that needs register spilling, this is the - * approximate cost of spilling this node. - */ - float spill_cost; -}; - -struct ra_graph { - struct ra_regs *regs; - /** - * the variables that need register allocation. - */ - struct ra_node *nodes; - unsigned int count; /**< count of nodes. */ - - unsigned int *stack; - unsigned int stack_count; -}; - -struct ra_regs * -ra_alloc_reg_set(unsigned int count) -{ - unsigned int i; - struct ra_regs *regs; - - regs = rzalloc(NULL, struct ra_regs); - regs->count = count; - regs->regs = rzalloc_array(regs, struct ra_reg, count); - - for (i = 0; i < count; i++) { - regs->regs[i].conflicts = rzalloc_array(regs->regs, GLboolean, count); - regs->regs[i].conflicts[i] = GL_TRUE; - - regs->regs[i].conflict_list = ralloc_array(regs->regs, unsigned int, 4); - regs->regs[i].conflict_list_size = 4; - regs->regs[i].conflict_list[0] = i; - regs->regs[i].num_conflicts = 1; - } - - return regs; -} - -static void -ra_add_conflict_list(struct ra_regs *regs, unsigned int r1, unsigned int r2) -{ - struct ra_reg *reg1 = ®s->regs[r1]; - - if (reg1->conflict_list_size == reg1->num_conflicts) { - reg1->conflict_list_size *= 2; - reg1->conflict_list = reralloc(regs->regs, reg1->conflict_list, - unsigned int, reg1->conflict_list_size); - } - reg1->conflict_list[reg1->num_conflicts++] = r2; - reg1->conflicts[r2] = GL_TRUE; -} - -void -ra_add_reg_conflict(struct ra_regs *regs, unsigned int r1, unsigned int r2) -{ - if (!regs->regs[r1].conflicts[r2]) { - ra_add_conflict_list(regs, r1, r2); - ra_add_conflict_list(regs, r2, r1); - } -} - -/** - * Adds a conflict between base_reg and reg, and also between reg and - * anything that base_reg conflicts with. - * - * This can simplify code for setting up multiple register classes - * which are aggregates of some base hardware registers, compared to - * explicitly using ra_add_reg_conflict. - */ -void -ra_add_transitive_reg_conflict(struct ra_regs *regs, - unsigned int base_reg, unsigned int reg) -{ - int i; - - ra_add_reg_conflict(regs, reg, base_reg); - - for (i = 0; i < regs->regs[base_reg].num_conflicts; i++) { - ra_add_reg_conflict(regs, reg, regs->regs[base_reg].conflict_list[i]); - } -} - -unsigned int -ra_alloc_reg_class(struct ra_regs *regs) -{ - struct ra_class *class; - - regs->classes = reralloc(regs->regs, regs->classes, struct ra_class *, - regs->class_count + 1); - - class = rzalloc(regs, struct ra_class); - regs->classes[regs->class_count] = class; - - class->regs = rzalloc_array(class, GLboolean, regs->count); - - return regs->class_count++; -} - -void -ra_class_add_reg(struct ra_regs *regs, unsigned int c, unsigned int r) -{ - struct ra_class *class = regs->classes[c]; - - class->regs[r] = GL_TRUE; - class->p++; -} - -/** - * Must be called after all conflicts and register classes have been - * set up and before the register set is used for allocation. - */ -void -ra_set_finalize(struct ra_regs *regs) -{ - unsigned int b, c; - - for (b = 0; b < regs->class_count; b++) { - regs->classes[b]->q = ralloc_array(regs, unsigned int, regs->class_count); - } - - /* Compute, for each class B and C, how many regs of B an - * allocation to C could conflict with. - */ - for (b = 0; b < regs->class_count; b++) { - for (c = 0; c < regs->class_count; c++) { - unsigned int rc; - int max_conflicts = 0; - - for (rc = 0; rc < regs->count; rc++) { - int conflicts = 0; - int i; - - if (!regs->classes[c]->regs[rc]) - continue; - - for (i = 0; i < regs->regs[rc].num_conflicts; i++) { - unsigned int rb = regs->regs[rc].conflict_list[i]; - if (regs->classes[b]->regs[rb]) - conflicts++; - } - max_conflicts = MAX2(max_conflicts, conflicts); - } - regs->classes[b]->q[c] = max_conflicts; - } - } -} - -static void -ra_add_node_adjacency(struct ra_graph *g, unsigned int n1, unsigned int n2) -{ - g->nodes[n1].adjacency[n2] = GL_TRUE; - g->nodes[n1].adjacency_list[g->nodes[n1].adjacency_count] = n2; - g->nodes[n1].adjacency_count++; -} - -struct ra_graph * -ra_alloc_interference_graph(struct ra_regs *regs, unsigned int count) -{ - struct ra_graph *g; - unsigned int i; - - g = rzalloc(regs, struct ra_graph); - g->regs = regs; - g->nodes = rzalloc_array(g, struct ra_node, count); - g->count = count; - - g->stack = rzalloc_array(g, unsigned int, count); - - for (i = 0; i < count; i++) { - g->nodes[i].adjacency = rzalloc_array(g, GLboolean, count); - g->nodes[i].adjacency_list = ralloc_array(g, unsigned int, count); - g->nodes[i].adjacency_count = 0; - ra_add_node_adjacency(g, i, i); - g->nodes[i].reg = NO_REG; - } - - return g; -} - -void -ra_set_node_class(struct ra_graph *g, - unsigned int n, unsigned int class) -{ - g->nodes[n].class = class; -} - -void -ra_add_node_interference(struct ra_graph *g, - unsigned int n1, unsigned int n2) -{ - if (!g->nodes[n1].adjacency[n2]) { - ra_add_node_adjacency(g, n1, n2); - ra_add_node_adjacency(g, n2, n1); - } -} - -static GLboolean pq_test(struct ra_graph *g, unsigned int n) -{ - unsigned int j; - unsigned int q = 0; - int n_class = g->nodes[n].class; - - for (j = 0; j < g->nodes[n].adjacency_count; j++) { - unsigned int n2 = g->nodes[n].adjacency_list[j]; - unsigned int n2_class = g->nodes[n2].class; - - if (n != n2 && !g->nodes[n2].in_stack) { - q += g->regs->classes[n_class]->q[n2_class]; - } - } - - return q < g->regs->classes[n_class]->p; -} - -/** - * Simplifies the interference graph by pushing all - * trivially-colorable nodes into a stack of nodes to be colored, - * removing them from the graph, and rinsing and repeating. - * - * Returns GL_TRUE if all nodes were removed from the graph. GL_FALSE - * means that either spilling will be required, or optimistic coloring - * should be applied. - */ -GLboolean -ra_simplify(struct ra_graph *g) -{ - GLboolean progress = GL_TRUE; - int i; - - while (progress) { - progress = GL_FALSE; - - for (i = g->count - 1; i >= 0; i--) { - if (g->nodes[i].in_stack || g->nodes[i].reg != NO_REG) - continue; - - if (pq_test(g, i)) { - g->stack[g->stack_count] = i; - g->stack_count++; - g->nodes[i].in_stack = GL_TRUE; - progress = GL_TRUE; - } - } - } - - for (i = 0; i < g->count; i++) { - if (!g->nodes[i].in_stack) - return GL_FALSE; - } - - return GL_TRUE; -} - -/** - * Pops nodes from the stack back into the graph, coloring them with - * registers as they go. - * - * If all nodes were trivially colorable, then this must succeed. If - * not (optimistic coloring), then it may return GL_FALSE; - */ -GLboolean -ra_select(struct ra_graph *g) -{ - int i; - - while (g->stack_count != 0) { - unsigned int r; - int n = g->stack[g->stack_count - 1]; - struct ra_class *c = g->regs->classes[g->nodes[n].class]; - - /* Find the lowest-numbered reg which is not used by a member - * of the graph adjacent to us. - */ - for (r = 0; r < g->regs->count; r++) { - if (!c->regs[r]) - continue; - - /* Check if any of our neighbors conflict with this register choice. */ - for (i = 0; i < g->nodes[n].adjacency_count; i++) { - unsigned int n2 = g->nodes[n].adjacency_list[i]; - - if (!g->nodes[n2].in_stack && - g->regs->regs[r].conflicts[g->nodes[n2].reg]) { - break; - } - } - if (i == g->nodes[n].adjacency_count) - break; - } - if (r == g->regs->count) - return GL_FALSE; - - g->nodes[n].reg = r; - g->nodes[n].in_stack = GL_FALSE; - g->stack_count--; - } - - return GL_TRUE; -} - -/** - * Optimistic register coloring: Just push the remaining nodes - * on the stack. They'll be colored first in ra_select(), and - * if they succeed then the locally-colorable nodes are still - * locally-colorable and the rest of the register allocation - * will succeed. - */ -void -ra_optimistic_color(struct ra_graph *g) -{ - unsigned int i; - - for (i = 0; i < g->count; i++) { - if (g->nodes[i].in_stack || g->nodes[i].reg != NO_REG) - continue; - - g->stack[g->stack_count] = i; - g->stack_count++; - g->nodes[i].in_stack = GL_TRUE; - } -} - -GLboolean -ra_allocate_no_spills(struct ra_graph *g) -{ - if (!ra_simplify(g)) { - ra_optimistic_color(g); - } - return ra_select(g); -} - -unsigned int -ra_get_node_reg(struct ra_graph *g, unsigned int n) -{ - return g->nodes[n].reg; -} - -/** - * Forces a node to a specific register. This can be used to avoid - * creating a register class containing one node when handling data - * that must live in a fixed location and is known to not conflict - * with other forced register assignment (as is common with shader - * input data). These nodes do not end up in the stack during - * ra_simplify(), and thus at ra_select() time it is as if they were - * the first popped off the stack and assigned their fixed locations. - * - * Must be called before ra_simplify(). - */ -void -ra_set_node_reg(struct ra_graph *g, unsigned int n, unsigned int reg) -{ - g->nodes[n].reg = reg; - g->nodes[n].in_stack = GL_FALSE; -} - -static float -ra_get_spill_benefit(struct ra_graph *g, unsigned int n) -{ - int j; - float benefit = 0; - int n_class = g->nodes[n].class; - - /* Define the benefit of eliminating an interference between n, n2 - * through spilling as q(C, B) / p(C). This is similar to the - * "count number of edges" approach of traditional graph coloring, - * but takes classes into account. - */ - for (j = 0; j < g->nodes[n].adjacency_count; j++) { - unsigned int n2 = g->nodes[n].adjacency_list[j]; - if (n != n2) { - unsigned int n2_class = g->nodes[n2].class; - benefit += ((float)g->regs->classes[n_class]->q[n2_class] / - g->regs->classes[n_class]->p); - } - } - - return benefit; -} - -/** - * Returns a node number to be spilled according to the cost/benefit using - * the pq test, or -1 if there are no spillable nodes. - */ -int -ra_get_best_spill_node(struct ra_graph *g) -{ - unsigned int best_node = -1; - unsigned int best_benefit = 0.0; - unsigned int n; - - for (n = 0; n < g->count; n++) { - float cost = g->nodes[n].spill_cost; - float benefit; - - if (cost <= 0.0) - continue; - - benefit = ra_get_spill_benefit(g, n); - - if (benefit / cost > best_benefit) { - best_benefit = benefit / cost; - best_node = n; - } - } - - return best_node; -} - -/** - * Only nodes with a spill cost set (cost != 0.0) will be considered - * for register spilling. - */ -void -ra_set_node_spill_cost(struct ra_graph *g, unsigned int n, float cost) -{ - g->nodes[n].spill_cost = cost; -} +/* + * Copyright © 2010 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt + * + */ + +/** @file register_allocate.c + * + * Graph-coloring register allocator. + * + * The basic idea of graph coloring is to make a node in a graph for + * every thing that needs a register (color) number assigned, and make + * edges in the graph between nodes that interfere (can't be allocated + * to the same register at the same time). + * + * During the "simplify" process, any any node with fewer edges than + * there are registers means that that edge can get assigned a + * register regardless of what its neighbors choose, so that node is + * pushed on a stack and removed (with its edges) from the graph. + * That likely causes other nodes to become trivially colorable as well. + * + * Then during the "select" process, nodes are popped off of that + * stack, their edges restored, and assigned a color different from + * their neighbors. Because they were pushed on the stack only when + * they were trivially colorable, any color chosen won't interfere + * with the registers to be popped later. + * + * The downside to most graph coloring is that real hardware often has + * limitations, like registers that need to be allocated to a node in + * pairs, or aligned on some boundary. This implementation follows + * the paper "Retargetable Graph-Coloring Register Allocation for + * Irregular Architectures" by Johan Runeson and Sven-Olof Nyström. + * + * In this system, there are register classes each containing various + * registers, and registers may interfere with other registers. For + * example, one might have a class of base registers, and a class of + * aligned register pairs that would each interfere with their pair of + * the base registers. Each node has a register class it needs to be + * assigned to. Define p(B) to be the size of register class B, and + * q(B,C) to be the number of registers in B that the worst choice + * register in C could conflict with. Then, this system replaces the + * basic graph coloring test of "fewer edges from this node than there + * are registers" with "For this node of class B, the sum of q(B,C) + * for each neighbor node of class C is less than pB". + * + * A nice feature of the pq test is that q(B,C) can be computed once + * up front and stored in a 2-dimensional array, so that the cost of + * coloring a node is constant with the number of registers. We do + * this during ra_set_finalize(). + */ + +#include + +#include "main/imports.h" +#include "main/macros.h" +#include "main/mtypes.h" +#include "register_allocate.h" + +#define NO_REG ~0 + +struct ra_reg { + GLboolean *conflicts; + unsigned int *conflict_list; + unsigned int conflict_list_size; + unsigned int num_conflicts; +}; + +struct