src/intel/compiler/brw_fs_cmod_propagation.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_cfg.h"
  26 #include "brw_eu.h"
  27
  28 /** @file brw_fs_cmod_propagation.cpp
  29  *
  30  * Implements a pass that propagates the conditional modifier from a CMP x 0.0
  31  * instruction into the instruction that generated x. For instance, in this
  32  * sequence
  33  *
  34  *    add(8)          g70<1>F    g69<8,8,1>F    4096F
  35  *    cmp.ge.f0(8)    null       g70<8,8,1>F    0F
  36  *
  37  * we can do the comparison as part of the ADD instruction directly:
  38  *
  39  *    add.ge.f0(8)    g70<1>F    g69<8,8,1>F    4096F
  40  *
  41  * If there had been a use of the flag register and another CMP using g70
  42  *
  43  *    add.ge.f0(8)    g70<1>F    g69<8,8,1>F    4096F
  44  *    (+f0) sel(8)    g71<F>     g72<8,8,1>F    g73<8,8,1>F
  45  *    cmp.ge.f0(8)    null       g70<8,8,1>F    0F
  46  *
  47  * we can recognize that the CMP is generating the flag value that already
  48  * exists and therefore remove the instruction.
  49  */
  50
  51 static bool
  52 opt_cmod_propagation_local(const gen_device_info *devinfo, bblock_t *block)
  53 {
  54    bool progress = false;
  55    int ip = block->end_ip + 1;
  56
  57    foreach_inst_in_block_reverse_safe(fs_inst, inst, block) {
  58       ip--;
  59
  60       if ((inst->opcode != BRW_OPCODE_AND &&
  61            inst->opcode != BRW_OPCODE_CMP &&
  62            inst->opcode != BRW_OPCODE_MOV) ||
  63           inst->predicate != BRW_PREDICATE_NONE ||
  64           !inst->dst.is_null() ||
  65           (inst->src[0].file != VGRF && inst->src[0].file != ATTR &&
  66            inst->src[0].file != UNIFORM))
  67          continue;
  68
  69       /* An ABS source modifier can only be handled when processing a compare
  70        * with a value other than zero.
  71        */
  72       if (inst->src[0].abs &&
  73           (inst->opcode != BRW_OPCODE_CMP || inst->src[1].is_zero()))
  74          continue;
  75
  76       /* Only an AND.NZ can be propagated.  Many AND.Z instructions are
  77        * generated (for ir_unop_not in fs_visitor::emit_bool_to_cond_code).
  78        * Propagating those would require inverting the condition on the CMP.
  79        * This changes both the flag value and the register destination of the
  80        * CMP.  That result may be used elsewhere, so we can't change its value
  81        * on a whim.
  82        */
  83       if (inst->opcode == BRW_OPCODE_AND &&
  84           !(inst->src[1].is_one() &&
  85             inst->conditional_mod == BRW_CONDITIONAL_NZ &&
  86             !inst->src[0].negate))
  87          continue;
  88
  89       if (inst->opcode == BRW_OPCODE_MOV &&
  90           inst->conditional_mod != BRW_CONDITIONAL_NZ)
  91          continue;
  92
  93       bool read_flag = false;
  94       foreach_inst_in_block_reverse_starting_from(fs_inst, scan_inst, inst) {
  95          /* A CMP with a second source of zero can match with anything.  A CMP
  96           * with a second source that is not zero can only match with an ADD
  97           * instruction.
  98           */
  99          if (inst->opcode == BRW_OPCODE_CMP && !inst->src[1].is_zero()) {
 100             bool negate;
 101
 102             if (scan_inst->opcode != BRW_OPCODE_ADD)
 103                goto not_match;
 104
 105             /* A CMP is basically a subtraction.  The result of the
 106              * subtraction must be the same as the result of the addition.
 107              * This means that one of the operands must be negated.  So (a +
 108              * b) vs (a == -b) or (a + -b) vs (a == b).
 109              */
 110             if ((inst->src[0].equals(scan_inst->src[0]) &&
 111                  inst->src[1].negative_equals(scan_inst->src[1])) ||
 112                 (inst->src[0].equals(scan_inst->src[1]) &&
 113                  inst->src[1].negative_equals(scan_inst->src[0]))) {
 114                negate = false;
 115             } else if ((inst->src[0].negative_equals(scan_inst->src[0]) &&
 116                         inst->src[1].equals(scan_inst->src[1])) ||
 117                        (inst->src[0].negative_equals(scan_inst->src[1]) &&
 118                         inst->src[1].equals(scan_inst->src[0]))) {
 119                negate = true;
 120             } else {
 121                goto not_match;
 122             }
 123
 124             if (scan_inst->is_partial_write() ||
 125                 scan_inst->exec_size != inst->exec_size)
 126                goto not_match;
 127
 128             /* From the Sky Lake PRM Vol. 7 "Assigning Conditional Mods":
 129              *
 130              *    * Note that the [post condition signal] bits generated at
 131              *      the output of a compute are before the .sat.
 132              *
 133              * So we don't have to bail if scan_inst has saturate.
