intel/compiler/brw_opt_cse.cpp

/*
 * Copyright © 2012 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#define XXH_INLINE_ALL
#include "util/xxhash.h"

#include "brw_fs.h"
#include "brw_builder.h"
#include "brw_cfg.h"

/** @file
 *
 * Support for SSA-based global Common Subexpression Elimination (CSE).
 */

using namespace brw;

struct remap_entry {
   fs_inst *inst;
   bblock_t *block;
   enum brw_reg_type type;
   unsigned nr;
   bool negate;
   bool still_used;
};

static bool
is_expression(const fs_visitor *v, const fs_inst *const inst)
{
   switch (inst->opcode) {
   case BRW_OPCODE_MOV:
   case BRW_OPCODE_SEL:
   case BRW_OPCODE_NOT:
   case BRW_OPCODE_AND:
   case BRW_OPCODE_OR:
   case BRW_OPCODE_XOR:
   case BRW_OPCODE_SHR:
   case BRW_OPCODE_SHL:
   case BRW_OPCODE_ASR:
   case BRW_OPCODE_ROR:
   case BRW_OPCODE_ROL:
   case BRW_OPCODE_CMP:
   case BRW_OPCODE_CMPN:
   case BRW_OPCODE_CSEL:
   case BRW_OPCODE_BFREV:
   case BRW_OPCODE_BFE:
   case BRW_OPCODE_BFI1:
   case BRW_OPCODE_BFI2:
   case BRW_OPCODE_ADD:
   case BRW_OPCODE_MUL:
   case SHADER_OPCODE_MULH:
   case BRW_OPCODE_AVG:
   case BRW_OPCODE_FRC:
   case BRW_OPCODE_LZD:
   case BRW_OPCODE_FBH:
   case BRW_OPCODE_FBL:
   case BRW_OPCODE_CBIT:
   case BRW_OPCODE_ADD3:
   case BRW_OPCODE_RNDU:
   case BRW_OPCODE_RNDD:
   case BRW_OPCODE_RNDE:
   case BRW_OPCODE_RNDZ:
   case BRW_OPCODE_LINE:
   case BRW_OPCODE_PLN:
   case BRW_OPCODE_MAD:
   case BRW_OPCODE_LRP:
   case FS_OPCODE_FB_READ_LOGICAL:
   case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
   case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL:
   case SHADER_OPCODE_FIND_LIVE_CHANNEL:
   case SHADER_OPCODE_FIND_LAST_LIVE_CHANNEL:
   case SHADER_OPCODE_LOAD_LIVE_CHANNELS:
   case FS_OPCODE_LOAD_LIVE_CHANNELS:
   case SHADER_OPCODE_BROADCAST:
   case SHADER_OPCODE_SHUFFLE:
   case SHADER_OPCODE_QUAD_SWIZZLE:
   case SHADER_OPCODE_CLUSTER_BROADCAST:
   case SHADER_OPCODE_MOV_INDIRECT:
   case SHADER_OPCODE_TEX_LOGICAL:
   case SHADER_OPCODE_TXD_LOGICAL:
   case SHADER_OPCODE_TXF_LOGICAL:
   case SHADER_OPCODE_TXL_LOGICAL:
   case SHADER_OPCODE_TXS_LOGICAL:
   case FS_OPCODE_TXB_LOGICAL:
   case SHADER_OPCODE_TXF_CMS_W_LOGICAL:
   case SHADER_OPCODE_TXF_CMS_W_GFX12_LOGICAL:
   case SHADER_OPCODE_TXF_MCS_LOGICAL:
   case SHADER_OPCODE_LOD_LOGICAL:
   case SHADER_OPCODE_TG4_LOGICAL:
   case SHADER_OPCODE_TG4_BIAS_LOGICAL:
   case SHADER_OPCODE_TG4_EXPLICIT_LOD_LOGICAL:
   case SHADER_OPCODE_TG4_IMPLICIT_LOD_LOGICAL:
   case SHADER_OPCODE_TG4_OFFSET_LOGICAL:
   case SHADER_OPCODE_TG4_OFFSET_LOD_LOGICAL:
   case SHADER_OPCODE_TG4_OFFSET_BIAS_LOGICAL:
   case SHADER_OPCODE_SAMPLEINFO_LOGICAL:
   case SHADER_OPCODE_IMAGE_SIZE_LOGICAL:
   case SHADER_OPCODE_GET_BUFFER_SIZE:
   case FS_OPCODE_PACK:
   case FS_OPCODE_PACK_HALF_2x16_SPLIT:
   case SHADER_OPCODE_RCP:
   case SHADER_OPCODE_RSQ:
   case SHADER_OPCODE_SQRT:
   case SHADER_OPCODE_EXP2:
   case SHADER_OPCODE_LOG2:
   case SHADER_OPCODE_POW:
   case SHADER_OPCODE_INT_QUOTIENT:
   case SHADER_OPCODE_INT_REMAINDER:
   case SHADER_OPCODE_SIN:
   case SHADER_OPCODE_COS:
   case SHADER_OPCODE_LOAD_SUBGROUP_INVOCATION:
      return true;
   case SHADER_OPCODE_MEMORY_LOAD_LOGICAL:
      return inst->src[MEMORY_LOGICAL_MODE].ud == MEMORY_MODE_CONSTANT;
   case SHADER_OPCODE_LOAD_PAYLOAD:
      return !is_coalescing_payload(v->devinfo, v->alloc, inst);
   default:
      return inst->is_send_from_grf() && !inst->has_side_effects() &&
         !inst->is_volatile();
   }
}

