src/mesa/drivers/dri/i965/brw_fs_register_coalesce.cpp

   1 /*
   2  * Copyright © 2012 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 /** @file brw_fs_register_coalesce.cpp
  25  *
  26  * Implements register coalescing: Checks if the two registers involved in a
  27  * raw move don't interfere, in which case they can both be stored in the same
  28  * place and the MOV removed.
  29  *
  30  * To do this, all uses of the source of the MOV in the shader are replaced
  31  * with the destination of the MOV. For example:
  32  *
  33  * add vgrf3:F, vgrf1:F, vgrf2:F
  34  * mov vgrf4:F, vgrf3:F
  35  * mul vgrf5:F, vgrf5:F, vgrf4:F
  36  *
  37  * becomes
  38  *
  39  * add vgrf4:F, vgrf1:F, vgrf2:F
  40  * mul vgrf5:F, vgrf5:F, vgrf4:F
  41  */
  42
  43 #include "brw_fs.h"
  44 #include "brw_fs_live_variables.h"
  45
  46 static bool
  47 is_nop_mov(const fs_inst *inst)
  48 {
  49    if (inst->opcode == BRW_OPCODE_MOV) {
  50       return inst->dst.equals(inst->src[0]);
  51    }
  52
  53    return false;
  54 }
  55
  56 static bool
  57 is_coalesce_candidate(const fs_inst *inst, const int *virtual_grf_sizes)
  58 {
  59    if (inst->opcode != BRW_OPCODE_MOV ||
  60        inst->is_partial_write() ||
  61        inst->saturate ||
  62        inst->src[0].file != GRF ||
  63        inst->src[0].negate ||
  64        inst->src[0].abs ||
  65        !inst->src[0].is_contiguous() ||
  66        inst->dst.file != GRF ||
  67        inst->dst.type != inst->src[0].type) {
  68       return false;
  69    }
  70
  71    if (virtual_grf_sizes[inst->src[0].reg] >
  72        virtual_grf_sizes[inst->dst.reg])
  73       return false;
  74
  75    return true;
  76 }
  77
  78 static bool
  79 can_coalesce_vars(brw::fs_live_variables *live_intervals,
  80                   const exec_list *instructions, const fs_inst *inst,
  81                   int var_to, int var_from)
  82 {
  83    if (!live_intervals->vars_interfere(var_from, var_to))
  84       return true;
  85
  86    /* We know that the live ranges of A (var_from) and B (var_to)
  87     * interfere because of the ->vars_interfere() call above. If the end
  88     * of B's live range is after the end of A's range, then we know two
  89     * things:
  90     *  - the start of B's live range must be in A's live range (since we
  91     *    already know the two ranges interfere, this is the only remaining
  92     *    possibility)
  93     *  - the interference isn't of the form we're looking for (where B is
  94     *    entirely inside A)
  95     */
  96    if (live_intervals->end[var_to] > live_intervals->end[var_from])
  97       return false;
  98
  99    int scan_ip = -1;
 100
 101    foreach_list(n, instructions) {
 102       fs_inst *scan_inst = (fs_inst *)n;
 103       scan_ip++;
 104
 105       if (scan_inst->is_control_flow())
 106          return false;
 107
 108       if (scan_ip <= live_intervals->start[var_to])
 109          continue;
 110
 111       if (scan_ip > live_intervals->end[var_to])
 112          break;
 113
 114       if (scan_inst->dst.equals(inst->dst) ||
 115           scan_inst->dst.equals(inst->src[0]))
 116          return false;
 117    }
 118
 119    return true;
 120 }
 121
 122 bool
 123 fs_visitor::register_coalesce()
 124 {
 125    bool progress = false;
 126
 127    calculate_live_intervals();
 128
 129    int src_size = 0;
 130    int channels_remaining = 0;
 131    int reg_from = -1, reg_to = -1;
 132    int reg_to_offset[MAX_SAMPLER_MESSAGE_SIZE];
 133    fs_inst *mov[MAX_SAMPLER_MESSAGE_SIZE];
 134    int var_to[MAX_SAMPLER_MESSAGE_SIZE];
 135    int var_from[MAX_SAMPLER_MESSAGE_SIZE];
 136
 137    foreach_list(node, &this->instructions) {
