src/mesa/drivers/dri/i965/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_UNIFORM_PULL_CONSTANT_LOAD:
  75    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
  76    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD:
  77    case FS_OPCODE_CINTERP:
  78    case FS_OPCODE_LINTERP:
  79    case SHADER_OPCODE_FIND_LIVE_CHANNEL:
  80    case SHADER_OPCODE_BROADCAST:
  81    case SHADER_OPCODE_MOV_INDIRECT:
  82       return true;
  83    case SHADER_OPCODE_RCP:
  84    case SHADER_OPCODE_RSQ:
  85    case SHADER_OPCODE_SQRT:
  86    case SHADER_OPCODE_EXP2:
  87    case SHADER_OPCODE_LOG2:
  88    case SHADER_OPCODE_POW:
  89    case SHADER_OPCODE_INT_QUOTIENT:
  90    case SHADER_OPCODE_INT_REMAINDER:
  91    case SHADER_OPCODE_SIN:
  92    case SHADER_OPCODE_COS:
  93       return inst->mlen < 2;
  94    case SHADER_OPCODE_LOAD_PAYLOAD:
  95       return !inst->is_copy_payload(v->alloc);
  96    default:
  97       return inst->is_send_from_grf() && !inst->has_side_effects() &&
  98          !inst->is_volatile();
  99    }
 100 }
 101
 102 static bool
 103 operands_match(const fs_inst *a, const fs_inst *b, bool *negate)
 104 {
 105    fs_reg *xs = a->src;
 106    fs_reg *ys = b->src;
 107
 108    if (a->opcode == BRW_OPCODE_MAD) {
 109       return xs[0].equals(ys[0]) &&
 110              ((xs[1].equals(ys[1]) && xs[2].equals(ys[2])) ||
 111               (xs[2].equals(ys[1]) && xs[1].equals(ys[2])));
 112    } else if (a->opcode == BRW_OPCODE_MUL && a->dst.type == BRW_REGISTER_TYPE_F) {
 113       bool xs0_negate = xs[0].negate;
 114       bool xs1_negate = xs[1].file == IMM ? xs[1].f < 0.0f
 115                                           : xs[1].negate;
 116       bool ys0_negate = ys[0].negate;
 117       bool ys1_negate = ys[1].file == IMM ? ys[1].f < 0.0f
 118                                           : ys[1].negate;
 119       float xs1_imm = xs[1].f;
 120       float ys1_imm = ys[1].f;
 121
 122       xs[0].negate = false;
 123       xs[1].negate = false;
 124       ys[0].negate = false;
 125       ys[1].negate = false;
 126       xs[1].f = fabsf(xs[1].f);
 127       ys[1].f = fabsf(ys[1].f);
 128
 129       bool ret = (xs[0].equals(ys[0]) && xs[1].equals(ys[1])) ||
 130                  (xs[1].equals(ys[0]) && xs[0].equals(ys[1]));
 131
 132       xs[0].negate = xs0_negate;
 133       xs[1].negate = xs[1].file == IMM ? false : xs1_negate;
 134       ys[0].negate = ys0_negate;
 135       ys[1].negate = ys[1].file == IMM ? false : ys1_negate;
 136       xs[1].f = xs1_imm;
 137       ys[1].f = ys1_imm;
 138
 139       *negate = (xs0_negate != xs1_negate) != (ys0_negate != ys1_negate);
 140       return ret;
 141    } else if (!a->is_commutative()) {
 142       bool match = true;
 143       for (int i = 0; i < a->sources; i++) {
 144          if (!xs[i].equals(ys[i])) {
 145             match = false;
 146             break;
 147          }
 148       }
 149       return match;
 150    } else {
 151       return (xs[0].equals(ys[0]) && xs[1].equals(ys[1])) ||
 152              (xs[1].equals(ys[0]) && xs[0].equals(ys[1]));
 153    }
 154 }
 155
 156 static bool
 157 instructions_match(fs_inst *a, fs_inst *b, bool *negate)
 158 {
 159    return a->opcode == b->opcode &&
 160           a->force_writemask_all == b->force_writemask_all &&
 161           a->exec_size == b->exec_size &&
 162           a->force_sechalf == b->force_sechalf &&
 163           a->saturate == b->saturate &&
 164           a->predicate == b->predicate &&
 165           a->predicate_inverse == b->predicate_inverse &&
 166           a->conditional_mod == b->conditional_mod &&
 167           a->flag_subreg == b->flag_subreg &&
 168           a->dst.type == b->dst.type &&
 169           a->offset == b->offset &&
 170           a->mlen == b->mlen &&
 171           a->regs_written == b->regs_written &&
 172           a->base_mrf == b->base_mrf &&
 173           a->eot == b->eot &&
 174           a->header_size == b->header_size &&
 175           a->shadow_compare == b->shadow_compare &&
 176           a->pi_noperspective == b->pi_noperspective &&
 177           a->sources == b->sources &&
 178           operands_match(a, b, negate);
 179 }
 180
 181 static void
 182 create_copy_instr(const fs_builder &bld, fs_inst *inst, fs_reg src, bool negate)
 183 {
 184    int written = inst->regs_written;
 185    int dst_width = inst->exec_size / 8;
 186    fs_inst *copy;
 187
 188    if (written > dst_width) {
 189       fs_reg *payload;
 190       int sources, header_size;
 191       if (inst->opcode == SHADER_OPCODE_LOAD_PAYLOAD) {
 192          sources = inst->sources;
 193          header_size = inst->header_size;
 194       } else {
 195          assert(written % dst_width == 0);
 196          sources = written / dst_width;
 197          header_size = 0;
 198       }
 199
 200       assert(src.file == VGRF);
 201       payload = ralloc_array(bld.shader->mem_ctx, fs_reg, sources);
 202       for (int i = 0; i < header_size; i++) {
 203          payload[i] = src;
 204          src.reg_offset++;
 205       }
 206       for (int i = header_size; i < sources; i++) {
 207          payload[i] = src;
 208          src = offset(src, bld, 1);
 209       }
 210       copy = bld.LOAD_PAYLOAD(inst->dst, payload, sources, header_size);
 211    } else {
 212       copy = bld.MOV(inst->dst, src);
