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_expression(const fs_inst *const inst)
  47 {
  48    switch (inst->opcode) {
  49    case BRW_OPCODE_SEL:
  50    case BRW_OPCODE_NOT:
  51    case BRW_OPCODE_AND:
  52    case BRW_OPCODE_OR:
  53    case BRW_OPCODE_XOR:
  54    case BRW_OPCODE_SHR:
  55    case BRW_OPCODE_SHL:
  56    case BRW_OPCODE_ASR:
  57    case BRW_OPCODE_CMP:
  58    case BRW_OPCODE_CMPN:
  59    case BRW_OPCODE_ADD:
  60    case BRW_OPCODE_MUL:
  61    case BRW_OPCODE_FRC:
  62    case BRW_OPCODE_RNDU:
  63    case BRW_OPCODE_RNDD:
  64    case BRW_OPCODE_RNDE:
  65    case BRW_OPCODE_RNDZ:
  66    case BRW_OPCODE_LINE:
  67    case BRW_OPCODE_PLN:
  68    case BRW_OPCODE_MAD:
  69    case BRW_OPCODE_LRP:
  70    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
  71    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
  72    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD:
  73    case FS_OPCODE_CINTERP:
  74    case FS_OPCODE_LINTERP:
  75       return true;
  76    case SHADER_OPCODE_RCP:
  77    case SHADER_OPCODE_RSQ:
  78    case SHADER_OPCODE_SQRT:
  79    case SHADER_OPCODE_EXP2:
  80    case SHADER_OPCODE_LOG2:
  81    case SHADER_OPCODE_POW:
  82    case SHADER_OPCODE_INT_QUOTIENT:
  83    case SHADER_OPCODE_INT_REMAINDER:
  84    case SHADER_OPCODE_SIN:
  85    case SHADER_OPCODE_COS:
  86       return inst->mlen == 0;
  87    default:
  88       return false;
  89    }
  90 }
  91
  92 static bool
  93 is_expression_commutative(enum opcode op)
  94 {
  95    switch (op) {
  96    case BRW_OPCODE_AND:
  97    case BRW_OPCODE_OR:
  98    case BRW_OPCODE_XOR:
  99    case BRW_OPCODE_ADD:
 100    case BRW_OPCODE_MUL:
 101       return true;
 102    default:
 103       return false;
 104    }
 105 }
 106
 107 static bool
 108 operands_match(fs_inst *a, fs_inst *b)
 109 {
 110    fs_reg *xs = a->src;
 111    fs_reg *ys = b->src;
 112
 113    if (!is_expression_commutative(a->opcode)) {
 114       bool match = true;
 115       for (int i = 0; i < a->sources; i++) {
 116          if (!xs[i].equals(ys[i])) {
 117             match = false;
 118             break;
 119          }
 120       }
 121       return match;
 122    } else {
 123       return (xs[0].equals(ys[0]) && xs[1].equals(ys[1])) ||
 124              (xs[1].equals(ys[0]) && xs[0].equals(ys[1]));
 125    }
 126 }
 127
 128 static bool
 129 instructions_match(fs_inst *a, fs_inst *b)
 130 {
 131    return a->opcode == b->opcode &&
 132           a->saturate == b->saturate &&
 133           a->predicate == b->predicate &&
 134           a->predicate_inverse == b->predicate_inverse &&
 135           a->conditional_mod == b->conditional_mod &&
 136           a->dst.type == b->dst.type &&
 137           a->sources == b->sources &&
 138           operands_match(a, b);
 139 }
 140
 141 bool
 142 fs_visitor::opt_cse_local(bblock_t *block, exec_list *aeb)
 143 {
 144    bool progress = false;
 145
 146    void *cse_ctx = ralloc_context(NULL);
 147
 148    int ip = block->start_ip;
 149    for (fs_inst *inst = (fs_inst *)block->start;
 150         inst != block->end->next;
 151         inst = (fs_inst *) inst->next) {
 152
 153       /* Skip some cases. */
 154       if (is_expression(inst) && !inst->is_partial_write() &&
 155           (inst->dst.file != HW_REG || inst->dst.is_null()))
 156       {
 157          bool found = false;
 158
 159          aeb_entry *entry;
 160          foreach_list(entry_node, aeb) {
 161             entry = (aeb_entry *) entry_node;
 162
 163             /* Match current instruction's expression against those in AEB. */
 164             if (instructions_match(inst, entry->generator)) {
 165                found = true;
 166                progress = true;
 167                break;
 168             }
 169          }
 170
 171          if (!found) {
 172             /* Our first sighting of this expression.  Create an entry. */
 173             aeb_entry *entry = ralloc(cse_ctx, aeb_entry);
 174             entry->tmp = reg_undef;
 175             entry->generator = inst;
 176             aeb->push_tail(entry);
 177          } else {
 178             /* This is at least our second sighting of this expression.
