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

   1 /*
   2  * Copyright © 2013 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_sel_peephole.cpp
  28  *
  29  * This file contains the opt_peephole_sel() optimization pass that replaces
  30  * MOV instructions to the same destination in the "then" and "else" bodies of
  31  * an if statement with SEL instructions.
  32  */
  33
  34 /* Four MOVs seems to be pretty typical, so I picked the next power of two in
  35  * the hopes that it would handle almost anything possible in a single
  36  * pass.
  37  */
  38 #define MAX_MOVS 8 /**< The maximum number of MOVs to attempt to match. */
  39
  40 using namespace brw;
  41
  42 /**
  43  * Scans forwards from an IF counting consecutive MOV instructions in the
  44  * "then" and "else" blocks of the if statement.
  45  *
  46  * A pointer to the bblock_t following the IF is passed as the <then_block>
  47  * argument. The function stores pointers to the MOV instructions in the
  48  * <then_mov> and <else_mov> arrays.
  49  *
  50  * \return the minimum number of MOVs found in the two branches or zero if
  51  *         an error occurred.
  52  *
  53  * E.g.:
  54  *                  IF ...
  55  *    then_mov[0] = MOV g4, ...
  56  *    then_mov[1] = MOV g5, ...
  57  *    then_mov[2] = MOV g6, ...
  58  *                  ELSE ...
  59  *    else_mov[0] = MOV g4, ...
  60  *    else_mov[1] = MOV g5, ...
  61  *    else_mov[2] = MOV g7, ...
  62  *                  ENDIF
  63  *    returns 3.
  64  */
  65 static int
  66 count_movs_from_if(fs_inst *then_mov[MAX_MOVS], fs_inst *else_mov[MAX_MOVS],
  67                    bblock_t *then_block, bblock_t *else_block)
  68 {
  69    int then_movs = 0;
  70    foreach_inst_in_block(fs_inst, inst, then_block) {
  71       if (then_movs == MAX_MOVS || inst->opcode != BRW_OPCODE_MOV)
  72          break;
  73
  74       then_mov[then_movs] = inst;
  75       then_movs++;
  76    }
  77
  78    int else_movs = 0;
  79    foreach_inst_in_block(fs_inst, inst, else_block) {
  80       if (else_movs == MAX_MOVS || inst->opcode != BRW_OPCODE_MOV)
  81          break;
  82
  83       else_mov[else_movs] = inst;
  84       else_movs++;
  85    }
  86
  87    return MIN2(then_movs, else_movs);
  88 }
  89
  90 /**
  91  * Try to replace IF/MOV+/ELSE/MOV+/ENDIF with SEL.
  92  *
  93  * Many GLSL shaders contain the following pattern:
  94  *
  95  *    x = condition ? foo : bar
  96  *
  97  * or
  98  *
  99  *    if (...) a.xyzw = foo.xyzw;
 100  *    else     a.xyzw = bar.xyzw;
 101  *
 102  * The compiler emits an ir_if tree for this, since each subexpression might be
 103  * a complex tree that could have side-effects or short-circuit logic.
 104  *
 105  * However, the common case is to simply select one of two constants or
 106  * variable values---which is exactly what SEL is for.  In this case, the
 107  * assembly looks like:
 108  *
 109  *    (+f0) IF
 110  *    MOV dst src0
 111  *    ...
 112  *    ELSE
 113  *    MOV dst src1
 114  *    ...
 115  *    ENDIF
 116  *
 117  * where each pair of MOVs to a common destination and can be easily translated
 118  * into
 119  *
 120  *    (+f0) SEL dst src0 src1
 121  *
 122  * If src0 is an immediate value, we promote it to a temporary GRF.
 123  */
 124 bool
 125 fs_visitor::opt_peephole_sel()
 126 {
 127    bool progress = false;
 128
 129    foreach_block (block, cfg) {
 130       /* IF instructions, by definition, can only be found at the ends of
 131        * basic blocks.
 132        */
 133       fs_inst *if_inst = (fs_inst *)block->end();
 134       if (if_inst->opcode != BRW_OPCODE_IF)
 135          continue;
 136
 137       fs_inst *else_mov[MAX_MOVS] = { NULL };
 138       fs_inst *then_mov[MAX_MOVS] = { NULL };
 139
 140       bblock_t *then_block = block->next();
 141       bblock_t *else_block = NULL;
 142       foreach_list_typed(bblock_link, child, link, &block->children) {
 143          if (child->block != then_block) {
 144             if (child->block->prev()->end()->opcode == BRW_OPCODE_ELSE) {
 145                else_block = child->block;
 146             }
 147             break;
 148          }
 149       }
 150       if (else_block == NULL)
 151          continue;
 152
 153       int movs = count_movs_from_if(then_mov, else_mov, then_block, else_block);
 154
 155       if (movs == 0)
 156          continue;
 157
 158       /* Generate SEL instructions for pairs of MOVs to a common destination. */
 159       for (int i = 0; i < movs; i++) {
 160          if (!then_mov[i] || !else_mov[i])
 161             break;
 162
 163          /* Check that the MOVs are the right form. */
 164          if (!then_mov[i]->dst.equals(else_mov[i]->dst) ||
 165              then_mov[i]->exec_size != else_mov[i]->exec_size ||
 166              then_mov[i]->force_sechalf != else_mov[i]->force_sechalf ||
 167              then_mov[i]->force_writemask_all != else_mov[i]->force_writemask_all ||
 168              then_mov[i]->is_partial_write() ||
 169              else_mov[i]->is_partial_write() ||
 170              then_mov[i]->conditional_mod != BRW_CONDITIONAL_NONE ||
 171              else_mov[i]->conditional_mod != BRW_CONDITIONAL_NONE) {
 172             movs = i;
 173             break;
 174          }
 175
 176          /* Check that source types for mov operations match. */
 177          if (then_mov[i]->src[0].type != else_mov[i]->src[0].type) {
 178             movs = i;
 179             break;
 180          }
 181       }
 182
 183       if (movs == 0)
 184          continue;
 185
 186       for (int i = 0; i < movs; i++) {
 187          const fs_builder ibld = fs_builder(this, then_block, then_mov[i])
 188                                  .at(block, if_inst);
 189
 190          if (then_mov[i]->src[0].equals(else_mov[i]->src[0])) {
 191             ibld.MOV(then_mov[i]->dst, then_mov[i]->src[0]);
 192          } else {
 193             /* Only the last source register can be a constant, so if the MOV
 194              * in the "then" clause uses a constant, we need to put it in a
 195              * temporary.
 196              */
 197             fs_reg src0(then_mov[i]->src[0]);
 198             if (src0.file == IMM) {
 199                src0 = vgrf(glsl_type::float_type);
 200                src0.type = then_mov[i]->src[0].type;
 201                ibld.MOV(src0, then_mov[i]->src[0]);
 202             }
 203
 204             set_predicate_inv(if_inst->predicate, if_inst->predicate_inverse,
 205                               ibld.SEL(then_mov[i]->dst, src0,
 206                                        else_mov[i]->src[0]));
 207          }
 208
 209          then_mov[i]->remove(then_block);
 210          else_mov[i]->remove(else_block);
 211       }
 212
 213       progress = true;
 214    }
 215
 216    if (progress)
 217       invalidate_live_intervals();
 218
 219    return progress;
 220 }