2 * Copyright © 2010 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
25 * \file opt_structure_splitting.cpp
27 * If a structure is only ever referenced by its components, then
28 * split those components out to individual variables so they can be
29 * handled normally by other optimization passes.
31 * This skips structures like uniforms, which need to be accessible as
32 * structures for their access by the GL.
36 #include "ir_visitor.h"
37 #include "ir_print_visitor.h"
38 #include "ir_rvalue_visitor.h"
39 #include "glsl_types.h"
43 static bool debug = false;
45 // XXX using variable_entry2 here to avoid collision (MSVC multiply-defined
46 // function) with the variable_entry class seen in ir_variable_refcount.h
47 // Perhaps we can use the one in ir_variable_refcount.h and make this class
49 class variable_entry2 : public exec_node
52 variable_entry2(ir_variable *var)
55 this->whole_structure_access = 0;
56 this->declaration = false;
57 this->components = NULL;
61 ir_variable *var; /* The key: the variable's pointer. */
63 /** Number of times the variable is referenced, including assignments. */
64 unsigned whole_structure_access;
66 /* If the variable had a decl we can work with in the instruction
67 * stream. We can't do splitting on function arguments, which
68 * don't get this variable set.
72 ir_variable **components;
74 /** ralloc_parent(this->var) -- the shader's ralloc context. */
79 class ir_structure_reference_visitor : public ir_hierarchical_visitor {
81 ir_structure_reference_visitor(void)
83 this->mem_ctx = ralloc_context(NULL);
84 this->variable_list.make_empty();
87 ~ir_structure_reference_visitor(void)
92 virtual ir_visitor_status visit(ir_variable *);
93 virtual ir_visitor_status visit(ir_dereference_variable *);
94 virtual ir_visitor_status visit_enter(ir_dereference_record *);
95 virtual ir_visitor_status visit_enter(ir_assignment *);
96 virtual ir_visitor_status visit_enter(ir_function_signature *);
98 variable_entry2 *get_variable_entry2(ir_variable *var);
100 /* List of variable_entry */
101 exec_list variable_list;
107 ir_structure_reference_visitor::get_variable_entry2(ir_variable *var)
111 if (!var->type->is_record() || var->mode == ir_var_uniform)
114 foreach_iter(exec_list_iterator, iter, this->variable_list) {
115 variable_entry2 *entry = (variable_entry2 *)iter.get();
116 if (entry->var == var)
120 variable_entry2 *entry = new(mem_ctx) variable_entry2(var);
121 this->variable_list.push_tail(entry);
127 ir_structure_reference_visitor::visit(ir_variable *ir)
129 variable_entry2 *entry = this->get_variable_entry2(ir);
132 entry->declaration = true;
134 return visit_continue;
138 ir_structure_reference_visitor::visit(ir_dereference_variable *ir)
140 ir_variable *const var = ir->variable_referenced();
141 variable_entry2 *entry = this->get_variable_entry2(var);
144 entry->whole_structure_access++;
146 return visit_continue;
150 ir_structure_reference_visitor::visit_enter(ir_dereference_record *ir)
153 /* Don't descend into the ir_dereference_variable below. */
154 return visit_continue_with_parent;
158 ir_structure_reference_visitor::visit_enter(ir_assignment *ir)
160 /* If there are no structure references yet, no need to bother with
161 * processing the expression tree.
163 if (this->variable_list.is_empty())
164 return visit_continue_with_parent;
166 if (ir->lhs->as_dereference_variable() &&
167 ir->rhs->as_dereference_variable() &&
169 /* We'll split copies of a structure to copies of components, so don't
170 * descend to the ir_dereference_variables.
