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          nir_const_value *c = nir_src_as_const_value((*p)->arr.index);
  57
  58          assert(c);     /* must not be indirect dereference */
  59
  60          unsigned size = glsl_count_attribute_slots((*p)->type, false);
  61          offset += size * c->u32[0];
  62
  63          unsigned num_elements = glsl_type_is_array((*p)->type) ?
  64             glsl_get_aoa_size((*p)->type) : 1;
  65
  66          num_elements *= glsl_type_is_matrix(glsl_without_array((*p)->type)) ?
  67             glsl_get_matrix_columns(glsl_without_array((*p)->type)) : 1;
  68
  69          *element_index += num_elements * c->u32[0];
  70       } else if ((*p)->deref_type == nir_deref_type_struct) {
  71          /* TODO: we could also add struct splitting support to this pass */
  72          break;
  73       }
  74    }
  75
  76    nir_deref_path_finish(&path);
  77
  78    return offset;
  79 }
  80
  81 static nir_variable **
  82 get_array_elements(struct hash_table *ht, nir_variable *var,
  83                    gl_shader_stage stage)
  84 {
  85    nir_variable **elements;
  86    struct hash_entry *entry = _mesa_hash_table_search(ht, var);
  87    if (!entry) {
  88       const struct glsl_type *type = var->type;
  89       if (nir_is_per_vertex_io(var, stage)) {
  90          assert(glsl_type_is_array(type));
  91          type = glsl_get_array_element(type);
  92       }
  93
  94       unsigned num_elements = glsl_type_is_array(type) ?
  95          glsl_get_aoa_size(type) : 1;
  96
  97       num_elements *= glsl_type_is_matrix(glsl_without_array(type)) ?
  98          glsl_get_matrix_columns(glsl_without_array(type)) : 1;
  99
 100       elements = (nir_variable **) calloc(num_elements, sizeof(nir_variable *));
 101       _mesa_hash_table_insert(ht, var, elements);
 102    } else {
 103       elements = (nir_variable **) entry->data;
 104    }
 105
 106    return elements;
 107 }
 108
 109 static void
 110 lower_array(nir_builder *b, nir_intrinsic_instr *intr, nir_variable *var,
 111             struct hash_table *varyings)
 112 {
 113    b->cursor = nir_before_instr(&intr->instr);
 114
 115    nir_variable **elements =
 116       get_array_elements(varyings, var, b->shader->info.stage);
 117
 118    nir_ssa_def *vertex_index = NULL;
 119    unsigned elements_index = 0;
 120    unsigned io_offset = get_io_offset(b, nir_src_as_deref(intr->src[0]),
 121                                       var, &elements_index, &vertex_index);
 122
 123    nir_variable *element = elements[elements_index];
 124    if (!element) {
 125          element = nir_variable_clone(var, b->shader);
 126          element->data.location =  var->data.location + io_offset;
 127
 128          const struct glsl_type *type = glsl_without_array(element->type);
 129
 130          /* This pass also splits matrices so we need give them a new type. */
 131          if (glsl_type_is_matrix(type)) {
 132             type = glsl_vector_type(glsl_get_base_type(type),
 133                                     glsl_get_vector_elements(type));
 134          }
 135
 136          if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 137             type = glsl_get_array_instance(type,
 138                                            glsl_get_length(element->type));
 139          }
 140
 141          element->type = type;
 142          elements[elements_index] = element;
 143
 144          nir_shader_add_variable(b->shader, element);
 145    }
 146
 147    nir_deref_instr *element_deref = nir_build_deref_var(b, element);
 148
 149    if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 150       assert(vertex_index);
 151       element_deref = nir_build_deref_array(b, element_deref, vertex_index);
 152    }
 153
 154    nir_intrinsic_instr *element_intr =
 155       nir_intrinsic_instr_create(b->shader, intr->intrinsic);
 156    element_intr->num_components = intr->num_components;
 157    element_intr->src[0] = nir_src_for_ssa(&element_deref->dest.ssa);
 158
 159    if (intr->intrinsic != nir_intrinsic_store_deref) {
 160       nir_ssa_dest_init(&element_intr->instr, &element_intr->dest,
