src/glsl/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    void *mem_ctx;
  66    void *dead_ctx;
  67 };
  68
  69 /* Recursively constructs deref chains to split a copy instruction into
  70  * multiple (if needed) copy instructions with full-length deref chains.
  71  * External callers of this function should pass the tail and head of the
  72  * deref chains found as the source and destination of the copy instruction
  73  * into this function.
  74  *
  75  * \param  old_copy  The copy instruction we are splitting
  76  * \param  dest_head The head of the destination deref chain we are building
  77  * \param  src_head  The head of the source deref chain we are building
  78  * \param  dest_tail The tail of the destination deref chain we are building
  79  * \param  src_tail  The tail of the source deref chain we are building
  80  * \param  state     The current split_var_copies_state object
  81  */
  82 static void
  83 split_var_copy_instr(nir_intrinsic_instr *old_copy,
  84                      nir_deref *dest_head, nir_deref *src_head,
  85                      nir_deref *dest_tail, nir_deref *src_tail,
  86                      struct split_var_copies_state *state)
  87 {
  88    assert(src_tail->type == dest_tail->type);
  89
  90    /* Make sure these really are the tails of the deref chains */
  91    assert(dest_tail->child == NULL);
  92    assert(src_tail->child == NULL);
  93
  94    switch (glsl_get_base_type(src_tail->type)) {
  95    case GLSL_TYPE_ARRAY: {
  96       /* Make a wildcard dereference */
  97       nir_deref_array *deref = nir_deref_array_create(state->dead_ctx);
  98       deref->deref.type = glsl_get_array_element(src_tail->type);
  99       deref->deref_array_type = nir_deref_array_type_wildcard;
 100
 101       /* Set the tail of both as the newly created wildcard deref.  It is
 102        * safe to use the same wildcard in both places because a) we will be
 103        * copying it before we put it in an actual instruction and b)
 104        * everything that will potentially add another link in the deref
 105        * chain will also add the same thing to both chains.
 106        */
 107       src_tail->child = &deref->deref;
 108       dest_tail->child = &deref->deref;
 109
 110       split_var_copy_instr(old_copy, dest_head, src_head,
 111                            dest_tail->child, src_tail->child, state);
 112
 113       /* Set it back to the way we found it */
 114       src_tail->child = NULL;
 115       dest_tail->child = NULL;
 116       break;
 117    }
 118
 119    case GLSL_TYPE_STRUCT:
 120       /* This is the only part that actually does any interesting
 121        * splitting.  For array types, we just use wildcards and resolve
 122        * them later.  For structure types, we need to emit one copy
 123        * instruction for every structure element.  Because we may have
 124        * structs inside structs, we just recurse and let the next level
 125        * take care of any additional structures.
 126        */
 127       for (unsigned i = 0; i < glsl_get_length(src_tail->type); i++) {
 128          nir_deref_struct *deref = nir_deref_struct_create(state->dead_ctx, i);
 129          deref->deref.type = glsl_get_struct_field(src_tail->type, i);
 130
 131          /* Set the tail of both as the newly created structure deref.  It
 132           * is safe to use the same wildcard in both places because a) we
 133           * will be copying it before we put it in an actual instruction
 134           * and b) everything that will potentially add another link in the
 135           * deref chain will also add the same thing to both chains.
 136           */
 137          src_tail->child = &deref->deref;
 138          dest_tail->child = &deref->deref;
 139
 140          split_var_copy_instr(old_copy, dest_head, src_head,
 141                               dest_tail->child, src_tail->child, state);
 142       }
 143       /* Set it back to the way we found it */
 144       src_tail->child = NULL;
 145       dest_tail->child = NULL;
 146       break;
 147
 148    case GLSL_TYPE_UINT:
 149    case GLSL_TYPE_INT:
 150    case GLSL_TYPE_FLOAT:
 151    case GLSL_TYPE_BOOL:
 152       if (glsl_type_is_matrix(src_tail->type)) {
 153          nir_deref_array *deref = nir_deref_array_create(state->dead_ctx);
 154          deref->deref.type = glsl_get_column_type(src_tail->type);
 155          deref->deref_array_type = nir_deref_array_type_wildcard;
 156
 157          /* Set the tail of both as the newly created wildcard deref.  It
 158           * is safe to use the same wildcard in both places because a) we
 159           * will be copying it before we put it in an actual instruction
 160           * and b) everything that will potentially add another link in the
 161           * deref chain will also add the same thing to both chains.
