src/intel/compiler/brw_fs_cse.cpp

   1 /*
   2  * Copyright © 2012 Intel Corporation
   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 #include "brw_fs.h"
  25 #include "brw_cfg.h"
  26
  27 /** @file brw_fs_cse.cpp
  28  *
  29  * Support for local common subexpression elimination.
  30  *
  31  * See Muchnick's Advanced Compiler Design and Implementation, section
  32  * 13.1 (p378).
  33  */
  34
  35 using namespace brw;
  36
  37 namespace {
  38 struct aeb_entry : public exec_node {
  39    /** The instruction that generates the expression value. */
  40    fs_inst *generator;
  41
  42    /** The temporary where the value is stored. */
  43    fs_reg tmp;
  44 };
  45 }
  46
  47 static bool
  48 is_expression(const fs_visitor *v, const fs_inst *const inst)
  49 {
  50    switch (inst->opcode) {
  51    case BRW_OPCODE_MOV:
  52    case BRW_OPCODE_SEL:
  53    case BRW_OPCODE_NOT:
  54    case BRW_OPCODE_AND:
  55    case BRW_OPCODE_OR:
  56    case BRW_OPCODE_XOR:
  57    case BRW_OPCODE_SHR:
  58    case BRW_OPCODE_SHL:
  59    case BRW_OPCODE_ASR:
  60    case BRW_OPCODE_CMP:
  61    case BRW_OPCODE_CMPN:
  62    case BRW_OPCODE_ADD:
  63    case BRW_OPCODE_MUL:
  64    case SHADER_OPCODE_MULH:
  65    case BRW_OPCODE_FRC:
  66    case BRW_OPCODE_RNDU:
  67    case BRW_OPCODE_RNDD:
  68    case BRW_OPCODE_RNDE:
  69    case BRW_OPCODE_RNDZ:
  70    case BRW_OPCODE_LINE:
  71    case BRW_OPCODE_PLN:
  72    case BRW_OPCODE_MAD:
  73    case BRW_OPCODE_LRP:
  74    case FS_OPCODE_FB_READ_LOGICAL:
  75    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
  76    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL:
  77    case FS_OPCODE_LINTERP:
  78    case SHADER_OPCODE_FIND_LIVE_CHANNEL:
  79    case SHADER_OPCODE_BROADCAST:
  80    case SHADER_OPCODE_MOV_INDIRECT:
  81    case SHADER_OPCODE_TEX_LOGICAL:
  82    case SHADER_OPCODE_TXD_LOGICAL:
  83    case SHADER_OPCODE_TXF_LOGICAL:
  84    case SHADER_OPCODE_TXL_LOGICAL:
  85    case SHADER_OPCODE_TXS_LOGICAL:
  86    case FS_OPCODE_TXB_LOGICAL:
  87    case SHADER_OPCODE_TXF_CMS_LOGICAL:
  88    case SHADER_OPCODE_TXF_CMS_W_LOGICAL:
  89    case SHADER_OPCODE_TXF_UMS_LOGICAL:
  90    case SHADER_OPCODE_TXF_MCS_LOGICAL:
  91    case SHADER_OPCODE_LOD_LOGICAL:
  92    case SHADER_OPCODE_TG4_LOGICAL:
  93    case SHADER_OPCODE_TG4_OFFSET_LOGICAL:
  94    case FS_OPCODE_PACK:
  95       return true;
  96    case SHADER_OPCODE_RCP:
  97    case SHADER_OPCODE_RSQ:
  98    case SHADER_OPCODE_SQRT:
  99    case SHADER_OPCODE_EXP2:
 100    case SHADER_OPCODE_LOG2:
 101    case SHADER_OPCODE_POW:
 102    case SHADER_OPCODE_INT_QUOTIENT:
 103    case SHADER_OPCODE_INT_REMAINDER:
 104    case SHADER_OPCODE_SIN:
 105    case SHADER_OPCODE_COS:
 106       return inst->mlen < 2;
 107    case SHADER_OPCODE_LOAD_PAYLOAD:
 108       return !inst->is_copy_payload(v->alloc);
 109    default:
 110       return inst->is_send_from_grf() && !inst->has_side_effects() &&
 111          !inst->is_volatile();
 112    }
 113 }
 114
 115 static bool
 116 operands_match(const fs_inst *a, const fs_inst *b, bool *negate)
 117 {
 118    fs_reg *xs = a->src;
 119    fs_reg *ys = b->src;
 120
 121    if (a->opcode == BRW_OPCODE_MAD) {
 122       return xs[0].equals(ys[0]) &&
 123              ((xs[1].equals(ys[1]) && xs[2].equals(ys[2])) ||
 124               (xs[2].equals(ys[1]) && xs[1].equals(ys[2])));
 125    } else if (a->opcode == BRW_OPCODE_MUL && a->dst.type == BRW_REGISTER_TYPE_F) {
 126       bool xs0_negate = xs[0].negate;
 127       bool xs1_negate = xs[1].file == IMM ? xs[1].f < 0.0f
 128                                           : xs[1].negate;
 129       bool ys0_negate = ys[0].negate;
 130       bool ys1_negate = ys[1].file == IMM ? ys[1].f < 0.0f
 131                                           : ys[1].negate;
 132       float xs1_imm = xs[1].f;
 133       float ys1_imm = ys[1].f;
 134
 135       xs[0].negate = false;
 136       xs[1].negate = false;
 137       ys[0].negate = false;
 138       ys[1].negate = false;
 139       xs[1].f = fabsf(xs[1].f);
 140       ys[1].f = fabsf(ys[1].f);
 141
 142       bool ret = (xs[0].equals(ys[0]) && xs[1].equals(ys[1])) ||
 143                  (xs[1].equals(ys[0]) && xs[0].equals(ys[1]));
 144
 145       xs[0].negate = xs0_negate;
 146       xs[1].negate = xs[1].file == IMM ? false : xs1_negate;
 147       ys[0].negate = ys0_negate;
