src/compiler/nir/nir_opt_find_array_copies.c

   1 /*
   2  * Copyright © 2018 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.h"
  25 #include "nir_builder.h"
  26 #include "nir_deref.h"
  27
  28 static bool
  29 index_ssa_def_cb(nir_ssa_def *def, void *state)
  30 {
  31    unsigned *index = (unsigned *) state;
  32    def->index = (*index)++;
  33
  34    return true;
  35 }
  36
  37 static nir_deref_instr *
  38 get_deref_for_load_src(nir_src src, unsigned first_valid_load)
  39 {
  40    if (!src.is_ssa)
  41       return NULL;
  42
  43    if (src.ssa->parent_instr->type != nir_instr_type_intrinsic)
  44       return NULL;
  45
  46    nir_intrinsic_instr *load = nir_instr_as_intrinsic(src.ssa->parent_instr);
  47    if (load->intrinsic != nir_intrinsic_load_deref)
  48       return NULL;
  49
  50    if (load->dest.ssa.index < first_valid_load)
  51       return NULL;
  52
  53    return nir_src_as_deref(load->src[0]);
  54 }
  55
  56 struct match_state {
  57    /* Index into the array of the last copy or -1 for no ongoing copy. */
  58    unsigned next_array_idx;
  59
  60    /* Length of the array we're copying */
  61    unsigned array_len;
  62
  63    /* Index into the deref path to the array we think is being copied */
  64    int src_deref_array_idx;
  65    int dst_deref_array_idx;
  66
  67    /* Deref paths of the first load/store pair or copy */
  68    nir_deref_path first_src_path;
  69    nir_deref_path first_dst_path;
  70 };
  71
  72 static void
  73 match_state_init(struct match_state *state)
  74 {
  75    state->next_array_idx = 0;
  76    state->array_len = 0;
  77    state->src_deref_array_idx = -1;
  78    state->dst_deref_array_idx = -1;
  79 }
  80
  81 static void
  82 match_state_finish(struct match_state *state)
  83 {
  84    if (state->next_array_idx > 0) {
  85       nir_deref_path_finish(&state->first_src_path);
  86       nir_deref_path_finish(&state->first_dst_path);
  87    }
  88 }
  89
  90 static void
  91 match_state_reset(struct match_state *state)
  92 {
  93    match_state_finish(state);
  94    match_state_init(state);
  95 }
  96
  97 static bool
  98 try_match_deref(nir_deref_path *base_path, int *path_array_idx,
  99                 nir_deref_instr *deref, int arr_idx, void *mem_ctx)
 100 {
 101    nir_deref_path deref_path;
 102    nir_deref_path_init(&deref_path, deref, mem_ctx);
 103
 104    bool found = false;
 105    for (int i = 0; ; i++) {
 106       nir_deref_instr *b = base_path->path[i];
 107       nir_deref_instr *d = deref_path.path[i];
 108       /* They have to be the same length */
 109       if ((b == NULL) != (d == NULL))
 110          goto fail;
 111
 112       if (b == NULL)
 113          break;
 114
 115       /* This can happen if one is a deref_array and the other a wildcard */
 116       if (b->deref_type != d->deref_type)
 117          goto fail;
 118
 119       switch (b->deref_type) {
 120       case nir_deref_type_var:
 121          if (b->var != d->var)
 122             goto fail;
 123          continue;
 124
 125       case nir_deref_type_array:
 126          assert(b->arr.index.is_ssa && d->arr.index.is_ssa);
 127          nir_const_value *const_b_idx = nir_src_as_const_value(b->arr.index);
 128          nir_const_value *const_d_idx = nir_src_as_const_value(d->arr.index);
 129
 130          /* If we don't have an index into the path yet or if this entry in
 131           * the path is at the array index, see if this is a candidate.  We're
 132           * looking for an index which is zero in the base deref and arr_idx
 133           * in the search deref.
 134           */
 135          if ((*path_array_idx < 0 || *path_array_idx == i) &&
 136              const_b_idx && const_b_idx->u32[0] == 0 &&
 137              const_d_idx && const_d_idx->u32[0] == arr_idx) {
 138             *path_array_idx = i;
 139             continue;
 140          }
 141
 142          /* We're at the array index but not a candidate */
 143          if (*path_array_idx == i)
 144             goto fail;
 145
 146          /* If we're not the path array index, we must match exactly.  We
 147           * could probably just compare SSA values and trust in copy
 148           * propagation but doing it ourselves means this pass can run a bit
 149           * earlier.
