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    bool vs_in = (b->shader->info.stage == MESA_SHADER_VERTEX) &&
  40                 (var->data.mode == nir_var_shader_in);
  41
  42    nir_deref_path path;
  43    nir_deref_path_init(&path, deref, NULL);
  44
  45    assert(path.path[0]->deref_type == nir_deref_type_var);
  46    nir_deref_instr **p = &path.path[1];
  47
  48    /* For per-vertex input arrays (i.e. geometry shader inputs), skip the
  49     * outermost array index.  Process the rest normally.
  50     */
  51    if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
  52       *vertex_index = nir_ssa_for_src(b, (*p)->arr.index, 1);
  53       p++;
  54    }
  55
  56    unsigned offset = 0;
  57    for (; *p; p++) {
  58       if ((*p)->deref_type == nir_deref_type_array) {
  59          nir_const_value *c = nir_src_as_const_value((*p)->arr.index);
  60
  61          assert(c);     /* must not be indirect dereference */
  62
  63          unsigned size = glsl_count_attribute_slots((*p)->type, vs_in);
  64          offset += size * c->u32[0];
  65
  66          unsigned num_elements = glsl_type_is_array((*p)->type) ?
  67             glsl_get_aoa_size((*p)->type) : 1;
  68
  69          num_elements *= glsl_type_is_matrix(glsl_without_array((*p)->type)) ?
  70             glsl_get_matrix_columns(glsl_without_array((*p)->type)) : 1;
  71
  72          *element_index += num_elements * c->u32[0];
  73       } else if ((*p)->deref_type == nir_deref_type_struct) {
  74          /* TODO: we could also add struct splitting support to this pass */
  75          break;
  76       }
  77    }
  78
  79    nir_deref_path_finish(&path);
  80
  81    return offset;
  82 }
  83
  84 static nir_variable **
  85 get_array_elements(struct hash_table *ht, nir_variable *var,
  86                    gl_shader_stage stage)
  87 {
  88    nir_variable **elements;
  89    struct hash_entry *entry = _mesa_hash_table_search(ht, var);
  90    if (!entry) {
  91       const struct glsl_type *type = var->type;
  92       if (nir_is_per_vertex_io(var, stage)) {
  93          assert(glsl_type_is_array(type));
  94          type = glsl_get_array_element(type);
  95       }
  96
  97       unsigned num_elements = glsl_type_is_array(type) ?
  98          glsl_get_aoa_size(type) : 1;
  99
 100       num_elements *= glsl_type_is_matrix(glsl_without_array(type)) ?
 101          glsl_get_matrix_columns(glsl_without_array(type)) : 1;
 102
 103       elements = (nir_variable **) calloc(num_elements, sizeof(nir_variable *));
 104       _mesa_hash_table_insert(ht, var, elements);
 105    } else {
 106       elements = (nir_variable **) entry->data;
 107    }
 108
 109    return elements;
 110 }
 111
 112 static void
 113 lower_array(nir_builder *b, nir_intrinsic_instr *intr, nir_variable *var,
 114             struct hash_table *varyings)
 115 {
 116    b->cursor = nir_before_instr(&intr->instr);
 117
 118    nir_variable **elements =
 119       get_array_elements(varyings, var, b->shader->info.stage);
 120
 121    nir_ssa_def *vertex_index = NULL;
 122    unsigned elements_index = 0;
 123    unsigned io_offset = get_io_offset(b, nir_src_as_deref(intr->src[0]),
 124                                       var, &elements_index, &vertex_index);
 125
 126    nir_variable *element = elements[elements_index];
 127    if (!element) {
 128          element = nir_variable_clone(var, b->shader);
 129          element->data.location =  var->data.location + io_offset;
 130
 131          const struct glsl_type *type = glsl_without_array(element->type);
 132
 133          /* This pass also splits matrices so we need give them a new type. */
 134          if (glsl_type_is_matrix(type)) {
 135             type = glsl_vector_type(glsl_get_base_type(type),
