--- /dev/null
- #include "shader_api.h"
+/*
+ * 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 <ian.d.romanick@intel.com>
+ */
+#include <cstdlib>
+#include <cstdio>
+#include <cstdarg>
+#include <climits>
+
+extern "C" {
+#include <talloc.h>
+}
+
+#include "main/mtypes.h"
+#include "main/macros.h"
++#include "main/shaderobj.h"
+#include "glsl_symbol_table.h"
+#include "ir.h"
+#include "program.h"
+#include "hash_table.h"
+#include "linker.h"
+#include "ir_optimization.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;
+ }
+
+ 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? */
+};
+
+
+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->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++) {
+ 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) {
+ 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;
+ }
+
+ /* 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(NULL);
+ }
+ } else
+ variables.add_variable(var->name, 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,
+ prog->_NumLinkedShaders, 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->name, 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) {
+ linker_error_printf(prog,
+ "%s shader output `%s' delcared 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->name, func);
+ } else if ((var = inst->as_variable()) != NULL) {
+ sh->symbols->add_variable(var->name, 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, glsl_symbol_table *symbols,
+ exec_list *instructions, hash_table *temps)
+{
+ class remap_visitor : public ir_hierarchical_visitor {
+ public:
+ remap_visitor(glsl_symbol_table *symbols, exec_list *instructions,
+ hash_table *temps)
+ {
+ this->symbols = symbols;
+ this->instructions = instructions;
+ 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(NULL);
+
+ this->symbols->add_variable(copy->name, copy);
+ this->instructions->push_head(copy);
+ ir->var = copy;
+ }
+
+ return visit_continue;
+ }
+
+ private:
+ glsl_symbol_table *symbols;
+ exec_list *instructions;
+ hash_table *temps;
+ };
+
+ remap_visitor v(symbols, instructions, 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 sentinal 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(NULL);
+
+ if (var != NULL)
+ hash_table_insert(temps, inst, var);
+ else
+ remap_variables(inst, target->symbols, target->ir, 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(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_built_in)
+ continue;
+
+ ir_function_signature *other_sig =
+ other->exact_matching_signature(& sig->parameters);
+
+ if ((other_sig != NULL) && other_sig->is_defined
+ && !other_sig->is_built_in) {
+ 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 *const linked = _mesa_new_shader(NULL, 0, main->Type);
+ linked->ir = new(linked) exec_list;
+ clone_ir_list(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);
+
+ link_function_calls(prog, linked, linking_shaders, num_linking_shaders);
+
+ free(linking_shaders);
+
+ return linked;
+}
+
+
+struct uniform_node {
+ exec_node link;
+ struct gl_uniform *u;
+ unsigned 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);
+
+ for (unsigned i = 0; i < prog->_NumLinkedShaders; i++) {
+ 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;
+
+ const unsigned vec4_slots = (var->component_slots() + 3) / 4;
+ assert(vec4_slots != 0);
+
+ uniform_node *n = (uniform_node *) hash_table_find(ht, var->name);
+ if (n == NULL) {
+ n = (uniform_node *) calloc(1, sizeof(struct uniform_node));
+ n->u = (gl_uniform *) calloc(vec4_slots, sizeof(struct gl_uniform));
+ n->slots = vec4_slots;
+
+ n->u[0].Name = strdup(var->name);
+ for (unsigned j = 1; j < vec4_slots; j++)
+ n->u[j].Name = n->u[0].Name;
+
+ hash_table_insert(ht, n, n->u[0].Name);
+ uniforms.push_tail(& n->link);
+ total_uniforms += vec4_slots;
+ }
+
+ if (var->constant_value != NULL)
+ for (unsigned j = 0; j < vec4_slots; j++)
+ n->u[j].Initialized = true;
+
+ var->location = next_position;
+
+ for (unsigned j = 0; j < vec4_slots; j++) {
+ switch (prog->_LinkedShaders[i]->Type) {
+ case GL_VERTEX_SHADER:
+ n->u[j].VertPos = next_position;
+ break;
+ case GL_FRAGMENT_SHADER:
+ n->u[j].FragPos = next_position;
+ break;
+ case GL_GEOMETRY_SHADER:
+ /* FINISHME: Support geometry shaders. */
+ assert(prog->_LinkedShaders[i]->Type != GL_GEOMETRY_SHADER);
+ break;
+ }
+
+ next_position++;
+ }
+ }
+ }
+
+ 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) * node->slots);
+ idx += node->slots;
+
+ 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;
+
+ /* 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.
+ */
+ 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;
+}
+
+
+void
+assign_varying_locations(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);
+
+ /* FINISHME: Location assignment will need some changes when arrays,
+ * FINISHME: matrices, and structures are allowed as shader inputs /
+ * FINISHME: outputs.
+ */
+ output_var->location = output_index;
+ input_var->location = input_index;
+
+ output_index++;
+ input_index++;
+ }
+
+ foreach_list(node, producer->ir) {
+ ir_variable *const var = ((ir_instruction *) node)->as_variable();
+
+ if ((var == NULL) || (var->mode != ir_var_out))
+ continue;
+
+ /* An 'out' variable is only really a shader output if its value is read
+ * by the following stage.