ra_regs { + struct ra_reg *regs; + unsigned int count; + + struct ra_class **classes; + unsigned int class_count; +}; + +struct ra_class { + GLboolean *regs; + + /** + * p(B) in Runeson/Nyström paper. + * + * This is "how many regs are in the set." + */ + unsigned int p; + + /** + * q(B,C) (indexed by C, B is this register class) in + * Runeson/Nyström paper. This is "how many registers of B could + * the worst choice register from C conflict with". + */ + unsigned int *q; +}; + +struct ra_node { + /** @{ + * + * List of which nodes this node interferes with. This should be + * symmetric with the other node. + */ + GLboolean *adjacency; + unsigned int *adjacency_list; + unsigned int adjacency_count; + /** @} */ + + unsigned int class; + + /* Register, if assigned, or NO_REG. */ + unsigned int reg; + + /** + * Set when the node is in the trivially colorable stack. When + * set, the adjacency to this node is ignored, to implement the + * "remove the edge from the graph" in simplification without + * having to actually modify the adjacency_list. + */ + GLboolean in_stack; + + /* For an implementation that needs register spilling, this is the + * approximate cost of spilling this node. + */ + float spill_cost; +}; + +struct ra_graph { + struct ra_regs *regs; + /** + * the variables that need register allocation. + */ + struct ra_node *nodes; + unsigned int count; /**< count of nodes. */ + + unsigned int *stack; + unsigned int stack_count; +}; + +struct ra_regs * +ra_alloc_reg_set(unsigned int count) +{ + unsigned int i; + struct ra_regs *regs; + + regs = rzalloc(NULL, struct ra_regs); + regs->count = count; + regs->regs = rzalloc_array(regs, struct ra_reg, count); + + for (i = 0; i < count; i++) { + regs->regs[i].conflicts = rzalloc_array(regs->regs, GLboolean, count); + regs->regs[i].conflicts[i] = GL_TRUE; + + regs->regs[i].conflict_list = ralloc_array(regs->regs, unsigned int, 4); + regs->regs[i].conflict_list_size = 4; + regs->regs[i].conflict_list[0] = i; + regs->regs[i].num_conflicts = 1; + } + + return regs; +} + +static void +ra_add_conflict_list(struct ra_regs *regs, unsigned int r1, unsigned int r2) +{ + struct ra_reg *reg1 = ®s->regs[r1]; + + if (reg1->conflict_list_size == reg1->num_conflicts) { + reg1->conflict_list_size *= 2; + reg1->conflict_list = reralloc(regs->regs, reg1->conflict_list, + unsigned int, reg1->conflict_list_size); + } + reg1->conflict_list[reg1->num_conflicts++] = r2; + reg1->conflicts[r2] = GL_TRUE; +} + +void +ra_add_reg_conflict(struct ra_regs *regs, unsigned int r1, unsigned int r2) +{ + if (!regs->regs[r1].conflicts[r2]) { + ra_add_conflict_list(regs, r1, r2); + ra_add_conflict_list(regs, r2, r1); + } +} + +/** + * Adds a conflict between base_reg and reg, and also between reg and + * anything that base_reg conflicts with. + * + * This can simplify code for setting up multiple register classes + * which are aggregates of some base hardware registers, compared to + * explicitly using ra_add_reg_conflict. + */ +void +ra_add_transitive_reg_conflict(struct ra_regs *regs, + unsigned int base_reg, unsigned int reg) +{ + int i; + + ra_add_reg_conflict(regs, reg, base_reg); + + for (i = 0; i < regs->regs[base_reg].num_conflicts; i++) { + ra_add_reg_conflict(regs, reg, regs->regs[base_reg].conflict_list[i]); + } +} + +unsigned int +ra_alloc_reg_class(struct ra_regs *regs) +{ + struct ra_class *class; + + regs->classes = reralloc(regs->regs, regs->classes, struct ra_class *, + regs->class_count + 1); + + class = rzalloc(regs, struct ra_class); + regs->classes[regs->class_count] = class; + + class->regs = rzalloc_array(class, GLboolean, regs->count); + + return regs->class_count++; +} + +void +ra_class_add_reg(struct ra_regs *regs, unsigned int c, unsigned int r) +{ + struct ra_class *class = regs->classes[c]; + + class->regs[r] = GL_TRUE; + class->p++; +} + +/** + * Must be called after all conflicts and register classes have been + * set up and before the register set is used for allocation. + */ +void +ra_set_finalize(struct ra_regs *regs) +{ + unsigned int b, c; + + for (b = 0; b < regs->class_count; b++) { + regs->classes[b]->q = ralloc_array(regs, unsigned int, regs->class_count); + } + + /* Compute, for each class B and C, how many regs of B an + * allocation to C could conflict with. + */ + for (b = 0; b < regs->class_count; b++) { + for (c = 0; c < regs->class_count; c++) { + unsigned int rc; + int max_conflicts = 0; + + for (rc = 0; rc < regs->count; rc++) { + int conflicts = 0; + int i; + + if (!regs->classes[c]->regs[rc]) + continue; + + for (i = 0; i < regs->regs[rc].num_conflicts; i++) { + unsigned int rb = regs->regs[rc].conflict_list[i]; + if (regs->classes[b]->regs[rb]) + conflicts++; + } + max_conflicts = MAX2(max_conflicts, conflicts); + } + regs->classes[b]->q[c] = max_conflicts; + } + } +} + +static void +ra_add_node_adjacency(struct ra_graph *g, unsigned int n1, unsigned int n2) +{ + g->nodes[n1].adjacency[n2] = GL_TRUE; + g->nodes[n1].adjacency_list[g->nodes[n1].adjacency_count] = n2; + g->nodes[n1].adjacency_count++; +} + +struct ra_graph * +ra_alloc_interference_graph(struct ra_regs *regs, unsigned int count) +{ + struct ra_graph *g; + unsigned int i; + + g = rzalloc(regs, struct ra_graph); + g->regs = regs; + g->nodes = rzalloc_array(g, struct ra_node, count); + g->count = count; + + g->stack = rzalloc_array(g, unsigned int, count); + + for (i = 0; i < count; i++) { + g->nodes[i].adjacency = rzalloc_array(g, GLboolean, count); + g->nodes[i].adjacency_list = ralloc_array(g, unsigned int, count); + g->nodes[i].adjacency_count = 0; + ra_add_node_adjacency(g, i, i); + g->nodes[i].reg = NO_REG; + } + + return g; +} + +void +ra_set_node_class(struct ra_graph *g, + unsigned int n, unsigned int class) +{ + g->nodes[n].class = class; +} + +void +ra_add_node_interference(struct ra_graph *g, + unsigned int n1, unsigned int n2) +{ + if (!g->nodes[n1].adjacency[n2]) { + ra_add_node_adjacency(g, n1, n2); + ra_add_node_adjacency(g, n2, n1); + } +} + +static GLboolean pq_test(struct ra_graph *g, unsigned int n) +{ + unsigned int j; + unsigned int q = 0; + int n_class = g->nodes[n].class; + + for (j = 0; j < g->nodes[n].adjacency_count; j++) { + unsigned int n2 = g->nodes[n].adjacency_list[j]; + unsigned int n2_class = g->nodes[n2].class; + + if (n != n2 && !g->nodes[n2].in_stack) { + q += g->regs->classes[n_class]->q[n2_class]; + } + } + + return q < g->regs->classes[n_class]->p; +} + +/** + * Simplifies the interference graph by pushing all + * trivially-colorable nodes into a stack of nodes to be colored, + * removing them from the graph, and rinsing and repeating. + * + * Returns GL_TRUE if all nodes were removed from the graph. GL_FALSE + * means that either spilling will be required, or optimistic coloring + * should be applied. + */ +GLboolean +ra_simplify(struct ra_graph *g) +{ + GLboolean progress = GL_TRUE; + int i; + + while (progress) { + progress = GL_FALSE; + + for (i = g->count - 1; i >= 0; i--) { + if (g->nodes[i].in_stack || g->nodes[i].reg != NO_REG) + continue; + + if (pq_test(g, i)) { + g->stack[g->stack_count] = i; + g->stack_count++; + g->nodes[i].in_stack = GL_TRUE; + progress = GL_TRUE; + } + } + } + + for (i = 0; i < g->count; i++) { + if (!g->nodes[i].in_stack) + return GL_FALSE; + } + + return GL_TRUE; +} + +/** + * Pops nodes from the stack back into the graph, coloring them with + * registers as they go. + * + * If all nodes were trivially colorable, then this must succeed. If + * not (optimistic coloring), then it may return GL_FALSE; + */ +GLboolean +ra_select(struct ra_graph *g) +{ + int i; + + while (g->stack_count != 0) { + unsigned int r; + int n = g->stack[g->stack_count - 1]; + struct ra_class *c = g->regs->classes[g->nodes[n].class]; + + /* Find the lowest-numbered reg which is not used by a member + * of the graph adjacent to us. + */ + for (r = 0; r < g->regs->count; r++) { + if (!c->regs[r]) + continue; + + /* Check if any of our neighbors conflict with this register choice. */ + for (i = 0; i < g->nodes[n].adjacency_count; i++) { + unsigned int n2 = g->nodes[n].adjacency_list[i]; + + if (!g->nodes[n2].in_stack && + g->regs->regs[r].conflicts[g->nodes[n2].reg]) { + break; + } + } + if (i == g->nodes[n].adjacency_count) + break; + } + if (r == g->regs->count) + return GL_FALSE; + + g->nodes[n].reg = r; + g->nodes[n].in_stack = GL_FALSE; + g->stack_count--; + } + + return GL_TRUE; +} + +/** + * Optimistic register coloring: Just push the remaining nodes + * on the stack. They'll be colored first in ra_select(), and + * if they succeed then the locally-colorable nodes are still + * locally-colorable and the rest of