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
37using namespace brw;
38
39bool
40vec4_visitor::dead_code_eliminate()
41{
42   bool progress = false;
43
44   const vec4_live_variables &live_vars = live_analysis.require();
45   int num_vars = live_vars.num_vars;
46   BITSET_WORD *live = rzalloc_array(NULL, BITSET_WORD, BITSET_WORDS(num_vars));
47   BITSET_WORD *flag_live = rzalloc_array(NULL, BITSET_WORD, 1);
48
49   foreach_block_reverse_safe(block, cfg) {
50      memcpy(live, live_vars.block_data[block->num].liveout,
51             sizeof(BITSET_WORD) * BITSET_WORDS(num_vars));
52      memcpy(flag_live, live_vars.block_data[block->num].flag_liveout,
53             sizeof(BITSET_WORD));
54
55      foreach_inst_in_block_reverse_safe(vec4_instruction, inst, block) {
56         if ((inst->dst.file == VGRF && !inst->has_side_effects()) ||
57             (inst->dst.is_null() && inst->writes_flag(devinfo))){
58            bool result_live[4] = { false };
59            if (inst->dst.file == VGRF) {
60               for (unsigned i = 0; i < DIV_ROUND_UP(inst->size_written, 16); i++) {
61                  for (int c = 0; c < 4; c++) {
62                     const unsigned v = var_from_reg(alloc, inst->dst, c, i);
63                     result_live[c] |= BITSET_TEST(live, v);
64                  }
65               }
66            } else {
67               for (unsigned c = 0; c < 4; c++)
68                  result_live[c] = BITSET_TEST(flag_live, c);
69            }
70
71            /* If the instruction can't do writemasking, then it's all or
72             * nothing.
73             */
74            if (!inst->can_do_writemask(devinfo)) {
75               bool result = result_live[0] | result_live[1] |
76                             result_live[2] | result_live[3];
77               result_live[0] = result;
78               result_live[1] = result;
79               result_live[2] = result;
80               result_live[3] = result;
81            }
82
83            if (inst->writes_flag(devinfo)) {
84               /* Independently calculate the usage of the flag components and
85                * the destination value components.
86                */
87               uint8_t flag_mask = inst->dst.writemask;
88               uint8_t dest_mask = inst->dst.writemask;
89
90               for (int c = 0; c < 4; c++) {
91                  if (!result_live[c] && dest_mask & (1 << c))
92                     dest_mask &= ~(1 << c);
93
94                  if (!BITSET_TEST(flag_live, c))
95                     flag_mask &= ~(1 << c);
96               }
97
98               if (inst->dst.writemask != (flag_mask | dest_mask)) {
99                  progress = true;
100                  inst->dst.writemask = flag_mask | dest_mask;
101               }
102
103               /* If none of the destination components are read, replace the
104                * destination register with the NULL register.
105                */
106               if (dest_mask == 0) {
107                  progress = true;
108                  inst->dst = dst_reg(retype(brw_null_reg(), inst->dst.type));
109               }
110            } else {
111               for (int c = 0; c < 4; c++) {
112                  if (!result_live[c] && inst->dst.writemask & (1 << c)) {
113                     inst->dst.writemask &= ~(1 << c);
114                     progress = true;
115
116                     if (inst->dst.writemask == 0) {
117                        if (inst->writes_accumulator) {
118                           inst->dst = dst_reg(retype(brw_null_reg(), inst->dst.type));
119                        } else {
120                           inst->opcode = BRW_OPCODE_NOP;
121                           break;
122                        }
123                     }
124                  }
125               }
126            }
127         }
128
129         if (inst->dst.is_null() && inst->writes_flag(devinfo)) {
130            bool combined_live = false;
131            for (unsigned c = 0; c < 4; c++)
132               combined_live |= BITSET_TEST(flag_live, c);
133
134            if (!combined_live) {
135               inst->opcode = BRW_OPCODE_NOP;
136               progress = true;
137            }
138         }
139
140         if (inst->dst.file == VGRF && !inst->predicate &&
141             !inst->is_align1_partial_write()) {
142            for (unsigned i = 0; i < DIV_ROUND_UP(inst->size_written, 16); i++) {
143               for (int c = 0; c < 4; c++) {
144                  if (inst->dst.writemask & (1 << c)) {
145                     const unsigned v = var_from_reg(alloc, inst->dst, c, i);
146                     BITSET_CLEAR(live, v);
147                  }
148               }
149            }
150         }
151
152         if (inst->writes_flag(devinfo) && !inst->predicate && inst->exec_size == 8) {
153            for (unsigned c = 0; c < 4; c++)
154               BITSET_CLEAR(flag_live, c);
155         }
156
157         if (inst->opcode == BRW_OPCODE_NOP) {
158            inst->remove(block);
159            continue;
160         }
161
162         for (int i = 0; i < 3; i++) {
163            if (inst->src[i].file == VGRF) {
164               for (unsigned j = 0; j < DIV_ROUND_UP(inst->size_read(i), 16); j++) {
165                  for (int c = 0; c < 4; c++) {
166                     const unsigned v = var_from_reg(alloc, inst->src[i], c, j);
167                     BITSET_SET(live, v);
168                  }
169               }
170            }
171         }
172
173         for (unsigned c = 0; c < 4; c++) {
174            if (inst->reads_flag(c)) {
175               BITSET_SET(flag_live, c);
176            }
177         }
178      }
179   }
180
181   ralloc_free(live);
182   ralloc_free(flag_live);
183
184   if (progress)
185      invalidate_analysis(DEPENDENCY_INSTRUCTIONS);
186
187   return progress;
188}
189