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, unsigned *xfb_offset,
  38               nir_ssa_def **vertex_index)
  39 {
  40    nir_deref_path path;
  41    nir_deref_path_init(&path, deref, NULL);
  42
  43    assert(path.path[0]->deref_type == nir_deref_type_var);
  44    nir_deref_instr **p = &path.path[1];
  45
  46    /* For per-vertex input arrays (i.e. geometry shader inputs), skip the
  47     * outermost array index.  Process the rest normally.
  48     */
  49    if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
  50       *vertex_index = nir_ssa_for_src(b, (*p)->arr.index, 1);
  51       p++;
  52    }
  53
  54    unsigned offset = 0;
  55    *xfb_offset = 0;
  56    for (; *p; p++) {
  57       if ((*p)->deref_type == nir_deref_type_array) {
  58          /* must not be indirect dereference */
  59          unsigned index = nir_src_as_uint((*p)->arr.index);
  60
  61          unsigned size = glsl_count_attribute_slots((*p)->type, false);
  62          offset += size * index;
  63
  64          xfb_offset += index * glsl_get_component_slots((*p)->type) * 4;
  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 * index;
  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 xfb_offset = 0;
 124    unsigned io_offset = get_io_offset(b, nir_src_as_deref(intr->src[0]),
 125                                       var, &elements_index, &xfb_offset,
 126                                       &vertex_index);
 127
 128    nir_variable *element = elements[elements_index];
 129    if (!element) {
 130          element = nir_variable_clone(var, b->shader);
 131          element->data.location =  var->data.location + io_offset;
 132
 133          if (var->data.explicit_offset)
 134             element->data.offset = var->data.offset + xfb_offset;
 135
 136          const struct glsl_type *type = glsl_without_array(element->type);
 137
 138          /* This pass also splits matrices so we need give them a new type. */
 139          if (glsl_type_is_matrix(type))
 140             type = glsl_get_column_type(type);
 141
 142          if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 143             type = glsl_array_type(type, glsl_get_length(element->type),
 144                                    glsl_get_explicit_stride(element->type));
 145          }
 146
 147          element->type = type;
 148          elements[elements_index] = element;
 149
 150          nir_shader_add_variable(b->shader, element);
 151    }
 152
 153    nir_deref_instr *element_deref = nir_build_deref_var(b, element);
 154
 155    if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 156       assert(vertex_index);
 157       element_deref = nir_build_deref_array(b, element_deref, vertex_index);
 158    }
 159
 160    nir_intrinsic_instr *element_intr =
 161       nir_intrinsic_instr_create(b->shader, intr->intrinsic);
 162    element_intr->num_components = intr->num_components;
 163    element_intr->src[0] = nir_src_for_ssa(&element_deref->dest.ssa);
 164
 165    if (intr->intrinsic != nir_intrinsic_store_deref) {
 166       nir_ssa_dest_init(&element_intr->instr, &element_intr->dest,
 167                         intr->num_components, intr->dest.ssa.bit_size, NULL);
 168
 169       if (intr->intrinsic == nir_intrinsic_interp_deref_at_offset ||
 170           intr->intrinsic == nir_intrinsic_interp_deref_at_sample ||
 171           intr->intrinsic == nir_intrinsic_interp_deref_at_vertex) {
 172          nir_src_copy(&element_intr->src[1], &intr->src[1],
 173                       &element_intr->instr);
 174       }
 175
 176       nir_ssa_def_rewrite_uses(&intr->dest.ssa,
 177                                nir_src_for_ssa(&element_intr->dest.ssa));
 178    } else {
 179       nir_intrinsic_set_write_mask(element_intr,
 180                                    nir_intrinsic_write_mask(intr));
 181       nir_src_copy(&element_intr->src[1], &intr->src[1],
 182                    &element_intr->instr);
 183    }
 184
 185    nir_builder_instr_insert(b, &element_intr->instr);
 186
 187    /* Remove the old load intrinsic */
 188    nir_instr_remove(&intr->instr);
 189 }
 190
 191 static bool
 192 deref_has_indirect(nir_builder *b, nir_variable *var, nir_deref_path *path)
 193 {
 194    assert(path->path[0]->deref_type == nir_deref_type_var);
 195    nir_deref_instr **p = &path->path[1];
 196
 197    if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
 198       p++;
 199    }
 200
 201    for (; *p; p++) {
 202       if ((*p)->deref_type != nir_deref_type_array)
 203          continue;
 204
 205       if (!nir_src_is_const((*p)->arr.index))
 206          return true;
 207    }
 208
 209    return false;
 210 }
 211
 212 /* Creates a mask of locations that contains arrays that are indexed via
 213  * indirect indexing.
