From 1e1e2c35b405bbb537a6bbe35d9f0831111e272f Mon Sep 17 00:00:00 2001 From: marha Date: Wed, 26 Jan 2011 07:24:15 +0000 Subject: mesa xkeyboard-config git update 26 jan 2011 --- mesalib/configure.ac | 3 +- mesalib/src/glsl/ir_reader.cpp | 1995 +++---- mesalib/src/glsl/link_functions.cpp | 528 +- mesalib/src/glsl/linker.cpp | 3348 ++++++----- mesalib/src/mesa/main/teximage.c | 33 +- mesalib/src/mesa/main/texparam.c | 137 +- mesalib/src/mesa/program/ir_to_mesa.cpp | 6424 +++++++++++----------- mesalib/src/mesa/state_tracker/st_cb_texture.c | 4043 +++++++------- mesalib/src/mesa/state_tracker/st_extensions.c | 6 + mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c | 2496 ++++----- 10 files changed, 9516 insertions(+), 9497 deletions(-) (limited to 'mesalib') diff --git a/mesalib/configure.ac b/mesalib/configure.ac index 1c6edb75b..a331174bd 100644 --- a/mesalib/configure.ac +++ b/mesalib/configure.ac @@ -20,6 +20,7 @@ AC_CANONICAL_HOST dnl Versions for external dependencies LIBDRM_REQUIRED=2.4.23 LIBDRM_RADEON_REQUIRED=2.4.23 +LIBDRM_INTEL_REQUIRED=2.4.23 DRI2PROTO_REQUIRED=2.1 GLPROTO_REQUIRED=1.4.11 LIBDRM_XORG_REQUIRED=2.4.23 @@ -997,7 +998,7 @@ AC_SUBST([DRI_LIB_DEPS]) case $DRI_DIRS in *i915*|*i965*) - PKG_CHECK_MODULES([INTEL], [libdrm_intel >= 2.4.23]) + PKG_CHECK_MODULES([INTEL], [libdrm_intel >= $LIBDRM_INTEL_REQUIRED]) ;; esac diff --git a/mesalib/src/glsl/ir_reader.cpp b/mesalib/src/glsl/ir_reader.cpp index 5b1eccb46..b78aa051b 100644 --- a/mesalib/src/glsl/ir_reader.cpp +++ b/mesalib/src/glsl/ir_reader.cpp @@ -1,996 +1,999 @@ -/* - * 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_reader.h" -#include "glsl_parser_extras.h" -#include "glsl_types.h" -#include "s_expression.h" - -const static bool debug = false; - -class ir_reader { -public: - ir_reader(_mesa_glsl_parse_state *); - - void read(exec_list *instructions, const char *src, bool scan_for_protos); - -private: - void *mem_ctx; - _mesa_glsl_parse_state *state; - - void ir_read_error(s_expression *, const char *fmt, ...); - - const glsl_type *read_type(s_expression *); - - void scan_for_prototypes(exec_list *, s_expression *); - ir_function *read_function(s_expression *, bool skip_body); - void read_function_sig(ir_function *, s_expression *, bool skip_body); - - void read_instructions(exec_list *, s_expression *, ir_loop *); - ir_instruction *read_instruction(s_expression *, ir_loop *); - ir_variable *read_declaration(s_expression *); - ir_if *read_if(s_expression *, ir_loop *); - ir_loop *read_loop(s_expression *); - ir_return *read_return(s_expression *); - ir_rvalue *read_rvalue(s_expression *); - ir_assignment *read_assignment(s_expression *); - ir_expression *read_expression(s_expression *); - ir_call *read_call(s_expression *); - ir_swizzle *read_swizzle(s_expression *); - ir_constant *read_constant(s_expression *); - ir_texture *read_texture(s_expression *); - - ir_dereference *read_dereference(s_expression *); -}; - -ir_reader::ir_reader(_mesa_glsl_parse_state *state) : state(state) -{ - this->mem_ctx = state; -} - -void -_mesa_glsl_read_ir(_mesa_glsl_parse_state *state, exec_list *instructions, - const char *src, bool scan_for_protos) -{ - ir_reader r(state); - r.read(instructions, src, scan_for_protos); -} - -void -ir_reader::read(exec_list *instructions, const char *src, bool scan_for_protos) -{ - s_expression *expr = s_expression::read_expression(mem_ctx, src); - if (expr == NULL) { - ir_read_error(NULL, "couldn't parse S-Expression."); - return; - } - - if (scan_for_protos) { - scan_for_prototypes(instructions, expr); - if (state->error) - return; - } - - read_instructions(instructions, expr, NULL); - talloc_free(expr); - - if (debug) - validate_ir_tree(instructions); -} - -void -ir_reader::ir_read_error(s_expression *expr, const char *fmt, ...) -{ - va_list ap; - - state->error = true; - - if (state->current_function != NULL) - state->info_log = talloc_asprintf_append(state->info_log, - "In function %s:\n", - state->current_function->function_name()); - state->info_log = talloc_strdup_append(state->info_log, "error: "); - - va_start(ap, fmt); - state->info_log = talloc_vasprintf_append(state->info_log, fmt, ap); - va_end(ap); - state->info_log = talloc_strdup_append(state->info_log, "\n"); - - if (expr != NULL) { - state->info_log = talloc_strdup_append(state->info_log, - "...in this context:\n "); - expr->print(); - state->info_log = talloc_strdup_append(state->info_log, "\n\n"); - } -} - -const glsl_type * -ir_reader::read_type(s_expression *expr) -{ - s_expression *s_base_type; - s_int *s_size; - - s_pattern pat[] = { "array", s_base_type, s_size }; - if (MATCH(expr, pat)) { - const glsl_type *base_type = read_type(s_base_type); - if (base_type == NULL) { - ir_read_error(NULL, "when reading base type of array type"); - return NULL; - } - - return glsl_type::get_array_instance(base_type, s_size->value()); - } - - s_symbol *type_sym = SX_AS_SYMBOL(expr); - if (type_sym == NULL) { - ir_read_error(expr, "expected "); - return NULL; - } - - const glsl_type *type = state->symbols->get_type(type_sym->value()); - if (type == NULL) - ir_read_error(expr, "invalid type: %s", type_sym->value()); - - return type; -} - - -void -ir_reader::scan_for_prototypes(exec_list *instructions, s_expression *expr) -{ - s_list *list = SX_AS_LIST(expr); - if (list == NULL) { - ir_read_error(expr, "Expected ( ...); found an atom."); - return; - } - - foreach_iter(exec_list_iterator, it, list->subexpressions) { - s_list *sub = SX_AS_LIST(it.get()); - if (sub == NULL) - continue; // not a (function ...); ignore it. - - s_symbol *tag = SX_AS_SYMBOL(sub->subexpressions.get_head()); - if (tag == NULL || strcmp(tag->value(), "function") != 0) - continue; // not a (function ...); ignore it. - - ir_function *f = read_function(sub, true); - if (f == NULL) - return; - instructions->push_tail(f); - } -} - -ir_function * -ir_reader::read_function(s_expression *expr, bool skip_body) -{ - bool added = false; - s_symbol *name; - - s_pattern pat[] = { "function", name }; - if (!PARTIAL_MATCH(expr, pat)) { - ir_read_error(expr, "Expected (function (signature ...) ...)"); - return NULL; - } - - ir_function *f = state->symbols->get_function(name->value()); - if (f == NULL) { - f = new(mem_ctx) ir_function(name->value()); - added = state->symbols->add_function(f); - assert(added); - } - - exec_list_iterator it = ((s_list *) expr)->subexpressions.iterator(); - it.next(); // skip "function" tag - it.next(); // skip function name - for (/* nothing */; it.has_next(); it.next()) { - s_expression *s_sig = (s_expression *) it.get(); - read_function_sig(f, s_sig, skip_body); - } - return added ? f : NULL; -} - -void -ir_reader::read_function_sig(ir_function *f, s_expression *expr, bool skip_body) -{ - s_expression *type_expr; - s_list *paramlist; - s_list *body_list; - - s_pattern pat[] = { "signature", type_expr, paramlist, body_list }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "Expected (signature (parameters ...) " - "( ...))"); - return; - } - - const glsl_type *return_type = read_type(type_expr); - if (return_type == NULL) - return; - - s_symbol *paramtag = SX_AS_SYMBOL(paramlist->subexpressions.get_head()); - if (paramtag == NULL || strcmp(paramtag->value(), "parameters") != 0) { - ir_read_error(paramlist, "Expected (parameters ...)"); - return; - } - - // Read the parameters list into a temporary place. - exec_list hir_parameters; - state->symbols->push_scope(); - - exec_list_iterator it = paramlist->subexpressions.iterator(); - for (it.next() /* skip "parameters" */; it.has_next(); it.next()) { - ir_variable *var = read_declaration((s_expression *) it.get()); - if (var == NULL) - return; - - hir_parameters.push_tail(var); - } - - ir_function_signature *sig = f->exact_matching_signature(&hir_parameters); - if (sig == NULL && skip_body) { - /* If scanning for prototypes, generate a new signature. */ - sig = new(mem_ctx) ir_function_signature(return_type); - sig->is_builtin = true; - f->add_signature(sig); - } else if (sig != NULL) { - const char *badvar = sig->qualifiers_match(&hir_parameters); - if (badvar != NULL) { - ir_read_error(expr, "function `%s' parameter `%s' qualifiers " - "don't match prototype", f->name, badvar); - return; - } - - if (sig->return_type != return_type) { - ir_read_error(expr, "function `%s' return type doesn't " - "match prototype", f->name); - return; - } - } else { - /* No prototype for this body exists - skip it. */ - state->symbols->pop_scope(); - return; - } - assert(sig != NULL); - - sig->replace_parameters(&hir_parameters); - - if (!skip_body && !body_list->subexpressions.is_empty()) { - if (sig->is_defined) { - ir_read_error(expr, "function %s redefined", f->name); - return; - } - state->current_function = sig; - read_instructions(&sig->body, body_list, NULL); - state->current_function = NULL; - sig->is_defined = true; - } - - state->symbols->pop_scope(); -} - -void -ir_reader::read_instructions(exec_list *instructions, s_expression *expr, - ir_loop *loop_ctx) -{ - // Read in a list of instructions - s_list *list = SX_AS_LIST(expr); - if (list == NULL) { - ir_read_error(expr, "Expected ( ...); found an atom."); - return; - } - - foreach_iter(exec_list_iterator, it, list->subexpressions) { - s_expression *sub = (s_expression*) it.get(); - ir_instruction *ir = read_instruction(sub, loop_ctx); - if (ir != NULL) { - /* Global variable declarations should be moved to the top, before - * any functions that might use them. Functions are added to the - * instruction stream when scanning for prototypes, so without this - * hack, they always appear before variable declarations. - */ - if (state->current_function == NULL && ir->as_variable() != NULL) - instructions->push_head(ir); - else - instructions->push_tail(ir); - } - } -} - - -ir_instruction * -ir_reader::read_instruction(s_expression *expr, ir_loop *loop_ctx) -{ - s_symbol *symbol = SX_AS_SYMBOL(expr); - if (symbol != NULL) { - if (strcmp(symbol->value(), "break") == 0 && loop_ctx != NULL) - return new(mem_ctx) ir_loop_jump(ir_loop_jump::jump_break); - if (strcmp(symbol->value(), "continue") == 0 && loop_ctx != NULL) - return new(mem_ctx) ir_loop_jump(ir_loop_jump::jump_continue); - } - - s_list *list = SX_AS_LIST(expr); - if (list == NULL || list->subexpressions.is_empty()) { - ir_read_error(expr, "Invalid instruction.\n"); - return NULL; - } - - s_symbol *tag = SX_AS_SYMBOL(list->subexpressions.get_head()); - if (tag == NULL) { - ir_read_error(expr, "expected instruction tag"); - return NULL; - } - - ir_instruction *inst = NULL; - if (strcmp(tag->value(), "declare") == 0) { - inst = read_declaration(list); - } else if (strcmp(tag->value(), "assign") == 0) { - inst = read_assignment(list); - } else if (strcmp(tag->value(), "if") == 0) { - inst = read_if(list, loop_ctx); - } else if (strcmp(tag->value(), "loop") == 0) { - inst = read_loop(list); - } else if (strcmp(tag->value(), "return") == 0) { - inst = read_return(list); - } else if (strcmp(tag->value(), "function") == 0) { - inst = read_function(list, false); - } else { - inst = read_rvalue(list); - if (inst == NULL) - ir_read_error(NULL, "when reading instruction"); - } - return inst; -} - -ir_variable * -ir_reader::read_declaration(s_expression *expr) -{ - s_list *s_quals; - s_expression *s_type; - s_symbol *s_name; - - s_pattern pat[] = { "declare", s_quals, s_type, s_name }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (declare () )"); - return NULL; - } - - const glsl_type *type = read_type(s_type); - if (type == NULL) - return NULL; - - ir_variable *var = new(mem_ctx) ir_variable(type, s_name->value(), - ir_var_auto); - - foreach_iter(exec_list_iterator, it, s_quals->subexpressions) { - s_symbol *qualifier = SX_AS_SYMBOL(it.get()); - if (qualifier == NULL) { - ir_read_error(expr, "qualifier list must contain only symbols"); - return NULL; - } - - // FINISHME: Check for duplicate/conflicting qualifiers. - if (strcmp(qualifier->value(), "centroid") == 0) { - var->centroid = 1; - } else if (strcmp(qualifier->value(), "invariant") == 0) { - var->invariant = 1; - } else if (strcmp(qualifier->value(), "uniform") == 0) { - var->mode = ir_var_uniform; - } else if (strcmp(qualifier->value(), "auto") == 0) { - var->mode = ir_var_auto; - } else if (strcmp(qualifier->value(), "in") == 0) { - var->mode = ir_var_in; - } else if (strcmp(qualifier->value(), "out") == 0) { - var->mode = ir_var_out; - } else if (strcmp(qualifier->value(), "inout") == 0) { - var->mode = ir_var_inout; - } else if (strcmp(qualifier->value(), "smooth") == 0) { - var->interpolation = ir_var_smooth; - } else if (strcmp(qualifier->value(), "flat") == 0) { - var->interpolation = ir_var_flat; - } else if (strcmp(qualifier->value(), "noperspective") == 0) { - var->interpolation = ir_var_noperspective; - } else { - ir_read_error(expr, "unknown qualifier: %s", qualifier->value()); - return NULL; - } - } - - // Add the variable to the symbol table - state->symbols->add_variable(var); - - return var; -} - - -ir_if * -ir_reader::read_if(s_expression *expr, ir_loop *loop_ctx) -{ - s_expression *s_cond; - s_expression *s_then; - s_expression *s_else; - - s_pattern pat[] = { "if", s_cond, s_then, s_else }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (if (...) (...))"); - return NULL; - } - - ir_rvalue *condition = read_rvalue(s_cond); - if (condition == NULL) { - ir_read_error(NULL, "when reading condition of (if ...)"); - return NULL; - } - - ir_if *iff = new(mem_ctx) ir_if(condition); - - read_instructions(&iff->then_instructions, s_then, loop_ctx); - read_instructions(&iff->else_instructions, s_else, loop_ctx); - if (state->error) { - delete iff; - iff = NULL; - } - return iff; -} - - -ir_loop * -ir_reader::read_loop(s_expression *expr) -{ - s_expression *s_counter, *s_from, *s_to, *s_inc, *s_body; - - s_pattern pat[] = { "loop", s_counter, s_from, s_to, s_inc, s_body }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (loop " - " )"); - return NULL; - } - - // FINISHME: actually read the count/from/to fields. - - ir_loop *loop = new(mem_ctx) ir_loop; - read_instructions(&loop->body_instructions, s_body, loop); - if (state->error) { - delete loop; - loop = NULL; - } - return loop; -} - - -ir_return * -ir_reader::read_return(s_expression *expr) -{ - s_expression *s_retval; - - s_pattern pat[] = { "return", s_retval}; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (return )"); - return NULL; - } - - ir_rvalue *retval = read_rvalue(s_retval); - if (retval == NULL) { - ir_read_error(NULL, "when reading return value"); - return NULL; - } - - return new(mem_ctx) ir_return(retval); -} - - -ir_rvalue * -ir_reader::read_rvalue(s_expression *expr) -{ - s_list *list = SX_AS_LIST(expr); - if (list == NULL || list->subexpressions.is_empty()) - return NULL; - - s_symbol *tag = SX_AS_SYMBOL(list->subexpressions.get_head()); - if (tag == NULL) { - ir_read_error(expr, "expected rvalue tag"); - return NULL; - } - - ir_rvalue *rvalue = read_dereference(list); - if (rvalue != NULL || state->error) - return rvalue; - else if (strcmp(tag->value(), "swiz") == 0) { - rvalue = read_swizzle(list); - } else if (strcmp(tag->value(), "expression") == 0) { - rvalue = read_expression(list); - } else if (strcmp(tag->value(), "call") == 0) { - rvalue = read_call(list); - } else if (strcmp(tag->value(), "constant") == 0) { - rvalue = read_constant(list); - } else { - rvalue = read_texture(list); - if (rvalue == NULL && !state->error) - ir_read_error(expr, "unrecognized rvalue tag: %s", tag->value()); - } - - return rvalue; -} - -ir_assignment * -ir_reader::read_assignment(s_expression *expr) -{ - s_expression *cond_expr = NULL; - s_expression *lhs_expr, *rhs_expr; - s_list *mask_list; - - s_pattern pat4[] = { "assign", mask_list, lhs_expr, rhs_expr }; - s_pattern pat5[] = { "assign", cond_expr, mask_list, lhs_expr, rhs_expr }; - if (!MATCH(expr, pat4) && !MATCH(expr, pat5)) { - ir_read_error(expr, "expected (assign [] () " - " )"); - return NULL; - } - - ir_rvalue *condition = NULL; - if (cond_expr != NULL) { - condition = read_rvalue(cond_expr); - if (condition == NULL) { - ir_read_error(NULL, "when reading condition of assignment"); - return NULL; - } - } - - unsigned mask = 0; - - s_symbol *mask_symbol; - s_pattern mask_pat[] = { mask_symbol }; - if (MATCH(mask_list, mask_pat)) { - const char *mask_str = mask_symbol->value(); - unsigned mask_length = strlen(mask_str); - if (mask_length > 4) { - ir_read_error(expr, "invalid write mask: %s", mask_str); - return NULL; - } - - const unsigned idx_map[] = { 3, 0, 1, 2 }; /* w=bit 3, x=0, y=1, z=2 */ - - for (unsigned i = 0; i < mask_length; i++) { - if (mask_str[i] < 'w' || mask_str[i] > 'z') { - ir_read_error(expr, "write mask contains invalid character: %c", - mask_str[i]); - return NULL; - } - mask |= 1 << idx_map[mask_str[i] - 'w']; - } - } else if (!mask_list->subexpressions.is_empty()) { - ir_read_error(mask_list, "expected () or ()"); - return NULL; - } - - ir_dereference *lhs = read_dereference(lhs_expr); - if (lhs == NULL) { - ir_read_error(NULL, "when reading left-hand side of assignment"); - return NULL; - } - - ir_rvalue *rhs = read_rvalue(rhs_expr); - if (rhs == NULL) { - ir_read_error(NULL, "when reading right-hand side of assignment"); - return NULL; - } - - if (mask == 0 && (lhs->type->is_vector() || lhs->type->is_scalar())) { - ir_read_error(expr, "non-zero write mask required."); - return NULL; - } - - return new(mem_ctx) ir_assignment(lhs, rhs, condition, mask); -} - -ir_call * -ir_reader::read_call(s_expression *expr) -{ - s_symbol *name; - s_list *params; - - s_pattern pat[] = { "call", name, params }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (call ( ...))"); - return NULL; - } - - exec_list parameters; - - foreach_iter(exec_list_iterator, it, params->subexpressions) { - s_expression *expr = (s_expression*) it.get(); - ir_rvalue *param = read_rvalue(expr); - if (param == NULL) { - ir_read_error(expr, "when reading parameter to function call"); - return NULL; - } - parameters.push_tail(param); - } - - ir_function *f = state->symbols->get_function(name->value()); - if (f == NULL) { - ir_read_error(expr, "found call to undefined function %s", - name->value()); - return NULL; - } - - ir_function_signature *callee = f->matching_signature(¶meters); - if (callee == NULL) { - ir_read_error(expr, "couldn't find matching signature for function " - "%s", name->value()); - return NULL; - } - - return new(mem_ctx) ir_call(callee, ¶meters); -} - -ir_expression * -ir_reader::read_expression(s_expression *expr) -{ - s_expression *s_type; - s_symbol *s_op; - s_expression *s_arg1; - - s_pattern pat[] = { "expression", s_type, s_op, s_arg1 }; - if (!PARTIAL_MATCH(expr, pat)) { - ir_read_error(expr, "expected (expression " - " [])"); - return NULL; - } - s_expression *s_arg2 = (s_expression *) s_arg1->next; // may be tail sentinel - - const glsl_type *type = read_type(s_type); - if (type == NULL) - return NULL; - - /* Read the operator */ - ir_expression_operation op = ir_expression::get_operator(s_op->value()); - if (op == (ir_expression_operation) -1) { - ir_read_error(expr, "invalid operator: %s", s_op->value()); - return NULL; - } - - unsigned num_operands = ir_expression::get_num_operands(op); - if (num_operands == 1 && !s_arg1->next->is_tail_sentinel()) { - ir_read_error(expr, "expected (expression %s )", - s_op->value()); - return NULL; - } - - ir_rvalue *arg1 = read_rvalue(s_arg1); - ir_rvalue *arg2 = NULL; - if (arg1 == NULL) { - ir_read_error(NULL, "when reading first operand of %s", s_op->value()); - return NULL; - } - - if (num_operands == 2) { - if (s_arg2->is_tail_sentinel() || !s_arg2->next->is_tail_sentinel()) { - ir_read_error(expr, "expected (expression %s " - ")", s_op->value()); - return NULL; - } - arg2 = read_rvalue(s_arg2); - if (arg2 == NULL) { - ir_read_error(NULL, "when reading second operand of %s", - s_op->value()); - return NULL; - } - } - - return new(mem_ctx) ir_expression(op, type, arg1, arg2); -} - -ir_swizzle * -ir_reader::read_swizzle(s_expression *expr) -{ - s_symbol *swiz; - s_expression *sub; - - s_pattern pat[] = { "swiz", swiz, sub }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (swiz )"); - return NULL; - } - - if (strlen(swiz->value()) > 4) { - ir_read_error(expr, "expected a valid swizzle; found %s", swiz->value()); - return NULL; - } - - ir_rvalue *rvalue = read_rvalue(sub); - if (rvalue == NULL) - return NULL; - - ir_swizzle *ir = ir_swizzle::create(rvalue, swiz->value(), - rvalue->type->vector_elements); - if (ir == NULL) - ir_read_error(expr, "invalid swizzle"); - - return ir; -} - -ir_constant * -ir_reader::read_constant(s_expression *expr) -{ - s_expression *type_expr; - s_list *values; - - s_pattern pat[] = { "constant", type_expr, values }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (constant (...))"); - return NULL; - } - - const glsl_type *type = read_type(type_expr); - if (type == NULL) - return NULL; - - if (values == NULL) { - ir_read_error(expr, "expected (constant (...))"); - return NULL; - } - - if (type->is_array()) { - unsigned elements_supplied = 0; - exec_list elements; - foreach_iter(exec_list_iterator, it, values->subexpressions) { - s_expression *elt = (s_expression *) it.get(); - ir_constant *ir_elt = read_constant(elt); - if (ir_elt == NULL) - return NULL; - elements.push_tail(ir_elt); - elements_supplied++; - } - - if (elements_supplied != type->length) { - ir_read_error(values, "expected exactly %u array elements, " - "given %u", type->length, elements_supplied); - return NULL; - } - return new(mem_ctx) ir_constant(type, &elements); - } - - const glsl_type *const base_type = type->get_base_type(); - - ir_constant_data data = { { 0 } }; - - // Read in list of values (at most 16). - int k = 0; - foreach_iter(exec_list_iterator, it, values->subexpressions) { - if (k >= 16) { - ir_read_error(values, "expected at most 16 numbers"); - return NULL; - } - - s_expression *expr = (s_expression*) it.get(); - - if (base_type->base_type == GLSL_TYPE_FLOAT) { - s_number *value = SX_AS_NUMBER(expr); - if (value == NULL) { - ir_read_error(values, "expected numbers"); - return NULL; - } - data.f[k] = value->fvalue(); - } else { - s_int *value = SX_AS_INT(expr); - if (value == NULL) { - ir_read_error(values, "expected integers"); - return NULL; - } - - switch (base_type->base_type) { - case GLSL_TYPE_UINT: { - data.u[k] = value->value(); - break; - } - case GLSL_TYPE_INT: { - data.i[k] = value->value(); - break; - } - case GLSL_TYPE_BOOL: { - data.b[k] = value->value(); - break; - } - default: - ir_read_error(values, "unsupported constant type"); - return NULL; - } - } - ++k; - } - - return new(mem_ctx) ir_constant(type, &data); -} - -ir_dereference * -ir_reader::read_dereference(s_expression *expr) -{ - s_symbol *s_var; - s_expression *s_subject; - s_expression *s_index; - s_symbol *s_field; - - s_pattern var_pat[] = { "var_ref", s_var }; - s_pattern array_pat[] = { "array_ref", s_subject, s_index }; - s_pattern record_pat[] = { "record_ref", s_subject, s_field }; - - if (MATCH(expr, var_pat)) { - ir_variable *var = state->symbols->get_variable(s_var->value()); - if (var == NULL) { - ir_read_error(expr, "undeclared variable: %s", s_var->value()); - return NULL; - } - return new(mem_ctx) ir_dereference_variable(var); - } else if (MATCH(expr, array_pat)) { - ir_rvalue *subject = read_rvalue(s_subject); - if (subject == NULL) { - ir_read_error(NULL, "when reading the subject of an array_ref"); - return NULL; - } - - ir_rvalue *idx = read_rvalue(s_index); - if (subject == NULL) { - ir_read_error(NULL, "when reading the index of an array_ref"); - return NULL; - } - return new(mem_ctx) ir_dereference_array(subject, idx); - } else if (MATCH(expr, record_pat)) { - ir_rvalue *subject = read_rvalue(s_subject); - if (subject == NULL) { - ir_read_error(NULL, "when reading the subject of a record_ref"); - return NULL; - } - return new(mem_ctx) ir_dereference_record(subject, s_field->value()); - } - return NULL; -} - -ir_texture * -ir_reader::read_texture(s_expression *expr) -{ - s_symbol *tag = NULL; - s_expression *s_sampler = NULL; - s_expression *s_coord = NULL; - s_list *s_offset = NULL; - s_expression *s_proj = NULL; - s_list *s_shadow = NULL; - s_expression *s_lod = NULL; - - ir_texture_opcode op; - - s_pattern tex_pattern[] = - { "tex", s_sampler, s_coord, s_offset, s_proj, s_shadow }; - s_pattern txf_pattern[] = - { "txf", s_sampler, s_coord, s_offset, s_lod }; - s_pattern other_pattern[] = - { tag, s_sampler, s_coord, s_offset, s_proj, s_shadow, s_lod }; - - if (MATCH(expr, tex_pattern)) { - op = ir_tex; - } else if (MATCH(expr, txf_pattern)) { - op = ir_txf; - } else if (MATCH(expr, other_pattern)) { - op = ir_texture::get_opcode(tag->value()); - if (op == -1) - return NULL; - } - - ir_texture *tex = new(mem_ctx) ir_texture(op); - - // Read sampler (must be a deref) - ir_dereference *sampler = read_dereference(s_sampler); - if (sampler == NULL) { - ir_read_error(NULL, "when reading sampler in (%s ...)", - tex->opcode_string()); - return NULL; - } - tex->set_sampler(sampler); - - // Read coordinate (any rvalue) - tex->coordinate = read_rvalue(s_coord); - if (tex->coordinate == NULL) { - ir_read_error(NULL, "when reading coordinate in (%s ...)", - tex->opcode_string()); - return NULL; - } - - // Read texel offset, i.e. (0 0 0) - s_int *offset_x; - s_int *offset_y; - s_int *offset_z; - s_pattern offset_pat[] = { offset_x, offset_y, offset_z }; - if (!MATCH(s_offset, offset_pat)) { - ir_read_error(s_offset, "expected ( )"); - return NULL; - } - tex->offsets[0] = offset_x->value(); - tex->offsets[1] = offset_y->value(); - tex->offsets[2] = offset_z->value(); - - if (op != ir_txf) { - s_int *proj_as_int = SX_AS_INT(s_proj); - if (proj_as_int && proj_as_int->value() == 1) { - tex->projector = NULL; - } else { - tex->projector = read_rvalue(s_proj); - if (tex->projector == NULL) { - ir_read_error(NULL, "when reading projective divide in (%s ..)", - tex->opcode_string()); - return NULL; - } - } - - if (s_shadow->subexpressions.is_empty()) { - tex->shadow_comparitor = NULL; - } else { - tex->shadow_comparitor = read_rvalue(s_shadow); - if (tex->shadow_comparitor == NULL) { - ir_read_error(NULL, "when reading shadow comparitor in (%s ..)", - tex->opcode_string()); - return NULL; - } - } - } - - switch (op) { - case ir_txb: - tex->lod_info.bias = read_rvalue(s_lod); - if (tex->lod_info.bias == NULL) { - ir_read_error(NULL, "when reading LOD bias in (txb ...)"); - return NULL; - } - break; - case ir_txl: - case ir_txf: - tex->lod_info.lod = read_rvalue(s_lod); - if (tex->lod_info.lod == NULL) { - ir_read_error(NULL, "when reading LOD in (%s ...)", - tex->opcode_string()); - return NULL; - } - break; - case ir_txd: { - s_expression *s_dx, *s_dy; - s_pattern dxdy_pat[] = { s_dx, s_dy }; - if (!MATCH(s_lod, dxdy_pat)) { - ir_read_error(s_lod, "expected (dPdx dPdy) in (txd ...)"); - return NULL; - } - tex->lod_info.grad.dPdx = read_rvalue(s_dx); - if (tex->lod_info.grad.dPdx == NULL) { - ir_read_error(NULL, "when reading dPdx in (txd ...)"); - return NULL; - } - tex->lod_info.grad.dPdy = read_rvalue(s_dy); - if (tex->lod_info.grad.dPdy == NULL) { - ir_read_error(NULL, "when reading dPdy in (txd ...)"); - return NULL; - } - break; - } - default: - // tex doesn't have any extra parameters. - break; - }; - return tex; -} +/* + * 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_reader.h" +#include "glsl_parser_extras.h" +#include "glsl_types.h" +#include "s_expression.h" + +const static bool debug = false; + +class ir_reader { +public: + ir_reader(_mesa_glsl_parse_state *); + + void read(exec_list *instructions, const char *src, bool scan_for_protos); + +private: + void *mem_ctx; + _mesa_glsl_parse_state *state; + + void ir_read_error(s_expression *, const char *fmt, ...); + + const glsl_type *read_type(s_expression *); + + void scan_for_prototypes(exec_list *, s_expression *); + ir_function *read_function(s_expression *, bool skip_body); + void read_function_sig(ir_function *, s_expression *, bool skip_body); + + void read_instructions(exec_list *, s_expression *, ir_loop *); + ir_instruction *read_instruction(s_expression *, ir_loop *); + ir_variable *read_declaration(s_expression *); + ir_if *read_if(s_expression *, ir_loop *); + ir_loop *read_loop(s_expression *); + ir_return *read_return(s_expression *); + ir_rvalue *read_rvalue(s_expression *); + ir_assignment *read_assignment(s_expression *); + ir_expression *read_expression(s_expression *); + ir_call *read_call(s_expression *); + ir_swizzle *read_swizzle(s_expression *); + ir_constant *read_constant(s_expression *); + ir_texture *read_texture(s_expression *); + + ir_dereference *read_dereference(s_expression *); +}; + +ir_reader::ir_reader(_mesa_glsl_parse_state *state) : state(state) +{ + this->mem_ctx = state; +} + +void +_mesa_glsl_read_ir(_mesa_glsl_parse_state *state, exec_list *instructions, + const char *src, bool scan_for_protos) +{ + ir_reader r(state); + r.read(instructions, src, scan_for_protos); +} + +void +ir_reader::read(exec_list *instructions, const char *src, bool scan_for_protos) +{ + s_expression *expr = s_expression::read_expression(mem_ctx, src); + if (expr == NULL) { + ir_read_error(NULL, "couldn't parse S-Expression."); + return; + } + + if (scan_for_protos) { + scan_for_prototypes(instructions, expr); + if (state->error) + return; + } + + read_instructions(instructions, expr, NULL); + talloc_free(expr); + + if (debug) + validate_ir_tree(instructions); +} + +void +ir_reader::ir_read_error(s_expression *expr, const char *fmt, ...) +{ + va_list ap; + + state->error = true; + + if (state->current_function != NULL) + state->info_log = talloc_asprintf_append(state->info_log, + "In function %s:\n", + state->current_function->function_name()); + state->info_log = talloc_strdup_append(state->info_log, "error: "); + + va_start(ap, fmt); + state->info_log = talloc_vasprintf_append(state->info_log, fmt, ap); + va_end(ap); + state->info_log = talloc_strdup_append(state->info_log, "\n"); + + if (expr != NULL) { + state->info_log = talloc_strdup_append(state->info_log, + "...in this context:\n "); + expr->print(); + state->info_log = talloc_strdup_append(state->info_log, "\n\n"); + } +} + +const glsl_type * +ir_reader::read_type(s_expression *expr) +{ + s_expression *s_base_type; + s_int *s_size; + + s_pattern pat[] = { "array", s_base_type, s_size }; + if (MATCH(expr, pat)) { + const glsl_type *base_type = read_type(s_base_type); + if (base_type == NULL) { + ir_read_error(NULL, "when reading base type of array type"); + return NULL; + } + + return glsl_type::get_array_instance(base_type, s_size->value()); + } + + s_symbol *type_sym = SX_AS_SYMBOL(expr); + if (type_sym == NULL) { + ir_read_error(expr, "expected "); + return NULL; + } + + const glsl_type *type = state->symbols->get_type(type_sym->value()); + if (type == NULL) + ir_read_error(expr, "invalid type: %s", type_sym->value()); + + return type; +} + + +void +ir_reader::scan_for_prototypes(exec_list *instructions, s_expression *expr) +{ + s_list *list = SX_AS_LIST(expr); + if (list == NULL) { + ir_read_error(expr, "Expected ( ...); found an atom."); + return; + } + + foreach_iter(exec_list_iterator, it, list->subexpressions) { + s_list *sub = SX_AS_LIST(it.get()); + if (sub == NULL) + continue; // not a (function ...); ignore it. + + s_symbol *tag = SX_AS_SYMBOL(sub->subexpressions.get_head()); + if (tag == NULL || strcmp(tag->value(), "function") != 0) + continue; // not a (function ...); ignore it. + + ir_function *f = read_function(sub, true); + if (f == NULL) + return; + instructions->push_tail(f); + } +} + +ir_function * +ir_reader::read_function(s_expression *expr, bool skip_body) +{ + bool added = false; + s_symbol *name; + + s_pattern pat[] = { "function", name }; + if (!PARTIAL_MATCH(expr, pat)) { + ir_read_error(expr, "Expected (function (signature ...) ...)"); + return NULL; + } + + ir_function *f = state->symbols->get_function(name->value()); + if (f == NULL) { + f = new(mem_ctx) ir_function(name->value()); + added = state->symbols->add_function(f); + assert(added); + } + + exec_list_iterator it = ((s_list *) expr)->subexpressions.iterator(); + it.next(); // skip "function" tag + it.next(); // skip function name + for (/* nothing */; it.has_next(); it.next()) { + s_expression *s_sig = (s_expression *) it.get(); + read_function_sig(f, s_sig, skip_body); + } + return added ? f : NULL; +} + +void +ir_reader::read_function_sig(ir_function *f, s_expression *expr, bool skip_body) +{ + s_expression *type_expr; + s_list *paramlist; + s_list *body_list; + + s_pattern pat[] = { "signature", type_expr, paramlist, body_list }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "Expected (signature (parameters ...) " + "( ...))"); + return; + } + + const glsl_type *return_type = read_type(type_expr); + if (return_type == NULL) + return; + + s_symbol *paramtag = SX_AS_SYMBOL(paramlist->subexpressions.get_head()); + if (paramtag == NULL || strcmp(paramtag->value(), "parameters") != 0) { + ir_read_error(paramlist, "Expected (parameters ...)"); + return; + } + + // Read the parameters list into a temporary place. + exec_list hir_parameters; + state->symbols->push_scope(); + + exec_list_iterator it = paramlist->subexpressions.iterator(); + for (it.next() /* skip "parameters" */; it.has_next(); it.next()) { + ir_variable *var = read_declaration((s_expression *) it.get()); + if (var == NULL) + return; + + hir_parameters.push_tail(var); + } + + ir_function_signature *sig = f->exact_matching_signature(&hir_parameters); + if (sig == NULL && skip_body) { + /* If scanning for prototypes, generate a new signature. */ + sig = new(mem_ctx) ir_function_signature(return_type); + sig->is_builtin = true; + f->add_signature(sig); + } else if (sig != NULL) { + const char *badvar = sig->qualifiers_match(&hir_parameters); + if (badvar != NULL) { + ir_read_error(expr, "function `%s' parameter `%s' qualifiers " + "don't match prototype", f->name, badvar); + return; + } + + if (sig->return_type != return_type) { + ir_read_error(expr, "function `%s' return type doesn't " + "match prototype", f->name); + return; + } + } else { + /* No prototype for this body exists - skip it. */ + state->symbols->pop_scope(); + return; + } + assert(sig != NULL); + + sig->replace_parameters(&hir_parameters); + + if (!skip_body && !body_list->subexpressions.is_empty()) { + if (sig->is_defined) { + ir_read_error(expr, "function %s redefined", f->name); + return; + } + state->current_function = sig; + read_instructions(&sig->body, body_list, NULL); + state->current_function = NULL; + sig->is_defined = true; + } + + state->symbols->pop_scope(); +} + +void +ir_reader::read_instructions(exec_list *instructions, s_expression *expr, + ir_loop *loop_ctx) +{ + // Read in a list of instructions + s_list *list = SX_AS_LIST(expr); + if (list == NULL) { + ir_read_error(expr, "Expected ( ...); found an atom."); + return; + } + + foreach_iter(exec_list_iterator, it, list->subexpressions) { + s_expression *sub = (s_expression*) it.get(); + ir_instruction *ir = read_instruction(sub, loop_ctx); + if (ir != NULL) { + /* Global variable declarations should be moved to the top, before + * any functions that might use them. Functions are added to the + * instruction stream when scanning for prototypes, so without this + * hack, they always appear before variable declarations. + */ + if (state->current_function == NULL && ir->as_variable() != NULL) + instructions->push_head(ir); + else + instructions->push_tail(ir); + } + } +} + + +ir_instruction * +ir_reader::read_instruction(s_expression *expr, ir_loop *loop_ctx) +{ + s_symbol *symbol = SX_AS_SYMBOL(expr); + if (symbol != NULL) { + if (strcmp(symbol->value(), "break") == 0 && loop_ctx != NULL) + return new(mem_ctx) ir_loop_jump(ir_loop_jump::jump_break); + if (strcmp(symbol->value(), "continue") == 0 && loop_ctx != NULL) + return new(mem_ctx) ir_loop_jump(ir_loop_jump::jump_continue); + } + + s_list *list = SX_AS_LIST(expr); + if (list == NULL || list->subexpressions.is_empty()) { + ir_read_error(expr, "Invalid instruction.\n"); + return NULL; + } + + s_symbol *tag = SX_AS_SYMBOL(list->subexpressions.get_head()); + if (tag == NULL) { + ir_read_error(expr, "expected instruction tag"); + return NULL; + } + + ir_instruction *inst = NULL; + if (strcmp(tag->value(), "declare") == 0) { + inst = read_declaration(list); + } else if (strcmp(tag->value(), "assign") == 0) { + inst = read_assignment(list); + } else if (strcmp(tag->value(), "if") == 0) { + inst = read_if(list, loop_ctx); + } else if (strcmp(tag->value(), "loop") == 0) { + inst = read_loop(list); + } else if (strcmp(tag->value(), "return") == 0) { + inst = read_return(list); + } else if (strcmp(tag->value(), "function") == 0) { + inst = read_function(list, false); + } else { + inst = read_rvalue(list); + if (inst == NULL) + ir_read_error(NULL, "when reading instruction"); + } + return inst; +} + +ir_variable * +ir_reader::read_declaration(s_expression *expr) +{ + s_list *s_quals; + s_expression *s_type; + s_symbol *s_name; + + s_pattern pat[] = { "declare", s_quals, s_type, s_name }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (declare () )"); + return NULL; + } + + const glsl_type *type = read_type(s_type); + if (type == NULL) + return NULL; + + ir_variable *var = new(mem_ctx) ir_variable(type, s_name->value(), + ir_var_auto); + + foreach_iter(exec_list_iterator, it, s_quals->subexpressions) { + s_symbol *qualifier = SX_AS_SYMBOL(it.get()); + if (qualifier == NULL) { + ir_read_error(expr, "qualifier list must contain only symbols"); + return NULL; + } + + // FINISHME: Check for duplicate/conflicting qualifiers. + if (strcmp(qualifier->value(), "centroid") == 0) { + var->centroid = 1; + } else if (strcmp(qualifier->value(), "invariant") == 0) { + var->invariant = 1; + } else if (strcmp(qualifier->value(), "uniform") == 0) { + var->mode = ir_var_uniform; + } else if (strcmp(qualifier->value(), "auto") == 0) { + var->mode = ir_var_auto; + } else if (strcmp(qualifier->value(), "in") == 0) { + var->mode = ir_var_in; + } else if (strcmp(qualifier->value(), "out") == 0) { + var->mode = ir_var_out; + } else if (strcmp(qualifier->value(), "inout") == 0) { + var->mode = ir_var_inout; + } else if (strcmp(qualifier->value(), "smooth") == 0) { + var->interpolation = ir_var_smooth; + } else if (strcmp(qualifier->value(), "flat") == 0) { + var->interpolation = ir_var_flat; + } else if (strcmp(qualifier->value(), "noperspective") == 0) { + var->interpolation = ir_var_noperspective; + } else { + ir_read_error(expr, "unknown qualifier: %s", qualifier->value()); + return NULL; + } + } + + // Add the variable to the symbol table + state->symbols->add_variable(var); + + return var; +} + + +ir_if * +ir_reader::read_if(s_expression *expr, ir_loop *loop_ctx) +{ + s_expression *s_cond; + s_expression *s_then; + s_expression *s_else; + + s_pattern pat[] = { "if", s_cond, s_then, s_else }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (if (...) (...))"); + return NULL; + } + + ir_rvalue *condition = read_rvalue(s_cond); + if (condition == NULL) { + ir_read_error(NULL, "when reading condition of (if ...)"); + return NULL; + } + + ir_if *iff = new(mem_ctx) ir_if(condition); + + read_instructions(&iff->then_instructions, s_then, loop_ctx); + read_instructions(&iff->else_instructions, s_else, loop_ctx); + if (state->error) { + delete iff; + iff = NULL; + } + return iff; +} + + +ir_loop * +ir_reader::read_loop(s_expression *expr) +{ + s_expression *s_counter, *s_from, *s_to, *s_inc, *s_body; + + s_pattern pat[] = { "loop", s_counter, s_from, s_to, s_inc, s_body }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (loop " + " )"); + return NULL; + } + + // FINISHME: actually read the count/from/to fields. + + ir_loop *loop = new(mem_ctx) ir_loop; + read_instructions(&loop->body_instructions, s_body, loop); + if (state->error) { + delete loop; + loop = NULL; + } + return loop; +} + + +ir_return * +ir_reader::read_return(s_expression *expr) +{ + s_expression *s_retval; + + s_pattern pat[] = { "return", s_retval}; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (return )"); + return NULL; + } + + ir_rvalue *retval = read_rvalue(s_retval); + if (retval == NULL) { + ir_read_error(NULL, "when reading return value"); + return NULL; + } + + return new(mem_ctx) ir_return(retval); +} + + +ir_rvalue * +ir_reader::read_rvalue(s_expression *expr) +{ + s_list *list = SX_AS_LIST(expr); + if (list == NULL || list->subexpressions.is_empty()) + return NULL; + + s_symbol *tag = SX_AS_SYMBOL(list->subexpressions.get_head()); + if (tag == NULL) { + ir_read_error(expr, "expected rvalue tag"); + return NULL; + } + + ir_rvalue *rvalue = read_dereference(list); + if (rvalue != NULL || state->error) + return rvalue; + else if (strcmp(tag->value(), "swiz") == 0) { + rvalue = read_swizzle(list); + } else if (strcmp(tag->value(), "expression") == 0) { + rvalue = read_expression(list); + } else if (strcmp(tag->value(), "call") == 0) { + rvalue = read_call(list); + } else if (strcmp(tag->value(), "constant") == 0) { + rvalue = read_constant(list); + } else { + rvalue = read_texture(list); + if (rvalue == NULL && !state->error) + ir_read_error(expr, "unrecognized rvalue tag: %s", tag->value()); + } + + return rvalue; +} + +ir_assignment * +ir_reader::read_assignment(s_expression *expr) +{ + s_expression *cond_expr = NULL; + s_expression *lhs_expr, *rhs_expr; + s_list *mask_list; + + s_pattern pat4[] = { "assign", mask_list, lhs_expr, rhs_expr }; + s_pattern pat5[] = { "assign", cond_expr, mask_list, lhs_expr, rhs_expr }; + if (!MATCH(expr, pat4) && !MATCH(expr, pat5)) { + ir_read_error(expr, "expected (assign [] () " + " )"); + return NULL; + } + + ir_rvalue *condition = NULL; + if (cond_expr != NULL) { + condition = read_rvalue(cond_expr); + if (condition == NULL) { + ir_read_error(NULL, "when reading condition of assignment"); + return NULL; + } + } + + unsigned mask = 0; + + s_symbol *mask_symbol; + s_pattern mask_pat[] = { mask_symbol }; + if (MATCH(mask_list, mask_pat)) { + const char *mask_str = mask_symbol->value(); + unsigned mask_length = strlen(mask_str); + if (mask_length > 4) { + ir_read_error(expr, "invalid write mask: %s", mask_str); + return NULL; + } + + const unsigned idx_map[] = { 3, 0, 1, 2 }; /* w=bit 3, x=0, y=1, z=2 */ + + for (unsigned i = 0; i < mask_length; i++) { + if (mask_str[i] < 'w' || mask_str[i] > 'z') { + ir_read_error(expr, "write mask contains invalid character: %c", + mask_str[i]); + return NULL; + } + mask |= 1 << idx_map[mask_str[i] - 'w']; + } + } else if (!mask_list->subexpressions.is_empty()) { + ir_read_error(mask_list, "expected () or ()"); + return NULL; + } + + ir_dereference *lhs = read_dereference(lhs_expr); + if (lhs == NULL) { + ir_read_error(NULL, "when reading left-hand side of assignment"); + return NULL; + } + + ir_rvalue *rhs = read_rvalue(rhs_expr); + if (rhs == NULL) { + ir_read_error(NULL, "when reading right-hand side of assignment"); + return NULL; + } + + if (mask == 0 && (lhs->type->is_vector() || lhs->type->is_scalar())) { + ir_read_error(expr, "non-zero write mask required."); + return NULL; + } + + return new(mem_ctx) ir_assignment(lhs, rhs, condition, mask); +} + +ir_call * +ir_reader::read_call(s_expression *expr) +{ + s_symbol *name; + s_list *params; + + s_pattern pat[] = { "call", name, params }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (call ( ...))"); + return NULL; + } + + exec_list parameters; + + foreach_iter(exec_list_iterator, it, params->subexpressions) { + s_expression *expr = (s_expression*) it.get(); + ir_rvalue *param = read_rvalue(expr); + if (param == NULL) { + ir_read_error(expr, "when reading parameter to function call"); + return NULL; + } + parameters.push_tail(param); + } + + ir_function *f = state->symbols->get_function(name->value()); + if (f == NULL) { + ir_read_error(expr, "found call to undefined function %s", + name->value()); + return NULL; + } + + ir_function_signature *callee = f->matching_signature(¶meters); + if (callee == NULL) { + ir_read_error(expr, "couldn't find matching signature for function " + "%s", name->value()); + return NULL; + } + + return new(mem_ctx) ir_call(callee, ¶meters); +} + +ir_expression * +ir_reader::read_expression(s_expression *expr) +{ + s_expression *s_type; + s_symbol *s_op; + s_expression *s_arg1; + + s_pattern pat[] = { "expression", s_type, s_op, s_arg1 }; + if (!PARTIAL_MATCH(expr, pat)) { + ir_read_error(expr, "expected (expression " + " [])"); + return NULL; + } + s_expression *s_arg2 = (s_expression *) s_arg1->next; // may be tail sentinel + + const glsl_type *type = read_type(s_type); + if (type == NULL) + return NULL; + + /* Read the operator */ + ir_expression_operation op = ir_expression::get_operator(s_op->value()); + if (op == (ir_expression_operation) -1) { + ir_read_error(expr, "invalid operator: %s", s_op->value()); + return NULL; + } + + unsigned num_operands = ir_expression::get_num_operands(op); + if (num_operands == 1 && !s_arg1->next->is_tail_sentinel()) { + ir_read_error(expr, "expected (expression %s )", + s_op->value()); + return NULL; + } + + ir_rvalue *arg1 = read_rvalue(s_arg1); + ir_rvalue *arg2 = NULL; + if (arg1 == NULL) { + ir_read_error(NULL, "when reading first operand of %s", s_op->value()); + return NULL; + } + + if (num_operands == 2) { + if (s_arg2->is_tail_sentinel() || !s_arg2->next->is_tail_sentinel()) { + ir_read_error(expr, "expected (expression %s " + ")", s_op->value()); + return NULL; + } + arg2 = read_rvalue(s_arg2); + if (arg2 == NULL) { + ir_read_error(NULL, "when reading second operand of %s", + s_op->value()); + return NULL; + } + } + + return new(mem_ctx) ir_expression(op, type, arg1, arg2); +} + +ir_swizzle * +ir_reader::read_swizzle(s_expression *expr) +{ + s_symbol *swiz; + s_expression *sub; + + s_pattern pat[] = { "swiz", swiz, sub }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (swiz )"); + return NULL; + } + + if (strlen(swiz->value()) > 4) { + ir_read_error(expr, "expected a valid swizzle; found %s", swiz->value()); + return NULL; + } + + ir_rvalue *rvalue = read_rvalue(sub); + if (rvalue == NULL) + return NULL; + + ir_swizzle *ir = ir_swizzle::create(rvalue, swiz->value(), + rvalue->type->vector_elements); + if (ir == NULL) + ir_read_error(expr, "invalid swizzle"); + + return ir; +} + +ir_constant * +ir_reader::read_constant(s_expression *expr) +{ + s_expression *type_expr; + s_list *values; + + s_pattern pat[] = { "constant", type_expr, values }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (constant (...))"); + return NULL; + } + + const glsl_type *type = read_type(type_expr); + if (type == NULL) + return NULL; + + if (values == NULL) { + ir_read_error(expr, "expected (constant (...))"); + return NULL; + } + + if (type->is_array()) { + unsigned elements_supplied = 0; + exec_list elements; + foreach_iter(exec_list_iterator, it, values->subexpressions) { + s_expression *elt = (s_expression *) it.get(); + ir_constant *ir_elt = read_constant(elt); + if (ir_elt == NULL) + return NULL; + elements.push_tail(ir_elt); + elements_supplied++; + } + + if (elements_supplied != type->length) { + ir_read_error(values, "expected exactly %u array elements, " + "given %u", type->length, elements_supplied); + return NULL; + } + return new(mem_ctx) ir_constant(type, &elements); + } + + const glsl_type *const base_type = type->get_base_type(); + + ir_constant_data data = { { 0 } }; + + // Read in list of values (at most 16). + int k = 0; + foreach_iter(exec_list_iterator, it, values->subexpressions) { + if (k >= 16) { + ir_read_error(values, "expected at most 16 numbers"); + return NULL; + } + + s_expression *expr = (s_expression*) it.get(); + + if (base_type->base_type == GLSL_TYPE_FLOAT) { + s_number *value = SX_AS_NUMBER(expr); + if (value == NULL) { + ir_read_error(values, "expected numbers"); + return NULL; + } + data.f[k] = value->fvalue(); + } else { + s_int *value = SX_AS_INT(expr); + if (value == NULL) { + ir_read_error(values, "expected integers"); + return NULL; + } + + switch (base_type->base_type) { + case GLSL_TYPE_UINT: { + data.u[k] = value->value(); + break; + } + case GLSL_TYPE_INT: { + data.i[k] = value->value(); + break; + } + case GLSL_TYPE_BOOL: { + data.b[k] = value->value(); + break; + } + default: + ir_read_error(values, "unsupported constant type"); + return NULL; + } + } + ++k; + } + + return new(mem_ctx) ir_constant(type, &data); +} + +ir_dereference * +ir_reader::read_dereference(s_expression *expr) +{ + s_symbol *s_var; + s_expression *s_subject; + s_expression *s_index; + s_symbol *s_field; + + s_pattern var_pat[] = { "var_ref", s_var }; + s_pattern array_pat[] = { "array_ref", s_subject, s_index }; + s_pattern record_pat[] = { "record_ref", s_subject, s_field }; + + if (MATCH(expr, var_pat)) { + ir_variable *var = state->symbols->get_variable(s_var->value()); + if (var == NULL) { + ir_read_error(expr, "undeclared variable: %s", s_var->value()); + return NULL; + } + return new(mem_ctx) ir_dereference_variable(var); + } else if (MATCH(expr, array_pat)) { + ir_rvalue *subject = read_rvalue(s_subject); + if (subject == NULL) { + ir_read_error(NULL, "when reading the subject of an array_ref"); + return NULL; + } + + ir_rvalue *idx = read_rvalue(s_index); + if (subject == NULL) { + ir_read_error(NULL, "when reading the index of an array_ref"); + return NULL; + } + return new(mem_ctx) ir_dereference_array(subject, idx); + } else if (MATCH(expr, record_pat)) { + ir_rvalue *subject = read_rvalue(s_subject); + if (subject == NULL) { + ir_read_error(NULL, "when reading the subject of a record_ref"); + return NULL; + } + return new(mem_ctx) ir_dereference_record(subject, s_field->value()); + } + return NULL; +} + +ir_texture * +ir_reader::read_texture(s_expression *expr) +{ + s_symbol *tag = NULL; + s_expression *s_sampler = NULL; + s_expression *s_coord = NULL; + s_list *s_offset = NULL; + s_expression *s_proj = NULL; + s_list *s_shadow = NULL; + s_expression *s_lod = NULL; + + ir_texture_opcode op = ir_tex; /* silence warning */ + + s_pattern tex_pattern[] = + { "tex", s_sampler, s_coord, s_offset, s_proj, s_shadow }; + s_pattern txf_pattern[] = + { "txf", s_sampler, s_coord, s_offset, s_lod }; + s_pattern other_pattern[] = + { tag, s_sampler, s_coord, s_offset, s_proj, s_shadow, s_lod }; + + if (MATCH(expr, tex_pattern)) { + op = ir_tex; + } else if (MATCH(expr, txf_pattern)) { + op = ir_txf; + } else if (MATCH(expr, other_pattern)) { + op = ir_texture::get_opcode(tag->value()); + if (op == -1) + return NULL; + } else { + ir_read_error(NULL, "unexpected texture pattern"); + return NULL; + } + + ir_texture *tex = new(mem_ctx) ir_texture(op); + + // Read sampler (must be a deref) + ir_dereference *sampler = read_dereference(s_sampler); + if (sampler == NULL) { + ir_read_error(NULL, "when reading sampler in (%s ...)", + tex->opcode_string()); + return NULL; + } + tex->set_sampler(sampler); + + // Read coordinate (any rvalue) + tex->coordinate = read_rvalue(s_coord); + if (tex->coordinate == NULL) { + ir_read_error(NULL, "when reading coordinate in (%s ...)", + tex->opcode_string()); + return NULL; + } + + // Read texel offset, i.e. (0 0 0) + s_int *offset_x; + s_int *offset_y; + s_int *offset_z; + s_pattern offset_pat[] = { offset_x, offset_y, offset_z }; + if (!MATCH(s_offset, offset_pat)) { + ir_read_error(s_offset, "expected ( )"); + return NULL; + } + tex->offsets[0] = offset_x->value(); + tex->offsets[1] = offset_y->value(); + tex->offsets[2] = offset_z->value(); + + if (op != ir_txf) { + s_int *proj_as_int = SX_AS_INT(s_proj); + if (proj_as_int && proj_as_int->value() == 1) { + tex->projector = NULL; + } else { + tex->projector = read_rvalue(s_proj); + if (tex->projector == NULL) { + ir_read_error(NULL, "when reading projective divide in (%s ..)", + tex->opcode_string()); + return NULL; + } + } + + if (s_shadow->subexpressions.is_empty()) { + tex->shadow_comparitor = NULL; + } else { + tex->shadow_comparitor = read_rvalue(s_shadow); + if (tex->shadow_comparitor == NULL) { + ir_read_error(NULL, "when reading shadow comparitor in (%s ..)", + tex->opcode_string()); + return NULL; + } + } + } + + switch (op) { + case ir_txb: + tex->lod_info.bias = read_rvalue(s_lod); + if (tex->lod_info.bias == NULL) { + ir_read_error(NULL, "when reading LOD bias in (txb ...)"); + return NULL; + } + break; + case ir_txl: + case ir_txf: + tex->lod_info.lod = read_rvalue(s_lod); + if (tex->lod_info.lod == NULL) { + ir_read_error(NULL, "when reading LOD in (%s ...)", + tex->opcode_string()); + return NULL; + } + break; + case ir_txd: { + s_expression *s_dx, *s_dy; + s_pattern