the register allocation + * will succeed. + */ +void +ra_optimistic_color(struct ra_graph *g) +{ + unsigned int i; + + for (i = 0; i < g->count; i++) { + if (g->nodes[i].in_stack || g->nodes[i].reg != NO_REG) + continue; + + g->stack[g->stack_count] = i; + g->stack_count++; + g->nodes[i].in_stack = GL_TRUE; + } +} + +GLboolean +ra_allocate_no_spills(struct ra_graph *g) +{ + if (!ra_simplify(g)) { + ra_optimistic_color(g); + } + return ra_select(g); +} + +unsigned int +ra_get_node_reg(struct ra_graph *g, unsigned int n) +{ + return g->nodes[n].reg; +} + +/** + * Forces a node to a specific register. This can be used to avoid + * creating a register class containing one node when handling data + * that must live in a fixed location and is known to not conflict + * with other forced register assignment (as is common with shader + * input data). These nodes do not end up in the stack during + * ra_simplify(), and thus at ra_select() time it is as if they were + * the first popped off the stack and assigned their fixed locations. + * + * Must be called before ra_simplify(). + */ +void +ra_set_node_reg(struct ra_graph *g, unsigned int n, unsigned int reg) +{ + g->nodes[n].reg = reg; + g->nodes[n].in_stack = GL_FALSE; +} + +static float +ra_get_spill_benefit(struct ra_graph *g, unsigned int n) +{ + int j; + float benefit = 0; + int n_class = g->nodes[n].class; + + /* Define the benefit of eliminating an interference between n, n2 + * through spilling as q(C, B) / p(C). This is similar to the + * "count number of edges" approach of traditional graph coloring, + * but takes classes into account. + */ + for (j = 0; j < g->nodes[n].adjacency_count; j++) { + unsigned int n2 = g->nodes[n].adjacency_list[j]; + if (n != n2) { + unsigned int n2_class = g->nodes[n2].class; + benefit += ((float)g->regs->classes[n_class]->q[n2_class] / + g->regs->classes[n_class]->p); + } + } + + return benefit; +} + +/** + * Returns a node number to be spilled according to the cost/benefit using + * the pq test, or -1 if there are no spillable nodes. + */ +int +ra_get_best_spill_node(struct ra_graph *g) +{ + unsigned int best_node = -1; + unsigned int best_benefit = 0.0; + unsigned int n; + + for (n = 0; n < g->count; n++) { + float cost = g->nodes[n].spill_cost; + float benefit; + + if (cost <= 0.0) + continue; + + benefit = ra_get_spill_benefit(g, n); + + if (benefit / cost > best_benefit) { + best_benefit = benefit / cost; + best_node = n; + } + } + + return best_node; +} + +/** + * Only nodes with a spill cost set (cost != 0.0) will be considered + * for register spilling. + */ +void +ra_set_node_spill_cost(struct ra_graph *g, unsigned int n, float cost) +{ + g->nodes[n].spill_cost = cost; +} diff --git a/mesalib/src/mesa/program/register_allocate.h b/mesalib/src/mesa/program/register_allocate.h index 2d5a630bf..ee2e58a47 100644 --- a/mesalib/src/mesa/program/register_allocate.h +++ b/mesalib/src/mesa/program/register_allocate.h @@ -1,74 +1,74 @@ -/* - * Copyright © 2010 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS - * IN THE SOFTWARE. - * - * Authors: - * Eric Anholt - * - */ - -struct ra_class; -struct ra_regs; - -/* @{ - * Register set setup. - * - * This should be done once at backend initializaion, as - * ra_set_finalize is O(r^2*c^2). The registers may be virtual - * registers, such as aligned register pairs that conflict with the - * two real registers from which they are composed. - */ -struct ra_regs *ra_alloc_reg_set(unsigned int count); -unsigned int ra_alloc_reg_class(struct ra_regs *regs); -void ra_add_reg_conflict(struct ra_regs *regs, - unsigned int r1, unsigned int r2); -void ra_add_transitive_reg_conflict(struct ra_regs *regs, - unsigned int base_reg, unsigned int reg); -void ra_class_add_reg(struct ra_regs *regs, unsigned int c, unsigned int reg); -void ra_set_finalize(struct ra_regs *regs); -/** @} */ - -/** @{ Interference graph setup. - * - * Each interference graph node is a virtual variable in the IL. It - * is up to the user to ra_set_node_class() for the virtual variable, - * and compute live ranges and ra_node_interfere() between conflicting - * live ranges. - */ -struct ra_graph *ra_alloc_interference_graph(struct ra_regs *regs, - unsigned int count); -void ra_set_node_class(struct ra_graph *g, unsigned int n, unsigned int c); -void