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_SUBROUTINE:
  66       return 1;
  67    case GLSL_TYPE_SAMPLER:
  68       return 0;
  69    case GLSL_TYPE_ATOMIC_UINT:
  70       return 0;
  71    case GLSL_TYPE_IMAGE:
  72    case GLSL_TYPE_VOID:
  73    case GLSL_TYPE_DOUBLE:
  74    case GLSL_TYPE_ERROR:
  75    case GLSL_TYPE_INTERFACE:
  76       unreachable("not reached");
  77    }
  78
  79    return 0;
  80 }
  81
  82 static unsigned
  83 type_size_scalar(const struct glsl_type *type)
  84 {
  85    unsigned int size, i;
  86
  87    switch (glsl_get_base_type(type)) {
  88    case GLSL_TYPE_UINT:
  89    case GLSL_TYPE_INT:
  90    case GLSL_TYPE_FLOAT:
  91    case GLSL_TYPE_BOOL:
  92       return glsl_get_components(type);
  93    case GLSL_TYPE_ARRAY:
  94       return type_size_scalar(glsl_get_array_element(type)) * glsl_get_length(type);
  95    case GLSL_TYPE_STRUCT:
  96       size = 0;
  97       for (i = 0; i < glsl_get_length(type); i++) {
  98          size += type_size_scalar(glsl_get_struct_field(type, i));
  99       }
 100       return size;
 101    case GLSL_TYPE_SUBROUTINE:
 102       return 1;
 103    case GLSL_TYPE_SAMPLER:
 104       return 0;
 105    case GLSL_TYPE_ATOMIC_UINT:
 106       return 0;
 107    case GLSL_TYPE_INTERFACE:
 108       return 0;
 109    case GLSL_TYPE_IMAGE:
 110       return 0;
 111    case GLSL_TYPE_VOID:
 112    case GLSL_TYPE_ERROR:
 113    case GLSL_TYPE_DOUBLE:
 114       unreachable("not reached");
 115    }
 116
 117    return 0;
 118 }
 119
 120 static unsigned
 121 type_size(const struct glsl_type *type, bool is_scalar)
 122 {
 123    if (is_scalar)
 124       return type_size_scalar(type);
 125    else
 126       return type_size_vec4(type);
 127 }
 128
 129 void
 130 nir_assign_var_locations(struct exec_list *var_list, unsigned *size, bool is_scalar)
 131 {
 132    unsigned location = 0;
 133
 134    foreach_list_typed(nir_variable, var, node, var_list) {
 135       /*
 136        * UBO's have their own address spaces, so don't count them towards the
 137        * number of global uniforms
 138        */
 139       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
 140           var->interface_type != NULL)
 141          continue;
 142
 143       var->data.driver_location = location;
 144       location += type_size(var->type, is_scalar);
 145    }
 146
 147    *size = location;
 148 }
 149
 150 static bool
 151 deref_has_indirect(nir_deref_var *deref)
 152 {
 153    for (nir_deref *tail = deref->deref.child; tail; tail = tail->child) {
 154       if (tail->deref_type == nir_deref_type_array) {
 155          nir_deref_array *arr = nir_deref_as_array(tail);
 156          if (arr->deref_array_type == nir_deref_array_type_indirect)
 157             return true;
 158       }
 159    }
 160
 161    return false;
 162 }
 163
 164 static bool
 165 mark_indirect_uses_block(nir_block *block, void *void_state)
 166 {
 167    struct set *indirect_set = void_state;
 168
 169    nir_foreach_instr(block, instr) {
 170       if (instr->type != nir_instr_type_intrinsic)
 171          continue;
 172
 173       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 174
 175       for (unsigned i = 0;
 176            i < nir_intrinsic_infos[intrin->intrinsic].num_variables; i++) {
 177          if (deref_has_indirect(intrin->variables[i]))
 178             _mesa_set_add(indirect_set, intrin->variables[i]->var);
 179       }
 180    }
 181
 182    return true;
 183 }
 184
 185 /* Identical to nir_assign_var_locations_packed except that it assigns
 186  * locations to the variables that are used 100% directly first and then
 187  * assigns locations to variables that are used indirectly.
