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
  36 struct lower_io_state {
  37    void *mem_ctx;
  38    bool is_scalar;
  39 };
  40
  41 static int
  42 type_size_vec4(const struct glsl_type *type)
  43 {
  44    unsigned int i;
  45    int size;
  46
  47    switch (glsl_get_base_type(type)) {
  48    case GLSL_TYPE_UINT:
  49    case GLSL_TYPE_INT:
  50    case GLSL_TYPE_FLOAT:
  51    case GLSL_TYPE_BOOL:
  52       if (glsl_type_is_matrix(type)) {
  53          return glsl_get_matrix_columns(type);
  54       } else {
  55          return 1;
  56       }
  57    case GLSL_TYPE_ARRAY:
  58       return type_size_vec4(glsl_get_array_element(type)) * glsl_get_length(type);
  59    case GLSL_TYPE_STRUCT:
  60       size = 0;
  61       for (i = 0; i <  glsl_get_length(type); i++) {
  62          size += type_size_vec4(glsl_get_struct_field(type, i));
  63       }
  64       return size;
  65    case GLSL_TYPE_SAMPLER:
  66       return 0;
  67    case GLSL_TYPE_ATOMIC_UINT:
  68       return 0;
  69    case GLSL_TYPE_IMAGE:
  70    case GLSL_TYPE_VOID:
  71    case GLSL_TYPE_DOUBLE:
  72    case GLSL_TYPE_ERROR:
  73    case GLSL_TYPE_INTERFACE:
  74       unreachable("not reached");
  75    }
  76
  77    return 0;
  78 }
  79
  80 static unsigned
  81 type_size_scalar(const struct glsl_type *type)
  82 {
  83    unsigned int size, i;
  84
  85    switch (glsl_get_base_type(type)) {
  86    case GLSL_TYPE_UINT:
  87    case GLSL_TYPE_INT:
  88    case GLSL_TYPE_FLOAT:
  89    case GLSL_TYPE_BOOL:
  90       return glsl_get_components(type);
  91    case GLSL_TYPE_ARRAY:
  92       return type_size_scalar(glsl_get_array_element(type)) * glsl_get_length(type);
  93    case GLSL_TYPE_STRUCT:
  94       size = 0;
  95       for (i = 0; i < glsl_get_length(type); i++) {
  96          size += type_size_scalar(glsl_get_struct_field(type, i));
  97       }
  98       return size;
  99    case GLSL_TYPE_SUBROUTINE:
 100       return 1;
 101    case GLSL_TYPE_SAMPLER:
 102       return 0;
 103    case GLSL_TYPE_ATOMIC_UINT:
 104       return 0;
 105    case GLSL_TYPE_INTERFACE:
 106       return 0;
 107    case GLSL_TYPE_IMAGE:
 108       return 0;
 109    case GLSL_TYPE_VOID:
 110    case GLSL_TYPE_ERROR:
 111    case GLSL_TYPE_DOUBLE:
 112       unreachable("not reached");
 113    }
 114
 115    return 0;
 116 }
 117
 118 static unsigned
 119 type_size(const struct glsl_type *type, bool is_scalar)
 120 {
 121    if (is_scalar)
 122       return type_size_scalar(type);
 123    else
 124       return type_size_vec4(type);
 125 }
 126
 127 void
 128 nir_assign_var_locations(struct exec_list *var_list, unsigned *size, bool is_scalar)
 129 {
 130    unsigned location = 0;
 131
 132    foreach_list_typed(nir_variable, var, node, var_list) {
 133       /*
 134        * UBO's have their own address spaces, so don't count them towards the
 135        * number of global uniforms
 136        */
 137       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
 138           var->interface_type != NULL)
 139          continue;
 140
 141       var->data.driver_location = location;
 142       location += type_size(var->type, is_scalar);
 143    }
 144
 145    *size = location;
 146 }
 147
 148 static bool
 149 deref_has_indirect(nir_deref_var *deref)
 150 {
 151    for (nir_deref *tail = deref->deref.child; tail; tail = tail->child) {
 152       if (tail->deref_type == nir_deref_type_array) {
 153          nir_deref_array *arr = nir_deref_as_array(tail);
 154          if (arr->deref_array_type == nir_deref_array_type_indirect)
 155             return true;
 156       }
 157    }
 158
 159    return false;
 160 }
 161
 162 static bool
 163 mark_indirect_uses_block(nir_block *block, void *void_state)
 164 {
 165    struct set *indirect_set = void_state;
 166
 167    nir_foreach_instr(block, instr) {
 168       if (instr->type != nir_instr_type_intrinsic)
 169          continue;
 170
 171       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 172
 173       for (unsigned i = 0;
 174            i < nir_intrinsic_infos[intrin->intrinsic].num_variables; i++) {
 175          if (deref_has_indirect(intrin->variables[i]))
 176             _mesa_set_add(indirect_set, intrin->variables[i]->var);
 177       }
 178    }
 179
 180    return true;
 181 }
 182
 183 /* Identical to nir_assign_var_locations_packed except that it assigns
 184  * locations to the variables that are used 100% directly first and then
 185  * assigns locations to variables that are used indirectly.