ra_add_node_interference(struct ra_graph *g, - unsigned int n1, unsigned int n2); -/** @} */ - -/** @{ Graph-coloring register allocation */ -GLboolean ra_simplify(struct ra_graph *g); -void ra_optimistic_color(struct ra_graph *g); -GLboolean ra_select(struct ra_graph *g); -GLboolean ra_allocate_no_spills(struct ra_graph *g); - -unsigned int ra_get_node_reg(struct ra_graph *g, unsigned int n); -void ra_set_node_reg(struct ra_graph * g, unsigned int n, unsigned int reg); -void ra_set_node_spill_cost(struct ra_graph *g, unsigned int n, float cost); -int ra_get_best_spill_node(struct ra_graph *g); -/** @} */ - +/* + * Copyright © 2010 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt + * + */ + +struct ra_class; +struct ra_regs; + +/* @{ + * Register set setup. + * + * This should be done once at backend initializaion, as + * ra_set_finalize is O(r^2*c^2). The registers may be virtual + * registers, such as aligned register pairs that conflict with the + * two real registers from which they are composed. + */ +struct ra_regs *ra_alloc_reg_set(unsigned int count); +unsigned int ra_alloc_reg_class(struct ra_regs *regs); +void ra_add_reg_conflict(struct ra_regs *regs, + unsigned int r1, unsigned int r2); +void ra_add_transitive_reg_conflict(struct ra_regs *regs, + unsigned int base_reg, unsigned int reg); +void ra_class_add_reg(struct ra_regs *regs, unsigned int c, unsigned int reg); +void ra_set_finalize(struct ra_regs *regs); +/** @} */ + +/** @{ Interference graph setup. + * + * Each interference graph node is a virtual variable in the IL. It + * is up to the user to ra_set_node_class() for the virtual variable, + * and compute live ranges and ra_node_interfere() between conflicting + * live ranges. + */ +struct ra_graph *ra_alloc_interference_graph(struct ra_regs *regs, + unsigned int count); +void ra_set_node_class(struct ra_graph *g, unsigned int n, unsigned int c); +void ra_add_node_interference(struct ra_graph *g, + unsigned int n1, unsigned int n2); +/** @} */ + +/** @{ Graph-coloring register allocation */ +GLboolean ra_simplify(struct ra_graph *g); +void ra_optimistic_color(struct ra_graph *g); +GLboolean ra_select(struct ra_graph *g); +GLboolean ra_allocate_no_spills(struct ra_graph *g); + +unsigned int ra_get_node_reg(struct ra_graph *g, unsigned int n); +void ra_set_node_reg(struct ra_graph * g, unsigned int n, unsigned int reg); +void ra_set_node_spill_cost(struct ra_graph *g, unsigned int n, float cost); +int ra_get_best_spill_node(struct ra_graph *g); +/** @} */ + diff --git a/mesalib/src/mesa/program/sampler.cpp b/mesalib/src/mesa/program/sampler.cpp index 9a2616743..e8d34c670 100644 --- a/mesalib/src/mesa/program/sampler.cpp +++ b/mesalib/src/mesa/program/sampler.cpp @@ -1,137 +1,137 @@ -/* - * Copyright (C) 2005-2007 Brian Paul All Rights Reserved. - * Copyright (C) 2008 VMware, Inc. All Rights Reserved. - * Copyright © 2010 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER - * DEALINGS IN THE SOFTWARE. - */ - -#include "ir.h" -#include "glsl_types.h" -#include "ir_visitor.h" - -extern "C" { -#include "main/compiler.h" -#include "main/mtypes.h" -#include "program/prog_parameter.h" -} - -static void fail_link(struct gl_shader_program *prog, const char *fmt, ...) PRINTFLIKE(2, 3); - -static void fail_link(struct gl_shader_program *prog, const char *fmt, ...) -{ - va_list args; - va_start(args, fmt); - ralloc_vasprintf_append(&prog->InfoLog, fmt, args); - va_end(args); - - prog->LinkStatus = GL_FALSE; -} - -class get_sampler_name : public ir_hierarchical_visitor -{ -public: - get_sampler_name(ir_dereference *last, - struct gl_shader_program *shader_program) - { - this->mem_ctx = ralloc_context(NULL); - this->shader_program = shader_program; - this->name = NULL; - this->offset = 0; - this->last = last; - } - - ~get_sampler_name() - { - ralloc_free(this->mem_ctx); - } - - virtual ir_visitor_status visit(ir_dereference_variable *ir) - { - this->name = ir->var->name; - return visit_continue; - } - - virtual ir_visitor_status visit_leave(ir_dereference_record *ir) - { - this->name = ralloc_asprintf(mem_ctx, "%s.