+ */
+ if (var->location == -1) {
+ var->shader_out = false;
+ var->mode = ir_var_auto;
+ }
+ }
+
+ foreach_list(node, consumer->ir) {
+ ir_variable *const var = ((ir_instruction *) node)->as_variable();
+
+ if ((var == NULL) || (var->mode != ir_var_in))
+ continue;
+
+ /* An 'in' variable is only really a shader input if its value is written
+ * by the previous stage.
+ */
+ var->shader_in = (var->location != -1);
+ }
+}
+
+
+void
+link_shaders(struct gl_shader_program *prog)
+{
+ 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 >= 110);
+ assert(max_version <= 130);
+ if ((max_version >= 130) && (min_version != max_version)) {
+ linker_error_printf(prog, "all shaders must use same shading "
+ "language version\n");
+ goto done;
+ }
+
+ prog->Version = max_version;
+
+ /* Link all shaders for a particular stage and validate the result.
+ */
+ prog->_NumLinkedShaders = 0;
+ if (num_vert_shaders > 0) {
+ gl_shader *const sh =
+ link_intrastage_shaders(prog, vert_shader_list, num_vert_shaders);
+
+ if (sh == NULL)
+ goto done;
+
+ if (!validate_vertex_shader_executable(prog, sh))
+ goto done;
+
+ prog->_LinkedShaders[prog->_NumLinkedShaders] = sh;
+ prog->_NumLinkedShaders++;
+ }
+
+ if (num_frag_shaders > 0) {
+ gl_shader *const sh =
+ link_intrastage_shaders(prog, frag_shader_list, num_frag_shaders);
+
+ if (sh == NULL)
+ goto done;
+
+ if (!validate_fragment_shader_executable(prog, sh))
+ goto done;
+
+ prog->_LinkedShaders[prog->_NumLinkedShaders] = sh;
+ prog->_NumLinkedShaders++;
+ }
+
+ /* 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)) {
+ /* Validate the inputs of each stage with the output of the preceeding
+ * stage.
+ */
+ for (unsigned i = 1; i < prog->_NumLinkedShaders; i++) {
+ if (!cross_validate_outputs_to_inputs(prog,
+ prog->_LinkedShaders[i - 1],
+ prog->_LinkedShaders[i]))
+ goto done;
+ }
+
+ prog->LinkStatus = true;
+ }
+
+ /* FINISHME: Perform whole-program optimization here. */
+ for (unsigned i = 0; i < prog->_NumLinkedShaders; i++) {
+ /* Optimization passes */
+ bool progress;
+ exec_list *ir = prog->_LinkedShaders[i]->ir;
+
+ /* Lowering */
+ do_mat_op_to_vec(ir);
+ do_mod_to_fract(ir);
+ do_div_to_mul_rcp(ir);
+
+ do {
+ progress = false;
+
+ progress = do_function_inlining(ir) || progress;
+ progress = do_if_simplification(ir) || progress;
+ progress = do_copy_propagation(ir) || progress;
+ progress = do_dead_code_local(ir) || progress;
+#if 0
+ progress = do_dead_code_unlinked(state, ir) || progress;
+#endif
+ progress = do_constant_variable_unlinked(ir) || progress;
+ progress = do_constant_folding(ir) || progress;
+ progress = do_if_return(ir) || progress;
+#if 0
+ if (ctx->Shader.EmitNoIfs)
+ progress = do_if_to_cond_assign(ir) || progress;
+#endif
+
+ progress = do_vec_index_to_swizzle(ir) || progress;
+ /* Do this one after the previous to let the easier pass handle
+ * constant vector indexing.
+ */
+ progress = do_vec_index_to_cond_assign(ir) || progress;
+
+ progress = do_swizzle_swizzle(ir) || progress;
+ } while (progress);
+ }
+
+ assign_uniform_locations(prog);
+
+ if (prog->_LinkedShaders[0]->Type == GL_VERTEX_SHADER)
+ /* 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))
+ goto done;
+
+ for (unsigned i = 1; i < prog->_NumLinkedShaders; i++)
+ assign_varying_locations(prog->_LinkedShaders[i - 1],
+ prog->_LinkedShaders[i]);
+
+ /* FINISHME: Assign fragment shader output locations. */
+
+done:
+ free(vert_shader_list);
+}
--- /dev/null
- #include "shader/prog_instruction.h"
- #include "shader/prog_optimize.h"
- #include "shader/prog_print.h"
- #include "shader/program.h"
- #include "shader/prog_uniform.h"
- #include "shader/prog_parameter.h"
- #include "shader/shader_api.h"
+/*
+ * 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
+ *
+ * Translates the IR to ARB_fragment_program text if possible,
+ * printing the result
+ */
+
+#include <stdio.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/shaderobj.h"
++#include "main/uniforms.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"
+}
+
+/**
+ * This struct is a corresponding struct to Mesa prog_src_register, with
+ * wider fields.
