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 /**
  41  * Scans forwards from an IF counting consecutive MOV instructions in the
  42  * "then" and "else" blocks of the if statement.
  43  *
  44  * A pointer to the fs_inst* for IF is passed as the <if_inst> argument. The
  45  * function stores pointers to the MOV instructions in the <then_mov> and
  46  * <else_mov> arrays.
  47  *
  48  * \return the minimum number of MOVs found in the two branches or zero if
  49  *         an error occurred.
  50  *
  51  * E.g.:
  52  *                  IF ...
  53  *    then_mov[0] = MOV g4, ...
  54  *    then_mov[1] = MOV g5, ...
  55  *    then_mov[2] = MOV g6, ...
  56  *                  ELSE ...
  57  *    else_mov[0] = MOV g4, ...
  58  *    else_mov[1] = MOV g5, ...
  59  *    else_mov[2] = MOV g7, ...
  60  *                  ENDIF
  61  *    returns 3.
  62  */
  63 static int
  64 count_movs_from_if(fs_inst *then_mov[MAX_MOVS], fs_inst *else_mov[MAX_MOVS],
  65                    fs_inst *if_inst, fs_inst *else_inst)
  66 {
  67    fs_inst *m = if_inst;
  68
  69    assert(m->opcode == BRW_OPCODE_IF);
  70    m = (fs_inst *) m->next;
  71
  72    int then_movs = 0;
  73    while (then_movs < MAX_MOVS && m->opcode == BRW_OPCODE_MOV) {
  74       then_mov[then_movs] = m;
  75       m = (fs_inst *) m->next;
  76       then_movs++;
  77    }
  78
  79    m = (fs_inst *) else_inst->next;
  80
  81    int else_movs = 0;
  82    while (else_movs < MAX_MOVS && m->opcode == BRW_OPCODE_MOV) {
  83       else_mov[else_movs] = m;
  84       m = (fs_inst *) m->next;
  85       else_movs++;
  86    }
  87
  88    return MIN2(then_movs, else_movs);
  89 }
  90
  91 /**
  92  * Try to replace IF/MOV+/ELSE/MOV+/ENDIF with SEL.
  93  *
  94  * Many GLSL shaders contain the following pattern:
  95  *
  96  *    x = condition ? foo : bar
  97  *
  98  * or
  99  *
 100  *    if (...) a.xyzw = foo.xyzw;
 101  *    else     a.xyzw = bar.xyzw;
 102  *
 103  * The compiler emits an ir_if tree for this, since each subexpression might be
 104  * a complex tree that could have side-effects or short-circuit logic.
 105  *
 106  * However, the common case is to simply select one of two constants or
 107  * variable values---which is exactly what SEL is for.  In this case, the
 108  * assembly looks like:
 109  *
 110  *    (+f0) IF
 111  *    MOV dst src0
 112  *    ...
 113  *    ELSE
 114  *    MOV dst src1
 115  *    ...
 116  *    ENDIF
 117  *
 118  * where each pair of MOVs to a common destination and can be easily translated
 119  * into
 120  *
 121  *    (+f0) SEL dst src0 src1
 122  *
 123  * If src0 is an immediate value, we promote it to a temporary GRF.
 124  */
 125 bool
 126 fs_visitor::opt_peephole_sel()
 127 {
 128    bool progress = false;
 129
 130    cfg_t cfg(&instructions);
 131
 132    for (int b = 0; b < cfg.num_blocks; b++) {
 133       bblock_t *block = cfg.blocks[b];
 134
 135       /* IF instructions, by definition, can only be found at the ends of
 136        * basic blocks.
 137        */
 138       fs_inst *if_inst = (fs_inst *) block->end;
 139       if (if_inst->opcode != BRW_OPCODE_IF)
 140          continue;
 141
 142       if (!block->else_inst)
 143          continue;
 144
 145       fs_inst *else_inst = (fs_inst *) block->else_inst;
 146       assert(else_inst->opcode == BRW_OPCODE_ELSE);
 147
 148       fs_inst *else_mov[MAX_MOVS] = { NULL };
 149       fs_inst *then_mov[MAX_MOVS] = { NULL };
 150
 151       int movs = count_movs_from_if(then_mov, else_mov, if_inst, else_inst);
 152
 153       if (movs == 0)
 154          continue;
 155
 156       fs_inst *sel_inst[MAX_MOVS] = { NULL };
 157       fs_inst *mov_imm_inst[MAX_MOVS] = { NULL };
 158
 159       /* Generate SEL instructions for pairs of MOVs to a common destination. */
 160       for (int i = 0; i < movs; i++) {
 161          if (!then_mov[i] || !else_mov[i])
 162             break;
 163
 164          /* Check that the MOVs are the right form. */
 165          if (!then_mov[i]->dst.equals(else_mov[i]->dst) ||
 166              then_mov[i]->is_partial_write() ||
 167              else_mov[i]->is_partial_write()) {
 168             movs = i;
 169             break;
 170          }
 171
 172          /* Check that source types for mov operations match. */
 173          if (then_mov[i]->src[0].type != else_mov[i]->src[0].type) {
 174             movs = i;
 175             break;
 176          }
 177
 178          if (!then_mov[i]->src[0].equals(else_mov[i]->src[0])) {
 179             /* Only the last source register can be a constant, so if the MOV
 180              * in the "then" clause uses a constant, we need to put it in a
 181              * temporary.
 182              */
 183             fs_reg src0(then_mov[i]->src[0]);
 184             if (src0.file == IMM) {
 185                src0 = fs_reg(this, glsl_type::float_type);
 186                src0.type = then_mov[i]->src[0].type;
 187                mov_imm_inst[i] = MOV(src0, then_mov[i]->src[0]);
 188             }
 189
 190             sel_inst[i] = SEL(then_mov[i]->dst, src0, else_mov[i]->src[0]);
 191
 192             if (brw->gen == 6 && if_inst->conditional_mod) {
 193                /* For Sandybridge with IF with embedded comparison */
 194                sel_inst[i]->predicate = BRW_PREDICATE_NORMAL;
 195             } else {
 196                /* Separate CMP and IF instructions */
 197                sel_inst[i]->predicate = if_inst->predicate;
 198                sel_inst[i]->predicate_inverse = if_inst->predicate_inverse;
 199             }
 200          } else {
 201             sel_inst[i] = MOV(then_mov[i]->dst, then_mov[i]->src[0]);
 202          }
 203       }
 204
 205       if (movs == 0)
 206          continue;
 207
 208       /* Emit a CMP if our IF used the embedded comparison */
 209       if (brw->gen == 6 && if_inst->conditional_mod) {
 210          fs_inst *cmp_inst = CMP(reg_null_d, if_inst->src[0], if_inst->src[1],
 211                                  if_inst->conditional_mod);
 212          if_inst->insert_before(cmp_inst);
 213       }
 214
 215       for (int i = 0; i < movs; i++) {
 216          if (mov_imm_inst[i])
 217             if_inst->insert_before(mov_imm_inst[i]);
 218          if_inst->insert_before(sel_inst[i]);
 219
 220          then_mov[i]->remove();
 221          else_mov[i]->remove();
 222       }
 223
 224       progress = true;
 225    }
 226
 227    if (progress)
 228       invalidate_live_intervals();
 229
 230    return progress;
 231 }