 188  */
 189 void
 190 nir_assign_var_locations_direct_first(nir_shader *shader,
 191                                       struct exec_list *var_list,
 192                                       unsigned *direct_size,
 193                                       unsigned *size,
 194                                       bool is_scalar)
 195 {
 196    struct set *indirect_set = _mesa_set_create(NULL, _mesa_hash_pointer,
 197                                                _mesa_key_pointer_equal);
 198
 199    nir_foreach_overload(shader, overload) {
 200       if (overload->impl)
 201          nir_foreach_block(overload->impl, mark_indirect_uses_block,
 202                            indirect_set);
 203    }
 204
 205    unsigned location = 0;
 206
 207    foreach_list_typed(nir_variable, var, node, var_list) {
 208       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
 209           var->interface_type != NULL)
 210          continue;
 211
 212       if (_mesa_set_search(indirect_set, var))
 213          continue;
 214
 215       var->data.driver_location = location;
 216       location += type_size(var->type, is_scalar);
 217    }
 218
 219    *direct_size = location;
 220
 221    foreach_list_typed(nir_variable, var, node, var_list) {
 222       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
 223           var->interface_type != NULL)
 224          continue;
 225
 226       if (!_mesa_set_search(indirect_set, var))
 227          continue;
 228
 229       var->data.driver_location = location;
 230       location += type_size(var->type, is_scalar);
 231    }
 232
 233    *size = location;
 234
 235    _mesa_set_destroy(indirect_set, NULL);
 236 }
 237
 238 static unsigned
 239 get_io_offset(nir_deref_var *deref, nir_instr *instr, nir_src *indirect,
 240               struct lower_io_state *state)
 241 {
 242    bool found_indirect = false;
 243    unsigned base_offset = 0;
 244
 245    nir_deref *tail = &deref->deref;
 246    while (tail->child != NULL) {
 247       const struct glsl_type *parent_type = tail->type;
 248       tail = tail->child;
 249
 250       if (tail->deref_type == nir_deref_type_array) {
 251          nir_deref_array *deref_array = nir_deref_as_array(tail);
 252          unsigned size = type_size(tail->type, state->is_scalar);
 253
 254          base_offset += size * deref_array->base_offset;
 255
 256          if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 257             nir_load_const_instr *load_const =
 258                nir_load_const_instr_create(state->mem_ctx, 1);
 259             load_const->value.u[0] = size;
 260             nir_instr_insert_before(instr, &load_const->instr);
 261
 262             nir_alu_instr *mul = nir_alu_instr_create(state->mem_ctx,
 263                                                       nir_op_imul);
 264             mul->src[0].src.is_ssa = true;
 265             mul->src[0].src.ssa = &load_const->def;
 266             nir_src_copy(&mul->src[1].src, &deref_array->indirect,
 267                          state->mem_ctx);
 268             mul->dest.write_mask = 1;
 269             nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, NULL);
 270             nir_instr_insert_before(instr, &mul->instr);
 271
 272             if (found_indirect) {
 273                nir_alu_instr *add = nir_alu_instr_create(state->mem_ctx,
 274                                                          nir_op_iadd);
 275                add->src[0].src = *indirect;
 276                add->src[1].src.is_ssa = true;
 277                add->src[1].src.ssa = &mul->dest.dest.ssa;
 278                add->dest.write_mask = 1;
 279                nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, NULL);
 280                nir_instr_insert_before(instr, &add->instr);
 281
 282                indirect->is_ssa = true;
 283                indirect->ssa = &add->dest.dest.ssa;
 284             } else {
 285                indirect->is_ssa = true;
 286                indirect->ssa = &mul->dest.dest.ssa;
 287                found_indirect = true;
 288             }
 289          }
 290       } else if (tail->deref_type == nir_deref_type_struct) {
 291          nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
 292
 293          for (unsigned i = 0; i < deref_struct->index; i++)
 294             base_offset += type_size(glsl_get_struct_field(parent_type, i),
 295                                      state->is_scalar);
 296       }
 297    }
 298
 299    return base_offset;
 300 }
 301
 302 static bool
 303 nir_lower_io_block(nir_block *block, void *void_state)
