src/mesa/drivers/dri/i965/brw_fs_peephole_predicated_break.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_peephole_predicated_break.cpp
  28  *
  29  * Loops are often structured as
  30  *
  31  * loop:
  32  *    CMP.f0
  33  *    (+f0) IF
  34  *    BREAK
  35  *    ENDIF
  36  *    ...
  37  *    WHILE loop
  38  *
  39  * This peephole pass removes the IF and ENDIF instructions and predicates the
  40  * BREAK, dropping two instructions from the loop body.
  41  *
  42  * If the loop was a DO { ... } WHILE loop, it looks like
  43  *
  44  * loop:
  45  *    ...
  46  *    CMP.f0
  47  *    (+f0) IF
  48  *    BREAK
  49  *    ENDIF
  50  *    WHILE loop
  51  *
  52  * and we can remove the BREAK instruction and predicate the WHILE.
  53  */
  54
  55 bool
  56 fs_visitor::opt_peephole_predicated_break()
  57 {
  58    bool progress = false;
  59
  60    foreach_block (block, cfg) {
  61       if (block->start_ip != block->end_ip)
  62          continue;
  63
  64       /* BREAK and CONTINUE instructions, by definition, can only be found at
  65        * the ends of basic blocks.
  66        */
  67       fs_inst *jump_inst = (fs_inst *)block->end();
  68       if (jump_inst->opcode != BRW_OPCODE_BREAK &&
  69           jump_inst->opcode != BRW_OPCODE_CONTINUE)
  70          continue;
  71
  72       fs_inst *if_inst = (fs_inst *)block->prev()->end();
  73       if (if_inst->opcode != BRW_OPCODE_IF)
  74          continue;
  75
  76       fs_inst *endif_inst = (fs_inst *)block->next()->start();
  77       if (endif_inst->opcode != BRW_OPCODE_ENDIF)
  78          continue;
  79
  80       bblock_t *jump_block = block;
  81       bblock_t *if_block = jump_block->prev();
  82       bblock_t *endif_block = jump_block->next();
  83
  84       /* For Sandybridge with IF with embedded comparison we need to emit an
  85        * instruction to set the flag register.
  86        */
  87       if (devinfo->gen == 6 && if_inst->conditional_mod) {
  88          bld.at(if_block, if_inst)
  89             .CMP(bld.null_reg_d(), if_inst->src[0], if_inst->src[1],
  90                  if_inst->conditional_mod);
  91          jump_inst->predicate = BRW_PREDICATE_NORMAL;
  92       } else {
  93          jump_inst->predicate = if_inst->predicate;
  94          jump_inst->predicate_inverse = if_inst->predicate_inverse;
  95       }
  96
  97       bblock_t *earlier_block = if_block;
  98       if (if_block->start_ip == if_block->end_ip) {
  99          earlier_block = if_block->prev();
 100       }
 101
 102       if_inst->remove(if_block);
 103
 104       bblock_t *later_block = endif_block;
 105       if (endif_block->start_ip == endif_block->end_ip) {
 106          later_block = endif_block->next();
 107       }
 108       endif_inst->remove(endif_block);
 109
 110       if (!earlier_block->ends_with_control_flow()) {
 111          earlier_block->children.make_empty();
 112          earlier_block->add_successor(cfg->mem_ctx, jump_block);
 113       }
 114
 115       if (!later_block->starts_with_control_flow()) {
 116          later_block->parents.make_empty();
 117       }
 118       jump_block->add_successor(cfg->mem_ctx, later_block);
 119
 120       if (earlier_block->can_combine_with(jump_block)) {
 121          earlier_block->combine_with(jump_block);
 122
 123          block = earlier_block;
 124       }
 125
 126       /* Now look at the first instruction of the block following the BREAK. If
 127        * it's a WHILE, we can delete the break, predicate the WHILE, and join
 128        * the two basic blocks.
 129        */
 130       bblock_t *while_block = earlier_block->next();
 131       fs_inst *while_inst = (fs_inst *)while_block->start();
 132
 133       if (jump_inst->opcode == BRW_OPCODE_BREAK &&
 134           while_inst->opcode == BRW_OPCODE_WHILE &&
 135           while_inst->predicate == BRW_PREDICATE_NONE) {
 136          jump_inst->remove(earlier_block);
 137          while_inst->predicate = jump_inst->predicate;
 138          while_inst->predicate_inverse = !jump_inst->predicate_inverse;
 139
 140          earlier_block->children.make_empty();
 141          earlier_block->add_successor(cfg->mem_ctx, while_block);
 142
 143          assert(earlier_block->can_combine_with(while_block));
 144          earlier_block->combine_with(while_block);
 145
 146          earlier_block->next()->parents.make_empty();
 147          earlier_block->add_successor(cfg->mem_ctx, earlier_block->next());
 148       }
 149
 150       progress = true;
 151    }
 152
 153    if (progress)
 154       invalidate_live_intervals();
 155
 156    return progress;
 157 }