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