%s", name, ir->field); - return visit_continue; - } - - virtual ir_visitor_status visit_leave(ir_dereference_array *ir) - { - ir_constant *index = ir->array_index->as_constant(); - int i; - - if (index) { - i = index->value.i[0]; - } else { - /* GLSL 1.10 and 1.20 allowed variable sampler array indices, - * while GLSL 1.30 requires that the array indices be - * constant integer expressions. We don't expect any driver - * to actually work with a really variable array index, so - * all that would work would be an unrolled loop counter that ends - * up being constant above. - */ - ralloc_strcat(&shader_program->InfoLog, - "warning: Variable sampler array index unsupported.\n" - "This feature of the language was removed in GLSL 1.20 " - "and is unlikely to be supported for 1.10 in Mesa.\n"); - i = 0; - } - if (ir != last) { - this->name = ralloc_asprintf(mem_ctx, "%s[%d]", name, i); - } else { - offset = i; - } - return visit_continue; - } - - struct gl_shader_program *shader_program; - const char *name; - void *mem_ctx; - int offset; - ir_dereference *last; -}; - -extern "C" { -int -_mesa_get_sampler_uniform_value(class ir_dereference *sampler, - struct gl_shader_program *shader_program, - const struct gl_program *prog) -{ - get_sampler_name getname(sampler, shader_program); - - sampler->accept(&getname); - - GLint index = _mesa_lookup_parameter_index(prog->Parameters, -1, - getname.name); - - if (index < 0) { - fail_link(shader_program, - "failed to find sampler named %s.\n", getname.name); - return 0; - } - - index += getname.offset; - - return prog->Parameters->ParameterValues[index][0].f; -} -} +/* + * Copyright (C) 2005-2007 Brian Paul All Rights Reserved. + * Copyright (C) 2008 VMware, Inc. All Rights Reserved. + * Copyright © 2010 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +#include "ir.h" +#include "glsl_types.h" +#include "ir_visitor.h" + +extern "C" { +#include "main/compiler.h" +#include "main/mtypes.h" +#include "program/prog_parameter.h" +} + +static void fail_link(struct gl_shader_program *prog, const char *fmt, ...) PRINTFLIKE(2, 3); + +static void fail_link(struct gl_shader_program *prog, const char *fmt, ...) +{ + va_list args; + va_start(args, fmt); + ralloc_vasprintf_append(&prog->InfoLog, fmt, args); + va_end(args); + + prog->LinkStatus = GL_FALSE; +} + +class get_sampler_name : public ir_hierarchical_visitor +{ +public: + get_sampler_name(ir_dereference *last, + struct gl_shader_program *shader_program) + { + this->mem_ctx = ralloc_context(NULL); + this->shader_program = shader_program; + this->name = NULL; + this->offset = 0; + this->last = last; + } + + ~get_sampler_name() + { + ralloc_free(this->mem_ctx); + } + + virtual ir_visitor_status visit(ir_dereference_variable *ir) + { + this->name = ir->var->name; + return visit_continue; + } + + virtual ir_visitor_status visit_leave(ir_dereference_record *ir) + { + this->name = ralloc_asprintf(mem_ctx, "%s.%s", name, ir->field); + return visit_continue; + } + + virtual ir_visitor_status visit_leave(ir_dereference_array *ir) + { + ir_constant *index = ir->array_index->as_constant(); + int i; + + if (index) { + i = index->value.i[0]; + } else { + /* GLSL 1.10 and 1.20 allowed variable sampler array indices, + * while GLSL 1.30 requires that the array indices be + * constant integer expressions. We don't expect any driver + * to actually work with a really variable array index, so + * all that would work would be an unrolled loop counter that ends + * up being constant above. + */ + ralloc_strcat(&shader_program->InfoLog, + "warning: Variable sampler array index unsupported.\n" + "This feature of the language was removed in GLSL 1.20 " + "and is unlikely to be supported for 1.10 in Mesa.\n"); + i = 0; + } + if (ir != last) { + this->name = ralloc_asprintf(mem_ctx, "%s[%d]", name, i); + } else { + offset = i; + } + return visit_continue; + } + + struct gl_shader_program *shader_program; + const char *name; + void *mem_ctx; + int offset; + ir_dereference *last; +}; + +extern "C" { +int +_mesa_get_sampler_uniform_value(class ir_dereference *sampler, + struct gl_shader_program *shader_program, + const struct gl_program *prog) +{ + get_sampler_name getname(sampler, shader_program); + + sampler->accept(&getname); + + GLint index = _mesa_lookup_parameter_index(prog->Parameters, -1, + getname.name); + + if (index < 0) { + fail_link(shader_program, + "failed to find sampler named %s.\n", getname.name); + return 0; + } + + index += getname.offset; + + return prog->Parameters->ParameterValues[index][0].f; +} +} -- cgit v1.2.3