 148       ys[1].negate = ys[1].file == IMM ? false : ys1_negate;
 149       xs[1].f = xs1_imm;
 150       ys[1].f = ys1_imm;
 151
 152       *negate = (xs0_negate != xs1_negate) != (ys0_negate != ys1_negate);
 153       if (*negate && (a->saturate || b->saturate))
 154          return false;
 155       return ret;
 156    } else if (!a->is_commutative()) {
 157       bool match = true;
 158       for (int i = 0; i < a->sources; i++) {
 159          if (!xs[i].equals(ys[i])) {
 160             match = false;
 161             break;
 162          }
 163       }
 164       return match;
 165    } else {
 166       return (xs[0].equals(ys[0]) && xs[1].equals(ys[1])) ||
 167              (xs[1].equals(ys[0]) && xs[0].equals(ys[1]));
 168    }
 169 }
 170
 171 static bool
 172 instructions_match(fs_inst *a, fs_inst *b, bool *negate)
 173 {
 174    return a->opcode == b->opcode &&
 175           a->force_writemask_all == b->force_writemask_all &&
 176           a->exec_size == b->exec_size &&
 177           a->group == b->group &&
 178           a->saturate == b->saturate &&
 179           a->predicate == b->predicate &&
 180           a->predicate_inverse == b->predicate_inverse &&
 181           a->conditional_mod == b->conditional_mod &&
 182           a->flag_subreg == b->flag_subreg &&
 183           a->dst.type == b->dst.type &&
 184           a->offset == b->offset &&
 185           a->mlen == b->mlen &&
 186           a->ex_mlen == b->ex_mlen &&
 187           a->sfid == b->sfid &&
 188           a->desc == b->desc &&
 189           a->size_written == b->size_written &&
 190           a->base_mrf == b->base_mrf &&
 191           a->check_tdr == b->check_tdr &&
 192           a->send_has_side_effects == b->send_has_side_effects &&
 193           a->eot == b->eot &&
 194           a->header_size == b->header_size &&
 195           a->shadow_compare == b->shadow_compare &&
 196           a->pi_noperspective == b->pi_noperspective &&
 197           a->target == b->target &&
 198           a->sources == b->sources &&
 199           operands_match(a, b, negate);
 200 }
 201
 202 static void
 203 create_copy_instr(const fs_builder &bld, fs_inst *inst, fs_reg src, bool negate)
 204 {
 205    unsigned written = regs_written(inst);
 206    unsigned dst_width =
 207       DIV_ROUND_UP(inst->dst.component_size(inst->exec_size), REG_SIZE);
 208    fs_inst *copy;
 209
 210    if (inst->opcode == SHADER_OPCODE_LOAD_PAYLOAD ||
 211        written != dst_width) {
 212       fs_reg *payload;
 213       int sources, header_size;
 214       if (inst->opcode == SHADER_OPCODE_LOAD_PAYLOAD) {
 215          sources = inst->sources;
 216          header_size = inst->header_size;
 217       } else {
 218          assert(written % dst_width == 0);
 219          sources = written / dst_width;
 220          header_size = 0;
 221       }
 222
 223       assert(src.file == VGRF);
 224       payload = ralloc_array(bld.shader->mem_ctx, fs_reg, sources);
 225       for (int i = 0; i < header_size; i++) {
 226          payload[i] = src;
 227          src.offset += REG_SIZE;
 228       }
 229       for (int i = header_size; i < sources; i++) {
 230          payload[i] = src;
 231          src = offset(src, bld, 1);
 232       }
 233       copy = bld.LOAD_PAYLOAD(inst->dst, payload, sources, header_size);
 234    } else {
 235       copy = bld.MOV(inst->dst, src);
 236       copy->group = inst->group;
 237       copy->force_writemask_all = inst->force_writemask_all;
 238       copy->src[0].negate = negate;
 239    }
 240    assert(regs_written(copy) == written);
 241 }
 242
 243 bool
 244 fs_visitor::opt_cse_local(bblock_t *block)
 245 {
 246    bool progress = false;
 247    exec_list aeb;
 248
 249    void *cse_ctx = ralloc_context(NULL);
 250
 251    int ip = block->start_ip;
 252    foreach_inst_in_block(fs_inst, inst, block) {
 253       /* Skip some cases. */
 254       if (is_expression(this, inst) && !inst->is_partial_write() &&
 255           ((inst->dst.file != ARF && inst->dst.file != FIXED_GRF) ||
 256            inst->dst.is_null()))
 257       {
 258          bool found = false;
 259          bool negate = false;
 260
 261          foreach_in_list_use_after(aeb_entry, entry, &aeb) {
 262             /* Match current instruction's expression against those in AEB. */
 263             if (!(entry->generator->dst.is_null() && !inst->dst.is_null()) &&
 264                 instructions_match(inst, entry->generator, &negate)) {
