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