/**
 * True if the instruction should only be CSE'd within their local block.
 */
bool
local_only(const fs_inst *inst)
{
   switch (inst->opcode) {
   case SHADER_OPCODE_FIND_LIVE_CHANNEL:
   case SHADER_OPCODE_FIND_LAST_LIVE_CHANNEL:
   case SHADER_OPCODE_LOAD_LIVE_CHANNELS:
   case FS_OPCODE_LOAD_LIVE_CHANNELS:
      /* These depend on the current channel enables, so the same opcode
       * in another block will likely return a different value.
       */
      return true;
   case BRW_OPCODE_MOV:
      /* Global CSE of MOVs is likely not worthwhile.  It can increase
       * register pressure by extending the lifetime of simple constants.
       */
      return true;
   case SHADER_OPCODE_LOAD_PAYLOAD:
      /* This is basically a MOV */
      return inst->sources == 1;
   case BRW_OPCODE_CMP:
      /* Seems to increase spilling a lot without much benefit */
      return true;
   default:
      return false;
   }
}

static bool
operands_match(const fs_inst *a, const fs_inst *b, bool *negate)
{
   brw_reg *xs = a->src;
   brw_reg *ys = b->src;

   if (a->opcode == BRW_OPCODE_MAD) {
      return xs[0].equals(ys[0]) &&
             ((xs[1].equals(ys[1]) && xs[2].equals(ys[2])) ||
              (xs[2].equals(ys[1]) && xs[1].equals(ys[2])));
   } else if (a->opcode == BRW_OPCODE_MUL && a->dst.type == BRW_TYPE_F) {
      bool xs0_negate = xs[0].negate;
      bool xs1_negate = xs[1].file == IMM ? xs[1].f < 0.0f
                                          : xs[1].negate;
      bool ys0_negate = ys[0].negate;
      bool ys1_negate = ys[1].file == IMM ? ys[1].f < 0.0f
                                          : ys[1].negate;
      float xs1_imm = xs[1].f;
      float ys1_imm = ys[1].f;

      xs[0].negate = false;
      xs[1].negate = false;
      ys[0].negate = false;
      ys[1].negate = false;
      xs[1].f = fabsf(xs[1].f);
      ys[1].f = fabsf(ys[1].f);

      bool ret = (xs[0].equals(ys[0]) && xs[1].equals(ys[1])) ||
                 (xs[1].equals(ys[0]) && xs[0].equals(ys[1]));

      xs[0].negate = xs0_negate;
      xs[1].negate = xs[1].file == IMM ? false : xs1_negate;
      ys[0].negate = ys0_negate;
      ys[1].negate = ys[1].file == IMM ? false : ys1_negate;
      xs[1].f = xs1_imm;
      ys[1].f = ys1_imm;

      *negate = (xs0_negate != xs1_negate) != (ys0_negate != ys1_negate);
      if (*negate && (a->saturate || b->saturate))
         return false;
      return ret;
   } else if (!a->is_commutative()) {
      bool match = true;
      for (int i = 0; i < a->sources; i++) {
         if (!xs[i].equals(ys[i])) {
            match = false;
            break;
         }
      }
      return match;
   } else if (a->sources == 3) {
      return (xs[0].equals(ys[0]) && xs[1].equals(ys[1]) && xs[2].equals(ys[2])) ||
             (xs[0].equals(ys[0]) && xs[1].equals(ys[2]) && xs[2].equals(ys[1])) ||
             (xs[0].equals(ys[1]) && xs[1].equals(ys[0]) && xs[2].equals(ys[2])) ||
             (xs[0].equals(ys[1]) && xs[1].equals(ys[2]) && xs[2].equals(ys[1])) ||
             (xs[0].equals(ys[2]) && xs[1].equals(ys[0]) && xs[2].equals(ys[1])) ||
             (xs[0].equals(ys[2]) && xs[1].equals(ys[1]) && xs[2].equals(ys[0]));
   } else {
      return (xs[0].equals(ys[0]) && xs[1].equals(ys[1])) ||
             (xs[1].equals(ys[0]) && xs[0].equals(ys[1]));
   }
}

