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_RSR:
  57    case BRW_OPCODE_RSL:
  58    case BRW_OPCODE_ASR:
  59    case BRW_OPCODE_ADD:
  60    case BRW_OPCODE_MUL:
  61    case BRW_OPCODE_FRC:
  62    case BRW_OPCODE_RNDU:
  63    case BRW_OPCODE_RNDD:
  64    case BRW_OPCODE_RNDE:
  65    case BRW_OPCODE_RNDZ:
  66    case BRW_OPCODE_LINE:
  67    case BRW_OPCODE_PLN:
  68    case BRW_OPCODE_MAD:
  69    case BRW_OPCODE_LRP:
  70    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
  71    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
  72    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD:
  73    case FS_OPCODE_CINTERP:
  74    case FS_OPCODE_LINTERP:
  75       return true;
  76    default:
  77       return false;
  78    }
  79 }
  80
  81 static bool
  82 operands_match(fs_reg *xs, fs_reg *ys)
  83 {
  84    return xs[0].equals(ys[0]) && xs[1].equals(ys[1]) && xs[2].equals(ys[2]);
  85 }
  86
  87 bool
  88 fs_visitor::opt_cse_local(bblock_t *block, exec_list *aeb)
  89 {
  90    bool progress = false;
  91
  92    void *mem_ctx = ralloc_context(this->mem_ctx);
  93
  94    int ip = block->start_ip;
  95    for (fs_inst *inst = (fs_inst *)block->start;
  96         inst != block->end->next;
  97         inst = (fs_inst *) inst->next) {
  98
  99       /* Skip some cases. */
 100       if (is_expression(inst) &&
 101           !inst->predicate &&
 102           !inst->is_partial_write() &&
 103           !inst->conditional_mod)
 104       {
 105          bool found = false;
 106
 107          aeb_entry *entry;
 108          foreach_list(entry_node, aeb) {
 109             entry = (aeb_entry *) entry_node;
 110
 111             /* Match current instruction's expression against those in AEB. */
 112             if (inst->opcode == entry->generator->opcode &&
 113                 inst->saturate == entry->generator->saturate &&
 114                 inst->dst.type == entry->generator->dst.type &&
 115                 operands_match(entry->generator->src, inst->src)) {
 116
 117                found = true;
 118                progress = true;
 119                break;
 120             }
 121          }
 122
 123          if (!found) {
 124             /* Our first sighting of this expression.  Create an entry. */
 125             aeb_entry *entry = ralloc(mem_ctx, aeb_entry);
 126             entry->tmp = reg_undef;
 127             entry->generator = inst;
 128             aeb->push_tail(entry);
 129          } else {
 130             /* This is at least our second sighting of this expression.
 131              * If we don't have a temporary already, make one.
 132              */
 133             bool no_existing_temp = entry->tmp.file == BAD_FILE;
 134             if (no_existing_temp) {
 135                int written = entry->generator->regs_written;
 136
 137                fs_reg orig_dst = entry->generator->dst;
 138                fs_reg tmp = fs_reg(GRF, virtual_grf_alloc(written),
 139                                    orig_dst.type);
 140                entry->tmp = tmp;
 141                entry->generator->dst = tmp;
 142
 143                for (int i = 0; i < written; i++) {
 144                   fs_inst *copy = MOV(orig_dst, tmp);
 145                   copy->force_writemask_all =
 146                      entry->generator->force_writemask_all;
 147                   entry->generator->insert_after(copy);
 148
 149                   orig_dst.reg_offset++;
 150                   tmp.reg_offset++;
 151                }
 152             }
 153
 154             /* dest <- temp */
 155             int written = inst->regs_written;
 156             assert(written == entry->generator->regs_written);
 157             assert(inst->dst.type == entry->tmp.type);
 158             fs_reg dst = inst->dst;
 159             fs_reg tmp = entry->tmp;
 160             fs_inst *copy = NULL;
 161             for (int i = 0; i < written; i++) {
 162                copy = MOV(dst, tmp);
 163                copy->force_writemask_all = inst->force_writemask_all;
 164                inst->insert_before(copy);
 165
 166                dst.reg_offset++;
 167                tmp.reg_offset++;
 168             }
 169             inst->remove();
 170
 171             /* Appending an instruction may have changed our bblock end. */
 172             if (inst == block->end) {
 173                block->end = copy;
 174             }
 175
 176             /* Continue iteration with copy->next */
 177             inst = copy;
 178          }
 179       }
 180
 181       foreach_list_safe(entry_node, aeb) {
 182          aeb_entry *entry = (aeb_entry *)entry_node;
 183
 184          for (int i = 0; i < 3; i++) {
 185             fs_reg *src_reg = &entry->generator->src[i];
 186
 187             /* Kill all AEB entries that use the destination we just
 188              * overwrote.
 189              */
 190             if (inst->overwrites_reg(entry->generator->src[i])) {
 191                entry->remove();
 192                ralloc_free(entry);
 193                break;
 194             }
 195
 196             /* Kill any AEB entries using registers that don't get reused any
 197              * more -- a sure sign they'll fail operands_match().
 198              */
 199             if (src_reg->file == GRF && virtual_grf_end[src_reg->reg] < ip) {
 200                entry->remove();
 201                ralloc_free(entry);
 202                break;
 203             }
 204          }
 205       }
 206
 207       ip++;
 208    }
 209
 210    ralloc_free(mem_ctx);
 211
 212    if (progress)
 213       this->live_intervals_valid = false;
 214
 215    return progress;
 216 }
 217
 218 bool
 219 fs_visitor::opt_cse()
 220 {
 221    bool progress = false;
 222
 223    calculate_live_intervals();
 224
 225    cfg_t cfg(this);
 226
 227    for (int b = 0; b < cfg.num_blocks; b++) {
 228       bblock_t *block = cfg.blocks[b];
 229       exec_list aeb;
 230
 231       progress = opt_cse_local(block, &aeb) || progress;
 232    }
 233
 234    return progress;
 235 }