src/compiler/nir/nir_split_var_copies.c

   1 /*
   2  * Copyright © 2014 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  * Authors:
  24  *    Jason Ekstrand (jason@jlekstrand.net)
  25  *
  26  */
  27
  28 #include "nir.h"
  29
  30 /*
  31  * Implements "copy splitting" which is similar to structure splitting only
  32  * it works on copy operations rather than the datatypes themselves.  The
  33  * GLSL language allows you to copy one variable to another an entire
  34  * structure (which may contain arrays or other structures) at a time.
  35  * Normally, in a language such as C this would be handled by a "structure
  36  * splitting" pass that breaks up the structures.  Unfortunately for us,
  37  * structures used in inputs or outputs can't be split.  Therefore,
  38  * regardlesss of what we do, we have to be able to copy to/from
  39  * structures.
  40  *
  41  * The primary purpose of structure splitting is to allow you to better
  42  * optimize variable access and lower things to registers where you can.
  43  * The primary issue here is that, if you lower the copy to a bunch of
  44  * loads and stores, you loose a lot of information about the copy
  45  * operation that you would like to keep around.  To solve this problem, we
  46  * have a "copy splitting" pass that, instead of splitting the structures
  47  * or lowering the copy into loads and storres, splits the copy operation
  48  * into a bunch of copy operations one for each leaf of the structure tree.
  49  * If an intermediate array is encountered, it is referenced with a
  50  * wildcard reference to indicate that the entire array is to be copied.
  51  *
  52  * As things become direct, array copies may be able to be losslessly
  53  * lowered to having fewer and fewer wildcards.  However, until that
  54  * happens we want to keep the information about the arrays intact.
  55  *
  56  * Prior to the copy splitting pass, there are no wildcard references but
  57  * there may be incomplete references where the tail of the deref chain is
  58  * an array or a structure and not a specific element.  After the copy
  59  * splitting pass has completed, every variable deref will be a full-length
  60  * dereference pointing to a single leaf in the structure type tree with
  61  * possibly a few wildcard array dereferences.
  62  */
  63
  64 struct split_var_copies_state {
  65    nir_shader *shader;
  66    void *dead_ctx;
  67    bool progress;
  68 };
  69
  70 /* Recursively constructs deref chains to split a copy instruction into
  71  * multiple (if needed) copy instructions with full-length deref chains.
  72  * External callers of this function should pass the tail and head of the
  73  * deref chains found as the source and destination of the copy instruction
  74  * into this function.
  75  *
  76  * \param  old_copy  The copy instruction we are splitting
  77  * \param  dest_head The head of the destination deref chain we are building
  78  * \param  src_head  The head of the source deref chain we are building
  79  * \param  dest_tail The tail of the destination deref chain we are building
  80  * \param  src_tail  The tail of the source deref chain we are building
  81  * \param  state     The current split_var_copies_state object
  82  */
  83 static void
  84 split_var_copy_instr(nir_intrinsic_instr *old_copy,
  85                      nir_deref_var *dest_head, nir_deref_var *src_head,
  86                      nir_deref *dest_tail, nir_deref *src_tail,
  87                      struct split_var_copies_state *state)
  88 {
  89    assert(src_tail->type == dest_tail->type);
  90
  91    /* Make sure these really are the tails of the deref chains */
  92    assert(dest_tail->child == NULL);
  93    assert(src_tail->child == NULL);
  94
  95    switch (glsl_get_base_type(src_tail->type)) {
  96    case GLSL_TYPE_ARRAY: {
  97       /* Make a wildcard dereference */
  98       nir_deref_array *deref = nir_deref_array_create(state->dead_ctx);
  99       deref->deref.type = glsl_get_array_element(src_tail->type);
 100       deref->deref_array_type = nir_deref_array_type_wildcard;
 101
 102       /* Set the tail of both as the newly created wildcard deref.  It is
 103        * safe to use the same wildcard in both places because a) we will be
 104        * copying it before we put it in an actual instruction and b)
 105        * everything that will potentially add another link in the deref
 106        * chain will also add the same thing to both chains.