 150           */
 151          if (b->arr.index.ssa == d->arr.index.ssa ||
 152              (const_b_idx && const_d_idx &&
 153               const_b_idx->u32[0] == const_d_idx->u32[0]))
 154             continue;
 155
 156          goto fail;
 157
 158       case nir_deref_type_array_wildcard:
 159          continue;
 160
 161       case nir_deref_type_struct:
 162          if (b->strct.index != d->strct.index)
 163             goto fail;
 164          continue;
 165
 166       default:
 167          unreachable("Invalid deref type in a path");
 168       }
 169    }
 170
 171    /* If we got here without failing, we've matched.  However, it isn't an
 172     * array match unless we found an altered array index.
 173     */
 174    found = *path_array_idx > 0;
 175
 176 fail:
 177    nir_deref_path_finish(&deref_path);
 178    return found;
 179 }
 180
 181 static nir_deref_instr *
 182 build_wildcard_deref(nir_builder *b, nir_deref_path *path,
 183                      unsigned wildcard_idx)
 184 {
 185    assert(path->path[wildcard_idx]->deref_type == nir_deref_type_array);
 186
 187    nir_deref_instr *tail =
 188       nir_build_deref_array_wildcard(b, path->path[wildcard_idx - 1]);
 189
 190    for (unsigned i = wildcard_idx + 1; path->path[i]; i++)
 191       tail = nir_build_deref_follower(b, tail, path->path[i]);
 192
 193    return tail;
 194 }
 195
 196 static bool
 197 opt_find_array_copies_block(nir_builder *b, nir_block *block,
 198                             unsigned *num_ssa_defs, void *mem_ctx)
 199 {
 200    bool progress = false;
 201
 202    struct match_state s;
 203    match_state_init(&s);
 204
 205    nir_variable *dst_var = NULL;
 206    unsigned prev_dst_var_last_write = *num_ssa_defs;
 207    unsigned dst_var_last_write = *num_ssa_defs;
 208
 209    nir_foreach_instr(instr, block) {
 210       /* Index the SSA defs before we do anything else. */
 211       nir_foreach_ssa_def(instr, index_ssa_def_cb, num_ssa_defs);
 212
 213       if (instr->type != nir_instr_type_intrinsic)
 214          continue;
 215
 216       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 217       if (intrin->intrinsic != nir_intrinsic_copy_deref &&
 218           intrin->intrinsic != nir_intrinsic_store_deref)
 219          continue;
 220
 221       nir_deref_instr *dst_deref = nir_src_as_deref(intrin->src[0]);
 222
 223       /* The destination must be local.  If we see a non-local store, we
 224        * continue on because it won't affect local stores or read-only
 225        * variables.
 226        */
 227       if (dst_deref->mode != nir_var_local)
 228          continue;
 229
 230       /* We keep track of the SSA indices where the two last-written
 231        * variables are written.  The prev_dst_var_last_write tells us when
 232        * the last store_deref to something other than dst happened.  If the
 233        * SSA def index from a load is greater than or equal to this number
 234        * then we know it happened afterwards and no writes to anything other
 235        * than dst occur between the load and the current instruction.
 236        */
 237       if (nir_deref_instr_get_variable(dst_deref) != dst_var) {
 238          prev_dst_var_last_write = dst_var_last_write;
 239          dst_var = nir_deref_instr_get_variable(dst_deref);
 240       }
 241       dst_var_last_write = *num_ssa_defs;
 242
 243       /* If it's a full variable store or copy, reset.  This will trigger
 244        * eventually because we'll fail to match an array element.  However,
 245        * it's a cheap early-exit.
 246        */
 247       if (dst_deref->deref_type == nir_deref_type_var)
 248          goto reset;
 249
 250       nir_deref_instr *src_deref;
 251       if (intrin->intrinsic == nir_intrinsic_copy_deref) {
 252          src_deref = nir_src_as_deref(intrin->src[1]);
 253       } else {
 254          assert(intrin->intrinsic == nir_intrinsic_store_deref);
 255          src_deref = get_deref_for_load_src(intrin->src[1],
 256                                             prev_dst_var_last_write);
 257
 258          /* We can only handle full writes */
 259          if (nir_intrinsic_write_mask(intrin) !=
 260              (1 << glsl_get_components(dst_deref->type)) - 1)
 261             goto reset;
 262       }
 263
 264       /* If we didn't find a valid src, then we have an unknown store and it
 265        * could mess things up.
 266        */
 267       if (src_deref == NULL)
 268          goto reset;
 269
 270       /* The source must be either local or something that's guaranteed to be
 271        * read-only.
 272        */
 273       const nir_variable_mode read_only_modes =
 274          nir_var_shader_in | nir_var_uniform | nir_var_system_value;
 275       if (!(src_deref->mode & (nir_var_local | read_only_modes)))
 276          goto reset;
 277
 278       /* If we don't yet have an active copy, then make this instruction the
 279        * active copy.
 280        */
 281       if (s.next_array_idx == 0) {
 282          /* We can't combine a copy if there is any chance the source and
 283           * destination will end up aliasing.  Just bail if they're the same
 284           * variable.