 214  */
 215 static void
 216 create_indirects_mask(nir_shader *shader,
 217                       BITSET_WORD *indirects, nir_variable_mode mode)
 218 {
 219    nir_foreach_function(function, shader) {
 220       if (function->impl) {
 221          nir_builder b;
 222          nir_builder_init(&b, function->impl);
 223
 224          nir_foreach_block(block, function->impl) {
 225             nir_foreach_instr_safe(instr, block) {
 226
 227                if (instr->type != nir_instr_type_intrinsic)
 228                   continue;
 229
 230                nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 231
 232                if (intr->intrinsic != nir_intrinsic_load_deref &&
 233                    intr->intrinsic != nir_intrinsic_store_deref &&
 234                    intr->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
 235                    intr->intrinsic != nir_intrinsic_interp_deref_at_sample &&
 236                    intr->intrinsic != nir_intrinsic_interp_deref_at_offset &&
 237                    intr->intrinsic != nir_intrinsic_interp_deref_at_vertex)
 238                   continue;
 239
 240                nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
 241                if (deref->mode != mode)
 242                   continue;
 243
 244                nir_variable *var = nir_deref_instr_get_variable(deref);
 245
 246                nir_deref_path path;
 247                nir_deref_path_init(&path, deref, NULL);
 248
 249                int loc = var->data.location * 4 + var->data.location_frac;
 250                if (deref_has_indirect(&b, var, &path))
 251                   BITSET_SET(indirects, loc);
 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                             BITSET_WORD *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                    intr->intrinsic != nir_intrinsic_interp_deref_at_vertex)
 284                   continue;
 285
 286                nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
 287                if (!(deref->mode & mask))
 288                   continue;
 289
 290                nir_variable *var = nir_deref_instr_get_variable(deref);
 291
 292                /* Drivers assume compact arrays are, in fact, arrays. */
 293                if (var->data.compact)
 294                   continue;
 295
 296                /* Skip indirects */
 297                int loc = var->data.location * 4 + var->data.location_frac;
 298                if (BITSET_TEST(indirects, loc))
 299                   continue;
 300
 301                nir_variable_mode mode = var->data.mode;
 302
 303                const struct glsl_type *type = var->type;
 304                if (nir_is_per_vertex_io(var, b.shader->info.stage)) {
 305                   assert(glsl_type_is_array(type));
 306                   type = glsl_get_array_element(type);
 307                }
 308
 309                /* Skip types we cannot split.
 310                 *
 311                 * TODO: Add support for struct splitting.
 312                 */
 313                if ((!glsl_type_is_array(type) && !glsl_type_is_matrix(type))||
 314                    glsl_type_is_struct_or_ifc(glsl_without_array(type)))
 315                   continue;
 316
 317                /* Skip builtins */
 318                if (!after_cross_stage_opts &&
 319                    var->data.location < VARYING_SLOT_VAR0 &&
 320                    var->data.location >= 0)
 321                   continue;
 322
 323                /* Don't bother splitting if we can't opt away any unused
 324                 * elements.
