summaryrefslogtreecommitdiff
path: root/src/mesa/drivers/dri/i965/brw_vec4_cse.cpp
blob: 30a4098b339257b68aff4c14009ad12611b2e9fd (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
/*
 * Copyright © 2012, 2013, 2014 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#include "brw_vec4.h"
#include "brw_cfg.h"

using namespace brw;

/** @file brw_vec4_cse.cpp
 *
 * Support for local common subexpression elimination.
 *
 * See Muchnick's Advanced Compiler Design and Implementation, section
 * 13.1 (p378).
 */

namespace {
struct aeb_entry : public exec_node {
   /** The instruction that generates the expression value. */
   vec4_instruction *generator;

   /** The temporary where the value is stored. */
   src_reg tmp;
};
}

static bool
is_expression(const vec4_instruction *const inst)
{
   switch (inst->opcode) {
   case BRW_OPCODE_MOV:
   case BRW_OPCODE_SEL:
   case BRW_OPCODE_NOT:
   case BRW_OPCODE_AND:
   case BRW_OPCODE_OR:
   case BRW_OPCODE_XOR:
   case BRW_OPCODE_SHR:
   case BRW_OPCODE_SHL:
   case BRW_OPCODE_ASR:
   case BRW_OPCODE_CMP:
   case BRW_OPCODE_CMPN:
   case BRW_OPCODE_ADD:
   case BRW_OPCODE_MUL:
   case BRW_OPCODE_FRC:
   case BRW_OPCODE_RNDU:
   case BRW_OPCODE_RNDD:
   case BRW_OPCODE_RNDE:
   case BRW_OPCODE_RNDZ:
   case BRW_OPCODE_LINE:
   case BRW_OPCODE_PLN:
   case BRW_OPCODE_MAD:
   case BRW_OPCODE_LRP:
   case VEC4_OPCODE_UNPACK_UNIFORM:
      return true;
   case SHADER_OPCODE_RCP:
   case SHADER_OPCODE_RSQ:
   case SHADER_OPCODE_SQRT:
   case SHADER_OPCODE_EXP2:
   case SHADER_OPCODE_LOG2:
   case SHADER_OPCODE_POW:
   case SHADER_OPCODE_INT_QUOTIENT:
   case SHADER_OPCODE_INT_REMAINDER:
   case SHADER_OPCODE_SIN:
   case SHADER_OPCODE_COS:
      return inst->mlen == 0;
   default:
      return false;
   }
}

static bool
is_expression_commutative(const vec4_instruction *inst)
{
   switch (inst->opcode) {
   case BRW_OPCODE_AND:
   case BRW_OPCODE_OR:
   case BRW_OPCODE_XOR:
   case BRW_OPCODE_ADD:
   case BRW_OPCODE_MUL:
      return true;
   case BRW_OPCODE_SEL:
      /* MIN and MAX are commutative. */
      if (inst->conditional_mod == BRW_CONDITIONAL_GE ||
          inst->conditional_mod == BRW_CONDITIONAL_L) {
         return true;
      }
      /* fallthrough */
   default:
      return false;
   }
}

static bool
operands_match(const vec4_instruction *a, const vec4_instruction *b)
{
   const src_reg *xs = a->src;
   const src_reg *ys = b->src;

   if (a->opcode == BRW_OPCODE_MAD) {
      return xs[0].equals(ys[0]) &&
             ((xs[1].equals(ys[1]) && xs[2].equals(ys[2])) ||
              (xs[2].equals(ys[1]) && xs[1].equals(ys[2])));
   } else if (!is_expression_commutative(a)) {
      return xs[0].equals(ys[0]) && xs[1].equals(ys[1]) && xs[2].equals(ys[2]);
   } else {
      return (xs[0].equals(ys[0]) && xs[1].equals(ys[1])) ||
             (xs[1].equals(ys[0]) && xs[0].equals(ys[1]));
   }
}

static bool
instructions_match(vec4_instruction *a, vec4_instruction *b)
{
   return a->opcode == b->opcode &&
          a->saturate == b->saturate &&
          a->conditional_mod == b->conditional_mod &&
          a->dst.type == b->dst.type &&
          a->dst.writemask == b->dst.writemask &&
          operands_match(a, b);
}