+ */
+typedef struct ir_to_mesa_src_reg {
+ int 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:
+ 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;
+ 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, int file, int index)
+ : file(file), index(index), var(var)
+ {
+ /* empty */
+ }
+
+ int 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();
+
+ function_entry *current_function;
+
+ GLcontext *ctx;
+ struct gl_program *prog;
+
+ 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);
+
+ 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);
+
+ GLboolean try_emit_mad(ir_expression *ir,
+ int mul_operand);
+
+ int *sampler_map;
+ int sampler_map_size;
+
+ void map_sampler(int location, int sampler);
+ int get_sampler_number(int location);
+
+ void *mem_ctx;
+};
+
+ir_to_mesa_src_reg ir_to_mesa_undef = {
+ PROGRAM_UNDEFINED, 0, SWIZZLE_NOOP, NEGATE_NONE, 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 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),
+ };
+
+ 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)
+{
+ 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::map_sampler(int location, int sampler)
+{
+ if (this->sampler_map_size <= location) {
+ this->sampler_map = talloc_realloc(this->mem_ctx, this->sampler_map,
+ int, location + 1);
+ this->sampler_map_size = location + 1;
+ }
+
+ this->sampler_map[location] = sampler;
+}
+
+int
+ir_to_mesa_visitor::get_sampler_number(int location)
+{
+ assert(location < this->sampler_map_size);
+ return this->sampler_map[location];
+}
+
+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)
+{
+ ir_to_mesa_src_reg src_reg;
+
+ src_reg.file = reg.file;
+ src_reg.index = reg.index;
+ src_reg.swizzle = SWIZZLE_XYZW;
+ src_reg.negate = 0;
+ src_reg.reladdr = reg.reladdr;
+
+ return src_reg;
+}
+
+/**
+ * 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 (!(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);
+}
+
+struct ir_to_mesa_src_reg
+ir_to_mesa_visitor::src_reg_for_float(float val)
+{
+ ir_to_mesa_src_reg src_reg;
+
+ src_reg.file = PROGRAM_CONSTANT;
+ src_reg.index = _mesa_add_unnamed_constant(this->prog->Parameters,
+ &val, 1, &src_reg.swizzle);
+ src_reg.reladdr = NULL;
+ src_reg.negate = 0;
+
+ 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;
+ default:
+ assert(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;
+
+ assert(!type->is_array());
+
+ src_reg.file = PROGRAM_TEMPORARY;
+ src_reg.index = next_temp;
+ src_reg.reladdr = NULL;
+ next_temp += type_size(type);
+
+ 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)
+{
+ (void)ir;
+}
+
+void
+ir_to_mesa_visitor::visit(ir_loop *ir)
+{
+ assert(!ir->from);
+ assert(!ir->to);
+ assert(!ir->increment);
+ assert(!ir->counter);
+
+ ir_to_mesa_emit_op0(NULL, OPCODE_BGNLOOP);
+ visit_exec_list(&ir->body_instructions, this);
+ 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;
+}
+
+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::visit(ir_expression *ir)
+{
+ unsigned int operand;
+ struct ir_to_mesa_src_reg op[2];
+ struct ir_to_mesa_src_reg result_src;
+ struct ir_to_mesa_dst_reg result_dst;
+ const glsl_type *vec4_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 4, 1);
+ const glsl_type *vec3_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 3, 1);
+ const glsl_type *vec2_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 2, 1);
+
+ /* 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;
+ }
+
+ 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());
+ }
+
+ 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_exp:
+ ir_to_mesa_emit_scalar_op2(ir, OPCODE_POW, result_dst,
+ src_reg_for_float(M_E), op[0]);
+ break;
+ case ir_unop_exp2:
+ ir_to_mesa_emit_scalar_op1(ir, OPCODE_EX2, result_dst, op[0]);
+ break;
+ case ir_unop_log:
+ ir_to_mesa_emit_scalar_op1(ir, OPCODE_LOG, result_dst, op[0]);
+ 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_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_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_logic_xor:
+ case ir_binop_nequal:
+ 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:
+ if (ir->operands[0]->type == vec4_type) {
+ assert(ir->operands[1]->type == vec4_type);
+ ir_to_mesa_emit_op2(ir, OPCODE_DP4,
+ result_dst,
+ op[0], op[1]);
+ } else if (ir->operands[0]->type == vec3_type) {
+ assert(ir->operands[1]->type == vec3_type);
+ ir_to_mesa_emit_op2(ir, OPCODE_DP3,
+ result_dst,
+ op[0], op[1]);
+ } else if (ir->operands[0]->type == vec2_type) {
+ assert(ir->operands[1]->type == vec2_type);
+ ir_to_mesa_emit_op2(ir, OPCODE_DP2,
+ result_dst,
+ op[0], op[1]);
+ }
+ break;
+
+ case ir_binop_cross:
+ ir_to_mesa_emit_op2(ir, OPCODE_XPD, result_dst, op[0], op[1]);
+ break;
+
+ case ir_unop_sqrt:
+ ir_to_mesa_emit_scalar_op1(ir, OPCODE_RSQ, result_dst, op[0]);
+ ir_to_mesa_emit_scalar_op1(ir, OPCODE_RCP, result_dst, result_src);
+ /* For incoming channels < 0, set the result to 0. */
+ ir_to_mesa_emit_op3(ir, OPCODE_CMP, result_dst,
+ op[0], src_reg_for_float(0.0), result_src);
+ 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,
+ result_src, 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:
+ assert(!"GLSL 1.30 features unsupported");
+ 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;
+}
+
+static int
+add_matrix_ref(struct gl_program *prog, int *tokens)
+{
+ int base_pos = -1;
+ int i;
+
+ /* Add a ref for each column. It looks like the reason we do
+ * it this way is that _mesa_add_state_reference doesn't work
+ * for things that aren't vec4s, so the tokens[2]/tokens[3]
+ * range has to be equal.
