src/compiler/nir/nir_lower_io_arrays_to_elements.c

   1 /*
   2  * Copyright © 2017 Timothy Arceri
   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 /** @file nir_lower_io_arrays_to_elements.c
  29  *
  30  * Split arrays/matrices with direct indexing into individual elements. This
  31  * will allow optimisation passes to better clean up unused elements.
  32  *
  33  */
  34
  35 static unsigned
  36 get_io_offset(nir_builder *b, nir_deref_instr *deref, nir_variable *var,
  37               unsigned *element_index, nir_ssa_def **vertex_index)
  38 {
  39    nir_deref_path path;
  40    nir_deref_path_init(&path, deref, NULL);
  41
  42    assert(path.path[0]->deref_type == nir_deref_type_var);
  43    nir_deref_instr **p = &path.path[1];
  44
  45    /* For per-vertex input arrays (i.e. geometry shader inputs), skip the
  46     * outermost array index.  Process the rest normally.
  47     */
  48    if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
  49       *vertex_index = nir_ssa_for_src(b, (*p)->arr.index, 1);
  50       p++;
  51    }
  52
  53    unsigned offset = 0;
  54    for (; *p; p++) {
  55       if ((*p)->deref_type == nir_deref_type_array) {
  56          /* must not be indirect dereference */
  57          unsigned index = nir_src_as_uint((*p)->arr.index);
  58
  59          unsigned size = glsl_count_attribute_slots((*p)->type, false);
  60          offset += size * index;
  61
  62          unsigned num_elements = glsl_type_is_array((*p)->type) ?
  63             glsl_get_aoa_size((*p)->type) : 1;
  64
  65          num_elements *= glsl_type_is_matrix(glsl_without_array((*p)->type)) ?
  66             glsl_get_matrix_columns(glsl_without_array((*p)->type)) : 1;
  67
  68          *element_index += num_elements * index;
  69       } else if ((*p)->deref_type == nir_deref_type_struct) {
  70          /* TODO: we could also add struct splitting support to this pass */
  71          break;
  72       }
  73    }
  74
  75    nir_deref_path_finish(&path);
  76
  77    return offset;
  78 }
  79
  80 static nir_variable **
  81 get_array_elements(struct hash_table *ht, nir_variable *var,
  82                    gl_shader_stage stage)
  83 {
  84    nir_variable **elements;
  85    struct hash_entry *entry = _mesa_hash_table_search(ht, var);
  86    if (!entry) {
  87       const struct glsl_type *type = var->type;
  88       if (nir_is_per_vertex_io(var, stage)) {
  89          assert(glsl_type_is_array(type));
  90          type = glsl_get_array_element(type);
  91       }
  92
  93       unsigned num_elements = glsl_type_is_array(type) ?
  94          glsl_get_aoa_size(type) : 1;
  95
  96       num_elements *= glsl_type_is_matrix(glsl_without_array(type)) ?
  97          glsl_get_matrix_columns(glsl_without_array(type)) : 1;
  98
  99       elements = (nir_variable **) calloc(num_elements, sizeof(nir_variable *));
 100       _mesa_hash_table_insert(ht, var, elements);
 101    } else {
 102       elements = (nir_variable **) entry->data;
 103    }
 104
 105    return elements;
 106 }
 107
 108 static void
 109 lower_array(nir_builder *b, nir_intrinsic_instr *intr, nir_variable *var,
 110             struct hash_table *varyings)
 111 {
 112    b->cursor = nir_before_instr(&intr->instr);
 113
 114    nir_variable **elements =
 115       get_array_elements(varyings, var, b->shader->info.stage);
 116
 117    nir_ssa_def *vertex_index = NULL;
 118    unsigned elements_index = 0;
 119    unsigned io_offset = get_io_offset(b, nir_src_as_deref(intr->src[0]),
 120                                       var, &elements_index, &vertex_index);
 121
 122    nir_variable *element = elements[elements_index];
 123    if (!element) {
 124          element = nir_variable_clone(var, b->shader);
 125          element->data.location =  var->data.location + io_offset;
 126
 127          const struct glsl_type *type = glsl_without_array(element->type);
 128
 129          /* This pass also splits matrices so we need give them a new type. */
 130          if (glsl_type_is_matrix(type))
 131             type = glsl_get_column_type(type);
 132
 133          if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 134             type = glsl_array_type(type, glsl_get_length(element->type),
 135                                    glsl_get_explicit_stride(element->type));
 136          }
 137
 138          element->type = type;
 139          elements[elements_index] = element;
 140
 141          nir_shader_add_variable(b->shader, element);
 142    }
 143
 144    nir_deref_instr *element_deref = nir_build_deref_var(b, element);
 145
 146    if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 147       assert(vertex_index);
 148       element_deref = nir_build_deref_array(b, element_deref, vertex_index);
 149    }
 150
 151    nir_intrinsic_instr *element_intr =
 152       nir_intrinsic_instr_create(b->shader, intr->intrinsic);
 153    element_intr->num_components = intr->num_components;
 154    element_intr->src[0] = nir_src_for_ssa(&element_deref->dest.ssa);
 155
 156    if (intr->intrinsic != nir_intrinsic_store_deref) {
 157       nir_ssa_dest_init(&element_intr->instr, &element_intr->dest,
 158                         intr->num_components, intr->dest.ssa.bit_size, NULL);
 159
 160       if (intr->intrinsic == nir_intrinsic_interp_deref_at_offset ||
 161           intr->intrinsic == nir_intrinsic_interp_deref_at_sample) {
 162          nir_src_copy(&element_intr->src[1], &intr->src[1],
 163                       &element_intr->instr);
