src/glsl/nir/nir_lower_io.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  *    Connor Abbott (cwabbott0@gmail.com)
  25  *    Jason Ekstrand (jason@jlekstrand.net)
  26  *
  27  */
  28
  29 /*
  30  * This lowering pass converts references to input/output variables with
  31  * loads/stores to actual input/output intrinsics.
  32  */
  33
  34 #include "nir.h"
  35 #include "nir_builder.h"
  36
  37 struct lower_io_state {
  38    nir_builder builder;
  39    void *mem_ctx;
  40    bool is_scalar;
  41 };
  42
  43 static int
  44 type_size_vec4(const struct glsl_type *type)
  45 {
  46    unsigned int i;
  47    int size;
  48
  49    switch (glsl_get_base_type(type)) {
  50    case GLSL_TYPE_UINT:
  51    case GLSL_TYPE_INT:
  52    case GLSL_TYPE_FLOAT:
  53    case GLSL_TYPE_BOOL:
  54       if (glsl_type_is_matrix(type)) {
  55          return glsl_get_matrix_columns(type);
  56       } else {
  57          return 1;
  58       }
  59    case GLSL_TYPE_ARRAY:
  60       return type_size_vec4(glsl_get_array_element(type)) * glsl_get_length(type);
  61    case GLSL_TYPE_STRUCT:
  62       size = 0;
  63       for (i = 0; i <  glsl_get_length(type); i++) {
  64          size += type_size_vec4(glsl_get_struct_field(type, i));
  65       }
  66       return size;
  67    case GLSL_TYPE_SUBROUTINE:
  68       return 1;
  69    case GLSL_TYPE_SAMPLER:
  70       return 0;
  71    case GLSL_TYPE_ATOMIC_UINT:
  72       return 0;
  73    case GLSL_TYPE_IMAGE:
  74    case GLSL_TYPE_VOID:
  75    case GLSL_TYPE_DOUBLE:
  76    case GLSL_TYPE_ERROR:
  77    case GLSL_TYPE_INTERFACE:
  78       unreachable("not reached");
  79    }
  80
  81    return 0;
  82 }
  83
  84 static unsigned
  85 type_size_scalar(const struct glsl_type *type)
  86 {
  87    unsigned int size, i;
  88
  89    switch (glsl_get_base_type(type)) {
  90    case GLSL_TYPE_UINT:
  91    case GLSL_TYPE_INT:
  92    case GLSL_TYPE_FLOAT:
  93    case GLSL_TYPE_BOOL:
  94       return glsl_get_components(type);
  95    case GLSL_TYPE_ARRAY:
  96       return type_size_scalar(glsl_get_array_element(type)) * glsl_get_length(type);
  97    case GLSL_TYPE_STRUCT:
  98       size = 0;
  99       for (i = 0; i < glsl_get_length(type); i++) {
 100          size += type_size_scalar(glsl_get_struct_field(type, i));
 101       }
 102       return size;
 103    case GLSL_TYPE_SUBROUTINE:
 104       return 1;
 105    case GLSL_TYPE_SAMPLER:
 106       return 0;
 107    case GLSL_TYPE_ATOMIC_UINT:
 108       return 0;
 109    case GLSL_TYPE_INTERFACE:
 110       return 0;
 111    case GLSL_TYPE_IMAGE:
 112       return 0;
 113    case GLSL_TYPE_VOID:
 114    case GLSL_TYPE_ERROR:
 115    case GLSL_TYPE_DOUBLE:
 116       unreachable("not reached");
 117    }
 118
 119    return 0;
 120 }
 121
 122 static unsigned
 123 type_size(const struct glsl_type *type, bool is_scalar)
 124 {
 125    if (is_scalar)
 126       return type_size_scalar(type);
 127    else
 128       return type_size_vec4(type);
 129 }
 130
 131 void
 132 nir_assign_var_locations(struct exec_list *var_list, unsigned *size, bool is_scalar)
 133 {
 134    unsigned location = 0;
 135
 136    foreach_list_typed(nir_variable, var, node, var_list) {
 137       /*
 138        * UBO's have their own address spaces, so don't count them towards the
 139        * number of global uniforms
 140        */
 141       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
 142           var->interface_type != NULL)
 143          continue;
 144
 145       var->data.driver_location = location;
 146       location += type_size(var->type, is_scalar);
 147    }
 148
 149    *size = location;
 150 }
 151
 152 static bool
 153 deref_has_indirect(nir_deref_var *deref)
 154 {
 155    for (nir_deref *tail = deref->deref.child; tail; tail = tail->child) {
 156       if (tail->deref_type == nir_deref_type_array) {
 157          nir_deref_array *arr = nir_deref_as_array(tail);
 158          if (arr->deref_array_type == nir_deref_array_type_indirect)
 159             return true;
 160       }
 161    }
 162
 163    return false;
 164 }
 165
 166 static bool
 167 mark_indirect_uses_block(nir_block *block, void *void_state)
 168 {
 169    struct set *indirect_set = void_state;
 170
 171    nir_foreach_instr(block, instr) {
 172       if (instr->type != nir_instr_type_intrinsic)
 173          continue;
 174
 175       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 176
 177       for (unsigned i = 0;
 178            i < nir_intrinsic_infos[intrin->intrinsic].num_variables; i++) {
 179          if (deref_has_indirect(intrin->variables[i]))
 180             _mesa_set_add(indirect_set, intrin->variables[i]->var);
 181       }
 182    }
 183
 184    return true;
 185 }
 186
 187 /* Identical to nir_assign_var_locations_packed except that it assigns
 188  * locations to the variables that are used 100% directly first and then
 189  * assigns locations to variables that are used indirectly.
