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 qinst **defs, struct qreg reg,
  64                   uint32_t val)
  65 {
  66         if (reg.file == QFILE_UNIF &&
  67             c->uniform_contents[reg.index] == QUNIFORM_CONSTANT &&
  68             c->uniform_data[reg.index] == val) {
  69                 return true;
  70         }
  71
  72         if (reg.file == QFILE_SMALL_IMM && reg.index == val)
  73                 return true;
  74
  75         return false;
  76 }
  77
  78 static bool
  79 is_zero(struct vc4_compile *c, struct qinst **defs, struct qreg reg)
  80 {
  81         reg = qir_follow_movs(defs, reg);
  82         return is_constant_value(c, defs, reg, 0);
  83 }
  84
  85 static bool
  86 is_1f(struct vc4_compile *c, struct qinst **defs, struct qreg reg)
  87 {
  88         reg = qir_follow_movs(defs, reg);
  89         return is_constant_value(c, defs, reg, fui(1.0));
  90 }
  91
  92 static void
  93 replace_with_mov(struct vc4_compile *c, struct qinst *inst, struct qreg arg)
  94 {
  95         dump_from(c, inst);
  96         inst->op = QOP_MOV;
  97         inst->src[0] = arg;
  98         inst->src[1] = c->undef;
  99         dump_to(c, inst);
 100 }
 101
 102 static bool
 103 add_replace_zero(struct vc4_compile *c,
 104                  struct qinst **defs,
 105                  struct qinst *inst,
 106                  int arg)
 107 {
 108         if (!is_zero(c, defs, inst->src[arg]))
 109                 return false;
 110         replace_with_mov(c, inst, inst->src[1 - arg]);
 111         return true;
 112 }
 113
 114 static bool
 115 fmul_replace_zero(struct vc4_compile *c,
 116                   struct qinst **defs,
 117                   struct qinst *inst,
 118                   int arg)
 119 {
 120         if (!is_zero(c, defs, inst->src[arg]))
 121                 return false;
 122         replace_with_mov(c, inst, inst->src[arg]);
 123         return true;
 124 }
 125
 126 static bool
 127 fmul_replace_one(struct vc4_compile *c,
 128                  struct qinst **defs,
 129                  struct qinst *inst,
 130                  int arg)
 131 {
 132         if (!is_1f(c, defs, inst->src[arg]))
 133                 return false;
 134         replace_with_mov(c, inst, inst->src[1 - arg]);
 135         return true;
 136 }
 137
 138 bool
 139 qir_opt_algebraic(struct vc4_compile *c)
 140 {
 141         bool progress = false;
 142         struct simple_node *node;
 143         struct qinst *defs[c->num_temps];
 144
 145         foreach(node, &c->instructions) {
 146                 struct qinst *inst = (struct qinst *)node;
 147
 148                 if (inst->dst.file == QFILE_TEMP)
 149                         defs[inst->dst.index] = inst;
 150
 151                 switch (inst->op) {
 152                 case QOP_SF:
 153                         /* SF just looks at the sign bit, or whether all the
 154                          * bits are 0.  This is preserved across an itof
 155                          * transformation.
 156                          */
 157                         if (inst->src[0].file == QFILE_TEMP &&
 158                             defs[inst->src[0].index]->op == QOP_ITOF) {
 159                                 dump_from(c, inst);
 160                                 inst->src[0] =
 161                                         defs[inst->src[0].index]->src[0];
 162                                 progress =  true;
 163                                 dump_to(c, inst);
 164                                 break;
 165                         }
 166                         break;
 167
 168                 case QOP_SEL_X_Y_ZS:
 169                 case QOP_SEL_X_Y_ZC:
 170                 case QOP_SEL_X_Y_NS:
 171                 case QOP_SEL_X_Y_NC:
 172                         if (qir_reg_equals(inst->src[0], inst->src[1])) {
 173                                 /* Turn "dst = (sf == x) ? a : a)" into
 174                                  * "dst = a"
 175                                  */
 176                                 replace_with_mov(c, inst, inst->src[1]);
 177                                 progress = true;
 178                                 break;
 179                         }
 180
 181                         if (is_zero(c, defs, inst->src[1])) {
 182                                 /* Replace references to a 0 uniform value
 183                                  * with the SEL_X_0 equivalent.
