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_function_inlining.cpp
27 * Replaces calls to functions with the body of the function.
32 #include "ir_visitor.h"
33 #include "ir_function_inlining.h"
34 #include "ir_expression_flattening.h"
35 #include "glsl_types.h"
36 #include "program/hash_table.h"
39 do_sampler_replacement(exec_list *instructions,
41 ir_dereference *deref);
43 class ir_function_inlining_visitor : public ir_hierarchical_visitor {
45 ir_function_inlining_visitor()
50 virtual ~ir_function_inlining_visitor()
55 virtual ir_visitor_status visit_enter(ir_expression *);
56 virtual ir_visitor_status visit_enter(ir_call *);
57 virtual ir_visitor_status visit_enter(ir_assignment *);
58 virtual ir_visitor_status visit_enter(ir_return *);
59 virtual ir_visitor_status visit_enter(ir_texture *);
60 virtual ir_visitor_status visit_enter(ir_swizzle *);
67 automatic_inlining_predicate(ir_instruction *ir)
69 ir_call *call = ir->as_call();
71 if (call && can_inline(call))
78 do_function_inlining(exec_list *instructions)
80 ir_function_inlining_visitor v;
82 do_expression_flattening(instructions, automatic_inlining_predicate);
90 replace_return_with_assignment(ir_instruction *ir, void *data)
92 void *ctx = ralloc_parent(ir);
93 ir_variable *retval = (ir_variable *)data;
94 ir_return *ret = ir->as_return();
98 ir_rvalue *lhs = new(ctx) ir_dereference_variable(retval);
99 ret->replace_with(new(ctx) ir_assignment(lhs, ret->value, NULL));
101 /* un-valued return has to be the last return, or we shouldn't
102 * have reached here. (see can_inline()).
104 assert(ret->next->is_tail_sentinel());
111 ir_call::generate_inline(ir_instruction *next_ir)
113 void *ctx = ralloc_parent(this);
114 ir_variable **parameters;
117 ir_variable *retval = NULL;
118 struct hash_table *ht;
120 ht = hash_table_ctor(0, hash_table_pointer_hash, hash_table_pointer_compare);
123 foreach_iter(exec_list_iterator, iter_sig, this->callee->parameters)
126 parameters = new ir_variable *[num_parameters];
128 /* Generate storage for the return value. */
129 if (!this->callee->return_type->is_void()) {
130 retval = new(ctx) ir_variable(this->callee->return_type, "_ret_val",
132 next_ir->insert_before(retval);
135 /* Generate the declarations for the parameters to our inlined code,
136 * and set up the mapping of real function body variables to ours.
139 exec_list_iterator sig_param_iter = this->callee->parameters.iterator();
140 exec_list_iterator param_iter = this->actual_parameters.iterator();
141 for (i = 0; i < num_parameters; i++) {
142 ir_variable *sig_param = (ir_variable *) sig_param_iter.get();
143 ir_rvalue *param = (ir_rvalue *) param_iter.get();
145 /* Generate a new variable for the parameter. */
146 if (sig_param->type->base_type == GLSL_TYPE_SAMPLER) {
147 /* For samplers, we want the inlined sampler references
148 * referencing the passed in sampler variable, since that
149 * will have the location information, which an assignment of
150 * a sampler wouldn't. Fix it up below.
152 parameters[i] = NULL;
154 parameters[i] = sig_param->clone(ctx, ht);
155 parameters[i]->mode = ir_var_auto;
157 /* Remove the read-only decoration becuase we're going to write
158 * directly to this variable. If the cloned variable is left
159 * read-only and the inlined function is inside a loop, the loop
160 * analysis code will get confused.
162 parameters[i]->read_only = false;
163 next_ir->insert_before(parameters[i]);
166 /* Move the actual param into our param variable if it's an 'in' type. */
167 if (parameters[i] && (sig_param->mode == ir_var_in ||
168 sig_param->mode == ir_var_const_in ||
169 sig_param->mode == ir_var_inout)) {
170 ir_assignment *assign;
172 assign = new(ctx) ir_assignment(new(ctx) ir_dereference_variable(parameters[i]),
174 next_ir->insert_before(assign);
177 sig_param_iter.next();
181 exec_list new_instructions;
183 /* Generate the inlined body of the function to a new list */
184 foreach_iter(exec_list_iterator, iter, callee->body) {
185 ir_instruction *ir = (ir_instruction *)iter.get();
186 ir_instruction *new_ir = ir->clone(ctx, ht);
188 new_instructions.push_tail(new_ir);
189 visit_tree(new_ir, replace_return_with_assignment, retval);
192 /* If any samplers were passed in, replace any deref of the sampler
193 * with a deref of the sampler argument.
195 param_iter = this->actual_parameters.iterator();
196 sig_param_iter = this->callee->parameters.iterator();
197 for (i = 0; i < num_parameters; i++) {
198 ir_instruction *const param = (ir_instruction *) param_iter.get();
199 ir_variable *sig_param = (ir_variable *) sig_param_iter.get();
201 if (sig_param->type->base_type == GLSL_TYPE_SAMPLER) {
202 ir_dereference *deref = param->as_dereference();
205 do_sampler_replacement(&new_instructions, sig_param, deref);
208 sig_param_iter.next();
211 /* Now push those new instructions in. */
212 next_ir->insert_before(&new_instructions);
214 /* Copy back the value of any 'out' parameters from the function body
215 * variables to our own.