 304 {
 305    struct lower_io_state *state = void_state;
 306
 307    nir_foreach_instr_safe(block, instr) {
 308       if (instr->type != nir_instr_type_intrinsic)
 309          continue;
 310
 311       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 312
 313       switch (intrin->intrinsic) {
 314       case nir_intrinsic_load_var: {
 315          nir_variable_mode mode = intrin->variables[0]->var->data.mode;
 316          if (mode != nir_var_shader_in && mode != nir_var_uniform)
 317             continue;
 318
 319          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 320
 321          /* Figure out the opcode */
 322          nir_intrinsic_op load_op;
 323          switch (mode) {
 324          case nir_var_shader_in:
 325             load_op = has_indirect ? nir_intrinsic_load_input_indirect :
 326                                      nir_intrinsic_load_input;
 327             break;
 328          case nir_var_uniform:
 329             load_op = has_indirect ? nir_intrinsic_load_uniform_indirect :
 330                                      nir_intrinsic_load_uniform;
 331             break;
 332          default:
 333             unreachable("Unknown variable mode");
 334          }
 335
 336          nir_intrinsic_instr *load = nir_intrinsic_instr_create(state->mem_ctx,
 337                                                                 load_op);
 338          load->num_components = intrin->num_components;
 339
 340          nir_src indirect;
 341          unsigned offset = get_io_offset(intrin->variables[0],
 342                                          &intrin->instr, &indirect, state);
 343          offset += intrin->variables[0]->var->data.driver_location;
 344
 345          load->const_index[0] = offset;
 346
 347          if (has_indirect)
 348             load->src[0] = indirect;
 349
 350          if (intrin->dest.is_ssa) {
 351             nir_ssa_dest_init(&load->instr, &load->dest,
 352                               intrin->num_components, NULL);
 353             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 354                                      nir_src_for_ssa(&load->dest.ssa),
 355                                      state->mem_ctx);
 356          } else {
 357             nir_dest_copy(&load->dest, &intrin->dest, state->mem_ctx);
 358          }
 359
 360          nir_instr_insert_before(&intrin->instr, &load->instr);
 361          nir_instr_remove(&intrin->instr);
 362          break;
 363       }
 364
 365       case nir_intrinsic_store_var: {
 366          if (intrin->variables[0]->var->data.mode != nir_var_shader_out)
 367             continue;
 368
 369          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 370
 371          nir_intrinsic_op store_op;
 372          if (has_indirect) {
 373             store_op = nir_intrinsic_store_output_indirect;
 374          } else {
 375             store_op = nir_intrinsic_store_output;
 376          }
 377
 378          nir_intrinsic_instr *store = nir_intrinsic_instr_create(state->mem_ctx,
 379                                                                  store_op);
 380          store->num_components = intrin->num_components;
 381
 382          nir_src indirect;
 383          unsigned offset = get_io_offset(intrin->variables[0],
 384                                          &intrin->instr, &indirect, state);
 385          offset += intrin->variables[0]->var->data.driver_location;
 386
 387          store->const_index[0] = offset;
 388
 389          nir_src_copy(&store->src[0], &intrin->src[0], state->mem_ctx);
 390
 391          if (has_indirect)
 392             store->src[1] = indirect;
 393
 394          nir_instr_insert_before(&intrin->instr, &store->instr);
 395          nir_instr_remove(&intrin->instr);
 396          break;
 397       }
 398
 399       default:
 400          break;
 401       }
 402    }
 403
 404    return true;
 405 }
 406
 407 static void
 408 nir_lower_io_impl(nir_function_impl *impl, bool is_scalar)
 409 {
 410    struct lower_io_state state;
 411
 412    state.mem_ctx = ralloc_parent(impl);
 413    state.is_scalar = is_scalar;
 414
 415    nir_foreach_block(impl, nir_lower_io_block, &state);
 416
 417    nir_metadata_preserve(impl, nir_metadata_block_index |
 418                                nir_metadata_dominance);
 419 }
 420
 421 void
 422 nir_lower_io(nir_shader *shader, bool is_scalar)
 423 {
 424    nir_foreach_overload(shader, overload) {
 425       if (overload->impl)
 426          nir_lower_io_impl(overload->impl, is_scalar);
 427    }
 428 }