 138       fs_inst *inst = (fs_inst *)node;
 139
 140       if (!is_coalesce_candidate(inst, virtual_grf_sizes))
 141          continue;
 142
 143       if (is_nop_mov(inst)) {
 144          inst->opcode = BRW_OPCODE_NOP;
 145          progress = true;
 146          continue;
 147       }
 148
 149       if (reg_from != inst->src[0].reg) {
 150          reg_from = inst->src[0].reg;
 151
 152          src_size = virtual_grf_sizes[inst->src[0].reg];
 153          assert(src_size <= MAX_SAMPLER_MESSAGE_SIZE);
 154
 155          channels_remaining = src_size;
 156          memset(mov, 0, sizeof(mov));
 157
 158          reg_to = inst->dst.reg;
 159       }
 160
 161       if (reg_to != inst->dst.reg)
 162          continue;
 163
 164       const int offset = inst->src[0].reg_offset;
 165       reg_to_offset[offset] = inst->dst.reg_offset;
 166       mov[offset] = inst;
 167       channels_remaining--;
 168
 169       if (channels_remaining)
 170          continue;
 171
 172       bool can_coalesce = true;
 173       for (int i = 0; i < src_size; i++) {
 174          var_to[i] = live_intervals->var_from_vgrf[reg_to] + reg_to_offset[i];
 175          var_from[i] = live_intervals->var_from_vgrf[reg_from] + i;
 176
 177          if (!can_coalesce_vars(live_intervals, &instructions, inst,
 178                                 var_to[i], var_from[i])) {
 179             can_coalesce = false;
 180             reg_from = -1;
 181             break;
 182          }
 183       }
 184
 185       if (!can_coalesce)
 186          continue;
 187
 188       progress = true;
 189
 190       for (int i = 0; i < src_size; i++) {
 191          if (mov[i]) {
 192             mov[i]->opcode = BRW_OPCODE_NOP;
 193             mov[i]->conditional_mod = BRW_CONDITIONAL_NONE;
 194             mov[i]->dst = reg_undef;
 195             mov[i]->src[0] = reg_undef;
 196             mov[i]->src[1] = reg_undef;
 197             mov[i]->src[2] = reg_undef;
 198          }
 199       }
 200
 201       foreach_list(node, &this->instructions) {
 202          fs_inst *scan_inst = (fs_inst *)node;
 203
 204          for (int i = 0; i < src_size; i++) {
 205             if (mov[i]) {
 206                if (scan_inst->dst.file == GRF &&
 207                    scan_inst->dst.reg == reg_from &&
 208                    scan_inst->dst.reg_offset == i) {
 209                   scan_inst->dst.reg = reg_to;
 210                   scan_inst->dst.reg_offset = reg_to_offset[i];
 211                }
 212                for (int j = 0; j < scan_inst->sources; j++) {
 213                   if (scan_inst->src[j].file == GRF &&
 214                       scan_inst->src[j].reg == reg_from &&
 215                       scan_inst->src[j].reg_offset == i) {
 216                      scan_inst->src[j].reg = reg_to;
 217                      scan_inst->src[j].reg_offset = reg_to_offset[i];
 218                   }
 219                }
 220             }
 221          }
 222       }
 223
 224       for (int i = 0; i < src_size; i++) {
 225          live_intervals->start[var_to[i]] =
 226             MIN2(live_intervals->start[var_to[i]],
 227                  live_intervals->start[var_from[i]]);
 228          live_intervals->end[var_to[i]] =
 229             MAX2(live_intervals->end[var_to[i]],
 230                  live_intervals->end[var_from[i]]);
 231       }
 232       reg_from = -1;
 233    }
 234
 235    if (progress) {
 236       foreach_list_safe(node, &this->instructions) {
 237          fs_inst *inst = (fs_inst *)node;
 238
 239          if (inst->opcode == BRW_OPCODE_NOP) {
 240             inst->remove();
 241          }
 242       }
 243
 244       invalidate_live_intervals();
 245    }
 246
 247    return progress;
 248 }