+ */
+ for (i = 0; i < 4; i++) {
+ tokens[2] = i;
+ tokens[3] = i;
+ int pos = _mesa_add_state_reference(prog->Parameters,
+ (gl_state_index *)tokens);
+ if (base_pos == -1)
+ base_pos = pos;
+ else
+ assert(base_pos + i == pos);
+ }
+
+ return base_pos;
+}
+
+static variable_storage *
+get_builtin_matrix_ref(void *mem_ctx, struct gl_program *prog, ir_variable *var,
+ ir_rvalue *array_index)
+{
+ /*
+ * NOTE: The ARB_vertex_program extension specified that matrices get
+ * loaded in registers in row-major order. With GLSL, we want column-
+ * major order. So, we need to transpose all matrices here...
+ */
+ static const struct {
+ const char *name;
+ int matrix;
+ int modifier;
+ } matrices[] = {
+ { "gl_ModelViewMatrix", STATE_MODELVIEW_MATRIX, STATE_MATRIX_TRANSPOSE },
+ { "gl_ModelViewMatrixInverse", STATE_MODELVIEW_MATRIX, STATE_MATRIX_INVTRANS },
+ { "gl_ModelViewMatrixTranspose", STATE_MODELVIEW_MATRIX, 0 },
+ { "gl_ModelViewMatrixInverseTranspose", STATE_MODELVIEW_MATRIX, STATE_MATRIX_INVERSE },
+
+ { "gl_ProjectionMatrix", STATE_PROJECTION_MATRIX, STATE_MATRIX_TRANSPOSE },
+ { "gl_ProjectionMatrixInverse", STATE_PROJECTION_MATRIX, STATE_MATRIX_INVTRANS },
+ { "gl_ProjectionMatrixTranspose", STATE_PROJECTION_MATRIX, 0 },
+ { "gl_ProjectionMatrixInverseTranspose", STATE_PROJECTION_MATRIX, STATE_MATRIX_INVERSE },
+
+ { "gl_ModelViewProjectionMatrix", STATE_MVP_MATRIX, STATE_MATRIX_TRANSPOSE },
+ { "gl_ModelViewProjectionMatrixInverse", STATE_MVP_MATRIX, STATE_MATRIX_INVTRANS },
+ { "gl_ModelViewProjectionMatrixTranspose", STATE_MVP_MATRIX, 0 },
+ { "gl_ModelViewProjectionMatrixInverseTranspose", STATE_MVP_MATRIX, STATE_MATRIX_INVERSE },
+
+ { "gl_TextureMatrix", STATE_TEXTURE_MATRIX, STATE_MATRIX_TRANSPOSE },
+ { "gl_TextureMatrixInverse", STATE_TEXTURE_MATRIX, STATE_MATRIX_INVTRANS },
+ { "gl_TextureMatrixTranspose", STATE_TEXTURE_MATRIX, 0 },
+ { "gl_TextureMatrixInverseTranspose", STATE_TEXTURE_MATRIX, STATE_MATRIX_INVERSE },
+
+ { "gl_NormalMatrix", STATE_MODELVIEW_MATRIX, STATE_MATRIX_INVERSE },
+
+ };
+ unsigned int i;
+ variable_storage *entry;
+
+ /* C++ gets angry when we try to use an int as a gl_state_index, so we use
+ * ints for gl_state_index. Make sure they're compatible.
