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 QOP_SFs have been
  49          * seen before this instruction, or if it depends on r4, how many r4
  50          * writes have been seen.
  51          */
  52         uint32_t implicit_arg_update_count;
  53 };
  54
  55 static bool
  56 inst_key_equals(const void *a, const void *b)
  57 {
  58         const struct inst_key *key_a = a;
  59         const struct inst_key *key_b = b;
  60
  61         return memcmp(key_a, key_b, sizeof(*key_a)) == 0;
  62 }
  63
  64 static struct qinst *
  65 vc4_find_cse(struct hash_table *ht, struct qinst *inst, uint32_t sf_count,
  66              uint32_t r4_count)
  67 {
  68         if (inst->dst.file != QFILE_TEMP ||
  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         if (qir_reads_r4(inst))
  82                 key.implicit_arg_update_count = r4_count;
  83
  84         uint32_t hash = _mesa_hash_data(&key, sizeof(key));
  85         struct hash_entry *entry =
  86                 _mesa_hash_table_search(ht, hash, &key);
  87
  88         if (entry) {
  89                 if (debug) {
  90                         fprintf(stderr, "CSE found match:\n");
  91
  92                         fprintf(stderr, "  Original inst: ");
  93                         qir_dump_inst(entry->data);
  94                         fprintf(stderr, "\n");
  95
  96                         fprintf(stderr, "  Our inst:      ");
  97                         qir_dump_inst(inst);
  98                         fprintf(stderr, "\n");
  99                 }
 100
 101                 return entry->data;
 102         }
 103
 104         struct inst_key *alloc_key = ralloc(ht, struct inst_key);
 105         if (!alloc_key)
 106                 return NULL;
 107         memcpy(alloc_key, &key, sizeof(*alloc_key));
 108         _mesa_hash_table_insert(ht, hash, alloc_key, inst);
 109
 110         if (debug) {
 111                 fprintf(stderr, "Added to CSE HT: ");
 112                 qir_dump_inst(inst);
 113                 fprintf(stderr, "\n");
 114         }
 115
 116         return NULL;
 117 }
 118
 119 bool
 120 qir_opt_cse(struct vc4_compile *c)
 121 {
 122         bool progress = false;
 123         struct simple_node *node, *t;
 124         struct qinst *last_sf = NULL;
 125         uint32_t sf_count = 0, r4_count = 0;
 126
 127         struct hash_table *ht = _mesa_hash_table_create(NULL, inst_key_equals);
 128         if (!ht)
 129                 return false;
 130
 131         foreach_s(node, t, &c->instructions) {
 132                 struct qinst *inst = (struct qinst *)node;
 133
 134                 if (qir_has_side_effects(inst)) {
 135                         if (inst->op == QOP_TLB_DISCARD_SETUP)
 136                                 last_sf = NULL;
 137                         continue;
 138                 }
 139
 140                 if (inst->op == QOP_SF) {
 141                         if (last_sf &&
 142                             qir_reg_equals(last_sf->src[0], inst->src[0])) {
 143                                 if (debug) {
 144                                         fprintf(stderr,
 145                                                 "Removing redundant SF: ");
 146                                         qir_dump_inst(inst);
 147                                         fprintf(stderr, "\n");
 148                                 }
 149                                 qir_remove_instruction(inst);
 150                                 progress = true;
 151                                 continue;
 152                         } else {
 153                                 last_sf = inst;
 154                                 sf_count++;
 155                         }
 156                 } else {
 157                         struct qinst *cse = vc4_find_cse(ht, inst,
 158                                                          sf_count, r4_count);
 159                         if (cse) {
 160                                 inst->src[0] = cse->dst;
 161                                 for (int i = 1; i < qir_get_op_nsrc(inst->op);
 162                                      i++)
 163                                         inst->src[i] = c->undef;
 164                                 inst->op = QOP_MOV;
 165                                 progress = true;
 166
 167                                 if (debug) {
 168                                         fprintf(stderr, "  Turned into:   ");
 169                                         qir_dump_inst(inst);
 170                                         fprintf(stderr, "\n");
 171                                 }
 172                         }
 173                 }
 174
 175                 if (qir_writes_r4(inst))
 176                         r4_count++;
 177         }
 178
 179         ralloc_free(ht);
 180
 181         return progress;
 182 }