dxdy_pat[] = { s_dx, s_dy }; + if (!MATCH(s_lod, dxdy_pat)) { + ir_read_error(s_lod, "expected (dPdx dPdy) in (txd ...)"); + return NULL; + } + tex->lod_info.grad.dPdx = read_rvalue(s_dx); + if (tex->lod_info.grad.dPdx == NULL) { + ir_read_error(NULL, "when reading dPdx in (txd ...)"); + return NULL; + } + tex->lod_info.grad.dPdy = read_rvalue(s_dy); + if (tex->lod_info.grad.dPdy == NULL) { + ir_read_error(NULL, "when reading dPdy in (txd ...)"); + return NULL; + } + break; + } + default: + // tex doesn't have any extra parameters. + break; + }; + return tex; +} diff --git a/mesalib/src/glsl/link_functions.cpp b/mesalib/src/glsl/link_functions.cpp index ea9129fe3..861fa39b5 100644 --- a/mesalib/src/glsl/link_functions.cpp +++ b/mesalib/src/glsl/link_functions.cpp @@ -1,258 +1,270 @@ -/* - * 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 -#include -#include - -#include "main/core.h" -#include "glsl_symbol_table.h" -#include "glsl_parser_extras.h" -#include "ir.h" -#include "program.h" -#include "program/hash_table.h" -#include "linker.h" - -static ir_function_signature * -find_matching_signature(const char *name, const exec_list *actual_parameters, - gl_shader **shader_list, unsigned num_shaders); - -class call_link_visitor : public ir_hierarchical_visitor { -public: - call_link_visitor(gl_shader_program *prog, gl_shader *linked, - gl_shader **shader_list, unsigned num_shaders) - { - this->prog = prog; - this->shader_list = shader_list; - this->num_shaders = num_shaders; - this->success = true; - this->linked = linked; - - this->locals = hash_table_ctor(0, hash_table_pointer_hash, - hash_table_pointer_compare); - } - - ~call_link_visitor() - { - hash_table_dtor(this->locals); - } - - virtual ir_visitor_status visit(ir_variable *ir) - { - hash_table_insert(locals, ir, ir); - return visit_continue; - } - - virtual ir_visitor_status visit_enter(ir_call *ir) - { - /* If ir is an ir_call from a function that was imported from another - * shader callee will point to an ir_function_signature in the original - * shader. In this case the function signature MUST NOT BE MODIFIED. - * Doing so will modify the original shader. This may prevent that - * shader from being linkable in other programs. - */ - const ir_function_signature *const callee = ir->get_callee(); - assert(callee != NULL); - const char *const name = callee->function_name(); - - /* Determine if the requested function signature already exists in the - * final linked shader. If it does, use it as the target of the call. - */ - ir_function_signature *sig = - find_matching_signature(name, &callee->parameters, &linked, 1); - if (sig != NULL) { - ir->set_callee(sig); - return visit_continue; - } - - /* Try to find the signature in one of the other shaders that is being - * linked. If it's not found there, return an error. - */ - sig = find_matching_signature(name, &ir->actual_parameters, shader_list, - num_shaders); - if (sig == NULL) { - /* FINISHME: Log the full signature of unresolved function. - */ - linker_error_printf(this->prog, "unresolved reference to function " - "`%s'\n", name); - this->success = false; - return visit_stop; - } - - /* Find the prototype information in the linked shader. Generate any - * details that may be missing. - */ - ir_function *f = linked->symbols->get_function(name); - if (f == NULL) - f = new(linked) ir_function(name); - - ir_function_signature *linked_sig = - f->exact_matching_signature(&callee->parameters); - if (linked_sig == NULL) { - linked_sig = new(linked) ir_function_signature(callee->return_type); - f->add_signature(linked_sig); - } - - /* At this point linked_sig and called may be the same. If ir is an - * ir_call from linked then linked_sig and callee will be - * ir_function_signatures that have no definitions (is_defined is false). - */ - assert(!linked_sig->is_defined); - assert(linked_sig->body.is_empty()); - - /* Create an in-place clone of the function definition. This multistep - * process introduces some complexity here, but it has some advantages. - * The parameter list and the and function body are cloned separately. - * The clone of the parameter list is used to prime the hashtable used - * to replace variable references in the cloned body. - * - * The big advantage is that the ir_function_signature does not change. - * This means that we don't have to process the rest of the IR tree to - * patch ir_call nodes. In addition, there is no way to remove or - * replace signature stored in a function. One could easily be added, - * but this avoids the need. - */ - struct hash_table *ht = hash_table_ctor(0, hash_table_pointer_hash, - hash_table_pointer_compare); - exec_list formal_parameters; - foreach_list_const(node, &sig->parameters) { - const ir_instruction *const original = (ir_instruction *) node; - assert(const_cast(original)->as_variable()); - - ir_instruction *copy = original->clone(linked, ht); - formal_parameters.push_tail(copy); - } - - linked_sig->replace_parameters(&formal_parameters); - - foreach_list_const(node, &sig->body) { - const ir_instruction *const original = (ir_instruction *) node; - - ir_instruction *copy = original->clone(linked, ht); - linked_sig->body.push_tail(copy); - } - - linked_sig->is_defined = true; - hash_table_dtor(ht); - - /* Patch references inside the function to things outside the function - * (i.e., function calls and global variables). - */ - linked_sig->accept(this); - - ir->set_callee(linked_sig); - - return visit_continue; - } - - virtual ir_visitor_status visit(ir_dereference_variable *ir) - { - if (hash_table_find(locals, ir->var) == NULL) { - /* The non-function variable must be a global, so try to find the - * variable in the shader's symbol table. If the variable is not - * found, then it's a global that *MUST* be defined in the original - * shader. - */ - ir_variable *var = linked->symbols->get_variable(ir->var->name); - if (var == NULL) { - /* Clone the ir_variable that the dereference already has and add - * it to the linked shader. - */ - var = ir->var->clone(linked, NULL); - linked->symbols->add_variable(var); - linked->ir->push_head(var); - } - - ir->var = var; - } - - return visit_continue; - } - - /** Was function linking successful? */ - bool success; - -private: - /** - * Shader program being linked - * - * This is only used for logging error messages. - */ - gl_shader_program *prog; - - /** List of shaders available for linking. */ - gl_shader **shader_list; - - /** Number of shaders available for linking. */ - unsigned num_shaders; - - /** - * Final linked shader - * - * This is used two ways. It is used to find global variables in the - * linked shader that are accessed by the function. It is also used to add - * global variables from the shader where the function originated. - */ - gl_shader *linked; - - /** - * Table of variables local to the function. - */ - hash_table *locals; -}; - - -/** - * Searches a list of shaders for a particular function definition - */ -ir_function_signature * -find_matching_signature(const char *name, const exec_list *actual_parameters, - gl_shader **shader_list, unsigned num_shaders) -{ - for (unsigned i = 0; i < num_shaders; i++) { - ir_function *const f = shader_list[i]->symbols->get_function(name); - - if (f == NULL) - continue; - - ir_function_signature *sig = f->matching_signature(actual_parameters); - - if ((sig == NULL) || !sig->is_defined) - continue; - - return sig; - } - - return NULL; -} - - -bool -link_function_calls(gl_shader_program *prog, gl_shader *main, - gl_shader **shader_list, unsigned num_shaders) -{ - call_link_visitor v(prog, main, shader_list, num_shaders); - - v.run(main->ir); - return v.success; -} +/* + * 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 +#include +#include + +#include "main/core.h" +#include "glsl_symbol_table.h" +#include "glsl_parser_extras.h" +#include "ir.h" +#include "program.h" +#include "program/hash_table.h" +#include "linker.h" + +static ir_function_signature * +find_matching_signature(const char *name, const exec_list *actual_parameters, + gl_shader **shader_list, unsigned num_shaders); + +class call_link_visitor : public ir_hierarchical_visitor { +public: + call_link_visitor(gl_shader_program *prog, gl_shader *linked, + gl_shader **shader_list, unsigned num_shaders) + { + this->prog = prog; + this->shader_list = shader_list; + this->num_shaders = num_shaders; + this->success = true; + this->linked = linked; + + this->locals = hash_table_ctor(0, hash_table_pointer_hash, + hash_table_pointer_compare); + } + + ~call_link_visitor() + { + hash_table_dtor(this->locals); + } + + virtual ir_visitor_status visit(ir_variable *ir) + { + hash_table_insert(locals, ir, ir); + return visit_continue; + } + + virtual ir_visitor_status visit_enter(ir_call *ir) + { + /* If ir is an ir_call from a function that was imported from another + * shader callee will point to an ir_function_signature in the original + * shader. In this case the function signature MUST NOT BE MODIFIED. + * Doing so will modify the original shader. This may prevent that + * shader from being linkable in other programs. + */ + const ir_function_signature *const callee = ir->get_callee(); + assert(callee != NULL); + const char *const name = callee->function_name(); + + /* Determine if the requested function signature already exists in the + * final linked shader. If it does, use it as the target of the call. + */ + ir_function_signature *sig = + find_matching_signature(name, &callee->parameters, &linked, 1); + if (sig != NULL) { + ir->set_callee(sig); + return visit_continue; + } + + /* Try to find the signature in one of the other shaders that is being + * linked. If it's not found there, return an error. + */ + sig = find_matching_signature(name, &ir->actual_parameters, shader_list, + num_shaders); + if (sig == NULL) { + /* FINISHME: Log the full signature of unresolved function. + */ + linker_error_printf(this->prog, "unresolved reference to function " + "`%s'\n", name); + this->success = false; + return visit_stop; + } + + /* Find the prototype information in the linked shader. Generate any + * details that may be missing. + */ + ir_function *f = linked->symbols->get_function(name); + if (f == NULL) + f = new(linked) ir_function(name); + + ir_function_signature *linked_sig = + f->exact_matching_signature(&callee->parameters); + if (linked_sig == NULL) { + linked_sig = new(linked) ir_function_signature(callee->return_type); + f->add_signature(linked_sig); + } + + /* At this point linked_sig and called may be the same. If ir is an + * ir_call from linked then linked_sig and callee will be + * ir_function_signatures that have no definitions (is_defined is false). + */ + assert(!linked_sig->is_defined); + assert(linked_sig->body.is_empty()); + + /* Create an in-place clone of the function definition. This multistep + * process introduces some complexity here, but it has some advantages. + * The parameter list and the and function body are cloned separately. + * The clone of the parameter list is used to prime the hashtable used + * to replace variable references in the cloned body. + * + * The big advantage is that the ir_function_signature does not change. + * This means that we don't have to process the rest of the IR tree to + * patch ir_call nodes. In addition, there is no way to remove or + * replace signature stored in a function. One could easily be added, + * but this avoids the need. + */ + struct hash_table *ht = hash_table_ctor(0, hash_table_pointer_hash, + hash_table_pointer_compare); + exec_list formal_parameters; + foreach_list_const(node, &sig->parameters) { + const ir_instruction *const original = (ir_instruction *) node; + assert(const_cast(original)->as_variable()); + + ir_instruction *copy = original->clone(linked, ht); + formal_parameters.push_tail(copy); + } + + linked_sig->replace_parameters(&formal_parameters); + + foreach_list_const(node, &sig->body) { + const ir_instruction *const original = (ir_instruction *) node; + + ir_instruction *copy = original->clone(linked, ht); + linked_sig->body.push_tail(copy); + } + + linked_sig->is_defined = true; + hash_table_dtor(ht); + + /* Patch references inside the function to things outside the function + * (i.e., function calls and global variables). + */ + linked_sig->accept(this); + + ir->set_callee(linked_sig); + + return visit_continue; + } + + virtual ir_visitor_status visit(ir_dereference_variable *ir) + { + if (hash_table_find(locals, ir->var) == NULL) { + /* The non-function variable must be a global, so try to find the + * variable in the shader's symbol table. If the variable is not + * found, then it's a global that *MUST* be defined in the original + * shader. + */ + ir_variable *var = linked->symbols->get_variable(ir->var->name); + if (var == NULL) { + /* Clone the ir_variable that the dereference already has and add + * it to the linked shader. + */ + var = ir->var->clone(linked, NULL); + linked->symbols->add_variable(var); + linked->ir->push_head(var); + } else if (var->type->is_array()) { + /* It is possible to have a global array declared in multiple + * shaders without a size. The array is implicitly sized by the + * maximal access to it in *any* shader. Because of this, we + * need to track the maximal access to the array as linking pulls + * more functions in that access the array. + */ + var->max_array_access = + MAX2(var->max_array_access, ir->var->max_array_access); + + if (var->type->length == 0 && ir->var->type->length != 0) + var->type = ir->var->type; + } + + ir->var = var; + } + + return visit_continue; + } + + /** Was function linking successful? */ + bool success; + +private: + /** + * Shader program being linked + * + * This is only used for logging error messages. + */ + gl_shader_program *prog; + + /** List of shaders available for linking. */ + gl_shader **shader_list; + + /** Number of shaders available for linking. */ + unsigned num_shaders; + + /** + * Final linked shader + * + * This is used two ways. It is used to find global variables in the + * linked shader that are accessed by the function. It is also used to add + * global variables from the shader where the function originated. + */ + gl_shader *linked; + + /** + * Table of variables local to the function. + */ + hash_table *locals; +}; + + +/** + * Searches a list of shaders for a particular function definition + */ +ir_function_signature * +find_matching_signature(const char *name, const exec_list *actual_parameters, + gl_shader **shader_list, unsigned num_shaders) +{ + for (unsigned i = 0; i < num_shaders; i++) { + ir_function *const f = shader_list[i]->symbols->get_function(name); + + if (f == NULL) + continue; + + ir_function_signature *sig = f->matching_signature(actual_parameters); + + if ((sig == NULL) || !sig->is_defined) + continue; + + return sig; + } + + return NULL; +} + + +bool +link_function_calls(gl_shader_program *prog, gl_shader *main, + gl_shader **shader_list, unsigned num_shaders) +{ + call_link_visitor v(prog, main, shader_list, num_shaders); + + v.run(main->ir); + return v.success; +} diff --git a/mesalib/src/glsl/linker.cpp b/mesalib/src/glsl/linker.cpp index 9f4b4149c..bf7a56353 100644 --- a/mesalib/src/glsl/linker.cpp +++ b/mesalib/src/glsl/linker.cpp @@ -1,1676 +1,1672 @@ -/* - * 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 linker.cpp - * GLSL linker implementation - * - * Given a set of shaders that are to be linked to generate a final program, - * there are three distinct stages. - * - * In the first stage shaders are partitioned into groups based on the shader - * type. All shaders of a particular type (e.g., vertex shaders) are linked - * together. - * - * - Undefined references in each shader are resolve to definitions in - * another shader. - * - Types and qualifiers of uniforms, outputs, and global variables defined - * in multiple shaders with the same name are verified to be the same. - * - Initializers for uniforms and global variables defined - * in multiple shaders with the same name are verified to be the same. - * - * The result, in the terminology of the GLSL spec, is a set of shader - * executables for each processing unit. - * - * After the first stage is complete, a series of semantic checks are performed - * on each of the shader executables. - * - * - Each shader executable must define a \c main function. - * - Each vertex shader executable must write to \c gl_Position. - * - Each fragment shader executable must write to either \c gl_FragData or - * \c gl_FragColor. - * - * In the final stage individual shader executables are linked to create a - * complete exectuable. - * - * - Types of uniforms defined in multiple shader stages with the same name - * are verified to be the same. - * - Initializers for uniforms defined in multiple shader stages with the - * same name are verified to be the same. - * - Types and qualifiers of outputs defined in one stage are verified to - * be the same as the types and qualifiers of inputs defined with the same - * name in a later stage. - * - * \author Ian Romanick - */ -#include -#include -#include -#include - -#include "main/core.h" -#include "glsl_symbol_table.h" -#include "ir.h" -#include "program.h" -#include "program/hash_table.h" -#include "linker.h" -#include "ir_optimization.h" - -extern "C" { -#include "main/shaderobj.h" -} - -/** - * Visitor that determines whether or not a variable is ever written. - */ -class find_assignment_visitor : public ir_hierarchical_visitor { -public: - find_assignment_visitor(const char *name) - : name(name), found(false) - { - /* empty */ - } - - virtual ir_visitor_status visit_enter(ir_assignment *ir) - { - ir_variable *const var = ir->lhs->variable_referenced(); - - if (strcmp(name, var->name) == 0) { - found = true; - return visit_stop; - } - - return visit_continue_with_parent; - } - - virtual ir_visitor_status visit_enter(ir_call *ir) - { - exec_list_iterator sig_iter = ir->get_callee()->parameters.iterator(); - foreach_iter(exec_list_iterator, iter, *ir) { - ir_rvalue *param_rval = (ir_rvalue *)iter.get(); - ir_variable *sig_param = (ir_variable *)sig_iter.get(); - - if (sig_param->mode == ir_var_out || - sig_param->mode == ir_var_inout) { - ir_variable *var = param_rval->variable_referenced(); - if (var && strcmp(name, var->name) == 0) { - found = true; - return visit_stop; - } - } - sig_iter.next(); - } - - return visit_continue_with_parent; - } - - bool variable_found() - { - return found; - } - -private: - const char *name; /**< Find writes to a variable with this name. */ - bool found; /**< Was a write to the variable found? */ -}; - - -/** - * Visitor that determines whether or not a variable is ever read. - */ -class find_deref_visitor : public ir_hierarchical_visitor { -public: - find_deref_visitor(const char *name) - : name(name), found(false) - { - /* empty */ - } - - virtual ir_visitor_status visit(ir_dereference_variable *ir) - { - if (strcmp(this->name, ir->var->name) == 0) { - this->found = true; - return visit_stop; - } - - return visit_continue; - } - - bool variable_found() const - { - return this->found; - } - -private: - const char *name; /**< Find writes to a variable with this name. */ - bool found; /**< Was a write to the variable found? */ -}; - - -void -linker_error_printf(gl_shader_program *prog, const char *fmt, ...) -{ - va_list ap; - - prog->InfoLog = talloc_strdup_append(prog->InfoLog, "error: "); - va_start(ap, fmt); - prog->InfoLog = talloc_vasprintf_append(prog->InfoLog, fmt, ap); - va_end(ap); -} - - -void -invalidate_variable_locations(gl_shader *sh, enum ir_variable_mode mode, - int generic_base) -{ - foreach_list(node, sh->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != (unsigned) mode)) - continue; - - /* Only assign locations for generic attributes / varyings / etc. - */ - if ((var->location >= generic_base) && !var->explicit_location) - var->location = -1; - } -} - - -/** - * Determine the number of attribute slots required for a particular type - * - * This code is here because it implements the language rules of a specific - * GLSL version. Since it's a property of the language and not a property of - * types in general, it doesn't really belong in glsl_type. - */ -unsigned -count_attribute_slots(const glsl_type *t) -{ - /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec: - * - * "A scalar input counts the same amount against this limit as a vec4, - * so applications may want to consider packing groups of four - * unrelated float inputs together into a vector to better utilize the - * capabilities of the underlying hardware. A matrix input will use up - * multiple locations. The number of locations used will equal the - * number of columns in the matrix." - * - * The spec does not explicitly say how arrays are counted. However, it - * should be safe to assume the total number of slots consumed by an array - * is the number of entries in the array multiplied by the number of slots - * consumed by a single element of the array. - */ - - if (t->is_array()) - return t->array_size() * count_attribute_slots(t->element_type()); - - if (t->is_matrix()) - return t->matrix_columns; - - return 1; -} - - -/** - * Verify that a vertex shader executable meets all semantic requirements - * - * \param shader Vertex shader executable to be verified - */ -bool -validate_vertex_shader_executable(struct gl_shader_program *prog, - struct gl_shader *shader) -{ - if (shader == NULL) - return true; - - find_assignment_visitor find("gl_Position"); - find.run(shader->ir); - if (!find.variable_found()) { - linker_error_printf(prog, - "vertex shader does not write to `gl_Position'\n"); - return false; - } - - return true; -} - - -/** - * Verify that a fragment shader executable meets all semantic requirements - * - * \param shader Fragment shader executable to be verified - */ -bool -validate_fragment_shader_executable(struct gl_shader_program *prog, - struct gl_shader *shader) -{ - if (shader == NULL) - return true; - - find_assignment_visitor frag_color("gl_FragColor"); - find_assignment_visitor frag_data("gl_FragData"); - - frag_color.run(shader->ir); - frag_data.run(shader->ir); - - if (frag_color.variable_found() && frag_data.variable_found()) { - linker_error_printf(prog, "fragment shader writes to both " - "`gl_FragColor' and `gl_FragData'\n"); - return false; - } - - return true; -} - - -/** - * Generate a string describing the mode of a variable - */ -static const char * -mode_string(const ir_variable *var) -{ - switch (var->mode) { - case ir_var_auto: - return (var->read_only) ? "global constant" : "global variable"; - - case ir_var_uniform: return "uniform"; - case ir_var_in: return "shader input"; - case ir_var_out: return "shader output"; - case ir_var_inout: return "shader inout"; - - case ir_var_temporary: - default: - assert(!"Should not get here."); - return "invalid variable"; - } -} - - -/** - * Perform validation of global variables used across multiple shaders - */ -bool -cross_validate_globals(struct gl_shader_program *prog, - struct gl_shader **shader_list, - unsigned num_shaders, - bool uniforms_only) -{ - /* Examine all of the uniforms in all of the shaders and cross validate - * them. - */ - glsl_symbol_table variables; - for (unsigned i = 0; i < num_shaders; i++) { - if (shader_list[i] == NULL) - continue; - - foreach_list(node, shader_list[i]->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if (var == NULL) - continue; - - if (uniforms_only && (var->mode != ir_var_uniform)) - continue; - - /* Don't cross validate temporaries that are at global scope. These - * will eventually get pulled into the shaders 'main'. - */ - if (var->mode == ir_var_temporary) - continue; - - /* If a global with this name has already been seen, verify that the - * new instance has the same type. In addition, if the globals have - * initializers, the values of the initializers must be the same. - */ - ir_variable *const existing = variables.get_variable(var->name); - if (existing != NULL) { - if (var->type != existing->type) { - /* Consider the types to be "the same" if both types are arrays - * of the same type and one of the arrays is implicitly sized. - * In addition, set the type of the linked variable to the - * explicitly sized array. - */ - if (var->type->is_array() - && existing->type->is_array() - && (var->type->fields.array == existing->type->fields.array) - && ((var->type->length == 0) - || (existing->type->length == 0))) { - if (existing->type->length == 0) { - existing->type = var->type; - existing->max_array_access = - MAX2(existing->max_array_access, - var->max_array_access); - } - } else { - linker_error_printf(prog, "%s `%s' declared as type " - "`%s' and type `%s'\n", - mode_string(var), - var->name, var->type->name, - existing->type->name); - return false; - } - } - - if (var->explicit_location) { - if (existing->explicit_location - && (var->location != existing->location)) { - linker_error_printf(prog, "explicit locations for %s " - "`%s' have differing values\n", - mode_string(var), var->name); - return false; - } - - existing->location = var->location; - existing->explicit_location = true; - } - - /* FINISHME: Handle non-constant initializers. - */ - if (var->constant_value != NULL) { - if (existing->constant_value != NULL) { - if (!var->constant_value->has_value(existing->constant_value)) { - linker_error_printf(prog, "initializers for %s " - "`%s' have differing values\n", - mode_string(var), var->name); - return false; - } - } else - /* If the first-seen instance of a particular uniform did not - * have an initializer but a later instance does, copy the - * initializer to the version stored in the symbol table. - */ - /* FINISHME: This is wrong. The constant_value field should - * FINISHME: not be modified! Imagine a case where a shader - * FINISHME: without an initializer is linked in two different - * FINISHME: programs with shaders that have differing - * FINISHME: initializers. Linking with the first will - * FINISHME: modify the shader, and linking with the second - * FINISHME: will fail. - */ - existing->constant_value = - var->constant_value->clone(talloc_parent(existing), NULL); - } - - if (existing->invariant != var->invariant) { - linker_error_printf(prog, "declarations for %s `%s' have " - "mismatching invariant qualifiers\n", - mode_string(var), var->name); - return false; - } - if (existing->centroid != var->centroid) { - linker_error_printf(prog, "declarations for %s `%s' have " - "mismatching centroid qualifiers\n", - mode_string(var), var->name); - return false; - } - } else - variables.add_variable(var); - } - } - - return true; -} - - -/** - * Perform validation of uniforms used across multiple shader stages - */ -bool -cross_validate_uniforms(struct gl_shader_program *prog) -{ - return cross_validate_globals(prog, prog->_LinkedShaders, - MESA_SHADER_TYPES, true); -} - - -/** - * Validate that outputs from one stage match inputs of another - */ -bool -cross_validate_outputs_to_inputs(struct gl_shader_program *prog, - gl_shader *producer, gl_shader *consumer) -{ - glsl_symbol_table parameters; - /* FINISHME: Figure these out dynamically. */ - const char *const producer_stage = "vertex"; - const char *const consumer_stage = "fragment"; - - /* Find all shader outputs in the "producer" stage. - */ - foreach_list(node, producer->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - /* FINISHME: For geometry shaders, this should also look for inout - * FINISHME: variables. - */ - if ((var == NULL) || (var->mode != ir_var_out)) - continue; - - parameters.add_variable(var); - } - - - /* Find all shader inputs in the "consumer" stage. Any variables that have - * matching outputs already in the symbol table must have the same type and - * qualifiers. - */ - foreach_list(node, consumer->ir) { - ir_variable *const input = ((ir_instruction *) node)->as_variable(); - - /* FINISHME: For geometry shaders, this should also look for inout - * FINISHME: variables. - */ - if ((input == NULL) || (input->mode != ir_var_in)) - continue; - - ir_variable *const output = parameters.get_variable(input->name); - if (output != NULL) { - /* Check that the types match between stages. - */ - if (input->type != output->type) { - /* There is a bit of a special case for gl_TexCoord. This - * built-in is unsized by default. Appliations that variable - * access it must redeclare it with a size. There is some - * language in the GLSL spec that implies the fragment shader - * and vertex shader do not have to agree on this size. Other - * driver behave this way, and one or two applications seem to - * rely on it. - * - * Neither declaration needs to be modified here because the array - * sizes are fixed later when update_array_sizes is called. - * - * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec: - * - * "Unlike user-defined varying variables, the built-in - * varying variables don't have a strict one-to-one - * correspondence between the vertex language and the - * fragment language." - */ - if (!output->type->is_array() - || (strncmp("gl_", output->name, 3) != 0)) { - linker_error_printf(prog, - "%s shader output `%s' declared as " - "type `%s', but %s shader input declared " - "as type `%s'\n", - producer_stage, output->name, - output->type->name, - consumer_stage, input->type->name); - return false; - } - } - - /* Check that all of the qualifiers match between stages. - */ - if (input->centroid != output->centroid) { - linker_error_printf(prog, - "%s shader output `%s' %s centroid qualifier, " - "but %s shader input %s centroid qualifier\n", - producer_stage, - output->name, - (output->centroid) ? "has" : "lacks", - consumer_stage, - (input->centroid) ? "has" : "lacks"); - return false; - } - - if (input->invariant != output->invariant) { - linker_error_printf(prog, - "%s shader output `%s' %s invariant qualifier, " - "but %s shader input %s invariant qualifier\n", - producer_stage, - output->name, - (output->invariant) ? "has" : "lacks", - consumer_stage, - (input->invariant) ? "has" : "lacks"); - return false; - } - - if (input->interpolation != output->interpolation) { - linker_error_printf(prog, - "%s shader output `%s' specifies %s " - "interpolation qualifier, " - "but %s shader input specifies %s " - "interpolation qualifier\n", - producer_stage, - output->name, - output->interpolation_string(), - consumer_stage, - input->interpolation_string()); - return false; - } - } - } - - return true; -} - - -/** - * Populates a shaders symbol table with all global declarations - */ -static void -populate_symbol_table(gl_shader *sh) -{ - sh->symbols = new(sh) glsl_symbol_table; - - foreach_list(node, sh->ir) { - ir_instruction *const inst = (ir_instruction *) node; - ir_variable *var; - ir_function *func; - - if ((func = inst->as_function()) != NULL) { - sh->symbols->add_function(func); - } else if ((var = inst->as_variable()) != NULL) { - sh->symbols->add_variable(var); - } - } -} - - -/** - * Remap variables referenced in an instruction tree - * - * This is used when instruction trees are cloned from one shader and placed in - * another. These trees will contain references to \c ir_variable nodes that - * do not exist in the target shader. This function finds these \c ir_variable - * references and replaces the references with matching variables in the target - * shader. - * - * If there is no matching variable in the target shader, a clone of the - * \c ir_variable is made and added to the target shader. The new variable is - * added to \b both the instruction stream and the symbol table. - * - * \param inst IR tree that is to be processed. - * \param symbols Symbol table containing global scope symbols in the - * linked shader. - * \param instructions Instruction stream where new variable declarations - * should be added. - */ -void -remap_variables(ir_instruction *inst, struct gl_shader *target, - hash_table *temps) -{ - class remap_visitor : public ir_hierarchical_visitor { - public: - remap_visitor(struct gl_shader *target, - hash_table *temps) - { - this->target = target; - this->symbols = target->symbols; - this->instructions = target->ir; - this->temps = temps; - } - - virtual ir_visitor_status visit(ir_dereference_variable *ir) - { - if (ir->var->mode == ir_var_temporary) { - ir_variable *var = (ir_variable *) hash_table_find(temps, ir->var); - - assert(var != NULL); - ir->var = var; - return visit_continue; - } - - ir_variable *const existing = - this->symbols->get_variable(ir->var->name); - if (existing != NULL) - ir->var = existing; - else { - ir_variable *copy = ir->var->clone(this->target, NULL); - - this->symbols->add_variable(copy); - this->instructions->push_head(copy); - ir->var = copy; - } - - return visit_continue; - } - - private: - struct gl_shader *target; - glsl_symbol_table *symbols; - exec_list *instructions; - hash_table *temps; - }; - - remap_visitor v(target, temps); - - inst->accept(&v); -} - - -/** - * Move non-declarations from one instruction stream to another - * - * The intended usage pattern of this function is to pass the pointer to the - * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node - * pointer) for \c last and \c false for \c make_copies on the first - * call. Successive calls pass the return value of the previous call for - * \c last and \c true for \c make_copies. - * - * \param instructions Source instruction stream - * \param last Instruction after which new instructions should be - * inserted in the target instruction stream - * \param make_copies Flag selecting whether instructions in \c instructions - * should be copied (via \c ir_instruction::clone) into the - * target list or moved. - * - * \return - * The new "last" instruction in the target instruction stream. This pointer - * is suitable for use as the \c last parameter of a later call to this - * function. - */ -exec_node * -move_non_declarations(exec_list *instructions, exec_node *last, - bool make_copies, gl_shader *target) -{ - hash_table *temps = NULL; - - if (make_copies) - temps = hash_table_ctor(0, hash_table_pointer_hash, - hash_table_pointer_compare); - - foreach_list_safe(node, instructions) { - ir_instruction *inst = (ir_instruction *) node; - - if (inst->as_function()) - continue; - - ir_variable *var = inst->as_variable(); - if ((var != NULL) && (var->mode != ir_var_temporary)) - continue; - - assert(inst->as_assignment() - || ((var != NULL) && (var->mode == ir_var_temporary))); - - if (make_copies) { - inst = inst->clone(target, NULL); - - if (var != NULL) - hash_table_insert(temps, inst, var); - else - remap_variables(inst, target, temps); - } else { - inst->remove(); - } - - last->insert_after(inst); - last = inst; - } - - if (make_copies) - hash_table_dtor(temps); - - return last; -} - -/** - * Get the function signature for main from a shader - */ -static ir_function_signature * -get_main_function_signature(gl_shader *sh) -{ - ir_function *const f = sh->symbols->get_function("main"); - if (f != NULL) { - exec_list void_parameters; - - /* Look for the 'void main()' signature and ensure that it's defined. - * This keeps the linker from accidentally pick a shader that just - * contains a prototype for main. - * - * We don't have to check for multiple definitions of main (in multiple - * shaders) because that would have already been caught above. - */ - ir_function_signature *sig = f->matching_signature(&void_parameters); - if ((sig != NULL) && sig->is_defined) { - return sig; - } - } - - return NULL; -} - - -/** - * Combine a group of shaders for a single stage to generate a linked shader - * - * \note - * If this function is supplied a single shader, it is cloned, and the new - * shader is returned. - */ -static struct gl_shader * -link_intrastage_shaders(void *mem_ctx, - struct gl_context *ctx, - struct gl_shader_program *prog, - struct gl_shader **shader_list, - unsigned num_shaders) -{ - /* Check that global variables defined in multiple shaders are consistent. - */ - if (!cross_validate_globals(prog, shader_list, num_shaders, false)) - return NULL; - - /* Check that there is only a single definition of each function signature - * across all shaders. - */ - for (unsigned i = 0; i < (num_shaders - 1); i++) { - foreach_list(node, shader_list[i]->ir) { - ir_function *const f = ((ir_instruction *) node)->as_function(); - - if (f == NULL) - continue; - - for (unsigned j = i + 1; j < num_shaders; j++) { - ir_function *const other = - shader_list[j]->symbols->get_function(f->name); - - /* If the other shader has no function (and therefore no function - * signatures) with the same name, skip to the next shader. - */ - if (other == NULL) - continue; - - foreach_iter (exec_list_iterator, iter, *f) { - ir_function_signature *sig = - (ir_function_signature *) iter.get(); - - if (!sig->is_defined || sig->is_builtin) - continue; - - ir_function_signature *other_sig = - other->exact_matching_signature(& sig->parameters); - - if ((other_sig != NULL) && other_sig->is_defined - && !other_sig->is_builtin) { - linker_error_printf(prog, - "function `%s' is multiply defined", - f->name); - return NULL; - } - } - } - } - } - - /* Find the shader that defines main, and make a clone of it. - * - * Starting with the clone, search for undefined references. If one is - * found, find the shader that defines it. Clone the reference and add - * it to the shader. Repeat until there are no undefined references or - * until a reference cannot be resolved. - */ - gl_shader *main = NULL; - for (unsigned i = 0; i < num_shaders; i++) { - if (get_main_function_signature(shader_list[i]) != NULL) { - main = shader_list[i]; - break; - } - } - - if (main == NULL) { - linker_error_printf(prog, "%s shader lacks `main'\n", - (shader_list[0]->Type == GL_VERTEX_SHADER) - ? "vertex" : "fragment"); - return NULL; - } - - gl_shader *linked = ctx->Driver.NewShader(NULL, 0, main->Type); - linked->ir = new(linked) exec_list; - clone_ir_list(mem_ctx, linked->ir, main->ir); - - populate_symbol_table(linked); - - /* The a pointer to the main function in the final linked shader (i.e., the - * copy of the original shader that contained the main function). - */ - ir_function_signature *const main_sig = get_main_function_signature(linked); - - /* Move any instructions other than variable declarations or function - * declarations into main. - */ - exec_node *insertion_point = - move_non_declarations(linked->ir, (exec_node *) &main_sig->body, false, - linked); - - for (unsigned i = 0; i < num_shaders; i++) { - if (shader_list[i] == main) - continue; - - insertion_point = move_non_declarations(shader_list[i]->ir, - insertion_point, true, linked); - } - - /* Resolve initializers for global variables in the linked shader. - */ - unsigned num_linking_shaders = num_shaders; - for (unsigned i = 0; i < num_shaders; i++) - num_linking_shaders += shader_list[i]->num_builtins_to_link; - - gl_shader **linking_shaders = - (gl_shader **) calloc(num_linking_shaders, sizeof(gl_shader *)); - - memcpy(linking_shaders, shader_list, - sizeof(linking_shaders[0]) * num_shaders); - - unsigned idx = num_shaders; - for (unsigned i = 0; i < num_shaders; i++) { - memcpy(&linking_shaders[idx], shader_list[i]->builtins_to_link, - sizeof(linking_shaders[0]) * shader_list[i]->num_builtins_to_link); - idx += shader_list[i]->num_builtins_to_link; - } - - assert(idx == num_linking_shaders); - - if (!link_function_calls(prog, linked, linking_shaders, - num_linking_shaders)) { - ctx->Driver.DeleteShader(ctx, linked); - linked = NULL; - } - - free(linking_shaders); - - /* Make a pass over all global variables to ensure that arrays with - * unspecified sizes have a size specified. The size is inferred from the - * max_array_access field. - */ - if (linked != NULL) { - foreach_list(node, linked->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if (var == NULL) - continue; - - if ((var->mode != ir_var_auto) && (var->mode != ir_var_temporary)) - continue; - - if (!var->type->is_array() || (var->type->length != 0)) - continue; - - const glsl_type *type = - glsl_type::get_array_instance(var->type->fields.array, - var->max_array_access); - - assert(type != NULL); - var->type = type; - } - } - - return linked; -} - - -struct uniform_node { - exec_node link; - struct gl_uniform *u; - unsigned slots; -}; - -/** - * Update the sizes of linked shader uniform arrays to the maximum - * array index used. - * - * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec: - * - * If one or more elements of an array are active, - * GetActiveUniform will return the name of the array in name, - * subject to the restrictions listed above. The type of the array - * is returned in type. The size parameter contains the highest - * array element index used, plus one. The compiler or linker - * determines the highest index used. There will be only one - * active uniform reported by the GL per uniform array. - - */ -static void -update_array_sizes(struct gl_shader_program *prog) -{ - for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - foreach_list(node, prog->_LinkedShaders[i]->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != ir_var_uniform && - var->mode != ir_var_in && - var->mode != ir_var_out) || - !var->type->is_array()) - continue; - - unsigned int size = var->max_array_access; - for (unsigned j = 0; j < MESA_SHADER_TYPES; j++) { - if (prog->_LinkedShaders[j] == NULL) - continue; - - foreach_list(node2, prog->_LinkedShaders[j]->ir) { - ir_variable *other_var = ((ir_instruction *) node2)->as_variable(); - if (!other_var) - continue; - - if (strcmp(var->name, other_var->name) == 0 && - other_var->max_array_access > size) { - size = other_var->max_array_access; - } - } - } - - if (size + 1 != var->type->fields.array->length) { - var->type = glsl_type::get_array_instance(var->type->fields.array, - size + 1); - /* FINISHME: We should update the types of array - * dereferences of this variable now. - */ - } - } - } -} - -static void -add_uniform(void *mem_ctx, exec_list *uniforms, struct hash_table *ht, - const char *name, const glsl_type *type, GLenum shader_type, - unsigned *next_shader_pos, unsigned *total_uniforms) -{ - if (type->is_record()) { - for (unsigned int i = 0; i < type->length; i++) { - const glsl_type *field_type = type->fields.structure[i].type; - char *field_name = talloc_asprintf(mem_ctx, "%s.%s", name, - type->fields.structure[i].name); - - add_uniform(mem_ctx, uniforms, ht, field_name, field_type, - shader_type, next_shader_pos, total_uniforms); - } - } else { - uniform_node *n = (uniform_node *) hash_table_find(ht, name); - unsigned int vec4_slots; - const glsl_type *array_elem_type = NULL; - - if (type->is_array()) { - array_elem_type = type->fields.array; - /* Array of structures. */ - if (array_elem_type->is_record()) { - for (unsigned int i = 0; i < type->length; i++) { - char *elem_name = talloc_asprintf(mem_ctx, "%s[%d]", name, i); - add_uniform(mem_ctx, uniforms, ht, elem_name, array_elem_type, - shader_type, next_shader_pos, total_uniforms); - } - return; - } - } - - /* Fix the storage size of samplers at 1 vec4 each. Be sure to pad out - * vectors to vec4 slots. - */ - if (type->is_array()) { - if (array_elem_type->is_sampler()) - vec4_slots = type->length; - else - vec4_slots = type->length * array_elem_type->matrix_columns; - } else if (type->is_sampler()) { - vec4_slots = 1; - } else { - vec4_slots = type->matrix_columns; - } - - if (n == NULL) { - n = (uniform_node *) calloc(1, sizeof(struct uniform_node)); - n->u = (gl_uniform *) calloc(1, sizeof(struct gl_uniform)); - n->slots = vec4_slots; - - n->u->Name = strdup(name); - n->u->Type = type; - n->u->VertPos = -1; - n->u->FragPos = -1; - n->u->GeomPos = -1; - (*total_uniforms)++; - - hash_table_insert(ht, n, name); - uniforms->push_tail(& n->link); - } - - switch (shader_type) { - case GL_VERTEX_SHADER: - n->u->VertPos = *next_shader_pos; - break; - case GL_FRAGMENT_SHADER: - n->u->FragPos = *next_shader_pos; - break; - case GL_GEOMETRY_SHADER: - n->u->GeomPos = *next_shader_pos; - break; - } - - (*next_shader_pos) += vec4_slots; - } -} - -void -assign_uniform_locations(struct gl_shader_program *prog) -{ - /* */ - exec_list uniforms; - unsigned total_uniforms = 0; - hash_table *ht = hash_table_ctor(32, hash_table_string_hash, - hash_table_string_compare); - void *mem_ctx = talloc_new(NULL); - - for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - unsigned next_position = 0; - - foreach_list(node, prog->_LinkedShaders[i]->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != ir_var_uniform)) - continue; - - if (strncmp(var->name, "gl_", 3) == 0) { - /* At the moment, we don't allocate uniform locations for - * builtin uniforms. It's permitted by spec, and we'll - * likely switch to doing that at some point, but not yet. - */ - continue; - } - - var->location = next_position; - add_uniform(mem_ctx, &uniforms, ht, var->name, var->type, - prog->_LinkedShaders[i]->Type, - &next_position, &total_uniforms); - } - } - - talloc_free(mem_ctx); - - gl_uniform_list *ul = (gl_uniform_list *) - calloc(1, sizeof(gl_uniform_list)); - - ul->Size = total_uniforms; - ul->NumUniforms = total_uniforms; - ul->Uniforms = (gl_uniform *) calloc(total_uniforms, sizeof(gl_uniform)); - - unsigned idx = 0; - uniform_node *next; - for (uniform_node *node = (uniform_node *) uniforms.head - ; node->link.next != NULL - ; node = next) { - next = (uniform_node *) node->link.next; - - node->link.remove(); - memcpy(&ul->Uniforms[idx], node->u, sizeof(gl_uniform)); - idx++; - - free(node->u); - free(node); - } - - hash_table_dtor(ht); - - prog->Uniforms = ul; -} - - -/** - * Find a contiguous set of available bits in a bitmask - * - * \param used_mask Bits representing used (1) and unused (0) locations - * \param needed_count Number of contiguous bits needed. - * - * \return - * Base location of the available bits on success or -1 on failure. - */ -int -find_available_slots(unsigned used_mask, unsigned needed_count) -{ - unsigned needed_mask = (1 << needed_count) - 1; - const int max_bit_to_test = (8 * sizeof(used_mask)) - needed_count; - - /* The comparison to 32 is redundant, but without it GCC emits "warning: - * cannot optimize possibly infinite loops" for the loop below. - */ - if ((needed_count == 0) || (max_bit_to_test < 0) || (max_bit_to_test > 32)) - return -1; - - for (int i = 0; i <= max_bit_to_test; i++) { - if ((needed_mask & ~used_mask) == needed_mask) - return i; - - needed_mask <<= 1; - } - - return -1; -} - - -bool -assign_attribute_locations(gl_shader_program *prog, unsigned max_attribute_index) -{ - /* Mark invalid attribute locations as being used. - */ - unsigned used_locations = (max_attribute_index >= 32) - ? ~0 : ~((1 << max_attribute_index) - 1); - - gl_shader *const sh = prog->_LinkedShaders[0]; - assert(sh->Type == GL_VERTEX_SHADER); - - /* Operate in a total of four passes. - * - * 1. Invalidate the location assignments for all vertex shader inputs. - * - * 2. Assign locations for inputs that have user-defined (via - * glBindVertexAttribLocation) locatoins. - * - * 3. Sort the attributes without assigned locations by number of slots - * required in decreasing order. Fragmentation caused by attribute - * locations assigned by the application may prevent large attributes - * from having enough contiguous space. - * - * 4. Assign locations to any inputs without assigned locations. - */ - - invalidate_variable_locations(sh, ir_var_in, VERT_ATTRIB_GENERIC0); - - if (prog->Attributes != NULL) { - for (unsigned i = 0; i < prog->Attributes->NumParameters; i++) { - ir_variable *const var = - sh->symbols->get_variable(prog->Attributes->Parameters[i].Name); - - /* Note: attributes that occupy multiple slots, such as arrays or - * matrices, may appear in the attrib array multiple times. - */ - if ((var == NULL) || (var->location != -1)) - continue; - - /* From page 61 of the OpenGL 4.0 spec: - * - * "LinkProgram will fail if the attribute bindings assigned by - * BindAttribLocation do not leave not enough space to assign a - * location for an active matrix attribute or an active attribute - * array, both of which require multiple contiguous generic - * attributes." - * - * Previous versions of the spec contain similar language but omit the - * bit about attribute arrays. - * - * Page 61 of the OpenGL 4.0 spec also says: - * - * "It is possible for an application to bind more than one - * attribute name to the same location. This is referred to as - * aliasing. This will only work if only one of the aliased - * attributes is active in the executable program, or if no path - * through the shader consumes more than one attribute of a set - * of attributes aliased to the same location. A link error can - * occur if the linker determines that every path through the - * shader consumes multiple aliased attributes, but - * implementations are not required to generate an error in this - * case." - * - * These two paragraphs are either somewhat contradictory, or I don't - * fully understand one or both of them. - */ - /* FINISHME: The code as currently written does not support attribute - * FINISHME: location aliasing (see comment above). - */ - const int attr = prog->Attributes->Parameters[i].StateIndexes[0]; - const unsigned slots = count_attribute_slots(var->type); - - /* Mask representing the contiguous slots that will be used by this - * attribute. - */ - const unsigned use_mask = (1 << slots) - 1; - - /* Generate a link error if the set of bits requested for this - * attribute overlaps any previously allocated bits. - */ - if ((~(use_mask << attr) & used_locations) != used_locations) { - linker_error_printf(prog, - "insufficient contiguous attribute locations " - "available for vertex shader input `%s'", - var->name); - return false; - } - - var->location = VERT_ATTRIB_GENERIC0 + attr; - used_locations |= (use_mask << attr); - } - } - - /* Temporary storage for the set of attributes that need locations assigned. - */ - struct temp_attr { - unsigned slots; - ir_variable *var; - - /* Used below in the call to qsort. */ - static int compare(const void *a, const void *b) - { - const temp_attr *const l = (const temp_attr *) a; - const temp_attr *const r = (const temp_attr *) b; - - /* Reversed because we want a descending order sort below. */ - return r->slots - l->slots; - } - } to_assign[16]; - - unsigned num_attr = 0; - - foreach_list(node, sh->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != ir_var_in)) - continue; - - if (var->explicit_location) { - const unsigned slots = count_attribute_slots(var->type); - const unsigned use_mask = (1 << slots) - 1; - const int attr = var->location - VERT_ATTRIB_GENERIC0; - - if ((var->location >= (int)(max_attribute_index + VERT_ATTRIB_GENERIC0)) - || (var->location < 0)) { - linker_error_printf(prog, - "invalid explicit location %d specified for " - "`%s'\n", - (var->location < 0) ? var->location : attr, - var->name); - return false; - } else if (var->location >= VERT_ATTRIB_GENERIC0) { - used_locations |= (use_mask << attr); - } - } - - /* The location was explicitly assigned, nothing to do here. - */ - if (var->location != -1) - continue; - - to_assign[num_attr].slots = count_attribute_slots(var->type); - to_assign[num_attr].var = var; - num_attr++; - } - - /* If all of the attributes were assigned locations by the application (or - * are built-in attributes with fixed locations), return early. This should - * be the common case. - */ - if (num_attr == 0) - return true; - - qsort(to_assign, num_attr, sizeof(to_assign[0]), temp_attr::compare); - - /* VERT_ATTRIB_GENERIC0 is a psdueo-alias for VERT_ATTRIB_POS. It can only - * be explicitly assigned by via glBindAttribLocation. Mark it as reserved - * to prevent it from being automatically allocated below. - */ - find_deref_visitor find("gl_Vertex"); - find.run(sh->ir); - if (find.variable_found()) - used_locations |= (1 << 0); - - for (unsigned i = 0; i < num_attr; i++) { - /* Mask representing the contiguous slots that will be used by this - * attribute. - */ - const unsigned use_mask = (1 << to_assign[i].slots) - 1; - - int location = find_available_slots(used_locations, to_assign[i].slots); - - if (location < 0) { - linker_error_printf(prog, - "insufficient contiguous attribute locations " - "available for vertex shader input `%s'", - to_assign[i].var->name); - return false; - } - - to_assign[i].var->location = VERT_ATTRIB_GENERIC0 + location; - used_locations |= (use_mask << location); - } - - return true; -} - - -/** - * Demote shader inputs and outputs that are not used in other stages - */ -void -demote_shader_inputs_and_outputs(gl_shader *sh, enum ir_variable_mode mode) -{ - foreach_list(node, sh->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != int(mode))) - continue; - - /* A shader 'in' or 'out' variable is only really an input or output if - * its value is used by other shader stages. This will cause the variable - * to have a location assigned. - */ - if (var->location == -1) { - var->mode = ir_var_auto; - } - } -} - - -void -assign_varying_locations(struct gl_shader_program *prog, - gl_shader *producer, gl_shader *consumer) -{ - /* FINISHME: Set dynamically when geometry shader support is added. */ - unsigned output_index = VERT_RESULT_VAR0; - unsigned input_index = FRAG_ATTRIB_VAR0; - - /* Operate in a total of three passes. - * - * 1. Assign locations for any matching inputs and outputs. - * - * 2. Mark output variables in the producer that do not have locations as - * not being outputs. This lets the optimizer eliminate them. - * - * 3. Mark input variables in the consumer that do not have locations as - * not being inputs. This lets the optimizer eliminate them. - */ - - invalidate_variable_locations(producer, ir_var_out, VERT_RESULT_VAR0); - invalidate_variable_locations(consumer, ir_var_in, FRAG_ATTRIB_VAR0); - - foreach_list(node, producer->ir) { - ir_variable *const output_var = ((ir_instruction *) node)->as_variable(); - - if ((output_var == NULL) || (output_var->mode != ir_var_out) - || (output_var->location != -1)) - continue; - - ir_variable *const input_var = - consumer->symbols->get_variable(output_var->name); - - if ((input_var == NULL) || (input_var->mode != ir_var_in)) - continue; - - assert(input_var->location == -1); - - output_var->location = output_index; - input_var->location = input_index; - - /* FINISHME: Support for "varying" records in GLSL 1.50. */ - assert(!output_var->type->is_record()); - - if (output_var->type->is_array()) { - const unsigned slots = output_var->type->length - * output_var->type->fields.array->matrix_columns; - - output_index += slots; - input_index += slots; - } else { - const unsigned slots = output_var->type->matrix_columns; - - output_index += slots; - input_index += slots; - } - } - - foreach_list(node, consumer->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != ir_var_in)) - continue; - - if (var->location == -1) { - if (prog->Version <= 120) { - /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec: - * - * Only those varying variables used (i.e. read) in - * the fragment shader executable must be written to - * by the vertex shader executable; declaring - * superfluous varying variables in a vertex shader is - * permissible. - * - * We interpret this text as meaning that the VS must - * write the variable for the FS to read it. See - * "glsl1-varying read but not written" in piglit. - */ - - linker_error_printf(prog, "fragment shader varying %s not written " - "by vertex shader\n.", var->name); - prog->LinkStatus = false; - } - - /* An 'in' variable is only really a shader input if its - * value is written by the previous stage. - */ - var->mode = ir_var_auto; - } - } -} - - -void -link_shaders(struct gl_context *ctx, struct gl_shader_program *prog) -{ - void *mem_ctx = talloc_init("temporary linker context"); - - prog->LinkStatus = false; - prog->Validated = false; - prog->_Used = false; - - if (prog->InfoLog != NULL) - talloc_free(prog->InfoLog); - - prog->InfoLog = talloc_strdup(NULL, ""); - - /* Separate the shaders into groups based on their type. - */ - struct gl_shader **vert_shader_list; - unsigned num_vert_shaders = 0; - struct gl_shader **frag_shader_list; - unsigned num_frag_shaders = 0; - - vert_shader_list = (struct gl_shader **) - calloc(2 * prog->NumShaders, sizeof(struct gl_shader *)); - frag_shader_list = &vert_shader_list[prog->NumShaders]; - - unsigned min_version = UINT_MAX; - unsigned max_version = 0; - for (unsigned i = 0; i < prog->NumShaders; i++) { - min_version = MIN2(min_version, prog->Shaders[i]->Version); - max_version = MAX2(max_version, prog->Shaders[i]->Version); - - switch (prog->Shaders[i]->Type) { - case GL_VERTEX_SHADER: - vert_shader_list[num_vert_shaders] = prog->Shaders[i]; - num_vert_shaders++; - break; - case GL_FRAGMENT_SHADER: - frag_shader_list[num_frag_shaders] = prog->Shaders[i]; - num_frag_shaders++; - break; - case GL_GEOMETRY_SHADER: - /* FINISHME: Support geometry shaders. */ - assert(prog->Shaders[i]->Type != GL_GEOMETRY_SHADER); - break; - } - } - - /* Previous to GLSL version 1.30, different compilation units could mix and - * match shading language versions. With GLSL 1.30 and later, the versions - * of all shaders must match. - */ - assert(min_version >= 100); - assert(max_version <= 130); - if ((max_version >= 130 || min_version == 100) - && min_version != max_version) { - linker_error_printf(prog, "all shaders must use same shading " - "language version\n"); - goto done; - } - - prog->Version = max_version; - - for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] != NULL) - ctx->Driver.DeleteShader(ctx, prog->_LinkedShaders[i]); - - prog->_LinkedShaders[i] = NULL; - } - - /* Link all shaders for a particular stage and validate the result. - */ - if (num_vert_shaders > 0) { - gl_shader *const sh = - link_intrastage_shaders(mem_ctx, ctx, prog, vert_shader_list, - num_vert_shaders); - - if (sh == NULL) - goto done; - - if (!validate_vertex_shader_executable(prog, sh)) - goto done; - - _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_VERTEX], - sh); - } - - if (num_frag_shaders > 0) { - gl_shader *const sh = - link_intrastage_shaders(mem_ctx, ctx, prog, frag_shader_list, - num_frag_shaders); - - if (sh == NULL) - goto done; - - if (!validate_fragment_shader_executable(prog, sh)) - goto done; - - _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_FRAGMENT], - sh); - } - - /* Here begins the inter-stage linking phase. Some initial validation is - * performed, then locations are assigned for uniforms, attributes, and - * varyings. - */ - if (cross_validate_uniforms(prog)) { - unsigned prev; - - for (prev = 0; prev < MESA_SHADER_TYPES; prev++) { - if (prog->_LinkedShaders[prev] != NULL) - break; - } - - /* Validate the inputs of each stage with the output of the preceeding - * stage. - */ - for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - if (!cross_validate_outputs_to_inputs(prog, - prog->_LinkedShaders[prev], - prog->_LinkedShaders[i])) - goto done; - - prev = i; - } - - prog->LinkStatus = true; - } - - /* Do common optimization before assigning storage for attributes, - * uniforms, and varyings. Later optimization could possibly make - * some of that unused. - */ - for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, 32)) - ; - } - - update_array_sizes(prog); - - assign_uniform_locations(prog); - - if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) { - /* FINISHME: The value of the max_attribute_index parameter is - * FINISHME: implementation dependent based on the value of - * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be - * FINISHME: at least 16, so hardcode 16 for now. - */ - if (!assign_attribute_locations(prog, 16)) { - prog->LinkStatus = false; - goto done; - } - } - - unsigned prev; - for (prev = 0; prev < MESA_SHADER_TYPES; prev++) { - if (prog->_LinkedShaders[prev] != NULL) - break; - } - - for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - assign_varying_locations(prog, - prog->_LinkedShaders[prev], - prog->_LinkedShaders[i]); - prev = i; - } - - if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) { - demote_shader_inputs_and_outputs(prog->_LinkedShaders[MESA_SHADER_VERTEX], - ir_var_out); - } - - if (prog->_LinkedShaders[MESA_SHADER_GEOMETRY] != NULL) { - gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_GEOMETRY]; - - demote_shader_inputs_and_outputs(sh, ir_var_in); - demote_shader_inputs_and_outputs(sh, ir_var_inout); - demote_shader_inputs_and_outputs(sh, ir_var_out); - } - - if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] != NULL) { - gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_FRAGMENT]; - - demote_shader_inputs_and_outputs(sh, ir_var_in); - } - - /* FINISHME: Assign fragment shader output locations. */ - -done: - free(vert_shader_list); - - for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - /* Retain any live IR, but trash the rest. */ - reparent_ir(prog->_LinkedShaders[i]->ir, prog->_LinkedShaders[i]->ir); - } - - talloc_free(mem_ctx); -} +/* + * 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 linker.cpp + * GLSL linker implementation + * + * Given a set of shaders that are to be linked to generate a final program, + * there are three distinct stages. + * + * In the first stage shaders are partitioned into groups based on the shader + * type. All shaders of a particular type (e.g., vertex shaders) are linked + * together. + * + * - Undefined references in each shader are resolve to definitions in + * another shader. + * - Types and qualifiers of uniforms, outputs, and global variables defined + * in multiple shaders with the same name are verified to be the same. + * - Initializers for uniforms and global variables defined + * in multiple shaders with the same name are verified to be the same. + * + * The result, in the terminology of the GLSL spec, is a set of shader + * executables for each processing unit. + * + * After the first stage is complete, a series of semantic checks are performed + * on each of the shader executables. + * + * - Each shader executable must define a \c main function. + * - Each vertex shader executable must write to \c gl_Position. + * - Each fragment shader executable must write to either \c gl_FragData or + * \c gl_FragColor. + * + * In the final stage individual shader executables are linked to create a + * complete exectuable. + * + * - Types of uniforms defined in multiple shader stages with the same name + * are verified to be the same. + * - Initializers for uniforms defined in multiple shader stages with the + * same name are verified to be the same. + * - Types and qualifiers of outputs defined in one stage are verified to + * be the same as the types and qualifiers of inputs defined with the same + * name in a later stage. + * + * \author Ian Romanick + */ +#include +#include +#include +#include + +#include "main/core.h" +#include "glsl_symbol_table.h" +#include "ir.h" +#include "program.h" +#include "program/hash_table.h" +#include "linker.h" +#include "ir_optimization.h" + +extern "C" { +#include "main/shaderobj.h" +} + +/** + * Visitor that determines whether or not a variable is ever written. + */ +class find_assignment_visitor : public ir_hierarchical_visitor { +public: + find_assignment_visitor(const char *name) + : name(name), found(false) + { + /* empty */ + } + + virtual ir_visitor_status visit_enter(ir_assignment *ir) + { + ir_variable *const var = ir->lhs->variable_referenced(); + + if (strcmp(name, var->name) == 0) { + found = true; + return visit_stop; + } + + return visit_continue_with_parent; + } + + virtual ir_visitor_status visit_enter(ir_call *ir) + { + exec_list_iterator sig_iter = ir->get_callee()->parameters.iterator(); + foreach_iter(exec_list_iterator, iter, *ir) { + ir_rvalue *param_rval = (ir_rvalue *)iter.get(); + ir_variable *sig_param = (ir_variable *)sig_iter.get(); + + if (sig_param->mode == ir_var_out || + sig_param->mode == ir_var_inout) { + ir_variable *var = param_rval->variable_referenced(); + if (var && strcmp(name, var->name) == 0) { + found = true; + return visit_stop; + } + } + sig_iter.next(); + } + + return visit_continue_with_parent; + } + + bool variable_found() + { + return found; + } + +private: + const char *name; /**< Find writes to a variable with this name. */ + bool found; /**< Was a write to the variable found? */ +}; + + +/** + * Visitor that determines whether or not a variable is ever read. + */ +class find_deref_visitor : public ir_hierarchical_visitor { +public: + find_deref_visitor(const char *name) + : name(name), found(false) + { + /* empty */ + } + + virtual ir_visitor_status visit(ir_dereference_variable *ir) + { + if (strcmp(this->name, ir->var->name) == 0) { + this->found = true; + return visit_stop; + } + + return visit_continue; + } + + bool variable_found() const + { + return this->found; + } + +private: + const char *name; /**< Find writes to a variable with this name. */ + bool found; /**< Was a write to the variable found? */ +}; + + +void +linker_error_printf(gl_shader_program *prog, const char *fmt, ...) +{ + va_list ap; + + prog->InfoLog = talloc_strdup_append(prog->InfoLog, "error: "); + va_start(ap, fmt); + prog->InfoLog = talloc_vasprintf_append(prog->InfoLog, fmt, ap); + va_end(ap); +} + + +void +invalidate_variable_locations(gl_shader *sh, enum ir_variable_mode mode, + int generic_base) +{ + foreach_list(node, sh->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != (unsigned) mode)) + continue; + + /* Only assign locations for generic attributes / varyings / etc. + */ + if ((var->location >= generic_base) && !var->explicit_location) + var->location = -1; + } +} + + +/** + * Determine the number of attribute slots required for a particular type + * + * This code is here because it implements the language rules of a specific + * GLSL version. Since it's a property of the language and not a property of + * types in general, it doesn't really belong in glsl_type. + */ +unsigned +count_attribute_slots(const glsl_type *t) +{ + /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec: + * + * "A scalar input counts the same amount against this limit as a vec4, + * so applications may want to consider packing groups of four + * unrelated float inputs together into a vector to better utilize the + * capabilities of the underlying hardware. A matrix input will use up + * multiple locations. The number of locations used will equal the + * number of columns in the matrix." + * + * The spec does not explicitly say how arrays are counted. However, it + * should be safe to assume the total number of slots consumed by an array + * is the number of entries in the array multiplied by the number of slots + * consumed by a single element of the array. + */ + + if (t->is_array()) + return t->array_size() * count_attribute_slots(t->element_type()); + + if (t->is_matrix()) + return t->matrix_columns; + + return 1; +} + + +/** + * Verify that a vertex shader executable meets all semantic requirements + * + * \param shader Vertex shader executable to be verified + */ +bool +validate_vertex_shader_executable(struct gl_shader_program *prog, + struct gl_shader *shader) +{ + if (shader == NULL) + return true; + + find_assignment_visitor find("gl_Position"); + find.run(shader->ir); + if (!find.variable_found()) { + linker_error_printf(prog, + "vertex shader does not write to `gl_Position'\n"); + return false; + } + + return true; +} + + +/** + * Verify that a fragment shader executable meets all semantic requirements + * + * \param shader Fragment shader executable to be verified + */ +bool +validate_fragment_shader_executable(struct gl_shader_program *prog, + struct gl_shader *shader) +{ + if (shader == NULL) + return true; + + find_assignment_visitor frag_color("gl_FragColor"); + find_assignment_visitor frag_data("gl_FragData"); + + frag_color.run(shader->ir); + frag_data.run(shader->ir); + + if (frag_color.variable_found() && frag_data.variable_found()) { + linker_error_printf(prog, "fragment shader writes to both " + "`gl_FragColor' and `gl_FragData'\n"); + return false; + } + + return true; +} + + +/** + * Generate a string describing the mode of a variable + */ +static const char * +mode_string(const ir_variable *var) +{ + switch (var->mode) { + case ir_var_auto: + return (var->read_only) ? "global constant" : "global variable"; + + case ir_var_uniform: return "uniform"; + case ir_var_in: return "shader input"; + case ir_var_out: return "shader output"; + case ir_var_inout: return "shader inout"; + + case ir_var_temporary: + default: + assert(!"Should not get here."); + return "invalid variable"; + } +} + + +/** + * Perform validation of global variables used across multiple shaders + */ +bool +cross_validate_globals(struct gl_shader_program *prog, + struct gl_shader **shader_list, + unsigned num_shaders, + bool uniforms_only) +{ + /* Examine all of the uniforms in all of the shaders and cross validate + * them. + */ + glsl_symbol_table variables; + for (unsigned i = 0; i < num_shaders; i++) { + if (shader_list[i] == NULL) + continue; + + foreach_list(node, shader_list[i]->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if (var == NULL) + continue; + + if (uniforms_only && (var->mode != ir_var_uniform)) + continue; + + /* Don't cross validate temporaries that are at global scope. These + * will eventually get pulled into the shaders 'main'. + */ + if (var->mode == ir_var_temporary) + continue; + + /* If a global with this name has already been seen, verify that the + * new instance has the same type. In addition, if the globals have + * initializers, the values of the initializers must be the same. + */ + ir_variable *const existing = variables.get_variable(var->name); + if (existing != NULL) { + if (var->type != existing->type) { + /* Consider the types to be "the same" if both types are arrays + * of the same type and one of the arrays is implicitly sized. + * In addition, set the type of the linked variable to the + * explicitly sized array. + */ + if (var->type->is_array() + && existing->type->is_array() + && (var->type->fields.array == existing->type->fields.array) + && ((var->type->length == 0) + || (existing->type->length == 0))) { + if (var->type->length != 0) { + existing->type = var->type; + } + } else { + linker_error_printf(prog, "%s `%s' declared as type " + "`%s' and type `%s'\n", + mode_string(var), + var->name, var->type->name, + existing->type->name); + return false; + } + } + + if (var->explicit_location) { + if (existing->explicit_location + && (var->location != existing->location)) { + linker_error_printf(prog, "explicit locations for %s " + "`%s' have differing values\n", + mode_string(var), var->name); + return false; + } + + existing->location = var->location; + existing->explicit_location = true; + } + + /* FINISHME: Handle non-constant initializers. + */ + if (var->constant_value != NULL) { + if (existing->constant_value != NULL) { + if (!var->constant_value->has_value(existing->constant_value)) { + linker_error_printf(prog, "initializers for %s " + "`%s' have differing values\n", + mode_string(var), var->name); + return false; + } + } else + /* If the first-seen instance of a particular uniform did not + * have an initializer but a later instance does, copy the + * initializer to the version stored in the symbol table. + */ + /* FINISHME: This is wrong. The constant_value field should + * FINISHME: not be modified! Imagine a case where a shader + * FINISHME: without an initializer is linked in two different + * FINISHME: programs with shaders that have differing + * FINISHME: initializers. Linking with the first will + * FINISHME: modify the shader, and linking with the second + * FINISHME: will fail. + */ + existing->constant_value = + var->constant_value->clone(talloc_parent(existing), NULL); + } + + if (existing->invariant != var->invariant) { + linker_error_printf(prog, "declarations for %s `%s' have " + "mismatching invariant qualifiers\n", + mode_string(var), var->name); + return false; + } + if (existing->centroid != var->centroid) { + linker_error_printf(prog, "declarations for %s `%s' have " + "mismatching centroid qualifiers\n", + mode_string(var), var->name); + return false; + } + } else + variables.add_variable(var); + } + } + + return true; +} + + +/** + * Perform validation of uniforms used across multiple shader stages + */ +bool +cross_validate_uniforms(struct gl_shader_program *prog) +{ + return cross_validate_globals(prog, prog->_LinkedShaders, + MESA_SHADER_TYPES, true); +} + + +/** + * Validate that outputs from one stage match inputs of another + */ +bool +cross_validate_outputs_to_inputs(struct gl_shader_program *prog, + gl_shader *producer, gl_shader *consumer) +{ + glsl_symbol_table parameters; + /* FINISHME: Figure these out dynamically. */ + const char *const producer_stage = "vertex"; + const char *const consumer_stage = "fragment"; + + /* Find all shader outputs in the "producer" stage. + */ + foreach_list(node, producer->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + /* FINISHME: For geometry shaders, this should also look for inout + * FINISHME: variables. + */ + if ((var == NULL) || (var->mode != ir_var_out)) + continue; + + parameters.add_variable(var); + } + + + /* Find all shader inputs in the "consumer" stage. Any variables that have + * matching outputs already in the symbol table must have the same type and + * qualifiers. + */ + foreach_list(node, consumer->ir) { + ir_variable *const input = ((ir_instruction *) node)->as_variable(); + + /* FINISHME: For geometry shaders, this should also look for inout + * FINISHME: variables. + */ + if ((input == NULL) || (input->mode != ir_var_in)) + continue; + + ir_variable *const output = parameters.get_variable(input->name); + if (output != NULL) { + /* Check that the types match between stages. + */ + if (input->type != output->type) { + /* There is a bit of a special case for gl_TexCoord. This + * built-in is unsized by default. Appliations that variable + * access it must redeclare it with a size. There is some + * language in the GLSL spec that implies the fragment shader + * and vertex shader do not have to agree on this size. Other + * driver behave this way, and one or two applications seem to + * rely on it. + * + * Neither declaration needs to be modified here because the array + * sizes are fixed later when update_array_sizes is called. + * + * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec: + * + * "Unlike user-defined varying variables, the built-in + * varying variables don't have a strict one-to-one + * correspondence between the vertex language and the + * fragment language." + */ + if (!output->type->is_array() + || (strncmp("gl_", output->name, 3) != 0)) { + linker_error_printf(prog, + "%s shader output `%s' declared as " + "type `%s', but %s shader input declared " + "as type `%s'\n", + producer_stage, output->name, + output->type->name, + consumer_stage, input->type->name); + return false; + } + } + + /* Check that all of the qualifiers match between stages. + */ + if (input->centroid != output->centroid) { + linker_error_printf(prog, + "%s shader output `%s' %s centroid qualifier, " + "but %s shader input %s centroid qualifier\n", + producer_stage, + output->name, + (output->centroid) ? "has" : "lacks", + consumer_stage, + (input->centroid) ? "has" : "lacks"); + return false; + } + + if (input->invariant != output->invariant) { + linker_error_printf(prog, + "%s shader output `%s' %s invariant qualifier, " + "but %s shader input %s invariant qualifier\n", + producer_stage, + output->name, + (output->invariant) ? "has" : "lacks", + consumer_stage, + (input->invariant) ? "has" : "lacks"); + return false; + } + + if (input->interpolation != output->interpolation) { + linker_error_printf(prog, + "%s shader output `%s' specifies %s " + "interpolation qualifier, " + "but %s shader input specifies %s " + "interpolation qualifier\n", + producer_stage, + output->name, + output->interpolation_string(), + consumer_stage, + input->interpolation_string()); + return false; + } + } + } + + return true; +} + + +/** + * Populates a shaders symbol table with all global declarations + */ +static void +populate_symbol_table(gl_shader *sh) +{ + sh->symbols = new(sh) glsl_symbol_table; + + foreach_list(node, sh->ir) { + ir_instruction *const inst = (ir_instruction *) node; + ir_variable *var; + ir_function *func; + + if ((func = inst->as_function()) != NULL) { + sh->symbols->add_function(func); + } else if ((var = inst->as_variable()) != NULL) { + sh->symbols->add_variable(var); + } + } +} + + +/** + * Remap variables referenced in an instruction tree + * + * This is used when instruction trees are cloned from one shader and placed in + * another. These trees will contain references to \c ir_variable nodes that + * do not exist in the target shader. This function finds these \c ir_variable + * references and replaces the references with matching variables in the target + * shader. + * + * If there is no matching variable in the target shader, a clone of the + * \c ir_variable is made and added to the target shader. The new variable is + * added to \b both the instruction stream and the symbol table. + * + * \param inst IR tree that is to be processed. + * \param symbols Symbol table containing global scope symbols in the + * linked shader. + * \param instructions Instruction stream where new variable declarations + * should be added. + */ +void +remap_variables(ir_instruction *inst, struct gl_shader *target, + hash_table *temps) +{ + class remap_visitor : public ir_hierarchical_visitor { + public: + remap_visitor(struct gl_shader *target, + hash_table *temps) + { + this->target = target; + this->symbols = target->symbols; + this->instructions = target->ir; + this->temps = temps; + } + + virtual ir_visitor_status visit(ir_dereference_variable *ir) + { + if (ir->var->mode == ir_var_temporary) { + ir_variable *var = (ir_variable *) hash_table_find(temps, ir->var); + + assert(var != NULL); + ir->var = var; + return visit_continue; + } + + ir_variable *const existing = + this->symbols->get_variable(ir->var->name); + if (existing != NULL) + ir->var = existing; + else { + ir_variable *copy = ir->var->clone(this->target, NULL); + + this->symbols->add_variable(copy); + this->instructions->push_head(copy); + ir->var = copy; + } + + return visit_continue; + } + + private: + struct gl_shader *target; + glsl_symbol_table *symbols; + exec_list *instructions; + hash_table *temps; + }; + + remap_visitor v(target, temps); + + inst->accept(&v); +} + + +/** + * Move non-declarations from one instruction stream to another + * + * The intended usage pattern of this function is to pass the pointer to the + * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node + * pointer) for \c last and \c false for \c make_copies on the first + * call. Successive calls pass the return value of the previous call for + * \c last and \c true for \c make_copies. + * + * \param instructions Source instruction stream + * \param last Instruction after which new instructions should be + * inserted in the target instruction stream + * \param make_copies Flag selecting whether instructions in \c instructions + * should be copied (via \c ir_instruction::clone) into the + * target list or moved. + * + * \return + * The new "last" instruction in the target instruction stream. This pointer + * is suitable for use as the \c last parameter of a later call to this + * function. + */ +exec_node * +move_non_declarations(exec_list *instructions, exec_node *last, + bool make_copies, gl_shader *target) +{ + hash_table *temps = NULL; + + if (make_copies) + temps = hash_table_ctor(0, hash_table_pointer_hash, + hash_table_pointer_compare); + + foreach_list_safe(node, instructions) { + ir_instruction *inst = (ir_instruction *) node; + + if (inst->as_function()) + continue; + + ir_variable *var = inst->as_variable(); + if ((var != NULL) && (var->mode != ir_var_temporary)) + continue; + + assert(inst->as_assignment() + || ((var != NULL) && (var->mode == ir_var_temporary))); + + if (make_copies) { + inst = inst->clone(target, NULL); + + if (var != NULL) + hash_table_insert(temps, inst, var); + else + remap_variables(inst, target, temps); + } else { + inst->remove(); + } + + last->insert_after(inst); + last = inst; + } + + if (make_copies) + hash_table_dtor(temps); + + return last; +} + +/** + * Get the function signature for main from a shader + */ +static ir_function_signature * +get_main_function_signature(gl_shader *sh) +{ + ir_function *const f = sh->symbols->get_function("main"); + if (f != NULL) { + exec_list void_parameters; + + /* Look for the 'void main()' signature and ensure that it's defined. + * This keeps the linker from accidentally pick a shader that just + * contains a prototype for main. + * + * We don't have to check for multiple definitions of main (in multiple + * shaders) because that would have already been caught above. + */ + ir_function_signature *sig = f->matching_signature(&void_parameters); + if ((sig != NULL) && sig->is_defined) { + return sig; + } + } + + return NULL; +} + + +/** + * Combine a group of shaders for a single stage to generate a linked shader + * + * \note + * If this function is supplied a single shader, it is cloned, and the new + * shader is returned. + */ +static struct gl_shader * +link_intrastage_shaders(void *mem_ctx, + struct gl_context *ctx, + struct gl_shader_program *prog, + struct gl_shader **shader_list, + unsigned num_shaders) +{ + /* Check that global variables defined in multiple shaders are consistent. + */ + if (!cross_validate_globals(prog, shader_list, num_shaders, false)) + return NULL; + + /* Check that there is only a single definition of each function signature + * across all shaders. + */ + for (unsigned i = 0; i < (num_shaders - 1); i++) { + foreach_list(node, shader_list[i]->ir) { + ir_function *const f = ((ir_instruction *) node)->as_function(); + + if (f == NULL) + continue; + + for (unsigned j = i + 1; j < num_shaders; j++) { + ir_function *const other = + shader_list[j]->symbols->get_function(f->name); + + /* If the other shader has no function (and therefore no function + * signatures) with the same name, skip to the next shader. + */ + if (other == NULL) + continue; + + foreach_iter (exec_list_iterator, iter, *f) { + ir_function_signature *sig = + (ir_function_signature *) iter.get(); + + if (!sig->is_defined || sig->is_builtin) + continue; + + ir_function_signature *other_sig = + other->exact_matching_signature(& sig->parameters); + + if ((other_sig != NULL) && other_sig->is_defined + && !other_sig->is_builtin) { + linker_error_printf(prog, + "function `%s' is multiply defined", + f->name); + return NULL; + } + } + } + } + } + + /* Find the shader that defines main, and make a clone of it. + * + * Starting with the clone, search for undefined references. If one is + * found, find the shader that defines it. Clone the reference and add + * it to the shader. Repeat until there are no undefined references or + * until a reference cannot be resolved. + */ + gl_shader *main = NULL; + for (unsigned i = 0; i < num_shaders; i++) { + if (get_main_function_signature(shader_list[i]) != NULL) { + main = shader_list[i]; + break; + } + } + + if (main == NULL) { + linker_error_printf(prog, "%s shader lacks `main'\n", + (shader_list[0]->Type == GL_VERTEX_SHADER) + ? "vertex" : "fragment"); + return NULL; + } + + gl_shader *linked = ctx->Driver.NewShader(NULL, 0, main->Type); + linked->ir = new(linked) exec_list; + clone_ir_list(mem_ctx, linked->ir, main->ir); + + populate_symbol_table(linked); + + /* The a pointer to the main function in the final linked shader (i.e., the + * copy of the original shader that contained the main function). + */ + ir_function_signature *const main_sig = get_main_function_signature(linked); + + /* Move any instructions other than variable declarations or function + * declarations into main. + */ + exec_node *insertion_point = + move_non_declarations(linked->ir, (exec_node *) &main_sig->body, false, + linked); + + for (unsigned i = 0; i < num_shaders; i++) { + if (shader_list[i] == main) + continue; + + insertion_point = move_non_declarations(shader_list[i]->ir, + insertion_point, true, linked); + } + + /* Resolve initializers for global variables in the linked shader. + */ + unsigned num_linking_shaders = num_shaders; + for (unsigned i = 0; i < num_shaders; i++) + num_linking_shaders += shader_list[i]->num_builtins_to_link; + + gl_shader **linking_shaders = + (gl_shader **) calloc(num_linking_shaders, sizeof(gl_shader *)); + + memcpy(linking_shaders, shader_list, + sizeof(linking_shaders[0]) * num_shaders); + + unsigned idx = num_shaders; + for (unsigned i = 0; i < num_shaders; i++) { + memcpy(&linking_shaders[idx], shader_list[i]->builtins_to_link, + sizeof(linking_shaders[0]) * shader_list[i]->num_builtins_to_link); + idx += shader_list[i]->num_builtins_to_link; + } + + assert(idx == num_linking_shaders); + + if (!link_function_calls(prog, linked, linking_shaders, + num_linking_shaders)) { + ctx->Driver.DeleteShader(ctx, linked); + linked = NULL; + } + + free(linking_shaders); + + /* 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. + */ + if (linked != NULL) { + class array_sizing_visitor : public ir_hierarchical_visitor { + public: + virtual ir_visitor_status visit(ir_variable *var) + { + if (var->type->is_array() && (var->type->length == 0)) { + const glsl_type *type = + glsl_type::get_array_instance(var->type->fields.array, + var->max_array_access); + + assert(type != NULL); + var->type = type; + } + + return visit_continue; + } + } v; + + v.run(linked->ir); + } + + return linked; +} + + +struct uniform_node { + exec_node link; + struct gl_uniform *u; + unsigned slots; +}; + +/** + * Update the sizes of linked shader uniform arrays to the maximum + * array index used. + * + * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec: + * + * If one or more elements of an array are active, + * GetActiveUniform will return the name of the array in name, + * subject to the restrictions listed above. The type of the array + * is returned in type. The size parameter contains the highest + * array element index used, plus one. The compiler or linker + * determines the highest index used. There will be only one + * active uniform reported by the GL per uniform array. + + */ +static void +update_array_sizes(struct gl_shader_program *prog) +{ + for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + foreach_list(node, prog->_LinkedShaders[i]->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != ir_var_uniform && + var->mode != ir_var_in && + var->mode != ir_var_out) || + !var->type->is_array()) + continue; + + unsigned int size = var->max_array_access; + for (unsigned j = 0; j < MESA_SHADER_TYPES; j++) { + if (prog->_LinkedShaders[j] == NULL) + continue; + + foreach_list(node2, prog->_LinkedShaders[j]->ir) { + ir_variable *other_var = ((ir_instruction *) node2)->as_variable(); + if (!other_var) + continue; + + if (strcmp(var->name, other_var->name) == 0 && + other_var->max_array_access > size) { + size = other_var->max_array_access; + } + } + } + + if (size + 1 != var->type->fields.array->length) { + var->type = glsl_type::get_array_instance(var->type->fields.array, + size + 1); + /* FINISHME: We should update the types of array + * dereferences of this variable now. + */ + } + } + } +} + +static void +add_uniform(void *mem_ctx, exec_list *uniforms, struct hash_table *ht, + const char *name, const glsl_type *type, GLenum shader_type, + unsigned *next_shader_pos, unsigned *total_uniforms) +{ + if (type->is_record()) { + for (unsigned int i = 0; i < type->length; i++) { + const glsl_type *field_type = type->fields.structure[i].type; + char *field_name = talloc_asprintf(mem_ctx, "%s.%s", name, + type->fields.structure[i].name); + + add_uniform(mem_ctx, uniforms, ht, field_name, field_type, + shader_type, next_shader_pos, total_uniforms); + } + } else { + uniform_node *n = (uniform_node *) hash_table_find(ht, name); + unsigned int vec4_slots; + const glsl_type *array_elem_type = NULL; + + if (type->is_array()) { + array_elem_type = type->fields.array; + /* Array of structures. */ + if (array_elem_type->is_record()) { + for (unsigned int i = 0; i < type->length; i++) { + char *elem_name = talloc_asprintf(mem_ctx, "%s[%d]", name, i); + add_uniform(mem_ctx, uniforms, ht, elem_name, array_elem_type, + shader_type, next_shader_pos, total_uniforms); + } + return; + } + } + + /* Fix the storage size of samplers at 1 vec4 each. Be sure to pad out + * vectors to vec4 slots. + */ + if (type->is_array()) { + if (array_elem_type->is_sampler()) + vec4_slots = type->length; + else + vec4_slots = type->length * array_elem_type->matrix_columns; + } else if (type->is_sampler()) { + vec4_slots = 1; + } else { + vec4_slots = type->matrix_columns; + } + + if (n == NULL) { + n = (uniform_node *) calloc(1, sizeof(struct uniform_node)); + n->u = (gl_uniform *) calloc(1, sizeof(struct gl_uniform)); + n->slots = vec4_slots; + + n->u->Name = strdup(name); + n->u->Type = type; + n->u->VertPos = -1; + n->u->FragPos = -1; + n->u->GeomPos = -1; + (*total_uniforms)++; + + hash_table_insert(ht, n, name); + uniforms->push_tail(& n->link); + } + + switch (shader_type) { + case GL_VERTEX_SHADER: + n->u->VertPos = *next_shader_pos; + break; + case GL_FRAGMENT_SHADER: + n->u->FragPos = *next_shader_pos; + break; + case GL_GEOMETRY_SHADER: + n->u->GeomPos = *next_shader_pos; + break; + } + + (*next_shader_pos) += vec4_slots; + } +} + +void +assign_uniform_locations(struct gl_shader_program *prog) +{ + /* */ + exec_list uniforms; + unsigned total_uniforms = 0; + hash_table *ht = hash_table_ctor(32, hash_table_string_hash, + hash_table_string_compare); + void *mem_ctx = talloc_new(NULL); + + for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + unsigned next_position = 0; + + foreach_list(node, prog->_LinkedShaders[i]->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != ir_var_uniform)) + continue; + + if (strncmp(var->name, "gl_", 3) == 0) { + /* At the moment, we don't allocate uniform locations for + * builtin uniforms. It's permitted by spec, and we'll + * likely switch to doing that at some point, but not yet. + */ + continue; + } + + var->location = next_position; + add_uniform(mem_ctx, &uniforms, ht, var->name, var->type, + prog->_LinkedShaders[i]->Type, + &next_position, &total_uniforms); + } + } + + talloc_free(mem_ctx); + + gl_uniform_list *ul = (gl_uniform_list *) + calloc(1, sizeof(gl_uniform_list)); + + ul->Size = total_uniforms; + ul->NumUniforms = total_uniforms; + ul->Uniforms = (gl_uniform *) calloc(total_uniforms, sizeof(gl_uniform)); + + unsigned idx = 0; + uniform_node *next; + for (uniform_node *node = (uniform_node *) uniforms.head + ; node->link.next != NULL + ; node = next) { + next = (uniform_node *) node->link.next; + + node->link.remove(); + memcpy(&ul->Uniforms[idx], node->u, sizeof(gl_uniform)); + idx++; + + free(node->u); + free(node); + } + + hash_table_dtor(ht); + + prog->Uniforms = ul; +} + + +/** + * Find a contiguous set of available bits in a bitmask + * + * \param used_mask Bits representing used (1) and unused (0) locations + * \param needed_count Number of contiguous bits needed. + * + * \return + * Base location of the available bits on success or -1 on failure. + */ +int +find_available_slots(unsigned used_mask, unsigned needed_count) +{ + unsigned needed_mask = (1 << needed_count) - 1; + const int max_bit_to_test = (8 * sizeof(used_mask)) - needed_count; + + /* The comparison to 32 is redundant, but without it GCC emits "warning: + * cannot optimize possibly infinite loops" for the loop below. + */ + if ((needed_count == 0) || (max_bit_to_test < 0) || (max_bit_to_test > 32)) + return -1; + + for (int i = 0; i <= max_bit_to_test; i++) { + if ((needed_mask & ~used_mask) == needed_mask) + return i; + + needed_mask <<= 1; + } + + return -1; +} + + +bool +assign_attribute_locations(gl_shader_program *prog, unsigned max_attribute_index) +{ + /* Mark invalid attribute locations as being used. + */ + unsigned used_locations = (max_attribute_index >= 32) + ? ~0 : ~((1 << max_attribute_index) - 1); + + gl_shader *const sh = prog->_LinkedShaders[0]; + assert(sh->Type == GL_VERTEX_SHADER); + + /* Operate in a total of four passes. + * + * 1. Invalidate the location assignments for all vertex shader inputs. + * + * 2. Assign locations for inputs that have user-defined (via + * glBindVertexAttribLocation) locatoins. + * + * 3. Sort the attributes without assigned locations by number of slots + * required in decreasing order. Fragmentation caused by attribute + * locations assigned by the application may prevent large attributes + * from having enough contiguous space. + * + * 4. Assign locations to any inputs without assigned locations. + */ + + invalidate_variable_locations(sh, ir_var_in, VERT_ATTRIB_GENERIC0); + + if (prog->Attributes != NULL) { + for (unsigned i = 0; i < prog->Attributes->NumParameters; i++) { + ir_variable *const var = + sh->symbols->get_variable(prog->Attributes->Parameters[i].Name); + + /* Note: attributes that occupy multiple slots, such as arrays or + * matrices, may appear in the attrib array multiple times. + */ + if ((var == NULL) || (var->location != -1)) + continue; + + /* From page 61 of the OpenGL 4.0 spec: + * + * "LinkProgram will fail if the attribute bindings assigned by + * BindAttribLocation do not leave not enough space to assign a + * location for an active matrix attribute or an active attribute + * array, both of which require multiple contiguous generic + * attributes." + * + * Previous versions of the spec contain similar language but omit the + * bit about attribute arrays. + * + * Page 61 of the OpenGL 4.0 spec also says: + * + * "It is possible for an application to bind more than one + * attribute name to the same location. This is referred to as + * aliasing. This will only work if only one of the aliased + * attributes is active in the executable program, or if no path + * through the shader consumes more than one attribute of a set + * of attributes aliased to the same location. A link error can + * occur if the linker determines that every path through the + * shader consumes multiple aliased attributes, but + * implementations are not required to generate an error in this + * case." + * + * These two paragraphs are either somewhat contradictory, or I don't + * fully understand one or both of them. + */ + /* FINISHME: The code as currently written does not support attribute + * FINISHME: location aliasing (see comment above). + */ + const int attr = prog->Attributes->Parameters[i].StateIndexes[0]; + const unsigned slots = count_attribute_slots(var->type); + + /* Mask representing the contiguous slots that will be used by this + * attribute. + */ + const unsigned use_mask = (1 << slots) - 1; + + /* Generate a link error if the set of bits requested for this + * attribute overlaps any previously allocated bits. + */ + if ((~(use_mask << attr) & used_locations) != used_locations) { + linker_error_printf(prog, + "insufficient contiguous attribute locations " + "available for vertex shader input `%s'", + var->name); + return false; + } + + var->location = VERT_ATTRIB_GENERIC0 + attr; + used_locations |= (use_mask << attr); + } + } + + /* Temporary storage for the set of attributes that need locations assigned. + */ + struct temp_attr { + unsigned slots; + ir_variable *var; + + /* Used below in the call to qsort. */ + static int compare(const void *a, const void *b) + { + const temp_attr *const l = (const temp_attr *) a; + const temp_attr *const r = (const temp_attr *) b; + + /* Reversed because we want a descending order sort below. */ + return r->slots - l->slots; + } + } to_assign[16]; + + unsigned num_attr = 0; + + foreach_list(node, sh->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != ir_var_in)) + continue; + + if (var->explicit_location) { + const unsigned slots = count_attribute_slots(var->type); + const unsigned use_mask = (1 << slots) - 1; + const int attr = var->location - VERT_ATTRIB_GENERIC0; + + if ((var->location >= (int)(max_attribute_index + VERT_ATTRIB_GENERIC0)) + || (var->location < 0)) { + linker_error_printf(prog, + "invalid explicit location %d specified for " + "`%s'\n", + (var->location < 0) ? var->location : attr, + var->name); + return false; + } else if (var->location >= VERT_ATTRIB_GENERIC0) { + used_locations |= (use_mask << attr); + } + } + + /* The location was explicitly assigned, nothing to do here. + */ + if (var->location != -1) + continue; + + to_assign[num_attr].slots = count_attribute_slots(var->type); + to_assign[num_attr].var = var; + num_attr++; + } + + /* If all of the attributes were assigned locations by the application (or + * are built-in attributes with fixed locations), return early. This should + * be the common case. + */ + if (num_attr == 0) + return true; + + qsort(to_assign, num_attr, sizeof(to_assign[0]), temp_attr::compare); + + /* VERT_ATTRIB_GENERIC0 is a psdueo-alias for VERT_ATTRIB_POS. It can only + * be explicitly assigned by via glBindAttribLocation. Mark it as reserved + * to prevent it from being automatically allocated below. + */ + find_deref_visitor find("gl_Vertex"); + find.run(sh->ir); + if (find.variable_found()) + used_locations |= (1 << 0); + + for (unsigned i = 0; i < num_attr; i++) { + /* Mask representing the contiguous slots that will be used by this + * attribute. + */ + const unsigned use_mask = (1 << to_assign[i].slots) - 1; + + int location = find_available_slots(used_locations, to_assign[i].slots); + + if (location < 0) { + linker_error_printf(prog, + "insufficient contiguous attribute locations " + "available for vertex shader input `%s'", + to_assign[i].var->name); + return false; + } + + to_assign[i].var->location = VERT_ATTRIB_GENERIC0 + location; + used_locations |= (use_mask << location); + } + + return true; +} + + +/** + * Demote shader inputs and outputs that are not used in other stages + */ +void +demote_shader_inputs_and_outputs(gl_shader *sh, enum ir_variable_mode mode) +{ + foreach_list(node, sh->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != int(mode))) + continue; + + /* A shader 'in' or 'out' variable is only really an input or output if + * its value is used by other shader stages. This will cause the variable + * to have a location assigned. + */ + if (var->location == -1) { + var->mode = ir_var_auto; + } + } +} + + +void +assign_varying_locations(struct gl_shader_program *prog, + gl_shader *producer, gl_shader *consumer) +{ + /* FINISHME: Set dynamically when geometry shader support is added. */ + unsigned output_index = VERT_RESULT_VAR0; + unsigned input_index = FRAG_ATTRIB_VAR0; + + /* Operate in a total of three passes. + * + * 1. Assign locations for any matching inputs and outputs. + * + * 2. Mark output variables in the producer that do not have locations as + * not being outputs. This lets the optimizer eliminate them. + * + * 3. Mark input variables in the consumer that do not have locations as + * not being inputs. This lets the optimizer eliminate them. + */ + + invalidate_variable_locations(producer, ir_var_out, VERT_RESULT_VAR0); + invalidate_variable_locations(consumer, ir_var_in, FRAG_ATTRIB_VAR0); + + foreach_list(node, producer->ir) { + ir_variable *const output_var = ((ir_instruction *) node)->as_variable(); + + if ((output_var == NULL) || (output_var->mode != ir_var_out) + || (output_var->location != -1)) + continue; + + ir_variable *const input_var = + consumer->symbols->get_variable(output_var->name); + + if ((input_var == NULL) || (input_var->mode != ir_var_in)) + continue; + + assert(input_var->location == -1); + + output_var->location = output_index; + input_var->location = input_index; + + /* FINISHME: Support for "varying" records in GLSL 1.50. */ + assert(!output_var->type->is_record()); + + if (output_var->type->is_array()) { + const unsigned slots = output_var->type->length + * output_var->type->fields.array->matrix_columns; + + output_index += slots; + input_index += slots; + } else { + const unsigned slots = output_var->type->matrix_columns; + + output_index += slots; + input_index += slots; + } + } + + foreach_list(node, consumer->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != ir_var_in)) + continue; + + if (var->location == -1) { + if (prog->Version <= 120) { + /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec: + * + * Only those varying variables used (i.e. read) in + * the fragment shader executable must be written to + * by the vertex shader executable; declaring + * superfluous varying variables in a vertex shader is + * permissible. + * + * We interpret this text as meaning that the VS must + * write the variable for the FS to read it. See + * "glsl1-varying read but not written" in piglit. + */ + + linker_error_printf(prog, "fragment shader varying %s not written " + "by vertex shader\n.", var->name); + prog->LinkStatus = false; + } + + /* An 'in' variable is only really a shader input if its + * value is written by the previous stage. + */ + var->mode = ir_var_auto; + } + } +} + + +void +link_shaders(struct gl_context *ctx, struct gl_shader_program *prog) +{ + void *mem_ctx = talloc_init("temporary linker context"); + + prog->LinkStatus = false; + prog->Validated = false; + prog->_Used = false; + + if (prog->InfoLog != NULL) + talloc_free(prog->InfoLog); + + prog->InfoLog = talloc_strdup(NULL, ""); + + /* Separate the shaders into groups based on their type. + */ + struct gl_shader **vert_shader_list; + unsigned num_vert_shaders = 0; + struct gl_shader **frag_shader_list; + unsigned num_frag_shaders = 0; + + vert_shader_list = (struct gl_shader **) + calloc(2 * prog->NumShaders, sizeof(struct gl_shader *)); + frag_shader_list = &vert_shader_list[prog->NumShaders]; + + unsigned min_version = UINT_MAX; + unsigned max_version = 0; + for (unsigned i = 0; i < prog->NumShaders; i++) { + min_version = MIN2(min_version, prog->Shaders[i]->Version); + max_version = MAX2(max_version, prog->Shaders[i]->Version); + + switch (prog->Shaders[i]->Type) { + case GL_VERTEX_SHADER: + vert_shader_list[num_vert_shaders] = prog->Shaders[i]; + num_vert_shaders++; + break; + case GL_FRAGMENT_SHADER: + frag_shader_list[num_frag_shaders] = prog->Shaders[i]; + num_frag_shaders++; + break; + case GL_GEOMETRY_SHADER: + /* FINISHME: Support geometry shaders. */ + assert(prog->Shaders[i]->Type != GL_GEOMETRY_SHADER); + break; + } + } + + /* Previous to GLSL version 1.30, different compilation units could mix and + * match shading language versions. With GLSL 1.30 and later, the versions + * of all shaders must match. + */ + assert(min_version >= 100); + assert(max_version <= 130); + if ((max_version >= 130 || min_version == 100) + && min_version != max_version) { + linker_error_printf(prog, "all shaders must use same shading " + "language version\n"); + goto done; + } + + prog->Version = max_version; + + for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] != NULL) + ctx->Driver.DeleteShader(ctx, prog->_LinkedShaders[i]); + + prog->_LinkedShaders[i] = NULL; + } + + /* Link all shaders for a particular stage and validate the result. + */ + if (num_vert_shaders > 0) { + gl_shader *const sh = + link_intrastage_shaders(mem_ctx, ctx, prog, vert_shader_list, + num_vert_shaders); + + if (sh == NULL) + goto done; + + if (!validate_vertex_shader_executable(prog, sh)) + goto done; + + _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_VERTEX], + sh); + } + + if (num_frag_shaders > 0) { + gl_shader *const sh = + link_intrastage_shaders(mem_ctx, ctx, prog, frag_shader_list, + num_frag_shaders); + + if (sh == NULL) + goto done; + + if (!validate_fragment_shader_executable(prog, sh)) + goto done; + + _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_FRAGMENT], + sh); + } + + /* Here begins the inter-stage linking phase. Some initial validation is + * performed, then locations are assigned for uniforms, attributes, and + * varyings. + */ + if (cross_validate_uniforms(prog)) { + unsigned prev; + + for (prev = 0; prev < MESA_SHADER_TYPES; prev++) { + if (prog->_LinkedShaders[prev] != NULL) + break; + } + + /* Validate the inputs of each stage with the output of the preceeding + * stage. + */ + for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + if (!cross_validate_outputs_to_inputs(prog, + prog->_LinkedShaders[prev], + prog->_LinkedShaders[i])) + goto done; + + prev = i; + } + + prog->LinkStatus = true; + } + + /* Do common optimization before assigning storage for attributes, + * uniforms, and varyings. Later optimization could possibly make + * some of that unused. + */ + for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, 32)) + ; + } + + update_array_sizes(prog); + + assign_uniform_locations(prog); + + if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) { + /* FINISHME: The value of the max_attribute_index parameter is + * FINISHME: implementation dependent based on the value of + * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be + * FINISHME: at least 16, so hardcode 16 for now. + */ + if (!assign_attribute_locations(prog, 16)) { + prog->LinkStatus = false; + goto done; + } + } + + unsigned prev; + for (prev = 0; prev < MESA_SHADER_TYPES; prev++) { + if (prog->_LinkedShaders[prev] != NULL) + break; + } + + for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + assign_varying_locations(prog, + prog->_LinkedShaders[prev], + prog->_LinkedShaders[i]); + prev = i; + } + + if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) { + demote_shader_inputs_and_outputs(prog->_LinkedShaders[MESA_SHADER_VERTEX], + ir_var_out); + } + + if (prog->_LinkedShaders[MESA_SHADER_GEOMETRY] != NULL) { + gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_GEOMETRY]; + + demote_shader_inputs_and_outputs(sh, ir_var_in); + demote_shader_inputs_and_outputs(sh, ir_var_inout); + demote_shader_inputs_and_outputs(sh, ir_var_out); + } + + if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] != NULL) { + gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_FRAGMENT]; + + demote_shader_inputs_and_outputs(sh, ir_var_in); + } + + /* FINISHME: Assign fragment shader output locations. */ + +done: + free(vert_shader_list); + + for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + /* Retain any live IR, but trash the rest. */ + reparent_ir(prog->_LinkedShaders[i]->ir, prog->_LinkedShaders[i]->ir); + } + + talloc_free(mem_ctx); +} diff --git a/mesalib/src/mesa/main/teximage.c b/mesalib/src/mesa/main/teximage.c index c3cd6b66a..7f630e23c 100644 --- a/mesalib/src/mesa/main/teximage.c +++ b/mesalib/src/mesa/main/teximage.c @@ -703,6 +703,9 @@ _mesa_select_tex_object(struct gl_context *ctx, const struct gl_texture_unit *texUnit, GLenum target) { + const GLboolean arrayTex = (ctx->Extensions.MESA_texture_array || + ctx->Extensions.EXT_texture_array); + switch (target) { case GL_TEXTURE_1D: return texUnit->CurrentTex[TEXTURE_1D_INDEX]; @@ -735,17 +738,13 @@ _mesa_select_tex_object(struct gl_context *ctx, return ctx->Extensions.NV_texture_rectangle ? ctx->Texture.ProxyTex[TEXTURE_RECT_INDEX] : NULL; case GL_TEXTURE_1D_ARRAY_EXT: - return ctx->Extensions.MESA_texture_array - ? texUnit->CurrentTex[TEXTURE_1D_ARRAY_INDEX] : NULL; + return arrayTex ? texUnit->CurrentTex[TEXTURE_1D_ARRAY_INDEX] : NULL; case GL_PROXY_TEXTURE_1D_ARRAY_EXT: - return ctx->Extensions.MESA_texture_array - ? ctx->Texture.ProxyTex[TEXTURE_1D_ARRAY_INDEX] : NULL; + return arrayTex ? ctx->Texture.ProxyTex[TEXTURE_1D_ARRAY_INDEX] : NULL; case GL_TEXTURE_2D_ARRAY_EXT: - return ctx->Extensions.MESA_texture_array - ? texUnit->CurrentTex[TEXTURE_2D_ARRAY_INDEX] : NULL; + return arrayTex ? texUnit->CurrentTex[TEXTURE_2D_ARRAY_INDEX] : NULL; case GL_PROXY_TEXTURE_2D_ARRAY_EXT: - return ctx->Extensions.MESA_texture_array - ? ctx->Texture.ProxyTex[TEXTURE_2D_ARRAY_INDEX] : NULL; + return arrayTex ? ctx->Texture.ProxyTex[TEXTURE_2D_ARRAY_INDEX] : NULL; default: _mesa_problem(NULL, "bad target in _mesa_select_tex_object()"); return NULL; @@ -930,7 +929,8 @@ _mesa_max_texture_levels(struct gl_context *ctx, GLenum target) case GL_PROXY_TEXTURE_1D_ARRAY_EXT: case GL_TEXTURE_2D_ARRAY_EXT: case GL_PROXY_TEXTURE_2D_ARRAY_EXT: - return ctx->Extensions.MESA_texture_array + return (ctx->Extensions.MESA_texture_array || + ctx->Extensions.EXT_texture_array) ? ctx->Const.MaxTextureLevels : 0; default: return 0; /* bad target */ @@ -1382,7 +1382,8 @@ target_can_be_compressed(const struct gl_context *ctx, GLenum target, return ctx->Extensions.ARB_texture_cube_map; case GL_PROXY_TEXTURE_2D_ARRAY_EXT: case GL_TEXTURE_2D_ARRAY_EXT: - return ctx->Extensions.MESA_texture_array; + return (ctx->Extensions.MESA_texture_array || + ctx->Extensions.MESA_texture_array); default: return GL_FALSE; } @@ -1423,7 +1424,8 @@ legal_teximage_target(struct gl_context *ctx, GLuint dims, GLenum target) return ctx->Extensions.NV_texture_rectangle; case GL_TEXTURE_1D_ARRAY_EXT: case GL_PROXY_TEXTURE_1D_ARRAY_EXT: - return ctx->Extensions.MESA_texture_array; + return (ctx->Extensions.MESA_texture_array || + ctx->Extensions.EXT_texture_array); default: return GL_FALSE; } @@ -1434,7 +1436,8 @@ legal_teximage_target(struct gl_context *ctx, GLuint dims, GLenum target) return GL_TRUE; case GL_TEXTURE_2D_ARRAY_EXT: case GL_PROXY_TEXTURE_2D_ARRAY_EXT: - return ctx->Extensions.MESA_texture_array; + return (ctx->Extensions.MESA_texture_array || + ctx->Extensions.EXT_texture_array); default: return GL_FALSE; } @@ -1471,7 +1474,8 @@ legal_texsubimage_target(struct gl_context *ctx, GLuint dims, GLenum target) case GL_TEXTURE_RECTANGLE_NV: return ctx->Extensions.NV_texture_rectangle; case GL_TEXTURE_1D_ARRAY_EXT: - return ctx->Extensions.MESA_texture_array; + return (ctx->Extensions.MESA_texture_array || + ctx->Extensions.EXT_texture_array); default: return GL_FALSE; } @@ -1480,7 +1484,8 @@ legal_texsubimage_target(struct gl_context *ctx, GLuint dims, GLenum target) case GL_TEXTURE_3D: return GL_TRUE; case GL_TEXTURE_2D_ARRAY_EXT: - return ctx->Extensions.MESA_texture_array; + return (ctx->Extensions.MESA_texture_array || + ctx->Extensions.EXT_texture_array); default: return GL_FALSE; } diff --git a/mesalib/src/mesa/main/texparam.c b/mesalib/src/mesa/main/texparam.c index 4909b76cb..6e14face4 100644 --- a/mesalib/src/mesa/main/texparam.c +++ b/mesalib/src/mesa/main/texparam.c @@ -33,6 +33,7 @@ #include "main/glheader.h" #include "main/colormac.h" #include "main/context.h" +#include "main/enums.h" #include "main/formats.h" #include "main/macros.h" #include "main/mfeatures.h" @@ -116,12 +117,14 @@ get_texobj(struct gl_context *ctx, GLenum target, GLboolean get) } break; case GL_TEXTURE_1D_ARRAY_EXT: - if (ctx->Extensions.MESA_texture_array) { + if (ctx->Extensions.MESA_texture_array || + ctx->Extensions.EXT_texture_array) { return texUnit->CurrentTex[TEXTURE_1D_ARRAY_INDEX]; } break; case GL_TEXTURE_2D_ARRAY_EXT: - if (ctx->Extensions.MESA_texture_array) { + if (ctx->Extensions.MESA_texture_array || + ctx->Extensions.EXT_texture_array) { return texUnit->CurrentTex[TEXTURE_2D_ARRAY_INDEX]; } break; @@ -231,8 +234,7 @@ set_tex_parameteri(struct gl_context *ctx, } /* fall-through */ default: - _mesa_error( ctx, GL_INVALID_ENUM, "glTexParameter(param=0x%x)", - params[0] ); + goto invalid_param; } return GL_FALSE; @@ -246,8 +248,7 @@ set_tex_parameteri(struct gl_context *ctx, texObj->MagFilter = params[0]; return GL_TRUE; default: - _mesa_error( ctx, GL_INVALID_ENUM, "glTexParameter(param=0x%x)", - params[0]); + goto invalid_param; } return GL_FALSE; @@ -315,21 +316,18 @@ set_tex_parameteri(struct gl_context *ctx, return GL_FALSE; case GL_TEXTURE_COMPARE_MODE_ARB: - if (ctx->Extensions.ARB_shadow && - (params[0] == GL_NONE || - params[0] == GL_COMPARE_R_TO_TEXTURE_ARB)) { - if (texObj->CompareMode != params[0]) { + if (ctx->Extensions.ARB_shadow) { + if (texObj->CompareMode == params[0]) + return GL_FALSE; + if (params[0] == GL_NONE || + params[0] == GL_COMPARE_R_TO_TEXTURE_ARB) { flush(ctx); texObj->CompareMode = params[0]; return GL_TRUE; } - return GL_FALSE; - } - else { - _mesa_error(ctx, GL_INVALID_ENUM, - "glTexParameter(GL_TEXTURE_COMPARE_MODE_ARB)"); + goto invalid_param; } - return GL_FALSE; + goto invalid_pname; case GL_TEXTURE_COMPARE_FUNC_ARB: if (ctx->Extensions.ARB_shadow) { @@ -354,32 +352,26 @@ set_tex_parameteri(struct gl_context *ctx, } /* fall-through */ default: - _mesa_error(ctx, GL_INVALID_ENUM, - "glTexParameter(GL_TEXTURE_COMPARE_FUNC_ARB)"); + goto invalid_param; } } - else { - _mesa_error(ctx, GL_INVALID_ENUM, "glTexParameter(pname=0x%x)", pname); - } - return GL_FALSE; + goto invalid_pname; case GL_DEPTH_TEXTURE_MODE_ARB: - if (ctx->Extensions.ARB_depth_texture && - (params[0] == GL_LUMINANCE || - params[0] == GL_INTENSITY || - params[0] == GL_ALPHA || - (ctx->Extensions.ARB_texture_rg && params[0] == GL_RED))) { - if (texObj->DepthMode != params[0]) { + if (ctx->Extensions.ARB_depth_texture) { + if (texObj->DepthMode == params[0]) + return GL_FALSE; + if (params[0] == GL_LUMINANCE || + params[0] == GL_INTENSITY || + params[0] == GL_ALPHA || + (ctx->Extensions.ARB_texture_rg && params[0] == GL_RED)) { flush(ctx); texObj->DepthMode = params[0]; return GL_TRUE; } + goto invalid_param; } - else { - _mesa_error(ctx, GL_INVALID_ENUM, - "glTexParameter(GL_DEPTH_TEXTURE_MODE_ARB)"); - } - return GL_FALSE; + goto invalid_pname; #if FEATURE_OES_draw_texture case GL_TEXTURE_CROP_RECT_OES: @@ -410,8 +402,7 @@ set_tex_parameteri(struct gl_context *ctx, return GL_TRUE; } } - _mesa_error(ctx, GL_INVALID_ENUM, "glTexParameter(pname=0x%x)", pname); - return GL_FALSE; + goto invalid_pname; case GL_TEXTURE_SWIZZLE_RGBA_EXT: if (ctx->Extensions.EXT_texture_swizzle) { @@ -431,8 +422,8 @@ set_tex_parameteri(struct gl_context *ctx, } return GL_TRUE; } - _mesa_error(ctx, GL_INVALID_ENUM, "glTexParameter(pname=0x%x)", pname); - return GL_FALSE; + goto invalid_pname; + case GL_TEXTURE_SRGB_DECODE_EXT: if (ctx->Extensions.EXT_texture_sRGB_decode) { GLenum decode = params[0]; @@ -445,11 +436,20 @@ set_tex_parameteri(struct gl_context *ctx, return GL_TRUE; } } - _mesa_error(ctx, GL_INVALID_ENUM, "glTexParameter(pname=0x%x)", pname); - return GL_FALSE; + goto invalid_pname; + default: - _mesa_error(ctx, GL_INVALID_ENUM, "glTexParameter(pname=0x%x)", pname); + goto invalid_pname; } + +invalid_pname: + _mesa_error(ctx, GL_INVALID_ENUM, "glTexParameter(pname=%s)", + _mesa_lookup_enum_by_nr(pname)); + return GL_FALSE; + +invalid_param: + _mesa_error(ctx, GL_INVALID_ENUM, "glTexParameter(param=%s)", + _mesa_lookup_enum_by_nr(params[0])); return GL_FALSE; } @@ -834,7 +834,6 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, const struct gl_texture_unit *texUnit; struct gl_texture_object *texObj; const struct gl_texture_image *img = NULL; - GLboolean isProxy; GLint maxLevels; gl_format texFormat; GET_CURRENT_CONTEXT(ctx); @@ -862,7 +861,6 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, } texObj = _mesa_select_tex_object(ctx, texUnit, target); - _mesa_lock_texture(ctx, texObj); img = _mesa_select_tex_image(ctx, texObj, target, level); if (!img || !img->TexFormat) { @@ -871,13 +869,11 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, *params = 1; else *params = 0; - goto out; + return; } texFormat = img->TexFormat; - isProxy = _mesa_is_proxy_texture(target); - switch (pname) { case GL_TEXTURE_WIDTH: *params = img->Width; @@ -889,9 +885,9 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, *params = img->Depth; break; case GL_TEXTURE_INTERNAL_FORMAT: - if (_mesa_is_format_compressed(img->TexFormat)) { + if (_mesa_is_format_compressed(texFormat)) { /* need to return the actual compressed format */ - *params = _mesa_compressed_format_to_glenum(ctx, img->TexFormat); + *params = _mesa_compressed_format_to_glenum(ctx, texFormat); } else { /* return the user's requested internal format */ @@ -962,8 +958,7 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, if (ctx->Extensions.ARB_depth_texture) *params = _mesa_get_format_bits(texFormat, pname); else - _mesa_error(ctx, GL_INVALID_ENUM, - "glGetTexLevelParameter[if]v(pname)"); + goto invalid_pname; break; case GL_TEXTURE_STENCIL_SIZE_EXT: if (ctx->Extensions.EXT_packed_depth_stencil || @@ -971,8 +966,7 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, *params = _mesa_get_format_bits(texFormat, pname); } else { - _mesa_error(ctx, GL_INVALID_ENUM, - "glGetTexLevelParameter[if]v(pname)"); + goto invalid_pname; } break; case GL_TEXTURE_SHARED_SIZE: @@ -983,14 +977,14 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, *params = 0; } else { - _mesa_error(ctx, GL_INVALID_ENUM, - "glGetTexLevelParameter[if]v(pname)"); + goto invalid_pname; } break; /* GL_ARB_texture_compression */ case GL_TEXTURE_COMPRESSED_IMAGE_SIZE: - if (_mesa_is_format_compressed(img->TexFormat) && !isProxy) { + if (_mesa_is_format_compressed(texFormat) && + !_mesa_is_proxy_texture(target)) { *params = _mesa_format_image_size(texFormat, img->Width, img->Height, img->Depth); } @@ -1000,7 +994,7 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, } break; case GL_TEXTURE_COMPRESSED: - *params = (GLint) _mesa_is_format_compressed(img->TexFormat); + *params = (GLint) _mesa_is_format_compressed(texFormat); break; /* GL_ARB_texture_float */ @@ -1010,8 +1004,7 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, _mesa_get_format_datatype(texFormat) : GL_NONE; } else { - _mesa_error(ctx, GL_INVALID_ENUM, - "glGetTexLevelParameter[if]v(pname)"); + goto invalid_pname; } break; case GL_TEXTURE_GREEN_TYPE_ARB: @@ -1020,8 +1013,7 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, _mesa_get_format_datatype(texFormat) : GL_NONE; } else { - _mesa_error(ctx, GL_INVALID_ENUM, - "glGetTexLevelParameter[if]v(pname)"); + goto invalid_pname; } break; case GL_TEXTURE_BLUE_TYPE_ARB: @@ -1030,8 +1022,7 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, _mesa_get_format_datatype(texFormat) : GL_NONE; } else { - _mesa_error(ctx, GL_INVALID_ENUM, - "glGetTexLevelParameter[if]v(pname)"); + goto invalid_pname; } break; case GL_TEXTURE_ALPHA_TYPE_ARB: @@ -1040,8 +1031,7 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, _mesa_get_format_datatype(texFormat) : GL_NONE; } else { - _mesa_error(ctx, GL_INVALID_ENUM, - "glGetTexLevelParameter[if]v(pname)"); + goto invalid_pname; } break; case GL_TEXTURE_LUMINANCE_TYPE_ARB: @@ -1050,8 +1040,7 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, _mesa_get_format_datatype(texFormat) : GL_NONE; } else { - _mesa_error(ctx, GL_INVALID_ENUM, - "glGetTexLevelParameter[if]v(pname)"); + goto invalid_pname; } break; case GL_TEXTURE_INTENSITY_TYPE_ARB: @@ -1060,8 +1049,7 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, _mesa_get_format_datatype(texFormat) : GL_NONE; } else { - _mesa_error(ctx, GL_INVALID_ENUM, - "glGetTexLevelParameter[if]v(pname)"); + goto invalid_pname; } break; case GL_TEXTURE_DEPTH_TYPE_ARB: @@ -1070,18 +1058,21 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, _mesa_get_format_datatype(texFormat) : GL_NONE; } else { - _mesa_error(ctx, GL_INVALID_ENUM, - "glGetTexLevelParameter[if]v(pname)"); + goto invalid_pname; } break; default: - _mesa_error(ctx, GL_INVALID_ENUM, - "glGetTexLevelParameter[if]v(pname)"); + goto invalid_pname; } - out: - _mesa_unlock_texture(ctx, texObj); + /* no error if we get here */ + return; + +invalid_pname: + _mesa_error(ctx, GL_INVALID_ENUM, + "glGetTexLevelParameter[if]v(pname=%s)", + _mesa_lookup_enum_by_nr(pname)); } diff --git a/mesalib/src/mesa/program/ir_to_mesa.cpp b/mesalib/src/mesa/program/ir_to_mesa.cpp index ae2742afd..ee3eae6cf 100644 --- a/mesalib/src/mesa/program/ir_to_mesa.cpp +++ b/mesalib/src/mesa/program/ir_to_mesa.cpp @@ -1,3210 +1,3214 @@ -/* - * 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" -} - -static int swizzle_for_size(int size); - -/** - * This struct is a corresponding struct to Mesa prog_src_register, with - * wider fields. - */ -typedef struct ir_to_mesa_src_reg { - ir_to_mesa_src_reg(int file, int index, const glsl_type *type) - { - this->file = (gl_register_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; - } - - ir_to_mesa_src_reg() - { - this->file = PROGRAM_UNDEFINED; - this->index = 0; - this->swizzle = 0; - this->negate = 0; - this->reladdr = NULL; - } - - 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. */ - ir_to_mesa_src_reg *reladdr; -} ir_to_mesa_src_reg; - -typedef struct ir_to_mesa_dst_reg { - int 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. */ - ir_to_mesa_src_reg *reladdr; -} ir_to_mesa_dst_reg; - -extern ir_to_mesa_src_reg ir_to_mesa_undef; - -class ir_to_mesa_instruction : public exec_node { -public: - /* Callers of this talloc-based new need not call delete. It's - * easier to just talloc_free 'ctx' (or any of its ancestors). */ - static void* operator new(size_t size, void *ctx) - { - void *node; - - node = talloc_zero_size(ctx, size); - assert(node != NULL); - - return node; - } - - enum prog_opcode op; - ir_to_mesa_dst_reg dst_reg; - ir_to_mesa_src_reg src_reg[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. */ - ir_to_mesa_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); - - ir_to_mesa_src_reg get_temp(const glsl_type *type); - void reladdr_to_temp(ir_instruction *ir, - ir_to_mesa_src_reg *reg, int *num_reladdr); - - struct ir_to_mesa_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 *); - /*@}*/ - - struct ir_to_mesa_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 *ir_to_mesa_emit_op0(ir_instruction *ir, - enum prog_opcode op); - - ir_to_mesa_instruction *ir_to_mesa_emit_op1(ir_instruction *ir, - enum prog_opcode op, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg src0); - - ir_to_mesa_instruction *ir_to_mesa_emit_op2(ir_instruction *ir, - enum prog_opcode op, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg src0, - ir_to_mesa_src_reg src1); - - ir_to_mesa_instruction *ir_to_mesa_emit_op3(ir_instruction *ir, - enum prog_opcode op, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg src0, - ir_to_mesa_src_reg src1, - ir_to_mesa_src_reg src2); - - /** - * Emit the correct dot-product instruction for the type of arguments - * - * \sa ir_to_mesa_emit_op2 - */ - void ir_to_mesa_emit_dp(ir_instruction *ir, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg src0, - ir_to_mesa_src_reg src1, - unsigned elements); - - void ir_to_mesa_emit_scalar_op1(ir_instruction *ir, - enum prog_opcode op, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg src0); - - void ir_to_mesa_emit_scalar_op2(ir_instruction *ir, - enum prog_opcode op, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg src0, - ir_to_mesa_src_reg src1); - - void emit_scs(ir_instruction *ir, enum prog_opcode op, - ir_to_mesa_dst_reg dst, - const ir_to_mesa_src_reg &src); - - GLboolean try_emit_mad(ir_expression *ir, - int mul_operand); - GLboolean 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; -}; - -ir_to_mesa_src_reg ir_to_mesa_undef = ir_to_mesa_src_reg(PROGRAM_UNDEFINED, 0, NULL); - -ir_to_mesa_dst_reg ir_to_mesa_undef_dst = { - PROGRAM_UNDEFINED, 0, SWIZZLE_NOOP, COND_TR, NULL, -}; - -ir_to_mesa_dst_reg ir_to_mesa_address_reg = { - PROGRAM_ADDRESS, 0, WRITEMASK_X, COND_TR, NULL -}; - -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); - prog->InfoLog = talloc_vasprintf_append(prog->InfoLog, fmt, args); - va_end(args); - - prog->LinkStatus = GL_FALSE; -} - -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::ir_to_mesa_emit_op3(ir_instruction *ir, - enum prog_opcode op, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg src0, - ir_to_mesa_src_reg src1, - ir_to_mesa_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) { - ir_to_mesa_emit_op1(ir, OPCODE_ARL, ir_to_mesa_address_reg, - *dst.reladdr); - - num_reladdr--; - } - assert(num_reladdr == 0); - - inst->op = op; - inst->dst_reg = dst; - inst->src_reg[0] = src0; - inst->src_reg[1] = src1; - inst->src_reg[2] = src2; - inst->ir = ir; - - inst->function = NULL; - - this->instructions.push_tail(inst); - - return inst; -} - - -ir_to_mesa_instruction * -ir_to_mesa_visitor::ir_to_mesa_emit_op2(ir_instruction *ir, - enum prog_opcode op, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg src0, - ir_to_mesa_src_reg src1) -{ - return ir_to_mesa_emit_op3(ir, op, dst, src0, src1, ir_to_mesa_undef); -} - -ir_to_mesa_instruction * -ir_to_mesa_visitor::ir_to_mesa_emit_op1(ir_instruction *ir, - enum prog_opcode op, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg src0) -{ - assert(dst.writemask != 0); - return ir_to_mesa_emit_op3(ir, op, dst, - src0, ir_to_mesa_undef, ir_to_mesa_undef); -} - -ir_to_mesa_instruction * -ir_to_mesa_visitor::ir_to_mesa_emit_op0(ir_instruction *ir, - enum prog_opcode op) -{ - return ir_to_mesa_emit_op3(ir, op, ir_to_mesa_undef_dst, - ir_to_mesa_undef, - ir_to_mesa_undef, - ir_to_mesa_undef); -} - -void -ir_to_mesa_visitor::ir_to_mesa_emit_dp(ir_instruction *ir, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg src0, - ir_to_mesa_src_reg src1, - unsigned elements) -{ - static const gl_inst_opcode dot_opcodes[] = { - OPCODE_DP2, OPCODE_DP3, OPCODE_DP4 - }; - - ir_to_mesa_emit_op3(ir, dot_opcodes[elements - 2], - dst, src0, src1, ir_to_mesa_undef); -} - -inline ir_to_mesa_dst_reg -ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg reg) -{ - ir_to_mesa_dst_reg dst_reg; - - dst_reg.file = reg.file; - dst_reg.index = reg.index; - dst_reg.writemask = WRITEMASK_XYZW; - dst_reg.cond_mask = COND_TR; - dst_reg.reladdr = reg.reladdr; - - return dst_reg; -} - -inline ir_to_mesa_src_reg -ir_to_mesa_src_reg_from_dst(ir_to_mesa_dst_reg reg) -{ - return ir_to_mesa_src_reg(reg.file, reg.index, NULL); -} - -/** - * 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::ir_to_mesa_emit_scalar_op2(ir_instruction *ir, - enum prog_opcode op, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg orig_src0, - ir_to_mesa_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; - ir_to_mesa_src_reg src0 = orig_src0; - ir_to_mesa_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 = ir_to_mesa_emit_op2(ir, op, - dst, - src0, - src1); - inst->dst_reg.writemask = this_mask; - done_mask |= this_mask; - } -} - -void -ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op1(ir_instruction *ir, - enum prog_opcode op, - ir_to_mesa_dst_reg dst, - ir_to_mesa_src_reg src0) -{ - ir_to_mesa_src_reg undef = ir_to_mesa_undef; - - undef.swizzle = SWIZZLE_XXXX; - - ir_to_mesa_emit_scalar_op2(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, - ir_to_mesa_dst_reg dst, - const ir_to_mesa_src_reg &src) -{ - /* Vertex programs cannot use the SCS opcode. - */ - if (this->prog->Target == GL_VERTEX_PROGRAM_ARB) { - ir_to_mesa_emit_scalar_op1(ir, op, dst, src); - return; - } - - const unsigned component = (op == OPCODE_SIN) ? 0 : 1; - const unsigned scs_mask = (1U << component); - int done_mask = ~dst.writemask; - ir_to_mesa_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); - ir_to_mesa_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; - ir_to_mesa_dst_reg tmp_dst = ir_to_mesa_dst_reg_from_src(tmp); - - /* Emit the SCS instruction. - */ - inst = ir_to_mesa_emit_op1(ir, OPCODE_SCS, tmp_dst, src0); - inst->dst_reg.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 = ir_to_mesa_emit_op1(ir, OPCODE_SCS, dst, tmp); - inst->dst_reg.writemask = this_mask; - } else { - /* Emit the SCS instruction to write directly to the destination. - */ - ir_to_mesa_instruction *inst = - ir_to_mesa_emit_op1(ir, OPCODE_SCS, dst, src0); - inst->dst_reg.writemask = scs_mask; - } - - done_mask |= this_mask; - } -} - -struct ir_to_mesa_src_reg -ir_to_mesa_visitor::src_reg_for_float(float val) -{ - ir_to_mesa_src_reg src_reg(PROGRAM_CONSTANT, -1, NULL); - - src_reg.index = _mesa_add_unnamed_constant(this->prog->Parameters, - &val, 1, &src_reg.swizzle); - - return src_reg; -} - -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: - 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. - */ -ir_to_mesa_src_reg -ir_to_mesa_visitor::get_temp(const glsl_type *type) -{ - ir_to_mesa_src_reg src_reg; - int swizzle[4]; - int i; - - src_reg.file = PROGRAM_TEMPORARY; - src_reg.index = next_temp; - src_reg.reladdr = NULL; - next_temp += type_size(type); - - if (type->is_array() || type->is_record()) { - src_reg.swizzle = SWIZZLE_NOOP; - } else { - for (i = 0; i < type->vector_elements; i++) - swizzle[i] = i; - for (; i < 4; i++) - swizzle[i] = type->vector_elements - 1; - src_reg.swizzle = MAKE_SWIZZLE4(swizzle[0], swizzle[1], - swizzle[2], swizzle[3]); - } - src_reg.negate = 0; - - return src_reg; -} - -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; - } - - if (ir->mode == ir_var_uniform && strncmp(ir->name, "gl_", 3) == 0) { - unsigned int i; - const struct gl_builtin_uniform_desc *statevar; - - for (i = 0; _mesa_builtin_uniform_desc[i].name; i++) { - if (strcmp(ir->name, _mesa_builtin_uniform_desc[i].name) == 0) - break; - } - - if (!_mesa_builtin_uniform_desc[i].name) { - fail_link(this->shader_program, - "Failed to find builtin uniform `%s'\n", ir->name); - return; - } - - statevar = &_mesa_builtin_uniform_desc[i]; - - int array_count; - if (ir->type->is_array()) { - array_count = ir->type->length; - } else { - array_count = 1; - } - - /* 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 < statevar->num_elements; i++) { - if (statevar->elements[i].swizzle != SWIZZLE_XYZW) { - break; - } - } - - struct variable_storage *storage; - ir_to_mesa_dst_reg dst; - if (i == statevar->num_elements) { - /* We'll set the index later. */ - storage = new(mem_ctx) variable_storage(ir, PROGRAM_STATE_VAR, -1); - this->variables.push_tail(storage); - - dst = ir_to_mesa_undef_dst; - } else { - 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 = ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg(PROGRAM_TEMPORARY, - storage->index, - NULL)); - } - - - for (int a = 0; a < array_count; a++) { - for (unsigned int i = 0; i < statevar->num_elements; i++) { - struct gl_builtin_uniform_element *element = &statevar->elements[i]; - int tokens[STATE_LENGTH]; - - memcpy(tokens, element->tokens, sizeof(element->tokens)); - if (ir->type->is_array()) { - tokens[1] = a; - } - - int index = _mesa_add_state_reference(this->prog->Parameters, - (gl_state_index *)tokens); - - if (storage->file == PROGRAM_STATE_VAR) { - if (storage->index == -1) { - storage->index = index; - } else { - assert(index == - (int)(storage->index + a * statevar->num_elements + i)); - } - } else { - ir_to_mesa_src_reg src(PROGRAM_STATE_VAR, index, NULL); - src.swizzle = element->swizzle; - ir_to_mesa_emit_op1(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 + type_size(ir->type)) { - fail_link(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(ir) ir_dereference_variable(ir->counter); - - if (ir->from != NULL) { - assert(ir->counter != NULL); - - ir_assignment *a = new(ir) ir_assignment(counter, ir->from, NULL); - - a->accept(this); - delete a; - } - - ir_to_mesa_emit_op0(NULL, OPCODE_BGNLOOP); - - if (ir->to) { - ir_expression *e = - new(ir) ir_expression(ir->cmp, glsl_type::bool_type, - counter, ir->to); - ir_if *if_stmt = new(ir) ir_if(e); - - ir_loop_jump *brk = new(ir) ir_loop_jump(ir_loop_jump::jump_break); - - if_stmt->then_instructions.push_tail(brk); - - if_stmt->accept(this); - - delete if_stmt; - delete e; - delete brk; - } - - visit_exec_list(&ir->body_instructions, this); - - if (ir->increment) { - ir_expression *e = - new(ir) ir_expression(ir_binop_add, counter->type, - counter, ir->increment); - - ir_assignment *a = new(ir) ir_assignment(counter, e, NULL); - - a->accept(this); - delete a; - delete e; - } - - ir_to_mesa_emit_op0(NULL, OPCODE_ENDLOOP); -} - -void -ir_to_mesa_visitor::visit(ir_loop_jump *ir) -{ - switch (ir->mode) { - case ir_loop_jump::jump_break: - ir_to_mesa_emit_op0(NULL, OPCODE_BRK); - break; - case ir_loop_jump::jump_continue: - ir_to_mesa_emit_op0(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); - } - } -} - -GLboolean -ir_to_mesa_visitor::try_emit_mad(ir_expression *ir, int mul_operand) -{ - int nonmul_operand = 1 - mul_operand; - ir_to_mesa_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); - ir_to_mesa_emit_op3(ir, OPCODE_MAD, - ir_to_mesa_dst_reg_from_src(this->result), a, b, c); - - return true; -} - -GLboolean -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); - ir_to_mesa_src_reg src = this->result; - - this->result = get_temp(ir->type); - ir_to_mesa_instruction *inst; - inst = ir_to_mesa_emit_op1(ir, OPCODE_MOV, - ir_to_mesa_dst_reg_from_src(this->result), - src); - inst->saturate = true; - - return true; -} - -void -ir_to_mesa_visitor::reladdr_to_temp(ir_instruction *ir, - ir_to_mesa_src_reg *reg, int *num_reladdr) -{ - if (!reg->reladdr) - return; - - ir_to_mesa_emit_op1(ir, OPCODE_ARL, ir_to_mesa_address_reg, *reg->reladdr); - - if (*num_reladdr != 1) { - ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type); - - ir_to_mesa_emit_op1(ir, OPCODE_MOV, - ir_to_mesa_dst_reg_from_src(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); - } - - ir_to_mesa_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 ir_to_mesa_src_reg result_src = get_temp(ir->type); - ir_to_mesa_dst_reg result_dst = ir_to_mesa_dst_reg_from_src(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; - - ir_to_mesa_emit_op1(ir, OPCODE_SWZ, result_dst, src); - this->result = result_src; -} - -void -ir_to_mesa_visitor::visit(ir_expression *ir) -{ - unsigned int operand; - struct ir_to_mesa_src_reg op[Elements(ir->operands)]; - struct ir_to_mesa_src_reg result_src; - struct ir_to_mesa_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; - } - 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 = ir_to_mesa_dst_reg_from_src(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: - ir_to_mesa_emit_op2(ir, OPCODE_SEQ, result_dst, - op[0], src_reg_for_float(0.0)); - break; - case ir_unop_neg: - op[0].negate = ~op[0].negate; - result_src = op[0]; - break; - case ir_unop_abs: - ir_to_mesa_emit_op1(ir, OPCODE_ABS, result_dst, op[0]); - break; - case ir_unop_sign: - ir_to_mesa_emit_op1(ir, OPCODE_SSG, result_dst, op[0]); - break; - case ir_unop_rcp: - ir_to_mesa_emit_scalar_op1(ir, OPCODE_RCP, result_dst, op[0]); - break; - - case ir_unop_exp2: - ir_to_mesa_emit_scalar_op1(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: - ir_to_mesa_emit_scalar_op1(ir, OPCODE_LG2, result_dst, op[0]); - break; - case ir_unop_sin: - ir_to_mesa_emit_scalar_op1(ir, OPCODE_SIN, result_dst, op[0]); - break; - case ir_unop_cos: - ir_to_mesa_emit_scalar_op1(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: - ir_to_mesa_emit_op1(ir, OPCODE_DDX, result_dst, op[0]); - break; - case ir_unop_dFdy: - ir_to_mesa_emit_op1(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)); - - ir_to_mesa_emit_op1(ir, opcode, result_dst, op[0]); - break; - } - - case ir_binop_add: - ir_to_mesa_emit_op2(ir, OPCODE_ADD, result_dst, op[0], op[1]); - break; - case ir_binop_sub: - ir_to_mesa_emit_op2(ir, OPCODE_SUB, result_dst, op[0], op[1]); - break; - - case ir_binop_mul: - ir_to_mesa_emit_op2(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: - ir_to_mesa_emit_op2(ir, OPCODE_SLT, result_dst, op[0], op[1]); - break; - case ir_binop_greater: - ir_to_mesa_emit_op2(ir, OPCODE_SGT, result_dst, op[0], op[1]); - break; - case ir_binop_lequal: - ir_to_mesa_emit_op2(ir, OPCODE_SLE, result_dst, op[0], op[1]); - break; - case ir_binop_gequal: - ir_to_mesa_emit_op2(ir, OPCODE_SGE, result_dst, op[0], op[1]); - break; - case ir_binop_equal: - ir_to_mesa_emit_op2(ir, OPCODE_SEQ, result_dst, op[0], op[1]); - break; - case ir_binop_nequal: - ir_to_mesa_emit_op2(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()) { - ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type); - ir_to_mesa_emit_op2(ir, OPCODE_SNE, - ir_to_mesa_dst_reg_from_src(temp), op[0], op[1]); - ir_to_mesa_emit_dp(ir, result_dst, temp, temp, vector_elements); - ir_to_mesa_emit_op2(ir, OPCODE_SEQ, - result_dst, result_src, src_reg_for_float(0.0)); - } else { - ir_to_mesa_emit_op2(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()) { - ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type); - ir_to_mesa_emit_op2(ir, OPCODE_SNE, - ir_to_mesa_dst_reg_from_src(temp), op[0], op[1]); - ir_to_mesa_emit_dp(ir, result_dst, temp, temp, vector_elements); - ir_to_mesa_emit_op2(ir, OPCODE_SNE, - result_dst, result_src, src_reg_for_float(0.0)); - } else { - ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]); - } - break; - - case ir_unop_any: - assert(ir->operands[0]->type->is_vector()); - ir_to_mesa_emit_dp(ir, result_dst, op[0], op[0], - ir->operands[0]->type->vector_elements); - ir_to_mesa_emit_op2(ir, OPCODE_SNE, - result_dst, result_src, src_reg_for_float(0.0)); - break; - - case ir_binop_logic_xor: - ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]); - break; - - case ir_binop_logic_or: - /* This could be a saturated add and skip the SNE. */ - ir_to_mesa_emit_op2(ir, OPCODE_ADD, - result_dst, - op[0], op[1]); - - ir_to_mesa_emit_op2(ir, OPCODE_SNE, - result_dst, - result_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". */ - ir_to_mesa_emit_op2(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); - ir_to_mesa_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). */ - ir_to_mesa_emit_scalar_op1(ir, OPCODE_RSQ, result_dst, op[0]); - ir_to_mesa_emit_op2(ir, OPCODE_MUL, result_dst, result_src, op[0]); - /* For incoming channels <= 0, set the result to 0. */ - op[0].negate = ~op[0].negate; - ir_to_mesa_emit_op3(ir, OPCODE_CMP, result_dst, - op[0], result_src, src_reg_for_float(0.0)); - break; - case ir_unop_rsq: - ir_to_mesa_emit_scalar_op1(ir, OPCODE_RSQ, result_dst, op[0]); - break; - case ir_unop_i2f: - case ir_unop_b2f: - case ir_unop_b2i: - /* Mesa IR lacks types, ints are stored as truncated floats. */ - result_src = op[0]; - break; - case ir_unop_f2i: - ir_to_mesa_emit_op1(ir, OPCODE_TRUNC, result_dst, op[0]); - break; - case ir_unop_f2b: - case ir_unop_i2b: - ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, - op[0], src_reg_for_float(0.0)); - break; - case ir_unop_trunc: - ir_to_mesa_emit_op1(ir, OPCODE_TRUNC, result_dst, op[0]); - break; - case ir_unop_ceil: - op[0].negate = ~op[0].negate; - ir_to_mesa_emit_op1(ir, OPCODE_FLR, result_dst, op[0]); - result_src.negate = ~result_src.negate; - break; - case ir_unop_floor: - ir_to_mesa_emit_op1(ir, OPCODE_FLR, result_dst, op[0]); - break; - case ir_unop_fract: - ir_to_mesa_emit_op1(ir, OPCODE_FRC, result_dst, op[0]); - break; - - case ir_binop_min: - ir_to_mesa_emit_op2(ir, OPCODE_MIN, result_dst, op[0], op[1]); - break; - case ir_binop_max: - ir_to_mesa_emit_op2(ir, OPCODE_MAX, result_dst, op[0], op[1]); - break; - case ir_binop_pow: - ir_to_mesa_emit_scalar_op2(ir, OPCODE_POW, result_dst, op[0], op[1]); - break; - - case ir_unop_bit_not: - case ir_unop_u2f: - 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) -{ - ir_to_mesa_src_reg src_reg; - 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_reg = this->result; - assert(src_reg.file != PROGRAM_UNDEFINED); - - for (i = 0; i < 4; i++) { - if (i < ir->type->vector_elements) { - switch (i) { - case 0: - swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.x); - break; - case 1: - swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.y); - break; - case 2: - swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.z); - break; - case 3: - swizzle[i] = GET_SWZ(src_reg.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_reg.swizzle = MAKE_SWIZZLE4(swizzle[0], - swizzle[1], - swizzle[2], - swizzle[3]); - - this->result = src_reg; -} - -void -ir_to_mesa_visitor::visit(ir_dereference_variable *ir) -{ - variable_storage *entry = find_variable_storage(ir->var); - - if (!entry) { - switch (ir->var->mode) { - case ir_var_uniform: - entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_UNIFORM, - ir->var->location); - this->variables.push_tail(entry); - break; - case ir_var_in: - case ir_var_out: - case ir_var_inout: - case ir_var_system_value: - /* The linker assigns locations for varyings and attributes, - * including deprecated builtins (like gl_Color), user-assign - * generic attributes (glBindVertexLocation), and - * user-defined varyings. - * - * FINISHME: We would hit this path for function arguments. Fix! - */ - assert(ir->var->location != -1); - if (ir->var->mode == ir_var_in || - ir->var->mode == ir_var_inout) { - entry = new(mem_ctx) variable_storage(ir->var, - PROGRAM_INPUT, - ir->var->location); - - if (this->prog->Target == GL_VERTEX_PROGRAM_ARB && - ir->var->location >= VERT_ATTRIB_GENERIC0) { - _mesa_add_attribute(prog->Attributes, - ir->var->name, - _mesa_sizeof_glsl_type(ir->var->type->gl_type), - ir->var->type->gl_type, - ir->var->location - VERT_ATTRIB_GENERIC0); - } - } else if (ir->var->mode == ir_var_system_value) { - entry = new(mem_ctx) variable_storage(ir->var, - PROGRAM_SYSTEM_VALUE, - ir->var->location); - } else { - entry = new(mem_ctx) variable_storage(ir->var, - PROGRAM_OUTPUT, - ir->var->location); - } - - break; - case ir_var_auto: - case ir_var_temporary: - entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_TEMPORARY, - this->next_temp); - this->variables.push_tail(entry); - - next_temp += type_size(ir->var->type); - break; - } - - if (!entry) { - printf("Failed to make storage for %s\n", ir->var->name); - exit(1); - } - } - - this->result = ir_to_mesa_src_reg(entry->file, entry->index, ir->var->type); -} - -void -ir_to_mesa_visitor::visit(ir_dereference_array *ir) -{ - ir_constant *index; - ir_to_mesa_src_reg src_reg; - int element_size = type_size(ir->type); - - index = ir->array_index->constant_expression_value(); - - ir->array->accept(this); - src_reg = this->result; - - if (index) { - src_reg.index += index->value.i[0] * element_size; - } else { - ir_to_mesa_src_reg array_base = this->result; - /* 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); - - ir_to_mesa_src_reg index_reg; - - if (element_size == 1) { - index_reg = this->result; - } else { - index_reg = get_temp(glsl_type::float_type); - - ir_to_mesa_emit_op2(ir, OPCODE_MUL, - ir_to_mesa_dst_reg_from_src(index_reg), - this->result, src_reg_for_float(element_size)); - } - - src_reg.reladdr = talloc(mem_ctx, ir_to_mesa_src_reg); - memcpy(src_reg.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_reg.swizzle = swizzle_for_size(ir->type->vector_elements); - else - src_reg.swizzle = SWIZZLE_NOOP; - - this->result = src_reg; -} - -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 struct ir_to_mesa_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 ir_to_mesa_dst_reg_from_src(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) -{ - struct ir_to_mesa_dst_reg l; - struct ir_to_mesa_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); - ir_to_mesa_src_reg condition = this->result; - - for (i = 0; i < type_size(ir->lhs->type); i++) { - if (switch_order) { - ir_to_mesa_emit_op3(ir, OPCODE_CMP, l, - condition, ir_to_mesa_src_reg_from_dst(l), r); - } else { - ir_to_mesa_emit_op3(ir, OPCODE_CMP, l, - condition, r, ir_to_mesa_src_reg_from_dst(l)); - } - - l.index++; - r.index++; - } - } else { - for (i = 0; i < type_size(ir->lhs->type); i++) { - ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r); - l.index++; - r.index++; - } - } -} - - -void -ir_to_mesa_visitor::visit(ir_constant *ir) -{ - ir_to_mesa_src_reg src_reg; - 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) { - ir_to_mesa_src_reg temp_base = get_temp(ir->type); - ir_to_mesa_dst_reg temp = ir_to_mesa_dst_reg_from_src(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_reg = this->result; - - for (i = 0; i < (unsigned int)size; i++) { - ir_to_mesa_emit_op1(ir, OPCODE_MOV, temp, src_reg); - - src_reg.index++; - temp.index++; - } - } - this->result = temp_base; - return; - } - - if (ir->type->is_array()) { - ir_to_mesa_src_reg temp_base = get_temp(ir->type); - ir_to_mesa_dst_reg temp = ir_to_mesa_dst_reg_from_src(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_reg = this->result; - for (int j = 0; j < size; j++) { - ir_to_mesa_emit_op1(ir, OPCODE_MOV, temp, src_reg); - - src_reg.index++; - temp.index++; - } - } - this->result = temp_base; - return; - } - - if (ir->type->is_matrix()) { - ir_to_mesa_src_reg mat = get_temp(ir->type); - ir_to_mesa_dst_reg mat_column = ir_to_mesa_dst_reg_from_src(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_reg = ir_to_mesa_src_reg(PROGRAM_CONSTANT, -1, NULL); - src_reg.index = _mesa_add_unnamed_constant(this->prog->Parameters, - values, - ir->type->vector_elements, - &src_reg.swizzle); - ir_to_mesa_emit_op1(ir, OPCODE_MOV, mat_column, src_reg); - - mat_column.index++; - } - - this->result = mat; - return; - } - - src_reg.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 = ir_to_mesa_src_reg(PROGRAM_CONSTANT, -1, ir->type); - this->result.index = _mesa_add_unnamed_constant(this->prog->Parameters, - 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 = talloc(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 = ir_to_mesa_undef; - } - - 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); - ir_to_mesa_src_reg r = this->result; - - ir_to_mesa_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++) { - ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r); - l.index++; - r.index++; - } - } - - sig_iter.next(); - } - assert(!sig_iter.has_next()); - - /* Emit call instruction */ - call_inst = ir_to_mesa_emit_op1(ir, OPCODE_CAL, - ir_to_mesa_undef_dst, ir_to_mesa_undef); - 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); - - ir_to_mesa_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); - ir_to_mesa_dst_reg l = ir_to_mesa_dst_reg_from_src(this->result); - - for (i = 0; i < type_size(param->type); i++) { - ir_to_mesa_emit_op1(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) -{ - ir_to_mesa_src_reg result_src, coord, lod_info, projector; - ir_to_mesa_dst_reg result_dst, coord_dst; - ir_to_mesa_instruction *inst = NULL; - prog_opcode opcode = OPCODE_NOP; - - 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 = ir_to_mesa_dst_reg_from_src(coord); - ir_to_mesa_emit_op1(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 = ir_to_mesa_dst_reg_from_src(result_src); - - switch (ir->op) { - case ir_tex: - 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: - case ir_txf: - assert(!"GLSL 1.30 features unsupported"); - break; - } - - if (ir->projector) { - if (opcode == OPCODE_TEX) { - /* Slot the projector in as the last component of the coord. */ - coord_dst.writemask = WRITEMASK_W; - ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst, projector); - coord_dst.writemask = WRITEMASK_XYZW; - opcode = OPCODE_TXP; - } else { - ir_to_mesa_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; - ir_to_mesa_emit_op1(ir, OPCODE_RCP, coord_dst, projector); - - coord_dst.writemask = WRITEMASK_XYZ; - ir_to_mesa_emit_op2(ir, OPCODE_MUL, coord_dst, coord, coord_w); - - coord_dst.writemask = WRITEMASK_XYZW; - coord.swizzle = SWIZZLE_XYZW; - } - } - - if (ir->shadow_comparitor) { - /* Slot the shadow value in as the second to last component of the - * coord. - */ - ir->shadow_comparitor->accept(this); - coord_dst.writemask = WRITEMASK_Z; - ir_to_mesa_emit_op1(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; - ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst, lod_info); - coord_dst.writemask = WRITEMASK_XYZW; - } - - inst = ir_to_mesa_emit_op1(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); - - const glsl_type *sampler_type = ir->sampler->type; - - 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()) { - ir_to_mesa_dst_reg l; - int i; - - assert(current_function); - - ir->get_value()->accept(this); - ir_to_mesa_src_reg r = this->result; - - l = ir_to_mesa_dst_reg_from_src(current_function->return_reg); - - for (i = 0; i < type_size(current_function->sig->return_type); i++) { - ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r); - l.index++; - r.index++; - } - } - - ir_to_mesa_emit_op0(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; - ir_to_mesa_emit_op1(ir, OPCODE_KIL, ir_to_mesa_undef_dst, this->result); - } else { - ir_to_mesa_emit_op0(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, *else_inst = NULL; - 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) { - ir_to_mesa_src_reg temp = get_temp(glsl_type::bool_type); - cond_inst = ir_to_mesa_emit_op1(ir->condition, OPCODE_MOV, - ir_to_mesa_dst_reg_from_src(temp), - result); - } - cond_inst->cond_update = GL_TRUE; - - if_inst = ir_to_mesa_emit_op0(ir->condition, OPCODE_IF); - if_inst->dst_reg.cond_mask = COND_NE; - } else { - if_inst = ir_to_mesa_emit_op1(ir->condition, - OPCODE_IF, ir_to_mesa_undef_dst, - this->result); - } - - this->instructions.push_tail(if_inst); - - visit_exec_list(&ir->then_instructions, this); - - if (!ir->else_instructions.is_empty()) { - else_inst = ir_to_mesa_emit_op0(ir->condition, OPCODE_ELSE); - visit_exec_list(&ir->else_instructions, this); - } - - if_inst = ir_to_mesa_emit_op1(ir->condition, OPCODE_ENDIF, - ir_to_mesa_undef_dst, ir_to_mesa_undef); -} - -ir_to_mesa_visitor::ir_to_mesa_visitor() -{ - result.file = PROGRAM_UNDEFINED; - next_temp = 1; - next_signature_id = 1; - current_function = NULL; - mem_ctx = talloc_new(NULL); -} - -ir_to_mesa_visitor::~ir_to_mesa_visitor() -{ - talloc_free(mem_ctx); -} - -static struct prog_src_register -mesa_src_reg_from_ir_src_reg(ir_to_mesa_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 = talloc_zero_array(v->mem_ctx, int, if_count); - loop_stack = talloc_zero_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); - } -} - -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); -} - -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 = talloc_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] = 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) { - fail_link(shader_program, "Allocation of uniform `%s' to target " - "failed (%d vs %d)\n", - uniform->Name, index, parameter_index); - } - } - } - - talloc_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 = talloc_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) { - fail_link(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 = talloc_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 = talloc_new(NULL); - - set_uniform_initializer(ctx, mem_ctx, shader_program, var->name, - var->type, var->constant_value); - } - } - - talloc_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 = talloc_zero_array(mem_ctx, - ir_to_mesa_instruction *, - this->next_temp * 4); - - foreach_iter(exec_list_iterator, iter, this->instructions) { - ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get(); - - /* 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_reg[r].index * 4; - - if (inst->src_reg[r].file != PROGRAM_TEMPORARY || - inst->src_reg[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_reg[r].swizzle, i); - ir_to_mesa_instruction *copy_chan = acp[acp_base + src_chan]; - - if (!copy_chan) { - good = false; - break; - } - - if (!first) { - first = copy_chan; - } else { - if (first->src_reg[0].file != copy_chan->src_reg[0].file || - first->src_reg[0].index != copy_chan->src_reg[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_reg[r].file = first->src_reg[0].file; - inst->src_reg[r].index = first->src_reg[0].index; - - int swizzle = 0; - for (int i = 0; i < 4; i++) { - int src_chan = GET_SWZ(inst->src_reg[r].swizzle, i); - ir_to_mesa_instruction *copy_inst = acp[acp_base + src_chan]; - swizzle |= (GET_SWZ(copy_inst->src_reg[0].swizzle, src_chan) << - (3 * i)); - } - inst->src_reg[r].swizzle = swizzle; - } - } - - switch (inst->op) { - case OPCODE_BGNLOOP: - case OPCODE_ENDLOOP: - case OPCODE_ELSE: - case OPCODE_ENDIF: - /* End of a basic block, clear the ACP entirely. */ - memset(acp, 0, sizeof(*acp) * this->next_temp * 4); - break; - - default: - /* Continuing the block, clear any written channels from - * the ACP. - */ - if (inst->dst_reg.file == PROGRAM_TEMPORARY) { - if (inst->dst_reg.reladdr) { - memset(acp, 0, sizeof(*acp) * this->next_temp * 4); - } else { - for (int i = 0; i < 4; i++) { - if (inst->dst_reg.writemask & (1 << i)) { - acp[4 * inst->dst_reg.index + i] = NULL; - } - } - } - } - break; - } - - /* If this is a copy, add it to the ACP. */ - if (inst->op == OPCODE_MOV && - inst->dst_reg.file == PROGRAM_TEMPORARY && - !inst->dst_reg.reladdr && - !inst->saturate && - !inst->src_reg[0].reladdr && - !inst->src_reg[0].negate) { - for (int i = 0; i < 4; i++) { - if (inst->dst_reg.writemask & (1 << i)) { - acp[4 * inst->dst_reg.index + i] = inst; - } - } - } - } - - talloc_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.ir_to_mesa_emit_op0(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.ir_to_mesa_emit_op0(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.ir_to_mesa_emit_op0(NULL, OPCODE_RET); - - ir_to_mesa_instruction *end; - end = v.ir_to_mesa_emit_op0(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 = talloc_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_reg.file; - mesa_inst->DstReg.Index = inst->dst_reg.index; - mesa_inst->DstReg.CondMask = inst->dst_reg.cond_mask; - mesa_inst->DstReg.WriteMask = inst->dst_reg.writemask; - mesa_inst->DstReg.RelAddr = inst->dst_reg.reladdr != NULL; - mesa_inst->SrcReg[0] = mesa_src_reg_from_ir_src_reg(inst->src_reg[0]); - mesa_inst->SrcReg[1] = mesa_src_reg_from_ir_src_reg(inst->src_reg[1]); - mesa_inst->SrcReg[2] = mesa_src_reg_from_ir_src_reg(inst->src_reg[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; - - if (options->EmitNoIfs && mesa_inst->Opcode == OPCODE_IF) { - fail_link(shader_program, "Couldn't flatten if statement\n"); - } - - switch (mesa_inst->Opcode) { - 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); - - _mesa_reference_program(ctx, &shader->Program, prog); - - if ((ctx->Shader.Flags & GLSL_NO_OPT) == 0) { - _mesa_optimize_program(ctx, prog); - } - - return prog; -} - -extern "C" { - -/** - * Called via ctx->Driver.CompilerShader(). - * This is a no-op. - * XXX can we remove the ctx->Driver.CompileShader() hook? - */ -GLboolean -_mesa_ir_compile_shader(struct gl_context *ctx, struct gl_shader *shader) -{ - assert(shader->CompileStatus); - (void) ctx; - - return GL_TRUE; -} - - -/** - * 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 - | ((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->EmitNoIfs) { - progress = lower_discard(ir) || progress; - progress = lower_if_to_cond_assign(ir) || 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 shader %d:\n", 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); - } - - talloc_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); - - talloc_free(state); - - if (shader->CompileStatus) { - if (!ctx->Driver.CompileShader(ctx, shader)) - shader->CompileStatus = GL_FALSE; - } -} - - -/** - * 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) { - fail_link(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" +} + +static int swizzle_for_size(int size); + +/** + * This struct is a corresponding struct to Mesa prog_src_register, with + * wider fields. + */ +typedef struct ir_to_mesa_src_reg { + ir_to_mesa_src_reg(int file, int index, const glsl_type *type) + { + this->file = (gl_register_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; + } + + ir_to_mesa_src_reg() + { + this->file = PROGRAM_UNDEFINED; + this->index = 0; + this->swizzle = 0; + this->negate = 0; + this->reladdr = NULL; + } + + 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. */ + ir_to_mesa_src_reg *reladdr; +} ir_to_mesa_src_reg; + +typedef struct ir_to_mesa_dst_reg { + int 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. */ + ir_to_mesa_src_reg *reladdr; +} ir_to_mesa_dst_reg; + +extern ir_to_mesa_src_reg ir_to_mesa_undef; + +class ir_to_mesa_instruction : public exec_node { +public: + /* Callers of this talloc-based new need not call delete. It's + * easier to just talloc_free 'ctx' (or any of its ancestors). */ + static void* operator new(size_t size, void *ctx) + { + void *node; + + node = talloc_zero_size(ctx, size); + assert(node != NULL); + + return node; + } + + enum prog_opcode op; + ir_to_mesa_dst_reg dst_reg; + ir_to_mesa_src_reg src_reg[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. */ + ir_to_mesa_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); + + ir_to_mesa_src_reg get_temp(const glsl_type *type); + void reladdr_to_temp(ir_instruction *ir, + ir_to_mesa_src_reg *reg, int *num_reladdr); + + struct ir_to_mesa_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 *); + /*@}*/ + + struct ir_to_mesa_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 *ir_to_mesa_emit_op0(ir_instruction *ir, + enum prog_opcode op); + + ir_to_mesa_instruction *ir_to_mesa_emit_op1(ir_instruction *ir, + enum prog_opcode op, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg src0); + + ir_to_mesa_instruction *ir_to_mesa_emit_op2(ir_instruction *ir, + enum prog_opcode op, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg src0, + ir_to_mesa_src_reg src1); + + ir_to_mesa_instruction *ir_to_mesa_emit_op3(ir_instruction *ir, + enum prog_opcode op, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg src0, + ir_to_mesa_src_reg src1, + ir_to_mesa_src_reg src2); + + /** + * Emit the correct dot-product instruction for the type of arguments + * + * \sa ir_to_mesa_emit_op2 + */ + void ir_to_mesa_emit_dp(ir_instruction *ir, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg src0, + ir_to_mesa_src_reg src1, + unsigned elements); + + void ir_to_mesa_emit_scalar_op1(ir_instruction *ir, + enum prog_opcode op, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg src0); + + void ir_to_mesa_emit_scalar_op2(ir_instruction *ir, + enum prog_opcode op, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg src0, + ir_to_mesa_src_reg src1); + + void emit_scs(ir_instruction *ir, enum prog_opcode op, + ir_to_mesa_dst_reg dst, + const ir_to_mesa_src_reg &src); + + GLboolean try_emit_mad(ir_expression *ir, + int mul_operand); + GLboolean 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; +}; + +ir_to_mesa_src_reg ir_to_mesa_undef = ir_to_mesa_src_reg(PROGRAM_UNDEFINED, 0, NULL); + +ir_to_mesa_dst_reg ir_to_mesa_undef_dst = { + PROGRAM_UNDEFINED, 0, SWIZZLE_NOOP, COND_TR, NULL, +}; + +ir_to_mesa_dst_reg ir_to_mesa_address_reg = { + PROGRAM_ADDRESS, 0, WRITEMASK_X, COND_TR, NULL +}; + +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); + prog->InfoLog = talloc_vasprintf_append(prog->InfoLog, fmt, args); + va_end(args); + + prog->LinkStatus = GL_FALSE; +} + +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::ir_to_mesa_emit_op3(ir_instruction *ir, + enum prog_opcode op, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg src0, + ir_to_mesa_src_reg src1, + ir_to_mesa_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) { + ir_to_mesa_emit_op1(ir, OPCODE_ARL, ir_to_mesa_address_reg, + *dst.reladdr); + + num_reladdr--; + } + assert(num_reladdr == 0); + + inst->op = op; + inst->dst_reg = dst; + inst->src_reg[0] = src0; + inst->src_reg[1] = src1; + inst->src_reg[2] = src2; + inst->ir = ir; + + inst->function = NULL; + + this->instructions.push_tail(inst); + + return inst; +} + + +ir_to_mesa_instruction * +ir_to_mesa_visitor::ir_to_mesa_emit_op2(ir_instruction *ir, + enum prog_opcode op, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg src0, + ir_to_mesa_src_reg src1) +{ + return ir_to_mesa_emit_op3(ir, op, dst, src0, src1, ir_to_mesa_undef); +} + +ir_to_mesa_instruction * +ir_to_mesa_visitor::ir_to_mesa_emit_op1(ir_instruction *ir, + enum prog_opcode op, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg src0) +{ + assert(dst.writemask != 0); + return ir_to_mesa_emit_op3(ir, op, dst, + src0, ir_to_mesa_undef, ir_to_mesa_undef); +} + +ir_to_mesa_instruction * +ir_to_mesa_visitor::ir_to_mesa_emit_op0(ir_instruction *ir, + enum prog_opcode op) +{ + return ir_to_mesa_emit_op3(ir, op, ir_to_mesa_undef_dst, + ir_to_mesa_undef, + ir_to_mesa_undef, + ir_to_mesa_undef); +} + +void +ir_to_mesa_visitor::ir_to_mesa_emit_dp(ir_instruction *ir, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg src0, + ir_to_mesa_src_reg src1, + unsigned elements) +{ + static const gl_inst_opcode dot_opcodes[] = { + OPCODE_DP2, OPCODE_DP3, OPCODE_DP4 + }; + + ir_to_mesa_emit_op3(ir, dot_opcodes[elements - 2], + dst, src0, src1, ir_to_mesa_undef); +} + +inline ir_to_mesa_dst_reg +ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg reg) +{ + ir_to_mesa_dst_reg dst_reg; + + dst_reg.file = reg.file; + dst_reg.index = reg.index; + dst_reg.writemask = WRITEMASK_XYZW; + dst_reg.cond_mask = COND_TR; + dst_reg.reladdr = reg.reladdr; + + return dst_reg; +} + +inline ir_to_mesa_src_reg +ir_to_mesa_src_reg_from_dst(ir_to_mesa_dst_reg reg) +{ + return ir_to_mesa_src_reg(reg.file, reg.index, NULL); +} + +/** + * 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::ir_to_mesa_emit_scalar_op2(ir_instruction *ir, + enum prog_opcode op, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg orig_src0, + ir_to_mesa_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; + ir_to_mesa_src_reg src0 = orig_src0; + ir_to_mesa_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 = ir_to_mesa_emit_op2(ir, op, + dst, + src0, + src1); + inst->dst_reg.writemask = this_mask; + done_mask |= this_mask; + } +} + +void +ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op1(ir_instruction *ir, + enum prog_opcode op, + ir_to_mesa_dst_reg dst, + ir_to_mesa_src_reg src0) +{ + ir_to_mesa_src_reg undef = ir_to_mesa_undef; + + undef.swizzle = SWIZZLE_XXXX; + + ir_to_mesa_emit_scalar_op2(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, + ir_to_mesa_dst_reg dst, + const ir_to_mesa_src_reg &src) +{ + /* Vertex programs cannot use the SCS opcode. + */ + if (this->prog->Target == GL_VERTEX_PROGRAM_ARB) { + ir_to_mesa_emit_scalar_op1(ir, op, dst, src); + return; + } + + const unsigned component = (op == OPCODE_SIN) ? 