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_SEL_X_Y_ZS:
 149                 case QOP_SEL_X_Y_ZC:
 150                 case QOP_SEL_X_Y_NS:
 151                 case QOP_SEL_X_Y_NC:
 152                 case QOP_SEL_X_Y_CS:
 153                 case QOP_SEL_X_Y_CC:
 154                         if (is_zero(c, inst->src[1])) {
 155                                 /* Replace references to a 0 uniform value
 156                                  * with the SEL_X_0 equivalent.
 157                                  */
 158                                 dump_from(c, inst);
 159                                 inst->op -= (QOP_SEL_X_Y_ZS - QOP_SEL_X_0_ZS);
 160                                 inst->src[1] = c->undef;
 161                                 progress = true;
 162                                 dump_to(c, inst);
 163                                 break;
 164                         }
 165
 166                         if (is_zero(c, inst->src[0])) {
 167                                 /* Replace references to a 0 uniform value
 168                                  * with the SEL_X_0 equivalent, flipping the
 169                                  * condition being evaluated since the operand
 170                                  * order is flipped.
 171                                  */
 172                                 dump_from(c, inst);
 173                                 inst->op -= QOP_SEL_X_Y_ZS;
 174                                 inst->op ^= 1;
 175                                 inst->op += QOP_SEL_X_0_ZS;
 176                                 inst->src[0] = inst->src[1];
 177                                 inst->src[1] = c->undef;
 178                                 progress = true;
 179                                 dump_to(c, inst);
 180                                 break;
 181                         }
 182
 183                         break;
 184
 185                 case QOP_FMIN:
 186                         if (is_1f(c, inst->src[1]) &&
 187                             inst->src[0].pack >= QPU_UNPACK_8D_REP &&
 188                             inst->src[0].pack <= QPU_UNPACK_8D) {
 189                                 replace_with_mov(c, inst, inst->src[0]);
 190                                 progress = true;
 191                         }
 192                         break;
 193
 194                 case QOP_FMAX:
 195                         if (is_zero(c, inst->src[1]) &&
 196                             inst->src[0].pack >= QPU_UNPACK_8D_REP &&
 197                             inst->src[0].pack <= QPU_UNPACK_8D) {
 198                                 replace_with_mov(c, inst, inst->src[0]);
 199                                 progress = true;
 200                         }
 201                         break;
 202
 203                 case QOP_FSUB:
 204                 case QOP_SUB:
 205                         if (is_zero(c, inst->src[1])) {
 206                                 replace_with_mov(c, inst, inst->src[0]);
 207                                 progress = true;
 208                         }
 209                         break;
 210
 211                 case QOP_ADD:
 212                         if (replace_x_0_with_x(c, inst, 0) ||
 213                             replace_x_0_with_x(c, inst, 1)) {
 214                                 progress = true;
 215                                 break;
 216                         }
 217                         break;
 218
 219                 case QOP_FADD:
 220                         if (replace_x_0_with_x(c, inst, 0) ||
 221                             replace_x_0_with_x(c, inst, 1)) {
 222                                 progress = true;
 223                                 break;
 224                         }
 225
 226                         /* FADD(a, FSUB(0, b)) -> FSUB(a, b) */
 227                         if (inst->src[1].file == QFILE_TEMP &&
 228                             c->defs[inst->src[1].index] &&
 229                             c->defs[inst->src[1].index]->op == QOP_FSUB) {
 230                                 struct qinst *fsub = c->defs[inst->src[1].index];
 231                                 if (is_zero(c, fsub->src[0])) {
 232                                         dump_from(c, inst);
 233                                         inst->op = QOP_FSUB;
 234                                         inst->src[1] = fsub->src[1];
 235                                         progress = true;
 236                                         dump_to(c, inst);
 237                                         break;
 238                                 }
 239                         }
 240
 241                         /* FADD(FSUB(0, b), a) -> FSUB(a, b) */
 242                         if (inst->src[0].file == QFILE_TEMP &&
 243                             c->defs[inst->src[0].index] &&
 244                             c->defs[inst->src[0].index]->op == QOP_FSUB) {
 245                                 struct qinst *fsub = c->defs[inst->src[0].index];
 246                                 if (is_zero(c, fsub->src[0])) {
 247                                         dump_from(c, inst);
 248                                         inst->op = QOP_FSUB;
 249                                         inst->src[0] = inst->src[1];
 250                                         inst->src[1] = fsub->src[1];
 251                                         dump_to(c, inst);
 252                                         progress = true;
 253                                         break;
 254                                 }
 255                         }
 256                         break;
 257
 258                 case QOP_FMUL:
 259                         if (!inst->dst.pack &&
 260                             (replace_x_0_with_0(c, inst, 0) ||
 261                              replace_x_0_with_0(c, inst, 1) ||
 262                              fmul_replace_one(c, inst, 0) ||
 263                              fmul_replace_one(c, inst, 1))) {
 264                                 progress = true;
 265                                 break;
 266                         }
 267                         break;
 268
 269                 case QOP_MUL24:
 270                         if (!inst->dst.pack &&
 271                             (replace_x_0_with_0(c, inst, 0) ||
 272                              replace_x_0_with_0(c, inst, 1))) {
 273                                 progress = true;
 274                                 break;
 275                         }
 276                         break;
 277
 278                 case QOP_AND:
 279                         if (replace_x_0_with_0(c, inst, 0) ||
 280                             replace_x_0_with_0(c, inst, 1)) {
 281                                 progress = true;
 282                                 break;
 283                         }
 284
 285                         if (is_constant_value(c, inst->src[0], ~0)) {
 286                                 replace_with_mov(c, inst, inst->src[1]);
 287                                 progress = true;
 288                                 break;
 289                         }
 290                         if (is_constant_value(c, inst->src[1], ~0)) {
 291                                 replace_with_mov(c, inst, inst->src[0]);
 292                                 progress = true;
 293                                 break;
 294                         }
 295                         break;
 296
 297                 case QOP_OR:
 298                         if (replace_x_0_with_x(c, inst, 0) ||
 299                             replace_x_0_with_x(c, inst, 1)) {
 300                                 progress = true;
 301                                 break;
 302                         }
 303                         break;
 304
 305                 case QOP_RCP:
 306                         if (is_1f(c, inst->src[0])) {
 307                                 replace_with_mov(c, inst, inst->src[0]);
 308                                 progress = true;
 309                                 break;
 310                         }
 311                         break;
 312
 313                 default:
 314                         break;
 315                 }
 316         }
 317
 318         return progress;
 319 }