172 return visit_continue_with_parent;
174 return visit_continue;
178 ir_structure_reference_visitor::visit_enter(ir_function_signature *ir)
180 /* We don't have logic for structure-splitting function arguments,
181 * so just look at the body instructions and not the parameter
184 visit_list_elements(this, &ir->body);
185 return visit_continue_with_parent;
188 class ir_structure_splitting_visitor : public ir_rvalue_visitor {
190 ir_structure_splitting_visitor(exec_list *vars)
192 this->variable_list = vars;
195 virtual ~ir_structure_splitting_visitor()
199 virtual ir_visitor_status visit_leave(ir_assignment *);
201 void split_deref(ir_dereference **deref);
202 void handle_rvalue(ir_rvalue **rvalue);
203 variable_entry2 *get_splitting_entry(ir_variable *var);
205 exec_list *variable_list;
209 ir_structure_splitting_visitor::get_splitting_entry(ir_variable *var)
213 if (!var->type->is_record())
216 foreach_iter(exec_list_iterator, iter, *this->variable_list) {
217 variable_entry2 *entry = (variable_entry2 *)iter.get();
218 if (entry->var == var) {
227 ir_structure_splitting_visitor::split_deref(ir_dereference **deref)
229 if ((*deref)->ir_type != ir_type_dereference_record)
232 ir_dereference_record *deref_record = (ir_dereference_record *)*deref;
233 ir_dereference_variable *deref_var = deref_record->record->as_dereference_variable();
237 variable_entry2 *entry = get_splitting_entry(deref_var->var);
242 for (i = 0; i < entry->var->type->length; i++) {
243 if (strcmp(deref_record->field,
244 entry->var->type->fields.structure[i].name) == 0)
247 assert(i != entry->var->type->length);
249 *deref = new(entry->mem_ctx) ir_dereference_variable(entry->components[i]);
253 ir_structure_splitting_visitor::handle_rvalue(ir_rvalue **rvalue)
258 ir_dereference *deref = (*rvalue)->as_dereference();
268 ir_structure_splitting_visitor::visit_leave(ir_assignment *ir)
270 ir_dereference_variable *lhs_deref = ir->lhs->as_dereference_variable();
271 ir_dereference_variable *rhs_deref = ir->rhs->as_dereference_variable();
272 variable_entry2 *lhs_entry = lhs_deref ? get_splitting_entry(lhs_deref->var) : NULL;
273 variable_entry2 *rhs_entry = rhs_deref ? get_splitting_entry(rhs_deref->var) : NULL;
274 const glsl_type *type = ir->rhs->type;
276 if ((lhs_entry || rhs_entry) && !ir->condition) {
277 for (unsigned int i = 0; i < type->length; i++) {
278 ir_dereference *new_lhs, *new_rhs;
279 void *mem_ctx = lhs_entry ? lhs_entry->mem_ctx : rhs_entry->mem_ctx;
282 new_lhs = new(mem_ctx) ir_dereference_variable(lhs_entry->components[i]);
284 new_lhs = new(mem_ctx)
285 ir_dereference_record(ir->lhs->clone(mem_ctx, NULL),
286 type->fields.structure[i].name);
290 new_rhs = new(mem_ctx) ir_dereference_variable(rhs_entry->components[i]);
292 new_rhs = new(mem_ctx)
293 ir_dereference_record(ir->rhs->clone(mem_ctx, NULL),
294 type->fields.structure[i].name);
297 ir->insert_before(new(mem_ctx) ir_assignment(new_lhs,
303 handle_rvalue(&ir->rhs);
304 split_deref(&ir->lhs);
307 handle_rvalue(&ir->condition);
309 return visit_continue;
312 } /* unnamed namespace */
315 do_structure_splitting(exec_list *instructions)
317 ir_structure_reference_visitor refs;
319 visit_list_elements(&refs, instructions);
321 /* Trim out variables we can't split. */
322 foreach_iter(exec_list_iterator, iter, refs.variable_list) {
323 variable_entry2 *entry = (variable_entry2 *)iter.get();
326 printf("structure %s@%p: decl %d, whole_access %d\n",
327 entry->var->name, (void *) entry->var, entry->declaration,
328 entry->whole_structure_access);
331 if (!entry->declaration || entry->whole_structure_access) {
336 if (refs.variable_list.is_empty())
339 void *mem_ctx = ralloc_context(NULL);
341 /* Replace the decls of the structures to be split with their split
344 foreach_iter(exec_list_iterator, iter, refs.variable_list) {
345 variable_entry2 *entry = (variable_entry2 *)iter.get();
346 const struct glsl_type *type = entry->var->type;
348 entry->mem_ctx = ralloc_parent(entry->var);
350 entry->components = ralloc_array(mem_ctx,
354 for (unsigned int i = 0; i < entry->var->type->length; i++) {
355 const char *name = ralloc_asprintf(mem_ctx, "%s_%s",
357 type->fields.structure[i].name);
359 entry->components[i] =
360 new(entry->mem_ctx) ir_variable(type->fields.structure[i].type,
363 entry->var->insert_before(entry->components[i]);
366 entry->var->remove();
369 ir_structure_splitting_visitor split(&refs.variable_list);
370 visit_list_elements(&split, instructions);
372 ralloc_free(mem_ctx);