diff options
Diffstat (limited to 'mesalib/src')
-rw-r--r-- | mesalib/src/glsl/ast.h | 3 | ||||
-rw-r--r-- | mesalib/src/glsl/ast_function.cpp | 18 | ||||
-rw-r--r-- | mesalib/src/glsl/ast_to_hir.cpp | 44 | ||||
-rw-r--r-- | mesalib/src/glsl/glsl_parser_extras.h | 6 | ||||
-rw-r--r-- | mesalib/src/glsl/ir_function.cpp | 2 | ||||
-rw-r--r-- | mesalib/src/glsl/link_functions.cpp | 6 | ||||
-rw-r--r-- | mesalib/src/glsl/linker.cpp | 8 | ||||
-rw-r--r-- | mesalib/src/glsl/opt_dead_functions.cpp | 319 | ||||
-rw-r--r-- | mesalib/src/mesa/program/prog_parameter.h | 13 | ||||
-rw-r--r-- | mesalib/src/mesa/state_tracker/st_program.c | 326 | ||||
-rw-r--r-- | mesalib/src/mesa/state_tracker/st_program.h | 15 |
11 files changed, 382 insertions, 378 deletions
diff --git a/mesalib/src/glsl/ast.h b/mesalib/src/glsl/ast.h index 878f48b20..d1de22718 100644 --- a/mesalib/src/glsl/ast.h +++ b/mesalib/src/glsl/ast.h @@ -730,7 +730,6 @@ _mesa_ast_field_selection_to_hir(const ast_expression *expr, struct _mesa_glsl_parse_state *state); void -emit_function(_mesa_glsl_parse_state *state, exec_list *instructions, - ir_function *f); +emit_function(_mesa_glsl_parse_state *state, ir_function *f); #endif /* AST_H */ diff --git a/mesalib/src/glsl/ast_function.cpp b/mesalib/src/glsl/ast_function.cpp index 8bcf48dfd..5b6ed3bc8 100644 --- a/mesalib/src/glsl/ast_function.cpp +++ b/mesalib/src/glsl/ast_function.cpp @@ -125,7 +125,7 @@ match_function_by_name(exec_list *instructions, const char *name, if (f == NULL) { f = new(ctx) ir_function(name); state->symbols->add_global_function(f); - emit_function(state, instructions, f); + emit_function(state, f); } f->add_signature(sig->clone_prototype(f, NULL)); @@ -199,6 +199,20 @@ match_function_by_name(exec_list *instructions, const char *name, */ ir_call *call = new(ctx) ir_call(sig, actual_parameters); if (!sig->return_type->is_void()) { + /* If the function call is a constant expression, don't + * generate the instructions to call it; just generate an + * ir_constant representing the constant value. + * + * Function calls can only be constant expressions starting + * in GLSL 1.20. + */ + if (state->language_version >= 120) { + ir_constant *const_val = call->constant_expression_value(); + if (const_val) { + return const_val; + } + } + ir_variable *var; ir_dereference_variable *deref; @@ -211,8 +225,6 @@ match_function_by_name(exec_list *instructions, const char *name, deref = new(ctx) ir_dereference_variable(var); ir_assignment *assign = new(ctx) ir_assignment(deref, call, NULL); instructions->push_tail(assign); - if (state->language_version >= 120) - var->constant_value = call->constant_expression_value(); deref = new(ctx) ir_dereference_variable(var); return deref; diff --git a/mesalib/src/glsl/ast_to_hir.cpp b/mesalib/src/glsl/ast_to_hir.cpp index 7da146119..2025911ac 100644 --- a/mesalib/src/glsl/ast_to_hir.cpp +++ b/mesalib/src/glsl/ast_to_hir.cpp @@ -66,6 +66,8 @@ _mesa_ast_to_hir(exec_list *instructions, struct _mesa_glsl_parse_state *state) state->current_function = NULL; + state->toplevel_ir = instructions; + /* Section 4.2 of the GLSL 1.20 specification states: * "The built-in functions are scoped in a scope outside the global scope * users declare global variables in. That is, a shader's global scope, @@ -85,6 +87,8 @@ _mesa_ast_to_hir(exec_list *instructions, struct _mesa_glsl_parse_state *state) ast->hir(instructions, state); detect_recursion_unlinked(state, instructions); + + state->toplevel_ir = NULL; } @@ -1765,11 +1769,6 @@ process_array_type(YYLTYPE *loc, const glsl_type *base, ast_node *array_size, ir_rvalue *const ir = array_size->hir(& dummy_instructions, state); YYLTYPE loc = array_size->get_location(); - /* FINISHME: Verify that the grammar forbids side-effects in array - * FINISHME: sizes. i.e., 'vec4 [x = 12] data' - */ - assert(dummy_instructions.is_empty()); - if (ir != NULL) { if (!ir->type->is_integer()) { _mesa_glsl_error(& loc, state, "array size must be integer type"); @@ -1786,6 +1785,14 @@ process_array_type(YYLTYPE *loc, const glsl_type *base, ast_node *array_size, } else { assert(size->type == ir->type); length = size->value.u[0]; + + /* If the array size is const (and we've verified that + * it is) then no instructions should have been emitted + * when we converted it to HIR. If they were emitted, + * then either the array size isn't const after all, or + * we are emitting unnecessary instructions. + */ + assert(dummy_instructions.is_empty()); } } } @@ -2926,23 +2933,16 @@ ast_parameter_declarator::parameters_to_hir(exec_list *ast_parameters, void -emit_function(_mesa_glsl_parse_state *state, exec_list *instructions, - ir_function *f) +emit_function(_mesa_glsl_parse_state *state, ir_function *f) { - /* Emit the new function header */ - if (state->current_function == NULL) { - instructions->push_tail(f); - } else { - /* IR invariants disallow function declarations or definitions nested - * within other function definitions. Insert the new ir_function - * block in the instruction sequence before the ir_function block - * containing the current ir_function_signature. - */ - ir_function *const curr = - const_cast<ir_function *>(state->current_function->function()); - - curr->insert_before(f); - } + /* IR invariants disallow function declarations or definitions + * nested within other function definitions. But there is no + * requirement about the relative order of function declarations + * and definitions with respect to one another. So simply insert + * the new ir_function block at the end of the toplevel instruction + * list. + */ + state->toplevel_ir->push_tail(f); } @@ -3069,7 +3069,7 @@ ast_function::hir(exec_list *instructions, return NULL; } - emit_function(state, instructions, f); + emit_function(state, f); } /* Verify the return type of main() */ diff --git a/mesalib/src/glsl/glsl_parser_extras.h b/mesalib/src/glsl/glsl_parser_extras.h index 2f4d3cba7..fc392da5b 100644 --- a/mesalib/src/glsl/glsl_parser_extras.h +++ b/mesalib/src/glsl/glsl_parser_extras.h @@ -129,6 +129,12 @@ struct _mesa_glsl_parse_state { */ class ir_function_signature *current_function; + /** + * During AST to IR conversion, pointer to the toplevel IR + * instruction list being generated. + */ + exec_list *toplevel_ir; + /** Have we found a return statement in this function? */ bool found_return; diff --git a/mesalib/src/glsl/ir_function.cpp b/mesalib/src/glsl/ir_function.cpp index 2a4de5b0d..51d32b46f 100644 --- a/mesalib/src/glsl/ir_function.cpp +++ b/mesalib/src/glsl/ir_function.cpp @@ -27,7 +27,7 @@ typedef enum { PARAMETER_LIST_NO_MATCH, PARAMETER_LIST_EXACT_MATCH, - PARAMETER_LIST_INEXACT_MATCH, /*< Match requires implicit conversion. */ + PARAMETER_LIST_INEXACT_MATCH /*< Match requires implicit