 161                         intr->num_components, intr->dest.ssa.bit_size, NULL);
 162
 163       if (intr->intrinsic == nir_intrinsic_interp_deref_at_offset ||
 164           intr->intrinsic == nir_intrinsic_interp_deref_at_sample) {
 165          nir_src_copy(&element_intr->src[1], &intr->src[1],
 166                       &element_intr->instr);
 167       }
 168
 169       nir_ssa_def_rewrite_uses(&intr->dest.ssa,
 170                                nir_src_for_ssa(&element_intr->dest.ssa));
 171    } else {
 172       nir_intrinsic_set_write_mask(element_intr,
 173                                    nir_intrinsic_write_mask(intr));
 174       nir_src_copy(&element_intr->src[1], &intr->src[1],
 175                    &element_intr->instr);
 176    }
 177
 178    nir_builder_instr_insert(b, &element_intr->instr);
 179
 180    /* Remove the old load intrinsic */
 181    nir_instr_remove(&intr->instr);
 182 }
 183
 184 static bool
 185 deref_has_indirect(nir_builder *b, nir_variable *var, nir_deref_path *path)
 186 {
 187    assert(path->path[0]->deref_type == nir_deref_type_var);
 188    nir_deref_instr **p = &path->path[1];
 189
 190    if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 191       p++;
 192    }
 193
 194    for (; *p; p++) {
 195       if ((*p)->deref_type != nir_deref_type_array)
 196          continue;
 197
 198       if (!nir_src_as_const_value((*p)->arr.index))
 199          return true;
 200    }
 201
 202    return false;
 203 }
 204
 205 /* Creates a mask of locations that contains arrays that are indexed via
 206  * indirect indexing.
 207  */
 208 static void
 209 create_indirects_mask(nir_shader *shader, uint64_t *indirects,
 210                       uint64_t *patch_indirects, nir_variable_mode mode)
 211 {
 212    nir_foreach_function(function, shader) {
 213       if (function->impl) {
 214          nir_builder b;
 215          nir_builder_init(&b, function->impl);
 216
 217          nir_foreach_block(block, function->impl) {
 218             nir_foreach_instr_safe(instr, block) {
 219
 220                if (instr->type != nir_instr_type_intrinsic)
 221                   continue;
 222
 223                nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 224
 225                if (intr->intrinsic != nir_intrinsic_load_deref &&
 226                    intr->intrinsic != nir_intrinsic_store_deref &&
 227                    intr->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
 228                    intr->intrinsic != nir_intrinsic_interp_deref_at_sample &&
 229                    intr->intrinsic != nir_intrinsic_interp_deref_at_offset)
 230                   continue;
 231
 232                nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
 233                nir_variable *var = nir_deref_instr_get_variable(deref);
 234
 235                if (var->data.mode != mode)
 236                   continue;
 237
 238                nir_deref_path path;
 239                nir_deref_path_init(&path, deref, NULL);
 240
 241                uint64_t loc_mask = ((uint64_t)1) << var->data.location;
 242                if (var->data.patch) {
 243                   if (deref_has_indirect(&b, var, &path))
 244                      patch_indirects[var->data.location_frac] |= loc_mask;
 245                } else {
 246                   if (deref_has_indirect(&b, var, &path))
 247                      indirects[var->data.location_frac] |= loc_mask;
 248                }
 249
 250                nir_deref_path_finish(&path);
 251             }
 252          }
 253       }
 254    }
 255 }
 256
 257 static void
 258 lower_io_arrays_to_elements(nir_shader *shader, nir_variable_mode mask,
 259                             uint64_t *indirects, uint64_t *patch_indirects,
 260                             struct hash_table *varyings,
 261                             bool after_cross_stage_opts)
 262 {
 263    nir_foreach_function(function, shader) {
 264       if (function->impl) {
 265          nir_builder b;
 266          nir_builder_init(&b, function->impl);
 267
 268          nir_foreach_block(block, function->impl) {
