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 static bool
  82 fill_block_array(nir_block *block, void *void_data)
  83 {
  84    nir_block **blocks = void_data;
  85    blocks[block->index] = block;
  86    return true;
  87 }
  88
  89 struct nir_phi_builder *
  90 nir_phi_builder_create(nir_function_impl *impl)
  91 {
  92    struct nir_phi_builder *pb = ralloc(NULL, struct nir_phi_builder);
  93
  94    pb->shader = impl->function->shader;
  95    pb->impl = impl;
  96
  97    assert(impl->valid_metadata & (nir_metadata_block_index |
  98                                   nir_metadata_dominance));
  99
 100    pb->num_blocks = impl->num_blocks;
 101    pb->blocks = ralloc_array(pb, nir_block *, pb->num_blocks);
 102    nir_foreach_block(impl, fill_block_array, pb->blocks);
 103
 104    exec_list_make_empty(&pb->values);
 105
 106    pb->iter_count = 0;
 107    pb->work = rzalloc_array(pb, unsigned, pb->num_blocks);
 108    pb->W = ralloc_array(pb, nir_block *, pb->num_blocks);
 109
 110    return pb;
 111 }
 112
 113 struct nir_phi_builder_value *
 114 nir_phi_builder_add_value(struct nir_phi_builder *pb, unsigned num_components,
 115                           unsigned bit_size, const BITSET_WORD *defs)
 116 {
 117    struct nir_phi_builder_value *val;
 118    unsigned i, w_start = 0, w_end = 0;
 119
 120    val = rzalloc_size(pb, sizeof(*val) + sizeof(val->defs[0]) * pb->num_blocks);
 121    val->builder = pb;
 122    val->num_components = num_components;
 123    val->bit_size = bit_size;
 124    exec_list_make_empty(&val->phis);
 125    exec_list_push_tail(&pb->values, &val->node);
 126
 127    pb->iter_count++;
 128
 129    BITSET_WORD tmp;
 130    BITSET_FOREACH_SET(i, tmp, defs, pb->num_blocks) {
 131       if (pb->work[i] < pb->iter_count)
 132          pb->W[w_end++] = pb->blocks[i];
 133       pb->work[i] = pb->iter_count;
 134    }
 135
 136    while (w_start != w_end) {
 137       nir_block *cur = pb->W[w_start++];
 138       struct set_entry *dom_entry;
 139       set_foreach(cur->dom_frontier, dom_entry) {
 140          nir_block *next = (nir_block *) dom_entry->key;
 141
 142          /* If there's more than one return statement, then the end block
 143           * can be a join point for some definitions. However, there are
 144           * no instructions in the end block, so nothing would use those
 145           * phi nodes. Of course, we couldn't place those phi nodes
 146           * anyways due to the restriction of having no instructions in the
 147           * end block...
 148           */
 149          if (next == pb->impl->end_block)
 150             continue;
 151
 152          if (val->defs[next->index] == NULL) {
 153             /* Instead of creating a phi node immediately, we simply set the
 154              * value to the magic value NEEDS_PHI.  Later, we create phi nodes
 155              * on demand in nir_phi_builder_value_get_block_def().
 156              */
 157             val->defs[next->index] = NEEDS_PHI;
 158
 159             if (pb->work[next->index] < pb->iter_count) {
 160                pb->work[next->index] = pb->iter_count;
 161                pb->W[w_end++] = next;
 162             }
 163          }
 164       }
 165    }
 166
 167    return val;
 168 }
 169
 170 void
 171 nir_phi_builder_value_set_block_def(struct nir_phi_builder_value *val,
 172                                     nir_block *block, nir_ssa_def *def)
 173 {
 174    val->defs[block->index] = def;
 175 }
 176
 177 nir_ssa_def *
 178 nir_phi_builder_value_get_block_def(struct nir_phi_builder_value *val,
 179                                     nir_block *block)
 180 {
 181    /* For each block, we have one of three types of values */
 182    if (val->defs[block->index] == NULL) {
 183       /* NULL indicates that we have no SSA def for this block. */
 184       if (block->imm_dom) {
 185          /* Grab it from our immediate dominator.  We'll stash it here for
 186           * easy access later.
 187           */
 188          val->defs[block->index] =
 189             nir_phi_builder_value_get_block_def(val, block->imm_dom);
 190          return val->defs[block->index];
 191       } else {
 192          /* No immediate dominator means that this block is either the
 193           * start block or unreachable.  In either case, the value is
 194           * undefined so we need an SSA undef.