 186  */
 187 void
 188 nir_assign_var_locations_direct_first(nir_shader *shader,
 189                                       struct exec_list *var_list,
 190                                       unsigned *direct_size,
 191                                       unsigned *size,
 192                                       bool is_scalar)
 193 {
 194    struct set *indirect_set = _mesa_set_create(NULL, _mesa_hash_pointer,
 195                                                _mesa_key_pointer_equal);
 196
 197    nir_foreach_overload(shader, overload) {
 198       if (overload->impl)
 199          nir_foreach_block(overload->impl, mark_indirect_uses_block,
 200                            indirect_set);
 201    }
 202
 203    unsigned location = 0;
 204
 205    foreach_list_typed(nir_variable, var, node, var_list) {
 206       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
 207           var->interface_type != NULL)
 208          continue;
 209
 210       if (_mesa_set_search(indirect_set, var))
 211          continue;
 212
 213       var->data.driver_location = location;
 214       location += type_size(var->type, is_scalar);
 215    }
 216
 217    *direct_size = location;
 218
 219    foreach_list_typed(nir_variable, var, node, var_list) {
 220       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
 221           var->interface_type != NULL)
 222          continue;
 223
 224       if (!_mesa_set_search(indirect_set, var))
 225          continue;
 226
 227       var->data.driver_location = location;
 228       location += type_size(var->type, is_scalar);
 229    }
 230
 231    *size = location;
 232
 233    _mesa_set_destroy(indirect_set, NULL);
 234 }
 235
 236 static unsigned
 237 get_io_offset(nir_deref_var *deref, nir_instr *instr, nir_src *indirect,
 238               struct lower_io_state *state)
 239 {
 240    bool found_indirect = false;
 241    unsigned base_offset = 0;
 242
 243    nir_deref *tail = &deref->deref;
 244    while (tail->child != NULL) {
 245       const struct glsl_type *parent_type = tail->type;
 246       tail = tail->child;
 247
 248       if (tail->deref_type == nir_deref_type_array) {
 249          nir_deref_array *deref_array = nir_deref_as_array(tail);
 250          unsigned size = type_size(tail->type, state->is_scalar);
 251
 252          base_offset += size * deref_array->base_offset;
 253
 254          if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 255             nir_load_const_instr *load_const =
 256                nir_load_const_instr_create(state->mem_ctx, 1);
 257             load_const->value.u[0] = size;
 258             nir_instr_insert_before(instr, &load_const->instr);
 259
 260             nir_alu_instr *mul = nir_alu_instr_create(state->mem_ctx,
 261                                                       nir_op_imul);
 262             mul->src[0].src.is_ssa = true;
 263             mul->src[0].src.ssa = &load_const->def;
 264             nir_src_copy(&mul->src[1].src, &deref_array->indirect,
 265                          state->mem_ctx);
 266             mul->dest.write_mask = 1;
 267             nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, NULL);
 268             nir_instr_insert_before(instr, &mul->instr);
 269
 270             if (found_indirect) {
 271                nir_alu_instr *add = nir_alu_instr_create(state->mem_ctx,
 272                                                          nir_op_iadd);
 273                add->src[0].src = *indirect;
 274                add->src[1].src.is_ssa = true;
 275                add->src[1].src.ssa = &mul->dest.dest.ssa;
 276                add->dest.write_mask = 1;
 277                nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, NULL);
 278                nir_instr_insert_before(instr, &add->instr);
 279
 280                indirect->is_ssa = true;
 281                indirect->ssa = &add->dest.dest.ssa;
 282             } else {
 283                indirect->is_ssa = true;
 284                indirect->ssa = &mul->dest.dest.ssa;
 285                found_indirect = true;
 286             }
 287          }
 288       } else if (tail->deref_type == nir_deref_type_struct) {
 289          nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
 290
 291          for (unsigned i = 0; i < deref_struct->index; i++)
 292             base_offset += type_size(glsl_get_struct_field(parent_type, i),
 293                                      state->is_scalar);
 294       }
 295    }
 296
 297    return base_offset;
 298 }
 299
 300 static bool
 301 nir_lower_io_block(nir_block *block, void *void_state)
