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 namespace {
  36 struct aeb_entry : public exec_node {
  37    /** The instruction that generates the expression value. */
  38    fs_inst *generator;
  39
  40    /** The temporary where the value is stored. */
  41    fs_reg tmp;
  42 };
  43 }
  44
  45 static bool
  46 is_copy_payload(const fs_inst *inst)
  47 {
  48    const int reg = inst->src[0].reg;
  49    if (inst->src[0].reg_offset != 0)
  50       return false;
  51
  52    for (int i = 1; i < inst->sources; i++) {
  53       if (inst->src[i].reg != reg ||
  54           inst->src[i].reg_offset != i) {
  55          return false;
  56       }
  57    }
  58    return true;
  59 }
  60
  61 static bool
  62 is_expression(const fs_inst *const inst)
  63 {
  64    switch (inst->opcode) {
  65    case BRW_OPCODE_MOV:
  66    case BRW_OPCODE_SEL:
  67    case BRW_OPCODE_NOT:
  68    case BRW_OPCODE_AND:
  69    case BRW_OPCODE_OR:
  70    case BRW_OPCODE_XOR:
  71    case BRW_OPCODE_SHR:
  72    case BRW_OPCODE_SHL:
  73    case BRW_OPCODE_ASR:
  74    case BRW_OPCODE_CMP:
  75    case BRW_OPCODE_CMPN:
  76    case BRW_OPCODE_ADD:
  77    case BRW_OPCODE_MUL:
  78    case BRW_OPCODE_FRC:
  79    case BRW_OPCODE_RNDU:
  80    case BRW_OPCODE_RNDD:
  81    case BRW_OPCODE_RNDE:
  82    case BRW_OPCODE_RNDZ:
  83    case BRW_OPCODE_LINE:
  84    case BRW_OPCODE_PLN:
  85    case BRW_OPCODE_MAD:
  86    case BRW_OPCODE_LRP:
  87    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
  88    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
  89    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD:
  90    case FS_OPCODE_CINTERP:
  91    case FS_OPCODE_LINTERP:
  92    case SHADER_OPCODE_FIND_LIVE_CHANNEL:
  93    case SHADER_OPCODE_BROADCAST:
  94       return true;
  95    case SHADER_OPCODE_RCP:
  96    case SHADER_OPCODE_RSQ:
  97    case SHADER_OPCODE_SQRT:
  98    case SHADER_OPCODE_EXP2:
  99    case SHADER_OPCODE_LOG2:
 100    case SHADER_OPCODE_POW:
 101    case SHADER_OPCODE_INT_QUOTIENT:
 102    case SHADER_OPCODE_INT_REMAINDER:
 103    case SHADER_OPCODE_SIN:
 104    case SHADER_OPCODE_COS:
 105       return inst->mlen < 2;
 106    case SHADER_OPCODE_LOAD_PAYLOAD:
 107       return !is_copy_payload(inst);
 108    default:
 109       return inst->is_send_from_grf() && !inst->has_side_effects();
 110    }
 111 }
 112
 113 static bool
 114 operands_match(const fs_inst *a, const fs_inst *b, bool *negate)
 115 {
 116    fs_reg *xs = a->src;
 117    fs_reg *ys = b->src;
 118
 119    if (a->opcode == BRW_OPCODE_MAD) {
 120       return xs[0].equals(ys[0]) &&
 121              ((xs[1].equals(ys[1]) && xs[2].equals(ys[2])) ||
 122               (xs[2].equals(ys[1]) && xs[1].equals(ys[2])));
 123    } else if (a->opcode == BRW_OPCODE_MUL && a->dst.type == BRW_REGISTER_TYPE_F) {
 124       bool xs0_negate = xs[0].negate;
 125       bool xs1_negate = xs[1].file == IMM ? xs[1].fixed_hw_reg.dw1.f < 0.0f
 126                                           : xs[1].negate;
 127       bool ys0_negate = ys[0].negate;
 128       bool ys1_negate = ys[1].file == IMM ? ys[1].fixed_hw_reg.dw1.f < 0.0f
 129                                           : ys[1].negate;
 130       float xs1_imm = xs[1].fixed_hw_reg.dw1.f;
 131       float ys1_imm = ys[1].fixed_hw_reg.dw1.f;
 132
 133       xs[0].negate = false;
 134       xs[1].negate = false;
 135       ys[0].negate = false;
 136       ys[1].negate = false;
 137       xs[1].fixed_hw_reg.dw1.f = fabsf(xs[1].fixed_hw_reg.dw1.f);
 138       ys[1].fixed_hw_reg.dw1.f = fabsf(ys[1].fixed_hw_reg.dw1.f);
 139
 140       bool ret = (xs[0].equals(ys[0]) && xs[1].equals(ys[1])) ||
 141                  (xs[1].equals(ys[0]) && xs[0].equals(ys[1]));
 142
 143       xs[0].negate = xs0_negate;
 144       xs[1].negate = xs[1].file == IMM ? false : xs1_negate;
 145       ys[0].negate = ys0_negate;
 146       ys[1].negate = ys[1].file == IMM ? false : ys1_negate;
 147       xs[1].fixed_hw_reg.dw1.f = xs1_imm;