 162           */
 163          src_tail->child = &deref->deref;
 164          dest_tail->child = &deref->deref;
 165
 166          split_var_copy_instr(old_copy, dest_head, src_head,
 167                               dest_tail->child, src_tail->child, state);
 168
 169          /* Set it back to the way we found it */
 170          src_tail->child = NULL;
 171          dest_tail->child = NULL;
 172       } else {
 173          /* At this point, we have fully built our deref chains and can
 174           * actually add the new copy instruction.
 175           */
 176          nir_intrinsic_instr *new_copy =
 177             nir_intrinsic_instr_create(state->mem_ctx, nir_intrinsic_copy_var);
 178
 179          /* We need to make copies because a) this deref chain actually
 180           * belongs to the copy instruction and b) the deref chains may
 181           * have some of the same links due to the way we constructed them
 182           */
 183          nir_deref *src = nir_copy_deref(new_copy, src_head);
 184          nir_deref *dest = nir_copy_deref(new_copy, dest_head);
 185
 186          new_copy->variables[0] = nir_deref_as_var(dest);
 187          new_copy->variables[1] = nir_deref_as_var(src);
 188
 189          /* Emit the copy instruction after the old instruction.  We'll
 190           * remove the old one later.
 191           */
 192          nir_instr_insert_after(&old_copy->instr, &new_copy->instr);
 193       }
 194       break;
 195
 196    case GLSL_TYPE_SAMPLER:
 197    case GLSL_TYPE_IMAGE:
 198    case GLSL_TYPE_ATOMIC_UINT:
 199    case GLSL_TYPE_INTERFACE:
 200    default:
 201       unreachable("Cannot copy these types");
 202    }
 203 }
 204
 205 static bool
 206 split_var_copies_block(nir_block *block, void *void_state)
 207 {
 208    struct split_var_copies_state *state = void_state;
 209
 210    nir_foreach_instr_safe(block, instr) {
 211       if (instr->type != nir_instr_type_intrinsic)
 212          continue;
 213
 214       nir_intrinsic_instr *intrinsic = nir_instr_as_intrinsic(instr);
 215       if (intrinsic->intrinsic != nir_intrinsic_copy_var)
 216          continue;
 217
 218       nir_deref *dest_head = &intrinsic->variables[0]->deref;
 219       nir_deref *src_head = &intrinsic->variables[1]->deref;
 220       nir_deref *dest_tail = nir_deref_tail(dest_head);
 221       nir_deref *src_tail = nir_deref_tail(src_head);
 222
 223       switch (glsl_get_base_type(src_tail->type)) {
 224       case GLSL_TYPE_ARRAY:
 225       case GLSL_TYPE_STRUCT:
 226          split_var_copy_instr(intrinsic, dest_head, src_head,
 227                               dest_tail, src_tail, state);
 228          nir_instr_remove(&intrinsic->instr);
 229          ralloc_steal(state->dead_ctx, instr);
 230          break;
 231       case GLSL_TYPE_FLOAT:
 232       case GLSL_TYPE_INT:
 233       case GLSL_TYPE_UINT:
 234       case GLSL_TYPE_BOOL:
 235          if (glsl_type_is_matrix(src_tail->type)) {
 236             split_var_copy_instr(intrinsic, dest_head, src_head,
 237                                  dest_tail, src_tail, state);
 238             nir_instr_remove(&intrinsic->instr);
 239             ralloc_steal(state->dead_ctx, instr);
 240          }
 241          break;
 242       default:
 243          unreachable("Invalid type");
 244          break;
 245       }
 246    }
 247
 248    return true;
 249 }
 250
 251 static void
 252 split_var_copies_impl(nir_function_impl *impl)
 253 {
 254    struct split_var_copies_state state;
 255
 256    state.mem_ctx = ralloc_parent(impl);
 257    state.dead_ctx = ralloc_context(NULL);
 258
 259    nir_foreach_block(impl, split_var_copies_block, &state);
 260
 261    ralloc_free(state.dead_ctx);
 262 }
 263
 264 void
 265 nir_split_var_copies(nir_shader *shader)
 266 {
 267    nir_foreach_overload(shader, overload) {
 268       if (overload->impl)
 269          split_var_copies_impl(overload->impl);
 270    }
 271 }