 265                found = true;
 266                progress = true;
 267                break;
 268             }
 269          }
 270
 271          if (!found) {
 272             if (inst->opcode != BRW_OPCODE_MOV ||
 273                 (inst->opcode == BRW_OPCODE_MOV &&
 274                  inst->src[0].file == IMM &&
 275                  inst->src[0].type == BRW_REGISTER_TYPE_VF)) {
 276                /* Our first sighting of this expression.  Create an entry. */
 277                aeb_entry *entry = ralloc(cse_ctx, aeb_entry);
 278                entry->tmp = reg_undef;
 279                entry->generator = inst;
 280                aeb.push_tail(entry);
 281             }
 282          } else {
 283             /* This is at least our second sighting of this expression.
 284              * If we don't have a temporary already, make one.
 285              */
 286             bool no_existing_temp = entry->tmp.file == BAD_FILE;
 287             if (no_existing_temp && !entry->generator->dst.is_null()) {
 288                const fs_builder ibld = fs_builder(this, block, entry->generator)
 289                                        .at(block, entry->generator->next);
 290                int written = regs_written(entry->generator);
 291
 292                entry->tmp = fs_reg(VGRF, alloc.allocate(written),
 293                                    entry->generator->dst.type);
 294
 295                create_copy_instr(ibld, entry->generator, entry->tmp, false);
 296
 297                entry->generator->dst = entry->tmp;
 298             }
 299
 300             /* dest <- temp */
 301             if (!inst->dst.is_null()) {
 302                assert(inst->size_written == entry->generator->size_written);
 303                assert(inst->dst.type == entry->tmp.type);
 304                const fs_builder ibld(this, block, inst);
 305
 306                create_copy_instr(ibld, inst, entry->tmp, negate);
 307             }
 308
 309             /* Set our iterator so that next time through the loop inst->next
 310              * will get the instruction in the basic block after the one we've
 311              * removed.
 312              */
 313             fs_inst *prev = (fs_inst *)inst->prev;
 314
 315             inst->remove(block);
 316             inst = prev;
 317          }
 318       }
 319
 320       foreach_in_list_safe(aeb_entry, entry, &aeb) {
 321          /* Kill all AEB entries that write a different value to or read from
 322           * the flag register if we just wrote it.
 323           */
 324          if (inst->flags_written()) {
 325             bool negate; /* dummy */
 326             if (entry->generator->flags_read(devinfo) ||
 327                 (entry->generator->flags_written() &&
 328                  !instructions_match(inst, entry->generator, &negate))) {
 329                entry->remove();
 330                ralloc_free(entry);
 331                continue;
 332             }
 333          }
 334
 335          for (int i = 0; i < entry->generator->sources; i++) {
 336             fs_reg *src_reg = &entry->generator->src[i];
 337
 338             /* Kill all AEB entries that use the destination we just
 339              * overwrote.
 340              */
 341             if (regions_overlap(inst->dst, inst->size_written,
 342                                 entry->generator->src[i],
 343                                 entry->generator->size_read(i))) {
 344                entry->remove();
 345                ralloc_free(entry);
 346                break;
 347             }
 348
 349             /* Kill any AEB entries using registers that don't get reused any
 350              * more -- a sure sign they'll fail operands_match().
 351              */
 352             if (src_reg->file == VGRF && virtual_grf_end[src_reg->nr] < ip) {
 353                entry->remove();
 354                ralloc_free(entry);
 355                break;
 356             }
 357          }
 358       }
 359
 360       ip++;
 361    }
 362
 363    ralloc_free(cse_ctx);
 364
 365    return progress;
 366 }
 367
 368 bool
 369 fs_visitor::opt_cse()
 370 {
 371    bool progress = false;
 372
 373    calculate_live_intervals();
 374
 375    foreach_block (block, cfg) {
 376       progress = opt_cse_local(block) || progress;
 377    }
 378
 379    if (progress)
 380       invalidate_live_intervals();
 381
 382    return progress;
 383 }