 136                                     glsl_get_vector_elements(type));
 137          }
 138
 139          if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 140             type = glsl_get_array_instance(type,
 141                                            glsl_get_length(element->type));
 142          }
 143
 144          element->type = type;
 145          elements[elements_index] = element;
 146
 147          nir_shader_add_variable(b->shader, element);
 148    }
 149
 150    nir_deref_instr *element_deref = nir_build_deref_var(b, element);
 151
 152    if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 153       assert(vertex_index);
 154       element_deref = nir_build_deref_array(b, element_deref, vertex_index);
 155    }
 156
 157    nir_intrinsic_instr *element_intr =
 158       nir_intrinsic_instr_create(b->shader, intr->intrinsic);
 159    element_intr->num_components = intr->num_components;
 160    element_intr->src[0] = nir_src_for_ssa(&element_deref->dest.ssa);
 161
 162    if (intr->intrinsic != nir_intrinsic_store_deref) {
 163       nir_ssa_dest_init(&element_intr->instr, &element_intr->dest,
 164                         intr->num_components, intr->dest.ssa.bit_size, NULL);
 165
 166       if (intr->intrinsic == nir_intrinsic_interp_deref_at_offset ||
 167           intr->intrinsic == nir_intrinsic_interp_deref_at_sample) {
 168          nir_src_copy(&element_intr->src[1], &intr->src[1],
 169                       &element_intr->instr);
 170       }
 171
 172       nir_ssa_def_rewrite_uses(&intr->dest.ssa,
 173                                nir_src_for_ssa(&element_intr->dest.ssa));
 174    } else {
 175       nir_intrinsic_set_write_mask(element_intr,
 176                                    nir_intrinsic_write_mask(intr));
 177       nir_src_copy(&element_intr->src[1], &intr->src[1],
 178                    &element_intr->instr);
 179    }
 180
 181    nir_builder_instr_insert(b, &element_intr->instr);
 182
 183    /* Remove the old load intrinsic */
 184    nir_instr_remove(&intr->instr);
 185 }
 186
 187 static bool
 188 deref_has_indirect(nir_builder *b, nir_variable *var, nir_deref_path *path)
 189 {
 190    assert(path->path[0]->deref_type == nir_deref_type_var);
 191    nir_deref_instr **p = &path->path[1];
 192
 193    if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 194       p++;
 195    }
 196
 197    for (; *p; p++) {
 198       if ((*p)->deref_type != nir_deref_type_array)
 199          continue;
 200
 201       if (!nir_src_as_const_value((*p)->arr.index))
 202          return true;
 203    }
 204
 205    return false;
 206 }
 207
 208 /* Creates a mask of locations that contains arrays that are indexed via
 209  * indirect indexing.
 210  */
 211 static void
 212 create_indirects_mask(nir_shader *shader, uint64_t *indirects,
 213                       uint64_t *patch_indirects, nir_variable_mode mode)
 214 {
 215    nir_foreach_function(function, shader) {
 216       if (function->impl) {
 217          nir_builder b;
 218          nir_builder_init(&b, function->impl);
 219
 220          nir_foreach_block(block, function->impl) {
 221             nir_foreach_instr_safe(instr, block) {
 222
 223                if (instr->type != nir_instr_type_intrinsic)
 224                   continue;
 225
 226                nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 227
 228                if (intr->intrinsic != nir_intrinsic_load_deref &&
 229                    intr->intrinsic != nir_intrinsic_store_deref &&
 230                    intr->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
 231                    intr->intrinsic != nir_intrinsic_interp_deref_at_sample &&
 232                    intr->intrinsic != nir_intrinsic_interp_deref_at_offset)
