src/broadcom/compiler/vir_opt_copy_propagate.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_copy_propagation.c
  26  *
  27  * This implements simple copy propagation for VIR without control flow.
  28  *
  29  * For each temp, it keeps a qreg of which source it was MOVed from, if it
  30  * was.  If we see that used later, we can just reuse the source value, since
  31  * we know we don't have control flow, and we have SSA for our values so
  32  * there's no killing to worry about.
  33  */
  34
  35 #include "v3d_compiler.h"
  36
  37 static bool
  38 is_copy_mov(struct qinst *inst)
  39 {
  40         if (!inst)
  41                 return false;
  42
  43         if (inst->qpu.type != V3D_QPU_INSTR_TYPE_ALU ||
  44             (inst->qpu.alu.mul.op != V3D_QPU_M_FMOV &&
  45              inst->qpu.alu.mul.op != V3D_QPU_M_MOV)) {
  46                 return false;
  47         }
  48
  49         if (inst->dst.file != QFILE_TEMP)
  50                 return false;
  51
  52         if (inst->src[0].file != QFILE_TEMP)
  53                 return false;
  54
  55         if (inst->qpu.alu.add.output_pack != V3D_QPU_PACK_NONE ||
  56             inst->qpu.alu.mul.output_pack != V3D_QPU_PACK_NONE) {
  57                 return false;
  58         }
  59
  60         if (inst->qpu.flags.ac != V3D_QPU_COND_NONE ||
  61             inst->qpu.flags.mc != V3D_QPU_COND_NONE) {
  62                 return false;
  63         }
  64
  65         switch (inst->src[0].file) {
  66         case QFILE_MAGIC:
  67                 /* No copy propagating from R3/R4/R5 -- the MOVs from those
  68                  * are there to register allocate values produced into R3/4/5
  69                  * to other regs (though hopefully r3/4/5).
  70                  */
  71                 switch (inst->src[0].index) {
  72                 case V3D_QPU_WADDR_R3:
  73                 case V3D_QPU_WADDR_R4:
  74                 case V3D_QPU_WADDR_R5:
  75                         return false;
  76                 default:
  77                         break;
  78                 }
  79                 break;
  80
  81         case QFILE_REG:
  82                 switch (inst->src[0].index) {
  83                 case 0:
  84                 case 1:
  85                 case 2:
  86                         /* MOVs from rf0/1/2 are only to track the live
  87                          * intervals for W/centroid W/Z.
  88                          */
  89                         return false;
  90                 }
  91                 break;
  92
  93         default:
  94                 break;
  95         }
  96
  97         return true;
  98 }
  99
 100 static bool
 101 vir_has_unpack(struct qinst *inst, int chan)
 102 {
 103         assert(chan == 0 || chan == 1);
 104
 105         if (vir_is_add(inst)) {
 106                 if (chan == 0)
 107                         return inst->qpu.alu.add.a_unpack != V3D_QPU_UNPACK_NONE;
 108                 else
 109                         return inst->qpu.alu.add.b_unpack != V3D_QPU_UNPACK_NONE;
 110         } else {
 111                 if (chan == 0)
 112                         return inst->qpu.alu.mul.a_unpack != V3D_QPU_UNPACK_NONE;
 113                 else
 114                         return inst->qpu.alu.mul.b_unpack != V3D_QPU_UNPACK_NONE;
 115         }
 116 }
 117
 118 static bool
 119 try_copy_prop(struct v3d_compile *c, struct qinst *inst, struct qinst **movs)
 120 {
 121         bool debug = false;
 122         bool progress = false;
 123
 124         for (int i = 0; i < vir_get_nsrc(inst); i++) {
 125                 if (inst->src[i].file != QFILE_TEMP)
 126                         continue;
 127
 128                 /* We have two ways of finding MOVs we can copy propagate
 129                  * from.  One is if it's an SSA def: then we can reuse it from
 130                  * any block in the program, as long as its source is also an
 131                  * SSA def.  Alternatively, if it's in the "movs" array
 132                  * tracked within the block, then we know the sources for it
 133                  * haven't been changed since we saw the instruction within
 134                  * our block.