 164       }
 165
 166       nir_ssa_def_rewrite_uses(&intr->dest.ssa,
 167                                nir_src_for_ssa(&element_intr->dest.ssa));
 168    } else {
 169       nir_intrinsic_set_write_mask(element_intr,
 170                                    nir_intrinsic_write_mask(intr));
 171       nir_src_copy(&element_intr->src[1], &intr->src[1],
 172                    &element_intr->instr);
 173    }
 174
 175    nir_builder_instr_insert(b, &element_intr->instr);
 176
 177    /* Remove the old load intrinsic */
 178    nir_instr_remove(&intr->instr);
 179 }
 180
 181 static bool
 182 deref_has_indirect(nir_builder *b, nir_variable *var, nir_deref_path *path)
 183 {
 184    assert(path->path[0]->deref_type == nir_deref_type_var);
 185    nir_deref_instr **p = &path->path[1];
 186
 187    if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 188       p++;
 189    }
 190
 191    for (; *p; p++) {
 192       if ((*p)->deref_type != nir_deref_type_array)
 193          continue;
 194
 195       if (!nir_src_is_const((*p)->arr.index))
 196          return true;
 197    }
 198
 199    return false;
 200 }
 201
 202 /* Creates a mask of locations that contains arrays that are indexed via
 203  * indirect indexing.
 204  */
 205 static void
 206 create_indirects_mask(nir_shader *shader, uint64_t *indirects,
 207                       uint64_t *patch_indirects, nir_variable_mode mode)
 208 {
 209    nir_foreach_function(function, shader) {
 210       if (function->impl) {
 211          nir_builder b;
 212          nir_builder_init(&b, function->impl);
 213
 214          nir_foreach_block(block, function->impl) {
 215             nir_foreach_instr_safe(instr, block) {
 216
 217                if (instr->type != nir_instr_type_intrinsic)
 218                   continue;
 219
 220                nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 221
 222                if (intr->intrinsic != nir_intrinsic_load_deref &&
 223                    intr->intrinsic != nir_intrinsic_store_deref &&
 224                    intr->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
 225                    intr->intrinsic != nir_intrinsic_interp_deref_at_sample &&
 226                    intr->intrinsic != nir_intrinsic_interp_deref_at_offset)
 227                   continue;
 228
 229                nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
 230                if (deref->mode != mode)
 231                   continue;
 232
 233                nir_variable *var = nir_deref_instr_get_variable(deref);
 234
 235                nir_deref_path path;
 236                nir_deref_path_init(&path, deref, NULL);
 237
 238                uint64_t loc_mask = ((uint64_t)1) << var->data.location;
 239                if (var->data.patch) {
 240                   if (deref_has_indirect(&b, var, &path))
 241                      patch_indirects[var->data.location_frac] |= loc_mask;
 242                } else {
 243                   if (deref_has_indirect(&b, var, &path))
 244                      indirects[var->data.location_frac] |= loc_mask;
 245                }
 246
 247                nir_deref_path_finish(&path);
 248             }
 249          }
 250       }
 251    }
 252 }
 253
 254 static void
 255 lower_io_arrays_to_elements(nir_shader *shader, nir_variable_mode mask,
 256                             uint64_t *indirects, uint64_t *patch_indirects,
 257                             struct hash_table *varyings,
 258                             bool after_cross_stage_opts)
 259 {
 260    nir_foreach_function(function, shader) {
 261       if (function->impl) {
 262          nir_builder b;
 263          nir_builder_init(&b, function->impl);
 264
 265          nir_foreach_block(block, function->impl) {
 266             nir_foreach_instr_safe(instr, block) {
 267                if (instr->type != nir_instr_type_intrinsic)
 268                   continue;
 269
 270                nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 271
 272                if (intr->intrinsic != nir_intrinsic_load_deref &&
 273                    intr->intrinsic != nir_intrinsic_store_deref &&
 274                    intr->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
 275                    intr->intrinsic != nir_intrinsic_interp_deref_at_sample &&
 276                    intr->intrinsic != nir_intrinsic_interp_deref_at_offset)
 277                   continue;
 278
 279                nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
 280                if (!(deref->mode & mask))
 281                   continue;
 282
 283                nir_variable *var = nir_deref_instr_get_variable(deref);
 284
 285                /* Skip indirects */
 286                uint64_t loc_mask = ((uint64_t)1) << var->data.location;
 287                if (var->data.patch) {
 288                   if (patch_indirects[var->data.location_frac] & loc_mask)
 289                      continue;
 290                } else {
 291                   if (indirects[var->data.location_frac] & loc_mask)
 292                      continue;
 293                }
 294
 295                nir_variable_mode mode = var->data.mode;
 296
 297                const struct glsl_type *type = var->type;
 298                if (nir_is_per_vertex_io(var, b.shader->info.stage)) {
 299                   assert(glsl_type_is_array(type));
 300                   type = glsl_get_array_element(type);
 301                }
 302
 303                /* Skip types we cannot split.