bool
vec4_visitor::opt_cse_local(bblock_t *block)
{
   bool progress = false;
   exec_list aeb;

   void *cse_ctx = ralloc_context(NULL);

   int ip = block->start_ip;
   foreach_inst_in_block (vec4_instruction, inst, block) {
      /* Skip some cases. */
      if (is_expression(inst) && !inst->predicate && inst->mlen == 0 &&
          (inst->dst.file != HW_REG || inst->dst.is_null()))
      {
         bool found = false;

         foreach_in_list_use_after(aeb_entry, entry, &aeb) {
            /* Match current instruction's expression against those in AEB. */
            if (instructions_match(inst, entry->generator)) {
               found = true;
               progress = true;
               break;
            }
         }

         if (!found) {
            if (inst->opcode != BRW_OPCODE_MOV ||
                (inst->opcode == BRW_OPCODE_MOV &&
                 inst->src[0].file == IMM &&
                 inst->src[0].type == BRW_REGISTER_TYPE_VF)) {
               /* Our first sighting of this expression.  Create an entry. */
               aeb_entry *entry = ralloc(cse_ctx, aeb_entry);
               entry->tmp = src_reg(); /* file will be BAD_FILE */
               entry->generator = inst;
               aeb.push_tail(entry);
            }
         } else {
            /* This is at least our second sighting of this expression.
             * If we don't have a temporary already, make one.
             */
            bool no_existing_temp = entry->tmp.file == BAD_FILE;
            if (no_existing_temp && !entry->generator->dst.is_null()) {
               entry->tmp = src_reg(this, glsl_type::float_type);
               entry->tmp.type = inst->dst.type;
               entry->tmp.swizzle = BRW_SWIZZLE_XYZW;

               vec4_instruction *copy = MOV(entry->generator->dst, entry->tmp);
               entry->generator->insert_after(block, copy);
               entry->generator->dst = dst_reg(entry->tmp);
            }

            /* dest <- temp */
            if (!inst->dst.is_null()) {
               assert(inst->dst.type == entry->tmp.type);
               vec4_instruction *copy = MOV(inst->dst, entry->tmp);
               copy->force_writemask_all = inst->force_writemask_all;
               inst->insert_before(block, copy);
            }

            /* Set our iterator so that next time through the loop inst->next
             * will get the instruction in the basic block after the one we've
             * removed.
             */
            vec4_instruction *prev = (vec4_instruction *)inst->prev;

            inst->remove(block);
            inst = prev;
         }
      }

      foreach_in_list_safe(aeb_entry, entry, &aeb) {
         /* Kill all AEB entries that write a different value to or read from
          * the flag register if we just wrote it.
          */
         if (inst->writes_flag()) {
            if (entry->generator->reads_flag() ||
                (entry->generator->writes_flag() &&
                 !instructions_match(inst, entry->generator))) {
               entry->remove();
               ralloc_free(entry);
               continue;
            }
         }

         for (int i = 0; i < 3; i++) {
            src_reg *src = &entry->generator->src[i];

            /* Kill all AEB entries that use the destination we just
             * overwrote.
             */
            if (inst->dst.file == entry->generator->src[i].file &&
                inst->dst.reg == entry->generator->src[i].reg) {
               entry->remove();
               ralloc_free(entry);
               break;
            }

            /* Kill any AEB entries using registers that don't get reused any
             * more -- a sure sign they'll fail operands_match().
             */
            if (src->file == GRF) {
               assert((src->reg * 4 + 3) < (virtual_grf_count * 4));

               int last_reg_use = MAX2(MAX2(virtual_grf_end[src->reg * 4 + 0],
                                            virtual_grf_end[src->reg * 4 + 1]),
                                       MAX2(virtual_grf_end[src->reg * 4 + 2],
                                            virtual_grf_end[src->reg * 4 + 3]));
               if (last_reg_use < ip) {
                  entry->remove();
                  ralloc_free(entry);
                  break;
               }
            }
         }
      }

      ip++;
   }

   ralloc_free(cse_ctx);

   return progress;
}

bool
vec4_visitor::opt_cse()
{
   bool progress = false;

   calculate_live_intervals();

   foreach_block (block, cfg) {
      progress = opt_cse_local(block) || progress;
   }

   if (progress)
      invalidate_live_intervals();

   return progress;
}