glsl/lower_if: move and rename found_control_flow
[mesa.git] / src / compiler / glsl / lower_if_to_cond_assign.cpp
1 /*
2 * Copyright © 2010 Intel Corporation
3 *
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:
10 *
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
13 * Software.
14 *
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.
22 */
23
24 /**
25 * \file lower_if_to_cond_assign.cpp
26 *
27 * This attempts to flatten if-statements to conditional assignments for
28 * GPUs with limited or no flow control support.
29 *
30 * It can't handle other control flow being inside of its block, such
31 * as calls or loops. Hopefully loop unrolling and inlining will take
32 * care of those.
33 *
34 * Drivers for GPUs with no control flow support should simply call
35 *
36 * lower_if_to_cond_assign(instructions)
37 *
38 * to attempt to flatten all if-statements.
39 *
40 * Some GPUs (such as i965 prior to gen6) do support control flow, but have a
41 * maximum nesting depth N. Drivers for such hardware can call
42 *
43 * lower_if_to_cond_assign(instructions, N)
44 *
45 * to attempt to flatten any if-statements appearing at depth > N.
46 */
47
48 #include "compiler/glsl_types.h"
49 #include "ir.h"
50 #include "util/set.h"
51 #include "util/hash_table.h" /* Needed for the hashing functions */
52
53 namespace {
54
55 class ir_if_to_cond_assign_visitor : public ir_hierarchical_visitor {
56 public:
57 ir_if_to_cond_assign_visitor(unsigned max_depth)
58 {
59 this->progress = false;
60 this->max_depth = max_depth;
61 this->depth = 0;
62
63 this->condition_variables =
64 _mesa_set_create(NULL, _mesa_hash_pointer,
65 _mesa_key_pointer_equal);
66 }
67
68 ~ir_if_to_cond_assign_visitor()
69 {
70 _mesa_set_destroy(this->condition_variables, NULL);
71 }
72
73 ir_visitor_status visit_enter(ir_if *);
74 ir_visitor_status visit_leave(ir_if *);
75
76 bool found_unsupported_op;
77 bool progress;
78 unsigned max_depth;
79 unsigned depth;
80
81 struct set *condition_variables;
82 };
83
84 } /* anonymous namespace */
85
86 bool
87 lower_if_to_cond_assign(exec_list *instructions, unsigned max_depth)
88 {
89 if (max_depth == UINT_MAX)
90 return false;
91
92 ir_if_to_cond_assign_visitor v(max_depth);
93
94 visit_list_elements(&v, instructions);
95
96 return v.progress;
97 }
98
99 void
100 check_control_flow(ir_instruction *ir, void *data)
101 {
102 ir_if_to_cond_assign_visitor *v = (ir_if_to_cond_assign_visitor *)data;
103
104 switch (ir->ir_type) {
105 case ir_type_call:
106 case ir_type_discard:
107 case ir_type_loop:
108 case ir_type_loop_jump:
109 case ir_type_return:
110 v->found_unsupported_op = true;
111 break;
112 default:
113 break;
114 }
115 }
116
117 void
118 move_block_to_cond_assign(void *mem_ctx,
119 ir_if *if_ir, ir_rvalue *cond_expr,
120 exec_list *instructions,
121 struct set *set)
122 {
123 foreach_in_list_safe(ir_instruction, ir, instructions) {
124 if (ir->ir_type == ir_type_assignment) {
125 ir_assignment *assign = (ir_assignment *)ir;
126
127 if (_mesa_set_search(set, assign) == NULL) {
128 _mesa_set_add(set, assign);
129
130 /* If the LHS of the assignment is a condition variable that was
131 * previously added, insert an additional assignment of false to
132 * the variable.