 325                 */
 326                if (!after_cross_stage_opts && var->data.always_active_io)
 327                   continue;
 328
 329                switch (intr->intrinsic) {
 330                case nir_intrinsic_interp_deref_at_centroid:
 331                case nir_intrinsic_interp_deref_at_sample:
 332                case nir_intrinsic_interp_deref_at_offset:
 333                case nir_intrinsic_interp_deref_at_vertex:
 334                case nir_intrinsic_load_deref:
 335                case nir_intrinsic_store_deref:
 336                   if ((mask & nir_var_shader_in && mode == nir_var_shader_in) ||
 337                       (mask & nir_var_shader_out && mode == nir_var_shader_out))
 338                      lower_array(&b, intr, var, varyings);
 339                   break;
 340                default:
 341                   break;
 342                }
 343             }
 344          }
 345       }
 346    }
 347 }
 348
 349 void
 350 nir_lower_io_arrays_to_elements_no_indirects(nir_shader *shader,
 351                                              bool outputs_only)
 352 {
 353    struct hash_table *split_inputs = _mesa_pointer_hash_table_create(NULL);
 354    struct hash_table *split_outputs = _mesa_pointer_hash_table_create(NULL);
 355
 356    BITSET_DECLARE(indirects, 4 * VARYING_SLOT_TESS_MAX) = {0};
 357
 358    lower_io_arrays_to_elements(shader, nir_var_shader_out,
 359                                indirects, split_outputs, true);
 360
 361    if (!outputs_only) {
 362       lower_io_arrays_to_elements(shader, nir_var_shader_in,
 363                                   indirects, split_inputs, true);
 364
 365       /* Remove old input from the shaders inputs list */
 366       hash_table_foreach(split_inputs, entry) {
 367          nir_variable *var = (nir_variable *) entry->key;
 368          exec_node_remove(&var->node);
 369
 370          free(entry->data);
 371       }
 372    }
 373
 374    /* Remove old output from the shaders outputs list */
 375    hash_table_foreach(split_outputs, entry) {
 376       nir_variable *var = (nir_variable *) entry->key;
 377       exec_node_remove(&var->node);
 378
 379       free(entry->data);
 380    }
 381
 382    _mesa_hash_table_destroy(split_inputs, NULL);
 383    _mesa_hash_table_destroy(split_outputs, NULL);
 384
 385    nir_remove_dead_derefs(shader);
 386 }
 387
 388 void
 389 nir_lower_io_arrays_to_elements(nir_shader *producer, nir_shader *consumer)
 390 {
 391    struct hash_table *split_inputs = _mesa_pointer_hash_table_create(NULL);
 392    struct hash_table *split_outputs = _mesa_pointer_hash_table_create(NULL);
 393
 394    BITSET_DECLARE(indirects, 4 * VARYING_SLOT_TESS_MAX) = {0};
 395
 396    create_indirects_mask(producer, indirects, nir_var_shader_out);
 397    create_indirects_mask(consumer, indirects, nir_var_shader_in);
 398
 399    lower_io_arrays_to_elements(producer, nir_var_shader_out,
 400                                indirects, split_outputs, false);
 401
 402    lower_io_arrays_to_elements(consumer, nir_var_shader_in,
 403                                indirects, split_inputs, false);
 404
 405    /* Remove old input from the shaders inputs list */
 406    hash_table_foreach(split_inputs, entry) {
 407       nir_variable *var = (nir_variable *) entry->key;
 408       exec_node_remove(&var->node);
 409
 410       free(entry->data);
 411    }
 412
 413    /* Remove old output from the shaders outputs list */
 414    hash_table_foreach(split_outputs, entry) {
 415       nir_variable *var = (nir_variable *) entry->key;
 416       exec_node_remove(&var->node);
 417
 418       free(entry->data);
 419    }
 420
 421    _mesa_hash_table_destroy(split_inputs, NULL);
 422    _mesa_hash_table_destroy(split_outputs, NULL);
 423
 424    nir_remove_dead_derefs(producer);
 425    nir_remove_dead_derefs(consumer);
 426 }