 148       ys[1].fixed_hw_reg.dw1.f = ys1_imm;
 149
 150       *negate = (xs0_negate != xs1_negate) != (ys0_negate != ys1_negate);
 151       return ret;
 152    } else if (!a->is_commutative()) {
 153       bool match = true;
 154       for (int i = 0; i < a->sources; i++) {
 155          if (!xs[i].equals(ys[i])) {
 156             match = false;
 157             break;
 158          }
 159       }
 160       return match;
 161    } else {
 162       return (xs[0].equals(ys[0]) && xs[1].equals(ys[1])) ||
 163              (xs[1].equals(ys[0]) && xs[0].equals(ys[1]));
 164    }
 165 }
 166
 167 static bool
 168 instructions_match(fs_inst *a, fs_inst *b, bool *negate)
 169 {
 170    return a->opcode == b->opcode &&
 171           a->saturate == b->saturate &&
 172           a->predicate == b->predicate &&
 173           a->predicate_inverse == b->predicate_inverse &&
 174           a->conditional_mod == b->conditional_mod &&
 175           a->dst.type == b->dst.type &&
 176           a->sources == b->sources &&
 177           (a->is_tex() ? (a->offset == b->offset &&
 178                           a->mlen == b->mlen &&
 179                           a->regs_written == b->regs_written &&
 180                           a->base_mrf == b->base_mrf &&
 181                           a->eot == b->eot &&
 182                           a->header_present == b->header_present &&
 183                           a->shadow_compare == b->shadow_compare)
 184                        : true) &&
 185           operands_match(a, b, negate);
 186 }
 187
 188 static fs_inst *
 189 create_copy_instr(fs_visitor *v, fs_inst *inst, fs_reg src, bool negate)
 190 {
 191    int written = inst->regs_written;
 192    int dst_width = inst->dst.width / 8;
 193    fs_reg dst = inst->dst;
 194    fs_inst *copy;
 195
 196    if (written > dst_width) {
 197       fs_reg *sources = ralloc_array(v->mem_ctx, fs_reg, written / dst_width);
 198       for (int i = 0; i < written / dst_width; i++)
 199          sources[i] = offset(src, i);
 200       copy = v->LOAD_PAYLOAD(dst, sources, written / dst_width);
 201    } else {
 202       copy = v->MOV(dst, src);
 203       copy->force_writemask_all = inst->force_writemask_all;
 204       copy->src[0].negate = negate;
 205    }
 206    assert(copy->regs_written == written);
 207
 208    return copy;
 209 }
 210
 211 bool
 212 fs_visitor::opt_cse_local(bblock_t *block)
 213 {
 214    bool progress = false;
 215    exec_list aeb;
 216
 217    void *cse_ctx = ralloc_context(NULL);
 218
 219    int ip = block->start_ip;
 220    foreach_inst_in_block(fs_inst, inst, block) {
 221       /* Skip some cases. */
 222       if (is_expression(inst) && !inst->is_partial_write() &&
 223           (inst->dst.file != HW_REG || inst->dst.is_null()))
 224       {
 225          bool found = false;
 226          bool negate = false;
 227
 228          foreach_in_list_use_after(aeb_entry, entry, &aeb) {
 229             /* Match current instruction's expression against those in AEB. */
 230             if (!(entry->generator->dst.is_null() && !inst->dst.is_null()) &&
 231                 instructions_match(inst, entry->generator, &negate)) {
 232                found = true;
 233                progress = true;
 234                break;
 235             }
 236          }
 237
 238          if (!found) {
 239             if (inst->opcode != BRW_OPCODE_MOV ||
 240                 (inst->opcode == BRW_OPCODE_MOV &&
 241                  inst->src[0].file == IMM &&
 242                  inst->src[0].type == BRW_REGISTER_TYPE_VF)) {
 243                /* Our first sighting of this expression.  Create an entry. */
 244                aeb_entry *entry = ralloc(cse_ctx, aeb_entry);
 245                entry->tmp = reg_undef;
 246                entry->generator = inst;
 247                aeb.push_tail(entry);
 248             }
 249          } else {
 250             /* This is at least our second sighting of this expression.