+ */
+ assert(sizeof(gl_state_index) == sizeof(int));
+
+ for (i = 0; i < Elements(matrices); i++) {
+ if (strcmp(var->name, matrices[i].name) == 0) {
+ int tokens[STATE_LENGTH];
+ int base_pos = -1;
+
+ tokens[0] = matrices[i].matrix;
+ tokens[4] = matrices[i].modifier;
+ if (matrices[i].matrix == STATE_TEXTURE_MATRIX) {
+ ir_constant *index = array_index->constant_expression_value();
+ if (index) {
+ tokens[1] = index->value.i[0];
+ base_pos = add_matrix_ref(prog, tokens);
+ } else {
+ for (i = 0; i < var->type->length; i++) {
+ tokens[1] = i;
+ int pos = add_matrix_ref(prog, tokens);
+ if (base_pos == -1)
+ base_pos = pos;
+ else
+ assert(base_pos + (int)i * 4 == pos);
+ }
+ }
+ } else {
+ tokens[1] = 0; /* unused array index */
+ base_pos = add_matrix_ref(prog, tokens);
+ }
+ tokens[4] = matrices[i].modifier;
+
+ entry = new(mem_ctx) variable_storage(var,
+ PROGRAM_STATE_VAR,
+ base_pos);
+
+ return entry;
+ }
+ }
+
+ return NULL;
+}
+
+void
+ir_to_mesa_visitor::visit(ir_dereference_variable *ir)
+{
+ ir_to_mesa_src_reg src_reg;
+ variable_storage *entry = find_variable_storage(ir->var);
+ unsigned int loc;
+
+ if (!entry) {
+ switch (ir->var->mode) {
+ case ir_var_uniform:
+ entry = get_builtin_matrix_ref(this->mem_ctx, this->prog, ir->var,
+ NULL);
+ if (entry)
+ break;
+
+ /* FINISHME: Fix up uniform name for arrays and things */
+ if (ir->var->type->base_type == GLSL_TYPE_SAMPLER) {
+ /* FINISHME: we whack the location of the var here, which
+ * is probably not expected. But we need to communicate
+ * mesa's sampler number to the tex instruction.
+ */
+ int sampler = _mesa_add_sampler(this->prog->Parameters,
+ ir->var->name,
+ ir->var->type->gl_type);
+ map_sampler(ir->var->location, sampler);
+
+ entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_SAMPLER,
+ sampler);
+ this->variables.push_tail(entry);
+ break;
+ }
+
+ assert(ir->var->type->gl_type != 0 &&
+ ir->var->type->gl_type != GL_INVALID_ENUM);
+ loc = _mesa_add_uniform(this->prog->Parameters,
+ ir->var->name,
+ type_size(ir->var->type) * 4,
+ ir->var->type->gl_type,
+ NULL);
+
+ /* Always mark the uniform used at this point. If it isn't
+ * used, dead code elimination should have nuked the decl already.
+ */
+ this->prog->Parameters->Parameters[loc].Used = GL_TRUE;
+
+ entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_UNIFORM, loc);
+ this->variables.push_tail(entry);
+ break;
+ case ir_var_in:
+ case ir_var_out:
+ case ir_var_inout:
+ /* 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,
+ type_size(ir->var->type) * 4,
+ ir->var->type->gl_type,
+ ir->var->location - VERT_ATTRIB_GENERIC0);
+ }
+ } 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);
+ }
+ }
+
+ src_reg.file = entry->file;
+ src_reg.index = entry->index;
+ /* 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->var->type->vector_elements);
+ else
+ src_reg.swizzle = SWIZZLE_NOOP;
+ src_reg.reladdr = NULL;
+ src_reg.negate = 0;
+
+ this->result = src_reg;
+}
+
+void
+ir_to_mesa_visitor::visit(ir_dereference_array *ir)
+{
+ ir_constant *index;
+ ir_to_mesa_src_reg src_reg;
+ ir_dereference_variable *deref_var = ir->array->as_dereference_variable();
+ int element_size = type_size(ir->type);
+
+ index = ir->array_index->constant_expression_value();
+
+ if (deref_var && strncmp(deref_var->var->name,
+ "gl_TextureMatrix",
+ strlen("gl_TextureMatrix")) == 0) {
+ ir_to_mesa_src_reg src_reg;
+ struct variable_storage *entry;
+
+ entry = get_builtin_matrix_ref(this->mem_ctx, this->prog, deref_var->var,
+ ir->array_index);
+ assert(entry);
+
+ src_reg.file = entry->file;
+ src_reg.index = entry->index;
+ src_reg.swizzle = swizzle_for_size(ir->type->vector_elements);
+ src_reg.negate = 0;
+
+ if (index) {
+ src_reg.reladdr = NULL;
+ } else {
+ ir_to_mesa_src_reg index_reg = get_temp(glsl_type::float_type);
+
+ ir->array_index->accept(this);
+ 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));
+ }
+
+ this->result = src_reg;
+ return;
+ }
+
+ 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);
+ }
+ this->result.swizzle = swizzle_for_size(ir->type->vector_elements);
+ 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.
+ *
+ * Thanks to ir_swizzle_swizzle, and ir_vec_index_to_swizzle, we
+ * should only see potentially one variable array index of a vector,
+ * and one swizzle, before getting to actual vec4 storage. So handle
+ * those, then go use ir_dereference to handle the rest.