conversion. */ } parameter_list_match_t; /** diff --git a/mesalib/src/glsl/link_functions.cpp b/mesalib/src/glsl/link_functions.cpp index d40f771e3..acee32712 100644 --- a/mesalib/src/glsl/link_functions.cpp +++ b/mesalib/src/glsl/link_functions.cpp @@ -104,10 +104,12 @@ public: if (f == NULL) { f = new(linked) ir_function(name); - /* Add the new function to the linked IR. + /* Add the new function to the linked IR. Put it at the end + * so that it comes after any global variable declarations + * that it refers to. */ linked->symbols->add_function(f); - linked->ir->push_head(f); + linked->ir->push_tail(f); } ir_function_signature *linked_sig = diff --git a/mesalib/src/glsl/linker.cpp b/mesalib/src/glsl/linker.cpp index 19eb9b5ff..b54ef4108 100644 --- a/mesalib/src/glsl/linker.cpp +++ b/mesalib/src/glsl/linker.cpp @@ -922,6 +922,14 @@ link_intrastage_shaders(void *mem_ctx, free(linking_shaders); +#ifdef DEBUG + /* At this point linked should contain all of the linked IR, so + * validate it to make sure nothing went wrong. + */ + if (linked) + validate_ir_tree(linked->ir); +#endif + /* Make a pass over all variable declarations to ensure that arrays with * unspecified sizes have a size specified. The size is inferred from the * max_array_access field. diff --git a/mesalib/src/glsl/opt_dead_functions.cpp b/mesalib/src/glsl/opt_dead_functions.cpp index 3c3eb8904..51c77e3b9 100644 --- a/mesalib/src/glsl/opt_dead_functions.cpp +++ b/mesalib/src/glsl/opt_dead_functions.cpp @@ -1,164 +1,155 @@ -/*
- * 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 opt_dead_functions.cpp
- *
- * Eliminates unused functions from the linked program.
- */
-
-#include "ir.h"
-#include "ir_visitor.h"
-#include "ir_expression_flattening.h"
-#include "glsl_types.h"
-
-class signature_entry : public exec_node
-{
-public:
- signature_entry(ir_function_signature *sig)
- {
- this->signature = sig;
- this->used = false;
- }
-
- ir_function_signature *signature;
- bool used;
-};
-
-class ir_dead_functions_visitor : public ir_hierarchical_visitor {
-public:
- ir_dead_functions_visitor()
- {
- this->mem_ctx = ralloc_context(NULL);
- this->seen_another_function_signature = false;
- }
-
- ~ir_dead_functions_visitor()
- {
- ralloc_free(this->mem_ctx);
- }
-
- virtual ir_visitor_status visit_enter(ir_function_signature *);
- virtual ir_visitor_status visit_enter(ir_call *);
-
- signature_entry *get_signature_entry(ir_function_signature *var);
-
- bool (*predicate)(ir_instruction *ir);
-
- bool seen_another_function_signature;
-
- /* List of signature_entry */
- exec_list signature_list;
- void *mem_ctx;
-};
-
-
-signature_entry *
-ir_dead_functions_visitor::get_signature_entry(ir_function_signature *sig)
-{
- foreach_iter(exec_list_iterator, iter, this->signature_list) {
- signature_entry *entry = (signature_entry *)iter.get();
- if (entry->signature == sig)
- return entry;
- }
-
- signature_entry *entry = new(mem_ctx) signature_entry(sig);
- this->signature_list.push_tail(entry);
- return entry;
-}
-
-
-ir_visitor_status
-ir_dead_functions_visitor::visit_enter(ir_function_signature *ir)
-{
- signature_entry *entry = this->get_signature_entry(ir);
-
- if (strcmp(ir->function_name(), "main") == 0) {
- entry->used = true;
- }
-
- /* If this is the first signature to look at, no need to descend to see
- * if it has calls to another function signature.
