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