 195           */
 196          nir_ssa_undef_instr *undef =
 197             nir_ssa_undef_instr_create(val->builder->shader,
 198                                        val->num_components);
 199          nir_instr_insert(nir_before_cf_list(&val->builder->impl->body),
 200                           &undef->instr);
 201          val->defs[block->index] = &undef->def;
 202          return &undef->def;
 203       }
 204    } else if (val->defs[block->index] == NEEDS_PHI) {
 205       /* The magic value NEEDS_PHI indicates that the block needs a phi node
 206        * but none has been created.  We need to create one now so we can
 207        * return it to the caller.
 208        *
 209        * Because a phi node may use SSA defs that it does not dominate (this
 210        * happens in loops), we do not yet have enough information to fully
 211        * fill out the phi node.  Instead, the phi nodes we create here will be
 212        * empty (have no sources) and won't actually be placed in the block's
 213        * instruction list yet.  Later, in nir_phi_builder_finish(), we walk
 214        * over all of the phi instructions, fill out the sources lists, and
 215        * place them at the top of their respective block's instruction list.
 216        *
 217        * Creating phi nodes on-demand allows us to avoid creating dead phi
 218        * nodes that will just get deleted later. While this probably isn't a
 219        * big win for a full into-SSA pass, other users may use the phi builder
 220        * to make small SSA form repairs where most of the phi nodes will never
 221        * be used.
 222        */
 223       nir_phi_instr *phi = nir_phi_instr_create(val->builder->shader);
 224       nir_ssa_dest_init(&phi->instr, &phi->dest, val->num_components,
 225                         val->bit_size, NULL);
 226       phi->instr.block = block;
 227       exec_list_push_tail(&val->phis, &phi->instr.node);
 228       val->defs[block->index] = &phi->dest.ssa;
 229       return &phi->dest.ssa;
 230    } else {
 231       /* In this case, we have an actual SSA def.  It's either the result of a
 232        * phi node created by the case above or one passed to us through
 233        * nir_phi_builder_value_set_block_def().
 234        */
 235       return val->defs[block->index];
 236    }
 237 }
 238
 239 static int
 240 compare_blocks(const void *_a, const void *_b)
 241 {
 242    nir_block * const * a = _a;
 243    nir_block * const * b = _b;
 244
 245    return (*a)->index - (*b)->index;
 246 }
 247
 248 void
 249 nir_phi_builder_finish(struct nir_phi_builder *pb)
 250 {
 251    const unsigned num_blocks = pb->num_blocks;
 252    NIR_VLA(nir_block *, preds, num_blocks);
 253
 254    foreach_list_typed(struct nir_phi_builder_value, val, node, &pb->values) {
 255       /* We treat the linked list of phi nodes like a worklist.  The list is
 256        * pre-populated by calls to nir_phi_builder_value_get_block_def() that
 257        * create phi nodes.  As we fill in the sources of phi nodes, more may
 258        * be created and are added to the end of the list.
 259        *
 260        * Because we are adding and removing phi nodes from the list as we go,
 261        * we can't iterate over it normally.  Instead, we just iterate until
 262        * the list is empty.
 263        */
 264       while (!exec_list_is_empty(&val->phis)) {
 265          struct exec_node *head = exec_list_get_head(&val->phis);
 266          nir_phi_instr *phi = exec_node_data(nir_phi_instr, head, instr.node);
 267          assert(phi->instr.type == nir_instr_type_phi);
 268
 269          exec_node_remove(&phi->instr.node);
 270
 271          /* Construct an array of predecessors.  We sort it to ensure
 272           * determinism in the phi insertion algorithm.
 273           *
 274           * XXX: Calling qsort this many times seems expensive.
 275           */
 276          int num_preds = 0;
 277          struct set_entry *entry;
 278          set_foreach(phi->instr.block->predecessors, entry)
 279             preds[num_preds++] = (nir_block *)entry->key;
 280          qsort(preds, num_preds, sizeof(*preds), compare_blocks);
 281
 282          for (unsigned i = 0; i < num_preds; i++) {
 283             nir_phi_src *src = ralloc(phi, nir_phi_src);
 284             src->pred = preds[i];
 285             src->src = nir_src_for_ssa(
 286                nir_phi_builder_value_get_block_def(val, preds[i]));
 287             exec_list_push_tail(&phi->srcs, &src->node);
 288          }
 289
 290          nir_instr_insert(nir_before_block(phi->instr.block), &phi->instr);
 291       }
 292    }
 293
 294    ralloc_free(pb);
 295 }