 179              * If we don't have a temporary already, make one.
 180              */
 181             bool no_existing_temp = entry->tmp.file == BAD_FILE;
 182             if (no_existing_temp && !entry->generator->dst.is_null()) {
 183                int written = entry->generator->regs_written;
 184
 185                fs_reg orig_dst = entry->generator->dst;
 186                fs_reg tmp = fs_reg(GRF, virtual_grf_alloc(written),
 187                                    orig_dst.type);
 188                entry->tmp = tmp;
 189                entry->generator->dst = tmp;
 190
 191                for (int i = 0; i < written; i++) {
 192                   fs_inst *copy = MOV(orig_dst, tmp);
 193                   copy->force_writemask_all =
 194                      entry->generator->force_writemask_all;
 195                   entry->generator->insert_after(copy);
 196
 197                   orig_dst.reg_offset++;
 198                   tmp.reg_offset++;
 199                }
 200             }
 201
 202             /* dest <- temp */
 203             if (!inst->dst.is_null()) {
 204                int written = inst->regs_written;
 205                assert(written == entry->generator->regs_written);
 206                assert(inst->dst.type == entry->tmp.type);
 207                fs_reg dst = inst->dst;
 208                fs_reg tmp = entry->tmp;
 209                fs_inst *copy = NULL;
 210                for (int i = 0; i < written; i++) {
 211                   copy = MOV(dst, tmp);
 212                   copy->force_writemask_all = inst->force_writemask_all;
 213                   inst->insert_before(copy);
 214
 215                   dst.reg_offset++;
 216                   tmp.reg_offset++;
 217                }
 218             }
 219
 220             /* Set our iterator so that next time through the loop inst->next
 221              * will get the instruction in the basic block after the one we've
 222              * removed.
 223              */
 224             fs_inst *prev = (fs_inst *)inst->prev;
 225
 226             inst->remove();
 227
 228             /* Appending an instruction may have changed our bblock end. */
 229             if (inst == block->end) {
 230                block->end = prev;
 231             }
 232
 233             inst = prev;
 234          }
 235       }
 236
 237       foreach_list_safe(entry_node, aeb) {
 238          aeb_entry *entry = (aeb_entry *)entry_node;
 239
 240          /* Kill all AEB entries that write a different value to or read from
 241           * the flag register if we just wrote it.
 242           */
 243          if (inst->writes_flag()) {
 244             if (entry->generator->reads_flag() ||
 245                 (entry->generator->writes_flag() &&
 246                  !instructions_match(inst, entry->generator))) {
 247                entry->remove();
 248                ralloc_free(entry);
 249                continue;
 250             }
 251          }
 252
 253          for (int i = 0; i < entry->generator->sources; i++) {
 254             fs_reg *src_reg = &entry->generator->src[i];
 255
 256             /* Kill all AEB entries that use the destination we just
 257              * overwrote.
 258              */
 259             if (inst->overwrites_reg(entry->generator->src[i])) {
 260                entry->remove();
 261                ralloc_free(entry);
 262                break;
 263             }
 264
 265             /* Kill any AEB entries using registers that don't get reused any
 266              * more -- a sure sign they'll fail operands_match().
 267              */
 268             if (src_reg->file == GRF && virtual_grf_end[src_reg->reg] < ip) {
 269                entry->remove();
 270                ralloc_free(entry);
 271                break;
 272             }
 273          }
 274       }
 275
 276       ip++;
 277    }
 278
 279    ralloc_free(cse_ctx);
 280
 281    if (progress)
 282       invalidate_live_intervals();
 283
 284    return progress;
 285 }
 286
 287 bool
 288 fs_visitor::opt_cse()
 289 {
 290    bool progress = false;
 291
 292    calculate_live_intervals();
 293
 294    cfg_t cfg(&instructions);
 295
 296    for (int b = 0; b < cfg.num_blocks; b++) {
 297       bblock_t *block = cfg.blocks[b];
 298       exec_list aeb;
 299
 300       progress = opt_cse_local(block, &aeb) || progress;
 301    }
 302
 303    return progress;
 304 }