- */
- if (!this->seen_another_function_signature) {
- this->seen_another_function_signature = true;
- return visit_continue_with_parent;
- }
-
- return visit_continue;
-}
-
-
-ir_visitor_status
-ir_dead_functions_visitor::visit_enter(ir_call *ir)
-{
- signature_entry *entry = this->get_signature_entry(ir->get_callee());
-
- entry->used = true;
-
- return visit_continue;
-}
-
-bool
-do_dead_functions(exec_list *instructions)
-{
- ir_dead_functions_visitor v;
- bool progress = false;
-
- visit_list_elements(&v, instructions);
-
- /* Now that we've figured out which function signatures are used, remove
- * the unused ones, and remove function definitions that have no more
- * signatures.
- */
- foreach_iter(exec_list_iterator, iter, v.signature_list) {
- signature_entry *entry = (signature_entry *)iter.get();
-
- if (!entry->used) {
- entry->signature->remove();
- delete entry->signature;
- progress = true;
- }
- delete(entry);
- }
-
- /* We don't just do this above when we nuked a signature because of
- * const pointers.
- */
- foreach_iter(exec_list_iterator, iter, *instructions) {
- ir_instruction *ir = (ir_instruction *)iter.get();
- ir_function *func = ir->as_function();
-
- if (func && func->signatures.is_empty()) {
- /* At this point (post-linking), the symbol table is no
- * longer in use, so not removing the function from the
- * symbol table should be OK.
- */
- func->remove();
- delete func;
- progress = true;
- }
- }
-
- return progress;
-}
+/* + * 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 opt_dead_functions.cpp + * + * Eliminates unused functions from the linked program. + */ + +#include "ir.h" +#include "ir_visitor.h" +#include "ir_expression_flattening.h" +#include "glsl_types.h" + +class signature_entry : public exec_node +{ +public: + signature_entry(ir_function_signature *sig) + { + this->signature = sig; + this->used = false; + } + + ir_function_signature *signature; + bool used; +}; + +class ir_dead_functions_visitor : public ir_hierarchical_visitor { +public: + ir_dead_functions_visitor() + { + this->mem_ctx = ralloc_context(NULL); + } + + ~ir_dead_functions_visitor() + { + ralloc_free(this->mem_ctx); + } + + virtual ir_visitor_status visit_enter(ir_function_signature *); + virtual ir_visitor_status visit_enter(ir_call *); + + signature_entry *get_signature_entry(ir_function_signature *var); + + bool (*predicate)(ir_instruction *ir); + + /* List of signature_entry */ + exec_list signature_list; + void *mem_ctx; +}; + + +signature_entry * +ir_dead_functions_visitor::get_signature_entry(ir_function_signature *sig) +{ + foreach_iter(exec_list_iterator, iter, this->signature_list) { + signature_entry *entry = (signature_entry *)iter.get(); + if (entry->signature == sig) + return entry; + } + + signature_entry *entry = new(mem_ctx) signature_entry(sig); + this->signature_list.push_tail(entry); + return entry; +} + + +ir_visitor_status +ir_dead_functions_visitor::visit_enter(ir_function_signature *ir) +{ + signature_entry *entry = this->get_signature_entry(ir); + + if (strcmp(ir->function_name(), "main") == 0) { + entry->used = true; + } + + + + return visit_continue; +} + + +ir_visitor_status +ir_dead_functions_visitor::visit_enter(ir_call *ir) +{ + signature_entry *entry = this->get_signature_entry(ir->get_callee()); + + entry->used = true; + + return visit_continue; +} + +bool +do_dead_functions(exec_list *instructions) +{ + ir_dead_functions_visitor v; + bool progress = false; + + visit_list_elements(&v, instructions); + + /* Now that we've figured out which function signatures are used, remove + * the unused ones, and remove function definitions that have no more + * signatures. + */ + foreach_iter(exec_list_iterator, iter, v.signature_list) { + signature_entry *entry = (signature_entry *)iter.get(); + + if (!entry->used) { + entry->signature->remove(); + delete entry->signature; + progress = true; + } + delete(entry); + } + + /* We don't just do this above when we nuked a signature because of + * const pointers. + */ + foreach_iter(exec_list_iterator, iter, *instructions) { + ir_instruction *ir = (ir_instruction *)iter.get(); + ir_function *func = ir->as_function(); + + if (func && func->signatures.is_empty()) { + /* At this point (post-linking), the symbol table is no + * longer in use, so not removing the function from the + * symbol table should be OK. + */ + func->remove(); + delete func; + progress = true; + } + } + + return progress; +} diff --git a/mesalib/src/mesa/program/prog_parameter.h b/mesalib/src/mesa/program/prog_parameter.h index f858cf0fa..1a5ed3439 100644 --- a/mesalib/src/mesa/program/prog_parameter.h +++ b/mesalib/src/mesa/program/prog_parameter.h @@ -46,16 +46,19 @@ #define PROG_PARAM_BIT_CYL_WRAP 0x10 /**< XXX gallium debug */ /*@}*/ + /** * Actual data for constant values of parameters. */ -typedef union gl_constant_value { - GLfloat f; - GLboolean b; - GLint i; - GLuint u; +typedef union gl_constant_value +{ + GLfloat f; + GLboolean b; + GLint i; + GLuint u; } gl_constant_value; + /** * Program parameter. * Used by shaders/programs for uniforms, constants, varying vars, etc. diff --git a/mesalib/src/mesa/state_tracker/st_program.c b/mesalib/src/mesa/state_tracker/st_program.c index ca01d2e19..a4f47edfc 100644 --- a/mesalib/src/mesa/state_tracker/st_program.c +++ b/mesalib/src/mesa/state_tracker/st_program.c @@ -416,151 +416,6 @@ st_get_vp_variant(struct st_context *st, return vpv; } -/** - * Translate Mesa fragment shader attributes to TGSI attributes. - * \return GL_TRUE if color output should be written to all render targets, - * GL_FALSE if not - */ -GLboolean -st_prepare_fragment_program(struct gl_context *ctx, - struct st_fragment_program *stfp) -{ - GLuint attr; - const GLbitfield inputsRead = stfp->Base.Base.InputsRead; - GLboolean write_all = GL_FALSE; - - /* - * Convert Mesa program inputs to TGSI input register semantics. - */ - for (attr = 0; attr < FRAG_ATTRIB_MAX; attr++) { - if (inputsRead & (1 << attr)) { - const GLuint slot = stfp->num_inputs++; - - stfp->input_to_index[attr] = slot; - - switch (attr) { - case FRAG_ATTRIB_WPOS: - stfp->input_semantic_name[slot] = TGSI_SEMANTIC_POSITION; - stfp->input_semantic_index[slot] = 0; - stfp->interp_mode[slot] = TGSI_INTERPOLATE_LINEAR; - break; - case FRAG_ATTRIB_COL0: - stfp->input_semantic_name[slot] = TGSI_SEMANTIC_COLOR; - stfp->input_semantic_index[slot] = 0; - stfp->interp_mode[slot] = TGSI_INTERPOLATE_LINEAR; - break; - case FRAG_ATTRIB_COL1: - stfp->input_semantic_name[slot] = TGSI_SEMANTIC_COLOR; - stfp->input_semantic_index[slot] = 1; - stfp->interp_mode[slot] = TGSI_INTERPOLATE_LINEAR; - break; - case FRAG_ATTRIB_FOGC: - stfp->input_semantic_name[slot] = TGSI_SEMANTIC_FOG; - stfp->input_semantic_index[slot] = 0; - stfp->interp_mode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; - break; - case