 233                   continue;
 234
 235                nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
 236                nir_variable *var = nir_deref_instr_get_variable(deref);
 237
 238                if (var->data.mode != mode)
 239                   continue;
 240
 241                nir_deref_path path;
 242                nir_deref_path_init(&path, deref, NULL);
 243
 244                uint64_t loc_mask = ((uint64_t)1) << var->data.location;
 245                if (var->data.patch) {
 246                   if (deref_has_indirect(&b, var, &path))
 247                      patch_indirects[var->data.location_frac] |= loc_mask;
 248                } else {
 249                   if (deref_has_indirect(&b, var, &path))
 250                      indirects[var->data.location_frac] |= loc_mask;
 251                }
 252
 253                nir_deref_path_finish(&path);
 254             }
 255          }
 256       }
 257    }
 258 }
 259
 260 static void
 261 lower_io_arrays_to_elements(nir_shader *shader, nir_variable_mode mask,
 262                             uint64_t *indirects, uint64_t *patch_indirects,
 263                             struct hash_table *varyings,
 264                             bool after_cross_stage_opts)
 265 {
 266    nir_foreach_function(function, shader) {
 267       if (function->impl) {
 268          nir_builder b;
 269          nir_builder_init(&b, function->impl);
 270
 271          nir_foreach_block(block, function->impl) {
 272             nir_foreach_instr_safe(instr, block) {
 273                if (instr->type != nir_instr_type_intrinsic)
 274                   continue;
 275
 276                nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 277
 278                if (intr->intrinsic != nir_intrinsic_load_deref &&
 279                    intr->intrinsic != nir_intrinsic_store_deref &&
 280                    intr->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
 281                    intr->intrinsic != nir_intrinsic_interp_deref_at_sample &&
 282                    intr->intrinsic != nir_intrinsic_interp_deref_at_offset)
 283                   continue;
 284
 285                nir_variable *var =
 286                   nir_deref_instr_get_variable(nir_src_as_deref(intr->src[0]));
 287
 288                /* Skip indirects */
 289                uint64_t loc_mask = ((uint64_t)1) << var->data.location;
 290                if (var->data.patch) {
 291                   if (patch_indirects[var->data.location_frac] & loc_mask)
 292                      continue;
 293                } else {
 294                   if (indirects[var->data.location_frac] & loc_mask)
 295                      continue;
 296                }
 297
 298                nir_variable_mode mode = var->data.mode;
 299
 300                const struct glsl_type *type = var->type;
 301                if (nir_is_per_vertex_io(var, b.shader->info.stage)) {
 302                   assert(glsl_type_is_array(type));
 303                   type = glsl_get_array_element(type);
 304                }
 305
 306                /* Skip types we cannot split.
 307                 *
 308                 * TODO: Add support for struct splitting.
 309                 */
 310                if ((!glsl_type_is_array(type) && !glsl_type_is_matrix(type))||
 311                    glsl_type_is_struct(glsl_without_array(type)))
 312                   continue;
 313
 314                /* Skip builtins */
 315                if (!after_cross_stage_opts &&
 316                    var->data.location < VARYING_SLOT_VAR0 &&
 317                    var->data.location >= 0)
 318                   continue;
 319
 320                /* Don't bother splitting if we can't opt away any unused
 321                 * elements.