 107        */
 108       src_tail->child = &deref->deref;
 109       dest_tail->child = &deref->deref;
 110
 111       split_var_copy_instr(old_copy, dest_head, src_head,
 112                            dest_tail->child, src_tail->child, state);
 113
 114       /* Set it back to the way we found it */
 115       src_tail->child = NULL;
 116       dest_tail->child = NULL;
 117       break;
 118    }
 119
 120    case GLSL_TYPE_STRUCT:
 121       /* This is the only part that actually does any interesting
 122        * splitting.  For array types, we just use wildcards and resolve
 123        * them later.  For structure types, we need to emit one copy
 124        * instruction for every structure element.  Because we may have
 125        * structs inside structs, we just recurse and let the next level
 126        * take care of any additional structures.
 127        */
 128       for (unsigned i = 0; i < glsl_get_length(src_tail->type); i++) {
 129          nir_deref_struct *deref = nir_deref_struct_create(state->dead_ctx, i);
 130          deref->deref.type = glsl_get_struct_field(src_tail->type, i);
 131
 132          /* Set the tail of both as the newly created structure deref.  It
 133           * is safe to use the same wildcard in both places because a) we
 134           * will be copying it before we put it in an actual instruction
 135           * and b) everything that will potentially add another link in the
 136           * deref chain will also add the same thing to both chains.
 137           */
 138          src_tail->child = &deref->deref;
 139          dest_tail->child = &deref->deref;
 140
 141          split_var_copy_instr(old_copy, dest_head, src_head,
 142                               dest_tail->child, src_tail->child, state);
 143       }
 144       /* Set it back to the way we found it */
 145       src_tail->child = NULL;
 146       dest_tail->child = NULL;
 147       break;
 148
 149    case GLSL_TYPE_UINT:
 150    case GLSL_TYPE_UINT16:
 151    case GLSL_TYPE_UINT64:
 152    case GLSL_TYPE_INT:
 153    case GLSL_TYPE_INT16:
 154    case GLSL_TYPE_INT64:
 155    case GLSL_TYPE_FLOAT:
 156    case GLSL_TYPE_FLOAT16:
 157    case GLSL_TYPE_DOUBLE:
 158    case GLSL_TYPE_BOOL:
 159       if (glsl_type_is_matrix(src_tail->type)) {
 160          nir_deref_array *deref = nir_deref_array_create(state->dead_ctx);
 161          deref->deref.type = glsl_get_column_type(src_tail->type);
 162          deref->deref_array_type = nir_deref_array_type_wildcard;
 163
 164          /* Set the tail of both as the newly created wildcard deref.  It
 165           * is safe to use the same wildcard in both places because a) we
 166           * will be copying it before we put it in an actual instruction
 167           * and b) everything that will potentially add another link in the
 168           * deref chain will also add the same thing to both chains.
 169           */
 170          src_tail->child = &deref->deref;
 171          dest_tail->child = &deref->deref;
 172
 173          split_var_copy_instr(old_copy, dest_head, src_head,
 174                               dest_tail->child, src_tail->child, state);
 175
 176          /* Set it back to the way we found it */
 177          src_tail->child = NULL;
 178          dest_tail->child = NULL;
 179       } else {
 180          /* At this point, we have fully built our deref chains and can
 181           * actually add the new copy instruction.
 182           */
 183          nir_intrinsic_instr *new_copy =
 184             nir_intrinsic_instr_create(state->shader, nir_intrinsic_copy_var);
 185
 186          /* We need to make copies because a) this deref chain actually
 187           * belongs to the copy instruction and b) the deref chains may
 188           * have some of the same links due to the way we constructed them
 189           */
 190          new_copy->variables[0] = nir_deref_var_clone(dest_head, new_copy);
 191          new_copy->variables[1] = nir_deref_var_clone(src_head, new_copy);
 192
 193          /* Emit the copy instruction after the old instruction.  We'll
 194           * remove the old one later.