 135                  */
 136                 struct qinst *mov = movs[inst->src[i].index];
 137                 if (!mov) {
 138                         if (!is_copy_mov(c->defs[inst->src[i].index]))
 139                                 continue;
 140                         mov = c->defs[inst->src[i].index];
 141
 142                         if (mov->src[0].file == QFILE_TEMP &&
 143                             !c->defs[mov->src[0].index])
 144                                 continue;
 145                 }
 146
 147                 if (vir_has_unpack(mov, 0)) {
 148                         /* Make sure that the meaning of the unpack
 149                          * would be the same between the two
 150                          * instructions.
 151                          */
 152                         if (v3d_qpu_unpacks_f32(&inst->qpu) !=
 153                             v3d_qpu_unpacks_f32(&mov->qpu) ||
 154                             v3d_qpu_unpacks_f16(&inst->qpu) !=
 155                             v3d_qpu_unpacks_f16(&mov->qpu)) {
 156                                 continue;
 157                         }
 158
 159                         /* No composing the unpacks. */
 160                         if (vir_has_unpack(inst, i))
 161                                 continue;
 162
 163                         /* these ops can't represent abs. */
 164                         if (mov->qpu.alu.mul.a_unpack == V3D_QPU_UNPACK_ABS) {
 165                                 switch (inst->qpu.alu.add.op) {
 166                                 case V3D_QPU_A_VFPACK:
 167                                 case V3D_QPU_A_FROUND:
 168                                 case V3D_QPU_A_FTRUNC:
 169                                 case V3D_QPU_A_FFLOOR:
 170                                 case V3D_QPU_A_FCEIL:
 171                                 case V3D_QPU_A_FDX:
 172                                 case V3D_QPU_A_FDY:
 173                                 case V3D_QPU_A_FTOIN:
 174                                 case V3D_QPU_A_FTOIZ:
 175                                 case V3D_QPU_A_FTOUZ:
 176                                 case V3D_QPU_A_FTOC:
 177                                         continue;
 178                                 default:
 179                                         break;
 180                                 }
 181                         }
 182                 }
 183
 184                 if (debug) {
 185                         fprintf(stderr, "Copy propagate: ");
 186                         vir_dump_inst(c, inst);
 187                         fprintf(stderr, "\n");
 188                 }
 189
 190                 inst->src[i] = mov->src[0];
 191                 if (vir_has_unpack(mov, 0)) {
 192                         enum v3d_qpu_input_unpack unpack = mov->qpu.alu.mul.a_unpack;
 193
 194                         vir_set_unpack(inst, i, unpack);
 195                 }
 196
 197                 if (debug) {
 198                         fprintf(stderr, "to: ");
 199                         vir_dump_inst(c, inst);
 200                         fprintf(stderr, "\n");
 201                 }
 202
 203                 progress = true;
 204         }
 205
 206         return progress;
 207 }
 208
 209 static void
 210 apply_kills(struct v3d_compile *c, struct qinst **movs, struct qinst *inst)
 211 {
 212         if (inst->dst.file != QFILE_TEMP)
 213                 return;
 214
 215         for (int i = 0; i < c->num_temps; i++) {
 216                 if (movs[i] &&
 217                     (movs[i]->dst.index == inst->dst.index ||
 218                      (movs[i]->src[0].file == QFILE_TEMP &&
 219                       movs[i]->src[0].index == inst->dst.index))) {
 220                         movs[i] = NULL;
 221                 }
 222         }
 223 }
 224
 225 bool
 226 vir_opt_copy_propagate(struct v3d_compile *c)
 227 {
 228         bool progress = false;
 229         struct qinst **movs;
 230
 231         movs = ralloc_array(c, struct qinst *, c->num_temps);
 232         if (!movs)
 233                 return false;
 234
 235         vir_for_each_block(block, c) {
 236                 /* The MOVs array tracks only available movs within the
 237                  * block.
 238                  */
 239                 memset(movs, 0, sizeof(struct qinst *) * c->num_temps);
 240
 241                 vir_for_each_inst(inst, block) {
 242                         progress = try_copy_prop(c, inst, movs) || progress;
 243
 244                         apply_kills(c, movs, inst);
 245
 246                         if (is_copy_mov(inst))
 247                                 movs[inst->dst.index] = inst;
 248                 }
 249         }
 250
 251         ralloc_free(movs);
 252
 253         return progress;
 254 }