 251              * If we don't have a temporary already, make one.
 252              */
 253             bool no_existing_temp = entry->tmp.file == BAD_FILE;
 254             if (no_existing_temp && !entry->generator->dst.is_null()) {
 255                int written = entry->generator->regs_written;
 256                assert((written * 8) % entry->generator->dst.width == 0);
 257
 258                entry->tmp = fs_reg(GRF, alloc.allocate(written),
 259                                    entry->generator->dst.type,
 260                                    entry->generator->dst.width);
 261
 262                fs_inst *copy = create_copy_instr(this, entry->generator,
 263                                                  entry->tmp, false);
 264                entry->generator->insert_after(block, copy);
 265
 266                entry->generator->dst = entry->tmp;
 267             }
 268
 269             /* dest <- temp */
 270             if (!inst->dst.is_null()) {
 271                assert(inst->regs_written == entry->generator->regs_written);
 272                assert(inst->dst.width == entry->generator->dst.width);
 273                assert(inst->dst.type == entry->tmp.type);
 274
 275                fs_inst *copy = create_copy_instr(this, inst,
 276                                                  entry->tmp, negate);
 277                inst->insert_before(block, copy);
 278             }
 279
 280             /* Set our iterator so that next time through the loop inst->next
 281              * will get the instruction in the basic block after the one we've
 282              * removed.
 283              */
 284             fs_inst *prev = (fs_inst *)inst->prev;
 285
 286             inst->remove(block);
 287             inst = prev;
 288          }
 289       }
 290
 291       foreach_in_list_safe(aeb_entry, entry, &aeb) {
 292          /* Kill all AEB entries that write a different value to or read from
 293           * the flag register if we just wrote it.
 294           */
 295          if (inst->writes_flag()) {
 296             bool negate; /* dummy */
 297             if (entry->generator->reads_flag() ||
 298                 (entry->generator->writes_flag() &&
 299                  !instructions_match(inst, entry->generator, &negate))) {
 300                entry->remove();
 301                ralloc_free(entry);
 302                continue;
 303             }
 304          }
 305
 306          for (int i = 0; i < entry->generator->sources; i++) {
 307             fs_reg *src_reg = &entry->generator->src[i];
 308
 309             /* Kill all AEB entries that use the destination we just
 310              * overwrote.
 311              */
 312             if (inst->overwrites_reg(entry->generator->src[i])) {
 313                entry->remove();
 314                ralloc_free(entry);
 315                break;
 316             }
 317
 318             /* Kill any AEB entries using registers that don't get reused any
 319              * more -- a sure sign they'll fail operands_match().
 320              */
 321             if (src_reg->file == GRF && virtual_grf_end[src_reg->reg] < ip) {
 322                entry->remove();
 323                ralloc_free(entry);
 324                break;
 325             }
 326          }
 327       }
 328
 329       ip++;
 330    }
 331
 332    ralloc_free(cse_ctx);
 333
 334    return progress;
 335 }
 336
 337 bool
 338 fs_visitor::opt_cse()
 339 {
 340    bool progress = false;
 341
 342    calculate_live_intervals();
 343
 344    foreach_block (block, cfg) {
 345       progress = opt_cse_local(block) || progress;
 346    }
 347
 348    if (progress)
 349       invalidate_live_intervals();
 350
 351    return progress;
 352 }