 213       copy->force_sechalf = inst->force_sechalf;
 214       copy->force_writemask_all = inst->force_writemask_all;
 215       copy->src[0].negate = negate;
 216    }
 217    assert(copy->regs_written == written);
 218 }
 219
 220 bool
 221 fs_visitor::opt_cse_local(bblock_t *block)
 222 {
 223    bool progress = false;
 224    exec_list aeb;
 225
 226    void *cse_ctx = ralloc_context(NULL);
 227
 228    int ip = block->start_ip;
 229    foreach_inst_in_block(fs_inst, inst, block) {
 230       /* Skip some cases. */
 231       if (is_expression(this, inst) && !inst->is_partial_write() &&
 232           ((inst->dst.file != ARF && inst->dst.file != FIXED_GRF) ||
 233            inst->dst.is_null()))
 234       {
 235          bool found = false;
 236          bool negate = false;
 237
 238          foreach_in_list_use_after(aeb_entry, entry, &aeb) {
 239             /* Match current instruction's expression against those in AEB. */
 240             if (!(entry->generator->dst.is_null() && !inst->dst.is_null()) &&
 241                 instructions_match(inst, entry->generator, &negate)) {
 242                found = true;
 243                progress = true;
 244                break;
 245             }
 246          }
 247
 248          if (!found) {
 249             if (inst->opcode != BRW_OPCODE_MOV ||
 250                 (inst->opcode == BRW_OPCODE_MOV &&
 251                  inst->src[0].file == IMM &&
 252                  inst->src[0].type == BRW_REGISTER_TYPE_VF)) {
 253                /* Our first sighting of this expression.  Create an entry. */
 254                aeb_entry *entry = ralloc(cse_ctx, aeb_entry);
 255                entry->tmp = reg_undef;
 256                entry->generator = inst;
 257                aeb.push_tail(entry);
 258             }
 259          } else {
 260             /* This is at least our second sighting of this expression.
 261              * If we don't have a temporary already, make one.
 262              */
 263             bool no_existing_temp = entry->tmp.file == BAD_FILE;
 264             if (no_existing_temp && !entry->generator->dst.is_null()) {
 265                const fs_builder ibld = fs_builder(this, block, entry->generator)
 266                                        .at(block, entry->generator->next);
 267                int written = entry->generator->regs_written;
 268
 269                entry->tmp = fs_reg(VGRF, alloc.allocate(written),
 270                                    entry->generator->dst.type);
 271
 272                create_copy_instr(ibld, entry->generator, entry->tmp, false);
 273
 274                entry->generator->dst = entry->tmp;
 275             }
 276
 277             /* dest <- temp */
 278             if (!inst->dst.is_null()) {
 279                assert(inst->regs_written == entry->generator->regs_written);
 280                assert(inst->dst.type == entry->tmp.type);
 281                const fs_builder ibld(this, block, inst);
 282
 283                create_copy_instr(ibld, inst, entry->tmp, negate);
 284             }
 285
 286             /* Set our iterator so that next time through the loop inst->next
 287              * will get the instruction in the basic block after the one we've
 288              * removed.
 289              */
 290             fs_inst *prev = (fs_inst *)inst->prev;
 291
 292             inst->remove(block);
 293             inst = prev;
 294          }
 295       }
 296
 297       foreach_in_list_safe(aeb_entry, entry, &aeb) {
 298          /* Kill all AEB entries that write a different value to or read from
 299           * the flag register if we just wrote it.
 300           */
 301          if (inst->writes_flag()) {
 302             bool negate; /* dummy */
 303             if (entry->generator->reads_flag() ||
 304                 (entry->generator->writes_flag() &&
 305                  !instructions_match(inst, entry->generator, &negate))) {
 306                entry->remove();
 307                ralloc_free(entry);
 308                continue;
 309             }
 310          }
 311
 312          for (int i = 0; i < entry->generator->sources; i++) {
 313             fs_reg *src_reg = &entry->generator->src[i];
 314
 315             /* Kill all AEB entries that use the destination we just
 316              * overwrote.
 317              */
 318             if (inst->overwrites_reg(entry->generator->src[i])) {
 319                entry->remove();
 320                ralloc_free(entry);
 321                break;
 322             }
 323
 324             /* Kill any AEB entries using registers that don't get reused any
 325              * more -- a sure sign they'll fail operands_match().
 326              */
 327             if (src_reg->file == VGRF && virtual_grf_end[src_reg->nr] < ip) {
 328                entry->remove();
 329                ralloc_free(entry);
 330                break;
 331             }
 332          }
 333       }
 334
 335       ip++;
 336    }
 337
 338    ralloc_free(cse_ctx);
 339
 340    return progress;
 341 }
 342
 343 bool
 344 fs_visitor::opt_cse()
 345 {
 346    bool progress = false;
 347
 348    calculate_live_intervals();
 349
 350    foreach_block (block, cfg) {
 351       progress = opt_cse_local(block) || progress;
 352    }
 353
 354    if (progress)
 355       invalidate_live_intervals();
 356
 357    return progress;
 358 }