0 : 1; + const unsigned scs_mask = (1U << component); + int done_mask = ~dst.writemask; + ir_to_mesa_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); + ir_to_mesa_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; + ir_to_mesa_dst_reg tmp_dst = ir_to_mesa_dst_reg_from_src(tmp); + + /* Emit the SCS instruction. + */ + inst = ir_to_mesa_emit_op1(ir, OPCODE_SCS, tmp_dst, src0); + inst->dst_reg.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 = ir_to_mesa_emit_op1(ir, OPCODE_SCS, dst, tmp); + inst->dst_reg.writemask = this_mask; + } else { + /* Emit the SCS instruction to write directly to the destination. + */ + ir_to_mesa_instruction *inst = + ir_to_mesa_emit_op1(ir, OPCODE_SCS, dst, src0); + inst->dst_reg.writemask = scs_mask; + } + + done_mask |= this_mask; + } +} + +struct ir_to_mesa_src_reg +ir_to_mesa_visitor::src_reg_for_float(float val) +{ + ir_to_mesa_src_reg src_reg(PROGRAM_CONSTANT, -1, NULL); + + src_reg.index = _mesa_add_unnamed_constant(this->prog->Parameters, + &val, 1, &src_reg.swizzle); + + return src_reg; +} + +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. + */ +ir_to_mesa_src_reg +ir_to_mesa_visitor::get_temp(const glsl_type *type) +{ + ir_to_mesa_src_reg src_reg; + int swizzle[4]; + int i; + + src_reg.file = PROGRAM_TEMPORARY; + src_reg.index = next_temp; + src_reg.reladdr = NULL; + next_temp += type_size(type); + + if (type->is_array() || type->is_record()) { + src_reg.swizzle = SWIZZLE_NOOP; + } else { + for (i = 0; i < type->vector_elements; i++) + swizzle[i] = i; + for (; i < 4; i++) + swizzle[i] = type->vector_elements - 1; + src_reg.swizzle = MAKE_SWIZZLE4(swizzle[0], swizzle[1], + swizzle[2], swizzle[3]); + } + src_reg.negate = 0; + + return src_reg; +} + +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; + } + + if (ir->mode == ir_var_uniform && strncmp(ir->name, "gl_", 3) == 0) { + unsigned int i; + const struct gl_builtin_uniform_desc *statevar; + + for (i = 0; _mesa_builtin_uniform_desc[i].name; i++) { + if (strcmp(ir->name, _mesa_builtin_uniform_desc[i].name) == 0) + break; + } + + if (!_mesa_builtin_uniform_desc[i].name) { + fail_link(this->shader_program, + "Failed to find builtin uniform `%s'\n", ir->name); + return; + } + + statevar = &_mesa_builtin_uniform_desc[i]; + + int array_count; + if (ir->type->is_array()) { + array_count = ir->type->length; + } else { + array_count = 1; + } + + /* 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 < statevar->num_elements; i++) { + if (statevar->elements[i].swizzle != SWIZZLE_XYZW) { + break; + } + } + + struct variable_storage *storage; + ir_to_mesa_dst_reg dst; + if (i == statevar->num_elements) { + /* We'll set the index later. */ + storage = new(mem_ctx) variable_storage(ir, PROGRAM_STATE_VAR, -1); + this->variables.push_tail(storage); + + dst = ir_to_mesa_undef_dst; + } else { + 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 = ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg(PROGRAM_TEMPORARY, + storage->index, + NULL)); + } + + + for (int a = 0; a < array_count; a++) { + for (unsigned int i = 0; i < statevar->num_elements; i++) { + struct gl_builtin_uniform_element *element = &statevar->elements[i]; + int tokens[STATE_LENGTH]; + + memcpy(tokens, element->tokens, sizeof(element->tokens)); + if (ir->type->is_array()) { + tokens[1] = a; + } + + int index = _mesa_add_state_reference(this->prog->Parameters, + (gl_state_index *)tokens); + + if (storage->file == PROGRAM_STATE_VAR) { + if (storage->index == -1) { + storage->index = index; + } else { + assert(index == + (int)(storage->index + a * statevar->num_elements + i)); + } + } else { + ir_to_mesa_src_reg src(PROGRAM_STATE_VAR, index, NULL); + src.swizzle = element->swizzle; + ir_to_mesa_emit_op1(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 + type_size(ir->type)) { + fail_link(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(ir) ir_dereference_variable(ir->counter); + + if (ir->from != NULL) { + assert(ir->counter != NULL); + + ir_assignment *a = new(ir) ir_assignment(counter, ir->from, NULL); + + a->accept(this); + delete a; + } + + ir_to_mesa_emit_op0(NULL, OPCODE_BGNLOOP); + + if (ir->to) { + ir_expression *e = + new(ir) ir_expression(ir->cmp, glsl_type::bool_type, + counter, ir->to); + ir_if *if_stmt = new(ir) ir_if(e); + + ir_loop_jump *brk = new(ir) ir_loop_jump(ir_loop_jump::jump_break); + + if_stmt->then_instructions.push_tail(brk); + + if_stmt->accept(this); + + delete if_stmt; + delete e; + delete brk; + } + + visit_exec_list(&ir->body_instructions, this); + + if (ir->increment) { + ir_expression *e = + new(ir) ir_expression(ir_binop_add, counter->type, + counter, ir->increment); + + ir_assignment *a = new(ir) ir_assignment(counter, e, NULL); + + a->accept(this); + delete a; + delete e; + } + + ir_to_mesa_emit_op0(NULL, OPCODE_ENDLOOP); +} + +void +ir_to_mesa_visitor::visit(ir_loop_jump *ir) +{ + switch (ir->mode) { + case ir_loop_jump::jump_break: + ir_to_mesa_emit_op0(NULL, OPCODE_BRK); + break; + case ir_loop_jump::jump_continue: + ir_to_mesa_emit_op0(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); + } + } +} + +GLboolean +ir_to_mesa_visitor::try_emit_mad(ir_expression *ir, int mul_operand) +{ + int nonmul_operand = 1 - mul_operand; + ir_to_mesa_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); + ir_to_mesa_emit_op3(ir, OPCODE_MAD, + ir_to_mesa_dst_reg_from_src(this->result), a, b, c); + + return true; +} + +GLboolean +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); + ir_to_mesa_src_reg src = this->result; + + this->result = get_temp(ir->type); + ir_to_mesa_instruction *inst; + inst = ir_to_mesa_emit_op1(ir, OPCODE_MOV, + ir_to_mesa_dst_reg_from_src(this->result), + src); + inst->saturate = true; + + return true; +} + +void +ir_to_mesa_visitor::reladdr_to_temp(ir_instruction *ir, + ir_to_mesa_src_reg *reg, int *num_reladdr) +{ + if (!reg->reladdr) + return; + + ir_to_mesa_emit_op1(ir, OPCODE_ARL, ir_to_mesa_address_reg, *reg->reladdr); + + if (*num_reladdr != 1) { + ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type); + + ir_to_mesa_emit_op1(ir, OPCODE_MOV, + ir_to_mesa_dst_reg_from_src(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); + } + + ir_to_mesa_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 ir_to_mesa_src_reg result_src = get_temp(ir->type); + ir_to_mesa_dst_reg result_dst = ir_to_mesa_dst_reg_from_src(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; + + ir_to_mesa_emit_op1(ir, OPCODE_SWZ, result_dst, src); + this->result = result_src; +} + +void +ir_to_mesa_visitor::visit(ir_expression *ir) +{ + unsigned int operand; + struct ir_to_mesa_src_reg op[Elements(ir->operands)]; + struct ir_to_mesa_src_reg result_src; + struct ir_to_mesa_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; + } + 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 = ir_to_mesa_dst_reg_from_src(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: + ir_to_mesa_emit_op2(ir, OPCODE_SEQ, result_dst, + op[0], src_reg_for_float(0.0)); + break; + case ir_unop_neg: + op[0].negate = ~op[0].negate; + result_src = op[0]; + break; + case ir_unop_abs: + ir_to_mesa_emit_op1(ir, OPCODE_ABS, result_dst, op[0]); + break; + case ir_unop_sign: + ir_to_mesa_emit_op1(ir, OPCODE_SSG, result_dst, op[0]); + break; + case ir_unop_rcp: + ir_to_mesa_emit_scalar_op1(ir, OPCODE_RCP, result_dst, op[0]); + break; + + case ir_unop_exp2: + ir_to_mesa_emit_scalar_op1(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: + ir_to_mesa_emit_scalar_op1(ir, OPCODE_LG2, result_dst, op[0]); + break; + case ir_unop_sin: + ir_to_mesa_emit_scalar_op1(ir, OPCODE_SIN, result_dst, op[0]); + break; + case ir_unop_cos: + ir_to_mesa_emit_scalar_op1(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: + ir_to_mesa_emit_op1(ir, OPCODE_DDX, result_dst, op[0]); + break; + case ir_unop_dFdy: + ir_to_mesa_emit_op1(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)); + + ir_to_mesa_emit_op1(ir, opcode, result_dst, op[0]); + break; + } + + case ir_binop_add: + ir_to_mesa_emit_op2(ir, OPCODE_ADD, result_dst, op[0], op[1]); + break; + case ir_binop_sub: + ir_to_mesa_emit_op2(ir, OPCODE_SUB, result_dst, op[0], op[1]); + break; + + case ir_binop_mul: + ir_to_mesa_emit_op2(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: + ir_to_mesa_emit_op2(ir, OPCODE_SLT, result_dst, op[0], op[1]); + break; + case ir_binop_greater: + ir_to_mesa_emit_op2(ir, OPCODE_SGT, result_dst, op[0], op[1]); + break; + case ir_binop_lequal: + ir_to_mesa_emit_op2(ir, OPCODE_SLE, result_dst, op[0], op[1]); + break; + case ir_binop_gequal: + ir_to_mesa_emit_op2(ir, OPCODE_SGE, result_dst, op[0], op[1]); + break; + case ir_binop_equal: + ir_to_mesa_emit_op2(ir, OPCODE_SEQ, result_dst, op[0], op[1]); + break; + case ir_binop_nequal: + ir_to_mesa_emit_op2(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()) { + ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type); + ir_to_mesa_emit_op2(ir, OPCODE_SNE, + ir_to_mesa_dst_reg_from_src(temp), op[0], op[1]); + ir_to_mesa_emit_dp(ir, result_dst, temp, temp, vector_elements); + ir_to_mesa_emit_op2(ir, OPCODE_SEQ, + result_dst, result_src, src_reg_for_float(0.0)); + } else { + ir_to_mesa_emit_op2(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()) { + ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type); + ir_to_mesa_emit_op2(ir, OPCODE_SNE, + ir_to_mesa_dst_reg_from_src(temp), op[0], op[1]); + ir_to_mesa_emit_dp(ir, result_dst, temp, temp, vector_elements); + ir_to_mesa_emit_op2(ir, OPCODE_SNE, + result_dst, result_src, src_reg_for_float(0.0)); + } else { + ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]); + } + break; + + case ir_unop_any: + assert(ir->operands[0]->type->is_vector()); + ir_to_mesa_emit_dp(ir, result_dst, op[0], op[0], + ir->operands[0]->type->vector_elements); + ir_to_mesa_emit_op2(ir, OPCODE_SNE, + result_dst, result_src, src_reg_for_float(0.0)); + break; + + case ir_binop_logic_xor: + ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]); + break; + + case ir_binop_logic_or: + /* This could be a saturated add and skip the SNE. */ + ir_to_mesa_emit_op2(ir, OPCODE_ADD, + result_dst, + op[0], op[1]); + + ir_to_mesa_emit_op2(ir, OPCODE_SNE, + result_dst, + result_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". */ + ir_to_mesa_emit_op2(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); + ir_to_mesa_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). */ + ir_to_mesa_emit_scalar_op1(ir, OPCODE_RSQ, result_dst, op[0]); + ir_to_mesa_emit_op2(ir, OPCODE_MUL, result_dst, result_src, op[0]); + /* For incoming channels <= 0, set the result to 0. */ + op[0].negate = ~op[0].negate; + ir_to_mesa_emit_op3(ir, OPCODE_CMP, result_dst, + op[0], result_src, src_reg_for_float(0.0)); + break; + case ir_unop_rsq: + ir_to_mesa_emit_scalar_op1(ir, OPCODE_RSQ, result_dst, op[0]); + break; + case ir_unop_i2f: + case ir_unop_b2f: + case ir_unop_b2i: + /* Mesa IR lacks types, ints are stored as truncated floats. */ + result_src = op[0]; + break; + case ir_unop_f2i: + ir_to_mesa_emit_op1(ir, OPCODE_TRUNC, result_dst, op[0]); + break; + case ir_unop_f2b: + case ir_unop_i2b: + ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, + op[0], src_reg_for_float(0.0)); + break; + case ir_unop_trunc: + ir_to_mesa_emit_op1(ir, OPCODE_TRUNC, result_dst, op[0]); + break; + case ir_unop_ceil: + op[0].negate = ~op[0].negate; + ir_to_mesa_emit_op1(ir, OPCODE_FLR, result_dst, op[0]); + result_src.negate = ~result_src.negate; + break; + case ir_unop_floor: + ir_to_mesa_emit_op1(ir, OPCODE_FLR, result_dst, op[0]); + break; + case ir_unop_fract: + ir_to_mesa_emit_op1(ir, OPCODE_FRC, result_dst, op[0]); + break; + + case ir_binop_min: + ir_to_mesa_emit_op2(ir, OPCODE_MIN, result_dst, op[0], op[1]); + break; + case ir_binop_max: + ir_to_mesa_emit_op2(ir, OPCODE_MAX, result_dst, op[0], op[1]); + break; + case ir_binop_pow: + ir_to_mesa_emit_scalar_op2(ir, OPCODE_POW, result_dst, op[0], op[1]); + break; + + case ir_unop_bit_not: + case ir_unop_u2f: + 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) +{ + ir_to_mesa_src_reg src_reg; + 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_reg = this->result; + assert(src_reg.file != PROGRAM_UNDEFINED); + + for (i = 0; i < 4; i++) { + if (i < ir->type->vector_elements) { + switch (i) { + case 0: + swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.x); + break; + case 1: + swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.y); + break; + case 2: + swizzle[i] = GET_SWZ(src_reg.swizzle, ir->mask.z); + break; + case 3: + swizzle[i] = GET_SWZ(src_reg.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_reg.swizzle = MAKE_SWIZZLE4(swizzle[0], + swizzle[1], + swizzle[2], + swizzle[3]); + + this->result = src_reg; +} + +void +ir_to_mesa_visitor::visit(ir_dereference_variable *ir) +{ + variable_storage *entry = find_variable_storage(ir->var); + + if (!entry) { + switch (ir->var->mode) { + case ir_var_uniform: + entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_UNIFORM, + ir->var->location); + this->variables.push_tail(entry); + break; + case ir_var_in: + case ir_var_out: + case ir_var_inout: + case ir_var_system_value: + /* The linker assigns locations for varyings and attributes, + * including deprecated builtins (like gl_Color), user-assign + * generic attributes (glBindVertexLocation), and + * user-defined varyings. + * + * FINISHME: We would hit this path for function arguments. Fix! + */ + assert(ir->var->location != -1); + if (ir->var->mode == ir_var_in || + ir->var->mode == ir_var_inout) { + entry = new(mem_ctx) variable_storage(ir->var, + PROGRAM_INPUT, + ir->var->location); + + if (this->prog->Target == GL_VERTEX_PROGRAM_ARB && + ir->var->location >= VERT_ATTRIB_GENERIC0) { + _mesa_add_attribute(prog->Attributes, + ir->var->name, + _mesa_sizeof_glsl_type(ir->var->type->gl_type), + ir->var->type->gl_type, + ir->var->location - VERT_ATTRIB_GENERIC0); + } + } else if (ir->var->mode == ir_var_system_value) { + entry = new(mem_ctx) variable_storage(ir->var, + PROGRAM_SYSTEM_VALUE, + ir->var->location); + } else { + entry = new(mem_ctx) variable_storage(ir->var, + PROGRAM_OUTPUT, + ir->var->location); + } + + break; + case ir_var_auto: + case ir_var_temporary: + entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_TEMPORARY, + this->next_temp); + this->variables.push_tail(entry); + + next_temp += type_size(ir->var->type); + break; + } + + if (!entry) { + printf("Failed to make storage for %s\n", ir->var->name); + exit(1); + } + } + + this->result = ir_to_mesa_src_reg(entry->file, entry->index, ir->var->type); +} + +void +ir_to_mesa_visitor::visit(ir_dereference_array *ir) +{ + ir_constant *index; + ir_to_mesa_src_reg src_reg; + int element_size = type_size(ir->type); + + index = ir->array_index->constant_expression_value(); + + ir->array->accept(this); + src_reg = this->result; + + if (index) { + src_reg.index += index->value.i[0] * element_size; + } else { + ir_to_mesa_src_reg array_base = this->result; + /* 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); + + ir_to_mesa_src_reg index_reg; + + if (element_size == 1) { + index_reg = this->result; + } else { + index_reg = get_temp(glsl_type::float_type); + + ir_to_mesa_emit_op2(ir, OPCODE_MUL, + ir_to_mesa_dst_reg_from_src(index_reg), + this->result, src_reg_for_float(element_size)); + } + + src_reg.reladdr = talloc(mem_ctx, ir_to_mesa_src_reg); + memcpy(src_reg.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_reg.swizzle = swizzle_for_size(ir->type->vector_elements); + else + src_reg.swizzle = SWIZZLE_NOOP; + + this->result = src_reg; +} + +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 struct ir_to_mesa_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 ir_to_mesa_dst_reg_from_src(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) +{ + struct ir_to_mesa_dst_reg l; + struct ir_to_mesa_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); + ir_to_mesa_src_reg condition = this->result; + + for (i = 0; i < type_size(ir->lhs->type); i++) { + if (switch_order) { + ir_to_mesa_emit_op3(ir, OPCODE_CMP, l, + condition, ir_to_mesa_src_reg_from_dst(l), r); + } else { + ir_to_mesa_emit_op3(ir, OPCODE_CMP, l, + condition, r, ir_to_mesa_src_reg_from_dst(l)); + } + + l.index++; + r.index++; + } + } else { + for (i = 0; i < type_size(ir->lhs->type); i++) { + ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r); + l.index++; + r.index++; + } + } +} + + +void +ir_to_mesa_visitor::visit(ir_constant *ir) +{ + ir_to_mesa_src_reg src_reg; + 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) { + ir_to_mesa_src_reg temp_base = get_temp(ir->type); + ir_to_mesa_dst_reg temp = ir_to_mesa_dst_reg_from_src(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_reg = this->result; + + for (i = 0; i < (unsigned int)size; i++) { + ir_to_mesa_emit_op1(ir, OPCODE_MOV, temp, src_reg); + + src_reg.index++; + temp.index++; + } + } + this->result = temp_base; + return; + } + + if (ir->type->is_array()) { + ir_to_mesa_src_reg temp_base = get_temp(ir->type); + ir_to_mesa_dst_reg temp = ir_to_mesa_dst_reg_from_src(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_reg = this->result; + for (int j = 0; j < size; j++) { + ir_to_mesa_emit_op1(ir, OPCODE_MOV, temp, src_reg); + + src_reg.index++; + temp.index++; + } + } + this->result = temp_base; + return; + } + + if (ir->type->is_matrix()) { + ir_to_mesa_src_reg mat = get_temp(ir->type); + ir_to_mesa_dst_reg mat_column = ir_to_mesa_dst_reg_from_src(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_reg = ir_to_mesa_src_reg(PROGRAM_CONSTANT, -1, NULL); + src_reg.index = _mesa_add_unnamed_constant(this->prog->Parameters, + values, + ir->type->vector_elements, + &src_reg.swizzle); + ir_to_mesa_emit_op1(ir, OPCODE_MOV, mat_column, src_reg); + + mat_column.index++; + } + + this->result = mat; + return; + } + + src_reg.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 = ir_to_mesa_src_reg(PROGRAM_CONSTANT, -1, ir->type); + this->result.index = _mesa_add_unnamed_constant(this->prog->Parameters, + 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 = talloc(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 = ir_to_mesa_undef; + } + + 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); + ir_to_mesa_src_reg r = this->result; + + ir_to_mesa_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++) { + ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r); + l.index++; + r.index++; + } + } + + sig_iter.next(); + } + assert(!sig_iter.has_next()); + + /* Emit call instruction */ + call_inst = ir_to_mesa_emit_op1(ir, OPCODE_CAL, + ir_to_mesa_undef_dst, ir_to_mesa_undef); + 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); + + ir_to_mesa_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); + ir_to_mesa_dst_reg l = ir_to_mesa_dst_reg_from_src(this->result); + + for (i = 0; i < type_size(param->type); i++) { + ir_to_mesa_emit_op1(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) +{ + ir_to_mesa_src_reg result_src, coord, lod_info, projector; + ir_to_mesa_dst_reg result_dst, coord_dst; + ir_to_mesa_instruction *inst = NULL; + prog_opcode opcode = OPCODE_NOP; + + 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 = ir_to_mesa_dst_reg_from_src(coord); + ir_to_mesa_emit_op1(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 = ir_to_mesa_dst_reg_from_src(result_src); + + switch (ir->op) { + case ir_tex: + 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: + case ir_txf: + assert(!"GLSL 1.30 features unsupported"); + break; + } + + if (ir->projector) { + if (opcode == OPCODE_TEX) { + /* Slot the projector in as the last component of the coord. */ + coord_dst.writemask = WRITEMASK_W; + ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst, projector); + coord_dst.writemask = WRITEMASK_XYZW; + opcode = OPCODE_TXP; + } else { + ir_to_mesa_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; + ir_to_mesa_emit_op1(ir, OPCODE_RCP, coord_dst, projector); + + coord_dst.writemask = WRITEMASK_XYZ; + ir_to_mesa_emit_op2(ir, OPCODE_MUL, coord_dst, coord, coord_w); + + coord_dst.writemask = WRITEMASK_XYZW; + coord.swizzle = SWIZZLE_XYZW; + } + } + + if (ir->shadow_comparitor) { + /* Slot the shadow value in as the second to last component of the + * coord. + */ + ir->shadow_comparitor->accept(this); + coord_dst.writemask = WRITEMASK_Z; + ir_to_mesa_emit_op1(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; + ir_to_mesa_emit_op1(ir, OPCODE_MOV, coord_dst, lod_info); + coord_dst.writemask = WRITEMASK_XYZW; + } + + inst = ir_to_mesa_emit_op1(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); + + const glsl_type *sampler_type = ir->sampler->type; + + 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()) { + ir_to_mesa_dst_reg l; + int i; + + assert(current_function); + + ir->get_value()->accept(this); + ir_to_mesa_src_reg r = this->result; + + l = ir_to_mesa_dst_reg_from_src(current_function->return_reg); + + for (i = 0; i < type_size(current_function->sig->return_type); i++) { + ir_to_mesa_emit_op1(ir, OPCODE_MOV, l, r); + l.index++; + r.index++; + } + } + + ir_to_mesa_emit_op0(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; + ir_to_mesa_emit_op1(ir, OPCODE_KIL, ir_to_mesa_undef_dst, this->result); + } else { + ir_to_mesa_emit_op0(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, *else_inst = NULL; + 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) { + ir_to_mesa_src_reg temp = get_temp(glsl_type::bool_type); + cond_inst = ir_to_mesa_emit_op1(ir->condition, OPCODE_MOV, + ir_to_mesa_dst_reg_from_src(temp), + result); + } + cond_inst->cond_update = GL_TRUE; + + if_inst = ir_to_mesa_emit_op0(ir->condition, OPCODE_IF); + if_inst->dst_reg.cond_mask = COND_NE; + } else { + if_inst = ir_to_mesa_emit_op1(ir->condition, + OPCODE_IF, ir_to_mesa_undef_dst, + this->result); + } + + this->instructions.push_tail(if_inst); + + visit_exec_list(&ir->then_instructions, this); + + if (!ir->else_instructions.is_empty()) { + else_inst = ir_to_mesa_emit_op0(ir->condition, OPCODE_ELSE); + visit_exec_list(&ir->else_instructions, this); + } + + if_inst = ir_to_mesa_emit_op1(ir->condition, OPCODE_ENDIF, + ir_to_mesa_undef_dst, ir_to_mesa_undef); +} + +ir_to_mesa_visitor::ir_to_mesa_visitor() +{ + result.file = PROGRAM_UNDEFINED; + next_temp = 1; + next_signature_id = 1; + current_function = NULL; + mem_ctx = talloc_new(NULL); +} + +ir_to_mesa_visitor::~ir_to_mesa_visitor() +{ + talloc_free(mem_ctx); +} + +static struct prog_src_register +mesa_src_reg_from_ir_src_reg(ir_to_mesa_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 = talloc_zero_array(v->mem_ctx, int, if_count); + loop_stack = talloc_zero_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); + } +} + +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); +} + +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 = talloc_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] = 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) { + fail_link(shader_program, "Allocation of uniform `%s' to target " + "failed (%d vs %d)\n", + uniform->Name, index, parameter_index); + } + } + } + + talloc_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 = talloc_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) { + fail_link(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 = talloc_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 = talloc_new(NULL); + + set_uniform_initializer(ctx, mem_ctx, shader_program, var->name, + var->type, var->constant_value); + } + } + + talloc_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 = talloc_zero_array(mem_ctx, + ir_to_mesa_instruction *, + this->next_temp * 4); + + foreach_iter(exec_list_iterator, iter, this->instructions) { + ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get(); + + assert(inst->dst_reg.file != PROGRAM_TEMPORARY + || inst->dst_reg.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_reg[r].index * 4; + + if (inst->src_reg[r].file != PROGRAM_TEMPORARY || + inst->src_reg[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_reg[r].swizzle, i); + ir_to_mesa_instruction *copy_chan = acp[acp_base + src_chan]; + + if (!copy_chan) { + good = false; + break; + } + + if (!first) { + first = copy_chan; + } else { + if (first->src_reg[0].file != copy_chan->src_reg[0].file || + first->src_reg[0].index != copy_chan->src_reg[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_reg[r].file = first->src_reg[0].file; + inst->src_reg[r].index = first->src_reg[0].index; + + int swizzle = 0; + for (int i = 0; i < 4; i++) { + int src_chan = GET_SWZ(inst->src_reg[r].swizzle, i); + ir_to_mesa_instruction *copy_inst = acp[acp_base + src_chan]; + swizzle |= (GET_SWZ(copy_inst->src_reg[0].swizzle, src_chan) << + (3 * i)); + } + inst->src_reg[r].swizzle = swizzle; + } + } + + switch (inst->op) { + case OPCODE_BGNLOOP: + case OPCODE_ENDLOOP: + case OPCODE_ELSE: + case OPCODE_ENDIF: + /* End of a basic block, clear the ACP entirely. */ + memset(acp, 0, sizeof(*acp) * this->next_temp * 4); + break; + + default: + /* Continuing the block, clear any written channels from + * the ACP. + */ + if (inst->dst_reg.file == PROGRAM_TEMPORARY) { + if (inst->dst_reg.reladdr) { + memset(acp, 0, sizeof(*acp) * this->next_temp * 4); + } else { + for (int i = 0; i < 4; i++) { + if (inst->dst_reg.writemask & (1 << i)) { + acp[4 * inst->dst_reg.index + i] = NULL; + } + } + } + } + break; + } + + /* If this is a copy, add it to the ACP. */ + if (inst->op == OPCODE_MOV && + inst->dst_reg.file == PROGRAM_TEMPORARY && + !inst->dst_reg.reladdr && + !inst->saturate && + !inst->src_reg[0].reladdr && + !inst->src_reg[0].negate) { + for (int i = 0; i < 4; i++) { + if (inst->dst_reg.writemask & (1 << i)) { + acp[4 * inst->dst_reg.index + i] = inst; + } + } + } + } + + talloc_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.ir_to_mesa_emit_op0(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.ir_to_mesa_emit_op0(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.ir_to_mesa_emit_op0(NULL, OPCODE_RET); + + ir_to_mesa_instruction *end; + end = v.ir_to_mesa_emit_op0(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 = talloc_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_reg.file; + mesa_inst->DstReg.Index = inst->dst_reg.index; + mesa_inst->DstReg.CondMask = inst->dst_reg.cond_mask; + mesa_inst->DstReg.WriteMask = inst->dst_reg.writemask; + mesa_inst->DstReg.RelAddr = inst->dst_reg.reladdr != NULL; + mesa_inst->SrcReg[0] = mesa_src_reg_from_ir_src_reg(inst->src_reg[0]); + mesa_inst->SrcReg[1] = mesa_src_reg_from_ir_src_reg(inst->src_reg[1]); + mesa_inst->SrcReg[2] = mesa_src_reg_from_ir_src_reg(inst->src_reg[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; + + if (options->EmitNoIfs && mesa_inst->Opcode == OPCODE_IF) { + fail_link(shader_program, "Couldn't flatten if statement\n"); + } + + switch (mesa_inst->Opcode) { + 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); + + _mesa_reference_program(ctx, &shader->Program, prog); + + if ((ctx->Shader.Flags & GLSL_NO_OPT) == 0) { + _mesa_optimize_program(ctx, prog); + } + + return prog; +} + +extern "C" { + +/** + * Called via ctx->Driver.CompilerShader(). + * This is a no-op. + * XXX can we remove the ctx->Driver.CompileShader() hook? + */ +GLboolean +_mesa_ir_compile_shader(struct gl_context *ctx, struct gl_shader *shader) +{ + assert(shader->CompileStatus); + (void) ctx; + + return GL_TRUE; +} + + +/** + * 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 + | ((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->EmitNoIfs) { + progress = lower_discard(ir) || progress; + progress = lower_if_to_cond_assign(ir) || 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 shader %d:\n", 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); + } + + talloc_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); + + talloc_free(state); + + if (shader->CompileStatus) { + if (!ctx->Driver.CompileShader(ctx, shader)) + shader->CompileStatus = GL_FALSE; + } +} + + +/** + * 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) { + fail_link(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/state_tracker/st_cb_texture.c b/mesalib/src/mesa/state_tracker/st_cb_texture.c index b60ac72e4..9653dcd65 100644 --- a/mesalib/src/mesa/state_tracker/st_cb_texture.c +++ b/mesalib/src/mesa/state_tracker/st_cb_texture.c @@ -1,2022 +1,2021 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * 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, sub license, 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 NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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/mfeatures.h" -#include "main/bufferobj.h" -#include "main/enums.h" -#include "main/fbobject.h" -#include "main/formats.h" -#include "main/image.h" -#include "main/imports.h" -#include "main/macros.h" -#include "main/mipmap.h" -#include "main/pack.h" -#include "main/pixeltransfer.h" -#include "main/texcompress.h" -#include "main/texfetch.h" -#include "main/texgetimage.h" -#include "main/teximage.h" -#include "main/texobj.h" -#include "main/texstore.h" - -#include "state_tracker/st_debug.h" -#include "state_tracker/st_context.h" -#include "state_tracker/st_cb_fbo.h" -#include "state_tracker/st_cb_flush.h" -#include "state_tracker/st_cb_texture.h" -#include "state_tracker/st_format.h" -#include "state_tracker/st_texture.h" -#include "state_tracker/st_gen_mipmap.h" -#include "state_tracker/st_atom.h" - -#include "pipe/p_context.h" -#include "pipe/p_defines.h" -#include "util/u_inlines.h" -#include "pipe/p_shader_tokens.h" -#include "util/u_tile.h" -#include "util/u_blit.h" -#include "util/u_format.h" -#include "util/u_surface.h" -#include "util/u_sampler.h" -#include "util/u_math.h" -#include "util/u_box.h" - -#define DBG if (0) printf - - -static enum pipe_texture_target -gl_target_to_pipe(GLenum target) -{ - switch (target) { - case GL_TEXTURE_1D: - return PIPE_TEXTURE_1D; - - case GL_TEXTURE_2D: - return PIPE_TEXTURE_2D; - - case GL_TEXTURE_RECTANGLE_NV: - return PIPE_TEXTURE_RECT; - - case GL_TEXTURE_3D: - return PIPE_TEXTURE_3D; - - case GL_TEXTURE_CUBE_MAP_ARB: - return PIPE_TEXTURE_CUBE; - - default: - assert(0); - return 0; - } -} - - -/** called via ctx->Driver.NewTextureImage() */ -static struct gl_texture_image * -st_NewTextureImage(struct gl_context * ctx) -{ - DBG("%s\n", __FUNCTION__); - (void) ctx; - return (struct gl_texture_image *) ST_CALLOC_STRUCT(st_texture_image); -} - - -/** called via ctx->Driver.NewTextureObject() */ -static struct gl_texture_object * -st_NewTextureObject(struct gl_context * ctx, GLuint name, GLenum target) -{ - struct st_texture_object *obj = ST_CALLOC_STRUCT(st_texture_object); - - DBG("%s\n", __FUNCTION__); - _mesa_initialize_texture_object(&obj->base, name, target); - - return &obj->base; -} - -/** called via ctx->Driver.DeleteTextureObject() */ -static void -st_DeleteTextureObject(struct gl_context *ctx, - struct gl_texture_object *texObj) -{ - struct st_context *st = st_context(ctx); - struct st_texture_object *stObj = st_texture_object(texObj); - if (stObj->pt) - pipe_resource_reference(&stObj->pt, NULL); - if (stObj->sampler_view) { - if (stObj->sampler_view->context != st->pipe) { - /* Take "ownership" of this texture sampler view by setting - * its context pointer to this context. This avoids potential - * crashes when the texture object is shared among contexts - * and the original/owner context has already been destroyed. - */ - stObj->sampler_view->context = st->pipe; - } - pipe_sampler_view_reference(&stObj->sampler_view, NULL); - } - _mesa_delete_texture_object(ctx, texObj); -} - - -/** called via ctx->Driver.FreeTexImageData() */ -static void -st_FreeTextureImageData(struct gl_context * ctx, struct gl_texture_image *texImage) -{ - struct st_texture_image *stImage = st_texture_image(texImage); - - DBG("%s\n", __FUNCTION__); - - if (stImage->pt) { - pipe_resource_reference(&stImage->pt, NULL); - } - - if (texImage->Data) { - _mesa_align_free(texImage->Data); - texImage->Data = NULL; - } -} - - -/** - * From linux kernel i386 header files, copes with odd sizes better - * than COPY_DWORDS would: - * XXX Put this in src/mesa/main/imports.h ??? - */ -#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86) -static INLINE void * -__memcpy(void *to, const void *from, size_t n) -{ - int d0, d1, d2; - __asm__ __volatile__("rep ; movsl\n\t" - "testb $2,%b4\n\t" - "je 1f\n\t" - "movsw\n" - "1:\ttestb $1,%b4\n\t" - "je 2f\n\t" - "movsb\n" "2:":"=&c"(d0), "=&D"(d1), "=&S"(d2) - :"0"(n / 4), "q"(n), "1"((long) to), "2"((long) from) - :"memory"); - return (to); -} -#else -#define __memcpy(a,b,c) memcpy(a,b,c) -#endif - - -/** - * The system memcpy (at least on ubuntu 5.10) has problems copying - * to agp (writecombined) memory from a source which isn't 64-byte - * aligned - there is a 4x performance falloff. - * - * The x86 __memcpy is immune to this but is slightly slower - * (10%-ish) than the system memcpy. - * - * The sse_memcpy seems to have a slight cliff at 64/32 bytes, but - * isn't much faster than x86_memcpy for agp copies. - * - * TODO: switch dynamically. - */ -static void * -do_memcpy(void *dest, const void *src, size_t n) -{ - if ((((unsigned long) src) & 63) || (((unsigned long) dest) & 63)) { - return __memcpy(dest, src, n); - } - else - return memcpy(dest, src, n); -} - - -/** - * Return default texture resource binding bitmask for the given format. - */ -static GLuint -default_bindings(struct st_context *st, enum pipe_format format) -{ - struct pipe_screen *screen = st->pipe->screen; - const unsigned target = PIPE_TEXTURE_2D; - const unsigned geom = 0x0; - unsigned bindings; - - if (util_format_is_depth_or_stencil(format)) - bindings = PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_DEPTH_STENCIL; - else - bindings = PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET; - - if (screen->is_format_supported(screen, format, target, 0, bindings, geom)) - return bindings; - else - return PIPE_BIND_SAMPLER_VIEW; -} - - -/** Return number of image dimensions (1, 2 or 3) for a texture target. */ -static GLuint -get_texture_dims(GLenum target) -{ - switch (target) { - case GL_TEXTURE_1D: - case GL_TEXTURE_1D_ARRAY_EXT: - return 1; - case GL_TEXTURE_2D: - case GL_TEXTURE_CUBE_MAP_ARB: - case GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_X_ARB: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Y_ARB: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y_ARB: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Z_ARB: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z_ARB: - case GL_TEXTURE_RECTANGLE_NV: - case GL_TEXTURE_2D_ARRAY_EXT: - return 2; - case GL_TEXTURE_3D: - return 3; - default: - assert(0 && "invalid texture target in get_texture_dims()"); - return 1; - } -} - - -/** - * Try to allocate a pipe_resource object for the given st_texture_object. - * - * We use the given st_texture_image as a clue to determine the size of the - * mipmap image at level=0. - * - * \return GL_TRUE for success, GL_FALSE if out of memory. - */ -static GLboolean -guess_and_alloc_texture(struct st_context *st, - struct st_texture_object *stObj, - const struct st_texture_image *stImage) -{ - const GLuint dims = get_texture_dims(stObj->base.Target); - GLuint level, lastLevel, width, height, depth; - GLuint bindings; - enum pipe_format fmt; - - DBG("%s\n", __FUNCTION__); - - assert(!stObj->pt); - - level = stImage->level; - width = stImage->base.Width2; /* size w/out border */ - height = stImage->base.Height2; - depth = stImage->base.Depth2; - - assert(width > 0); - assert(height > 0); - assert(depth > 0); - - /* Depending on the image's size, we can't always make a guess here. - */ - if (level > 0) { - if ( (dims >= 1 && width == 1) || - (dims >= 2 && height == 1) || - (dims >= 3 && depth == 1) ) { - /* we can't determine the image size at level=0 */ - stObj->width0 = stObj->height0 = stObj->depth0 = 0; - /* this is not an out of memory error */ - return GL_TRUE; - } - } - - /* grow the image size until we hit level = 0 */ - while (level > 0) { - if (width != 1) - width <<= 1; - if (height != 1) - height <<= 1; - if (depth != 1) - depth <<= 1; - level--; - } - - assert(level == 0); - - /* At this point, (width x height x depth) is the expected size of - * the level=0 mipmap image. - */ - - /* Guess a reasonable value for lastLevel. With OpenGL we have no - * idea how many mipmap levels will be in a texture until we start - * to render with it. Make an educated guess here but be prepared - * to re-allocating a texture buffer with space for more (or fewer) - * mipmap levels later. - */ - if ((stObj->base.MinFilter == GL_NEAREST || - stObj->base.MinFilter == GL_LINEAR || - stImage->base._BaseFormat == GL_DEPTH_COMPONENT || - stImage->base._BaseFormat == GL_DEPTH_STENCIL_EXT) && - !stObj->base.GenerateMipmap && - stImage->level == 0) { - /* only alloc space for a single mipmap level */ - lastLevel = 0; - } - else { - /* alloc space for a full mipmap */ - GLuint l2width = util_logbase2(width); - GLuint l2height = util_logbase2(height); - GLuint l2depth = util_logbase2(depth); - lastLevel = MAX2(MAX2(l2width, l2height), l2depth); - } - - /* Save the level=0 dimensions */ - stObj->width0 = width; - stObj->height0 = height; - stObj->depth0 = depth; - - fmt = st_mesa_format_to_pipe_format(stImage->base.TexFormat); - - bindings = default_bindings(st, fmt); - - stObj->pt = st_texture_create(st, - gl_target_to_pipe(stObj->base.Target), - fmt, - lastLevel, - width, - height, - depth, - bindings); - - DBG("%s returning %d\n", __FUNCTION__, (stObj->pt != NULL)); - - return stObj->pt != NULL; -} - - -/** - * Adjust pixel unpack params and image dimensions to strip off the - * texture border. - * Gallium doesn't support texture borders. They've seldem been used - * and seldom been implemented correctly anyway. - * \param unpackNew returns the new pixel unpack parameters - */ -static void -strip_texture_border(GLint border, - GLint *width, GLint *height, GLint *depth, - const struct gl_pixelstore_attrib *unpack, - struct gl_pixelstore_attrib *unpackNew) -{ - assert(border > 0); /* sanity check */ - - *unpackNew = *unpack; - - if (unpackNew->RowLength == 0) - unpackNew->RowLength = *width; - - if (depth && unpackNew->ImageHeight == 0) - unpackNew->ImageHeight = *height; - - unpackNew->SkipPixels += border; - if (height) - unpackNew->SkipRows += border; - if (depth) - unpackNew->SkipImages += border; - - assert(*width >= 3); - *width = *width - 2 * border; - if (height && *height >= 3) - *height = *height - 2 * border; - if (depth && *depth >= 3) - *depth = *depth - 2 * border; -} - - -/** - * Try to do texture compression via rendering. If the Gallium driver - * can render into a compressed surface this will allow us to do texture - * compression. - * \return GL_TRUE for success, GL_FALSE for failure - */ -static GLboolean -compress_with_blit(struct gl_context * ctx, - GLenum target, GLint level, - GLint xoffset, GLint yoffset, GLint zoffset, - GLint width, GLint height, GLint depth, - GLenum format, GLenum type, const void *pixels, - const struct gl_pixelstore_attrib *unpack, - struct gl_texture_image *texImage) -{ - const GLuint dstImageOffsets[1] = {0}; - struct st_texture_image *stImage = st_texture_image(texImage); - struct st_context *st = st_context(ctx); - struct pipe_context *pipe = st->pipe; - struct pipe_screen *screen = pipe->screen; - gl_format mesa_format; - struct pipe_resource templ; - struct pipe_resource *src_tex; - struct pipe_sampler_view view_templ; - struct pipe_sampler_view *src_view; - struct pipe_surface *dst_surface, surf_tmpl; - struct pipe_transfer *tex_xfer; - void *map; - - if (!stImage->pt) { - /* XXX: Can this happen? Should we assert? */ - return GL_FALSE; - } - - /* get destination surface (in the compressed texture) */ - memset(&surf_tmpl, 0, sizeof(surf_tmpl)); - surf_tmpl.format = stImage->pt->format; - surf_tmpl.usage = PIPE_BIND_RENDER_TARGET; - surf_tmpl.u.tex.level = stImage->level; - surf_tmpl.u.tex.first_layer = stImage->face; - surf_tmpl.u.tex.last_layer = stImage->face; - dst_surface = pipe->create_surface(pipe, stImage->pt, &surf_tmpl); - if (!dst_surface) { - /* can't render into this format (or other problem) */ - return GL_FALSE; - } - - /* Choose format for the temporary RGBA texture image. - */ - mesa_format = st_ChooseTextureFormat(ctx, GL_RGBA, format, type); - assert(mesa_format); - if (!mesa_format) - return GL_FALSE; - - /* Create the temporary source texture - */ - memset(&templ, 0, sizeof(templ)); - templ.target = st->internal_target; - templ.format = st_mesa_format_to_pipe_format(mesa_format); - templ.width0 = width; - templ.height0 = height; - templ.depth0 = 1; - templ.array_size = 1; - templ.last_level = 0; - templ.usage = PIPE_USAGE_DEFAULT; - templ.bind = PIPE_BIND_SAMPLER_VIEW; - src_tex = screen->resource_create(screen, &templ); - - if (!src_tex) - return GL_FALSE; - - /* Put user's tex data into the temporary texture - */ - tex_xfer = pipe_get_transfer(st_context(ctx)->pipe, src_tex, - 0, 0, /* layer, level are zero */ - PIPE_TRANSFER_WRITE, - 0, 0, width, height); /* x, y, w, h */ - map = pipe_transfer_map(pipe, tex_xfer); - - _mesa_texstore(ctx, 2, GL_RGBA, mesa_format, - map, /* dest ptr */ - 0, 0, 0, /* dest x/y/z offset */ - tex_xfer->stride, /* dest row stride (bytes) */ - dstImageOffsets, /* image offsets (for 3D only) */ - width, height, 1, /* size */ - format, type, /* source format/type */ - pixels, /* source data */ - unpack); /* source data packing */ - - pipe_transfer_unmap(pipe, tex_xfer); - pipe->transfer_destroy(pipe, tex_xfer); - - /* Create temporary sampler view */ - u_sampler_view_default_template(&view_templ, - src_tex, - src_tex->format); - src_view = pipe->create_sampler_view(pipe, src_tex, &view_templ); - - - /* copy / compress image */ - util_blit_pixels_tex(st->blit, - src_view, /* sampler view (src) */ - 0, 0, /* src x0, y0 */ - width, height, /* src x1, y1 */ - dst_surface, /* pipe_surface (dst) */ - xoffset, yoffset, /* dst x0, y0 */ - xoffset + width, /* dst x1 */ - yoffset + height, /* dst y1 */ - 0.0, /* z */ - PIPE_TEX_MIPFILTER_NEAREST); - - pipe_surface_reference(&dst_surface, NULL); - pipe_resource_reference(&src_tex, NULL); - pipe_sampler_view_reference(&src_view, NULL); - - return GL_TRUE; -} - - -/** - * Do glTexImage1/2/3D(). - */ -static void -st_TexImage(struct gl_context * ctx, - GLint dims, - GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint depth, - GLint border, - GLenum format, GLenum type, const void *pixels, - const struct gl_pixelstore_attrib *unpack, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage, - GLsizei imageSize, GLboolean compressed_src) -{ - struct st_context *st = st_context(ctx); - struct pipe_screen *screen = st->pipe->screen; - struct st_texture_object *stObj = st_texture_object(texObj); - struct st_texture_image *stImage = st_texture_image(texImage); - GLuint dstRowStride = 0; - struct gl_pixelstore_attrib unpackNB; - enum pipe_transfer_usage transfer_usage = 0; - - DBG("%s target %s level %d %dx%dx%d border %d\n", __FUNCTION__, - _mesa_lookup_enum_by_nr(target), level, width, height, depth, border); - - /* switch to "normal" */ - if (stObj->surface_based) { - gl_format texFormat; - - _mesa_clear_texture_object(ctx, texObj); - pipe_resource_reference(&stObj->pt, NULL); - - /* oops, need to init this image again */ - texFormat = _mesa_choose_texture_format(ctx, texObj, target, level, - internalFormat, format, type); - - _mesa_init_teximage_fields(ctx, target, texImage, - width, height, depth, border, - internalFormat, texFormat); - - stObj->surface_based = GL_FALSE; - } - - /* gallium does not support texture borders, strip it off */ - if (border) { - strip_texture_border(border, &width, &height, &depth, unpack, &unpackNB); - unpack = &unpackNB; - texImage->Width = width; - texImage->Height = height; - texImage->Depth = depth; - texImage->Border = 0; - border = 0; - } - else { - assert(texImage->Width == width); - assert(texImage->Height == height); - assert(texImage->Depth == depth); - } - - stImage->face = _mesa_tex_target_to_face(target); - stImage->level = level; - - _mesa_set_fetch_functions(texImage, dims); - - /* Release the reference to a potentially orphaned buffer. - * Release any old malloced memory. - */ - if (stImage->pt) { - pipe_resource_reference(&stImage->pt, NULL); - assert(!texImage->Data); - } - else if (texImage->Data) { - _mesa_align_free(texImage->Data); - } - - /* - * See if the new image is somehow incompatible with the existing - * mipmap. If so, free the old mipmap. - */ - if (stObj->pt) { - if (level > (GLint) stObj->pt->last_level || - !st_texture_match_image(stObj->pt, &stImage->base, - stImage->face, stImage->level)) { - DBG("release it\n"); - pipe_resource_reference(&stObj->pt, NULL); - assert(!stObj->pt); - pipe_sampler_view_reference(&stObj->sampler_view, NULL); - } - } - - if (width == 0 || height == 0 || depth == 0) { - /* stop after freeing old image */ - return; - } - - if (!stObj->pt) { - if (!guess_and_alloc_texture(st, stObj, stImage)) { - /* Probably out of memory. - * Try flushing any pending rendering, then retry. - */ - st_finish(st); - if (!guess_and_alloc_texture(st, stObj, stImage)) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); - return; - } - } - } - - assert(!stImage->pt); - - /* Check if this texture image can live inside the texture object's buffer. - * If so, store the image there. Otherwise the image will temporarily live - * in its own buffer. - */ - if (stObj->pt && - st_texture_match_image(stObj->pt, &stImage->base, - stImage->face, stImage->level)) { - - pipe_resource_reference(&stImage->pt, stObj->pt); - assert(stImage->pt); - } - - if (!stImage->pt) - DBG("XXX: Image did not fit into texture - storing in local memory!\n"); - - /* Pixel data may come from regular user memory or a PBO. For the later, - * do bounds checking and map the PBO to read pixels data from it. - * - * XXX we should try to use a GPU-accelerated path to copy the image data - * from the PBO to the texture. - */ - if (compressed_src) { - pixels = _mesa_validate_pbo_compressed_teximage(ctx, imageSize, pixels, - unpack, - "glCompressedTexImage"); - } - else { - pixels = _mesa_validate_pbo_teximage(ctx, dims, width, height, 1, - format, type, - pixels, unpack, "glTexImage"); - } - - /* See if we can do texture compression with a blit/render. - */ - if (!compressed_src && - !ctx->Mesa_DXTn && - _mesa_is_format_compressed(texImage->TexFormat) && - screen->is_format_supported(screen, - stImage->pt->format, - stImage->pt->target, 0, - PIPE_BIND_RENDER_TARGET, 0)) { - if (!pixels) - goto done; - - if (compress_with_blit(ctx, target, level, 0, 0, 0, width, height, depth, - format, type, pixels, unpack, texImage)) { - goto done; - } - } - - /* - * Prepare to store the texture data. Either map the gallium texture buffer - * memory or malloc space for it. - */ - if (stImage->pt) { - /* Store the image in the gallium texture memory buffer */ - if (format == GL_DEPTH_COMPONENT && - util_format_is_depth_and_stencil(stImage->pt->format)) - transfer_usage = PIPE_TRANSFER_READ_WRITE; - else - transfer_usage = PIPE_TRANSFER_WRITE; - - texImage->Data = st_texture_image_map(st, stImage, 0, - transfer_usage, 0, 0, width, height); - if(stImage->transfer) - dstRowStride = stImage->transfer->stride; - } - else { - /* Allocate regular memory and store the image there temporarily. */ - GLuint imageSize = _mesa_format_image_size(texImage->TexFormat, - width, height, depth); - dstRowStride = _mesa_format_row_stride(texImage->TexFormat, width); - - texImage->Data = _mesa_align_malloc(imageSize, 16); - } - - if (!texImage->Data) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); - return; - } - - if (!pixels) { - /* We've allocated texture memory, but have no pixel data - all done. */ - goto done; - } - - DBG("Upload image %dx%dx%d row_len %x pitch %x\n", - width, height, depth, width, dstRowStride); - - /* Copy user texture image into the texture buffer. - */ - if (compressed_src) { - const GLuint srcRowStride = - _mesa_format_row_stride(texImage->TexFormat, width); - if (dstRowStride == srcRowStride) { - memcpy(texImage->Data, pixels, imageSize); - } - else { - char *dst = texImage->Data; - const char *src = pixels; - GLuint i, bw, bh, lines; - _mesa_get_format_block_size(texImage->TexFormat, &bw, &bh); - lines = (height + bh - 1) / bh; - - for (i = 0; i < lines; ++i) { - memcpy(dst, src, srcRowStride); - dst += dstRowStride; - src += srcRowStride; - } - } - } - else { - const GLuint srcImageStride = - _mesa_image_image_stride(unpack, width, height, format, type); - GLint i; - const GLubyte *src = (const GLubyte *) pixels; - - for (i = 0; i < depth; i++) { - if (!