+ */
+static struct ir_to_mesa_dst_reg
+get_assignment_lhs(ir_instruction *ir, ir_to_mesa_visitor *v,
+ ir_to_mesa_src_reg *r)
+{
+ struct ir_to_mesa_dst_reg dst_reg;
+ ir_swizzle *swiz;
+
+ ir_dereference_array *deref_array = ir->as_dereference_array();
+ /* This should have been handled by ir_vec_index_to_cond_assign */
+ 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);
+ dst_reg = ir_to_mesa_dst_reg_from_src(v->result);
+
+ if ((swiz = ir->as_swizzle())) {
+ int swizzles[4] = {
+ swiz->mask.x,
+ swiz->mask.y,
+ swiz->mask.z,
+ swiz->mask.w
+ };
+ int new_r_swizzle[4];
+ int orig_r_swizzle = r->swizzle;
+ int i;
+
+ for (i = 0; i < 4; i++) {
+ new_r_swizzle[i] = GET_SWZ(orig_r_swizzle, 0);
+ }
+
+ dst_reg.writemask = 0;
+ for (i = 0; i < 4; i++) {
+ if (i < swiz->mask.num_components) {
+ dst_reg.writemask |= 1 << swizzles[i];
+ new_r_swizzle[swizzles[i]] = GET_SWZ(orig_r_swizzle, i);
+ }
+ }
+
+ r->swizzle = MAKE_SWIZZLE4(new_r_swizzle[0],
+ new_r_swizzle[1],
+ new_r_swizzle[2],
+ new_r_swizzle[3]);
+ }
+
+ return dst_reg;
+}
+
+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;
+
+ assert(!ir->lhs->type->is_array());
+
+ ir->rhs->accept(this);
+ r = this->result;
+
+ l = get_assignment_lhs(ir->lhs, this, &r);
+
+ assert(l.file != PROGRAM_UNDEFINED);
+ assert(r.file != PROGRAM_UNDEFINED);
+
+ if (ir->condition) {
+ ir_to_mesa_src_reg condition;
+
+ ir->condition->accept(this);
+ condition = this->result;
+
+ /* 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 computing the condition.
+ */
+ condition.negate = ~condition.negate;
+ for (i = 0; i < type_size(ir->lhs->type); i++) {
+ 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];
+ GLfloat *values = stack_vals;
+ unsigned int i;
+
+ if (ir->type->is_array()) {
+ ir->print();
+ printf("\n");
+ assert(!"FINISHME: array constants");
+ }
+
+ if (ir->type->is_matrix()) {
+ /* Unfortunately, 4 floats is all we can get into
+ * _mesa_add_unnamed_constant. So, make a temp to store the
+ * matrix and move each constant value into it. If we get
+ * lucky, copy propagation will eliminate the extra moves.
+ */
+ ir_to_mesa_src_reg mat = get_temp(glsl_type::vec4_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++) {
+ src_reg.file = PROGRAM_CONSTANT;
+
+ assert(ir->type->base_type == GLSL_TYPE_FLOAT);
+ values = &ir->value.f[i * ir->type->vector_elements];
+
+ src_reg.index = _mesa_add_unnamed_constant(this->prog->Parameters,
+ values,
+ ir->type->vector_elements,
+ &src_reg.swizzle);
+ src_reg.reladdr = NULL;
+ src_reg.negate = 0;
+ ir_to_mesa_emit_op1(ir, OPCODE_MOV, mat_column, src_reg);
+
+ mat_column.index++;
+ }
+
+ this->result = mat;
+ }
+
+ 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");
+ }
+
+ src_reg.index = _mesa_add_unnamed_constant(this->prog->Parameters,
+ values, ir->type->vector_elements,
+ &src_reg.swizzle);
+ src_reg.reladdr = NULL;
+ src_reg.negate = 0;
+
+ this->result = src_reg;
+}
+
+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);
+ break;
+ }
+
+ if (sig->return_type) {
+ 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 = { 0 }, 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;
+
+ ir_dereference_variable *sampler = ir->sampler->as_dereference_variable();
+ assert(sampler); /* FINISHME: sampler arrays */
+ /* generate the mapping, remove when we generate storage at
+ * declaration time
+ */
+ sampler->accept(this);
+
+ inst->sampler = get_sampler_number(sampler->var->location);
+
+ switch (sampler->type->sampler_dimensionality) {