 134              */
 135
 136             /* Otherwise, try propagating the conditional. */
 137             const enum brw_conditional_mod cond =
 138                negate ? brw_swap_cmod(inst->conditional_mod)
 139                       : inst->conditional_mod;
 140
 141             if (scan_inst->can_do_cmod() &&
 142                 ((!read_flag && scan_inst->conditional_mod == BRW_CONDITIONAL_NONE) ||
 143                  scan_inst->conditional_mod == cond)) {
 144                scan_inst->conditional_mod = cond;
 145                inst->remove(block);
 146                progress = true;
 147             }
 148             break;
 149          }
 150
 151          if (regions_overlap(scan_inst->dst, scan_inst->size_written,
 152                              inst->src[0], inst->size_read(0))) {
 153             if (scan_inst->is_partial_write() ||
 154                 scan_inst->dst.offset != inst->src[0].offset ||
 155                 scan_inst->exec_size != inst->exec_size)
 156                break;
 157
 158             /* CMP's result is the same regardless of dest type. */
 159             if (inst->conditional_mod == BRW_CONDITIONAL_NZ &&
 160                 scan_inst->opcode == BRW_OPCODE_CMP &&
 161                 (inst->dst.type == BRW_REGISTER_TYPE_D ||
 162                  inst->dst.type == BRW_REGISTER_TYPE_UD)) {
 163                inst->remove(block);
 164                progress = true;
 165                break;
 166             }
 167
 168             /* If the AND wasn't handled by the previous case, it isn't safe
 169              * to remove it.
 170              */
 171             if (inst->opcode == BRW_OPCODE_AND)
 172                break;
 173
 174             /* Comparisons operate differently for ints and floats */
 175             if (scan_inst->dst.type != inst->dst.type &&
 176                 (scan_inst->dst.type == BRW_REGISTER_TYPE_F ||
 177                  inst->dst.type == BRW_REGISTER_TYPE_F))
 178                break;
 179
 180             /* If the instruction generating inst's source also wrote the
 181              * flag, and inst is doing a simple .nz comparison, then inst
 182              * is redundant - the appropriate value is already in the flag
 183              * register.  Delete inst.
 184              */
 185             if (inst->conditional_mod == BRW_CONDITIONAL_NZ &&
 186                 !inst->src[0].negate &&
 187                 scan_inst->flags_written()) {
 188                inst->remove(block);
 189                progress = true;
 190                break;
 191             }
 192
 193             /* The conditional mod of the CMP/CMPN instructions behaves
 194              * specially because the flag output is not calculated from the
 195              * result of the instruction, but the other way around, which
 196              * means that even if the condmod to propagate and the condmod
 197              * from the CMP instruction are the same they will in general give
 198              * different results because they are evaluated based on different
 199              * inputs.
 200              */
 201             if (scan_inst->opcode == BRW_OPCODE_CMP ||
 202                 scan_inst->opcode == BRW_OPCODE_CMPN)
 203                break;
 204
 205             /* From the Sky Lake PRM Vol. 7 "Assigning Conditional Mods":
 206              *
 207              *    * Note that the [post condition signal] bits generated at
 208              *      the output of a compute are before the .sat.
 209              */
 210             if (scan_inst->saturate)
 211                break;
 212
 213             /* From the Sky Lake PRM, Vol 2a, "Multiply":
 214              *
 215              *    "When multiplying integer data types, if one of the sources
 216              *     is a DW, the resulting full precision data is stored in
 217              *     the accumulator. However, if the destination data type is
 218              *     either W or DW, the low bits of the result are written to
 219              *     the destination register and the remaining high bits are
 220              *     discarded. This results in undefined Overflow and Sign
 221              *     flags. Therefore, conditional modifiers and saturation
 222              *     (.sat) cannot be used in this case."
 223              *
 224              * We just disallow cmod propagation on all integer multiplies.
 225              */
 226             if (!brw_reg_type_is_floating_point(scan_inst->dst.type) &&
 227                 scan_inst->opcode == BRW_OPCODE_MUL)
 228                break;
 229
 230             /* Otherwise, try propagating the conditional. */
 231             enum brw_conditional_mod cond =
 232                inst->src[0].negate ? brw_swap_cmod(inst->conditional_mod)
 233                                    : inst->conditional_mod;
 234
 235             if (scan_inst->can_do_cmod() &&
 236                 ((!read_flag && scan_inst->conditional_mod == BRW_CONDITIONAL_NONE) ||
 237                  scan_inst->conditional_mod == cond)) {
 238                scan_inst->conditional_mod = cond;
 239                inst->remove(block);
 240                progress = true;
 241             }
 242             break;
 243          }
 244
 245       not_match:
 246          if (scan_inst->flags_written())
 247             break;
 248
 249          read_flag = read_flag || scan_inst->flags_read(devinfo);
 250       }
 251    }
 252
 253    return progress;
 254 }
 255
 256 bool
 257 fs_visitor::opt_cmod_propagation()
 258 {
 259    bool progress = false;
 260
 261    foreach_block_reverse(block, cfg) {
 262       progress = opt_cmod_propagation_local(devinfo, block) || progress;
 263    }
 264
 265    if (progress)
 266       invalidate_live_intervals();
 267
 268    return progress;
 269 }