 269             nir_foreach_instr_safe(instr, block) {
 270                if (instr->type != nir_instr_type_intrinsic)
 271                   continue;
 272
 273                nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 274
 275                if (intr->intrinsic != nir_intrinsic_load_deref &&
 276                    intr->intrinsic != nir_intrinsic_store_deref &&
 277                    intr->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
 278                    intr->intrinsic != nir_intrinsic_interp_deref_at_sample &&
 279                    intr->intrinsic != nir_intrinsic_interp_deref_at_offset)
 280                   continue;
 281
 282                nir_variable *var =
 283                   nir_deref_instr_get_variable(nir_src_as_deref(intr->src[0]));
 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 =
 347       _mesa_hash_table_create(NULL, _mesa_hash_pointer,
 348                               _mesa_key_pointer_equal);
 349    struct hash_table *split_outputs =
 350       _mesa_hash_table_create(NULL, _mesa_hash_pointer,
 351                               _mesa_key_pointer_equal);
 352
 353    uint64_t indirects[4] = {0}, patch_indirects[4] = {0};
 354
 355    lower_io_arrays_to_elements(shader, nir_var_shader_out, indirects,
 356                                patch_indirects, split_outputs, true);
 357
 358    if (!outputs_only) {
 359       lower_io_arrays_to_elements(shader, nir_var_shader_in, indirects,
 360                                   patch_indirects, split_inputs, true);
 361
 362       /* Remove old input from the shaders inputs list */
 363       struct hash_entry *entry;
 364       hash_table_foreach(split_inputs, entry) {
 365          nir_variable *var = (nir_variable *) entry->key;
 366          exec_node_remove(&var->node);
 367
 368          free(entry->data);
 369       }
 370    }
 371
 372    /* Remove old output from the shaders outputs list */
 373    struct hash_entry *entry;
 374    hash_table_foreach(split_outputs, entry) {
 375       nir_variable *var = (nir_variable *) entry->key;
 376       exec_node_remove(&var->node);
 377
 378       free(entry->data);
 379    }
 380
 381    _mesa_hash_table_destroy(split_inputs, NULL);
 382    _mesa_hash_table_destroy(split_outputs, NULL);
 383
 384    nir_remove_dead_derefs(shader);
 385 }
 386
 387 void
 388 nir_lower_io_arrays_to_elements(nir_shader *producer, nir_shader *consumer)
 389 {
 390    struct hash_table *split_inputs =
 391       _mesa_hash_table_create(NULL, _mesa_hash_pointer,
 392                               _mesa_key_pointer_equal);
 393    struct hash_table *split_outputs =
 394       _mesa_hash_table_create(NULL, _mesa_hash_pointer,
 395                               _mesa_key_pointer_equal);
 396
 397    uint64_t indirects[4] = {0}, patch_indirects[4] = {0};
 398    create_indirects_mask(producer, indirects, patch_indirects,
 399                          nir_var_shader_out);
 400    create_indirects_mask(consumer, indirects, patch_indirects,
 401                          nir_var_shader_in);
 402
 403    lower_io_arrays_to_elements(producer, nir_var_shader_out, indirects,
 404                                patch_indirects, split_outputs, false);
 405
 406    lower_io_arrays_to_elements(consumer, nir_var_shader_in, indirects,
 407                                patch_indirects, split_inputs, false);
 408
 409    /* Remove old input from the shaders inputs list */
 410    struct hash_entry *entry;
 411    hash_table_foreach(split_inputs, entry) {
 412       nir_variable *var = (nir_variable *) entry->key;
 413       exec_node_remove(&var->node);
 414
 415       free(entry->data);
 416    }
 417
 418    /* Remove old output from the shaders outputs list */
 419    hash_table_foreach(split_outputs, entry) {
 420       nir_variable *var = (nir_variable *) entry->key;
 421       exec_node_remove(&var->node);
 422
 423       free(entry->data);
 424    }
 425
 426    _mesa_hash_table_destroy(split_inputs, NULL);
 427    _mesa_hash_table_destroy(split_outputs, NULL);
 428
 429    nir_remove_dead_derefs(producer);
 430    nir_remove_dead_derefs(consumer);
 431 }