src/mesa/drivers/dri/i965/brw_fs_cse.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 #include "brw_fs.h"
  25 #include "brw_cfg.h"
  26
  27 /** @file brw_fs_cse.cpp
  28  *
  29  * Support for local common subexpression elimination.
  30  *
  31  * See Muchnik's Advanced Compiler Design and Implementation, section
  32  * 13.1 (p378).
  33  */
  34
  35 namespace {
  36 struct aeb_entry : public exec_node {
  37    /** The instruction that generates the expression value. */
  38    fs_inst *generator;
  39
  40    /** The temporary where the value is stored. */
  41    fs_reg tmp;
  42 };
  43 }
  44
  45 static bool
  46 is_expression(const fs_inst *const inst)
  47 {
  48    switch (inst->opcode) {
  49    case BRW_OPCODE_SEL:
  50    case BRW_OPCODE_NOT:
  51    case BRW_OPCODE_AND:
  52    case BRW_OPCODE_OR:
  53    case BRW_OPCODE_XOR:
  54    case BRW_OPCODE_SHR:
  55    case BRW_OPCODE_SHL:
  56    case BRW_OPCODE_ASR:
  57    case BRW_OPCODE_ADD:
  58    case BRW_OPCODE_MUL:
  59    case BRW_OPCODE_FRC:
  60    case BRW_OPCODE_RNDU:
  61    case BRW_OPCODE_RNDD:
  62    case BRW_OPCODE_RNDE:
  63    case BRW_OPCODE_RNDZ:
  64    case BRW_OPCODE_LINE:
  65    case BRW_OPCODE_PLN:
  66    case BRW_OPCODE_MAD:
  67    case BRW_OPCODE_LRP:
  68    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
  69    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
  70    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD:
  71    case FS_OPCODE_CINTERP:
  72    case FS_OPCODE_LINTERP:
  73       return true;
  74    default:
  75       return false;
  76    }
  77 }
  78
  79 static bool
  80 operands_match(fs_reg *xs, fs_reg *ys)
  81 {
  82    return xs[0].equals(ys[0]) && xs[1].equals(ys[1]) && xs[2].equals(ys[2]);
  83 }
  84
  85 bool
  86 fs_visitor::opt_cse_local(bblock_t *block, exec_list *aeb)
  87 {
  88    bool progress = false;
  89
  90    void *mem_ctx = ralloc_context(this->mem_ctx);
  91
  92    int ip = block->start_ip;
  93    for (fs_inst *inst = (fs_inst *)block->start;
  94         inst != block->end->next;
  95         inst = (fs_inst *) inst->next) {
  96
  97       /* Skip some cases. */
  98       if (is_expression(inst) &&
  99           !inst->predicate &&
 100           !inst->is_partial_write() &&
 101           !inst->conditional_mod &&
 102           inst->dst.file != HW_REG)
 103       {
 104          bool found = false;
 105
 106          aeb_entry *entry;
 107          foreach_list(entry_node, aeb) {
 108             entry = (aeb_entry *) entry_node;
 109
 110             /* Match current instruction's expression against those in AEB. */
 111             if (inst->opcode == entry->generator->opcode &&
 112                 inst->saturate == entry->generator->saturate &&
 113                 inst->dst.type == entry->generator->dst.type &&
 114                 operands_match(entry->generator->src, inst->src)) {
 115
 116                found = true;
 117                progress = true;
 118                break;
 119             }
 120          }
 121
 122          if (!found) {
 123             /* Our first sighting of this expression.  Create an entry. */
 124             aeb_entry *entry = ralloc(mem_ctx, aeb_entry);
 125             entry->tmp = reg_undef;
 126             entry->generator = inst;
 127             aeb->push_tail(entry);
 128          } else {
 129             /* This is at least our second sighting of this expression.
 130              * If we don't have a temporary already, make one.
 131              */
 132             bool no_existing_temp = entry->tmp.file == BAD_FILE;
 133             if (no_existing_temp) {
 134                int written = entry->generator->regs_written;
 135
 136                fs_reg orig_dst = entry->generator->dst;
 137                fs_reg tmp = fs_reg(GRF, virtual_grf_alloc(written),
 138                                    orig_dst.type);
 139                entry->tmp = tmp;
 140                entry->generator->dst = tmp;
 141
 142                for (int i = 0; i < written; i++) {
 143                   fs_inst *copy = MOV(orig_dst, tmp);
 144                   copy->force_writemask_all =
 145                      entry->generator->force_writemask_all;
 146                   entry->generator->insert_after(copy);
 147
 148                   orig_dst.reg_offset++;
 149                   tmp.reg_offset++;
 150                }
 151             }
 152
 153             /* dest <- temp */
 154             int written = inst->regs_written;
 155             assert(written == entry->generator->regs_written);
 156             assert(inst->dst.type == entry->tmp.type);
 157             fs_reg dst = inst->dst;
 158             fs_reg tmp = entry->tmp;
 159             fs_inst *copy = NULL;
 160             for (int i = 0; i < written; i++) {
 161                copy = MOV(dst, tmp);
 162                copy->force_writemask_all = inst->force_writemask_all;
 163                inst->insert_before(copy);
 164
 165                dst.reg_offset++;
 166                tmp.reg_offset++;
 167             }
 168             inst->remove();
 169
 170             /* Appending an instruction may have changed our bblock end. */
 171             if (inst == block->end) {
 172                block->end = copy;
 173             }
 174
 175             /* Continue iteration with copy->next */
 176             inst = copy;
 177          }
 178       }
 179
 180       foreach_list_safe(entry_node, aeb) {
 181          aeb_entry *entry = (aeb_entry *)entry_node;
 182
 183          for (int i = 0; i < 3; i++) {
 184             fs_reg *src_reg = &entry->generator->src[i];
 185
 186             /* Kill all AEB entries that use the destination we just
 187              * overwrote.
 188              */
 189             if (inst->overwrites_reg(entry->generator->src[i])) {
 190                entry->remove();
 191                ralloc_free(entry);
 192                break;
 193             }
 194
 195             /* Kill any AEB entries using registers that don't get reused any
 196              * more -- a sure sign they'll fail operands_match().
 197              */
 198             if (src_reg->file == GRF && virtual_grf_end[src_reg->reg] < ip) {
 199                entry->remove();
 200                ralloc_free(entry);
 201                break;
 202             }
 203          }
 204       }
 205
 206       ip++;
 207    }
 208
 209    ralloc_free(mem_ctx);
 210
 211    if (progress)
 212       invalidate_live_intervals();
 213
 214    return progress;
 215 }
 216
 217 bool
 218 fs_visitor::opt_cse()
 219 {
 220    bool progress = false;
 221
 222    calculate_live_intervals();
 223
 224    cfg_t cfg(this);
 225
 226    for (int b = 0; b < cfg.num_blocks; b++) {
 227       bblock_t *block = cfg.blocks[b];
 228       exec_list aeb;
 229
 230       progress = opt_cse_local(block, &aeb) || progress;
 231    }
 232
 233    return progress;
 234 }