_mesa_texstore(ctx, dims, - texImage->_BaseFormat, - texImage->TexFormat, - texImage->Data, - 0, 0, 0, /* dstX/Y/Zoffset */ - dstRowStride, - texImage->ImageOffsets, - width, height, 1, - format, type, src, unpack)) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); - } - - if (stImage->pt && i + 1 < depth) { - /* unmap this slice */ - st_texture_image_unmap(st, stImage); - /* map next slice of 3D texture */ - texImage->Data = st_texture_image_map(st, stImage, i + 1, - transfer_usage, 0, 0, - width, height); - src += srcImageStride; - } - } - } - -done: - _mesa_unmap_teximage_pbo(ctx, unpack); - - if (stImage->pt && texImage->Data) { - st_texture_image_unmap(st, stImage); - texImage->Data = NULL; - } -} - - -static void -st_TexImage3D(struct gl_context * ctx, - GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint depth, - GLint border, - GLenum format, GLenum type, const void *pixels, - const struct gl_pixelstore_attrib *unpack, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - st_TexImage(ctx, 3, target, level, internalFormat, width, height, depth, - border, format, type, pixels, unpack, texObj, texImage, - 0, GL_FALSE); -} - - -static void -st_TexImage2D(struct gl_context * ctx, - GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint border, - GLenum format, GLenum type, const void *pixels, - const struct gl_pixelstore_attrib *unpack, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - st_TexImage(ctx, 2, target, level, internalFormat, width, height, 1, border, - format, type, pixels, unpack, texObj, texImage, 0, GL_FALSE); -} - - -static void -st_TexImage1D(struct gl_context * ctx, - GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint border, - GLenum format, GLenum type, const void *pixels, - const struct gl_pixelstore_attrib *unpack, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - st_TexImage(ctx, 1, target, level, internalFormat, width, 1, 1, border, - format, type, pixels, unpack, texObj, texImage, 0, GL_FALSE); -} - - -static void -st_CompressedTexImage2D(struct gl_context *ctx, GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint border, - GLsizei imageSize, const GLvoid *data, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - st_TexImage(ctx, 2, target, level, internalFormat, width, height, 1, border, - 0, 0, data, &ctx->Unpack, texObj, texImage, imageSize, GL_TRUE); -} - - - -/** - * glGetTexImage() helper: decompress a compressed texture by rendering - * a textured quad. Store the results in the user's buffer. - */ -static void -decompress_with_blit(struct gl_context * ctx, GLenum target, GLint level, - GLenum format, GLenum type, GLvoid *pixels, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - struct st_context *st = st_context(ctx); - struct pipe_context *pipe = st->pipe; - struct st_texture_image *stImage = st_texture_image(texImage); - struct st_texture_object *stObj = st_texture_object(texObj); - struct pipe_sampler_view *src_view = - st_get_texture_sampler_view(stObj, pipe); - const GLuint width = texImage->Width; - const GLuint height = texImage->Height; - struct pipe_surface *dst_surface; - struct pipe_resource *dst_texture; - struct pipe_transfer *tex_xfer; - unsigned bind = (PIPE_BIND_RENDER_TARGET | /* util_blit may choose to render */ - PIPE_BIND_TRANSFER_READ); - - /* create temp / dest surface */ - if (!util_create_rgba_surface(pipe, width, height, bind, - &dst_texture, &dst_surface)) { - _mesa_problem(ctx, "util_create_rgba_surface() failed " - "in decompress_with_blit()"); - return; - } - - /* blit/render/decompress */ - util_blit_pixels_tex(st->blit, - src_view, /* pipe_resource (src) */ - 0, 0, /* src x0, y0 */ - width, height, /* src x1, y1 */ - dst_surface, /* pipe_surface (dst) */ - 0, 0, /* dst x0, y0 */ - width, height, /* dst x1, y1 */ - 0.0, /* z */ - PIPE_TEX_MIPFILTER_NEAREST); - - /* map the dst_surface so we can read from it */ - tex_xfer = pipe_get_transfer(st_context(ctx)->pipe, - dst_texture, 0, 0, - PIPE_TRANSFER_READ, - 0, 0, width, height); - - pixels = _mesa_map_pbo_dest(ctx, &ctx->Pack, pixels); - - /* copy/pack data into user buffer */ - if (st_equal_formats(stImage->pt->format, format, type)) { - /* memcpy */ - const uint bytesPerRow = width * util_format_get_blocksize(stImage->pt->format); - ubyte *map = pipe_transfer_map(pipe, tex_xfer); - GLuint row; - for (row = 0; row < height; row++) { - GLvoid *dest = _mesa_image_address2d(&ctx->Pack, pixels, width, - height, format, type, row, 0); - memcpy(dest, map, bytesPerRow); - map += tex_xfer->stride; - } - pipe_transfer_unmap(pipe, tex_xfer); - } - else { - /* format translation via floats */ - GLuint row; - enum pipe_format format = util_format_linear(dst_texture->format); - for (row = 0; row < height; row++) { - const GLbitfield transferOps = 0x0; /* bypassed for glGetTexImage() */ - GLfloat rgba[4 * MAX_WIDTH]; - GLvoid *dest = _mesa_image_address2d(&ctx->Pack, pixels, width, - height, format, type, row, 0); - - if (ST_DEBUG & DEBUG_FALLBACK) - debug_printf("%s: fallback format translation\n", __FUNCTION__); - - /* get float[4] rgba row from surface */ - pipe_get_tile_rgba_format(pipe, tex_xfer, 0, row, width, 1, - format, rgba); - - _mesa_pack_rgba_span_float(ctx, width, (GLfloat (*)[4]) rgba, format, - type, dest, &ctx->Pack, transferOps); - } - } - - _mesa_unmap_pbo_dest(ctx, &ctx->Pack); - - pipe->transfer_destroy(pipe, tex_xfer); - - /* destroy the temp / dest surface */ - util_destroy_rgba_surface(dst_texture, dst_surface); -} - - - -/** - * Need to map texture image into memory before copying image data, - * then unmap it. - */ -static void -st_get_tex_image(struct gl_context * ctx, GLenum target, GLint level, - GLenum format, GLenum type, GLvoid * pixels, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage, GLboolean compressed_dst) -{ - struct st_context *st = st_context(ctx); - struct st_texture_image *stImage = st_texture_image(texImage); - const GLuint dstImageStride = - _mesa_image_image_stride(&ctx->Pack, texImage->Width, texImage->Height, - format, type); - GLuint depth, i; - GLubyte *dest; - - if (stImage->pt && - util_format_is_s3tc(stImage->pt->format) && - !compressed_dst) { - /* Need to decompress the texture. - * We'll do this by rendering a textured quad. - * Note that we only expect RGBA formats (no Z/depth formats). - */ - decompress_with_blit(ctx, target, level, format, type, pixels, - texObj, texImage); - return; - } - - /* Map */ - if (stImage->pt) { - /* Image is stored in hardware format in a buffer managed by the - * kernel. Need to explicitly map and unmap it. - */ - texImage->Data = st_texture_image_map(st, stImage, 0, - PIPE_TRANSFER_READ, 0, 0, - stImage->base.Width, - stImage->base.Height); - /* compute stride in texels from stride in bytes */ - texImage->RowStride = stImage->transfer->stride - * util_format_get_blockwidth(stImage->pt->format) - / util_format_get_blocksize(stImage->pt->format); - } - else { - /* Otherwise, the image should actually be stored in - * texImage->Data. This is pretty confusing for - * everybody, I'd much prefer to separate the two functions of - * texImage->Data - storage for texture images in main memory - * and access (ie mappings) of images. In other words, we'd - * create a new texImage->Map field and leave Data simply for - * storage. - */ - assert(texImage->Data); - } - - depth = texImage->Depth; - texImage->Depth = 1; - - dest = (GLubyte *) pixels; - - _mesa_set_fetch_functions(texImage, get_texture_dims(target)); - - for (i = 0; i < depth; i++) { - if (compressed_dst) { - _mesa_get_compressed_teximage(ctx, target, level, dest, - texObj, texImage); - } - else { - _mesa_get_teximage(ctx, target, level, format, type, dest, - texObj, texImage); - } - - if (stImage->pt && i + 1 < depth) { - /* unmap this slice */ - st_texture_image_unmap(st, stImage); - /* map next slice of 3D texture */ - texImage->Data = st_texture_image_map(st, stImage, i + 1, - PIPE_TRANSFER_READ, 0, 0, - stImage->base.Width, - stImage->base.Height); - dest += dstImageStride; - } - } - - texImage->Depth = depth; - - /* Unmap */ - if (stImage->pt) { - st_texture_image_unmap(st, stImage); - texImage->Data = NULL; - } -} - - -static void -st_GetTexImage(struct gl_context * ctx, GLenum target, GLint level, - GLenum format, GLenum type, GLvoid * pixels, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - st_get_tex_image(ctx, target, level, format, type, pixels, texObj, texImage, - GL_FALSE); -} - - -static void -st_GetCompressedTexImage(struct gl_context *ctx, GLenum target, GLint level, - GLvoid *pixels, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - st_get_tex_image(ctx, target, level, 0, 0, pixels, texObj, texImage, - GL_TRUE); -} - - - -static void -st_TexSubimage(struct gl_context *ctx, GLint dims, GLenum target, GLint level, - GLint xoffset, GLint yoffset, GLint zoffset, - GLint width, GLint height, GLint depth, - GLenum format, GLenum type, const void *pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - struct st_context *st = st_context(ctx); - struct pipe_screen *screen = st->pipe->screen; - struct st_texture_image *stImage = st_texture_image(texImage); - GLuint dstRowStride; - const GLuint srcImageStride = - _mesa_image_image_stride(packing, width, height, format, type); - GLint i; - const GLubyte *src; - /* init to silence warning only: */ - enum pipe_transfer_usage transfer_usage = PIPE_TRANSFER_WRITE; - - DBG("%s target %s level %d offset %d,%d %dx%d\n", __FUNCTION__, - _mesa_lookup_enum_by_nr(target), - level, xoffset, yoffset, width, height); - - pixels = - _mesa_validate_pbo_teximage(ctx, dims, width, height, depth, format, - type, pixels, packing, "glTexSubImage2D"); - if (!pixels) - return; - - /* See if we can do texture compression with a blit/render. - */ - if (!ctx->Mesa_DXTn && - _mesa_is_format_compressed(texImage->TexFormat) && - screen->is_format_supported(screen, - stImage->pt->format, - stImage->pt->target, 0, - PIPE_BIND_RENDER_TARGET, 0)) { - if (compress_with_blit(ctx, target, level, - xoffset, yoffset, zoffset, - width, height, depth, - format, type, pixels, packing, texImage)) { - goto done; - } - } - - /* Map buffer if necessary. Need to lock to prevent other contexts - * from uploading the buffer under us. - */ - if (stImage->pt) { - if (format == GL_DEPTH_COMPONENT && - util_format_is_depth_and_stencil(stImage->pt->format)) - transfer_usage = PIPE_TRANSFER_READ_WRITE; - else - transfer_usage = PIPE_TRANSFER_WRITE; - - texImage->Data = st_texture_image_map(st, stImage, zoffset, - transfer_usage, - xoffset, yoffset, - width, height); - } - - if (!texImage->Data) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage"); - goto done; - } - - src = (const GLubyte *) pixels; - dstRowStride = stImage->transfer->stride; - - for (i = 0; i < depth; i++) { - if (!_mesa_texstore(ctx, dims, texImage->_BaseFormat, - texImage->TexFormat, - texImage->Data, - 0, 0, 0, - dstRowStride, - texImage->ImageOffsets, - width, height, 1, - format, type, src, packing)) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage"); - } - - if (stImage->pt && i + 1 < depth) { - /* unmap this slice */ - st_texture_image_unmap(st, stImage); - /* map next slice of 3D texture */ - texImage->Data = st_texture_image_map(st, stImage, - zoffset + i + 1, - transfer_usage, - xoffset, yoffset, - width, height); - src += srcImageStride; - } - } - -done: - _mesa_unmap_teximage_pbo(ctx, packing); - - if (stImage->pt && texImage->Data) { - st_texture_image_unmap(st, stImage); - texImage->Data = NULL; - } -} - - - -static void -st_TexSubImage3D(struct gl_context *ctx, GLenum target, GLint level, - GLint xoffset, GLint yoffset, GLint zoffset, - GLsizei width, GLsizei height, GLsizei depth, - GLenum format, GLenum type, const GLvoid *pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - st_TexSubimage(ctx, 3, target, level, xoffset, yoffset, zoffset, - width, height, depth, format, type, - pixels, packing, texObj, texImage); -} - - -static void -st_TexSubImage2D(struct gl_context *ctx, GLenum target, GLint level, - GLint xoffset, GLint yoffset, - GLsizei width, GLsizei height, - GLenum format, GLenum type, const GLvoid * pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - st_TexSubimage(ctx, 2, target, level, xoffset, yoffset, 0, - width, height, 1, format, type, - pixels, packing, texObj, texImage); -} - - -static void -st_TexSubImage1D(struct gl_context *ctx, GLenum target, GLint level, - GLint xoffset, GLsizei width, GLenum format, GLenum type, - const GLvoid * pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - st_TexSubimage(ctx, 1, target, level, xoffset, 0, 0, width, 1, 1, - format, type, pixels, packing, texObj, texImage); -} - - -static void -st_CompressedTexSubImage1D(struct gl_context *ctx, GLenum target, GLint level, - GLint xoffset, GLsizei width, - GLenum format, - GLsizei imageSize, const GLvoid *data, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - assert(0); -} - - -static void -st_CompressedTexSubImage2D(struct gl_context *ctx, GLenum target, GLint level, - GLint xoffset, GLint yoffset, - GLsizei width, GLint height, - GLenum format, - GLsizei imageSize, const GLvoid *data, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - struct st_context *st = st_context(ctx); - struct st_texture_image *stImage = st_texture_image(texImage); - int srcBlockStride; - int dstBlockStride; - int y; - enum pipe_format pformat; - - if (stImage->pt) { - pformat = stImage->pt->format; - - texImage->Data = st_texture_image_map(st, stImage, 0, - PIPE_TRANSFER_WRITE, - xoffset, yoffset, - width, height); - - srcBlockStride = util_format_get_stride(pformat, width); - dstBlockStride = stImage->transfer->stride; - } else { - assert(stImage->pt); - /* TODO find good values for block and strides */ - /* TODO also adjust texImage->data for yoffset/xoffset */ - return; - } - - if (!texImage->Data) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexSubImage"); - return; - } - - assert(xoffset % util_format_get_blockwidth(pformat) == 0); - assert(yoffset % util_format_get_blockheight(pformat) == 0); - - for (y = 0; y < height; y += util_format_get_blockheight(pformat)) { - /* don't need to adjust for xoffset and yoffset as st_texture_image_map does that */ - const char *src = (const char*)data + srcBlockStride * util_format_get_nblocksy(pformat, y); - char *dst = (char*)texImage->Data + dstBlockStride * util_format_get_nblocksy(pformat, y); - memcpy(dst, src, util_format_get_stride(pformat, width)); - } - - if (stImage->pt) { - st_texture_image_unmap(st, stImage); - texImage->Data = NULL; - } -} - - -static void -st_CompressedTexSubImage3D(struct gl_context *ctx, GLenum target, GLint level, - GLint xoffset, GLint yoffset, GLint zoffset, - GLsizei width, GLint height, GLint depth, - GLenum format, - GLsizei imageSize, const GLvoid *data, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - assert(0); -} - - - -/** - * Do a CopyTexSubImage operation using a read transfer from the source, - * a write transfer to the destination and get_tile()/put_tile() to access - * the pixels/texels. - * - * Note: srcY=0=TOP of renderbuffer - */ -static void -fallback_copy_texsubimage(struct gl_context *ctx, GLenum target, GLint level, - struct st_renderbuffer *strb, - struct st_texture_image *stImage, - GLenum baseFormat, - GLint destX, GLint destY, GLint destZ, - GLint srcX, GLint srcY, - GLsizei width, GLsizei height) -{ - struct st_context *st = st_context(ctx); - struct pipe_context *pipe = st->pipe; - struct pipe_transfer *src_trans; - GLvoid *texDest; - enum pipe_transfer_usage transfer_usage; - - if (ST_DEBUG & DEBUG_FALLBACK) - debug_printf("%s: fallback processing\n", __FUNCTION__); - - assert(width <= MAX_WIDTH); - - if (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP) { - srcY = strb->Base.Height - srcY - height; - } - - src_trans = pipe_get_transfer(st_context(ctx)->pipe, - strb->texture, - 0, 0, - PIPE_TRANSFER_READ, - srcX, srcY, - width, height); - - if ((baseFormat == GL_DEPTH_COMPONENT || - baseFormat == GL_DEPTH_STENCIL) && - util_format_is_depth_and_stencil(stImage->pt->format)) - transfer_usage = PIPE_TRANSFER_READ_WRITE; - else - transfer_usage = PIPE_TRANSFER_WRITE; - - /* XXX this used to ignore destZ param */ - texDest = st_texture_image_map(st, stImage, destZ, transfer_usage, - destX, destY, width, height); - - if (baseFormat == GL_DEPTH_COMPONENT || - baseFormat == GL_DEPTH_STENCIL) { - const GLboolean scaleOrBias = (ctx->Pixel.DepthScale != 1.0F || - ctx->Pixel.DepthBias != 0.0F); - GLint row, yStep; - - /* determine bottom-to-top vs. top-to-bottom order for src buffer */ - if (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP) { - srcY = height - 1; - yStep = -1; - } - else { - srcY = 0; - yStep = 1; - } - - /* To avoid a large temp memory allocation, do copy row by row */ - for (row = 0; row < height; row++, srcY += yStep) { - uint data[MAX_WIDTH]; - pipe_get_tile_z(pipe, src_trans, 0, srcY, width, 1, data); - if (scaleOrBias) { - _mesa_scale_and_bias_depth_uint(ctx, width, data); - } - pipe_put_tile_z(pipe, stImage->transfer, 0, row, width, 1, data); - } - } - else { - /* RGBA format */ - GLfloat *tempSrc = - (GLfloat *) malloc(width * height * 4 * sizeof(GLfloat)); - - if (tempSrc && texDest) { - const GLint dims = 2; - const GLint dstRowStride = stImage->transfer->stride; - struct gl_texture_image *texImage = &stImage->base; - struct gl_pixelstore_attrib unpack = ctx->DefaultPacking; - - if (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP) { - unpack.Invert = GL_TRUE; - } - - /* get float/RGBA image from framebuffer */ - /* XXX this usually involves a lot of int/float conversion. - * try to avoid that someday. - */ - pipe_get_tile_rgba_format(pipe, src_trans, 0, 0, width, height, - util_format_linear(strb->texture->format), - tempSrc); - - /* Store into texture memory. - * Note that this does some special things such as pixel transfer - * ops and format conversion. In particular, if the dest tex format - * is actually RGBA but the user created the texture as GL_RGB we - * need to fill-in/override the alpha channel with 1.0. - */ - _mesa_texstore(ctx, dims, - texImage->_BaseFormat, - texImage->TexFormat, - texDest, - 0, 0, 0, - dstRowStride, - texImage->ImageOffsets, - width, height, 1, - GL_RGBA, GL_FLOAT, tempSrc, /* src */ - &unpack); - } - else { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage"); - } - - if (tempSrc) - free(tempSrc); - } - - st_texture_image_unmap(st, stImage); - pipe->transfer_destroy(pipe, src_trans); -} - - - -/** - * If the format of the src renderbuffer and the format of the dest - * texture are compatible (in terms of blitting), return a TGSI writemask - * to be used during the blit. - * If the src/dest are incompatible, return 0. - */ -static unsigned -compatible_src_dst_formats(struct gl_context *ctx, - const struct gl_renderbuffer *src, - const struct gl_texture_image *dst) -{ - /* Get logical base formats for the src and dest. - * That is, use the user-requested formats and not the actual, device- - * chosen formats. - * For example, the user may have requested an A8 texture but the - * driver may actually be using an RGBA texture format. When we - * copy/blit to that texture, we only want to copy the Alpha channel - * and not the RGB channels. - * - * Similarly, when the src FBO was created an RGB format may have been - * requested but the driver actually chose an RGBA format. In that case, - * we don't want to copy the undefined Alpha channel to the dest texture - * (it should be 1.0). - */ - const GLenum srcFormat = _mesa_base_fbo_format(ctx, src->InternalFormat); - const GLenum dstFormat = _mesa_base_tex_format(ctx, dst->InternalFormat); - - /** - * XXX when we have red-only and red/green renderbuffers we'll need - * to add more cases here (or implement a general-purpose routine that - * queries the existance of the R,G,B,A channels in the src and dest). - */ - if (srcFormat == dstFormat) { - /* This is the same as matching_base_formats, which should - * always pass, as it did previously. - */ - return TGSI_WRITEMASK_XYZW; - } - else if (srcFormat == GL_RGB && dstFormat == GL_RGBA) { - /* Make sure that A in the dest is 1. The actual src format - * may be RGBA and have undefined A values. - */ - return TGSI_WRITEMASK_XYZ; - } - else if (srcFormat == GL_RGBA && dstFormat == GL_RGB) { - /* Make sure that A in the dest is 1. The actual dst format - * may be RGBA and will need A=1 to provide proper alpha values - * when sampled later. - */ - return TGSI_WRITEMASK_XYZ; - } - else { - if (ST_DEBUG & DEBUG_FALLBACK) - debug_printf("%s failed for src %s, dst %s\n", - __FUNCTION__, - _mesa_lookup_enum_by_nr(srcFormat), - _mesa_lookup_enum_by_nr(dstFormat)); - - /* Otherwise fail. - */ - return 0; - } -} - - - -/** - * Do a CopyTex[Sub]Image1/2/3D() using a hardware (blit) path if possible. - * Note that the region to copy has already been clipped so we know we - * won't read from outside the source renderbuffer's bounds. - * - * Note: srcY=0=Bottom of renderbuffer (GL convention) - */ -static void -st_copy_texsubimage(struct gl_context *ctx, - GLenum target, GLint level, - GLint destX, GLint destY, GLint destZ, - GLint srcX, GLint srcY, - GLsizei width, GLsizei height) -{ - struct gl_texture_unit *texUnit = - &ctx->Texture.Unit[ctx->Texture.CurrentUnit]; - struct gl_texture_object *texObj = - _mesa_select_tex_object(ctx, texUnit, target); - struct gl_texture_image *texImage = - _mesa_select_tex_image(ctx, texObj, target, level); - struct st_texture_image *stImage = st_texture_image(texImage); - const GLenum texBaseFormat = texImage->_BaseFormat; - struct gl_framebuffer *fb = ctx->ReadBuffer; - struct st_renderbuffer *strb; - struct st_context *st = st_context(ctx); - struct pipe_context *pipe = st->pipe; - struct pipe_screen *screen = pipe->screen; - enum pipe_format dest_format, src_format; - GLboolean use_fallback = GL_TRUE; - GLboolean matching_base_formats; - GLuint format_writemask, sample_count; - struct pipe_surface *dest_surface = NULL; - GLboolean do_flip = (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP); - - /* make sure finalize_textures has been called? - */ - if (0) st_validate_state(st); - - /* determine if copying depth or color data */ - if (texBaseFormat == GL_DEPTH_COMPONENT || - texBaseFormat == GL_DEPTH_STENCIL) { - strb = st_renderbuffer(fb->_DepthBuffer); - if (strb->Base.Wrapped) { - strb = st_renderbuffer(strb->Base.Wrapped); - } - } - else { - /* texBaseFormat == GL_RGB, GL_RGBA, GL_ALPHA, etc */ - strb = st_renderbuffer(fb->_ColorReadBuffer); - } - - if (!strb || !strb->surface || !stImage->pt) { - debug_printf("%s: null strb or stImage\n", __FUNCTION__); - return; - } - - sample_count = strb->surface->texture->nr_samples; - /* I believe this would be legal, presumably would need to do a resolve - for color, and for depth/stencil spec says to just use one of the - depth/stencil samples per pixel? Need some transfer clarifications. */ - assert(sample_count < 2); - - if (srcX < 0) { - width -= -srcX; - destX += -srcX; - srcX = 0; - } - - if (srcY < 0) { - height -= -srcY; - destY += -srcY; - srcY = 0; - } - - if (destX < 0) { - width -= -destX; - srcX += -destX; - destX = 0; - } - - if (destY < 0) { - height -= -destY; - srcY += -destY; - destY = 0; - } - - if (width < 0 || height < 0) - return; - - - assert(strb); - assert(strb->surface); - assert(stImage->pt); - - src_format = strb->surface->format; - dest_format = stImage->pt->format; - - /* - * Determine if the src framebuffer and dest texture have the same - * base format. We need this to detect a case such as the framebuffer - * being GL_RGBA but the texture being GL_RGB. If the actual hardware - * texture format stores RGBA we need to set A=1 (overriding the - * framebuffer's alpha values). We can't do that with the blit or - * textured-quad paths. - */ - matching_base_formats = - (_mesa_get_format_base_format(strb->Base.Format) == - _mesa_get_format_base_format(texImage->TexFormat)); - format_writemask = compatible_src_dst_formats(ctx, &strb->Base, texImage); - - if (ctx->_ImageTransferState == 0x0) { - - if (matching_base_formats && - src_format == dest_format && - !do_flip) - { - /* use surface_copy() / blit */ - struct pipe_box src_box; - u_box_2d_zslice(srcX, srcY, strb->surface->u.tex.first_layer, - width, height, &src_box); - - /* for resource_copy_region(), y=0=top, always */ - pipe->resource_copy_region(pipe, - /* dest */ - stImage->pt, - stImage->level, - destX, destY, destZ + stImage->face, - /* src */ - strb->texture, - strb->surface->u.tex.level, - &src_box); - use_fallback = GL_FALSE; - } - else if (format_writemask && - texBaseFormat != GL_DEPTH_COMPONENT && - texBaseFormat != GL_DEPTH_STENCIL && - screen->is_format_supported(screen, src_format, - PIPE_TEXTURE_2D, sample_count, - PIPE_BIND_SAMPLER_VIEW, - 0) && - screen->is_format_supported(screen, dest_format, - PIPE_TEXTURE_2D, 0, - PIPE_BIND_RENDER_TARGET, - 0)) { - /* draw textured quad to do the copy */ - GLint srcY0, srcY1; - struct pipe_surface surf_tmpl; - memset(&surf_tmpl, 0, sizeof(surf_tmpl)); - surf_tmpl.format = stImage->pt->format; - surf_tmpl.usage = PIPE_BIND_RENDER_TARGET; - surf_tmpl.u.tex.level = stImage->level; - surf_tmpl.u.tex.first_layer = stImage->face + destZ; - surf_tmpl.u.tex.last_layer = stImage->face + destZ; - - dest_surface = pipe->create_surface(pipe, stImage->pt, - &surf_tmpl); - - if (do_flip) { - srcY1 = strb->Base.Height - srcY - height; - srcY0 = srcY1 + height; - } - else { - srcY0 = srcY; - srcY1 = srcY0 + height; - } - - util_blit_pixels_writemask(st->blit, - strb->texture, - strb->surface->u.tex.level, - srcX, srcY0, - srcX + width, srcY1, - strb->surface->u.tex.first_layer, - dest_surface, - destX, destY, - destX + width, destY + height, - 0.0, PIPE_TEX_MIPFILTER_NEAREST, - format_writemask); - use_fallback = GL_FALSE; - } - - if (dest_surface) - pipe_surface_reference(&dest_surface, NULL); - } - - if (use_fallback) { - /* software fallback */ - fallback_copy_texsubimage(ctx, target, level, - strb, stImage, texBaseFormat, - destX, destY, destZ, - srcX, srcY, width, height); - } -} - - - -static void -st_CopyTexImage1D(struct gl_context * ctx, GLenum target, GLint level, - GLenum internalFormat, - GLint x, GLint y, GLsizei width, GLint border) -{ - struct gl_texture_unit *texUnit = - &ctx->Texture.Unit[ctx->Texture.CurrentUnit]; - struct gl_texture_object *texObj = - _mesa_select_tex_object(ctx, texUnit, target); - struct gl_texture_image *texImage = - _mesa_select_tex_image(ctx, texObj, target, level); - - /* Setup or redefine the texture object, texture and texture - * image. Don't populate yet. - */ - ctx->Driver.TexImage1D(ctx, target, level, internalFormat, - width, border, - GL_RGBA, CHAN_TYPE, NULL, - &ctx->DefaultPacking, texObj, texImage); - - st_copy_texsubimage(ctx, target, level, - 0, 0, 0, /* destX,Y,Z */ - x, y, width, 1); /* src X, Y, size */ -} - - -static void -st_CopyTexImage2D(struct gl_context * ctx, GLenum target, GLint level, - GLenum internalFormat, - GLint x, GLint y, GLsizei width, GLsizei height, - GLint border) -{ - struct gl_texture_unit *texUnit = - &ctx->Texture.Unit[ctx->Texture.CurrentUnit]; - struct gl_texture_object *texObj = - _mesa_select_tex_object(ctx, texUnit, target); - struct gl_texture_image *texImage = - _mesa_select_tex_image(ctx, texObj, target, level); - - /* Setup or redefine the texture object, texture and texture - * image. Don't populate yet. - */ - ctx->Driver.TexImage2D(ctx, target, level, internalFormat, - width, height, border, - GL_RGBA, CHAN_TYPE, NULL, - &ctx->DefaultPacking, texObj, texImage); - - st_copy_texsubimage(ctx, target, level, - 0, 0, 0, /* destX,Y,Z */ - x, y, width, height); /* src X, Y, size */ -} - - -static void -st_CopyTexSubImage1D(struct gl_context * ctx, GLenum target, GLint level, - GLint xoffset, GLint x, GLint y, GLsizei width) -{ - const GLint yoffset = 0, zoffset = 0; - const GLsizei height = 1; - st_copy_texsubimage(ctx, target, level, - xoffset, yoffset, zoffset, /* destX,Y,Z */ - x, y, width, height); /* src X, Y, size */ -} - - -static void -st_CopyTexSubImage2D(struct gl_context * ctx, GLenum target, GLint level, - GLint xoffset, GLint yoffset, - GLint x, GLint y, GLsizei width, GLsizei height) -{ - const GLint zoffset = 0; - st_copy_texsubimage(ctx, target, level, - xoffset, yoffset, zoffset, /* destX,Y,Z */ - x, y, width, height); /* src X, Y, size */ -} - - -static void -st_CopyTexSubImage3D(struct gl_context * ctx, GLenum target, GLint level, - GLint xoffset, GLint yoffset, GLint zoffset, - GLint x, GLint y, GLsizei width, GLsizei height) -{ - st_copy_texsubimage(ctx, target, level, - xoffset, yoffset, zoffset, /* destX,Y,Z */ - x, y, width, height); /* src X, Y, size */ -} - - -/** - * Copy image data from stImage into the texture object 'stObj' at level - * 'dstLevel'. - */ -static void -copy_image_data_to_texture(struct st_context *st, - struct st_texture_object *stObj, - GLuint dstLevel, - struct st_texture_image *stImage) -{ - /* debug checks */ - { - const struct gl_texture_image *dstImage = - stObj->base.Image[stImage->face][stImage->level]; - assert(dstImage); - assert(dstImage->Width == stImage->base.Width); - assert(dstImage->Height == stImage->base.Height); - assert(dstImage->Depth == stImage->base.Depth); - } - - if (stImage->pt) { - /* Copy potentially with the blitter: - */ - st_texture_image_copy(st->pipe, - stObj->pt, dstLevel, /* dest texture, level */ - stImage->pt, stImage->level, /* src texture, level */ - stImage->face); - - pipe_resource_reference(&stImage->pt, NULL); - } - else if (stImage->base.Data) { - st_texture_image_data(st, - stObj->pt, - stImage->face, - dstLevel, - stImage->base.Data, - stImage->base.RowStride * - util_format_get_blocksize(stObj->pt->format), - stImage->base.RowStride * - stImage->base.Height * - util_format_get_blocksize(stObj->pt->format)); - _mesa_align_free(stImage->base.Data); - stImage->base.Data = NULL; - } - - pipe_resource_reference(&stImage->pt, stObj->pt); -} - - -/** - * Called during state validation. When this function is finished, - * the texture object should be ready for rendering. - * \return GL_TRUE for success, GL_FALSE for failure (out of mem) - */ -GLboolean -st_finalize_texture(struct gl_context *ctx, - struct pipe_context *pipe, - struct gl_texture_object *tObj) -{ - struct st_context *st = st_context(ctx); - struct st_texture_object *stObj = st_texture_object(tObj); - const GLuint nr_faces = (stObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1; - GLuint face; - struct st_texture_image *firstImage; - enum pipe_format firstImageFormat; - - if (stObj->base._Complete) { - /* The texture is complete and we know exactly how many mipmap levels - * are present/needed. This is conditional because we may be called - * from the st_generate_mipmap() function when the texture object is - * incomplete. In that case, we'll have set stObj->lastLevel before - * we get here. - */ - if (stObj->base.MinFilter == GL_LINEAR || - stObj->base.MinFilter == GL_NEAREST) - stObj->lastLevel = stObj->base.BaseLevel; - else - stObj->lastLevel = stObj->base._MaxLevel; - } - - firstImage = st_texture_image(stObj->base.Image[0][stObj->base.BaseLevel]); - assert(firstImage); - - /* If both firstImage and stObj point to a texture which can contain - * all active images, favour firstImage. Note that because of the - * completeness requirement, we know that the image dimensions - * will match. - */ - if (firstImage->pt && - firstImage->pt != stObj->pt && - (!stObj->pt || firstImage->pt->last_level >= stObj->pt->last_level)) { - pipe_resource_reference(&stObj->pt, firstImage->pt); - pipe_sampler_view_reference(&stObj->sampler_view, NULL); - } - - /* Find gallium format for the Mesa texture */ - firstImageFormat = st_mesa_format_to_pipe_format(firstImage->base.TexFormat); - - /* If we already have a gallium texture, check that it matches the texture - * object's format, target, size, num_levels, etc. - */ - if (stObj->pt) { - if (stObj->pt->target != gl_target_to_pipe(stObj->base.Target) || - !st_sampler_compat_formats(stObj->pt->format, firstImageFormat) || - stObj->pt->last_level < stObj->lastLevel || - stObj->pt->width0 != stObj->width0 || - stObj->pt->height0 != stObj->height0 || - stObj->pt->depth0 != stObj->depth0) - { - /* The gallium texture does not match the Mesa texture so delete the - * gallium texture now. We'll make a new one below. - */ - pipe_resource_reference(&stObj->pt, NULL); - pipe_sampler_view_reference(&stObj->sampler_view, NULL); - st->dirty.st |= ST_NEW_FRAMEBUFFER; - } - } - - /* May need to create a new gallium texture: - */ - if (!stObj->pt) { - GLuint bindings = default_bindings(st, firstImageFormat); - - stObj->pt = st_texture_create(st, - gl_target_to_pipe(stObj->base.Target), - firstImageFormat, - stObj->lastLevel, - stObj->width0, - stObj->height0, - stObj->depth0, - bindings); - - if (!stObj->pt) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); - return GL_FALSE; - } - } - - /* Pull in any images not in the object's texture: - */ - for (face = 0; face < nr_faces; face++) { - GLuint level; - for (level = stObj->base.BaseLevel; level <= stObj->lastLevel; level++) { - struct st_texture_image *stImage = - st_texture_image(stObj->base.Image[face][level]); - - /* Need to import images in main memory or held in other textures. - */ - if (stImage && stObj->pt != stImage->pt) { - copy_image_data_to_texture(st, stObj, level, stImage); - } - } - } - - return GL_TRUE; -} - - -/** - * Returns pointer to a default/dummy texture. - * This is typically used when the current shader has tex/sample instructions - * but the user has not provided a (any) texture(s). - */ -struct gl_texture_object * -st_get_default_texture(struct st_context *st) -{ - if (!st->default_texture) { - static const GLenum target = GL_TEXTURE_2D; - GLubyte pixels[16][16][4]; - struct gl_texture_object *texObj; - struct gl_texture_image *texImg; - GLuint i, j; - - /* The ARB_fragment_program spec says (0,0,0,1) should be returned - * when attempting to sample incomplete textures. - */ - for (i = 0; i < 16; i++) { - for (j = 0; j < 16; j++) { - pixels[i][j][0] = 0; - pixels[i][j][1] = 0; - pixels[i][j][2] = 0; - pixels[i][j][3] = 255; - } - } - - texObj = st->ctx->Driver.NewTextureObject(st->ctx, 0, target); - - texImg = _mesa_get_tex_image(st->ctx, texObj, target, 0); - - _mesa_init_teximage_fields(st->ctx, target, texImg, - 16, 16, 1, 0, /* w, h, d, border */ - GL_RGBA, MESA_FORMAT_RGBA8888); - - st_TexImage(st->ctx, 2, target, - 0, GL_RGBA, /* level, intformat */ - 16, 16, 1, 0, /* w, h, d, border */ - GL_RGBA, GL_UNSIGNED_BYTE, pixels, - &st->ctx->DefaultPacking, - texObj, texImg, - 0, 0); - - texObj->MinFilter = GL_NEAREST; - texObj->MagFilter = GL_NEAREST; - texObj->_Complete = GL_TRUE; - - st->default_texture = texObj; - } - return st->default_texture; -} - - -void -st_init_texture_functions(struct dd_function_table *functions) -{ - functions->ChooseTextureFormat = st_ChooseTextureFormat; - functions->TexImage1D = st_TexImage1D; - functions->TexImage2D = st_TexImage2D; - functions->TexImage3D = st_TexImage3D; - functions->TexSubImage1D = st_TexSubImage1D; - functions->TexSubImage2D = st_TexSubImage2D; - functions->TexSubImage3D = st_TexSubImage3D; - functions->CompressedTexSubImage1D = st_CompressedTexSubImage1D; - functions->CompressedTexSubImage2D = st_CompressedTexSubImage2D; - functions->CompressedTexSubImage3D = st_CompressedTexSubImage3D; - functions->CopyTexImage1D = st_CopyTexImage1D; - functions->CopyTexImage2D = st_CopyTexImage2D; - functions->CopyTexSubImage1D = st_CopyTexSubImage1D; - functions->CopyTexSubImage2D = st_CopyTexSubImage2D; - functions->CopyTexSubImage3D = st_CopyTexSubImage3D; - functions->GenerateMipmap = st_generate_mipmap; - - functions->GetTexImage = st_GetTexImage; - - /* compressed texture functions */ - functions->CompressedTexImage2D = st_CompressedTexImage2D; - functions->GetCompressedTexImage = st_GetCompressedTexImage; - - functions->NewTextureObject = st_NewTextureObject; - functions->NewTextureImage = st_NewTextureImage; - functions->DeleteTexture = st_DeleteTextureObject; - functions->FreeTexImageData = st_FreeTextureImageData; - - functions->TextureMemCpy = do_memcpy; - - /* XXX Temporary until we can query pipe's texture sizes */ - functions->TestProxyTexImage = _mesa_test_proxy_teximage; -} +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * 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, sub license, 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 NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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/mfeatures.h" +#include "main/bufferobj.h" +#include "main/enums.h" +#include "main/fbobject.h" +#include "main/formats.h" +#include "main/image.h" +#include "main/imports.h" +#include "main/macros.h" +#include "main/mipmap.h" +#include "main/pack.h" +#include "main/pixeltransfer.h" +#include "main/texcompress.h" +#include "main/texfetch.h" +#include "main/texgetimage.h" +#include "main/teximage.h" +#include "main/texobj.h" +#include "main/texstore.h" + +#include "state_tracker/st_debug.h" +#include "state_tracker/st_context.h" +#include "state_tracker/st_cb_fbo.h" +#include "state_tracker/st_cb_flush.h" +#include "state_tracker/st_cb_texture.h" +#include "state_tracker/st_format.h" +#include "state_tracker/st_texture.h" +#include "state_tracker/st_gen_mipmap.h" +#include "state_tracker/st_atom.h" + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "util/u_inlines.h" +#include "pipe/p_shader_tokens.h" +#include "util/u_tile.h" +#include "util/u_blit.h" +#include "util/u_format.h" +#include "util/u_surface.h" +#include "util/u_sampler.h" +#include "util/u_math.h" +#include "util/u_box.h" + +#define DBG if (0) printf + + +static enum pipe_texture_target +gl_target_to_pipe(GLenum target) +{ + switch (target) { + case GL_TEXTURE_1D: + return PIPE_TEXTURE_1D; + case GL_TEXTURE_2D: + return PIPE_TEXTURE_2D; + case GL_TEXTURE_RECTANGLE_NV: + return PIPE_TEXTURE_RECT; + case GL_TEXTURE_3D: + return PIPE_TEXTURE_3D; + case GL_TEXTURE_CUBE_MAP_ARB: + return PIPE_TEXTURE_CUBE; + case GL_TEXTURE_1D_ARRAY_EXT: + return PIPE_TEXTURE_1D_ARRAY; + case GL_TEXTURE_2D_ARRAY_EXT: + return PIPE_TEXTURE_2D_ARRAY; + default: + assert(0); + return 0; + } +} + + +/** called via ctx->Driver.NewTextureImage() */ +static struct gl_texture_image * +st_NewTextureImage(struct gl_context * ctx) +{ + DBG("%s\n", __FUNCTION__); + (void) ctx; + return (struct gl_texture_image *) ST_CALLOC_STRUCT(st_texture_image); +} + + +/** called via ctx->Driver.NewTextureObject() */ +static struct gl_texture_object * +st_NewTextureObject(struct gl_context * ctx, GLuint name, GLenum target) +{ + struct st_texture_object *obj = ST_CALLOC_STRUCT(st_texture_object); + + DBG("%s\n", __FUNCTION__); + _mesa_initialize_texture_object(&obj->base, name, target); + + return &obj->base; +} + +/** called via ctx->Driver.DeleteTextureObject() */ +static void +st_DeleteTextureObject(struct gl_context *ctx, + struct gl_texture_object *texObj) +{ + struct st_context *st = st_context(ctx); + struct st_texture_object *stObj = st_texture_object(texObj); + if (stObj->pt) + pipe_resource_reference(&stObj->pt, NULL); + if (stObj->sampler_view) { + if (stObj->sampler_view->context != st->pipe) { + /* Take "ownership" of this texture sampler view by setting + * its context pointer to this context. This avoids potential + * crashes when the texture object is shared among contexts + * and the original/owner context has already been destroyed. + */ + stObj->sampler_view->context = st->pipe; + } + pipe_sampler_view_reference(&stObj->sampler_view, NULL); + } + _mesa_delete_texture_object(ctx, texObj); +} + + +/** called via ctx->Driver.FreeTexImageData() */ +static void +st_FreeTextureImageData(struct gl_context * ctx, struct gl_texture_image *texImage) +{ + struct st_texture_image *stImage = st_texture_image(texImage); + + DBG("%s\n", __FUNCTION__); + + if (stImage->pt) { + pipe_resource_reference(&stImage->pt, NULL); + } + + if (texImage->Data) { + _mesa_align_free(texImage->Data); + texImage->Data = NULL; + } +} + + +/** + * From linux kernel i386 header files, copes with odd sizes better + * than COPY_DWORDS would: + * XXX Put this in src/mesa/main/imports.h ??? + */ +#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86) +static INLINE void * +__memcpy(void *to, const void *from, size_t n) +{ + int d0, d1, d2; + __asm__ __volatile__("rep ; movsl\n\t" + "testb $2,%b4\n\t" + "je 1f\n\t" + "movsw\n" + "1:\ttestb $1,%b4\n\t" + "je 2f\n\t" + "movsb\n" "2:":"=&c"(d0), "=&D"(d1), "=&S"(d2) + :"0"(n / 4), "q"(n), "1"((long) to), "2"((long) from) + :"memory"); + return (to); +} +#else +#define __memcpy(a,b,c) memcpy(a,b,c) +#endif + + +/** + * The system memcpy (at least on ubuntu 5.10) has problems copying + * to agp (writecombined) memory from a source which isn't 64-byte + * aligned - there is a 4x performance falloff. + * + * The x86 __memcpy is immune to this but is slightly slower + * (10%-ish) than the system memcpy. + * + * The sse_memcpy seems to have a slight cliff at 64/32 bytes, but + * isn't much faster than x86_memcpy for agp copies. + * + * TODO: switch dynamically. + */ +static void * +do_memcpy(void *dest, const void *src, size_t n) +{ + if ((((unsigned long) src) & 63) || (((unsigned long) dest) & 63)) { + return __memcpy(dest, src, n); + } + else + return memcpy(dest, src, n); +} + + +/** + * Return default texture resource binding bitmask for the given format. + */ +static GLuint +default_bindings(struct st_context *st, enum pipe_format format) +{ + struct pipe_screen *screen = st->pipe->screen; + const unsigned target = PIPE_TEXTURE_2D; + const unsigned geom = 0x0; + unsigned bindings; + + if (util_format_is_depth_or_stencil(format)) + bindings = PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_DEPTH_STENCIL; + else + bindings = PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET; + + if (screen->is_format_supported(screen, format, target, 0, bindings, geom)) + return bindings; + else + return PIPE_BIND_SAMPLER_VIEW; +} + + +/** Return number of image dimensions (1, 2 or 3) for a texture target. */ +static GLuint +get_texture_dims(GLenum target) +{ + switch (target) { + case GL_TEXTURE_1D: + case GL_TEXTURE_1D_ARRAY_EXT: + return 1; + case GL_TEXTURE_2D: + case GL_TEXTURE_CUBE_MAP_ARB: + case GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_X_ARB: + case GL_TEXTURE_CUBE_MAP_POSITIVE_Y_ARB: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y_ARB: + case GL_TEXTURE_CUBE_MAP_POSITIVE_Z_ARB: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z_ARB: + case GL_TEXTURE_RECTANGLE_NV: + case GL_TEXTURE_2D_ARRAY_EXT: + return 2; + case GL_TEXTURE_3D: + return 3; + default: + assert(0 && "invalid texture target in get_texture_dims()"); + return 1; + } +} + + +/** + * Try to allocate a pipe_resource object for the given st_texture_object. + * + * We use the given st_texture_image as a clue to determine the size of the + * mipmap image at level=0. + * + * \return GL_TRUE for success, GL_FALSE if out of memory. + */ +static GLboolean +guess_and_alloc_texture(struct st_context *st, + struct st_texture_object *stObj, + const struct st_texture_image *stImage) +{ + const GLuint dims = get_texture_dims(stObj->base.Target); + GLuint level, lastLevel, width, height, depth; + GLuint bindings; + enum pipe_format fmt; + + DBG("%s\n", __FUNCTION__); + + assert(!stObj->pt); + + level = stImage->level; + width = stImage->base.Width2; /* size w/out border */ + height = stImage->base.Height2; + depth = stImage->base.Depth2; + + assert(width > 0); + assert(height > 0); + assert(depth > 0); + + /* Depending on the image's size, we can't always make a guess here. + */ + if (level > 0) { + if ( (dims >= 1 && width == 1) || + (dims >= 2 && height == 1) || + (dims >= 3 && depth == 1) ) { + /* we can't determine the image size at level=0 */ + stObj->width0 = stObj->height0 = stObj->depth0 = 0; + /* this is not an out of memory error */ + return GL_TRUE; + } + } + + /* grow the image size until we hit level = 0 */ + while (level > 0) { + if (width != 1) + width <<= 1; + if (height != 1) + height <<= 1; + if (depth != 1) + depth <<= 1; + level--; + } + + assert(level == 0); + + /* At this point, (width x height x depth) is the expected size of + * the level=0 mipmap image. + */ + + /* Guess a reasonable value for lastLevel. With OpenGL we have no + * idea how many mipmap levels will be in a texture until we start + * to render with it. Make an educated guess here but be prepared + * to re-allocating a texture buffer with space for more (or fewer) + * mipmap levels later. + */ + if ((stObj->base.MinFilter == GL_NEAREST || + stObj->base.MinFilter == GL_LINEAR || + stImage->base._BaseFormat == GL_DEPTH_COMPONENT || + stImage->base._BaseFormat == GL_DEPTH_STENCIL_EXT) && + !stObj->base.GenerateMipmap && + stImage->level == 0) { + /* only alloc space for a single mipmap level */ + lastLevel = 0; + } + else { + /* alloc space for a full mipmap */ + GLuint l2width = util_logbase2(width); + GLuint l2height = util_logbase2(height); + GLuint l2depth = util_logbase2(depth); + lastLevel = MAX2(MAX2(l2width, l2height), l2depth); + } + + /* Save the level=0 dimensions */ + stObj->width0 = width; + stObj->height0 = height; + stObj->depth0 = depth; + + fmt = st_mesa_format_to_pipe_format(stImage->base.TexFormat); + + bindings = default_bindings(st, fmt); + + stObj->pt = st_texture_create(st, + gl_target_to_pipe(stObj->base.Target), + fmt, + lastLevel, + width, + height, + depth, + bindings); + + DBG("%s returning %d\n", __FUNCTION__, (stObj->pt != NULL)); + + return stObj->pt != NULL; +} + + +/** + * Adjust pixel unpack params and image dimensions to strip off the + * texture border. + * Gallium doesn't support texture borders. They've seldem been used + * and seldom been implemented correctly anyway. + * \param unpackNew returns the new pixel unpack parameters + */ +static void +strip_texture_border(GLint border, + GLint *width, GLint *height, GLint *depth, + const struct gl_pixelstore_attrib *unpack, + struct gl_pixelstore_attrib *unpackNew) +{ + assert(border > 0); /* sanity check */ + + *unpackNew = *unpack; + + if (unpackNew->RowLength == 0) + unpackNew->RowLength = *width; + + if (depth && unpackNew->ImageHeight == 0) + unpackNew->ImageHeight = *height; + + unpackNew->SkipPixels += border; + if (height) + unpackNew->SkipRows += border; + if (depth) + unpackNew->SkipImages += border; + + assert(*width >= 3); + *width = *width - 2 * border; + if (height && *height >= 3) + *height = *height - 2 * border; + if (depth && *depth >= 3) + *depth = *depth - 2 * border; +} + + +/** + * Try to do texture compression via rendering. If the Gallium driver + * can render into a compressed surface this will allow us to do texture + * compression. + * \return GL_TRUE for success, GL_FALSE for failure + */ +static GLboolean +compress_with_blit(struct gl_context * ctx, + GLenum target, GLint level, + GLint xoffset, GLint yoffset, GLint zoffset, + GLint width, GLint height, GLint depth, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *unpack, + struct gl_texture_image *texImage) +{ + const GLuint dstImageOffsets[1] = {0}; + struct st_texture_image *stImage = st_texture_image(texImage); + struct st_context *st = st_context(ctx); + struct pipe_context *pipe = st->pipe; + struct pipe_screen *screen = pipe->screen; + gl_format mesa_format; + struct pipe_resource templ; + struct pipe_resource *src_tex; + struct pipe_sampler_view view_templ; + struct pipe_sampler_view *src_view; + struct pipe_surface *dst_surface, surf_tmpl; + struct pipe_transfer *tex_xfer; + void *map; + + if (!stImage->pt) { + /* XXX: Can this happen? Should we assert? */ + return GL_FALSE; + } + + /* get destination surface (in the compressed texture) */ + memset(&surf_tmpl, 0, sizeof(surf_tmpl)); + surf_tmpl.format = stImage->pt->format; + surf_tmpl.usage = PIPE_BIND_RENDER_TARGET; + surf_tmpl.u.tex.level = stImage->level; + surf_tmpl.u.tex.first_layer = stImage->face; + surf_tmpl.u.tex.last_layer = stImage->face; + dst_surface = pipe->create_surface(pipe, stImage->pt, &surf_tmpl); + if (!dst_surface) { + /* can't render into this format (or other problem) */ + return GL_FALSE; + } + + /* Choose format for the temporary RGBA texture image. + */ + mesa_format = st_ChooseTextureFormat(ctx, GL_RGBA, format, type); + assert(mesa_format); + if (!mesa_format) + return GL_FALSE; + + /* Create the temporary source texture + */ + memset(&templ, 0, sizeof(templ)); + templ.target = st->internal_target; + templ.format = st_mesa_format_to_pipe_format(mesa_format); + templ.width0 = width; + templ.height0 = height; + templ.depth0 = 1; + templ.array_size = 1; + templ.last_level = 0; + templ.usage = PIPE_USAGE_DEFAULT; + templ.bind = PIPE_BIND_SAMPLER_VIEW; + src_tex = screen->resource_create(screen, &templ); + + if (!src_tex) + return GL_FALSE; + + /* Put user's tex data into the temporary texture + */ + tex_xfer = pipe_get_transfer(st_context(ctx)->pipe, src_tex, + 0, 0, /* layer, level are zero */ + PIPE_TRANSFER_WRITE, + 0, 0, width, height); /* x, y, w, h */ + map = pipe_transfer_map(pipe, tex_xfer); + + _mesa_texstore(ctx, 2, GL_RGBA, mesa_format, + map, /* dest ptr */ + 0, 0, 0, /* dest x/y/z offset */ + tex_xfer->stride, /* dest row stride (bytes) */ + dstImageOffsets, /* image offsets (for 3D only) */ + width, height, 1, /* size */ + format, type, /* source format/type */ + pixels, /* source data */ + unpack); /* source data packing */ + + pipe_transfer_unmap(pipe, tex_xfer); + pipe->transfer_destroy(pipe, tex_xfer); + + /* Create temporary sampler view */ + u_sampler_view_default_template(&view_templ, + src_tex, + src_tex->format); + src_view = pipe->create_sampler_view(pipe, src_tex, &view_templ); + + + /* copy / compress image */ + util_blit_pixels_tex(st->blit, + src_view, /* sampler view (src) */ + 0, 0, /* src x0, y0 */ + width, height, /* src x1, y1 */ + dst_surface, /* pipe_surface (dst) */ + xoffset, yoffset, /* dst x0, y0 */ + xoffset + width, /* dst x1 */ + yoffset + height, /* dst y1 */ + 0.0, /* z */ + PIPE_TEX_MIPFILTER_NEAREST); + + pipe_surface_reference(&dst_surface, NULL); + pipe_resource_reference(&src_tex, NULL); + pipe_sampler_view_reference(&src_view, NULL); + + return GL_TRUE; +} + + +/** + * Do glTexImage1/2/3D(). + */ +static void +st_TexImage(struct gl_context * ctx, + GLint dims, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint depth, + GLint border, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *unpack, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage, + GLsizei imageSize, GLboolean compressed_src) +{ + struct st_context *st = st_context(ctx); + struct pipe_screen *screen = st->pipe->screen; + struct st_texture_object *stObj = st_texture_object(texObj); + struct st_texture_image *stImage = st_texture_image(texImage); + GLuint dstRowStride = 0; + struct gl_pixelstore_attrib unpackNB; + enum pipe_transfer_usage transfer_usage = 0; + + DBG("%s target %s level %d %dx%dx%d border %d\n", __FUNCTION__, + _mesa_lookup_enum_by_nr(target), level, width, height, depth, border); + + /* switch to "normal" */ + if (stObj->surface_based) { + gl_format texFormat; + + _mesa_clear_texture_object(ctx, texObj); + pipe_resource_reference(&stObj->pt, NULL); + + /* oops, need to init this image again */ + texFormat = _mesa_choose_texture_format(ctx, texObj, target, level, + internalFormat, format, type); + + _mesa_init_teximage_fields(ctx, target, texImage, + width, height, depth, border, + internalFormat, texFormat); + + stObj->surface_based = GL_FALSE; + } + + /* gallium does not support texture borders, strip it off */ + if (border) { + strip_texture_border(border, &width, &height, &depth, unpack, &unpackNB); + unpack = &unpackNB; + texImage->Width = width; + texImage->Height = height; + texImage->Depth = depth; + texImage->Border = 0; + border = 0; + } + else { + assert(texImage->Width == width); + assert(texImage->Height == height); + assert(texImage->Depth == depth); + } + + stImage->face = _mesa_tex_target_to_face(target); + stImage->level = level; + + _mesa_set_fetch_functions(texImage, dims); + + /* Release the reference to a potentially orphaned buffer. + * Release any old malloced memory. + */ + if (stImage->pt) { + pipe_resource_reference(&stImage->pt, NULL); + assert(!texImage->Data); + } + else if (texImage->Data) { + _mesa_align_free(texImage->Data); + } + + /* + * See if the new image is somehow incompatible with the existing + * mipmap. If so, free the old mipmap. + */ + if (stObj->pt) { + if (level > (GLint) stObj->pt->last_level || + !st_texture_match_image(stObj->pt, &stImage->base, + stImage->face, stImage->level)) { + DBG("release it\n"); + pipe_resource_reference(&stObj->pt, NULL); + assert(!stObj->pt); + pipe_sampler_view_reference(&stObj->sampler_view, NULL); + } + } + + if (width == 0 || height == 0 || depth == 0) { + /* stop after freeing old image */ + return; + } + + if (!stObj->pt) { + if (!guess_and_alloc_texture(st, stObj, stImage)) { + /* Probably out of memory. + * Try flushing any pending rendering, then retry. + */ + st_finish(st); + if (!guess_and_alloc_texture(st, stObj, stImage)) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); + return; + } + } + } + + assert(!stImage->pt); + + /* Check if this texture image can live inside the texture object's buffer. + * If so, store the image there. Otherwise the image will temporarily live + * in its own buffer. + */ + if (stObj->pt && + st_texture_match_image(stObj->pt, &stImage->base, + stImage->face, stImage->level)) { + + pipe_resource_reference(&stImage->pt, stObj->pt); + assert(stImage->pt); + } + + if (!stImage->pt) + DBG("XXX: Image did not fit into texture - storing in local memory!