+ case GLSL_SAMPLER_DIM_1D:
+ inst->tex_target = TEXTURE_1D_INDEX;
+ break;
+ case GLSL_SAMPLER_DIM_2D:
+ inst->tex_target = 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;
+ default:
+ assert(!"FINISHME: other texture targets");
+ }
+
+ this->result = result_src;
+}
+
+void
+ir_to_mesa_visitor::visit(ir_return *ir)
+{
+ assert(current_function);
+
+ if (ir->get_value()) {
+ ir_to_mesa_dst_reg l;
+ int i;
+
+ 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)
+{
+ assert(ir->condition == NULL); /* FINISHME */
+
+ ir_to_mesa_emit_op0(ir, OPCODE_KIL_NV);
+}
+
+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 (ctx->Shader.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;
+ sampler_map = NULL;
+ sampler_map_size = 0;
+ current_function = NULL;
+}
+
+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) - 1);
+ 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;
+
+ 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 = (int *)calloc(if_count, sizeof(*if_stack));
+ loop_stack = (int *)calloc(loop_count, sizeof(*loop_stack));
+
+ 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;
+ }
+ }
+
+ free(if_stack);
+}
+
+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
+mark_input(struct gl_program *prog,
+ int index,
+ GLboolean reladdr)
+{
+ prog->InputsRead |= BITFIELD64_BIT(index);
+ int i;
+
+ if (reladdr) {
+ if (index >= FRAG_ATTRIB_TEX0 && index <= FRAG_ATTRIB_TEX7) {
+ for (i = 0; i < 8; i++) {
+ prog->InputsRead |= BITFIELD64_BIT(FRAG_ATTRIB_TEX0 + i);
+ }
+ } else {
+ assert(!"FINISHME: Mark InputsRead for varying arrays");
+ }
+ }
+}
+
+static void
+mark_output(struct gl_program *prog,
+ int index,
+ GLboolean reladdr)
+{
+ prog->OutputsWritten |= BITFIELD64_BIT(index);
+ int i;
+
+ if (reladdr) {
+ if (index >= VERT_RESULT_TEX0 && index <= VERT_RESULT_TEX7) {
+ for (i = 0; i < 8; i++) {
+ prog->OutputsWritten |= BITFIELD64_BIT(FRAG_ATTRIB_TEX0 + i);
+ }
+ } else {
+ assert(!"FINISHME: Mark OutputsWritten for varying arrays");
+ }
+ }
+}
+
+static void
+count_resources(struct gl_program *prog)
+{
+ unsigned int i;
+
+ prog->InputsRead = 0;
+ prog->OutputsWritten = 0;
+ prog->SamplersUsed = 0;
+
+ for (i = 0; i < prog->NumInstructions; i++) {
+ struct prog_instruction *inst = &prog->Instructions[i];
+ unsigned int reg;
+
+ switch (inst->DstReg.File) {
+ case PROGRAM_OUTPUT:
+ mark_output(prog, inst->DstReg.Index, inst->DstReg.RelAddr);
+ break;
+ case PROGRAM_INPUT:
+ mark_input(prog, inst->DstReg.Index, inst->DstReg.RelAddr);
+ break;
+ default:
+ break;
+ }
+
+ for (reg = 0; reg < _mesa_num_inst_src_regs(inst->Opcode); reg++) {
+ switch (inst->SrcReg[reg].File) {
+ case PROGRAM_OUTPUT:
+ mark_output(prog, inst->SrcReg[reg].Index,
+ inst->SrcReg[reg].RelAddr);
+ break;
+ case PROGRAM_INPUT:
+ mark_input(prog, inst->SrcReg[reg].Index, inst->SrcReg[reg].RelAddr);
+ break;
+ default:
+ break;
+ }
+ }
+
+ /* Instead of just using the uniform's value to map to a
+ * sampler, Mesa first allocates a separate number for the
+ * sampler (_mesa_add_sampler), then we reindex it down to a
+ * small integer (sampler_map[], SamplersUsed), then that gets
+ * mapped to the uniform's value, and we get an actual sampler.
+ */
+ 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);
+}
+
+/* Each stage has some uniforms in its Parameters list. The Uniforms
+ * list for the linked shader program has a pointer to these uniforms
+ * in each of the stage's Parameters list, so that their values can be
+ * updated when a uniform is set.
