1 /*
2 * Copyright © 2014 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 DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #include "brw_vec4.h"
25 #include "brw_vec4_live_variables.h"
26 #include "brw_cfg.h"
27
28 /** @file brw_vec4_dead_code_eliminate.cpp
29 *
30 * Dataflow-aware dead code elimination.
31 *
32 * Walks the instruction list from the bottom, removing instructions that
33 * have results that both aren't used in later blocks and haven't been read
34 * yet in the tail end of this block.
35 */
36
37 using namespace brw;
38
39 bool
dead_code_eliminate()40 vec4_visitor::dead_code_eliminate()
41 {
42 bool progress = false;
43
44 calculate_live_intervals();
45
46 int num_vars = live_intervals->num_vars;
47 BITSET_WORD *live = rzalloc_array(NULL, BITSET_WORD, BITSET_WORDS(num_vars));
48 BITSET_WORD *flag_live = rzalloc_array(NULL, BITSET_WORD, 1);
49
50 foreach_block_reverse_safe(block, cfg) {
51 memcpy(live, live_intervals->block_data[block->num].liveout,
52 sizeof(BITSET_WORD) * BITSET_WORDS(num_vars));
53 memcpy(flag_live, live_intervals->block_data[block->num].flag_liveout,
54 sizeof(BITSET_WORD));
55
56 foreach_inst_in_block_reverse_safe(vec4_instruction, inst, block) {
57 if ((inst->dst.file == VGRF && !inst->has_side_effects()) ||
58 (inst->dst.is_null() && inst->writes_flag())){
59 bool result_live[4] = { false };
60 if (inst->dst.file == VGRF) {
61 for (unsigned i = 0; i < DIV_ROUND_UP(inst->size_written, 16); i++) {
62 for (int c = 0; c < 4; c++) {
63 const unsigned v = var_from_reg(alloc, inst->dst, c, i);
64 result_live[c] |= BITSET_TEST(live, v);
65 }
66 }
67 } else {
68 for (unsigned c = 0; c < 4; c++)
69 result_live[c] = BITSET_TEST(flag_live, c);
70 }
71
72 /* If the instruction can't do writemasking, then it's all or
73 * nothing.
74 */
75 if (!inst->can_do_writemask(devinfo)) {
76 bool result = result_live[0] | result_live[1] |
77 result_live[2] | result_live[3];
78 result_live[0] = result;
79 result_live[1] = result;
80 result_live[2] = result;
81 result_live[3] = result;
82 }
83
84 for (int c = 0; c < 4; c++) {
85 if (!result_live[c] && inst->dst.writemask & (1 << c)) {
86 inst->dst.writemask &= ~(1 << c);
87 progress = true;
88
89 if (inst->dst.writemask == 0) {
90 if (inst->writes_accumulator || inst->writes_flag()) {
91 inst->dst = dst_reg(retype(brw_null_reg(), inst->dst.type));
92 } else {
93 inst->opcode = BRW_OPCODE_NOP;
94 break;
95 }
96 }
97 }
98 }
99 }
100
101 if (inst->dst.is_null() && inst->writes_flag()) {
102 bool combined_live = false;
103 for (unsigned c = 0; c < 4; c++)
104 combined_live |= BITSET_TEST(flag_live, c);
105
106 if (!combined_live) {
107 inst->opcode = BRW_OPCODE_NOP;
108 progress = true;
109 }
110 }
111
112 if (inst->dst.file == VGRF && !inst->predicate &&
113 !inst->is_align1_partial_write()) {
114 for (unsigned i = 0; i < DIV_ROUND_UP(inst->size_written, 16); i++) {
115 for (int c = 0; c < 4; c++) {
116 if (inst->dst.writemask & (1 << c)) {
117 const unsigned v = var_from_reg(alloc, inst->dst, c, i);
118 BITSET_CLEAR(live, v);
119 }
120 }
121 }
122 }
123
124 if (inst->writes_flag() && !inst->predicate && inst->exec_size == 8) {
125 for (unsigned c = 0; c < 4; c++)
126 BITSET_CLEAR(flag_live, c);
127 }
128
129 if (inst->opcode == BRW_OPCODE_NOP) {
130 inst->remove(block);
131 continue;
132 }
133
134 for (int i = 0; i < 3; i++) {
135 if (inst->src[i].file == VGRF) {
136 for (unsigned j = 0; j < DIV_ROUND_UP(inst->size_read(i), 16); j++) {
137 for (int c = 0; c < 4; c++) {
138 const unsigned v = var_from_reg(alloc, inst->src[i], c, j);
139 BITSET_SET(live, v);
140 }
141 }
142 }
143 }
144
145 for (unsigned c = 0; c < 4; c++) {
146 if (inst->reads_flag(c)) {
147 BITSET_SET(flag_live, c);
148 }
149 }
150 }
151 }
152
153 ralloc_free(live);
154 ralloc_free(flag_live);
155
156 if (progress)
157 invalidate_live_intervals();
158
159 return progress;
160 }
161