 184                                  */
 185                                 dump_from(c, inst);
 186                                 inst->op -= (QOP_SEL_X_Y_ZS - QOP_SEL_X_0_ZS);
 187                                 inst->src[1] = c->undef;
 188                                 progress = true;
 189                                 dump_to(c, inst);
 190                                 break;
 191                         }
 192
 193                         if (is_zero(c, defs, inst->src[0])) {
 194                                 /* Replace references to a 0 uniform value
 195                                  * with the SEL_X_0 equivalent, flipping the
 196                                  * condition being evaluated since the operand
 197                                  * order is flipped.
 198                                  */
 199                                 dump_from(c, inst);
 200                                 inst->op -= QOP_SEL_X_Y_ZS;
 201                                 inst->op ^= 1;
 202                                 inst->op += QOP_SEL_X_0_ZS;
 203                                 inst->src[0] = inst->src[1];
 204                                 inst->src[1] = c->undef;
 205                                 progress = true;
 206                                 dump_to(c, inst);
 207                                 break;
 208                         }
 209
 210                         break;
 211
 212                 case QOP_FSUB:
 213                 case QOP_SUB:
 214                         if (is_zero(c, defs, inst->src[1])) {
 215                                 replace_with_mov(c, inst, inst->src[0]);
 216                         }
 217                         break;
 218
 219                 case QOP_ADD:
 220                         if (add_replace_zero(c, defs, inst, 0) ||
 221                             add_replace_zero(c, defs, inst, 1)) {
 222                                 progress = true;
 223                                 break;
 224                         }
 225                         break;
 226
 227                 case QOP_FADD:
 228                         if (add_replace_zero(c, defs, inst, 0) ||
 229                             add_replace_zero(c, defs, inst, 1)) {
 230                                 progress = true;
 231                                 break;
 232                         }
 233
 234                         /* FADD(a, FSUB(0, b)) -> FSUB(a, b) */
 235                         if (inst->src[1].file == QFILE_TEMP &&
 236                             defs[inst->src[1].index]->op == QOP_FSUB) {
 237                                 struct qinst *fsub = defs[inst->src[1].index];
 238                                 if (is_zero(c, defs, fsub->src[0])) {
 239                                         dump_from(c, inst);
 240                                         inst->op = QOP_FSUB;
 241                                         inst->src[1] = fsub->src[1];
 242                                         progress = true;
 243                                         dump_to(c, inst);
 244                                         break;
 245                                 }
 246                         }
 247
 248                         /* FADD(FSUB(0, b), a) -> FSUB(a, b) */
 249                         if (inst->src[0].file == QFILE_TEMP &&
 250                             defs[inst->src[0].index]->op == QOP_FSUB) {
 251                                 struct qinst *fsub = defs[inst->src[0].index];
 252                                 if (is_zero(c, defs, fsub->src[0])) {
 253                                         dump_from(c, inst);
 254                                         inst->op = QOP_FSUB;
 255                                         inst->src[0] = inst->src[1];
 256                                         inst->src[1] = fsub->src[1];
 257                                         dump_to(c, inst);
 258                                         progress = true;
 259                                         break;
 260                                 }
 261                         }
 262                         break;
 263
 264                 case QOP_FMUL:
 265                         if (fmul_replace_zero(c, defs, inst, 0) ||
 266                             fmul_replace_zero(c, defs, inst, 1) ||
 267                             fmul_replace_one(c, defs, inst, 0) ||
 268                             fmul_replace_one(c, defs, inst, 1)) {
 269                                 progress = true;
 270                                 break;
 271                         }
 272                         break;
 273
 274                 default:
 275                         break;
 276                 }
 277         }
 278
 279         return progress;
 280 }