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 ir_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 "hash_table.h"
38 class ir_function_inlining_visitor : public ir_hierarchical_visitor {
40 ir_function_inlining_visitor()
45 virtual ~ir_function_inlining_visitor()
50 virtual ir_visitor_status visit_enter(ir_expression *);
51 virtual ir_visitor_status visit_enter(ir_call *);
52 virtual ir_visitor_status visit_enter(ir_assignment *);
53 virtual ir_visitor_status visit_enter(ir_return *);
54 virtual ir_visitor_status visit_enter(ir_texture *);
55 virtual ir_visitor_status visit_enter(ir_swizzle *);
61 unsigned int hash_func(const void *key)
63 return (unsigned int)(uintptr_t)key;
66 int hash_compare_func(const void *key1, const void *key2)
68 return key1 == key2 ? 0 : 1;
72 automatic_inlining_predicate(ir_instruction *ir)
74 ir_call *call = ir->as_call();
76 if (call && can_inline(call))
83 do_function_inlining(exec_list *instructions)
85 ir_function_inlining_visitor v;
87 do_expression_flattening(instructions, automatic_inlining_predicate);
95 replace_return_with_assignment(ir_instruction *ir, void *data)
97 void *ctx = talloc_parent(ir);
98 ir_variable *retval = (ir_variable *)data;
99 ir_return *ret = ir->as_return();
103 ir_rvalue *lhs = new(ctx) ir_dereference_variable(retval);
104 ret->insert_before(new(ctx) ir_assignment(lhs, ret->value, NULL));
107 /* un-valued return has to be the last return, or we shouldn't
108 * have reached here. (see can_inline()).
110 assert(!ret->next->is_tail_sentinal());
116 ir_call::generate_inline(ir_instruction *next_ir)
118 void *ctx = talloc_parent(this);
119 ir_variable **parameters;
122 ir_variable *retval = NULL;
123 struct hash_table *ht;
125 ht = hash_table_ctor(0, hash_func, hash_compare_func);
128 foreach_iter(exec_list_iterator, iter_sig, this->callee->parameters)
131 parameters = new ir_variable *[num_parameters];
133 /* Generate storage for the return value. */
134 if (this->callee->return_type) {
135 retval = new(ctx) ir_variable(this->callee->return_type, "__retval");
136 next_ir->insert_before(retval);
139 /* Generate the declarations for the parameters to our inlined code,
140 * and set up the mapping of real function body variables to ours.
143 exec_list_iterator sig_param_iter = this->callee->parameters.iterator();
144 exec_list_iterator param_iter = this->actual_parameters.iterator();
145 for (i = 0; i < num_parameters; i++) {
146 const ir_variable *const sig_param = (ir_variable *) sig_param_iter.get();
147 ir_rvalue *param = (ir_rvalue *) param_iter.get();
149 /* Generate a new variable for the parameter. */
150 parameters[i] = (ir_variable *)sig_param->clone(ht);
151 parameters[i]->mode = ir_var_auto;
152 next_ir->insert_before(parameters[i]);
154 /* Move the actual param into our param variable if it's an 'in' type. */
155 if (sig_param->mode == ir_var_in ||
156 sig_param->mode == ir_var_inout) {
157 ir_assignment *assign;
159 assign = new(ctx) ir_assignment(new(ctx) ir_dereference_variable(parameters[i]),
161 next_ir->insert_before(assign);
164 sig_param_iter.next();
168 /* Generate the inlined body of the function. */
169 foreach_iter(exec_list_iterator, iter, callee->body) {
170 ir_instruction *ir = (ir_instruction *)iter.get();
171 ir_instruction *new_ir = ir->clone(ht);
173 next_ir->insert_before(new_ir);
174 visit_tree(new_ir, replace_return_with_assignment, retval);
177 /* Copy back the value of any 'out' parameters from the function body
178 * variables to our own.
181 param_iter = this->actual_parameters.iterator();
182 for (i = 0; i < num_parameters; i++) {
183 ir_instruction *const param = (ir_instruction *) param_iter.get();
185 /* Move our param variable into the actual param if it's an 'out' type. */
186 if (parameters[i]->mode == ir_var_out ||
187 parameters[i]->mode == ir_var_inout) {
188 ir_assignment *assign;
190 assign = new(ctx) ir_assignment(param->as_rvalue(),
191 new(ctx) ir_dereference_variable(parameters[i]),
193 next_ir->insert_before(assign);
199 delete [] parameters;
204 return new(ctx) ir_dereference_variable(retval);
211 ir_function_inlining_visitor::visit_enter(ir_expression *ir)
214 return visit_continue_with_parent;
219 ir_function_inlining_visitor::visit_enter(ir_return *ir)
222 return visit_continue_with_parent;
227 ir_function_inlining_visitor::visit_enter(ir_texture *ir)
230 return visit_continue_with_parent;
235 ir_function_inlining_visitor::visit_enter(ir_swizzle *ir)
238 return visit_continue_with_parent;
243 ir_function_inlining_visitor::visit_enter(ir_call *ir)
245 if (can_inline(ir)) {
246 (void) ir->generate_inline(ir);
248 this->progress = true;
251 return visit_continue;
256 ir_function_inlining_visitor::visit_enter(ir_assignment *ir)
258 ir_call *call = ir->rhs->as_call();
259 if (!call || !can_inline(call))
260 return visit_continue;
262 /* generates the parameter setup, function body, and returns the return
263 * value of the function
265 ir_rvalue *rhs = call->generate_inline(ir);
269 this->progress = true;
271 return visit_continue;