\n"); + + /* Pixel data may come from regular user memory or a PBO. For the later, + * do bounds checking and map the PBO to read pixels data from it. + * + * XXX we should try to use a GPU-accelerated path to copy the image data + * from the PBO to the texture. + */ + if (compressed_src) { + pixels = _mesa_validate_pbo_compressed_teximage(ctx, imageSize, pixels, + unpack, + "glCompressedTexImage"); + } + else { + pixels = _mesa_validate_pbo_teximage(ctx, dims, width, height, 1, + format, type, + pixels, unpack, "glTexImage"); + } + + /* See if we can do texture compression with a blit/render. + */ + if (!compressed_src && + !ctx->Mesa_DXTn && + _mesa_is_format_compressed(texImage->TexFormat) && + screen->is_format_supported(screen, + stImage->pt->format, + stImage->pt->target, 0, + PIPE_BIND_RENDER_TARGET, 0)) { + if (!pixels) + goto done; + + if (compress_with_blit(ctx, target, level, 0, 0, 0, width, height, depth, + format, type, pixels, unpack, texImage)) { + goto done; + } + } + + /* + * Prepare to store the texture data. Either map the gallium texture buffer + * memory or malloc space for it. + */ + if (stImage->pt) { + /* Store the image in the gallium texture memory buffer */ + if (format == GL_DEPTH_COMPONENT && + util_format_is_depth_and_stencil(stImage->pt->format)) + transfer_usage = PIPE_TRANSFER_READ_WRITE; + else + transfer_usage = PIPE_TRANSFER_WRITE; + + texImage->Data = st_texture_image_map(st, stImage, 0, + transfer_usage, 0, 0, width, height); + if(stImage->transfer) + dstRowStride = stImage->transfer->stride; + } + else { + /* Allocate regular memory and store the image there temporarily. */ + GLuint imageSize = _mesa_format_image_size(texImage->TexFormat, + width, height, depth); + dstRowStride = _mesa_format_row_stride(texImage->TexFormat, width); + + texImage->Data = _mesa_align_malloc(imageSize, 16); + } + + if (!texImage->Data) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); + return; + } + + if (!pixels) { + /* We've allocated texture memory, but have no pixel data - all done. */ + goto done; + } + + DBG("Upload image %dx%dx%d row_len %x pitch %x\n", + width, height, depth, width, dstRowStride); + + /* Copy user texture image into the texture buffer. + */ + if (compressed_src) { + const GLuint srcRowStride = + _mesa_format_row_stride(texImage->TexFormat, width); + if (dstRowStride == srcRowStride) { + memcpy(texImage->Data, pixels, imageSize); + } + else { + char *dst = texImage->Data; + const char *src = pixels; + GLuint i, bw, bh, lines; + _mesa_get_format_block_size(texImage->TexFormat, &bw, &bh); + lines = (height + bh - 1) / bh; + + for (i = 0; i < lines; ++i) { + memcpy(dst, src, srcRowStride); + dst += dstRowStride; + src += srcRowStride; + } + } + } + else { + const GLuint srcImageStride = + _mesa_image_image_stride(unpack, width, height, format, type); + GLint i; + const GLubyte *src = (const GLubyte *) pixels; + + for (i = 0; i < depth; i++) { + if (!_mesa_texstore(ctx, dims, + texImage->_BaseFormat, + texImage->TexFormat, + texImage->Data, + 0, 0, 0, /* dstX/Y/Zoffset */ + dstRowStride, + texImage->ImageOffsets, + width, height, 1, + format, type, src, unpack)) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); + } + + if (stImage->pt && i + 1 < depth) { + /* unmap this slice */ + st_texture_image_unmap(st, stImage); + /* map next slice of 3D texture */ + texImage->Data = st_texture_image_map(st, stImage, i + 1, + transfer_usage, 0, 0, + width, height); + src += srcImageStride; + } + } + } + +done: + _mesa_unmap_teximage_pbo(ctx, unpack); + + if (stImage->pt && texImage->Data) { + st_texture_image_unmap(st, stImage); + texImage->Data = NULL; + } +} + + +static void +st_TexImage3D(struct gl_context * ctx, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint depth, + GLint border, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *unpack, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + st_TexImage(ctx, 3, target, level, internalFormat, width, height, depth, + border, format, type, pixels, unpack, texObj, texImage, + 0, GL_FALSE); +} + + +static void +st_TexImage2D(struct gl_context * ctx, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint border, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *unpack, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + st_TexImage(ctx, 2, target, level, internalFormat, width, height, 1, border, + format, type, pixels, unpack, texObj, texImage, 0, GL_FALSE); +} + + +static void +st_TexImage1D(struct gl_context * ctx, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint border, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *unpack, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + st_TexImage(ctx, 1, target, level, internalFormat, width, 1, 1, border, + format, type, pixels, unpack, texObj, texImage, 0, GL_FALSE); +} + + +static void +st_CompressedTexImage2D(struct gl_context *ctx, GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint border, + GLsizei imageSize, const GLvoid *data, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + st_TexImage(ctx, 2, target, level, internalFormat, width, height, 1, border, + 0, 0, data, &ctx->Unpack, texObj, texImage, imageSize, GL_TRUE); +} + + + +/** + * glGetTexImage() helper: decompress a compressed texture by rendering + * a textured quad. Store the results in the user's buffer. + */ +static void +decompress_with_blit(struct gl_context * ctx, GLenum target, GLint level, + GLenum format, GLenum type, GLvoid *pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + struct st_context *st = st_context(ctx); + struct pipe_context *pipe = st->pipe; + struct st_texture_image *stImage = st_texture_image(texImage); + struct st_texture_object *stObj = st_texture_object(texObj); + struct pipe_sampler_view *src_view = + st_get_texture_sampler_view(stObj, pipe); + const GLuint width = texImage->Width; + const GLuint height = texImage->Height; + struct pipe_surface *dst_surface; + struct pipe_resource *dst_texture; + struct pipe_transfer *tex_xfer; + unsigned bind = (PIPE_BIND_RENDER_TARGET | /* util_blit may choose to render */ + PIPE_BIND_TRANSFER_READ); + + /* create temp / dest surface */ + if (!util_create_rgba_surface(pipe, width, height, bind, + &dst_texture, &dst_surface)) { + _mesa_problem(ctx, "util_create_rgba_surface() failed " + "in decompress_with_blit()"); + return; + } + + /* blit/render/decompress */ + util_blit_pixels_tex(st->blit, + src_view, /* pipe_resource (src) */ + 0, 0, /* src x0, y0 */ + width, height, /* src x1, y1 */ + dst_surface, /* pipe_surface (dst) */ + 0, 0, /* dst x0, y0 */ + width, height, /* dst x1, y1 */ + 0.0, /* z */ + PIPE_TEX_MIPFILTER_NEAREST); + + /* map the dst_surface so we can read from it */ + tex_xfer = pipe_get_transfer(st_context(ctx)->pipe, + dst_texture, 0, 0, + PIPE_TRANSFER_READ, + 0, 0, width, height); + + pixels = _mesa_map_pbo_dest(ctx, &ctx->Pack, pixels); + + /* copy/pack data into user buffer */ + if (st_equal_formats(stImage->pt->format, format, type)) { + /* memcpy */ + const uint bytesPerRow = width * util_format_get_blocksize(stImage->pt->format); + ubyte *map = pipe_transfer_map(pipe, tex_xfer); + GLuint row; + for (row = 0; row < height; row++) { + GLvoid *dest = _mesa_image_address2d(&ctx->Pack, pixels, width, + height, format, type, row, 0); + memcpy(dest, map, bytesPerRow); + map += tex_xfer->stride; + } + pipe_transfer_unmap(pipe, tex_xfer); + } + else { + /* format translation via floats */ + GLuint row; + enum pipe_format format = util_format_linear(dst_texture->format); + for (row = 0; row < height; row++) { + const GLbitfield transferOps = 0x0; /* bypassed for glGetTexImage() */ + GLfloat rgba[4 * MAX_WIDTH]; + GLvoid *dest = _mesa_image_address2d(&ctx->Pack, pixels, width, + height, format, type, row, 0); + + if (ST_DEBUG & DEBUG_FALLBACK) + debug_printf("%s: fallback format translation\n", __FUNCTION__); + + /* get float[4] rgba row from surface */ + pipe_get_tile_rgba_format(pipe, tex_xfer, 0, row, width, 1, + format, rgba); + + _mesa_pack_rgba_span_float(ctx, width, (GLfloat (*)[4]) rgba, format, + type, dest, &ctx->Pack, transferOps); + } + } + + _mesa_unmap_pbo_dest(ctx, &ctx->Pack); + + pipe->transfer_destroy(pipe, tex_xfer); + + /* destroy the temp / dest surface */ + util_destroy_rgba_surface(dst_texture, dst_surface); +} + + + +/** + * Need to map texture image into memory before copying image data, + * then unmap it. + */ +static void +st_get_tex_image(struct gl_context * ctx, GLenum target, GLint level, + GLenum format, GLenum type, GLvoid * pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage, GLboolean compressed_dst) +{ + struct st_context *st = st_context(ctx); + struct st_texture_image *stImage = st_texture_image(texImage); + const GLuint dstImageStride = + _mesa_image_image_stride(&ctx->Pack, texImage->Width, texImage->Height, + format, type); + GLuint depth, i; + GLubyte *dest; + + if (stImage->pt && + util_format_is_s3tc(stImage->pt->format) && + !compressed_dst) { + /* Need to decompress the texture. + * We'll do this by rendering a textured quad. + * Note that we only expect RGBA formats (no Z/depth formats). + */ + decompress_with_blit(ctx, target, level, format, type, pixels, + texObj, texImage); + return; + } + + /* Map */ + if (stImage->pt) { + /* Image is stored in hardware format in a buffer managed by the + * kernel. Need to explicitly map and unmap it. + */ + texImage->Data = st_texture_image_map(st, stImage, 0, + PIPE_TRANSFER_READ, 0, 0, + stImage->base.Width, + stImage->base.Height); + /* compute stride in texels from stride in bytes */ + texImage->RowStride = stImage->transfer->stride + * util_format_get_blockwidth(stImage->pt->format) + / util_format_get_blocksize(stImage->pt->format); + } + else { + /* Otherwise, the image should actually be stored in + * texImage->Data. This is pretty confusing for + * everybody, I'd much prefer to separate the two functions of + * texImage->Data - storage for texture images in main memory + * and access (ie mappings) of images. In other words, we'd + * create a new texImage->Map field and leave Data simply for + * storage. + */ + assert(texImage->Data); + } + + depth = texImage->Depth; + texImage->Depth = 1; + + dest = (GLubyte *) pixels; + + _mesa_set_fetch_functions(texImage, get_texture_dims(target)); + + for (i = 0; i < depth; i++) { + if (compressed_dst) { + _mesa_get_compressed_teximage(ctx, target, level, dest, + texObj, texImage); + } + else { + _mesa_get_teximage(ctx, target, level, format, type, dest, + texObj, texImage); + } + + if (stImage->pt && i + 1 < depth) { + /* unmap this slice */ + st_texture_image_unmap(st, stImage); + /* map next slice of 3D texture */ + texImage->Data = st_texture_image_map(st, stImage, i + 1, + PIPE_TRANSFER_READ, 0, 0, + stImage->base.Width, + stImage->base.Height); + dest += dstImageStride; + } + } + + texImage->Depth = depth; + + /* Unmap */ + if (stImage->pt) { + st_texture_image_unmap(st, stImage); + texImage->Data = NULL; + } +} + + +static void +st_GetTexImage(struct gl_context * ctx, GLenum target, GLint level, + GLenum format, GLenum type, GLvoid * pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + st_get_tex_image(ctx, target, level, format, type, pixels, texObj, texImage, + GL_FALSE); +} + + +static void +st_GetCompressedTexImage(struct gl_context *ctx, GLenum target, GLint level, + GLvoid *pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + st_get_tex_image(ctx, target, level, 0, 0, pixels, texObj, texImage, + GL_TRUE); +} + + + +static void +st_TexSubimage(struct gl_context *ctx, GLint dims, GLenum target, GLint level, + GLint xoffset, GLint yoffset, GLint zoffset, + GLint width, GLint height, GLint depth, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + struct st_context *st = st_context(ctx); + struct pipe_screen *screen = st->pipe->screen; + struct st_texture_image *stImage = st_texture_image(texImage); + GLuint dstRowStride; + const GLuint srcImageStride = + _mesa_image_image_stride(packing, width, height, format, type); + GLint i; + const GLubyte *src; + /* init to silence warning only: */ + enum pipe_transfer_usage transfer_usage = PIPE_TRANSFER_WRITE; + + DBG("%s target %s level %d offset %d,%d %dx%d\n", __FUNCTION__, + _mesa_lookup_enum_by_nr(target), + level, xoffset, yoffset, width, height); + + pixels = + _mesa_validate_pbo_teximage(ctx, dims, width, height, depth, format, + type, pixels, packing, "glTexSubImage2D"); + if (!pixels) + return; + + /* See if we can do texture compression with a blit/render. + */ + if (!ctx->Mesa_DXTn && + _mesa_is_format_compressed(texImage->TexFormat) && + screen->is_format_supported(screen, + stImage->pt->format, + stImage->pt->target, 0, + PIPE_BIND_RENDER_TARGET, 0)) { + if (compress_with_blit(ctx, target, level, + xoffset, yoffset, zoffset, + width, height, depth, + format, type, pixels, packing, texImage)) { + goto done; + } + } + + /* Map buffer if necessary. Need to lock to prevent other contexts + * from uploading the buffer under us. + */ + if (stImage->pt) { + if (format == GL_DEPTH_COMPONENT && + util_format_is_depth_and_stencil(stImage->pt->format)) + transfer_usage = PIPE_TRANSFER_READ_WRITE; + else + transfer_usage = PIPE_TRANSFER_WRITE; + + texImage->Data = st_texture_image_map(st, stImage, zoffset, + transfer_usage, + xoffset, yoffset, + width, height); + } + + if (!texImage->Data) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage"); + goto done; + } + + src = (const GLubyte *) pixels; + dstRowStride = stImage->transfer->stride; + + for (i = 0; i < depth; i++) { + if (!_mesa_texstore(ctx, dims, texImage->_BaseFormat, + texImage->TexFormat, + texImage->Data, + 0, 0, 0, + dstRowStride, + texImage->ImageOffsets, + width, height, 1, + format, type, src, packing)) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage"); + } + + if (stImage->pt && i + 1 < depth) { + /* unmap this slice */ + st_texture_image_unmap(st, stImage); + /* map next slice of 3D texture */ + texImage->Data = st_texture_image_map(st, stImage, + zoffset + i + 1, + transfer_usage, + xoffset, yoffset, + width, height); + src += srcImageStride; + } + } + +done: + _mesa_unmap_teximage_pbo(ctx, packing); + + if (stImage->pt && texImage->Data) { + st_texture_image_unmap(st, stImage); + texImage->Data = NULL; + } +} + + + +static void +st_TexSubImage3D(struct gl_context *ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, GLint zoffset, + GLsizei width, GLsizei height, GLsizei depth, + GLenum format, GLenum type, const GLvoid *pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + st_TexSubimage(ctx, 3, target, level, xoffset, yoffset, zoffset, + width, height, depth, format, type, + pixels, packing, texObj, texImage); +} + + +static void +st_TexSubImage2D(struct gl_context *ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, + GLsizei width, GLsizei height, + GLenum format, GLenum type, const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + st_TexSubimage(ctx, 2, target, level, xoffset, yoffset, 0, + width, height, 1, format, type, + pixels, packing, texObj, texImage); +} + + +static void +st_TexSubImage1D(struct gl_context *ctx, GLenum target, GLint level, + GLint xoffset, GLsizei width, GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + st_TexSubimage(ctx, 1, target, level, xoffset, 0, 0, width, 1, 1, + format, type, pixels, packing, texObj, texImage); +} + + +static void +st_CompressedTexSubImage1D(struct gl_context *ctx, GLenum target, GLint level, + GLint xoffset, GLsizei width, + GLenum format, + GLsizei imageSize, const GLvoid *data, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + assert(0); +} + + +static void +st_CompressedTexSubImage2D(struct gl_context *ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, + GLsizei width, GLint height, + GLenum format, + GLsizei imageSize, const GLvoid *data, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + struct st_context *st = st_context(ctx); + struct st_texture_image *stImage = st_texture_image(texImage); + int srcBlockStride; + int dstBlockStride; + int y; + enum pipe_format pformat; + + if (stImage->pt) { + pformat = stImage->pt->format; + + texImage->Data = st_texture_image_map(st, stImage, 0, + PIPE_TRANSFER_WRITE, + xoffset, yoffset, + width, height); + + srcBlockStride = util_format_get_stride(pformat, width); + dstBlockStride = stImage->transfer->stride; + } else { + assert(stImage->pt); + /* TODO find good values for block and strides */ + /* TODO also adjust texImage->data for yoffset/xoffset */ + return; + } + + if (!texImage->Data) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexSubImage"); + return; + } + + assert(xoffset % util_format_get_blockwidth(pformat) == 0); + assert(yoffset % util_format_get_blockheight(pformat) == 0); + + for (y = 0; y < height; y += util_format_get_blockheight(pformat)) { + /* don't need to adjust for xoffset and yoffset as st_texture_image_map does that */ + const char *src = (const char*)data + srcBlockStride * util_format_get_nblocksy(pformat, y); + char *dst = (char*)texImage->Data + dstBlockStride * util_format_get_nblocksy(pformat, y); + memcpy(dst, src, util_format_get_stride(pformat, width)); + } + + if (stImage->pt) { + st_texture_image_unmap(st, stImage); + texImage->Data = NULL; + } +} + + +static void +st_CompressedTexSubImage3D(struct gl_context *ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, GLint zoffset, + GLsizei width, GLint height, GLint depth, + GLenum format, + GLsizei imageSize, const GLvoid *data, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + assert(0); +} + + + +/** + * Do a CopyTexSubImage operation using a read transfer from the source, + * a write transfer to the destination and get_tile()/put_tile() to access + * the pixels/texels. + * + * Note: srcY=0=TOP of renderbuffer + */ +static void +fallback_copy_texsubimage(struct gl_context *ctx, GLenum target, GLint level, + struct st_renderbuffer *strb, + struct st_texture_image *stImage, + GLenum baseFormat, + GLint destX, GLint destY, GLint destZ, + GLint srcX, GLint srcY, + GLsizei width, GLsizei height) +{ + struct st_context *st = st_context(ctx); + struct pipe_context *pipe = st->pipe; + struct pipe_transfer *src_trans; + GLvoid *texDest; + enum pipe_transfer_usage transfer_usage; + + if (ST_DEBUG & DEBUG_FALLBACK) + debug_printf("%s: fallback processing\n", __FUNCTION__); + + assert(width <= MAX_WIDTH); + + if (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP) { + srcY = strb->Base.Height - srcY - height; + } + + src_trans = pipe_get_transfer(st_context(ctx)->pipe, + strb->texture, + 0, 0, + PIPE_TRANSFER_READ, + srcX, srcY, + width, height); + + if ((baseFormat == GL_DEPTH_COMPONENT || + baseFormat == GL_DEPTH_STENCIL) && + util_format_is_depth_and_stencil(stImage->pt->format)) + transfer_usage = PIPE_TRANSFER_READ_WRITE; + else + transfer_usage = PIPE_TRANSFER_WRITE; + + /* XXX this used to ignore destZ param */ + texDest = st_texture_image_map(st, stImage, destZ, transfer_usage, + destX, destY, width, height); + + if (baseFormat == GL_DEPTH_COMPONENT || + baseFormat == GL_DEPTH_STENCIL) { + const GLboolean scaleOrBias = (ctx->Pixel.DepthScale != 1.0F || + ctx->Pixel.DepthBias != 0.0F); + GLint row, yStep; + + /* determine bottom-to-top vs. top-to-bottom order for src buffer */ + if (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP) { + srcY = height - 1; + yStep = -1; + } + else { + srcY = 0; + yStep = 1; + } + + /* To avoid a large temp memory allocation, do copy row by row */ + for (row = 0; row < height; row++, srcY += yStep) { + uint data[MAX_WIDTH]; + pipe_get_tile_z(pipe, src_trans, 0, srcY, width, 1, data); + if (scaleOrBias) { + _mesa_scale_and_bias_depth_uint(ctx, width, data); + } + pipe_put_tile_z(pipe, stImage->transfer, 0, row, width, 1, data); + } + } + else { + /* RGBA format */ + GLfloat *tempSrc = + (GLfloat *) malloc(width * height * 4 * sizeof(GLfloat)); + + if (tempSrc && texDest) { + const GLint dims = 2; + const GLint dstRowStride = stImage->transfer->stride; + struct gl_texture_image *texImage = &stImage->base; + struct gl_pixelstore_attrib unpack = ctx->DefaultPacking; + + if (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP) { + unpack.Invert = GL_TRUE; + } + + /* get float/RGBA image from framebuffer */ + /* XXX this usually involves a lot of int/float conversion. + * try to avoid that someday. + */ + pipe_get_tile_rgba_format(pipe, src_trans, 0, 0, width, height, + util_format_linear(strb->texture->format), + tempSrc); + + /* Store into texture memory. + * Note that this does some special things such as pixel transfer + * ops and format conversion. In particular, if the dest tex format + * is actually RGBA but the user created the texture as GL_RGB we + * need to fill-in/override the alpha channel with 1.0. + */ + _mesa_texstore(ctx, dims, + texImage->_BaseFormat, + texImage->TexFormat, + texDest, + 0, 0, 0, + dstRowStride, + texImage->ImageOffsets, + width, height, 1, + GL_RGBA, GL_FLOAT, tempSrc, /* src */ + &unpack); + } + else { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage"); + } + + if (tempSrc) + free(tempSrc); + } + + st_texture_image_unmap(st, stImage); + pipe->transfer_destroy(pipe, src_trans); +} + + + +/** + * If the format of the src renderbuffer and the format of the dest + * texture are compatible (in terms of blitting), return a TGSI writemask + * to be used during the blit. + * If the src/dest are incompatible, return 0. + */ +static unsigned +compatible_src_dst_formats(struct gl_context *ctx, + const struct gl_renderbuffer *src, + const struct gl_texture_image *dst) +{ + /* Get logical base formats for the src and dest. + * That is, use the user-requested formats and not the actual, device- + * chosen formats. + * For example, the user may have requested an A8 texture but the + * driver may actually be using an RGBA texture format. When we + * copy/blit to that texture, we only want to copy the Alpha channel + * and not the RGB channels. + * + * Similarly, when the src FBO was created an RGB format may have been + * requested but the driver actually chose an RGBA format. In that case, + * we don't want to copy the undefined Alpha channel to the dest texture + * (it should be 1.0). + */ + const GLenum srcFormat = _mesa_base_fbo_format(ctx, src->InternalFormat); + const GLenum dstFormat = _mesa_base_tex_format(ctx, dst->InternalFormat); + + /** + * XXX when we have red-only and red/green renderbuffers we'll need + * to add more cases here (or implement a general-purpose routine that + * queries the existance of the R,G,B,A channels in the src and dest). + */ + if (srcFormat == dstFormat) { + /* This is the same as matching_base_formats, which should + * always pass, as it did previously. + */ + return TGSI_WRITEMASK_XYZW; + } + else if (srcFormat == GL_RGB && dstFormat == GL_RGBA) { + /* Make sure that A in the dest is 1. The actual src format + * may be RGBA and have undefined A values. + */ + return TGSI_WRITEMASK_XYZ; + } + else if (srcFormat == GL_RGBA && dstFormat == GL_RGB) { + /* Make sure that A in the dest is 1. The actual dst format + * may be RGBA and will need A=1 to provide proper alpha values + * when sampled later. + */ + return TGSI_WRITEMASK_XYZ; + } + else { + if (ST_DEBUG & DEBUG_FALLBACK) + debug_printf("%s failed for src %s, dst %s\n", + __FUNCTION__, + _mesa_lookup_enum_by_nr(srcFormat), + _mesa_lookup_enum_by_nr(dstFormat)); + + /* Otherwise fail. + */ + return 0; + } +} + + + +/** + * Do a CopyTex[Sub]Image1/2/3D() using a hardware (blit) path if possible. + * Note that the region to copy has already been clipped so we know we + * won't read from outside the source renderbuffer's bounds. + * + * Note: srcY=0=Bottom of renderbuffer (GL convention) + */ +static void +st_copy_texsubimage(struct gl_context *ctx, + GLenum target, GLint level, + GLint destX, GLint destY, GLint destZ, + GLint srcX, GLint srcY, + GLsizei width, GLsizei height) +{ + struct gl_texture_unit *texUnit = + &ctx->Texture.Unit[ctx->Texture.CurrentUnit]; + struct gl_texture_object *texObj = + _mesa_select_tex_object(ctx, texUnit, target); + struct gl_texture_image *texImage = + _mesa_select_tex_image(ctx, texObj, target, level); + struct st_texture_image *stImage = st_texture_image(texImage); + const GLenum texBaseFormat = texImage->_BaseFormat; + struct gl_framebuffer *fb = ctx->ReadBuffer; + struct st_renderbuffer *strb; + struct st_context *st = st_context(ctx); + struct pipe_context *pipe = st->pipe; + struct pipe_screen *screen = pipe->screen; + enum pipe_format dest_format, src_format; + GLboolean use_fallback = GL_TRUE; + GLboolean matching_base_formats; + GLuint format_writemask, sample_count; + struct pipe_surface *dest_surface = NULL; + GLboolean do_flip = (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP); + + /* make sure finalize_textures has been called? + */ + if (0) st_validate_state(st); + + /* determine if copying depth or color data */ + if (texBaseFormat == GL_DEPTH_COMPONENT || + texBaseFormat == GL_DEPTH_STENCIL) { + strb = st_renderbuffer(fb->_DepthBuffer); + if (strb->Base.Wrapped) { + strb = st_renderbuffer(strb->Base.Wrapped); + } + } + else { + /* texBaseFormat == GL_RGB, GL_RGBA, GL_ALPHA, etc */ + strb = st_renderbuffer(fb->_ColorReadBuffer); + } + + if (!strb || !strb->surface || !stImage->pt) { + debug_printf("%s: null strb or stImage\n", __FUNCTION__); + return; + } + + sample_count = strb->surface->texture->nr_samples; + /* I believe this would be legal, presumably would need to do a resolve + for color, and for depth/stencil spec says to just use one of the + depth/stencil samples per pixel? Need some transfer clarifications. */ + assert(sample_count < 2); + + if (srcX < 0) { + width -= -srcX; + destX += -srcX; + srcX = 0; + } + + if (srcY < 0) { + height -= -srcY; + destY += -srcY; + srcY = 0; + } + + if (destX < 0) { + width -= -destX; + srcX += -destX; + destX = 0; + } + + if (destY < 0) { + height -= -destY; + srcY += -destY; + destY = 0; + } + + if (width < 0 || height < 0) + return; + + + assert(strb); + assert(strb->surface); + assert(stImage->pt); + + src_format = strb->surface->format; + dest_format = stImage->pt->format; + + /* + * Determine if the src framebuffer and dest texture have the same + * base format. We need this to detect a case such as the framebuffer + * being GL_RGBA but the texture being GL_RGB. If the actual hardware + * texture format stores RGBA we need to set A=1 (overriding the + * framebuffer's alpha values). We can't do that with the blit or + * textured-quad paths. + */ + matching_base_formats = + (_mesa_get_format_base_format(strb->Base.Format) == + _mesa_get_format_base_format(texImage->TexFormat)); + format_writemask = compatible_src_dst_formats(ctx, &strb->Base, texImage); + + if (ctx->_ImageTransferState == 0x0) { + + if (matching_base_formats && + src_format == dest_format && + !do_flip) + { + /* use surface_copy() / blit */ + struct pipe_box src_box; + u_box_2d_zslice(srcX, srcY, strb->surface->u.tex.first_layer, + width, height, &src_box); + + /* for resource_copy_region(), y=0=top, always */ + pipe->resource_copy_region(pipe, + /* dest */ + stImage->pt, + stImage->level, + destX, destY, destZ + stImage->face, + /* src */ + strb->texture, + strb->surface->u.tex.level, + &src_box); + use_fallback = GL_FALSE; + } + else if (format_writemask && + texBaseFormat != GL_DEPTH_COMPONENT && + texBaseFormat != GL_DEPTH_STENCIL && + screen->is_format_supported(screen, src_format, + PIPE_TEXTURE_2D, sample_count, + PIPE_BIND_SAMPLER_VIEW, + 0) && + screen->is_format_supported(screen, dest_format, + PIPE_TEXTURE_2D, 0, + PIPE_BIND_RENDER_TARGET, + 0)) { + /* draw textured quad to do the copy */ + GLint srcY0, srcY1; + struct pipe_surface surf_tmpl; + memset(&surf_tmpl, 0, sizeof(surf_tmpl)); + surf_tmpl.format = stImage->pt->format; + surf_tmpl.usage = PIPE_BIND_RENDER_TARGET; + surf_tmpl.u.tex.level = stImage->level; + surf_tmpl.u.tex.first_layer = stImage->face + destZ; + surf_tmpl.u.tex.last_layer = stImage->face + destZ; + + dest_surface = pipe->create_surface(pipe, stImage->pt, + &surf_tmpl); + + if (do_flip) { + srcY1 = strb->Base.Height - srcY - height; + srcY0 = srcY1 + height; + } + else { + srcY0 = srcY; + srcY1 = srcY0 + height; + } + + util_blit_pixels_writemask(st->blit, + strb->texture, + strb->surface->u.tex.level, + srcX, srcY0, + srcX + width, srcY1, + strb->surface->u.tex.first_layer, + dest_surface, + destX, destY, + destX + width, destY + height, + 0.0, PIPE_TEX_MIPFILTER_NEAREST, + format_writemask); + use_fallback = GL_FALSE; + } + + if (dest_surface) + pipe_surface_reference(&dest_surface, NULL); + } + + if (use_fallback) { + /* software fallback */ + fallback_copy_texsubimage(ctx, target, level, + strb, stImage, texBaseFormat, + destX, destY, destZ, + srcX, srcY, width, height); + } +} + + + +static void +st_CopyTexImage1D(struct gl_context * ctx, GLenum target, GLint level, + GLenum internalFormat, + GLint x, GLint y, GLsizei width, GLint border) +{ + struct gl_texture_unit *texUnit = + &ctx->Texture.Unit[ctx->Texture.CurrentUnit]; + struct gl_texture_object *texObj = + _mesa_select_tex_object(ctx, texUnit, target); + struct gl_texture_image *texImage = + _mesa_select_tex_image(ctx, texObj, target, level); + + /* Setup or redefine the texture object, texture and texture + * image. Don't populate yet. + */ + ctx->Driver.TexImage1D(ctx, target, level, internalFormat, + width, border, + GL_RGBA, CHAN_TYPE, NULL, + &ctx->DefaultPacking, texObj, texImage); + + st_copy_texsubimage(ctx, target, level, + 0, 0, 0, /* destX,Y,Z */ + x, y, width, 1); /* src X, Y, size */ +} + + +static void +st_CopyTexImage2D(struct gl_context * ctx, GLenum target, GLint level, + GLenum internalFormat, + GLint x, GLint y, GLsizei width, GLsizei height, + GLint border) +{ + struct gl_texture_unit *texUnit = + &ctx->Texture.Unit[ctx->Texture.CurrentUnit]; + struct gl_texture_object *texObj = + _mesa_select_tex_object(ctx, texUnit, target); + struct gl_texture_image *texImage = + _mesa_select_tex_image(ctx, texObj, target, level); + + /* Setup or redefine the texture object, texture and texture + * image. Don't populate yet. + */ + ctx->Driver.TexImage2D(ctx, target, level, internalFormat, + width, height, border, + GL_RGBA, CHAN_TYPE, NULL, + &ctx->DefaultPacking, texObj, texImage); + + st_copy_texsubimage(ctx, target, level, + 0, 0, 0, /* destX,Y,Z */ + x, y, width, height); /* src X, Y, size */ +} + + +static void +st_CopyTexSubImage1D(struct gl_context * ctx, GLenum target, GLint level, + GLint xoffset, GLint x, GLint y, GLsizei width) +{ + const GLint yoffset = 0, zoffset = 0; + const GLsizei height = 1; + st_copy_texsubimage(ctx, target, level, + xoffset, yoffset, zoffset, /* destX,Y,Z */ + x, y, width, height); /* src X, Y, size */ +} + + +static void +st_CopyTexSubImage2D(struct gl_context * ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, + GLint x, GLint y, GLsizei width, GLsizei height) +{ + const GLint zoffset = 0; + st_copy_texsubimage(ctx, target, level, + xoffset, yoffset, zoffset, /* destX,Y,Z */ + x, y, width, height); /* src X, Y, size */ +} + + +static void +st_CopyTexSubImage3D(struct gl_context * ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, GLint zoffset, + GLint x, GLint y, GLsizei width, GLsizei height) +{ + st_copy_texsubimage(ctx, target, level, + xoffset, yoffset, zoffset, /* destX,Y,Z */ + x, y, width, height); /* src X, Y, size */ +} + + +/** + * Copy image data from stImage into the texture object 'stObj' at level + * 'dstLevel'. + */ +static void +copy_image_data_to_texture(struct st_context *st, + struct st_texture_object *stObj, + GLuint dstLevel, + struct st_texture_image *stImage) +{ + /* debug checks */ + { + const struct gl_texture_image *dstImage = + stObj->base.Image[stImage->face][stImage->level]; + assert(dstImage); + assert(dstImage->Width == stImage->base.Width); + assert(dstImage->Height == stImage->base.Height); + assert(dstImage->Depth == stImage->base.Depth); + } + + if (stImage->pt) { + /* Copy potentially with the blitter: + */ + st_texture_image_copy(st->pipe, + stObj->pt, dstLevel, /* dest texture, level */ + stImage->pt, stImage->level, /* src texture, level */ + stImage->face); + + pipe_resource_reference(&stImage->pt, NULL); + } + else if (stImage->base.Data) { + st_texture_image_data(st, + stObj->pt, + stImage->face, + dstLevel, + stImage->base.Data, + stImage->base.RowStride * + util_format_get_blocksize(stObj->pt->format), + stImage->base.RowStride * + stImage->base.Height * + util_format_get_blocksize(stObj->pt->format)); + _mesa_align_free(stImage->base.Data); + stImage->base.Data = NULL; + } + + pipe_resource_reference(&stImage->pt, stObj->pt); +} + + +/** + * Called during state validation. When this function is finished, + * the texture object should be ready for rendering. + * \return GL_TRUE for success, GL_FALSE for failure (out of mem) + */ +GLboolean +st_finalize_texture(struct gl_context *ctx, + struct pipe_context *pipe, + struct gl_texture_object *tObj) +{ + struct st_context *st = st_context(ctx); + struct st_texture_object *stObj = st_texture_object(tObj); + const GLuint nr_faces = (stObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1; + GLuint face; + struct st_texture_image *firstImage; + enum pipe_format firstImageFormat; + + if (stObj->base._Complete) { + /* The texture is complete and we know exactly how many mipmap levels + * are present/needed. This is conditional because we may be called + * from the st_generate_mipmap() function when the texture object is + * incomplete. In that case, we'll have set stObj->lastLevel before + * we get here. + */ + if (stObj->base.MinFilter == GL_LINEAR || + stObj->base.MinFilter == GL_NEAREST) + stObj->lastLevel = stObj->base.BaseLevel; + else + stObj->lastLevel = stObj->base._MaxLevel; + } + + firstImage = st_texture_image(stObj->base.Image[0][stObj->base.BaseLevel]); + assert(firstImage); + + /* If both firstImage and stObj point to a texture which can contain + * all active images, favour firstImage. Note that because of the + * completeness requirement, we know that the image dimensions + * will match. + */ + if (firstImage->pt && + firstImage->pt != stObj->pt && + (!stObj->pt || firstImage->pt->last_level >= stObj->pt->last_level)) { + pipe_resource_reference(&stObj->pt, firstImage->pt); + pipe_sampler_view_reference(&stObj->sampler_view, NULL); + } + + /* Find gallium format for the Mesa texture */ + firstImageFormat = st_mesa_format_to_pipe_format(firstImage->base.TexFormat); + + /* If we already have a gallium texture, check that it matches the texture + * object's format, target, size, num_levels, etc. + */ + if (stObj->pt) { + if (stObj->pt->target != gl_target_to_pipe(stObj->base.Target) || + !st_sampler_compat_formats(stObj->pt->format, firstImageFormat) || + stObj->pt->last_level < stObj->lastLevel || + stObj->pt->width0 != stObj->width0 || + stObj->pt->height0 != stObj->height0 || + stObj->pt->depth0 != stObj->depth0) + { + /* The gallium texture does not match the Mesa texture so delete the + * gallium texture now. We'll make a new one below. + */ + pipe_resource_reference(&stObj->pt, NULL); + pipe_sampler_view_reference(&stObj->sampler_view, NULL); + st->dirty.st |= ST_NEW_FRAMEBUFFER; + } + } + + /* May need to create a new gallium texture: + */ + if (!stObj->pt) { + GLuint bindings = default_bindings(st, firstImageFormat); + + stObj->pt = st_texture_create(st, + gl_target_to_pipe(stObj->base.Target), + firstImageFormat, + stObj->lastLevel, + stObj->width0, + stObj->height0, + stObj->depth0, + bindings); + + if (!stObj->pt) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); + return GL_FALSE; + } + } + + /* Pull in any images not in the object's texture: + */ + for (face = 0; face < nr_faces; face++) { + GLuint level; + for (level = stObj->base.BaseLevel; level <= stObj->lastLevel; level++) { + struct st_texture_image *stImage = + st_texture_image(stObj->base.Image[face][level]); + + /* Need to import images in main memory or held in other textures. + */ + if (stImage && stObj->pt != stImage->pt) { + copy_image_data_to_texture(st, stObj, level, stImage); + } + } + } + + return GL_TRUE; +} + + +/** + * Returns pointer to a default/dummy texture. + * This is typically used when the current shader has tex/sample instructions + * but the user has not provided a (any) texture(s). + */ +struct gl_texture_object * +st_get_default_texture(struct st_context *st) +{ + if (!st->default_texture) { + static const GLenum target = GL_TEXTURE_2D; + GLubyte pixels[16][16][4]; + struct gl_texture_object *texObj; + struct gl_texture_image *texImg; + GLuint i, j; + + /* The ARB_fragment_program spec says (0,0,0,1) should be returned + * when attempting to sample incomplete textures. + */ + for (i = 0; i < 16; i++) { + for (j = 0; j < 16; j++) { + pixels[i][j][0] = 0; + pixels[i][j][1] = 0; + pixels[i][j][2] = 0; + pixels[i][j][3] = 255; + } + } + + texObj = st->ctx->Driver.NewTextureObject(st->ctx, 0, target); + + texImg = _mesa_get_tex_image(st->ctx, texObj, target, 0); + + _mesa_init_teximage_fields(st->ctx, target, texImg, + 16, 16, 1, 0, /* w, h, d, border */ + GL_RGBA, MESA_FORMAT_RGBA8888); + + st_TexImage(st->ctx, 2, target, + 0, GL_RGBA, /* level, intformat */ + 16, 16, 1, 0, /* w, h, d, border */ + GL_RGBA, GL_UNSIGNED_BYTE, pixels, + &st->ctx->DefaultPacking, + texObj, texImg, + 0, 0); + + texObj->MinFilter = GL_NEAREST; + texObj->MagFilter = GL_NEAREST; + texObj->_Complete = GL_TRUE; + + st->default_texture = texObj; + } + return st->default_texture; +} + + +void +st_init_texture_functions(struct dd_function_table *functions) +{ + functions->ChooseTextureFormat = st_ChooseTextureFormat; + functions->TexImage1D = st_TexImage1D; + functions->TexImage2D = st_TexImage2D; + functions->TexImage3D = st_TexImage3D; + functions->TexSubImage1D = st_TexSubImage1D; + functions->TexSubImage2D = st_TexSubImage2D; + functions->TexSubImage3D = st_TexSubImage3D; + functions->CompressedTexSubImage1D = st_CompressedTexSubImage1D; + functions->CompressedTexSubImage2D = st_CompressedTexSubImage2D; + functions->CompressedTexSubImage3D = st_CompressedTexSubImage3D; + functions->CopyTexImage1D = st_CopyTexImage1D; + functions->CopyTexImage2D = st_CopyTexImage2D; + functions->CopyTexSubImage1D = st_CopyTexSubImage1D; + functions->CopyTexSubImage2D = st_CopyTexSubImage2D; + functions->CopyTexSubImage3D = st_CopyTexSubImage3D; + functions->GenerateMipmap = st_generate_mipmap; + + functions->GetTexImage = st_GetTexImage; + + /* compressed texture functions */ + functions->CompressedTexImage2D = st_CompressedTexImage2D; + functions->GetCompressedTexImage = st_GetCompressedTexImage; + + functions->NewTextureObject = st_NewTextureObject; + functions->NewTextureImage = st_NewTextureImage; + functions->DeleteTexture = st_DeleteTextureObject; + functions->FreeTexImageData = st_FreeTextureImageData; + + functions->TextureMemCpy = do_memcpy; + + /* XXX Temporary until we can query pipe's texture sizes */ + functions->TestProxyTexImage = _mesa_test_proxy_teximage; +} diff --git a/mesalib/src/mesa/state_tracker/st_extensions.c b/mesalib/src/mesa/state_tracker/st_extensions.c index d7ed7ea3a..a81cbe85d 100644 --- a/mesalib/src/mesa/state_tracker/st_extensions.c +++ b/mesalib/src/mesa/state_tracker/st_extensions.c @@ -420,6 +420,12 @@ void st_init_extensions(struct st_context *st) ctx->Extensions.MESA_ycbcr_texture = GL_TRUE; } + /* GL_EXT_texture_array */ + if (screen->get_param(screen, PIPE_CAP_ARRAY_TEXTURES)) { + ctx->Extensions.EXT_texture_array = GL_TRUE; + ctx->Extensions.MESA_texture_array = GL_TRUE; + } + /* GL_ARB_framebuffer_object */ if (ctx->Extensions.EXT_packed_depth_stencil) { /* we support always support GL_EXT_framebuffer_blit */ diff --git a/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c b/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c index 87c69e4dd..5c68fd78c 100644 --- a/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c +++ b/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c @@ -1,1247 +1,1249 @@ -/************************************************************************** - * - * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas. - * 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, sub license, 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 NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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. - * - **************************************************************************/ - -/* - * \author - * Michal Krol, - * Keith Whitwell - */ - -#include "pipe/p_compiler.h" -#include "pipe/p_context.h" -#include "pipe/p_screen.h" -#include "pipe/p_shader_tokens.h" -#include "pipe/p_state.h" -#include "tgsi/tgsi_ureg.h" -#include "st_mesa_to_tgsi.h" -#include "st_context.h" -#include "program/prog_instruction.h" -#include "program/prog_parameter.h" -#include "util/u_debug.h" -#include "util/u_math.h" -#include "util/u_memory.h" - - -#define PROGRAM_ANY_CONST ((1 << PROGRAM_LOCAL_PARAM) | \ - (1 << PROGRAM_ENV_PARAM) | \ - (1 << PROGRAM_STATE_VAR) | \ - (1 << PROGRAM_NAMED_PARAM) | \ - (1 << PROGRAM_CONSTANT) | \ - (1 << PROGRAM_UNIFORM)) - - -struct label { - unsigned branch_target; - unsigned token; -}; - - -/** - * Intermediate state used during shader translation. - */ -struct st_translate { - struct ureg_program *ureg; - - struct ureg_dst temps[MAX_PROGRAM_TEMPS]; - struct ureg_src *constants; - struct ureg_dst outputs[PIPE_MAX_SHADER_OUTPUTS]; - struct ureg_src inputs[PIPE_MAX_SHADER_INPUTS]; - struct ureg_dst address[1]; - struct ureg_src samplers[PIPE_MAX_SAMPLERS]; - struct ureg_src systemValues[SYSTEM_VALUE_MAX]; - - /* Extra info for handling point size clamping in vertex shader */ - struct ureg_dst pointSizeResult; /**< Actual point size output register */ - struct ureg_src pointSizeConst; /**< Point size range constant register */ - GLint pointSizeOutIndex; /**< Temp point size output register */ - GLboolean prevInstWrotePointSize; - - const GLuint *inputMapping; - const GLuint *outputMapping; - - /* For every instruction that contains a label (eg CALL), keep - * details so that we can go back afterwards and emit the correct - * tgsi instruction number for each label. - */ - struct label *labels; - unsigned labels_size; - unsigned labels_count; - - /* Keep a record of the tgsi instruction number that each mesa - * instruction starts at, will be used to fix up labels after - * translation. - */ - unsigned *insn; - unsigned insn_size; - unsigned insn_count; - - unsigned procType; /**< TGSI_PROCESSOR_VERTEX/FRAGMENT */ - - boolean error; -}; - - -/** Map Mesa's SYSTEM_VALUE_x to TGSI_SEMANTIC_x */ -static unsigned mesa_sysval_to_semantic[SYSTEM_VALUE_MAX] = { - TGSI_SEMANTIC_FACE, - TGSI_SEMANTIC_INSTANCEID -}; - - -/** - * Make note of a branch to a label in the TGSI code. - * After we've emitted all instructions, we'll go over the list - * of labels built here and patch the TGSI code with the actual - * location of each label. - */ -static unsigned *get_label( struct st_translate *t, - unsigned branch_target ) -{ - unsigned i; - - if (t->labels_count + 1 >= t->labels_size) { - unsigned old_size = t->labels_size; - t->labels_size = 1 << (util_logbase2(t->labels_size) + 1); - t->labels = REALLOC( t->labels, - old_size * sizeof t->labels[0], - t->labels_size * sizeof t->labels[0] ); - if (t->labels == NULL) { - static unsigned dummy; - t->error = TRUE; - return &dummy; - } - } - - i = t->labels_count++; - t->labels[i].branch_target = branch_target; - return &t->labels[i].token; -} - - -/** - * Called prior to emitting the TGSI code for each Mesa instruction. - * Allocate additional space for instructions if needed. - * Update the insn[] array so the next Mesa instruction points to - * the next TGSI instruction. - */ -static void set_insn_start( struct st_translate *t, - unsigned start ) -{ - if (t->insn_count + 1 >= t->insn_size) { - unsigned old_size = t->insn_size; - t->insn_size = 1 << (util_logbase2(t->insn_size) + 1); - t->insn = REALLOC( t->insn, - old_size * sizeof t->insn[0], - t->insn_size * sizeof t->insn[0] ); - if (t->insn == NULL) { - t->error = TRUE; - return; - } - } - - t->insn[t->insn_count++] = start; -} - - -/** - * Map a Mesa dst register to a TGSI ureg_dst register. - */ -static struct ureg_dst -dst_register( struct st_translate *t, - gl_register_file file, - GLuint index ) -{ - switch( file ) { - case PROGRAM_UNDEFINED: - return ureg_dst_undef(); - - case PROGRAM_TEMPORARY: - if (ureg_dst_is_undef(t->temps[index])) - t->temps[index] = ureg_DECL_temporary( t->ureg ); - - return t->temps[index]; - - case PROGRAM_OUTPUT: - if (t->procType == TGSI_PROCESSOR_VERTEX && index == VERT_RESULT_PSIZ) - t->prevInstWrotePointSize = GL_TRUE; - - if (t->procType == TGSI_PROCESSOR_VERTEX) - assert(index < VERT_RESULT_MAX); - else if (t->procType == TGSI_PROCESSOR_FRAGMENT) - assert(index < FRAG_RESULT_MAX); - else - assert(index < GEOM_RESULT_MAX); - - assert(t->outputMapping[index] < Elements(t->outputs)); - - return t->outputs[t->outputMapping[index]]; - - case PROGRAM_ADDRESS: - return t->address[index]; - - default: - debug_assert( 0 ); - return ureg_dst_undef(); - } -} - - -/** - * Map a Mesa src register to a TGSI ureg_src register. - */ -static struct ureg_src -src_register( struct st_translate *t, - gl_register_file file, - GLint index ) -{ - switch( file ) { - case PROGRAM_UNDEFINED: - return ureg_src_undef(); - - case PROGRAM_TEMPORARY: - assert(index >= 0); - if (ureg_dst_is_undef(t->temps[index])) - t->temps[index] = ureg_DECL_temporary( t->ureg ); - assert(index < Elements(t->temps)); - return ureg_src(t->temps[index]); - - case PROGRAM_NAMED_PARAM: - case PROGRAM_ENV_PARAM: - case PROGRAM_LOCAL_PARAM: - case PROGRAM_UNIFORM: - assert(index >= 0); - return t->constants[index]; - case PROGRAM_STATE_VAR: - case PROGRAM_CONSTANT: /* ie, immediate */ - if (index < 0) - return ureg_DECL_constant( t->ureg, 0 ); - else - return t->constants[index]; - - case PROGRAM_INPUT: - assert(t->inputMapping[index] < Elements(t->inputs)); - return t->inputs[t->inputMapping[index]]; - - case PROGRAM_OUTPUT: - assert(t->outputMapping[index] < Elements(t->outputs)); - return ureg_src(t->outputs[t->outputMapping[index]]); /* not needed? */ - - case PROGRAM_ADDRESS: - return ureg_src(t->address[index]); - - case PROGRAM_SYSTEM_VALUE: - assert(index < Elements(t->systemValues)); - return t->systemValues[index]; - - default: - debug_assert( 0 ); - return ureg_src_undef(); - } -} - - -/** - * Map mesa texture target to TGSI texture target. - */ -static unsigned -translate_texture_target( GLuint textarget, - GLboolean shadow ) -{ - if (shadow) { - switch( textarget ) { - case TEXTURE_1D_INDEX: return TGSI_TEXTURE_SHADOW1D; - case TEXTURE_2D_INDEX: return TGSI_TEXTURE_SHADOW2D; - case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_SHADOWRECT; - default: break; - } - } - - switch( textarget ) { - case TEXTURE_1D_INDEX: return TGSI_TEXTURE_1D; - case TEXTURE_2D_INDEX: return TGSI_TEXTURE_2D; - case TEXTURE_3D_INDEX: return TGSI_TEXTURE_3D; - case TEXTURE_CUBE_INDEX: return TGSI_TEXTURE_CUBE; - case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_RECT; - default: - debug_assert( 0 ); - return TGSI_TEXTURE_1D; - } -} - - -/** - * Create a TGSI ureg_dst register from a Mesa dest register. - */ -static struct ureg_dst -translate_dst( struct st_translate *t, - const struct prog_dst_register *DstReg, - boolean saturate ) -{ - struct ureg_dst dst = dst_register( t, - DstReg->File, - DstReg->Index ); - - dst = ureg_writemask( dst, - DstReg->WriteMask ); - - if (saturate) - dst = ureg_saturate( dst ); - - if (DstReg->RelAddr) - dst = ureg_dst_indirect( dst, ureg_src(t->address[0]) ); - - return dst; -} - - -/** - * Create a TGSI ureg_src register from a Mesa src register. - */ -static struct ureg_src -translate_src( struct st_translate *t, - const struct prog_src_register *SrcReg ) -{ - struct ureg_src src = src_register( t, SrcReg->File, SrcReg->Index ); - - if (t->procType == TGSI_PROCESSOR_GEOMETRY && SrcReg->HasIndex2) { - src = src_register( t, SrcReg->File, SrcReg->Index2 ); - if (SrcReg->RelAddr2) - src = ureg_src_dimension_indirect( src, ureg_src(t->address[0]), - SrcReg->Index); - else - src = ureg_src_dimension( src, SrcReg->Index); - } - - src = ureg_swizzle( src, - GET_SWZ( SrcReg->Swizzle, 0 ) & 0x3, - GET_SWZ( SrcReg->Swizzle, 1 ) & 0x3, - GET_SWZ( SrcReg->Swizzle, 2 ) & 0x3, - GET_SWZ( SrcReg->Swizzle, 3 ) & 0x3); - - if (SrcReg->Negate == NEGATE_XYZW) - src = ureg_negate(src); - - if (SrcReg->Abs) - src = ureg_abs(src); - - if (SrcReg->RelAddr) { - src = ureg_src_indirect( src, ureg_src(t->address[0])); - if (SrcReg->File != PROGRAM_INPUT && - SrcReg->File != PROGRAM_OUTPUT) { - /* If SrcReg->Index was negative, it was set to zero in - * src_register(). Reassign it now. But don't do this - * for input/output regs since they get remapped while - * const buffers don't. - */ - src.Index = SrcReg->Index; - } - } - - return src; -} - - -static struct ureg_src swizzle_4v( struct ureg_src src, - const unsigned *swz ) -{ - return ureg_swizzle( src, swz[0], swz[1], swz[2], swz[3] ); -} - - -/** - * Translate a SWZ instruction into a MOV, MUL or MAD instruction. EG: - * - * SWZ dst, src.x-y10 - * - * becomes: - * - * MAD dst {1,-1,0,0}, src.xyxx, {0,0,1,0} - */ -static void emit_swz( struct st_translate *t, - struct ureg_dst dst, - const struct prog_src_register *SrcReg ) -{ - struct ureg_program *ureg = t->ureg; - struct ureg_src src = src_register( t, SrcReg->File, SrcReg->Index ); - - unsigned negate_mask = SrcReg->Negate; - - unsigned one_mask = ((GET_SWZ(SrcReg->Swizzle, 0) == SWIZZLE_ONE) << 0 | - (GET_SWZ(SrcReg->Swizzle, 1) == SWIZZLE_ONE) << 1 | - (GET_SWZ(SrcReg->Swizzle, 2) == SWIZZLE_ONE) << 2 | - (GET_SWZ(SrcReg->Swizzle, 3) == SWIZZLE_ONE) << 3); - - unsigned zero_mask = ((GET_SWZ(SrcReg->Swizzle, 0) == SWIZZLE_ZERO) << 0 | - (GET_SWZ(SrcReg->Swizzle, 1) == SWIZZLE_ZERO) << 1 | - (GET_SWZ(SrcReg->Swizzle, 2) == SWIZZLE_ZERO) << 2 | - (GET_SWZ(SrcReg->Swizzle, 3) == SWIZZLE_ZERO) << 3); - - unsigned negative_one_mask = one_mask & negate_mask; - unsigned positive_one_mask = one_mask & ~negate_mask; - - struct ureg_src imm; - unsigned i; - unsigned mul_swizzle[4] = {0,0,0,0}; - unsigned add_swizzle[4] = {0,0,0,0}; - unsigned src_swizzle[4] = {0,0,0,0}; - boolean need_add = FALSE; - boolean need_mul = FALSE; - - if (dst.WriteMask == 0) - return; - - /* Is this just a MOV? - */ - if (zero_mask == 0 && - one_mask == 0 && - (negate_mask == 0 || negate_mask == TGSI_WRITEMASK_XYZW)) - { - ureg_MOV( ureg, dst, translate_src( t, SrcReg )); - return; - } - -#define IMM_ZERO 0 -#define IMM_ONE 1 -#define IMM_NEG_ONE 2 - - imm = ureg_imm3f( ureg, 0, 1, -1 ); - - for (i = 0; i < 4; i++) { - unsigned bit = 1 << i; - - if (dst.WriteMask & bit) { - if (positive_one_mask & bit) { - mul_swizzle[i] = IMM_ZERO; - add_swizzle[i] = IMM_ONE; - need_add = TRUE; - } - else if (negative_one_mask & bit) { - mul_swizzle[i] = IMM_ZERO; - add_swizzle[i] = IMM_NEG_ONE; - need_add = TRUE; - } - else if (zero_mask & bit) { - mul_swizzle[i] = IMM_ZERO; - add_swizzle[i] = IMM_ZERO; - need_add = TRUE; - } - else { - add_swizzle[i] = IMM_ZERO; - src_swizzle[i] = GET_SWZ(SrcReg->Swizzle, i); - need_mul = TRUE; - if (negate_mask & bit) { - mul_swizzle[i] = IMM_NEG_ONE; - } - else { - mul_swizzle[i] = IMM_ONE; - } - } - } - } - - if (need_mul && need_add) { - ureg_MAD( ureg, - dst, - swizzle_4v( src, src_swizzle ), - swizzle_4v( imm, mul_swizzle ), - swizzle_4v( imm, add_swizzle ) ); - } - else if (need_mul) { - ureg_MUL( ureg, - dst, - swizzle_4v( src, src_swizzle ), - swizzle_4v( imm, mul_swizzle ) ); - } - else if (need_add) { - ureg_MOV( ureg, - dst, - swizzle_4v( imm, add_swizzle ) ); - } - else { - debug_assert(0); - } - -#undef IMM_ZERO -#undef IMM_ONE -#undef IMM_NEG_ONE -} - - -/** - * Negate the value of DDY to match