src/compiler/nir/nir_phi_builder.c

   1 /*
   2  * Copyright © 2016 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 "nir_phi_builder.h"
  25 #include "nir/nir_vla.h"
  26
  27 struct nir_phi_builder {
  28    nir_shader *shader;
  29    nir_function_impl *impl;
  30
  31    /* Copied from the impl for easy access */
  32    unsigned num_blocks;
  33
  34    /* Array of all blocks indexed by block->index. */
  35    nir_block **blocks;
  36
  37    /* Hold on to the values so we can easily iterate over them. */
  38    struct exec_list values;
  39
  40    /* Worklist for phi adding */
  41    unsigned iter_count;
  42    unsigned *work;
  43    nir_block **W;
  44 };
  45
  46 #define NEEDS_PHI ((nir_ssa_def *)(intptr_t)-1)
  47
  48 struct nir_phi_builder_value {
  49    struct exec_node node;
  50
  51    struct nir_phi_builder *builder;
  52
  53    /* Needed so we can create phis and undefs */
  54    unsigned num_components;
  55    unsigned bit_size;
  56
  57    /* The list of phi nodes associated with this value.  Phi nodes are not
  58     * added directly.  Instead, they are created, the instr->block pointer
  59     * set, and then added to this list.  Later, in phi_builder_finish, we
  60     * set up their sources and add them to the top of their respective
  61     * blocks.
  62     */
  63    struct exec_list phis;
  64
  65    /* Array of SSA defs, indexed by block.  For each block, this array has has
  66     * one of three types of values:
  67     *
  68     *  - NULL. Indicates that there is no known definition in this block.  If
  69     *    you need to find one, look at the block's immediate dominator.
  70     *
  71     *  - NEEDS_PHI. Indicates that the block may need a phi node but none has
  72     *    been created yet.  If a def is requested for a block, a phi will need
  73     *    to be created.
  74     *
  75     *  - A regular SSA def.  This will be either the result of a phi node or
  76     *    one of the defs provided by nir_phi_builder_value_set_blocK_def().
  77     */
  78    nir_ssa_def *defs[0];
  79 };
  80
  81 struct nir_phi_builder *
  82 nir_phi_builder_create(nir_function_impl *impl)
  83 {
  84    struct nir_phi_builder *pb = rzalloc(NULL, struct nir_phi_builder);
  85
  86    pb->shader = impl->function->shader;
  87    pb->impl = impl;
  88
  89    assert(impl->valid_metadata & (nir_metadata_block_index |
  90                                   nir_metadata_dominance));
  91
  92    pb->num_blocks = impl->num_blocks;
  93    pb->blocks = ralloc_array(pb, nir_block *, pb->num_blocks);
  94    nir_foreach_block(block, impl) {
  95       pb->blocks[block->index] = block;
  96    }
  97
  98    exec_list_make_empty(&pb->values);
  99
 100    pb->iter_count = 0;
 101    pb->work = rzalloc_array(pb, unsigned, pb->num_blocks);
 102    pb->W = ralloc_array(pb, nir_block *, pb->num_blocks);
 103
 104    return pb;
 105 }
 106
 107 struct nir_phi_builder_value *
 108 nir_phi_builder_add_value(struct nir_phi_builder *pb, unsigned num_components,
 109                           unsigned bit_size, const BITSET_WORD *defs)
 110 {
 111    struct nir_phi_builder_value *val;
 112    unsigned i, w_start = 0, w_end = 0;
 113
 114    val = rzalloc_size(pb, sizeof(*val) + sizeof(val->defs[0]) * pb->num_blocks);
 115    val->builder = pb;
 116    val->num_components = num_components;
 117    val->bit_size = bit_size;
 118    exec_list_make_empty(&val->phis);
 119    exec_list_push_tail(&pb->values, &val->node);
 120
 121    pb->iter_count++;
 122
 123    BITSET_WORD tmp;
 124    BITSET_FOREACH_SET(i, tmp, defs, pb->num_blocks) {
 125       if (pb->work[i] < pb->iter_count)
 126          pb->W[w_end++] = pb->blocks[i];
 127       pb->work[i] = pb->iter_count;
 128    }
 129
 130    while (w_start != w_end) {
 131       nir_block *cur = pb->W[w_start++];
 132       struct set_entry *dom_entry;
 133       set_foreach(cur->dom_frontier, dom_entry) {
 134          nir_block *next = (nir_block *) dom_entry->key;
 135
 136          /* If there's more than one return statement, then the end block
 137           * can be a join point for some definitions. However, there are
 138           * no instructions in the end block, so nothing would use those
 139           * phi nodes. Of course, we couldn't place those phi nodes
 140           * anyways due to the restriction of having no instructions in the
 141           * end block...
 142           */
 143          if (next == pb->impl->end_block)
 144             continue;
 145
 146          if (val->defs[next->index] == NULL) {
 147             /* Instead of creating a phi node immediately, we simply set the
 148              * value to the magic value NEEDS_PHI.  Later, we create phi nodes
 149              * on demand in nir_phi_builder_value_get_block_def().
 150              */
 151             val->defs[next->index] = NEEDS_PHI;
 152
 153             if (pb->work[next->index] < pb->iter_count) {
 154                pb->work[next->index] = pb->iter_count;
 155                pb->W[w_end++] = next;
 156             }
 157          }
 158       }
 159    }
 160
 161    return val;
 162 }
 163
 164 void
 165 nir_phi_builder_value_set_block_def(struct nir_phi_builder_value *val,
 166                                     nir_block *block, nir_ssa_def *def)
 167 {
 168    val->defs[block->index] = def;
 169 }
 170
 171 nir_ssa_def *
 172 nir_phi_builder_value_get_block_def(struct nir_phi_builder_value *val,
 173                                     nir_block *block)
 174 {
 175    /* Crawl up the dominance tree and find the closest dominator for which we
 176     * have a valid ssa_def, if any.