 190  */
 191 void
 192 nir_assign_var_locations_direct_first(nir_shader *shader,
 193                                       struct exec_list *var_list,
 194                                       unsigned *direct_size,
 195                                       unsigned *size,
 196                                       bool is_scalar)
 197 {
 198    struct set *indirect_set = _mesa_set_create(NULL, _mesa_hash_pointer,
 199                                                _mesa_key_pointer_equal);
 200
 201    nir_foreach_overload(shader, overload) {
 202       if (overload->impl)
 203          nir_foreach_block(overload->impl, mark_indirect_uses_block,
 204                            indirect_set);
 205    }
 206
 207    unsigned location = 0;
 208
 209    foreach_list_typed(nir_variable, var, node, var_list) {
 210       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
 211           var->interface_type != NULL)
 212          continue;
 213
 214       if (_mesa_set_search(indirect_set, var))
 215          continue;
 216
 217       var->data.driver_location = location;
 218       location += type_size(var->type, is_scalar);
 219    }
 220
 221    *direct_size = location;
 222
 223    foreach_list_typed(nir_variable, var, node, var_list) {
 224       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
 225           var->interface_type != NULL)
 226          continue;
 227
 228       if (!_mesa_set_search(indirect_set, var))
 229          continue;
 230
 231       var->data.driver_location = location;
 232       location += type_size(var->type, is_scalar);
 233    }
 234
 235    *size = location;
 236
 237    _mesa_set_destroy(indirect_set, NULL);
 238 }
 239
 240 static unsigned
 241 get_io_offset(nir_deref_var *deref, nir_instr *instr, nir_src *indirect,
 242               struct lower_io_state *state)
 243 {
 244    bool found_indirect = false;
 245    unsigned base_offset = 0;
 246
 247    nir_builder *b = &state->builder;
 248    nir_builder_insert_before_instr(b, instr);
 249
 250    nir_deref *tail = &deref->deref;
 251    while (tail->child != NULL) {
 252       const struct glsl_type *parent_type = tail->type;
 253       tail = tail->child;
 254
 255       if (tail->deref_type == nir_deref_type_array) {
 256          nir_deref_array *deref_array = nir_deref_as_array(tail);
 257          unsigned size = type_size(tail->type, state->is_scalar);
 258
 259          base_offset += size * deref_array->base_offset;
 260
 261          if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 262             nir_ssa_def *mul =
 263                nir_imul(b, nir_imm_int(b, size),
 264                         nir_ssa_for_src(b, deref_array->indirect, 1));
 265
 266             if (found_indirect) {
 267                indirect->ssa =
 268                   nir_iadd(b, nir_ssa_for_src(b, *indirect, 1), mul);
 269             } else {
 270                indirect->ssa = mul;
 271             }
 272             indirect->is_ssa = true;
 273             found_indirect = true;
 274          }
 275       } else if (tail->deref_type == nir_deref_type_struct) {
 276          nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
 277
 278          for (unsigned i = 0; i < deref_struct->index; i++)
 279             base_offset += type_size(glsl_get_struct_field(parent_type, i),
 280                                      state->is_scalar);
 281       }
 282    }
 283
 284    return base_offset;
 285 }
 286
 287 static nir_intrinsic_op
 288 load_op(nir_variable_mode mode, bool has_indirect)
 289 {
 290    nir_intrinsic_op op;
 291    switch (mode) {
 292    case nir_var_shader_in:
 293       op = has_indirect ? nir_intrinsic_load_input_indirect :
 294                           nir_intrinsic_load_input;
 295       break;
 296    case nir_var_uniform:
 297       op = has_indirect ? nir_intrinsic_load_uniform_indirect :
 298                           nir_intrinsic_load_uniform;
 299       break;
 300    default:
 301       unreachable("Unknown variable mode");
 302    }
 303    return op;
 304 }
 305
 306 static bool