 302 {
 303    struct lower_io_state *state = void_state;
 304
 305    nir_foreach_instr_safe(block, instr) {
 306       if (instr->type != nir_instr_type_intrinsic)
 307          continue;
 308
 309       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 310
 311       switch (intrin->intrinsic) {
 312       case nir_intrinsic_load_var: {
 313          nir_variable_mode mode = intrin->variables[0]->var->data.mode;
 314          if (mode != nir_var_shader_in && mode != nir_var_uniform)
 315             continue;
 316
 317          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 318
 319          /* Figure out the opcode */
 320          nir_intrinsic_op load_op;
 321          switch (mode) {
 322          case nir_var_shader_in:
 323             load_op = has_indirect ? nir_intrinsic_load_input_indirect :
 324                                      nir_intrinsic_load_input;
 325             break;
 326          case nir_var_uniform:
 327             load_op = has_indirect ? nir_intrinsic_load_uniform_indirect :
 328                                      nir_intrinsic_load_uniform;
 329             break;
 330          default:
 331             unreachable("Unknown variable mode");
 332          }
 333
 334          nir_intrinsic_instr *load = nir_intrinsic_instr_create(state->mem_ctx,
 335                                                                 load_op);
 336          load->num_components = intrin->num_components;
 337
 338          nir_src indirect;
 339          unsigned offset = get_io_offset(intrin->variables[0],
 340                                          &intrin->instr, &indirect, state);
 341          offset += intrin->variables[0]->var->data.driver_location;
 342
 343          load->const_index[0] = offset;
 344
 345          if (has_indirect)
 346             load->src[0] = indirect;
 347
 348          if (intrin->dest.is_ssa) {
 349             nir_ssa_dest_init(&load->instr, &load->dest,
 350                               intrin->num_components, NULL);
 351             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 352                                      nir_src_for_ssa(&load->dest.ssa),
 353                                      state->mem_ctx);
 354          } else {
 355             nir_dest_copy(&load->dest, &intrin->dest, state->mem_ctx);
 356          }
 357
 358          nir_instr_insert_before(&intrin->instr, &load->instr);
 359          nir_instr_remove(&intrin->instr);
 360          break;
 361       }
 362
 363       case nir_intrinsic_store_var: {
 364          if (intrin->variables[0]->var->data.mode != nir_var_shader_out)
 365             continue;
 366
 367          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 368
 369          nir_intrinsic_op store_op;
 370          if (has_indirect) {
 371             store_op = nir_intrinsic_store_output_indirect;
 372          } else {
 373             store_op = nir_intrinsic_store_output;
 374          }
 375
 376          nir_intrinsic_instr *store = nir_intrinsic_instr_create(state->mem_ctx,
 377                                                                  store_op);
 378          store->num_components = intrin->num_components;
 379
 380          nir_src indirect;
 381          unsigned offset = get_io_offset(intrin->variables[0],
 382                                          &intrin->instr, &indirect, state);
 383          offset += intrin->variables[0]->var->data.driver_location;
 384
 385          store->const_index[0] = offset;
 386
 387          nir_src_copy(&store->src[0], &intrin->src[0], state->mem_ctx);
 388
 389          if (has_indirect)
 390             store->src[1] = indirect;
 391
 392          nir_instr_insert_before(&intrin->instr, &store->instr);
 393          nir_instr_remove(&intrin->instr);
 394          break;
 395       }
 396
 397       default:
 398          break;
 399       }
 400    }
 401
 402    return true;
 403 }
 404
 405 static void
 406 nir_lower_io_impl(nir_function_impl *impl, bool is_scalar)
 407 {
 408    struct lower_io_state state;
 409
 410    state.mem_ctx = ralloc_parent(impl);
 411    state.is_scalar = is_scalar;
 412
 413    nir_foreach_block(impl, nir_lower_io_block, &state);
 414
 415    nir_metadata_preserve(impl, nir_metadata_block_index |
 416                                nir_metadata_dominance);
 417 }
 418
 419 void
 420 nir_lower_io(nir_shader *shader, bool is_scalar)
 421 {
 422    nir_foreach_overload(shader, overload) {
 423       if (overload->impl)
 424          nir_lower_io_impl(overload->impl, is_scalar);
 425    }
 426 }