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       /* We know that the live ranges of A (var_from) and B (var_to)
  85        * interfere because of the ->vars_interfere() call above. If the end
  86        * of B's live range is after the end of A's range, then we know two
  87        * things:
  88        *  - the start of B's live range must be in A's live range (since we
  89        *    already know the two ranges interfere, this is the only remaining
  90        *    possibility)
  91        *  - the interference isn't of the form we're looking for (where B is
  92        *    entirely inside A)
  93        */
  94       if (live_intervals->end[var_to] > live_intervals->end[var_from])
  95          return false;
  96
  97       int scan_ip = -1;
  98
  99       foreach_list(n, instructions) {
 100          fs_inst *scan_inst = (fs_inst *)n;
 101          scan_ip++;
 102
 103          if (scan_inst->is_control_flow())
 104             return false;
 105
 106          if (scan_ip <= live_intervals->start[var_to])
 107             continue;
 108
 109          if (scan_ip > live_intervals->end[var_to])
 110             break;
 111
 112          if (scan_inst->dst.equals(inst->dst) ||
 113              scan_inst->dst.equals(inst->src[0]))
 114             return false;
 115       }
 116    }
 117
 118    return true;
 119 }
 120
 121 bool
 122 fs_visitor::register_coalesce()
 123 {
 124    bool progress = false;
 125
 126    calculate_live_intervals();
 127
 128    int src_size = 0;
 129    int channels_remaining = 0;
 130    int reg_from = -1, reg_to = -1;
 131    int reg_to_offset[MAX_SAMPLER_MESSAGE_SIZE];
 132    fs_inst *mov[MAX_SAMPLER_MESSAGE_SIZE];
 133    int var_to[MAX_SAMPLER_MESSAGE_SIZE];
 134    int var_from[MAX_SAMPLER_MESSAGE_SIZE];
 135
 136    foreach_list(node, &this->instructions) {
 137       fs_inst *inst = (fs_inst *)node;
 138
 139       if (!is_coalesce_candidate(inst, virtual_grf_sizes))
 140          continue;
 141
 142       if (is_nop_mov(inst)) {
 143          inst->opcode = BRW_OPCODE_NOP;
 144          progress = true;
 145          continue;
 146       }
 147
 148       if (reg_from != inst->src[0].reg) {
 149          reg_from = inst->src[0].reg;
 150
 151          src_size = virtual_grf_sizes[inst->src[0].reg];
 152          assert(src_size <= MAX_SAMPLER_MESSAGE_SIZE);
 153
 154          channels_remaining = src_size;
 155          memset(mov, 0, sizeof(mov));
 156
 157          reg_to = inst->dst.reg;
 158       }
 159
 160       if (reg_to != inst->dst.reg)
 161          continue;
 162
 163       const int offset = inst->src[0].reg_offset;
 164       reg_to_offset[offset] = inst->dst.reg_offset;
 165       mov[offset] = inst;
 166       channels_remaining--;
 167
 168       if (channels_remaining)
 169          continue;
 170
 171       bool can_coalesce = true;
 172       for (int i = 0; i < src_size; i++) {
 173          var_to[i] = live_intervals->var_from_vgrf[reg_to] + reg_to_offset[i];
 174          var_from[i] = live_intervals->var_from_vgrf[reg_from] + i;
 175
 176          if (!can_coalesce_vars(live_intervals, &instructions, inst,
 177                                 var_to[i], var_from[i])) {
 178             can_coalesce = false;
 179             reg_from = -1;
 180             break;
 181          }
 182       }
 183
 184       if (!can_coalesce)
 185          continue;
 186
 187       progress = true;
 188
 189       for (int i = 0; i < src_size; i++) {
 190          if (mov[i]) {
 191             mov[i]->opcode = BRW_OPCODE_NOP;
 192             mov[i]->conditional_mod = BRW_CONDITIONAL_NONE;
 193             mov[i]->dst = reg_undef;
 194             mov[i]->src[0] = reg_undef;
 195             mov[i]->src[1] = reg_undef;
 196             mov[i]->src[2] = reg_undef;
 197          }
 198       }
 199
 200       foreach_list(node, &this->instructions) {
 201          fs_inst *scan_inst = (fs_inst *)node;
 202
 203          for (int i = 0; i < src_size; i++) {
 204             if (mov[i]) {
 205                if (scan_inst->dst.file == GRF &&
 206                    scan_inst->dst.reg == reg_from &&
 207                    scan_inst->dst.reg_offset == i) {
 208                   scan_inst->dst.reg = reg_to;
 209                   scan_inst->dst.reg_offset = reg_to_offset[i];
 210                }
 211                for (int j = 0; j < 3; j++) {
 212                   if (scan_inst->src[j].file == GRF &&
 213                       scan_inst->src[j].reg == reg_from &&
 214                       scan_inst->src[j].reg_offset == i) {
 215                      scan_inst->src[j].reg = reg_to;
 216                      scan_inst->src[j].reg_offset = reg_to_offset[i];
 217                   }
 218                }
 219             }
 220          }
 221       }
 222
 223       for (int i = 0; i < src_size; i++) {
 224          live_intervals->start[var_to[i]] =
 225             MIN2(live_intervals->start[var_to[i]],
 226                  live_intervals->start[var_from[i]]);
 227          live_intervals->end[var_to[i]] =
 228             MAX2(live_intervals->end[var_to[i]],
 229                  live_intervals->end[var_from[i]]);
 230       }
 231       reg_from = -1;
 232    }
 233
 234    if (progress) {
 235       foreach_list_safe(node, &this->instructions) {
 236          fs_inst *inst = (fs_inst *)node;
 237
 238          if (inst->opcode == BRW_OPCODE_NOP) {
 239             inst->remove();
 240          }
 241       }
 242
 243       invalidate_live_intervals();
 244    }
 245
 246    return progress;
 247 }