static bool
instructions_match(fs_inst *a, fs_inst *b, bool *negate)
{
   return a->opcode == b->opcode &&
          a->exec_size == b->exec_size &&
          a->group == b->group &&
          a->predicate == b->predicate &&
          a->conditional_mod == b->conditional_mod &&
          a->dst.type == b->dst.type &&
          a->offset == b->offset &&
          a->mlen == b->mlen &&
          a->ex_mlen == b->ex_mlen &&
          a->sfid == b->sfid &&
          a->desc == b->desc &&
          a->ex_desc == b->ex_desc &&
          a->size_written == b->size_written &&
          a->check_tdr == b->check_tdr &&
          a->header_size == b->header_size &&
          a->target == b->target &&
          a->sources == b->sources &&
          a->bits == b->bits &&
          operands_match(a, b, negate);
}

/* -------------------------------------------------------------------- */

#define HASH(hash, data) XXH32(&(data), sizeof(data), hash)

uint32_t
hash_reg(uint32_t hash, const brw_reg &r)
{
   struct {
      uint64_t u64;
      uint32_t u32;
      uint16_t u16a;
      uint16_t u16b;
   } data = {
      .u64 = r.u64, .u32 = r.bits, .u16a = r.offset, .u16b = r.stride
   };
   STATIC_ASSERT(sizeof(data) == 16); /* ensure there's no padding */
   hash = HASH(hash, data);
   return hash;
}

static uint32_t
hash_inst(const void *v)
{
   const fs_inst *inst = static_cast<const fs_inst *>(v);
   uint32_t hash = 0;

   /* Skip dst - that would make nothing ever match */

   /* Skip ir and annotation - we don't care for equivalency purposes. */

   const uint8_t u8data[] = {
      inst->sources,
      inst->exec_size,
      inst->group,
      inst->mlen,
      inst->ex_mlen,
      inst->sfid,
      inst->header_size,
      inst->target,

      inst->conditional_mod,
      inst->predicate,
   };
   const uint32_t u32data[] = {
      inst->desc,
      inst->ex_desc,
      inst->offset,
      inst->size_written,
      inst->opcode,
      inst->bits,
   };

   hash = HASH(hash, u8data);
   hash = HASH(hash, u32data);

   /* Skip hashing sched - we shouldn't be CSE'ing after that SWSB */

   if (inst->opcode == BRW_OPCODE_MAD) {
      /* Commutatively combine the hashes for the multiplicands */
      hash = hash_reg(hash, inst->src[0]);
      uint32_t hash1 = hash_reg(hash, inst->src[1]);
      uint32_t hash2 = hash_reg(hash, inst->src[2]);
      hash = hash1 * hash2;
   } else if (inst->opcode == BRW_OPCODE_MUL &&
              inst->dst.type == BRW_TYPE_F) {
      /* Canonicalize negations on either source (or both) and commutatively
       * combine the hashes for both sources.
       */
      brw_reg src[2] = { inst->src[0], inst->src[1] };
      uint32_t src_hash[2];

      for (int i = 0; i < 2; i++) {
         src[i].negate = false;
         if (src[i].file == IMM)
            src[i].f = fabs(src[i].f);

         src_hash[i] = hash_reg(hash, src[i]);
      }

      hash = src_hash[0] * src_hash[1];
   } else if (inst->is_commutative()) {
      /* Commutatively combine the sources */
      uint32_t hash0 = hash_reg(hash, inst->src[0]);
      uint32_t hash1 = hash_reg(hash, inst->src[1]);
      uint32_t hash2 = inst->sources > 2 ? hash_reg(hash, inst->src[2]) : 1;
      hash = hash0 * hash1 * hash2;
   } else {
      /* Just hash all the sources */
      for (int i = 0; i < inst->sources; i++)
         hash = hash_reg(hash, inst->src[i]);
   }

   return hash;
}

/* -------------------------------------------------------------------- */

static bool
cmp_func(const void *data1, const void *data2)
{
   bool negate;
   return instructions_match((fs_inst *) data1, (fs_inst *) data2, &negate);
}

static bool
remap_sources(fs_visitor &s, const brw::def_analysis &defs,
              fs_inst *inst, struct remap_entry *remap_table)
{
   bool progress = false;

   for (int i = 0; i < inst->sources; i++) {
      if (inst->src[i].file == VGRF &&
          inst->src[i].nr < defs.count() &&
          remap_table[inst->src[i].nr].inst != NULL) {
         const unsigned old_nr = inst->src[i].nr;
         const unsigned new_nr = remap_table[old_nr].nr;
         const bool need_negate = remap_table[old_nr].negate;