GL semantics where (0,0) is the - * lower-left corner of the window. - * Note that the GL_ARB_fragment_coord_conventions extension will - * effect this someday. - */ -static void emit_ddy( struct st_translate *t, - struct ureg_dst dst, - const struct prog_src_register *SrcReg ) -{ - struct ureg_program *ureg = t->ureg; - struct ureg_src src = translate_src( t, SrcReg ); - src = ureg_negate( src ); - ureg_DDY( ureg, dst, src ); -} - - - -static unsigned -translate_opcode( unsigned op ) -{ - switch( op ) { - case OPCODE_ARL: - return TGSI_OPCODE_ARL; - case OPCODE_ABS: - return TGSI_OPCODE_ABS; - case OPCODE_ADD: - return TGSI_OPCODE_ADD; - case OPCODE_BGNLOOP: - return TGSI_OPCODE_BGNLOOP; - case OPCODE_BGNSUB: - return TGSI_OPCODE_BGNSUB; - case OPCODE_BRA: - return TGSI_OPCODE_BRA; - case OPCODE_BRK: - return TGSI_OPCODE_BRK; - case OPCODE_CAL: - return TGSI_OPCODE_CAL; - case OPCODE_CMP: - return TGSI_OPCODE_CMP; - case OPCODE_CONT: - return TGSI_OPCODE_CONT; - case OPCODE_COS: - return TGSI_OPCODE_COS; - case OPCODE_DDX: - return TGSI_OPCODE_DDX; - case OPCODE_DDY: - return TGSI_OPCODE_DDY; - case OPCODE_DP2: - return TGSI_OPCODE_DP2; - case OPCODE_DP2A: - return TGSI_OPCODE_DP2A; - case OPCODE_DP3: - return TGSI_OPCODE_DP3; - case OPCODE_DP4: - return TGSI_OPCODE_DP4; - case OPCODE_DPH: - return TGSI_OPCODE_DPH; - case OPCODE_DST: - return TGSI_OPCODE_DST; - case OPCODE_ELSE: - return TGSI_OPCODE_ELSE; - case OPCODE_EMIT_VERTEX: - return TGSI_OPCODE_EMIT; - case OPCODE_END_PRIMITIVE: - return TGSI_OPCODE_ENDPRIM; - case OPCODE_ENDIF: - return TGSI_OPCODE_ENDIF; - case OPCODE_ENDLOOP: - return TGSI_OPCODE_ENDLOOP; - case OPCODE_ENDSUB: - return TGSI_OPCODE_ENDSUB; - case OPCODE_EX2: - return TGSI_OPCODE_EX2; - case OPCODE_EXP: - return TGSI_OPCODE_EXP; - case OPCODE_FLR: - return TGSI_OPCODE_FLR; - case OPCODE_FRC: - return TGSI_OPCODE_FRC; - case OPCODE_IF: - return TGSI_OPCODE_IF; - case OPCODE_TRUNC: - return TGSI_OPCODE_TRUNC; - case OPCODE_KIL: - return TGSI_OPCODE_KIL; - case OPCODE_KIL_NV: - return TGSI_OPCODE_KILP; - case OPCODE_LG2: - return TGSI_OPCODE_LG2; - case OPCODE_LOG: - return TGSI_OPCODE_LOG; - case OPCODE_LIT: - return TGSI_OPCODE_LIT; - case OPCODE_LRP: - return TGSI_OPCODE_LRP; - case OPCODE_MAD: - return TGSI_OPCODE_MAD; - case OPCODE_MAX: - return TGSI_OPCODE_MAX; - case OPCODE_MIN: - return TGSI_OPCODE_MIN; - case OPCODE_MOV: - return TGSI_OPCODE_MOV; - case OPCODE_MUL: - return TGSI_OPCODE_MUL; - case OPCODE_NOP: - return TGSI_OPCODE_NOP; - case OPCODE_NRM3: - return TGSI_OPCODE_NRM; - case OPCODE_NRM4: - return TGSI_OPCODE_NRM4; - case OPCODE_POW: - return TGSI_OPCODE_POW; - case OPCODE_RCP: - return TGSI_OPCODE_RCP; - case OPCODE_RET: - return TGSI_OPCODE_RET; - case OPCODE_RSQ: - return TGSI_OPCODE_RSQ; - case OPCODE_SCS: - return TGSI_OPCODE_SCS; - case OPCODE_SEQ: - return TGSI_OPCODE_SEQ; - case OPCODE_SGE: - return TGSI_OPCODE_SGE; - case OPCODE_SGT: - return TGSI_OPCODE_SGT; - case OPCODE_SIN: - return TGSI_OPCODE_SIN; - case OPCODE_SLE: - return TGSI_OPCODE_SLE; - case OPCODE_SLT: - return TGSI_OPCODE_SLT; - case OPCODE_SNE: - return TGSI_OPCODE_SNE; - case OPCODE_SSG: - return TGSI_OPCODE_SSG; - case OPCODE_SUB: - return TGSI_OPCODE_SUB; - case OPCODE_TEX: - return TGSI_OPCODE_TEX; - case OPCODE_TXB: - return TGSI_OPCODE_TXB; - case OPCODE_TXD: - return TGSI_OPCODE_TXD; - case OPCODE_TXL: - return TGSI_OPCODE_TXL; - case OPCODE_TXP: - return TGSI_OPCODE_TXP; - case OPCODE_XPD: - return TGSI_OPCODE_XPD; - case OPCODE_END: - return TGSI_OPCODE_END; - default: - debug_assert( 0 ); - return TGSI_OPCODE_NOP; - } -} - - -static void -compile_instruction( - struct st_translate *t, - const struct prog_instruction *inst ) -{ - struct ureg_program *ureg = t->ureg; - GLuint i; - struct ureg_dst dst[1]; - struct ureg_src src[4]; - unsigned num_dst; - unsigned num_src; - - num_dst = _mesa_num_inst_dst_regs( inst->Opcode ); - num_src = _mesa_num_inst_src_regs( inst->Opcode ); - - if (num_dst) - dst[0] = translate_dst( t, - &inst->DstReg, - inst->SaturateMode ); - - for (i = 0; i < num_src; i++) - src[i] = translate_src( t, &inst->SrcReg[i] ); - - switch( inst->Opcode ) { - case OPCODE_SWZ: - emit_swz( t, dst[0], &inst->SrcReg[0] ); - return; - - case OPCODE_BGNLOOP: - case OPCODE_CAL: - case OPCODE_ELSE: - case OPCODE_ENDLOOP: - case OPCODE_IF: - debug_assert(num_dst == 0); - ureg_label_insn( ureg, - translate_opcode( inst->Opcode ), - src, num_src, - get_label( t, inst->BranchTarget )); - return; - - case OPCODE_TEX: - case OPCODE_TXB: - case OPCODE_TXD: - case OPCODE_TXL: - case OPCODE_TXP: - src[num_src++] = t->samplers[inst->TexSrcUnit]; - ureg_tex_insn( ureg, - translate_opcode( inst->Opcode ), - dst, num_dst, - translate_texture_target( inst->TexSrcTarget, - inst->TexShadow ), - src, num_src ); - return; - - case OPCODE_SCS: - dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XY ); - ureg_insn( ureg, - translate_opcode( inst->Opcode ), - dst, num_dst, - src, num_src ); - break; - - case OPCODE_XPD: - dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XYZ ); - ureg_insn( ureg, - translate_opcode( inst->Opcode ), - dst, num_dst, - src, num_src ); - break; - - case OPCODE_NOISE1: - case OPCODE_NOISE2: - case OPCODE_NOISE3: - case OPCODE_NOISE4: - /* At some point, a motivated person could add a better - * implementation of noise. Currently not even the nvidia - * binary drivers do anything more than this. In any case, the - * place to do this is in the GL state tracker, not the poor - * driver. - */ - ureg_MOV( ureg, dst[0], ureg_imm1f(ureg, 0.5) ); - break; - - case OPCODE_DDY: - emit_ddy( t, dst[0], &inst->SrcReg[0] ); - break; - - default: - ureg_insn( ureg, - translate_opcode( inst->Opcode ), - dst, num_dst, - src, num_src ); - break; - } -} - - -/** - * Emit the TGSI instructions to adjust the WPOS pixel center convention - * Basically, add (adjX, adjY) to the fragment position. - */ -static void -emit_adjusted_wpos( struct st_translate *t, - const struct gl_program *program, - GLfloat adjX, GLfloat adjY) -{ - struct ureg_program *ureg = t->ureg; - struct ureg_dst wpos_temp = ureg_DECL_temporary(ureg); - struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]]; - - /* Note that we bias X and Y and pass Z and W through unchanged. - * The shader might also use gl_FragCoord.w and .z. - */ - ureg_ADD(ureg, wpos_temp, wpos_input, - ureg_imm4f(ureg, adjX, adjY, 0.0f, 0.0f)); - - t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]] = ureg_src(wpos_temp); -} - - -/** - * Emit the TGSI instructions for inverting the WPOS y coordinate. - * This code is unavoidable because it also depends on whether - * a FBO is bound (STATE_FB_WPOS_Y_TRANSFORM). - */ -static void -emit_wpos_inversion( struct st_translate *t, - const struct gl_program *program, - boolean invert) -{ - struct ureg_program *ureg = t->ureg; - - /* Fragment program uses fragment position input. - * Need to replace instances of INPUT[WPOS] with temp T - * where T = INPUT[WPOS] by y is inverted. - */ - static const gl_state_index wposTransformState[STATE_LENGTH] - = { STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM, 0, 0, 0 }; - - /* XXX: note we are modifying the incoming shader here! Need to - * do this before emitting the constant decls below, or this - * will be missed: - */ - unsigned wposTransConst = _mesa_add_state_reference(program->Parameters, - wposTransformState); - - struct ureg_src wpostrans = ureg_DECL_constant( ureg, wposTransConst ); - struct ureg_dst wpos_temp; - struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]]; - - /* MOV wpos_temp, input[wpos] - */ - if (wpos_input.File == TGSI_FILE_TEMPORARY) - wpos_temp = ureg_dst(wpos_input); - else { - wpos_temp = ureg_DECL_temporary( ureg ); - ureg_MOV( ureg, wpos_temp, wpos_input ); - } - - if (invert) { - /* MAD wpos_temp.y, wpos_input, wpostrans.xxxx, wpostrans.yyyy - */ - ureg_MAD( ureg, - ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ), - wpos_input, - ureg_scalar(wpostrans, 0), - ureg_scalar(wpostrans, 1)); - } else { - /* MAD wpos_temp.y, wpos_input, wpostrans.zzzz, wpostrans.wwww - */ - ureg_MAD( ureg, - ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ), - wpos_input, - ureg_scalar(wpostrans, 2), - ureg_scalar(wpostrans, 3)); - } - - /* Use wpos_temp as position input from here on: - */ - t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]] = ureg_src(wpos_temp); -} - - -/** - * Emit fragment position/ooordinate code. - */ -static void -emit_wpos(struct st_context *st, - struct st_translate *t, - const struct gl_program *program, - struct ureg_program *ureg) -{ - const struct gl_fragment_program *fp = - (const struct gl_fragment_program *) program; - struct pipe_screen *pscreen = st->pipe->screen; - boolean invert = FALSE; - - if (fp->OriginUpperLeft) { - /* Fragment shader wants origin in upper-left */ - if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT)) { - /* the driver supports upper-left origin */ - } - else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT)) { - /* the driver supports lower-left origin, need to invert Y */ - ureg_property_fs_coord_origin(ureg, TGSI_FS_COORD_ORIGIN_LOWER_LEFT); - invert = TRUE; - } - else - assert(0); - } - else { - /* Fragment shader wants origin in lower-left */ - if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT)) - /* the driver supports lower-left origin */ - ureg_property_fs_coord_origin(ureg, TGSI_FS_COORD_ORIGIN_LOWER_LEFT); - else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT)) - /* the driver supports upper-left origin, need to invert Y */ - invert = TRUE; - else - assert(0); - } - - if (fp->PixelCenterInteger) { - /* Fragment shader wants pixel center integer */ - if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) - /* the driver supports pixel center integer */ - ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER); - else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) - /* the driver supports pixel center half integer, need to bias X,Y */ - emit_adjusted_wpos(t, program, 0.5f, invert ? 0.5f : -0.5f); - else - assert(0); - } - else { - /* Fragment shader wants pixel center half integer */ - if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) { - /* the driver supports pixel center half integer */ - } - else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) { - /* the driver supports pixel center integer, need to bias X,Y */ - ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER); - emit_adjusted_wpos(t, program, 0.5f, invert ? -0.5f : 0.5f); - } - else - assert(0); - } - - /* we invert after adjustment so that we avoid the MOV to temporary, - * and reuse the adjustment ADD instead */ - emit_wpos_inversion(t, program, invert); -} - - -/** - * OpenGL's fragment gl_FrontFace input is 1 for front-facing, 0 for back. - * TGSI uses +1 for front, -1 for back. - * This function converts the TGSI value to the GL value. Simply clamping/ - * saturating the value to [0,1] does the job. - */ -static void -emit_face_var( struct st_translate *t, - const struct gl_program *program ) -{ - struct ureg_program *ureg = t->ureg; - struct ureg_dst face_temp = ureg_DECL_temporary( ureg ); - struct ureg_src face_input = t->inputs[t->inputMapping[FRAG_ATTRIB_FACE]]; - - /* MOV_SAT face_temp, input[face] - */ - face_temp = ureg_saturate( face_temp ); - ureg_MOV( ureg, face_temp, face_input ); - - /* Use face_temp as face input from here on: - */ - t->inputs[t->inputMapping[FRAG_ATTRIB_FACE]] = ureg_src(face_temp); -} - - -static void -emit_edgeflags( struct st_translate *t, - const struct gl_program *program ) -{ - struct ureg_program *ureg = t->ureg; - struct ureg_dst edge_dst = t->outputs[t->outputMapping[VERT_RESULT_EDGE]]; - struct ureg_src edge_src = t->inputs[t->inputMapping[VERT_ATTRIB_EDGEFLAG]]; - - ureg_MOV( ureg, edge_dst, edge_src ); -} - - -/** - * Translate Mesa program to TGSI format. - * \param program the program to translate - * \param numInputs number of input registers used - * \param inputMapping maps Mesa fragment program inputs to TGSI generic - * input indexes - * \param inputSemanticName the TGSI_SEMANTIC flag for each input - * \param inputSemanticIndex the semantic index (ex: which texcoord) for - * each input - * \param interpMode the TGSI_INTERPOLATE_LINEAR/PERSP mode for each input - * \param numOutputs number of output registers used - * \param outputMapping maps Mesa fragment program outputs to TGSI - * generic outputs - * \param outputSemanticName the TGSI_SEMANTIC flag for each output - * \param outputSemanticIndex the semantic index (ex: which texcoord) for - * each output - * - * \return PIPE_OK or PIPE_ERROR_OUT_OF_MEMORY - */ -enum pipe_error -st_translate_mesa_program( - struct gl_context *ctx, - uint procType, - struct ureg_program *ureg, - const struct gl_program *program, - GLuint numInputs, - const GLuint inputMapping[], - const ubyte inputSemanticName[], - const ubyte inputSemanticIndex[], - const GLuint interpMode[], - GLuint numOutputs, - const GLuint outputMapping[], - const ubyte outputSemanticName[], - const ubyte outputSemanticIndex[], - boolean passthrough_edgeflags ) -{ - struct st_translate translate, *t; - unsigned i; - enum pipe_error ret = PIPE_OK; - - assert(numInputs <= Elements(t->inputs)); - assert(numOutputs <= Elements(t->outputs)); - - t = &translate; - memset(t, 0, sizeof *t); - - t->procType = procType; - t->inputMapping = inputMapping; - t->outputMapping = outputMapping; - t->ureg = ureg; - t->pointSizeOutIndex = -1; - t->prevInstWrotePointSize = GL_FALSE; - - /*_mesa_print_program(program);*/ - - /* - * Declare input attributes. - */ - if (procType == TGSI_PROCESSOR_FRAGMENT) { - for (i = 0; i < numInputs; i++) { - if (program->InputFlags[0] & PROG_PARAM_BIT_CYL_WRAP) { - t->inputs[i] = ureg_DECL_fs_input_cyl(ureg, - inputSemanticName[i], - inputSemanticIndex[i], - interpMode[i], - TGSI_CYLINDRICAL_WRAP_X); - } - else { - t->inputs[i] = ureg_DECL_fs_input(ureg, - inputSemanticName[i], - inputSemanticIndex[i], - interpMode[i]); - } - } - - if (program->InputsRead & FRAG_BIT_WPOS) { - /* Must do this after setting up t->inputs, and before - * emitting constant references, below: - */ - emit_wpos(st_context(ctx), t, program, ureg); - } - - if (program->InputsRead & FRAG_BIT_FACE) { - emit_face_var( t, program ); - } - - /* - * Declare output attributes. - */ - for (i = 0; i < numOutputs; i++) { - switch (outputSemanticName[i]) { - case TGSI_SEMANTIC_POSITION: - t->outputs[i] = ureg_DECL_output( ureg, - TGSI_SEMANTIC_POSITION, /* Z / Depth */ - outputSemanticIndex[i] ); - - t->outputs[i] = ureg_writemask( t->outputs[i], - TGSI_WRITEMASK_Z ); - break; - case TGSI_SEMANTIC_STENCIL: - t->outputs[i] = ureg_DECL_output( ureg, - TGSI_SEMANTIC_STENCIL, /* Stencil */ - outputSemanticIndex[i] ); - t->outputs[i] = ureg_writemask( t->outputs[i], - TGSI_WRITEMASK_Y ); - break; - case TGSI_SEMANTIC_COLOR: - t->outputs[i] = ureg_DECL_output( ureg, - TGSI_SEMANTIC_COLOR, - outputSemanticIndex[i] ); - break; - default: - debug_assert(0); - return 0; - } - } - } - else if (procType == TGSI_PROCESSOR_GEOMETRY) { - for (i = 0; i < numInputs; i++) { - t->inputs[i] = ureg_DECL_gs_input(ureg, - i, - inputSemanticName[i], - inputSemanticIndex[i]); - } - - for (i = 0; i < numOutputs; i++) { - t->outputs[i] = ureg_DECL_output( ureg, - outputSemanticName[i], - outputSemanticIndex[i] ); - } - } - else { - assert(procType == TGSI_PROCESSOR_VERTEX); - - for (i = 0; i < numInputs; i++) { - t->inputs[i] = ureg_DECL_vs_input(ureg, i); - } - - for (i = 0; i < numOutputs; i++) { - t->outputs[i] = ureg_DECL_output( ureg, - outputSemanticName[i], - outputSemanticIndex[i] ); - if ((outputSemanticName[i] == TGSI_SEMANTIC_PSIZE) && program->Id) { - /* Writing to the point size result register requires special - * handling to implement clamping. - */ - static const gl_state_index pointSizeClampState[STATE_LENGTH] - = { STATE_INTERNAL, STATE_POINT_SIZE_IMPL_CLAMP, 0, 0, 0 }; - /* XXX: note we are modifying the incoming shader here! Need to - * do this before emitting the constant decls below, or this - * will be missed: - */ - unsigned pointSizeClampConst = - _mesa_add_state_reference(program->Parameters, - pointSizeClampState); - struct ureg_dst psizregtemp = ureg_DECL_temporary( ureg ); - t->pointSizeConst = ureg_DECL_constant( ureg, pointSizeClampConst ); - t->pointSizeResult = t->outputs[i]; - t->pointSizeOutIndex = i; - t->outputs[i] = psizregtemp; - } - } - if (passthrough_edgeflags) - emit_edgeflags( t, program ); - } - - /* Declare address register. - */ - if (program->NumAddressRegs > 0) { - debug_assert( program->NumAddressRegs == 1 ); - t->address[0] = ureg_DECL_address( ureg ); - } - - /* Declare misc input registers - */ - { - GLbitfield sysInputs = program->SystemValuesRead; - unsigned numSys = 0; - for (i = 0; sysInputs; i++) { - if (sysInputs & (1 << i)) { - unsigned semName = mesa_sysval_to_semantic[i]; - t->systemValues[i] = ureg_DECL_system_value(ureg, numSys, semName, 0); - numSys++; - sysInputs &= ~(1 << i); - } - } - } - - if (program->IndirectRegisterFiles & (1 << PROGRAM_TEMPORARY)) { - /* If temps are accessed with indirect addressing, declare temporaries - * in sequential order. Else, we declare them on demand elsewhere. - */ - for (i = 0; i < program->NumTemporaries; i++) { - /* XXX use TGSI_FILE_TEMPORARY_ARRAY when it's supported by ureg */ - t->temps[i] = ureg_DECL_temporary( t->ureg ); - } - } - - /* Emit constants and immediates. Mesa uses a single index space - * for these, so we put all the translated regs in t->constants. - */ - if (program->Parameters) { - t->constants = CALLOC( program->Parameters->NumParameters, - sizeof t->constants[0] ); - if (t->constants == NULL) { - ret = PIPE_ERROR_OUT_OF_MEMORY; - goto out; - } - - for (i = 0; i < program->Parameters->NumParameters; i++) { - switch (program->Parameters->Parameters[i].Type) { - case PROGRAM_ENV_PARAM: - case PROGRAM_LOCAL_PARAM: - case PROGRAM_STATE_VAR: - case PROGRAM_NAMED_PARAM: - case PROGRAM_UNIFORM: - t->constants[i] = ureg_DECL_constant( ureg, i ); - break; - - /* Emit immediates only when there's no indirect addressing of - * the const buffer. - * FIXME: Be smarter and recognize param arrays: - * indirect addressing is only valid within the referenced - * array. - */ - case PROGRAM_CONSTANT: - if (program->IndirectRegisterFiles & PROGRAM_ANY_CONST) - t->constants[i] = ureg_DECL_constant( ureg, i ); - else - t->constants[i] = - ureg_DECL_immediate( ureg, - program->Parameters->ParameterValues[i], - 4 ); - break; - default: - break; - } - } - } - - /* texture samplers */ - for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) { - if (program->SamplersUsed & (1 << i)) { - t->samplers[i] = ureg_DECL_sampler( ureg, i ); - } - } - - /* Emit each instruction in turn: - */ - for (i = 0; i < program->NumInstructions; i++) { - set_insn_start( t, ureg_get_instruction_number( ureg )); - compile_instruction( t, &program->Instructions[i] ); - - if (t->prevInstWrotePointSize && program->Id) { - /* The previous instruction wrote to the (fake) vertex point size - * result register. Now we need to clamp that value to the min/max - * point size range, putting the result into the real point size - * register. - * Note that we can't do this easily at the end of program due to - * possible early return. - */ - set_insn_start( t, ureg_get_instruction_number( ureg )); - ureg_MAX( t->ureg, - ureg_writemask(t->outputs[t->pointSizeOutIndex], WRITEMASK_X), - ureg_src(t->outputs[t->pointSizeOutIndex]), - ureg_swizzle(t->pointSizeConst, 1,1,1,1)); - ureg_MIN( t->ureg, ureg_writemask(t->pointSizeResult, WRITEMASK_X), - ureg_src(t->outputs[t->pointSizeOutIndex]), - ureg_swizzle(t->pointSizeConst, 2,2,2,2)); - } - t->prevInstWrotePointSize = GL_FALSE; - } - - /* Fix up all emitted labels: - */ - for (i = 0; i < t->labels_count; i++) { - ureg_fixup_label( ureg, - t->labels[i].token, - t->insn[t->labels[i].branch_target] ); - } - -out: - FREE(t->insn); - FREE(t->labels); - FREE(t->constants); - - if (t->error) { - debug_printf("%s: translate error flag set\n", __FUNCTION__); - } - - return ret; -} - - -/** - * Tokens cannot be free with free otherwise the builtin gallium - * malloc debugging will get confused. - */ -void -st_free_tokens(const struct tgsi_token *tokens) -{ - FREE((void *)tokens); -} +/************************************************************************** + * + * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * 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, sub license, 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 NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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. + * + **************************************************************************/ + +/* + * \author + * Michal Krol, + * Keith Whitwell + */ + +#include "pipe/p_compiler.h" +#include "pipe/p_context.h" +#include "pipe/p_screen.h" +#include "pipe/p_shader_tokens.h" +#include "pipe/p_state.h" +#include "tgsi/tgsi_ureg.h" +#include "st_mesa_to_tgsi.h" +#include "st_context.h" +#include "program/prog_instruction.h" +#include "program/prog_parameter.h" +#include "util/u_debug.h" +#include "util/u_math.h" +#include "util/u_memory.h" + + +#define PROGRAM_ANY_CONST ((1 << PROGRAM_LOCAL_PARAM) | \ + (1 << PROGRAM_ENV_PARAM) | \ + (1 << PROGRAM_STATE_VAR) | \ + (1 << PROGRAM_NAMED_PARAM) | \ + (1 << PROGRAM_CONSTANT) | \ + (1 << PROGRAM_UNIFORM)) + + +struct label { + unsigned branch_target; + unsigned token; +}; + + +/** + * Intermediate state used during shader translation. + */ +struct st_translate { + struct ureg_program *ureg; + + struct ureg_dst temps[MAX_PROGRAM_TEMPS]; + struct ureg_src *constants; + struct ureg_dst outputs[PIPE_MAX_SHADER_OUTPUTS]; + struct ureg_src inputs[PIPE_MAX_SHADER_INPUTS]; + struct ureg_dst address[1]; + struct ureg_src samplers[PIPE_MAX_SAMPLERS]; + struct ureg_src systemValues[SYSTEM_VALUE_MAX]; + + /* Extra info for handling point size clamping in vertex shader */ + struct ureg_dst pointSizeResult; /**< Actual point size output register */ + struct ureg_src pointSizeConst; /**< Point size range constant register */ + GLint pointSizeOutIndex; /**< Temp point size output register */ + GLboolean prevInstWrotePointSize; + + const GLuint *inputMapping; + const GLuint *outputMapping; + + /* For every instruction that contains a label (eg CALL), keep + * details so that we can go back afterwards and emit the correct + * tgsi instruction number for each label. + */ + struct label *labels; + unsigned labels_size; + unsigned labels_count; + + /* Keep a record of the tgsi instruction number that each mesa + * instruction starts at, will be used to fix up labels after + * translation. + */ + unsigned *insn; + unsigned insn_size; + unsigned insn_count; + + unsigned procType; /**< TGSI_PROCESSOR_VERTEX/FRAGMENT */ + + boolean error; +}; + + +/** Map Mesa's SYSTEM_VALUE_x to TGSI_SEMANTIC_x */ +static unsigned mesa_sysval_to_semantic[SYSTEM_VALUE_MAX] = { + TGSI_SEMANTIC_FACE, + TGSI_SEMANTIC_INSTANCEID +}; + + +/** + * Make note of a branch to a label in the TGSI code. + * After we've emitted all instructions, we'll go over the list + * of labels built here and patch the TGSI code with the actual + * location of each label. + */ +static unsigned *get_label( struct st_translate *t, + unsigned branch_target ) +{ + unsigned i; + + if (t->labels_count + 1 >= t->labels_size) { + unsigned old_size = t->labels_size; + t->labels_size = 1 << (util_logbase2(t->labels_size) + 1); + t->labels = REALLOC( t->labels, + old_size * sizeof t->labels[0], + t->labels_size * sizeof t->labels[0] ); + if (t->labels == NULL) { + static unsigned dummy; + t->error = TRUE; + return &dummy; + } + } + + i = t->labels_count++; + t->labels[i].branch_target = branch_target; + return &t->labels[i].token; +} + + +/** + * Called prior to emitting the TGSI code for each Mesa instruction. + * Allocate additional space for instructions if needed. + * Update the insn[] array so the next Mesa instruction points to + * the next TGSI instruction. + */ +static void set_insn_start( struct st_translate *t, + unsigned start ) +{ + if (t->insn_count + 1 >= t->insn_size) { + unsigned old_size = t->insn_size; + t->insn_size = 1 << (util_logbase2(t->insn_size) + 1); + t->insn = REALLOC( t->insn, + old_size * sizeof t->insn[0], + t->insn_size * sizeof t->insn[0] ); + if (t->insn == NULL) { + t->error = TRUE; + return; + } + } + + t->insn[t->insn_count++] = start; +} + + +/** + * Map a Mesa dst register to a TGSI ureg_dst register. + */ +static struct ureg_dst +dst_register( struct st_translate *t, + gl_register_file file, + GLuint index ) +{ + switch( file ) { + case PROGRAM_UNDEFINED: + return ureg_dst_undef(); + + case PROGRAM_TEMPORARY: + if (ureg_dst_is_undef(t->temps[index])) + t->temps[index] = ureg_DECL_temporary( t->ureg ); + + return t->temps[index]; + + case PROGRAM_OUTPUT: + if (t->procType == TGSI_PROCESSOR_VERTEX && index == VERT_RESULT_PSIZ) + t->prevInstWrotePointSize = GL_TRUE; + + if (t->procType == TGSI_PROCESSOR_VERTEX) + assert(index < VERT_RESULT_MAX); + else if (t->procType == TGSI_PROCESSOR_FRAGMENT) + assert(index < FRAG_RESULT_MAX); + else + assert(index < GEOM_RESULT_MAX); + + assert(t->outputMapping[index] < Elements(t->outputs)); + + return t->outputs[t->outputMapping[index]]; + + case PROGRAM_ADDRESS: + return t->address[index]; + + default: + debug_assert( 0 ); + return ureg_dst_undef(); + } +} + + +/** + * Map a Mesa src register to a TGSI ureg_src register. + */ +static struct ureg_src +src_register( struct st_translate *t, + gl_register_file file, + GLint index ) +{ + switch( file ) { + case PROGRAM_UNDEFINED: + return ureg_src_undef(); + + case PROGRAM_TEMPORARY: + assert(index >= 0); + if (ureg_dst_is_undef(t->temps[index])) + t->temps[index] = ureg_DECL_temporary( t->ureg ); + assert(index < Elements(t->temps)); + return ureg_src(t->temps[index]); + + case PROGRAM_NAMED_PARAM: + case PROGRAM_ENV_PARAM: + case PROGRAM_LOCAL_PARAM: + case PROGRAM_UNIFORM: + assert(index >= 0); + return t->constants[index]; + case PROGRAM_STATE_VAR: + case PROGRAM_CONSTANT: /* ie, immediate */ + if (index < 0) + return ureg_DECL_constant( t->ureg, 0 ); + else + return t->constants[index]; + + case PROGRAM_INPUT: + assert(t->inputMapping[index] < Elements(t->inputs)); + return t->inputs[t->inputMapping[index]]; + + case PROGRAM_OUTPUT: + assert(t->outputMapping[index] < Elements(t->outputs)); + return ureg_src(t->outputs[t->outputMapping[index]]); /* not needed? */ + + case PROGRAM_ADDRESS: + return ureg_src(t->address[index]); + + case PROGRAM_SYSTEM_VALUE: + assert(index < Elements(t->systemValues)); + return t->systemValues[index]; + + default: + debug_assert( 0 ); + return ureg_src_undef(); + } +} + + +/** + * Map mesa texture target to TGSI texture target. + */ +static unsigned +translate_texture_target( GLuint textarget, + GLboolean shadow ) +{ + if (shadow) { + switch( textarget ) { + case TEXTURE_1D_INDEX: return TGSI_TEXTURE_SHADOW1D; + case TEXTURE_2D_INDEX: return TGSI_TEXTURE_SHADOW2D; + case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_SHADOWRECT; + default: break; + } + } + + switch( textarget ) { + case TEXTURE_1D_INDEX: return TGSI_TEXTURE_1D; + case TEXTURE_2D_INDEX: return TGSI_TEXTURE_2D; + case TEXTURE_3D_INDEX: return TGSI_TEXTURE_3D; + case TEXTURE_CUBE_INDEX: return TGSI_TEXTURE_CUBE; + case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_RECT; + case TEXTURE_1D_ARRAY_INDEX: return TGSI_TEXTURE_1D_ARRAY; + case TEXTURE_2D_ARRAY_INDEX: return TGSI_TEXTURE_2D_ARRAY; + default: + debug_assert( 0 ); + return TGSI_TEXTURE_1D; + } +} + + +/** + * Create a TGSI ureg_dst register from a Mesa dest register. + */ +static struct ureg_dst +translate_dst( struct st_translate *t, + const struct prog_dst_register *DstReg, + boolean saturate ) +{ + struct ureg_dst dst = dst_register( t, + DstReg->File, + DstReg->Index ); + + dst = ureg_writemask( dst, + DstReg->WriteMask ); + + if (saturate) + dst = ureg_saturate( dst ); + + if (DstReg->RelAddr) + dst = ureg_dst_indirect( dst, ureg_src(t->address[0]) ); + + return dst; +} + + +/** + * Create a TGSI ureg_src register from a Mesa src register. + */ +static struct ureg_src +translate_src( struct st_translate *t, + const struct prog_src_register *SrcReg ) +{ + struct ureg_src src = src_register( t, SrcReg->File, SrcReg->Index ); + + if (t->procType == TGSI_PROCESSOR_GEOMETRY && SrcReg->HasIndex2) { + src = src_register( t, SrcReg->File, SrcReg->Index2 ); + if (SrcReg->RelAddr2) + src = ureg_src_dimension_indirect( src, ureg_src(t->address[0]), + SrcReg->Index); + else + src = ureg_src_dimension( src, SrcReg->Index); + } + + src = ureg_swizzle( src, + GET_SWZ( SrcReg->Swizzle, 0 ) & 0x3, + GET_SWZ( SrcReg->Swizzle, 1 ) & 0x3, + GET_SWZ( SrcReg->Swizzle, 2 ) & 0x3, + GET_SWZ( SrcReg->Swizzle, 3 ) & 0x3); + + if (SrcReg->Negate == NEGATE_XYZW) + src = ureg_negate(src); + + if (SrcReg->Abs) + src = ureg_abs(src); + + if (SrcReg->RelAddr) { + src = ureg_src_indirect( src, ureg_src(t->address[0])); + if (SrcReg->File != PROGRAM_INPUT && + SrcReg->File != PROGRAM_OUTPUT) { + /* If SrcReg->Index was negative, it was set to zero in + * src_register(). Reassign it now. But don't do this + * for input/output regs since they get remapped while + * const buffers don't. + */ + src.Index = SrcReg->Index; + } + } + + return src; +} + + +static struct ureg_src swizzle_4v( struct ureg_src src, + const unsigned *swz ) +{ + return ureg_swizzle( src, swz[0], swz[1], swz[2], swz[3] ); +} + + +/** + * Translate a SWZ instruction into a MOV, MUL or MAD instruction. EG: + * + * SWZ dst, src.x-y10 + * + * becomes: + * + * MAD dst {1,-1,0,0}, src.xyxx, {0,0,1,0} + */ +static void emit_swz( struct st_translate *t, + struct ureg_dst dst, + const struct prog_src_register *SrcReg ) +{ + struct ureg_program *ureg = t->ureg; + struct ureg_src src = src_register( t, SrcReg->File, SrcReg->Index ); + + unsigned negate_mask = SrcReg->Negate; + + unsigned one_mask = ((GET_SWZ(SrcReg->Swizzle, 0) == SWIZZLE_ONE) << 0 | + (GET_SWZ(SrcReg->Swizzle, 1) == SWIZZLE_ONE) << 1 | + (GET_SWZ(SrcReg->Swizzle, 2) == SWIZZLE_ONE) << 2 | + (GET_SWZ(SrcReg->Swizzle, 3) == SWIZZLE_ONE) << 3); + + unsigned zero_mask = ((GET_SWZ(SrcReg->Swizzle, 0) == SWIZZLE_ZERO) << 0 | + (GET_SWZ(SrcReg->Swizzle, 1) == SWIZZLE_ZERO) << 1 | + (GET_SWZ(SrcReg->Swizzle, 2) == SWIZZLE_ZERO) << 2 | + (GET_SWZ(SrcReg->Swizzle, 3) == SWIZZLE_ZERO) << 3); + + unsigned negative_one_mask = one_mask & negate_mask; + unsigned positive_one_mask = one_mask & ~negate_mask; + + struct ureg_src imm; + unsigned i; + unsigned mul_swizzle[4] = {0,0,0,0}; + unsigned add_swizzle[4] = {0,0,0,0}; + unsigned src_swizzle[4] = {0,0,0,0}; + boolean need_add = FALSE; + boolean need_mul = FALSE; + + if (dst.WriteMask == 0) + return; + + /* Is this just a MOV? + */ + if (zero_mask == 0 && + one_mask == 0 && + (negate_mask == 0 || negate_mask == TGSI_WRITEMASK_XYZW)) + { + ureg_MOV( ureg, dst, translate_src( t, SrcReg )); + return; + } + +#define IMM_ZERO 0 +#define IMM_ONE 1 +#define IMM_NEG_ONE 2 + + imm = ureg_imm3f( ureg, 0, 1, -1 ); + + for (i = 0; i < 4; i++) { + unsigned bit = 1 << i; + + if (dst.WriteMask & bit) { + if (positive_one_mask & bit) { + mul_swizzle[i] = IMM_ZERO; + add_swizzle[i] = IMM_ONE; + need_add = TRUE; + } + else if (negative_one_mask & bit) { + mul_swizzle[i] = IMM_ZERO; + add_swizzle[i] = IMM_NEG_ONE; + need_add = TRUE; + } + else if (zero_mask & bit) { + mul_swizzle[i] = IMM_ZERO; + add_swizzle[i] = IMM_ZERO; + need_add = TRUE; + } + else { + add_swizzle[i] = IMM_ZERO; + src_swizzle[i] = GET_SWZ(SrcReg->Swizzle, i); + need_mul = TRUE; + if (negate_mask & bit) { + mul_swizzle[i] = IMM_NEG_ONE; + } + else { + mul_swizzle[i] = IMM_ONE; + } + } + } + } + + if (need_mul && need_add) { + ureg_MAD( ureg, + dst, + swizzle_4v( src, src_swizzle ), + swizzle_4v( imm, mul_swizzle ), + swizzle_4v( imm, add_swizzle ) ); + } + else if (need_mul) { + ureg_MUL( ureg, + dst, + swizzle_4v( src, src_swizzle ), + swizzle_4v( imm, mul_swizzle ) ); + } + else if (need_add) { + ureg_MOV( ureg, + dst, + swizzle_4v( imm, add_swizzle ) ); + } + else { + debug_assert(0); + } + +#undef IMM_ZERO +#undef IMM_ONE +#undef IMM_NEG_ONE +} + + +/** + * Negate the value of DDY to match GL semantics where (0,0) is the + * lower-left corner of the window. + * Note that the GL_ARB_fragment_coord_conventions extension will + * effect this someday. + */ +static void emit_ddy( struct st_translate *t, + struct ureg_dst dst, + const struct prog_src_register *SrcReg ) +{ + struct ureg_program *ureg = t->ureg; + struct ureg_src src = translate_src( t, SrcReg ); + src = ureg_negate( src ); + ureg_DDY( ureg, dst, src ); +} + + + +static unsigned +translate_opcode( unsigned op ) +{ + switch( op ) { + case OPCODE_ARL: + return TGSI_OPCODE_ARL; + case OPCODE_ABS: + return TGSI_OPCODE_ABS; + case OPCODE_ADD: + return TGSI_OPCODE_ADD; + case OPCODE_BGNLOOP: + return TGSI_OPCODE_BGNLOOP; + case OPCODE_BGNSUB: + return TGSI_OPCODE_BGNSUB; + case OPCODE_BRA: + return TGSI_OPCODE_BRA; + case OPCODE_BRK: + return TGSI_OPCODE_BRK; + case OPCODE_CAL: + return TGSI_OPCODE_CAL; + case OPCODE_CMP: + return TGSI_OPCODE_CMP; + case OPCODE_CONT: + return TGSI_OPCODE_CONT; + case OPCODE_COS: + return TGSI_OPCODE_COS; + case OPCODE_DDX: + return TGSI_OPCODE_DDX; + case OPCODE_DDY: + return TGSI_OPCODE_DDY; + case OPCODE_DP2: + return TGSI_OPCODE_DP2; + case OPCODE_DP2A: + return TGSI_OPCODE_DP2A; + case OPCODE_DP3: + return TGSI_OPCODE_DP3; + case OPCODE_DP4: + return TGSI_OPCODE_DP4; + case OPCODE_DPH: + return TGSI_OPCODE_DPH; + case OPCODE_DST: + return TGSI_OPCODE_DST; + case OPCODE_ELSE: + return TGSI_OPCODE_ELSE; + case OPCODE_EMIT_VERTEX: + return TGSI_OPCODE_EMIT; + case OPCODE_END_PRIMITIVE: + return TGSI_OPCODE_ENDPRIM; + case OPCODE_ENDIF: + return TGSI_OPCODE_ENDIF; + case OPCODE_ENDLOOP: + return TGSI_OPCODE_ENDLOOP; + case OPCODE_ENDSUB: + return TGSI_OPCODE_ENDSUB; + case OPCODE_EX2: + return TGSI_OPCODE_EX2; + case OPCODE_EXP: + return TGSI_OPCODE_EXP; + case OPCODE_FLR: + return TGSI_OPCODE_FLR; + case OPCODE_FRC: + return TGSI_OPCODE_FRC; + case OPCODE_IF: + return TGSI_OPCODE_IF; + case OPCODE_TRUNC: + return TGSI_OPCODE_TRUNC; + case OPCODE_KIL: + return TGSI_OPCODE_KIL; + case OPCODE_KIL_NV: + return TGSI_OPCODE_KILP; + case OPCODE_LG2: + return TGSI_OPCODE_LG2; + case OPCODE_LOG: + return TGSI_OPCODE_LOG; + case OPCODE_LIT: + return TGSI_OPCODE_LIT; + case OPCODE_LRP: + return TGSI_OPCODE_LRP; + case OPCODE_MAD: + return TGSI_OPCODE_MAD; + case OPCODE_MAX: + return TGSI_OPCODE_MAX; + case OPCODE_MIN: + return TGSI_OPCODE_MIN; + case OPCODE_MOV: + return TGSI_OPCODE_MOV; + case OPCODE_MUL: + return TGSI_OPCODE_MUL; + case OPCODE_NOP: + return TGSI_OPCODE_NOP; + case OPCODE_NRM3: + return TGSI_OPCODE_NRM; + case OPCODE_NRM4: + return TGSI_OPCODE_NRM4; + case OPCODE_POW: + return TGSI_OPCODE_POW; + case OPCODE_RCP: + return TGSI_OPCODE_RCP; + case OPCODE_RET: + return TGSI_OPCODE_RET; + case OPCODE_RSQ: + return TGSI_OPCODE_RSQ; + case OPCODE_SCS: + return TGSI_OPCODE_SCS; + case OPCODE_SEQ: + return TGSI_OPCODE_SEQ; + case OPCODE_SGE: + return TGSI_OPCODE_SGE; + case OPCODE_SGT: + return TGSI_OPCODE_SGT; + case OPCODE_SIN: + return TGSI_OPCODE_SIN; + case OPCODE_SLE: + return TGSI_OPCODE_SLE; + case OPCODE_SLT: + return TGSI_OPCODE_SLT; + case OPCODE_SNE: + return TGSI_OPCODE_SNE; + case OPCODE_SSG: + return TGSI_OPCODE_SSG; + case OPCODE_SUB: + return TGSI_OPCODE_SUB; + case OPCODE_TEX: + return TGSI_OPCODE_TEX; + case OPCODE_TXB: + return TGSI_OPCODE_TXB; + case OPCODE_TXD: + return TGSI_OPCODE_TXD; + case OPCODE_TXL: + return TGSI_OPCODE_TXL; + case OPCODE_TXP: + return TGSI_OPCODE_TXP; + case OPCODE_XPD: + return TGSI_OPCODE_XPD; + case OPCODE_END: + return TGSI_OPCODE_END; + default: + debug_assert( 0 ); + return TGSI_OPCODE_NOP; + } +} + + +static void +compile_instruction( + struct st_translate *t, + const struct prog_instruction *inst ) +{ + struct ureg_program *ureg = t->ureg; + GLuint i; + struct ureg_dst dst[1]; + struct ureg_src src[4]; + unsigned num_dst; + unsigned num_src; + + num_dst = _mesa_num_inst_dst_regs( inst->Opcode ); + num_src = _mesa_num_inst_src_regs( inst->Opcode ); + + if (num_dst) + dst[0] = translate_dst( t, + &inst->DstReg, + inst->SaturateMode ); + + for (i = 0; i < num_src; i++) + src[i] = translate_src( t, &inst->SrcReg[i] ); + + switch( inst->Opcode ) { + case OPCODE_SWZ: + emit_swz( t, dst[0], &inst->SrcReg[0] ); + return; + + case OPCODE_BGNLOOP: + case OPCODE_CAL: + case OPCODE_ELSE: + case OPCODE_ENDLOOP: + case OPCODE_IF: + debug_assert(num_dst == 0); + ureg_label_insn( ureg, + translate_opcode( inst->Opcode ), + src, num_src, + get_label( t, inst->BranchTarget )); + return; + + case OPCODE_TEX: + case OPCODE_TXB: + case OPCODE_TXD: + case OPCODE_TXL: + case OPCODE_TXP: + src[num_src++] = t->samplers[inst->TexSrcUnit]; + ureg_tex_insn( ureg, + translate_opcode( inst->Opcode ), + dst, num_dst, + translate_texture_target( inst->TexSrcTarget, + inst->TexShadow ), + src, num_src ); + return; + + case OPCODE_SCS: + dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XY ); + ureg_insn( ureg, + translate_opcode( inst->Opcode ), + dst, num_dst, + src, num_src ); + break; + + case OPCODE_XPD: + dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XYZ ); + ureg_insn( ureg, + translate_opcode( inst->Opcode ), + dst, num_dst, + src, num_src ); + break; + + case OPCODE_NOISE1: + case OPCODE_NOISE2: + case OPCODE_NOISE3: + case OPCODE_NOISE4: + /* At some point, a motivated person could add a better + * implementation of noise. Currently not even the nvidia + * binary drivers do anything more than this. In any case, the + * place to do this is in the GL state tracker, not the poor + * driver. + */ + ureg_MOV( ureg, dst[0], ureg_imm1f(ureg, 0.5) ); + break; + + case OPCODE_DDY: + emit_ddy( t, dst[0], &inst->SrcReg[0] ); + break; + + default: + ureg_insn( ureg, + translate_opcode( inst->Opcode ), + dst, num_dst, + src, num_src ); + break; + } +} + + +/** + * Emit the TGSI instructions to adjust the WPOS pixel center convention + * Basically, add (adjX, adjY) to the fragment position. + */ +static void +emit_adjusted_wpos( struct st_translate *t, + const struct gl_program *program, + GLfloat adjX, GLfloat adjY) +{ + struct ureg_program *ureg = t->ureg; + struct ureg_dst wpos_temp = ureg_DECL_temporary(ureg); + struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]]; + + /* Note that we bias X and Y and pass Z and W through unchanged. + * The shader might also use gl_FragCoord.w and .z. + */ + ureg_ADD(ureg, wpos_temp, wpos_input, + ureg_imm4f(ureg, adjX, adjY, 0.0f, 0.0f)); + + t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]] = ureg_src(wpos_temp); +} + + +/** + * Emit the TGSI instructions for inverting the WPOS y coordinate. + * This code is unavoidable because it also depends on whether + * a FBO is bound (STATE_FB_WPOS_Y_TRANSFORM). + */ +static void +emit_wpos_inversion( struct st_translate *t, + const struct gl_program *program, + boolean invert) +{ + struct ureg_program *ureg = t->ureg; + + /* Fragment program uses fragment position input. + * Need to replace instances of INPUT[WPOS] with temp T + * where T = INPUT[WPOS] by y is inverted. + */ + static const gl_state_index wposTransformState[STATE_LENGTH] + = { STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM, 0, 0, 0 }; + + /* XXX: note we are modifying the incoming shader here! Need to + * do this before emitting the constant decls below, or this + * will be missed: + */ + unsigned wposTransConst = _mesa_add_state_reference(program->Parameters, + wposTransformState); + + struct ureg_src wpostrans = ureg_DECL_constant( ureg, wposTransConst ); + struct ureg_dst wpos_temp; + struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]]; + + /* MOV wpos_temp, input[wpos] + */ + if (wpos_input.File == TGSI_FILE_TEMPORARY) + wpos_temp = ureg_dst(wpos_input); + else { + wpos_temp = ureg_DECL_temporary( ureg ); + ureg_MOV( ureg, wpos_temp, wpos_input ); + } + + if (invert) { + /* MAD wpos_temp.y, wpos_input, wpostrans.xxxx, wpostrans.yyyy + */ + ureg_MAD( ureg, + ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ), + wpos_input, + ureg_scalar(wpostrans, 0), + ureg_scalar(wpostrans, 1)); + } else { + /* MAD wpos_temp.y, wpos_input, wpostrans.zzzz, wpostrans.wwww + */ + ureg_MAD( ureg, + ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ), + wpos_input, + ureg_scalar(wpostrans, 2), + ureg_scalar(wpostrans, 3)); + } + + /* Use wpos_temp as position input from here on: + */ + t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]] = ureg_src(wpos_temp); +} + + +/** + * Emit fragment position/ooordinate code. + */ +static void +emit_wpos(struct st_context *st, + struct st_translate *t, + const struct gl_program *program, + struct ureg_program *ureg) +{ + const struct gl_fragment_program *fp = + (const struct gl_fragment_program *) program; + struct pipe_screen *pscreen = st->pipe->screen; + boolean invert = FALSE; + + if (fp->OriginUpperLeft) { + /* Fragment shader wants origin in upper-left */ + if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT)) { + /* the driver supports upper-left origin */ + } + else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT)) { + /* the driver supports lower-left origin, need to invert Y */ + ureg_property_fs_coord_origin(ureg, TGSI_FS_COORD_ORIGIN_LOWER_LEFT); + invert = TRUE; + } + else + assert(0); + } + else { + /* Fragment shader wants origin in lower-left */ + if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT)) + /* the driver supports lower-left origin */ + ureg_property_fs_coord_origin(ureg, TGSI_FS_COORD_ORIGIN_LOWER_LEFT); + else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT)) + /* the driver supports upper-left origin, need to invert Y */ + invert = TRUE; + else + assert(0); + } + + if (fp->PixelCenterInteger) { + /* Fragment shader wants pixel center integer */ + if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) + /* the driver supports pixel center integer */ + ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER); + else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) + /* the driver supports pixel center half integer, need to bias X,Y */ + emit_adjusted_wpos(t, program, 0.5f, invert ? 0.5f : -0.5f); + else + assert(0); + } + else { + /* Fragment shader wants pixel center half integer */ + if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) { + /* the driver supports pixel center half integer */ + } + else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) { + /* the driver supports pixel center integer, need to bias X,Y */ + ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER); + emit_adjusted_wpos(t, program, 0.5f, invert ? -0.5f : 0.5f); + } + else + assert(0); + } + + /* we invert after adjustment so that we avoid the MOV to temporary, + * and reuse the adjustment ADD instead */ + emit_wpos_inversion(t, program, invert); +} + + +/** + * OpenGL's fragment gl_FrontFace input is 1 for front-facing, 0 for back. + * TGSI uses +1 for front, -1 for back. + * This function converts the TGSI value to the GL value. Simply clamping/ + * saturating the value to [0,1] does the job. + */ +static void +emit_face_var( struct st_translate *t, + const struct gl_program *program ) +{ + struct ureg_program *ureg = t->ureg; + struct ureg_dst face_temp = ureg_DECL_temporary( ureg ); + struct ureg_src face_input = t->inputs[t->inputMapping[FRAG_ATTRIB_FACE]]; + + /* MOV_SAT face_temp, input[face] + */ + face_temp = ureg_saturate( face_temp ); + ureg_MOV( ureg, face_temp, face_input ); + + /* Use face_temp as face input from here on: + */ + t->inputs[t->inputMapping[FRAG_ATTRIB_FACE]] = ureg_src(face_temp); +} + + +static void +emit_edgeflags( struct st_translate *t, + const struct gl_program *program ) +{ + struct ureg_program *ureg = t->ureg; + struct ureg_dst edge_dst = t->outputs[t->outputMapping[VERT_RESULT_EDGE]]; + struct ureg_src edge_src = t->inputs[t->inputMapping[VERT_ATTRIB_EDGEFLAG]]; + + ureg_MOV( ureg, edge_dst, edge_src ); +} + + +/** + * Translate Mesa program to TGSI format. + * \param program the program to translate + * \param numInputs number of input registers used + * \param inputMapping maps Mesa fragment program inputs to TGSI generic + * input indexes + * \param inputSemanticName the TGSI_SEMANTIC flag for each input + * \param inputSemanticIndex the semantic index (ex: which texcoord) for + * each input + * \param interpMode the TGSI_INTERPOLATE_LINEAR/PERSP mode for each input + * \param numOutputs number of output registers used + * \param outputMapping maps Mesa fragment program outputs to TGSI + * generic outputs + * \param outputSemanticName the TGSI_SEMANTIC flag for each output + * \param outputSemanticIndex the semantic index (ex: which texcoord) for + * each output + * + * \return PIPE_OK or PIPE_ERROR_OUT_OF_MEMORY + */ +enum pipe_error +st_translate_mesa_program( + struct gl_context *ctx, + uint procType, + struct ureg_program *ureg, + const struct gl_program *program, + GLuint numInputs, + const GLuint inputMapping[], + const ubyte inputSemanticName[], + const ubyte inputSemanticIndex[], + const GLuint interpMode[], + GLuint numOutputs, + const GLuint outputMapping[], + const ubyte outputSemanticName[], + const ubyte outputSemanticIndex[], + boolean passthrough_edgeflags ) +{ + struct st_translate translate, *t; + unsigned i; + enum pipe_error ret = PIPE_OK; + + assert(numInputs <= Elements(t->inputs)); + assert(numOutputs <= Elements(t->outputs)); + + t = &translate; + memset(t, 0, sizeof *t); + + t->procType = procType; + t->inputMapping = inputMapping; + t->outputMapping = outputMapping; + t->ureg = ureg; + t->pointSizeOutIndex = -1; + t->prevInstWrotePointSize = GL_FALSE; + + /*_mesa_print_program(program);*/ + + /* + * Declare input attributes. + */ + if (procType == TGSI_PROCESSOR_FRAGMENT) { + for (i = 0; i < numInputs; i++) { + if (program->InputFlags[0] & PROG_PARAM_BIT_CYL_WRAP) { + t->inputs[i] = ureg_DECL_fs_input_cyl(ureg, + inputSemanticName[i], + inputSemanticIndex[i], + interpMode[i], + TGSI_CYLINDRICAL_WRAP_X); + } + else { + t->inputs[i] = ureg_DECL_fs_input(ureg, + inputSemanticName[i], + inputSemanticIndex[i], + interpMode[i]); + } + } + + if (program->InputsRead & FRAG_BIT_WPOS) { + /* Must do this after setting up t->inputs, and before + * emitting constant references, below: + */ + emit_wpos(st_context(ctx), t, program, ureg); + } + + if (program->InputsRead & FRAG_BIT_FACE) { + emit_face_var( t, program ); + } + + /* + * Declare output attributes. + */ + for (i = 0; i < numOutputs; i++) { + switch (outputSemanticName[i]) { + case TGSI_SEMANTIC_POSITION: + t->outputs[i] = ureg_DECL_output( ureg, + TGSI_SEMANTIC_POSITION, /* Z / Depth */ + outputSemanticIndex[i] ); + + t->outputs[i] = ureg_writemask( t->outputs[i], + TGSI_WRITEMASK_Z ); + break; + case TGSI_SEMANTIC_STENCIL: + t->outputs[i] = ureg_DECL_output( ureg, + TGSI_SEMANTIC_STENCIL, /* Stencil */ + outputSemanticIndex[i] ); + t->outputs[i] = ureg_writemask( t->outputs[i], + TGSI_WRITEMASK_Y ); + break; + case TGSI_SEMANTIC_COLOR: + t->outputs[i] = ureg_DECL_output( ureg, + TGSI_SEMANTIC_COLOR, + outputSemanticIndex[i] ); + break; + default: + debug_assert(0); + return 0; + } + } + } + else if (procType == TGSI_PROCESSOR_GEOMETRY) { + for (i = 0; i < numInputs; i++) { + t->inputs[i] = ureg_DECL_gs_input(ureg, + i, + inputSemanticName[i], + inputSemanticIndex[i]); + } + + for (i = 0; i < numOutputs; i++) { + t->outputs[i] = ureg_DECL_output( ureg, + outputSemanticName[i], + outputSemanticIndex[i] ); + } + } + else { + assert(procType == TGSI_PROCESSOR_VERTEX); + + for (i = 0; i < numInputs; i++) { + t->inputs[i] = ureg_DECL_vs_input(ureg, i); + } + + for (i = 0; i < numOutputs; i++) { + t->outputs[i] = ureg_DECL_output( ureg, + outputSemanticName[i], + outputSemanticIndex[i] ); + if ((outputSemanticName[i] == TGSI_SEMANTIC_PSIZE) && program->Id) { + /* Writing to the point size result register requires special + * handling to implement clamping. + */ + static const gl_state_index pointSizeClampState[STATE_LENGTH] + = { STATE_INTERNAL, STATE_POINT_SIZE_IMPL_CLAMP, 0, 0, 0 }; + /* XXX: note we are modifying the incoming shader here! Need to + * do this before emitting the constant decls below, or this + * will be missed: + */ + unsigned pointSizeClampConst = + _mesa_add_state_reference(program->Parameters, + pointSizeClampState); + struct ureg_dst psizregtemp = ureg_DECL_temporary( ureg ); + t->pointSizeConst = ureg_DECL_constant( ureg, pointSizeClampConst ); + t->pointSizeResult = t->outputs[i]; + t->pointSizeOutIndex = i; + t->outputs[i] = psizregtemp; + } + } + if (passthrough_edgeflags) + emit_edgeflags( t, program ); + } + + /* Declare address register. + */ + if (program->NumAddressRegs > 0) { + debug_assert( program->NumAddressRegs == 1 ); + t->address[0] = ureg_DECL_address( ureg ); + } + + /* Declare misc input registers + */ + { + GLbitfield sysInputs = program->SystemValuesRead; + unsigned numSys = 0; + for (i = 0; sysInputs; i++) { + if (sysInputs & (1 << i)) { + unsigned semName = mesa_sysval_to_semantic[i]; + t->systemValues[i] = ureg_DECL_system_value(ureg, numSys, semName, 0); + numSys++; + sysInputs &= ~(1 << i); + } + } + } + + if (program->IndirectRegisterFiles & (1 << PROGRAM_TEMPORARY)) { + /* If temps are accessed with indirect addressing, declare temporaries + * in sequential order. Else, we declare them on demand elsewhere. + */ + for (i = 0; i < program->NumTemporaries; i++) { + /* XXX use TGSI_FILE_TEMPORARY_ARRAY when it's supported by ureg */ + t->temps[i] = ureg_DECL_temporary( t->ureg ); + } + } + + /* Emit constants and immediates. Mesa uses a single index space + * for these, so we put all the translated regs in t->constants. + */ + if (program->Parameters) { + t->constants = CALLOC( program->Parameters->NumParameters, + sizeof t->constants[0] ); + if (t->constants == NULL) { + ret = PIPE_ERROR_OUT_OF_MEMORY; + goto out; + } + + for (i = 0; i < program->Parameters->NumParameters; i++) { + switch (program->Parameters->Parameters[i].Type) { + case PROGRAM_ENV_PARAM: + case PROGRAM_LOCAL_PARAM: + case PROGRAM_STATE_VAR: + case PROGRAM_NAMED_PARAM: + case PROGRAM_UNIFORM: + t->constants[i] = ureg_DECL_constant( ureg, i ); + break; + + /* Emit immediates only when there's no indirect addressing of + * the const buffer. + * FIXME: Be smarter and recognize param arrays: + * indirect addressing is only valid within the referenced + * array. + */ + case PROGRAM_CONSTANT: + if (program->IndirectRegisterFiles & PROGRAM_ANY_CONST) + t->constants[i] = ureg_DECL_constant( ureg, i ); + else + t->constants[i] = + ureg_DECL_immediate( ureg, + program->Parameters->ParameterValues[i], + 4 ); + break; + default: + break; + } + } + } + + /* texture samplers */ + for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) { + if (program->SamplersUsed & (1 << i)) { + t->samplers[i] = ureg_DECL_sampler( ureg, i ); + } + } + + /* Emit each instruction in turn: + */ + for (i = 0; i < program->NumInstructions; i++) { + set_insn_start( t, ureg_get_instruction_number( ureg )); + compile_instruction( t, &program->Instructions[i] ); + + if (t->prevInstWrotePointSize && program->Id) { + /* The previous instruction wrote to the (fake) vertex point size + * result register. Now we need to clamp that value to the min/max + * point size range, putting the result into the real point size + * register. + * Note that we can't do this easily at the end of program due to + * possible early return. + */ + set_insn_start( t, ureg_get_instruction_number( ureg )); + ureg_MAX( t->ureg, + ureg_writemask(t->outputs[t->pointSizeOutIndex], WRITEMASK_X), + ureg_src(t->outputs[t->pointSizeOutIndex]), + ureg_swizzle(t->pointSizeConst, 1,1,1,1)); + ureg_MIN( t->ureg, ureg_writemask(t->pointSizeResult, WRITEMASK_X), + ureg_src(t->outputs[t->pointSizeOutIndex]), + ureg_swizzle(t->pointSizeConst, 2,2,2,2)); + } + t->prevInstWrotePointSize = GL_FALSE; + } + + /* Fix up all emitted labels: + */ + for (i = 0; i < t->labels_count; i++) { + ureg_fixup_label( ureg, + t->labels[i].token, + t->insn[t->labels[i].branch_target] ); + } + +out: + FREE(t->insn); + FREE(t->labels); + FREE(t->constants); + + if (t->error) { + debug_printf("%s: translate error flag set\n", __FUNCTION__); + } + + return ret; +} + + +/** + * Tokens cannot be free with free otherwise the builtin gallium + * malloc debugging will get confused. + */ +void +st_free_tokens(const struct tgsi_token *tokens) +{ + FREE((void *)tokens); +} -- cgit v1.2.3