 285           */
 286          if (nir_deref_instr_get_variable(src_deref) == dst_var)
 287             goto reset;
 288
 289          /* The load/store pair is enough to guarantee the same bit size and
 290           * number of components but a copy_var requires the actual types to
 291           * match.
 292           */
 293          if (dst_deref->type != src_deref->type)
 294             continue;
 295
 296          /* The first time we see a store, we don't know which array in the
 297           * deref path is the one being copied so we just record the paths
 298           * as-is and continue.  On the next iteration, it will try to match
 299           * based on which array index changed.
 300           */
 301          nir_deref_path_init(&s.first_dst_path, dst_deref, mem_ctx);
 302          nir_deref_path_init(&s.first_src_path, src_deref, mem_ctx);
 303          s.next_array_idx = 1;
 304          continue;
 305       }
 306
 307       if (!try_match_deref(&s.first_dst_path, &s.dst_deref_array_idx,
 308                            dst_deref, s.next_array_idx, mem_ctx) ||
 309           !try_match_deref(&s.first_src_path, &s.src_deref_array_idx,
 310                            src_deref, s.next_array_idx, mem_ctx))
 311          goto reset;
 312
 313       if (s.next_array_idx == 1) {
 314          /* This is our first non-trivial match.  We now have indices into
 315           * the search paths so we can do a couple more checks.
 316           */
 317          assert(s.dst_deref_array_idx > 0 && s.src_deref_array_idx > 0);
 318          const struct glsl_type *dst_arr_type =
 319             s.first_dst_path.path[s.dst_deref_array_idx - 1]->type;
 320          const struct glsl_type *src_arr_type =
 321             s.first_src_path.path[s.src_deref_array_idx - 1]->type;
 322
 323          assert(glsl_type_is_array(dst_arr_type) ||
 324                 glsl_type_is_matrix(dst_arr_type));
 325          assert(glsl_type_is_array(src_arr_type) ||
 326                 glsl_type_is_matrix(src_arr_type));
 327
 328          /* They must be the same length */
 329          s.array_len = glsl_get_length(dst_arr_type);
 330          if (s.array_len != glsl_get_length(src_arr_type))
 331             goto reset;
 332       }
 333
 334       s.next_array_idx++;
 335
 336       if (s.next_array_idx == s.array_len) {
 337          /* Hooray, We found a copy! */
 338          b->cursor = nir_after_instr(instr);
 339          nir_copy_deref(b, build_wildcard_deref(b, &s.first_dst_path,
 340                                                 s.dst_deref_array_idx),
 341                            build_wildcard_deref(b, &s.first_src_path,
 342                                                 s.src_deref_array_idx));
 343          match_state_reset(&s);
 344          progress = true;
 345       }
 346
 347       continue;
 348
 349    reset:
 350       match_state_reset(&s);
 351    }
 352
 353    return progress;
 354 }
 355
 356 static bool
 357 opt_find_array_copies_impl(nir_function_impl *impl)
 358 {
 359    nir_builder b;
 360    nir_builder_init(&b, impl);
 361
 362    bool progress = false;
 363
 364    void *mem_ctx = ralloc_context(NULL);
 365
 366    /* We re-index the SSA defs as we go; it makes it easier to handle
 367     * resetting the state machine.
 368     */
 369    unsigned num_ssa_defs = 0;
 370
 371    nir_foreach_block(block, impl) {
 372       if (opt_find_array_copies_block(&b, block, &num_ssa_defs, mem_ctx))
 373          progress = true;
 374    }
 375
 376    impl->ssa_alloc = num_ssa_defs;
 377
 378    ralloc_free(mem_ctx);
 379
 380    if (progress) {
 381       nir_metadata_preserve(impl, nir_metadata_block_index |
 382                                   nir_metadata_dominance);
 383    }
 384
 385    return progress;
 386 }
 387
 388 /**
 389  * This peephole optimization looks for a series of load/store_deref or
 390  * copy_deref instructions that copy an array from one variable to another and
 391  * turns it into a copy_deref that copies the entire array.  The pattern it
 392  * looks for is extremely specific but it's good enough to pick up on the
 393  * input array copies in DXVK and should also be able to pick up the sequence
 394  * generated by spirv_to_nir for a OpLoad of a large composite followed by
 395  * OpStore.
 396  *
 397  * TODO: Use a hash table approach to support out-of-order and interleaved
 398  * copies.
 399  */
 400 bool
 401 nir_opt_find_array_copies(nir_shader *shader)
 402 {
 403    bool progress = false;
 404
 405    nir_foreach_function(function, shader) {
 406       if (function->impl && opt_find_array_copies_impl(function->impl))
 407          progress = true;
 408    }
 409
 410    return progress;
 411 }