src/gallium/drivers/vc4/vc4_opt_algebraic.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 vc4_opt_algebraic.c
  26  *
  27  * This is the optimization pass for miscellaneous changes to instructions
  28  * where we can simplify the operation by some knowledge about the specific
  29  * operations.
  30  *
  31  * Mostly this will be a matter of turning things into MOVs so that they can
  32  * later be copy-propagated out.
  33  */
  34
  35 #include "vc4_qir.h"
  36 #include "util/u_math.h"
  37
  38 static bool debug;
  39
  40 static void
  41 dump_from(struct vc4_compile *c, struct qinst *inst)
  42 {
  43         if (!debug)
  44                 return;
  45
  46         fprintf(stderr, "optimizing: ");
  47         qir_dump_inst(c, inst);
  48         fprintf(stderr, "\n");
  49 }
  50
  51 static void
  52 dump_to(struct vc4_compile *c, struct qinst *inst)
  53 {
  54         if (!debug)
  55                 return;
  56
  57         fprintf(stderr, "to: ");
  58         qir_dump_inst(c, inst);
  59         fprintf(stderr, "\n");
  60 }
  61
  62 static bool
  63 is_constant_value(struct vc4_compile *c, struct qreg reg,
  64                   uint32_t val)
  65 {
  66         if (reg.file == QFILE_UNIF &&
  67             !reg.pack &&
  68             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT &&
  69             c->uniform_data[reg.index] == val) {
  70                 return true;
  71         }
  72
  73         if (reg.file == QFILE_SMALL_IMM && reg.index == val)
  74                 return true;
  75
  76         return false;
  77 }
  78
  79 static bool
  80 is_zero(struct vc4_compile *c, struct qreg reg)
  81 {
  82         reg = qir_follow_movs(c, reg);
  83         return is_constant_value(c, reg, 0);
  84 }
  85
  86 static bool
  87 is_1f(struct vc4_compile *c, struct qreg reg)
  88 {
  89         reg = qir_follow_movs(c, reg);
  90         return is_constant_value(c, reg, fui(1.0));
  91 }
  92
  93 static void
  94 replace_with_mov(struct vc4_compile *c, struct qinst *inst, struct qreg arg)
  95 {
  96         dump_from(c, inst);
  97         if (qir_is_mul(inst))
  98                 inst->op = QOP_MMOV;
  99         else if (qir_is_float_input(inst))
 100                 inst->op = QOP_FMOV;
 101         else
 102                 inst->op = QOP_MOV;
 103         inst->src[0] = arg;
 104         inst->src[1] = c->undef;
 105         dump_to(c, inst);
 106 }
 107
 108 static bool
 109 replace_x_0_with_x(struct vc4_compile *c,
 110                  struct qinst *inst,
 111                  int arg)
 112 {
 113         if (!is_zero(c, inst->src[arg]))
 114                 return false;
 115         replace_with_mov(c, inst, inst->src[1 - arg]);
 116         return true;
 117 }
 118
 119 static bool
 120 replace_x_0_with_0(struct vc4_compile *c,
 121                   struct qinst *inst,
 122                   int arg)
 123 {
 124         if (!is_zero(c, inst->src[arg]))
 125                 return false;
 126         replace_with_mov(c, inst, inst->src[arg]);
 127         return true;
 128 }
 129
 130 static bool
 131 fmul_replace_one(struct vc4_compile *c,
 132                  struct qinst *inst,
 133                  int arg)
 134 {
 135         if (!is_1f(c, inst->src[arg]))
 136                 return false;
 137         replace_with_mov(c, inst, inst->src[1 - arg]);
 138         return true;
 139 }
 140
 141 bool
 142 qir_opt_algebraic(struct vc4_compile *c)
 143 {
 144         bool progress = false;
 145
 146         list_for_each_entry(struct qinst, inst, &c->instructions, link) {
 147                 switch (inst->op) {
 148                 case QOP_FMIN:
 149                         if (is_1f(c, inst->src[1]) &&
 150                             inst->src[0].pack >= QPU_UNPACK_8D_REP &&
 151                             inst->src[0].pack <= QPU_UNPACK_8D) {
 152                                 replace_with_mov(c, inst, inst->src[0]);
 153                                 progress = true;
 154                         }
 155                         break;
 156
 157                 case QOP_FMAX:
 158                         if (is_zero(c, inst->src[1]) &&
 159                             inst->src[0].pack >= QPU_UNPACK_8D_REP &&
 160                             inst->src[0].pack <= QPU_UNPACK_8D) {
 161                                 replace_with_mov(c, inst, inst->src[0]);
 162                                 progress = true;
 163                         }
 164                         break;
 165
 166                 case QOP_FSUB:
 167                 case QOP_SUB:
 168                         if (is_zero(c, inst->src[1])) {
 169                                 replace_with_mov(c, inst, inst->src[0]);
 170                                 progress = true;
 171                         }
 172                         break;
 173
 174                 case QOP_ADD:
 175                         if (replace_x_0_with_x(c, inst, 0) ||
 176                             replace_x_0_with_x(c, inst, 1)) {
