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 "glsl_types.h"
49 #include "ir.h"
50
51 class ir_if_to_cond_assign_visitor : public ir_hierarchical_visitor {
52 public:
ir_if_to_cond_assign_visitor(unsigned max_depth)53 ir_if_to_cond_assign_visitor(unsigned max_depth)
54 {
55 this->progress = false;
56 this->max_depth = max_depth;
57 this->depth = 0;
58 }
59
60 ir_visitor_status visit_enter(ir_if *);
61 ir_visitor_status visit_leave(ir_if *);
62
63 bool progress;
64 unsigned max_depth;
65 unsigned depth;
66 };
67
68 bool
lower_if_to_cond_assign(exec_list * instructions,unsigned max_depth)69 lower_if_to_cond_assign(exec_list *instructions, unsigned max_depth)
70 {
71 ir_if_to_cond_assign_visitor v(max_depth);
72
73 visit_list_elements(&v, instructions);
74
75 return v.progress;
76 }
77
78 void
check_control_flow(ir_instruction * ir,void * data)79 check_control_flow(ir_instruction *ir, void *data)
80 {
81 bool *found_control_flow = (bool *)data;
82 switch (ir->ir_type) {
83 case ir_type_call:
84 case ir_type_discard:
85 case ir_type_loop:
86 case ir_type_loop_jump:
87 case ir_type_return:
88 *found_control_flow = true;
89 break;
90 default:
91 break;
92 }
93 }
94
95 void
move_block_to_cond_assign(void * mem_ctx,ir_if * if_ir,ir_variable * cond_var,bool then)96 move_block_to_cond_assign(void *mem_ctx,
97 ir_if *if_ir, ir_variable *cond_var, bool then)
98 {
99 exec_list *instructions;
100
101 if (then) {
102 instructions = &if_ir->then_instructions;
103 } else {
104 instructions = &if_ir->else_instructions;
105 }
106
107 foreach_iter(exec_list_iterator, iter, *instructions) {
108 ir_instruction *ir = (ir_instruction *)iter.get();
109
110 if (ir->ir_type == ir_type_assignment) {
111 ir_assignment *assign = (ir_assignment *)ir;
112 ir_rvalue *cond_expr;
113 ir_dereference *deref = new(mem_ctx) ir_dereference_variable(cond_var);
114
115 if (then) {
116 cond_expr = deref;
117 } else {
118 cond_expr = new(mem_ctx) ir_expression(ir_unop_logic_not,
119 glsl_type::bool_type,
120 deref,
121 NULL);
122 }
123
124 if (!assign->condition) {
125 assign->condition = cond_expr;
126 } else {
127 assign->condition = new(mem_ctx) ir_expression(ir_binop_logic_and,
128 glsl_type::bool_type,
129 cond_expr,
130 assign->condition);
131 }
132 }
133
134 /* Now, move from the if block to the block surrounding it. */
135 ir->remove();
136 if_ir->insert_before(ir);
137 }
138 }
139
140 ir_visitor_status
visit_enter(ir_if * ir)141 ir_if_to_cond_assign_visitor::visit_enter(ir_if *ir)
142 {
143 (void) ir;
144 this->depth++;
145 return visit_continue;
146 }
147
148 ir_visitor_status
visit_leave(ir_if * ir)149 ir_if_to_cond_assign_visitor::visit_leave(ir_if *ir)
150 {
151 /* Only flatten when beyond the GPU's maximum supported nesting depth. */
152 if (this->depth <= this->max_depth)
153 return visit_continue;
154
155 this->depth--;
156
157 bool found_control_flow = false;
158 ir_variable *cond_var;
159 ir_assignment *assign;
160 ir_dereference_variable *deref;
161
162 /* Check that both blocks don't contain anything we can't support. */
163 foreach_iter(exec_list_iterator, then_iter, ir->then_instructions) {
164 ir_instruction *then_ir = (ir_instruction *)then_iter.get();
165 visit_tree(then_ir, check_control_flow, &found_control_flow);
166 }
167 foreach_iter(exec_list_iterator, else_iter, ir->else_instructions) {
168 ir_instruction *else_ir = (ir_instruction *)else_iter.get();
169 visit_tree(else_ir, check_control_flow, &found_control_flow);
170 }
171 if (found_control_flow)
172 return visit_continue;
173
174 void *mem_ctx = hieralloc_parent(ir);
175
176 /* Store the condition to a variable so the assignment conditions are
177 * simpler.
178 */
179 cond_var = new(mem_ctx) ir_variable(glsl_type::bool_type,
180 "if_to_cond_assign_condition",
181 ir_var_temporary);
182 ir->insert_before(cond_var);
183
184 deref = new(mem_ctx) ir_dereference_variable(cond_var);
185 assign = new(mem_ctx) ir_assignment(deref,
186 ir->condition, NULL);
187 ir->insert_before(assign);
188
189 /* Now, move all of the instructions out of the if blocks, putting
190 * conditions on assignments.
191 */
192 move_block_to_cond_assign(mem_ctx, ir, cond_var, true);
193 move_block_to_cond_assign(mem_ctx, ir, cond_var, false);
194
195 ir->remove();
196
197 this->progress = true;
198
199 return visit_continue;
200 }
201