src/broadcom/compiler/vir_opt_dead_code.c

   1 /*
   2  * Copyright © 2014 Broadcom
   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 /**
  25  * @file v3d_opt_dead_code.c
  26  *
  27  * This is a simple dead code eliminator for SSA values in VIR.
  28  *
  29  * It walks all the instructions finding what temps are used, then walks again
  30  * to remove instructions writing unused temps.
  31  *
  32  * This is an inefficient implementation if you have long chains of
  33  * instructions where the entire chain is dead, but we expect those to have
  34  * been eliminated at the NIR level, and here we're just cleaning up small
  35  * problems produced by NIR->VIR.
  36  */
  37
  38 #include "v3d_compiler.h"
  39
  40 static bool debug;
  41
  42 static void
  43 dce(struct v3d_compile *c, struct qinst *inst)
  44 {
  45         if (debug) {
  46                 fprintf(stderr, "Removing: ");
  47                 vir_dump_inst(c, inst);
  48                 fprintf(stderr, "\n");
  49         }
  50         assert(!v3d_qpu_writes_flags(&inst->qpu));
  51         vir_remove_instruction(c, inst);
  52 }
  53
  54 static bool
  55 has_nonremovable_reads(struct v3d_compile *c, struct qinst *inst)
  56 {
  57         for (int i = 0; i < vir_get_nsrc(inst); i++) {
  58                 if (inst->src[i].file == QFILE_VPM) {
  59                         /* Instance ID, Vertex ID: Should have been removed at
  60                          * the NIR level
  61                          */
  62                         if (inst->src[i].index == ~0)
  63                                 return true;
  64
  65                         uint32_t attr = inst->src[i].index / 4;
  66                         uint32_t offset = inst->src[i].index % 4;
  67
  68                         if (c->vattr_sizes[attr] != offset)
  69                                 return true;
  70
  71                         /* Can't get rid of the last VPM read, or the
  72                          * simulator (at least) throws an error.
  73                          */
  74                         uint32_t total_size = 0;
  75                         for (uint32_t i = 0; i < ARRAY_SIZE(c->vattr_sizes); i++)
  76                                 total_size += c->vattr_sizes[i];
  77                         if (total_size == 1)
  78                                 return true;
  79                 }
  80         }
  81
  82         return false;
  83 }
  84
  85 static bool
  86 can_write_to_null(struct v3d_compile *c, struct qinst *inst)
  87 {
  88         /* The SFU instructions must write to a physical register. */
  89         if (c->devinfo->ver >= 41 && v3d_qpu_uses_sfu(&inst->qpu))
  90                 return false;
  91
  92         return true;
  93 }
  94
  95 static void
  96 vir_dce_flags(struct v3d_compile *c, struct qinst *inst)
  97 {
  98         if (debug) {
  99                 fprintf(stderr,
 100                         "Removing flags write from: ");
 101                 vir_dump_inst(c, inst);
 102                 fprintf(stderr, "\n");
 103         }
 104
 105         assert(inst->qpu.type == V3D_QPU_INSTR_TYPE_ALU);
 106
 107         inst->qpu.flags.apf = V3D_QPU_PF_NONE;
 108         inst->qpu.flags.mpf = V3D_QPU_PF_NONE;
 109         inst->qpu.flags.auf = V3D_QPU_UF_NONE;
 110         inst->qpu.flags.muf = V3D_QPU_UF_NONE;
 111 }
 112
 113 bool
 114 vir_opt_dead_code(struct v3d_compile *c)
 115 {
 116         bool progress = false;
 117         bool *used = calloc(c->num_temps, sizeof(bool));
 118
 119         /* Defuse the "are you removing the cursor?" assertion in the core.
 120          * You'll need to set up a new cursor for any new instructions after
 121          * doing DCE (which we would expect, anyway).
 122          */
 123         c->cursor.link = NULL;
 124
 125         vir_for_each_inst_inorder(inst, c) {
 126                 for (int i = 0; i < vir_get_nsrc(inst); i++) {
 127                         if (inst->src[i].file == QFILE_TEMP)
 128                                 used[inst->src[i].index] = true;
 129                 }
 130         }
 131
 132         vir_for_each_block(block, c) {
 133                 struct qinst *last_flags_write = NULL;
 134
 135                 vir_for_each_inst_safe(inst, block) {
 136                         /* If this instruction reads the flags, we can't
 137                          * remove the flags generation for it.
 138                          */
 139                         if (v3d_qpu_reads_flags(&inst->qpu))
 140                                 last_flags_write = NULL;
 141
 142                         if (inst->dst.file != QFILE_NULL &&
 143                             !(inst->dst.file == QFILE_TEMP &&
 144                               !used[inst->dst.index])) {
 145                                 continue;
 146                         }
 147
 148                         if (vir_has_side_effects(c, inst))
 149                                 continue;
 150
 151                         if (v3d_qpu_writes_flags(&inst->qpu)) {
 152                                 /* If we obscure a previous flags write,
 153                                  * drop it.
 154                                  */
 155                                 if (last_flags_write &&
 156                                     (inst->qpu.flags.apf != V3D_QPU_PF_NONE ||
 157                                      inst->qpu.flags.mpf != V3D_QPU_PF_NONE)) {
 158                                         vir_dce_flags(c, last_flags_write);
 159                                         progress = true;
 160                                 }
 161
 162                                 last_flags_write = inst;
 163                         }
 164
 165                         if (v3d_qpu_writes_flags(&inst->qpu) ||
 166                             has_nonremovable_reads(c, inst)) {
 167                                 /* If we can't remove the instruction, but we
 168                                  * don't need its destination value, just
 169                                  * remove the destination.  The register
 170                                  * allocator would trivially color it and it
 171                                  * wouldn't cause any register pressure, but
 172                                  * it's nicer to read the VIR code without
 173                                  * unused destination regs.
 174                                  */
 175                                 if (inst->dst.file == QFILE_TEMP &&
 176                                     can_write_to_null(c, inst)) {
 177                                         if (debug) {
 178                                                 fprintf(stderr,
 179                                                         "Removing dst from: ");
 180                                                 vir_dump_inst(c, inst);
 181                                                 fprintf(stderr, "\n");
 182                                         }
 183                                         c->defs[inst->dst.index] = NULL;
 184                                         inst->dst.file = QFILE_NULL;
 185                                         progress = true;
 186                                 }
 187                                 continue;
 188                         }
 189
 190                         for (int i = 0; i < vir_get_nsrc(inst); i++) {
 191                                 if (inst->src[i].file != QFILE_VPM)
 192                                         continue;
 193                                 uint32_t attr = inst->src[i].index / 4;
 194                                 uint32_t offset = (inst->src[i].index % 4);
 195
 196                                 if (c->vattr_sizes[attr] == offset) {
 197                                         c->vattr_sizes[attr]--;
 198                                 }
 199                         }
 200
 201                         assert(inst != last_flags_write);
 202                         dce(c, inst);
 203                         progress = true;
 204                         continue;
 205                 }
 206         }
 207
 208         free(used);
 209
 210         return progress;
 211 }