 177                                 progress = true;
 178                                 break;
 179                         }
 180                         break;
 181
 182                 case QOP_FADD:
 183                         if (replace_x_0_with_x(c, inst, 0) ||
 184                             replace_x_0_with_x(c, inst, 1)) {
 185                                 progress = true;
 186                                 break;
 187                         }
 188
 189                         /* FADD(a, FSUB(0, b)) -> FSUB(a, b) */
 190                         if (inst->src[1].file == QFILE_TEMP &&
 191                             c->defs[inst->src[1].index] &&
 192                             c->defs[inst->src[1].index]->op == QOP_FSUB) {
 193                                 struct qinst *fsub = c->defs[inst->src[1].index];
 194                                 if (is_zero(c, fsub->src[0])) {
 195                                         dump_from(c, inst);
 196                                         inst->op = QOP_FSUB;
 197                                         inst->src[1] = fsub->src[1];
 198                                         progress = true;
 199                                         dump_to(c, inst);
 200                                         break;
 201                                 }
 202                         }
 203
 204                         /* FADD(FSUB(0, b), a) -> FSUB(a, b) */
 205                         if (inst->src[0].file == QFILE_TEMP &&
 206                             c->defs[inst->src[0].index] &&
 207                             c->defs[inst->src[0].index]->op == QOP_FSUB) {
 208                                 struct qinst *fsub = c->defs[inst->src[0].index];
 209                                 if (is_zero(c, fsub->src[0])) {
 210                                         dump_from(c, inst);
 211                                         inst->op = QOP_FSUB;
 212                                         inst->src[0] = inst->src[1];
 213                                         inst->src[1] = fsub->src[1];
 214                                         dump_to(c, inst);
 215                                         progress = true;
 216                                         break;
 217                                 }
 218                         }
 219                         break;
 220
 221                 case QOP_FMUL:
 222                         if (!inst->dst.pack &&
 223                             (replace_x_0_with_0(c, inst, 0) ||
 224                              replace_x_0_with_0(c, inst, 1) ||
 225                              fmul_replace_one(c, inst, 0) ||
 226                              fmul_replace_one(c, inst, 1))) {
 227                                 progress = true;
 228                                 break;
 229                         }
 230                         break;
 231
 232                 case QOP_MUL24:
 233                         if (!inst->dst.pack &&
 234                             (replace_x_0_with_0(c, inst, 0) ||
 235                              replace_x_0_with_0(c, inst, 1))) {
 236                                 progress = true;
 237                                 break;
 238                         }
 239                         break;
 240
 241                 case QOP_AND:
 242                         if (replace_x_0_with_0(c, inst, 0) ||
 243                             replace_x_0_with_0(c, inst, 1)) {
 244                                 progress = true;
 245                                 break;
 246                         }
 247
 248                         if (is_constant_value(c, inst->src[0], ~0)) {
 249                                 replace_with_mov(c, inst, inst->src[1]);
 250                                 progress = true;
 251                                 break;
 252                         }
 253                         if (is_constant_value(c, inst->src[1], ~0)) {
 254                                 replace_with_mov(c, inst, inst->src[0]);
 255                                 progress = true;
 256                                 break;
 257                         }
 258                         break;
 259
 260                 case QOP_OR:
 261                         if (replace_x_0_with_x(c, inst, 0) ||
 262                             replace_x_0_with_x(c, inst, 1)) {
 263                                 progress = true;
 264                                 break;
 265                         }
 266                         break;
 267
 268                 case QOP_RCP:
 269                         if (is_1f(c, inst->src[0])) {
 270                                 replace_with_mov(c, inst, inst->src[0]);
 271                                 progress = true;
 272                                 break;
 273                         }
 274                         break;
 275
 276                 default:
 277                         break;
 278                 }
 279         }
 280
 281         return progress;
 282 }