FRAG_ATTRIB_FACE: - stfp->input_semantic_name[slot] = TGSI_SEMANTIC_FACE; - stfp->input_semantic_index[slot] = 0; - stfp->interp_mode[slot] = TGSI_INTERPOLATE_CONSTANT; - break; - /* In most cases, there is nothing special about these - * inputs, so adopt a convention to use the generic - * semantic name and the mesa FRAG_ATTRIB_ number as the - * index. - * - * All that is required is that the vertex shader labels - * its own outputs similarly, and that the vertex shader - * generates at least every output required by the - * fragment shader plus fixed-function hardware (such as - * BFC). - * - * There is no requirement that semantic indexes start at - * zero or be restricted to a particular range -- nobody - * should be building tables based on semantic index. - */ - case FRAG_ATTRIB_PNTC: - case FRAG_ATTRIB_TEX0: - case FRAG_ATTRIB_TEX1: - case FRAG_ATTRIB_TEX2: - case FRAG_ATTRIB_TEX3: - case FRAG_ATTRIB_TEX4: - case FRAG_ATTRIB_TEX5: - case FRAG_ATTRIB_TEX6: - case FRAG_ATTRIB_TEX7: - case FRAG_ATTRIB_VAR0: - default: - /* Actually, let's try and zero-base this just for - * readability of the generated TGSI. - */ - assert(attr >= FRAG_ATTRIB_TEX0); - stfp->input_semantic_index[slot] = (attr - FRAG_ATTRIB_TEX0); - stfp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC; - if (attr == FRAG_ATTRIB_PNTC) - stfp->interp_mode[slot] = TGSI_INTERPOLATE_LINEAR; - else - stfp->interp_mode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; - break; - } - } - else { - stfp->input_to_index[attr] = -1; - } - } - - /* - * Semantics and mapping for outputs - */ - { - uint numColors = 0; - GLbitfield64 outputsWritten = stfp->Base.Base.OutputsWritten; - - /* if z is written, emit that first */ - if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) { - stfp->output_semantic_name[stfp->num_outputs] = TGSI_SEMANTIC_POSITION; - stfp->output_semantic_index[stfp->num_outputs] = 0; - stfp->result_to_output[FRAG_RESULT_DEPTH] = stfp->num_outputs; - stfp->num_outputs++; - outputsWritten &= ~(1 << FRAG_RESULT_DEPTH); - } - - if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) { - stfp->output_semantic_name[stfp->num_outputs] = TGSI_SEMANTIC_STENCIL; - stfp->output_semantic_index[stfp->num_outputs] = 0; - stfp->result_to_output[FRAG_RESULT_STENCIL] = stfp->num_outputs; - stfp->num_outputs++; - outputsWritten &= ~(1 << FRAG_RESULT_STENCIL); - } - - /* handle remaning outputs (color) */ - for (attr = 0; attr < FRAG_RESULT_MAX; attr++) { - if (outputsWritten & BITFIELD64_BIT(attr)) { - switch (attr) { - case FRAG_RESULT_DEPTH: - case FRAG_RESULT_STENCIL: - /* handled above */ - assert(0); - break; - case FRAG_RESULT_COLOR: - write_all = GL_TRUE; /* fallthrough */ - default: - assert(attr == FRAG_RESULT_COLOR || - (FRAG_RESULT_DATA0 <= attr && attr < FRAG_RESULT_MAX)); - stfp->output_semantic_name[stfp->num_outputs] = TGSI_SEMANTIC_COLOR; - stfp->output_semantic_index[stfp->num_outputs] = numColors; - stfp->result_to_output[attr] = stfp->num_outputs; - numColors++; - break; - } - - stfp->num_outputs++; - } - } - } - - return write_all; -} - /** * Translate a Mesa fragment shader into a TGSI shader using extra info in @@ -613,12 +468,155 @@ st_translate_fragment_program(struct st_context *st, if (!stfp->tgsi.tokens) { /* need to translate Mesa instructions to TGSI now */ + GLuint outputMapping[FRAG_RESULT_MAX]; + GLuint inputMapping[FRAG_ATTRIB_MAX]; + GLuint