218 param_iter = this->actual_parameters.iterator();
219 sig_param_iter = this->callee->parameters.iterator();
220 for (i = 0; i < num_parameters; i++) {
221 ir_instruction *const param = (ir_instruction *) param_iter.get();
222 const ir_variable *const sig_param = (ir_variable *) sig_param_iter.get();
224 /* Move our param variable into the actual param if it's an 'out' type. */
225 if (parameters[i] && (sig_param->mode == ir_var_out ||
226 sig_param->mode == ir_var_inout)) {
227 ir_assignment *assign;
229 assign = new(ctx) ir_assignment(param->clone(ctx, NULL)->as_rvalue(),
230 new(ctx) ir_dereference_variable(parameters[i]),
232 next_ir->insert_before(assign);
236 sig_param_iter.next();
239 delete [] parameters;
244 return new(ctx) ir_dereference_variable(retval);
251 ir_function_inlining_visitor::visit_enter(ir_expression *ir)
254 return visit_continue_with_parent;
259 ir_function_inlining_visitor::visit_enter(ir_return *ir)
262 return visit_continue_with_parent;
267 ir_function_inlining_visitor::visit_enter(ir_texture *ir)
270 return visit_continue_with_parent;
275 ir_function_inlining_visitor::visit_enter(ir_swizzle *ir)
278 return visit_continue_with_parent;
283 ir_function_inlining_visitor::visit_enter(ir_call *ir)
285 if (can_inline(ir)) {
286 /* If the call was part of some tree, then it should have been
287 * flattened out or we shouldn't have seen it because of a
288 * visit_continue_with_parent in this visitor.
290 assert(ir == base_ir);
292 (void) ir->generate_inline(ir);
294 this->progress = true;
297 return visit_continue;
302 ir_function_inlining_visitor::visit_enter(ir_assignment *ir)
304 ir_call *call = ir->rhs->as_call();
305 if (!call || !can_inline(call))
306 return visit_continue;
308 /* generates the parameter setup, function body, and returns the return
309 * value of the function
311 ir_rvalue *rhs = call->generate_inline(ir);
315 this->progress = true;
317 return visit_continue;
321 * Replaces references to the "sampler" variable with a clone of "deref."
323 * From the spec, samplers can appear in the tree as function
324 * (non-out) parameters and as the result of array indexing and
325 * structure field selection. In our builtin implementation, they
326 * also appear in the sampler field of an ir_tex instruction.
329 class ir_sampler_replacement_visitor : public ir_hierarchical_visitor {
331 ir_sampler_replacement_visitor(ir_variable *sampler, ir_dereference *deref)
333 this->sampler = sampler;
337 virtual ~ir_sampler_replacement_visitor()
341 virtual ir_visitor_status visit_leave(ir_call *);
342 virtual ir_visitor_status visit_leave(ir_dereference_array *);
343 virtual ir_visitor_status visit_leave(ir_dereference_record *);
344 virtual ir_visitor_status visit_leave(ir_texture *);
346 void replace_deref(ir_dereference **deref);
347 void replace_rvalue(ir_rvalue **rvalue);
349 ir_variable *sampler;
350 ir_dereference *deref;
354 ir_sampler_replacement_visitor::replace_deref(ir_dereference **deref)
356 ir_dereference_variable *deref_var = (*deref)->as_dereference_variable();
357 if (deref_var && deref_var->var == this->sampler) {
358 *deref = this->deref->clone(ralloc_parent(*deref), NULL);
363 ir_sampler_replacement_visitor::replace_rvalue(ir_rvalue **rvalue)
368 ir_dereference *deref = (*rvalue)->as_dereference();
373 replace_deref(&deref);
378 ir_sampler_replacement_visitor::visit_leave(ir_texture *ir)
380 replace_deref(&ir->sampler);
382 return visit_continue;
386 ir_sampler_replacement_visitor::visit_leave(ir_dereference_array *ir)
388 replace_rvalue(&ir->array);
389 return visit_continue;
393 ir_sampler_replacement_visitor::visit_leave(ir_dereference_record *ir)
395 replace_rvalue(&ir->record);
396 return visit_continue;
400 ir_sampler_replacement_visitor::visit_leave(ir_call *ir)
402 foreach_iter(exec_list_iterator, iter, *ir) {
403 ir_rvalue *param = (ir_rvalue *)iter.get();
404 ir_rvalue *new_param = param;
405 replace_rvalue(&new_param);
407 if (new_param != param) {
408 param->replace_with(new_param);
411 return visit_continue;
415 do_sampler_replacement(exec_list *instructions,
416 ir_variable *sampler,
417 ir_dereference *deref)
419 ir_sampler_replacement_visitor v(sampler, deref);
421 visit_list_elements(&v, instructions);