 304                 *
 305                 * TODO: Add support for struct splitting.
 306                 */
 307                if ((!glsl_type_is_array(type) && !glsl_type_is_matrix(type))||
 308                    glsl_type_is_struct(glsl_without_array(type)))
 309                   continue;
 310
 311                /* Skip builtins */
 312                if (!after_cross_stage_opts &&
 313                    var->data.location < VARYING_SLOT_VAR0 &&
 314                    var->data.location >= 0)
 315                   continue;
 316
 317                /* Don't bother splitting if we can't opt away any unused
 318                 * elements.
 319                 */
 320                if (!after_cross_stage_opts && var->data.always_active_io)
 321                   continue;
 322
 323                switch (intr->intrinsic) {
 324                case nir_intrinsic_interp_deref_at_centroid:
 325                case nir_intrinsic_interp_deref_at_sample:
 326                case nir_intrinsic_interp_deref_at_offset:
 327                case nir_intrinsic_load_deref:
 328                case nir_intrinsic_store_deref:
 329                   if ((mask & nir_var_shader_in && mode == nir_var_shader_in) ||
 330                       (mask & nir_var_shader_out && mode == nir_var_shader_out))
 331                      lower_array(&b, intr, var, varyings);
 332                   break;
 333                default:
 334                   break;
 335                }
 336             }
 337          }
 338       }
 339    }
 340 }
 341
 342 void
 343 nir_lower_io_arrays_to_elements_no_indirects(nir_shader *shader,
 344                                              bool outputs_only)
 345 {
 346    struct hash_table *split_inputs = _mesa_pointer_hash_table_create(NULL);
 347    struct hash_table *split_outputs = _mesa_pointer_hash_table_create(NULL);
 348
 349    uint64_t indirects[4] = {0}, patch_indirects[4] = {0};
 350
 351    lower_io_arrays_to_elements(shader, nir_var_shader_out, indirects,
 352                                patch_indirects, split_outputs, true);
 353
 354    if (!outputs_only) {
 355       lower_io_arrays_to_elements(shader, nir_var_shader_in, indirects,
 356                                   patch_indirects, split_inputs, true);
 357
 358       /* Remove old input from the shaders inputs list */
 359       hash_table_foreach(split_inputs, entry) {
 360          nir_variable *var = (nir_variable *) entry->key;
 361          exec_node_remove(&var->node);
 362
 363          free(entry->data);
 364       }
 365    }
 366
 367    /* Remove old output from the shaders outputs list */
 368    hash_table_foreach(split_outputs, entry) {
 369       nir_variable *var = (nir_variable *) entry->key;
 370       exec_node_remove(&var->node);
 371
 372       free(entry->data);
 373    }
 374
 375    _mesa_hash_table_destroy(split_inputs, NULL);
 376    _mesa_hash_table_destroy(split_outputs, NULL);
 377
 378    nir_remove_dead_derefs(shader);
 379 }
 380
 381 void
 382 nir_lower_io_arrays_to_elements(nir_shader *producer, nir_shader *consumer)
 383 {
 384    struct hash_table *split_inputs = _mesa_pointer_hash_table_create(NULL);
 385    struct hash_table *split_outputs = _mesa_pointer_hash_table_create(NULL);
 386
 387    uint64_t indirects[4] = {0}, patch_indirects[4] = {0};
 388    create_indirects_mask(producer, indirects, patch_indirects,
 389                          nir_var_shader_out);
 390    create_indirects_mask(consumer, indirects, patch_indirects,
 391                          nir_var_shader_in);
 392
 393    lower_io_arrays_to_elements(producer, nir_var_shader_out, indirects,
 394                                patch_indirects, split_outputs, false);
 395
 396    lower_io_arrays_to_elements(consumer, nir_var_shader_in, indirects,
 397                                patch_indirects, split_inputs, false);
 398
 399    /* Remove old input from the shaders inputs list */
 400    hash_table_foreach(split_inputs, entry) {
 401       nir_variable *var = (nir_variable *) entry->key;
 402       exec_node_remove(&var->node);
 403
 404       free(entry->data);
 405    }
 406
 407    /* Remove old output from the shaders outputs list */
 408    hash_table_foreach(split_outputs, entry) {
 409       nir_variable *var = (nir_variable *) entry->key;
 410       exec_node_remove(&var->node);
 411
 412       free(entry->data);
 413    }
 414
 415    _mesa_hash_table_destroy(split_inputs, NULL);
 416    _mesa_hash_table_destroy(split_outputs, NULL);
 417
 418    nir_remove_dead_derefs(producer);
 419    nir_remove_dead_derefs(consumer);
 420 }