 322                 */
 323                if (!after_cross_stage_opts && var->data.always_active_io)
 324                   continue;
 325
 326                switch (intr->intrinsic) {
 327                case nir_intrinsic_interp_deref_at_centroid:
 328                case nir_intrinsic_interp_deref_at_sample:
 329                case nir_intrinsic_interp_deref_at_offset:
 330                case nir_intrinsic_load_deref:
 331                case nir_intrinsic_store_deref:
 332                   if ((mask & nir_var_shader_in && mode == nir_var_shader_in) ||
 333                       (mask & nir_var_shader_out && mode == nir_var_shader_out))
 334                      lower_array(&b, intr, var, varyings);
 335                   break;
 336                default:
 337                   break;
 338                }
 339             }
 340          }
 341       }
 342    }
 343 }
 344
 345 void
 346 nir_lower_io_arrays_to_elements_no_indirects(nir_shader *shader,
 347                                              bool outputs_only)
 348 {
 349    struct hash_table *split_inputs =
 350       _mesa_hash_table_create(NULL, _mesa_hash_pointer,
 351                               _mesa_key_pointer_equal);
 352    struct hash_table *split_outputs =
 353       _mesa_hash_table_create(NULL, _mesa_hash_pointer,
 354                               _mesa_key_pointer_equal);
 355
 356    nir_assert_unlowered_derefs(shader, nir_lower_load_store_derefs | nir_lower_interp_derefs);
 357
 358    uint64_t indirects[4] = {0}, patch_indirects[4] = {0};
 359
 360    lower_io_arrays_to_elements(shader, nir_var_shader_out, indirects,
 361                                patch_indirects, split_outputs, true);
 362
 363    if (!outputs_only) {
 364       lower_io_arrays_to_elements(shader, nir_var_shader_in, indirects,
 365                                   patch_indirects, split_inputs, true);
 366
 367       /* Remove old input from the shaders inputs list */
 368       struct hash_entry *entry;
 369       hash_table_foreach(split_inputs, entry) {
 370          nir_variable *var = (nir_variable *) entry->key;
 371          exec_node_remove(&var->node);
 372
 373          free(entry->data);
 374       }
 375    }
 376
 377    /* Remove old output from the shaders outputs list */
 378    struct hash_entry *entry;
 379    hash_table_foreach(split_outputs, entry) {
 380       nir_variable *var = (nir_variable *) entry->key;
 381       exec_node_remove(&var->node);
 382
 383       free(entry->data);
 384    }
 385
 386    _mesa_hash_table_destroy(split_inputs, NULL);
 387    _mesa_hash_table_destroy(split_outputs, NULL);
 388
 389    nir_remove_dead_derefs(shader);
 390 }
 391
 392 void
 393 nir_lower_io_arrays_to_elements(nir_shader *producer, nir_shader *consumer)
 394 {
 395    struct hash_table *split_inputs =
 396       _mesa_hash_table_create(NULL, _mesa_hash_pointer,
 397                               _mesa_key_pointer_equal);
 398    struct hash_table *split_outputs =
 399       _mesa_hash_table_create(NULL, _mesa_hash_pointer,
 400                               _mesa_key_pointer_equal);
 401
 402    nir_assert_unlowered_derefs(producer, nir_lower_load_store_derefs | nir_lower_interp_derefs);
 403    nir_assert_unlowered_derefs(consumer, nir_lower_load_store_derefs | nir_lower_interp_derefs);
 404
 405    uint64_t indirects[4] = {0}, patch_indirects[4] = {0};
 406    create_indirects_mask(producer, indirects, patch_indirects,
 407                          nir_var_shader_out);
 408    create_indirects_mask(consumer, indirects, patch_indirects,
 409                          nir_var_shader_in);
 410
 411    lower_io_arrays_to_elements(producer, nir_var_shader_out, indirects,
 412                                patch_indirects, split_outputs, false);
 413
 414    lower_io_arrays_to_elements(consumer, nir_var_shader_in, indirects,
 415                                patch_indirects, split_inputs, false);
 416
 417    /* Remove old input from the shaders inputs list */
 418    struct hash_entry *entry;
 419    hash_table_foreach(split_inputs, entry) {
 420       nir_variable *var = (nir_variable *) entry->key;
 421       exec_node_remove(&var->node);
 422
 423       free(entry->data);
 424    }
 425
 426    /* Remove old output from the shaders outputs list */
 427    hash_table_foreach(split_outputs, entry) {
 428       nir_variable *var = (nir_variable *) entry->key;
 429       exec_node_remove(&var->node);
 430
 431       free(entry->data);
 432    }
 433
 434    _mesa_hash_table_destroy(split_inputs, NULL);
 435    _mesa_hash_table_destroy(split_outputs, NULL);
 436
 437    nir_remove_dead_derefs(producer);
 438    nir_remove_dead_derefs(consumer);
 439 }