         if (need_negate &&
             (remap_table[old_nr].type != inst->src[i].type ||
              !inst->can_do_source_mods(s.devinfo))) {
            remap_table[old_nr].still_used = true;
            continue;
         }

         inst->src[i].nr = new_nr;

         if (!inst->src[i].abs)
            inst->src[i].negate ^= need_negate;

         progress = true;
      }
   }

   return progress;
}

bool
brw_opt_cse_defs(fs_visitor &s)
{
   const intel_device_info *devinfo = s.devinfo;
   const idom_tree &idom = s.idom_analysis.require();
   const brw::def_analysis &defs = s.def_analysis.require();
   bool progress = false;
   bool need_remaps = false;

   struct remap_entry *remap_table = new remap_entry[defs.count()];
   memset(remap_table, 0, defs.count() * sizeof(struct remap_entry));
   struct set *set = _mesa_set_create(NULL, NULL, cmp_func);

   foreach_block(block, s.cfg) {
      fs_inst *last_flag_write = NULL;
      fs_inst *last = NULL;

      foreach_inst_in_block_safe(fs_inst, inst, block) {
         if (need_remaps)
            progress |= remap_sources(s, defs, inst, remap_table);

         /* Updating last_flag_written should be at the bottom of the loop,
          * but doing it this way lets us use "continue" more easily.
          */
         if (last && last->flags_written(devinfo))
            last_flag_write = last;
         last = inst;

         if (inst->dst.is_null()) {
            bool ignored;
            if (last_flag_write && !inst->writes_accumulator &&
                instructions_match(last_flag_write, inst, &ignored)) {
               /* This instruction has no destination but has a flag write
                * which is redundant with the previous flag write in our
                * basic block.  So we can simply remove it.
                */
               inst->remove(block, true);
               last = NULL;
               progress = true;
            }
         } else if (is_expression(&s, inst) && defs.get(inst->dst)) {
            assert(!inst->writes_accumulator);
            assert(!inst->reads_accumulator_implicitly());

            uint32_t hash = hash_inst(inst);
            if (inst->flags_read(devinfo)) {
               hash = last_flag_write ? HASH(hash, last_flag_write)
                                      : HASH(hash, block);
            }

            struct set_entry *e =
               _mesa_set_search_or_add_pre_hashed(set, hash, inst, NULL);
            if (!e) goto out; /* out of memory error */
            fs_inst *match = (fs_inst *) e->key;

            /* If there was no match, move on */
            if (match == inst)
               continue;

            bblock_t *def_block = defs.get_block(match->dst);
            if (block != def_block && (local_only(inst) ||
                !idom.dominates(def_block, block))) {
               /* If `match` doesn't dominate `inst` then remove it from
                * the set and add `inst` instead so future lookups see that.
                */
               e->key = inst;
               continue;
            }

            /* We can replace inst with match or negate(match). */
            bool negate = false;
            if (inst->opcode == BRW_OPCODE_MUL &&
                inst->dst.type == BRW_TYPE_F) {
               /* Determine whether inst is actually negate(match) */
               bool ops_must_match = operands_match(inst, match, &negate);
               assert(ops_must_match);
            }

            /* Some later instruction could depend on the flags written by
             * this instruction. It can only be removed if the previous
             * instruction that write the flags is identical.
             */
            if (inst->flags_written(devinfo)) {
               bool ignored;

               if (last_flag_write == NULL ||
                   !instructions_match(last_flag_write, inst, &ignored)) {
                  continue;
               }
            }

            need_remaps = true;
            remap_table[inst->dst.nr].inst = inst;
            remap_table[inst->dst.nr].block = block;
            remap_table[inst->dst.nr].type = match->dst.type;
            remap_table[inst->dst.nr].nr = match->dst.nr;
            remap_table[inst->dst.nr].negate = negate;
            remap_table[inst->dst.nr].still_used = false;
         }
      }
   }

   /* Remove instruction now unused */
   for (unsigned i = 0; i < defs.count(); i++) {
      if (!remap_table[i].inst)
         continue;

      if (!remap_table[i].still_used) {
         remap_table[i].inst->remove(remap_table[i].block, true);
         progress = true;
      }
   }

out:
   delete [] remap_table;
   _mesa_set_destroy(set, NULL);

   if (progress) {
      s.cfg->adjust_block_ips();
      s.invalidate_analysis(DEPENDENCY_INSTRUCTION_DATA_FLOW |
                            DEPENDENCY_INSTRUCTION_DETAIL);
   }

   return progress;
}

#undef HASH