src/intel/compiler/brw_fs_dead_code_eliminate.cpp

   1 /*
   2  * Copyright © 2014 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_fs_live_variables.h"
  26 #include "brw_cfg.h"
  27
  28 /** @file brw_fs_dead_code_eliminate.cpp
  29  *
  30  * Dataflow-aware dead code elimination.
  31  *
  32  * Walks the instruction list from the bottom, removing instructions that
  33  * have results that both aren't used in later blocks and haven't been read
  34  * yet in the tail end of this block.
  35  */
  36
  37 using namespace brw;
  38
  39 /**
  40  * Is it safe to eliminate the instruction?
  41  */
  42 static bool
  43 can_eliminate(const fs_inst *inst, BITSET_WORD *flag_live)
  44 {
  45     return !inst->is_control_flow() &&
  46            !inst->has_side_effects() &&
  47            !(flag_live[0] & inst->flags_written()) &&
  48            !inst->writes_accumulator;
  49 }
  50
  51 /**
  52  * Is it safe to omit the write, making the destination ARF null?
  53  */
  54 static bool
  55 can_omit_write(const fs_inst *inst)
  56 {
  57    switch (inst->opcode) {
  58    case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
  59    case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL:
  60    case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
  61       return true;
  62    default:
  63       /* We can eliminate the destination write for ordinary instructions,
  64        * but not most SENDs.
  65        */
  66       if (inst->opcode < 128 && inst->mlen == 0)
  67          return true;
  68
  69       /* It might not be safe for other virtual opcodes. */
  70       return false;
  71    }
  72 }
  73
  74 bool
  75 fs_visitor::dead_code_eliminate()
  76 {
  77    bool progress = false;
  78
  79    const fs_live_variables &live_vars = live_analysis.require();
  80    int num_vars = live_vars.num_vars;
  81    BITSET_WORD *live = rzalloc_array(NULL, BITSET_WORD, BITSET_WORDS(num_vars));
  82    BITSET_WORD *flag_live = rzalloc_array(NULL, BITSET_WORD, 1);
  83
  84    foreach_block_reverse_safe(block, cfg) {
  85       memcpy(live, live_vars.block_data[block->num].liveout,
  86              sizeof(BITSET_WORD) * BITSET_WORDS(num_vars));
  87       memcpy(flag_live, live_vars.block_data[block->num].flag_liveout,
  88              sizeof(BITSET_WORD));
  89
  90       foreach_inst_in_block_reverse_safe(fs_inst, inst, block) {
  91          if (inst->dst.file == VGRF) {
  92             const unsigned var = live_vars.var_from_reg(inst->dst);
  93             bool result_live = false;
  94
  95             for (unsigned i = 0; i < regs_written(inst); i++)
  96                result_live |= BITSET_TEST(live, var + i);
  97
  98             if (!result_live &&
  99                 (can_omit_write(inst) || can_eliminate(inst, flag_live))) {
 100                inst->dst = fs_reg(spread(retype(brw_null_reg(), inst->dst.type),
 101                                          inst->dst.stride));
 102                progress = true;
 103             }
 104          }
 105
 106          if (inst->dst.is_null() && can_eliminate(inst, flag_live)) {
 107             inst->opcode = BRW_OPCODE_NOP;
 108             progress = true;
 109          }
 110
 111          if (inst->dst.file == VGRF) {
 112             if (!inst->is_partial_write()) {
 113                const unsigned var = live_vars.var_from_reg(inst->dst);
 114                for (unsigned i = 0; i < regs_written(inst); i++) {
 115                   BITSET_CLEAR(live, var + i);
 116                }
 117             }
 118          }
 119
 120          if (!inst->predicate && inst->exec_size >= 8)
 121             flag_live[0] &= ~inst->flags_written();
 122
 123          if (inst->opcode == BRW_OPCODE_NOP) {
 124             inst->remove(block);
 125             continue;
 126          }
 127
 128          for (int i = 0; i < inst->sources; i++) {
 129             if (inst->src[i].file == VGRF) {
 130                int var = live_vars.var_from_reg(inst->src[i]);
 131
 132                for (unsigned j = 0; j < regs_read(inst, i); j++) {
 133                   BITSET_SET(live, var + j);
 134                }
 135             }
 136          }
 137
 138          flag_live[0] |= inst->flags_read(devinfo);
 139       }
 140    }
 141
 142    ralloc_free(live);
 143    ralloc_free(flag_live);
 144
 145    if (progress)
 146       invalidate_analysis(DEPENDENCY_INSTRUCTIONS);
 147
 148    return progress;
 149 }