 195           */
 196          nir_instr_insert_after(&old_copy->instr, &new_copy->instr);
 197          state->progress = true;
 198       }
 199       break;
 200
 201    case GLSL_TYPE_SAMPLER:
 202    case GLSL_TYPE_IMAGE:
 203    case GLSL_TYPE_ATOMIC_UINT:
 204    case GLSL_TYPE_INTERFACE:
 205    default:
 206       unreachable("Cannot copy these types");
 207    }
 208 }
 209
 210 static bool
 211 split_var_copies_block(nir_block *block, struct split_var_copies_state *state)
 212 {
 213    nir_foreach_instr_safe(instr, block) {
 214       if (instr->type != nir_instr_type_intrinsic)
 215          continue;
 216
 217       nir_intrinsic_instr *intrinsic = nir_instr_as_intrinsic(instr);
 218       if (intrinsic->intrinsic != nir_intrinsic_copy_var)
 219          continue;
 220
 221       nir_deref_var *dest_head = intrinsic->variables[0];
 222       nir_deref_var *src_head = intrinsic->variables[1];
 223       nir_deref *dest_tail = nir_deref_tail(&dest_head->deref);
 224       nir_deref *src_tail = nir_deref_tail(&src_head->deref);
 225
 226       switch (glsl_get_base_type(src_tail->type)) {
 227       case GLSL_TYPE_ARRAY:
 228       case GLSL_TYPE_STRUCT:
 229          split_var_copy_instr(intrinsic, dest_head, src_head,
 230                               dest_tail, src_tail, state);
 231          nir_instr_remove(&intrinsic->instr);
 232          ralloc_steal(state->dead_ctx, instr);
 233          break;
 234       case GLSL_TYPE_FLOAT:
 235       case GLSL_TYPE_FLOAT16:
 236       case GLSL_TYPE_DOUBLE:
 237          if (glsl_type_is_matrix(src_tail->type)) {
 238             split_var_copy_instr(intrinsic, dest_head, src_head,
 239                                  dest_tail, src_tail, state);
 240             nir_instr_remove(&intrinsic->instr);
 241             ralloc_steal(state->dead_ctx, instr);
 242          }
 243          break;
 244       case GLSL_TYPE_INT:
 245       case GLSL_TYPE_UINT:
 246       case GLSL_TYPE_INT16:
 247       case GLSL_TYPE_UINT16:
 248       case GLSL_TYPE_INT64:
 249       case GLSL_TYPE_UINT64:
 250       case GLSL_TYPE_BOOL:
 251          assert(!glsl_type_is_matrix(src_tail->type));
 252          break;
 253       default:
 254          unreachable("Invalid type");
 255          break;
 256       }
 257    }
 258
 259    return true;
 260 }
 261
 262 static bool
 263 split_var_copies_impl(nir_function_impl *impl)
 264 {
 265    struct split_var_copies_state state;
 266
 267    state.shader = impl->function->shader;
 268    state.dead_ctx = ralloc_context(NULL);
 269    state.progress = false;
 270
 271    nir_foreach_block(block, impl) {
 272       split_var_copies_block(block, &state);
 273    }
 274
 275    ralloc_free(state.dead_ctx);
 276
 277    if (state.progress) {
 278       nir_metadata_preserve(impl, nir_metadata_block_index |
 279                                   nir_metadata_dominance);
 280    }
 281
 282    return state.progress;
 283 }
 284
 285 bool
 286 nir_split_var_copies(nir_shader *shader)
 287 {
 288    bool progress = false;
 289
 290    nir_foreach_function(function, shader) {
 291       if (function->impl)
 292          progress = split_var_copies_impl(function->impl) || progress;
 293    }
 294
 295    return progress;
 296 }