src/gallium/drivers/vc4/vc4_opt_cse.c

   1 /*
   2  * Copyright © 2014 Broadcom
   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 /**
  25  * @file vc4_opt_cse.c
  26  *
  27  * Implements CSE for QIR without control flow.
  28  *
  29  * For each operation that writes a destination (and isn't just a MOV), put it
  30  * in the hash table of all instructions that do so.  When faced with another
  31  * one, look it up in the hash table by its opcode and operands.  If there's
  32  * an entry in the table, then just reuse the entry's destination as the
  33  * source of a MOV instead of reproducing the computation.  That MOV will then
  34  * get cleaned up by copy propagation.
  35  */
  36
  37 #include "vc4_qir.h"
  38
  39 #include "util/hash_table.h"
  40 #include "util/ralloc.h"
  41
  42 static bool debug;
  43
  44 struct inst_key {
  45         enum qop op;
  46         struct qreg src[4];
  47         /**
  48          * If the instruction depends on the flags, how many SFs have been
  49          * seen before this instruction.
  50          */
  51         uint32_t implicit_arg_update_count;
  52 };
  53
  54 static bool
  55 inst_key_equals(const void *a, const void *b)
  56 {
  57         const struct inst_key *key_a = a;
  58         const struct inst_key *key_b = b;
  59
  60         return memcmp(key_a, key_b, sizeof(*key_a)) == 0;
  61 }
  62
  63 static struct qinst *
  64 vc4_find_cse(struct vc4_compile *c, struct hash_table *ht,
  65              struct qinst *inst, uint32_t sf_count)
  66 {
  67         if (inst->dst.file != QFILE_TEMP ||
  68             !c->defs[inst->dst.index] ||
  69             inst->op == QOP_MOV ||
  70             qir_get_op_nsrc(inst->op) > 4) {
  71                 return NULL;
  72         }
  73
  74         struct inst_key key;
  75         memset(&key, 0, sizeof(key));
  76         key.op = inst->op;
  77         memcpy(key.src, inst->src,
  78                qir_get_op_nsrc(inst->op) * sizeof(key.src[0]));
  79         if (qir_depends_on_flags(inst))
  80                 key.implicit_arg_update_count = sf_count;
  81
  82         uint32_t hash = _mesa_hash_data(&key, sizeof(key));
  83         struct hash_entry *entry =
  84                 _mesa_hash_table_search_pre_hashed(ht, hash, &key);
  85
  86         if (entry) {
  87                 if (debug) {
  88                         fprintf(stderr, "CSE found match:\n");
  89
  90                         fprintf(stderr, "  Original inst: ");
  91                         qir_dump_inst(c, entry->data);
  92                         fprintf(stderr, "\n");
  93
  94                         fprintf(stderr, "  Our inst:      ");
  95                         qir_dump_inst(c, inst);
  96                         fprintf(stderr, "\n");
  97                 }
  98
  99                 return entry->data;
 100         }
 101
 102         struct inst_key *alloc_key = ralloc(ht, struct inst_key);
 103         if (!alloc_key)
 104                 return NULL;
 105         memcpy(alloc_key, &key, sizeof(*alloc_key));
 106         _mesa_hash_table_insert_pre_hashed(ht, hash, alloc_key, inst);
 107
 108         if (debug) {
 109                 fprintf(stderr, "Added to CSE HT: ");
 110                 qir_dump_inst(c, inst);
 111                 fprintf(stderr, "\n");
 112         }
 113
 114         return NULL;
 115 }
 116
 117 bool
 118 qir_opt_cse(struct vc4_compile *c)
 119 {
 120         bool progress = false;
 121         uint32_t sf_count = 0;
 122
 123         struct hash_table *ht = _mesa_hash_table_create(NULL, NULL,
 124                                                         inst_key_equals);
 125         if (!ht)
 126                 return false;
 127
 128         list_for_each_entry(struct qinst, inst, &c->instructions, link) {
 129                 if (qir_has_side_effects(c, inst) ||
 130                     qir_has_side_effect_reads(c, inst) ||
 131                     inst->op == QOP_TLB_COLOR_READ) {
 132                         continue;
 133                 }
 134
 135                 if (inst->sf) {
 136                         sf_count++;
 137                 } else {
 138                         struct qinst *cse = vc4_find_cse(c, ht, inst, sf_count);
 139                         if (cse) {
 140                                 inst->src[0] = cse->dst;
 141                                 for (int i = 1; i < qir_get_op_nsrc(inst->op);
 142                                      i++)
 143                                         inst->src[i] = c->undef;
 144                                 inst->op = QOP_MOV;
 145                                 progress = true;
 146
 147                                 if (debug) {
 148                                         fprintf(stderr, "  Turned into:   ");
 149                                         qir_dump_inst(c, inst);
 150                                         fprintf(stderr, "\n");
 151                                 }
 152                         }
 153                 }
 154         }
 155
 156         ralloc_free(ht);
 157
 158         return progress;
 159 }