 307 nir_lower_io_block(nir_block *block, void *void_state)
 308 {
 309    struct lower_io_state *state = void_state;
 310
 311    nir_foreach_instr_safe(block, instr) {
 312       if (instr->type != nir_instr_type_intrinsic)
 313          continue;
 314
 315       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 316
 317       switch (intrin->intrinsic) {
 318       case nir_intrinsic_load_var: {
 319          nir_variable_mode mode = intrin->variables[0]->var->data.mode;
 320          if (mode != nir_var_shader_in && mode != nir_var_uniform)
 321             continue;
 322
 323          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 324
 325          nir_intrinsic_instr *load =
 326             nir_intrinsic_instr_create(state->mem_ctx,
 327                                        load_op(mode, has_indirect));
 328          load->num_components = intrin->num_components;
 329
 330          nir_src indirect;
 331          unsigned offset = get_io_offset(intrin->variables[0],
 332                                          &intrin->instr, &indirect, state);
 333          offset += intrin->variables[0]->var->data.driver_location;
 334
 335          load->const_index[0] = offset;
 336
 337          if (has_indirect)
 338             load->src[0] = indirect;
 339
 340          if (intrin->dest.is_ssa) {
 341             nir_ssa_dest_init(&load->instr, &load->dest,
 342                               intrin->num_components, NULL);
 343             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 344                                      nir_src_for_ssa(&load->dest.ssa),
 345                                      state->mem_ctx);
 346          } else {
 347             nir_dest_copy(&load->dest, &intrin->dest, state->mem_ctx);
 348          }
 349
 350          nir_instr_insert_before(&intrin->instr, &load->instr);
 351          nir_instr_remove(&intrin->instr);
 352          break;
 353       }
 354
 355       case nir_intrinsic_store_var: {
 356          if (intrin->variables[0]->var->data.mode != nir_var_shader_out)
 357             continue;
 358
 359          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 360
 361          nir_intrinsic_op store_op;
 362          if (has_indirect) {
 363             store_op = nir_intrinsic_store_output_indirect;
 364          } else {
 365             store_op = nir_intrinsic_store_output;
 366          }
 367
 368          nir_intrinsic_instr *store = nir_intrinsic_instr_create(state->mem_ctx,
 369                                                                  store_op);
 370          store->num_components = intrin->num_components;
 371
 372          nir_src indirect;
 373          unsigned offset = get_io_offset(intrin->variables[0],
 374                                          &intrin->instr, &indirect, state);
 375          offset += intrin->variables[0]->var->data.driver_location;
 376
 377          store->const_index[0] = offset;
 378
 379          nir_src_copy(&store->src[0], &intrin->src[0], state->mem_ctx);
 380
 381          if (has_indirect)
 382             store->src[1] = indirect;
 383
 384          nir_instr_insert_before(&intrin->instr, &store->instr);
 385          nir_instr_remove(&intrin->instr);
 386          break;
 387       }
 388
 389       default:
 390          break;
 391       }
 392    }
 393
 394    return true;
 395 }
 396
 397 static void
 398 nir_lower_io_impl(nir_function_impl *impl, bool is_scalar)
 399 {
 400    struct lower_io_state state;
 401
 402    nir_builder_init(&state.builder, impl);
 403    state.mem_ctx = ralloc_parent(impl);
 404    state.is_scalar = is_scalar;
 405
 406    nir_foreach_block(impl, nir_lower_io_block, &state);
 407
 408    nir_metadata_preserve(impl, nir_metadata_block_index |
 409                                nir_metadata_dominance);
 410 }
 411
 412 void
 413 nir_lower_io(nir_shader *shader, bool is_scalar)
 414 {
 415    nir_foreach_overload(shader, overload) {
 416       if (overload->impl)
 417          nir_lower_io_impl(overload->impl, is_scalar);
 418    }
 419 }