interpMode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */ + GLuint attr; + const GLbitfield inputsRead = stfp->Base.Base.InputsRead; struct ureg_program *ureg; - GLboolean write_all = st_prepare_fragment_program(st->ctx, stfp); + + GLboolean write_all = GL_FALSE; + + ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS]; + ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS]; + uint fs_num_inputs = 0; + + ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; + ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS]; + uint fs_num_outputs = 0; if (!stfp->glsl_to_tgsi) _mesa_remove_output_reads(&stfp->Base.Base, PROGRAM_OUTPUT); + /* + * Convert Mesa program inputs to TGSI input register semantics. + */ + for (attr = 0; attr < FRAG_ATTRIB_MAX; attr++) { + if (inputsRead & (1 << attr)) { + const GLuint slot = fs_num_inputs++; + + inputMapping[attr] = slot; + + switch (attr) { + case FRAG_ATTRIB_WPOS: + input_semantic_name[slot] = TGSI_SEMANTIC_POSITION; + input_semantic_index[slot] = 0; + interpMode[slot] = TGSI_INTERPOLATE_LINEAR; + break; + case FRAG_ATTRIB_COL0: + input_semantic_name[slot] = TGSI_SEMANTIC_COLOR; + input_semantic_index[slot] = 0; + interpMode[slot] = TGSI_INTERPOLATE_LINEAR; + break; + case FRAG_ATTRIB_COL1: + input_semantic_name[slot] = TGSI_SEMANTIC_COLOR; + input_semantic_index[slot] = 1; + interpMode[slot] = TGSI_INTERPOLATE_LINEAR; + break; + case FRAG_ATTRIB_FOGC: + input_semantic_name[slot] = TGSI_SEMANTIC_FOG; + input_semantic_index[slot] = 0; + interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; + break; + case FRAG_ATTRIB_FACE: + input_semantic_name[slot] = TGSI_SEMANTIC_FACE; + input_semantic_index[slot] = 0; + interpMode[slot] = TGSI_INTERPOLATE_CONSTANT; + break; + /* In most cases, there is nothing special about these + * inputs, so adopt a convention to use the generic + * semantic name and the mesa FRAG_ATTRIB_ number as the + * index. + * + * All that is required is that the vertex shader labels + * its own outputs similarly, and that the vertex shader + * generates at least every output required by the + * fragment shader plus fixed-function hardware (such as + * BFC). + * + * There is no requirement that semantic indexes start at + * zero or be restricted to a particular range -- nobody + * should be building tables based on semantic index. + */ + case FRAG_ATTRIB_PNTC: + case FRAG_ATTRIB_TEX0: + case FRAG_ATTRIB_TEX1: + case FRAG_ATTRIB_TEX2: + case FRAG_ATTRIB_TEX3: + case FRAG_ATTRIB_TEX4: + case FRAG_ATTRIB_TEX5: + case FRAG_ATTRIB_TEX6: + case FRAG_ATTRIB_TEX7: + case FRAG_ATTRIB_VAR0: + default: + /* Actually, let's try and zero-base this just for + * readability of the generated TGSI. + */ + assert(attr >= FRAG_ATTRIB_TEX0); + input_semantic_index[slot] = (attr - FRAG_ATTRIB_TEX0); + input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC; + if (attr == FRAG_ATTRIB_PNTC) + interpMode[slot] = TGSI_INTERPOLATE_LINEAR; + else + interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; + break; + } + } + else { + inputMapping[attr] = -1; + } + } + + /* + * Semantics and mapping for outputs + */ + { + uint numColors = 0; + GLbitfield64 outputsWritten = stfp->Base.Base.OutputsWritten; + + /* if z is written, emit that first */ + if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) { + fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION; + fs_output_semantic_index[fs_num_outputs] = 0; + outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs; + fs_num_outputs++; + outputsWritten &= ~(1 << FRAG_RESULT_DEPTH); + } + + if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) { + fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL; + fs_output_semantic_index[fs_num_outputs] = 0; + outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs; + fs_num_outputs++; + outputsWritten &= ~(1 << FRAG_RESULT_STENCIL); + } + + /* handle remaning outputs (color) */ + for (attr = 0; attr < FRAG_RESULT_MAX; attr++) { + if (outputsWritten & BITFIELD64_BIT(attr)) { + switch (attr) { + case FRAG_RESULT_DEPTH: + case FRAG_RESULT_STENCIL: + /* handled above */ + assert(0); + break; + case FRAG_RESULT_COLOR: + write_all = GL_TRUE; /* fallthrough */ + default: + assert(attr == FRAG_RESULT_COLOR || + (FRAG_RESULT_DATA0 <= attr && attr < FRAG_RESULT_MAX)); + fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR; + fs_output_semantic_index[fs_num_outputs] = numColors; + outputMapping[attr] = fs_num_outputs; + numColors++; + break; + } + + fs_num_outputs++; + } + } + } + ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT ); if (ureg == NULL) return NULL; @@ -638,32 +636,32 @@ st_translate_fragment_program(struct st_context *st, stfp->glsl_to_tgsi, &stfp->Base.Base, /* inputs */ - stfp->num_inputs, - stfp->input_to_index, - stfp->input_semantic_name, - stfp->input_semantic_index, - stfp->interp_mode, + fs_num_inputs, + inputMapping, + input_semantic_name, + input_semantic_index, + interpMode, /* outputs */ - stfp->num_outputs, - stfp->result_to_output, - stfp->output_semantic_name, - stfp->output_semantic_index, FALSE ); + fs_num_outputs, + outputMapping, + fs_output_semantic_name, + fs_output_semantic_index, FALSE ); else st_translate_mesa_program(st->ctx, TGSI_PROCESSOR_FRAGMENT, ureg, &stfp->Base.Base, /* inputs */ - stfp->num_inputs, - stfp->input_to_index, - stfp->input_semantic_name, - stfp->input_semantic_index, - stfp->interp_mode, + fs_num_inputs, + inputMapping, + input_semantic_name, + input_semantic_index, + interpMode, /* outputs */ - stfp->num_outputs, - stfp->result_to_output, - stfp->output_semantic_name, - stfp->output_semantic_index, FALSE ); + fs_num_outputs, + outputMapping, + fs_output_semantic_name, + fs_output_semantic_index, FALSE ); stfp->tgsi.tokens = ureg_get_tokens( ureg, NULL ); ureg_destroy( ureg ); diff --git a/mesalib/src/mesa/state_tracker/st_program.h b/mesalib/src/mesa/state_tracker/st_program.h index 67723de6d..699b6e8cc 100644 --- a/mesalib/src/mesa/state_tracker/st_program.h +++ b/mesalib/src/mesa/state_tracker/st_program.h @@ -85,21 +85,6 @@ struct st_fragment_program { struct gl_fragment_program Base; struct glsl_to_tgsi_visitor* glsl_to_tgsi; - - /** maps a Mesa FRAG_ATTRIB_x to a packed TGSI input index */ - GLuint input_to_index[FRAG_ATTRIB_MAX]; - /** maps a TGSI input index back to a Mesa FRAG_ATTRIB_x */ - GLuint index_to_input[PIPE_MAX_SHADER_INPUTS]; - ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS]; - ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS]; - GLuint num_inputs; - GLuint interp_mode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */ - - /** Maps FRAG_RESULT_x to slot */ - GLuint result_to_output[FRAG_RESULT_MAX]; - ubyte output_semantic_name[FRAG_RESULT_MAX]; - ubyte output_semantic_index[FRAG_RESULT_MAX]; - GLuint num_outputs; struct pipe_shader_state tgsi; |