133 */
134 const bool assign_to_cv =
135 _mesa_set_search(
136 set, assign->lhs->variable_referenced()) != NULL;
137
138 if (!assign->condition) {
139 if (assign_to_cv) {
140 assign->rhs =
141 new(mem_ctx) ir_expression(ir_binop_logic_and,
142 glsl_type::bool_type,
143 cond_expr->clone(mem_ctx, NULL),
144 assign->rhs);
145 } else {
146 assign->condition = cond_expr->clone(mem_ctx, NULL);
147 }
148 } else {
149 assign->condition =
150 new(mem_ctx) ir_expression(ir_binop_logic_and,
151 glsl_type::bool_type,
152 cond_expr->clone(mem_ctx, NULL),
153 assign->condition);
154 }
155 }
156 }
157
158 /* Now, move from the if block to the block surrounding it. */
159 ir->remove();
160 if_ir->insert_before(ir);
161 }
162 }
163
164 ir_visitor_status
165 ir_if_to_cond_assign_visitor::visit_enter(ir_if *ir)
166 {
167 (void) ir;
168 this->depth++;
169
170 return visit_continue;
171 }
172
173 ir_visitor_status
174 ir_if_to_cond_assign_visitor::visit_leave(ir_if *ir)
175 {
176 /* Only flatten when beyond the GPU's maximum supported nesting depth. */
177 if (this->depth-- <= this->max_depth)
178 return visit_continue;
179
180 this->found_unsupported_op = false;
181
182 ir_assignment *assign;
183
184 /* Check that both blocks don't contain anything we can't support. */
185 foreach_in_list(ir_instruction, then_ir, &ir->then_instructions) {
186 visit_tree(then_ir, check_control_flow, this);
187 }
188 foreach_in_list(ir_instruction, else_ir, &ir->else_instructions) {
189 visit_tree(else_ir, check_control_flow, this);
190 }
191 if (this->found_unsupported_op)
192 return visit_continue; /* can't handle inner unsupported opcodes */
193
194 void *mem_ctx = ralloc_parent(ir);
195
196 /* Store the condition to a variable. Move all of the instructions from
197 * the then-clause of the if-statement. Use the condition variable as a
198 * condition for all assignments.
199 */
200 ir_variable *const then_var =
201 new(mem_ctx) ir_variable(glsl_type::bool_type,
202 "if_to_cond_assign_then",
203 ir_var_temporary);
204 ir->insert_before(then_var);
205
206 ir_dereference_variable *then_cond =
207 new(mem_ctx) ir_dereference_variable(then_var);
208
209 assign = new(mem_ctx) ir_assignment(then_cond, ir->condition);
210 ir->insert_before(assign);
211
212 move_block_to_cond_assign(mem_ctx, ir, then_cond,
213 &ir->then_instructions,
214 this->condition_variables);
215
216 /* Add the new condition variable to the hash table. This allows us to
217 * find this variable when lowering other (enclosing) if-statements.
218 */
219 _mesa_set_add(this->condition_variables, then_var);
220
221 /* If there are instructions in the else-clause, store the inverse of the
222 * condition to a variable. Move all of the instructions from the
223 * else-clause if the if-statement. Use the (inverse) condition variable
224 * as a condition for all assignments.
225 */
226 if (!ir->else_instructions.is_empty()) {
227 ir_variable *const else_var =
228 new(mem_ctx) ir_variable(glsl_type::bool_type,
229 "if_to_cond_assign_else",
230 ir_var_temporary);
231 ir->insert_before(else_var);
232
233 ir_dereference_variable *else_cond =
234 new(mem_ctx) ir_dereference_variable(else_var);
235
236 ir_rvalue *inverse =
237 new(mem_ctx) ir_expression(ir_unop_logic_not,
238 then_cond->clone(mem_ctx, NULL));
239
240 assign = new(mem_ctx) ir_assignment(else_cond, inverse);
241 ir->insert_before(assign);
242
243 move_block_to_cond_assign(mem_ctx, ir, else_cond,
244 &ir->else_instructions,
245 this->condition_variables);
246
247 /* Add the new condition variable to the hash table. This allows us to
248 * find this variable when lowering other (enclosing) if-statements.
249 */
250 _mesa_set_add(this->condition_variables, else_var);
251 }
252
253 ir->remove();
254
255 this->progress = true;
256
257 return visit_continue;
258 }