 177     */
 178    nir_block *dom = block;
 179    while (dom && val->defs[dom->index] == NULL)
 180       dom = dom->imm_dom;
 181
 182    nir_ssa_def *def;
 183    if (dom == NULL) {
 184       /* No dominator means either that we crawled to the top without ever
 185        * finding a definition or that this block is unreachable.  In either
 186        * case, the value is undefined so we need an SSA undef.
 187        */
 188       nir_ssa_undef_instr *undef =
 189          nir_ssa_undef_instr_create(val->builder->shader,
 190                                     val->num_components,
 191                                     val->bit_size);
 192       nir_instr_insert(nir_before_cf_list(&val->builder->impl->body),
 193                        &undef->instr);
 194       def = &undef->def;
 195    } else if (val->defs[dom->index] == NEEDS_PHI) {
 196       /* The magic value NEEDS_PHI indicates that the block needs a phi node
 197        * but none has been created.  We need to create one now so we can
 198        * return it to the caller.
 199        *
 200        * Because a phi node may use SSA defs that it does not dominate (this
 201        * happens in loops), we do not yet have enough information to fully
 202        * fill out the phi node.  Instead, the phi nodes we create here will be
 203        * empty (have no sources) and won't actually be placed in the block's
 204        * instruction list yet.  Later, in nir_phi_builder_finish(), we walk
 205        * over all of the phi instructions, fill out the sources lists, and
 206        * place them at the top of their respective block's instruction list.
 207        *
 208        * Creating phi nodes on-demand allows us to avoid creating dead phi
 209        * nodes that will just get deleted later. While this probably isn't a
 210        * big win for a full into-SSA pass, other users may use the phi builder
 211        * to make small SSA form repairs where most of the phi nodes will never
 212        * be used.
 213        */
 214       nir_phi_instr *phi = nir_phi_instr_create(val->builder->shader);
 215       nir_ssa_dest_init(&phi->instr, &phi->dest, val->num_components,
 216                         val->bit_size, NULL);
 217       phi->instr.block = dom;
 218       exec_list_push_tail(&val->phis, &phi->instr.node);
 219       def = val->defs[dom->index] = &phi->dest.ssa;
 220    } else {
 221       /* In this case, we have an actual SSA def.  It's either the result of a
 222        * phi node created by the case above or one passed to us through
 223        * nir_phi_builder_value_set_block_def().
 224        */
 225       def = val->defs[dom->index];
 226    }
 227
 228    /* Walk the chain and stash the def in all of the applicable blocks.  We do
 229     * this for two reasons:
 230     *
 231     *  1) To speed up lookup next time even if the next time is called from a
 232     *     block that is not dominated by this one.
 233     *  2) To avoid unneeded recreation of phi nodes and undefs.
 234     */
 235    for (dom = block; dom && val->defs[dom->index] == NULL; dom = dom->imm_dom)
 236       val->defs[dom->index] = def;
 237
 238    return def;
 239 }
 240
 241 static int
 242 compare_blocks(const void *_a, const void *_b)
 243 {
 244    const nir_block * const * a = _a;
 245    const nir_block * const * b = _b;
 246
 247    return (*a)->index - (*b)->index;
 248 }
 249
 250 void
 251 nir_phi_builder_finish(struct nir_phi_builder *pb)
 252 {
 253    const unsigned num_blocks = pb->num_blocks;
 254    NIR_VLA(nir_block *, preds, num_blocks);
 255
 256    foreach_list_typed(struct nir_phi_builder_value, val, node, &pb->values) {
 257       /* We treat the linked list of phi nodes like a worklist.  The list is
 258        * pre-populated by calls to nir_phi_builder_value_get_block_def() that
 259        * create phi nodes.  As we fill in the sources of phi nodes, more may
 260        * be created and are added to the end of the list.
 261        *
 262        * Because we are adding and removing phi nodes from the list as we go,
 263        * we can't iterate over it normally.  Instead, we just iterate until
 264        * the list is empty.
 265        */
 266       while (!exec_list_is_empty(&val->phis)) {
 267          struct exec_node *head = exec_list_get_head(&val->phis);
 268          nir_phi_instr *phi = exec_node_data(nir_phi_instr, head, instr.node);
 269          assert(phi->instr.type == nir_instr_type_phi);
 270
 271          exec_node_remove(&phi->instr.node);
 272
 273          /* Construct an array of predecessors.  We sort it to ensure
 274           * determinism in the phi insertion algorithm.
 275           *
 276           * XXX: Calling qsort this many times seems expensive.
 277           */
 278          int num_preds = 0;
 279          struct set_entry *entry;
 280          set_foreach(phi->instr.block->predecessors, entry)
 281             preds[num_preds++] = (nir_block *)entry->key;
 282          qsort(preds, num_preds, sizeof(*preds), compare_blocks);
 283
 284          for (unsigned i = 0; i < num_preds; i++) {
 285             nir_phi_src *src = ralloc(phi, nir_phi_src);
 286             src->pred = preds[i];
 287             src->src = nir_src_for_ssa(
 288                nir_phi_builder_value_get_block_def(val, preds[i]));
 289             exec_list_push_tail(&phi->srcs, &src->node);
 290          }
 291
 292          nir_instr_insert(nir_before_block(phi->instr.block), &phi->instr);
 293       }
 294    }
 295
 296    ralloc_free(pb);
 297 }