+ */
+static void
+link_uniforms_to_shared_uniform_list(struct gl_uniform_list *uniforms,
+ struct gl_program *prog)
+{
+ unsigned int i;
+
+ for (i = 0; i < prog->Parameters->NumParameters; i++) {
+ const struct gl_program_parameter *p = prog->Parameters->Parameters + i;
+
+ if (p->Type == PROGRAM_UNIFORM || p->Type == PROGRAM_SAMPLER) {
+ struct gl_uniform *uniform =
+ _mesa_append_uniform(uniforms, p->Name, prog->Target, i);
+ if (uniform)
+ uniform->Initialized = p->Initialized;
+ }
+ }
+}
+
+struct gl_program *
+get_mesa_program(GLcontext *ctx, struct gl_shader_program *shader_program,
+ struct gl_shader *shader)
+{
+ void *mem_ctx = shader_program;
+ 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;
+
+ 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;
+ default:
+ assert(!"should not be reached");
+ break;
+ }
+
+ validate_ir_tree(shader->ir);
+
+ prog = ctx->Driver.NewProgram(ctx, target, 1);
+ 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.mem_ctx = talloc_new(NULL);
+
+ /* 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);
+
+ entry->bgn_inst = v.ir_to_mesa_emit_op0(NULL, OPCODE_RET);
+ entry->bgn_inst = v.ir_to_mesa_emit_op0(NULL, OPCODE_ENDSUB);
+ 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(mem_ctx, ir_instruction *,
+ num_instructions);
+
+ mesa_inst = mesa_instructions;
+ i = 0;
+ foreach_iter(exec_list_iterator, iter, v.instructions) {
+ ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get();
+
+ mesa_inst->Opcode = inst->op;
+ mesa_inst->CondUpdate = inst->cond_update;
+ 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;
+
+ if (ctx->Shader.EmitNoIfs && mesa_inst->Opcode == OPCODE_IF) {
+ shader_program->InfoLog =
+ talloc_asprintf_append(shader_program->InfoLog,
+ "Couldn't flatten if statement\n");
+ shader_program->LinkStatus = false;
+ }
+
+ if (mesa_inst->Opcode == OPCODE_BGNSUB)
+ inst->function->inst = i;
+ else if (mesa_inst->Opcode == OPCODE_CAL)
+ mesa_inst->BranchTarget = inst->function->sig_id; /* rewritten later */
+ else if (mesa_inst->Opcode == OPCODE_ARL)
+ prog->NumAddressRegs = 1;
+
+ mesa_inst++;
+ i++;
+ }
+
+ set_branchtargets(&v, mesa_instructions, num_instructions);
+ if (ctx->Shader.Flags & GLSL_DUMP) {
+ printf("Mesa %s program:\n", target_string);
+ print_program(mesa_instructions, mesa_instruction_annotation,
+ num_instructions);
+ }
+
+ prog->Instructions = mesa_instructions;
+ prog->NumInstructions = num_instructions;
+
+ _mesa_reference_program(ctx, &shader->Program, prog);
+
+ if ((ctx->Shader.Flags & GLSL_NO_OPT) == 0) {
+ _mesa_optimize_program(ctx, prog);
+ }
+
+ return prog;
+}
+
+extern "C" {
+
+void
+_mesa_glsl_compile_shader(GLcontext *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;
+ state->error = preprocess(state, &source, &state->info_log,
+ &ctx->Extensions);
+
+ if (!state->error) {
+ _mesa_glsl_lexer_ctor(state, source);
+ _mesa_glsl_parse(state);
+ _mesa_glsl_lexer_dtor(state);
+ }
+
+ 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);
+
+ /* Lowering */
+ do_mat_op_to_vec(shader->ir);
+ do_mod_to_fract(shader->ir);
+ do_div_to_mul_rcp(shader->ir);
+
+ /* Optimization passes */
+ bool progress;
+ do {
+ progress = false;
+
+ progress = do_function_inlining(shader->ir) || progress;
+ progress = do_if_simplification(shader->ir) || progress;
+ progress = do_copy_propagation(shader->ir) || progress;
+ progress = do_dead_code_local(shader->ir) || progress;
+ progress = do_dead_code_unlinked(state, shader->ir) || progress;
+ progress = do_constant_variable_unlinked(shader->ir) || progress;
+ progress = do_constant_folding(shader->ir) || progress;
+ progress = do_if_return(shader->ir) || progress;
+ if (ctx->Shader.EmitNoIfs)
+ progress = do_if_to_cond_assign(shader->ir) || progress;
+
+ progress = do_vec_index_to_swizzle(shader->ir) || progress;
+ /* Do this one after the previous to let the easier pass handle
+ * constant vector indexing.
+ */
+ progress = do_vec_index_to_cond_assign(shader->ir) || progress;
+
+ progress = do_swizzle_swizzle(shader->ir) || progress;
+ } while (progress);
+
+ 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;
+
+ /* Retain any live IR, but trash the rest. */
+ reparent_ir(shader->ir, shader);
+
+ talloc_free(state);
+ }
+
+void
+_mesa_glsl_link_shader(GLcontext *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) {
+ prog->InfoLog =
+ talloc_asprintf_append(prog->InfoLog,
+ "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);
+
+ if (prog->LinkStatus) {
+ link_shaders(prog);
+
+ /* We don't use the linker's uniforms list, and cook up our own at
+ * generate time.
+ */
+ free(prog->Uniforms);
+ prog->Uniforms = _mesa_new_uniform_list();
+ }
+
+ if (prog->LinkStatus) {
+ for (i = 0; i < prog->_NumLinkedShaders; i++) {
+ struct gl_program *linked_prog;
+
+ linked_prog = get_mesa_program(ctx, prog,
+ prog->_LinkedShaders[i]);
+ count_resources(linked_prog);
+
+ link_uniforms_to_shared_uniform_list(prog->Uniforms, linked_prog);
+
+ switch (prog->_LinkedShaders[i]->Type) {
+ case GL_VERTEX_SHADER:
+ _mesa_reference_vertprog(ctx, &prog->VertexProgram,
+ (struct gl_vertex_program *)linked_prog);
+ 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);
+ ctx->Driver.ProgramStringNotify(ctx, GL_FRAGMENT_PROGRAM_ARB,
+ linked_prog);
+ break;
+ }
+ }
+ }
+}
+
+} /* extern "C" */