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_VOID:
  71    case GLSL_TYPE_ERROR:
  72       unreachable("not reached");
  73    }
  74
  75    return 0;
  76 }
  77
  78 static void
  79 assign_var_locations(struct hash_table *ht, unsigned *size)
  80 {
  81    unsigned location = 0;
  82
  83    struct hash_entry *entry;
  84    hash_table_foreach(ht, entry) {
  85       nir_variable *var = (nir_variable *) entry->data;
  86
  87       /*
  88        * UBO's have their own address spaces, so don't count them towards the
  89        * number of global uniforms
  90        */
  91       if (var->data.mode == nir_var_uniform && var->interface_type != NULL)
  92          continue;
  93
  94       var->data.driver_location = location;
  95       location += type_size(var->type);
  96    }
  97
  98    *size = location;
  99 }
 100
 101 static void
 102 assign_var_locations_shader(nir_shader *shader)
 103 {
 104    assign_var_locations(shader->inputs, &shader->num_inputs);
 105    assign_var_locations(shader->outputs, &shader->num_outputs);
 106    assign_var_locations(shader->uniforms, &shader->num_uniforms);
 107 }
 108
 109 static bool
 110 deref_has_indirect(nir_deref_var *deref)
 111 {
 112    for (nir_deref *tail = deref->deref.child; tail; tail = tail->child) {
 113       if (tail->deref_type == nir_deref_type_array) {
 114          nir_deref_array *arr = nir_deref_as_array(tail);
 115          if (arr->deref_array_type == nir_deref_array_type_indirect)
 116             return true;
 117       }
 118    }
 119
 120    return false;
 121 }
 122
 123 static unsigned
 124 get_io_offset(nir_deref_var *deref, nir_instr *instr, nir_src *indirect,
 125               struct lower_io_state *state)
 126 {
 127    bool found_indirect = false;
 128    unsigned base_offset = 0;
 129
 130    nir_deref *tail = &deref->deref;
 131    while (tail->child != NULL) {
 132       const struct glsl_type *parent_type = tail->type;
 133       tail = tail->child;
 134
 135       if (tail->deref_type == nir_deref_type_array) {
 136          nir_deref_array *deref_array = nir_deref_as_array(tail);
 137          unsigned size = type_size(tail->type);
 138
 139          base_offset += size * deref_array->base_offset;
 140
 141          if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 142             nir_load_const_instr *load_const =
 143                nir_load_const_instr_create(state->mem_ctx, 1);
 144             load_const->value.u[0] = size;
 145             nir_instr_insert_before(instr, &load_const->instr);
 146
 147             nir_alu_instr *mul = nir_alu_instr_create(state->mem_ctx,
 148                                                       nir_op_imul);
 149             mul->src[0].src.is_ssa = true;
 150             mul->src[0].src.ssa = &load_const->def;
 151             mul->src[1].src = nir_src_copy(deref_array->indirect,
 152                                            state->mem_ctx);
 153             mul->dest.write_mask = 1;
 154             mul->dest.dest.is_ssa = true;
 155             nir_ssa_def_init(&mul->instr, &mul->dest.dest.ssa, 1, NULL);
 156             nir_instr_insert_before(instr, &mul->instr);
 157
 158             if (found_indirect) {
 159                nir_alu_instr *add = nir_alu_instr_create(state->mem_ctx,
 160                                                          nir_op_iadd);
 161                add->src[0].src = *indirect;
 162                add->src[1].src.is_ssa = true;
 163                add->src[1].src.ssa = &mul->dest.dest.ssa;
 164                add->dest.write_mask = 1;
 165                add->dest.dest.is_ssa = true;
 166                nir_ssa_def_init(&add->instr, &add->dest.dest.ssa, 1, NULL);
 167                nir_instr_insert_before(instr, &add->instr);
 168
 169                indirect->is_ssa = true;
 170                indirect->ssa = &add->dest.dest.ssa;
 171             } else {
 172                indirect->is_ssa = true;
 173                indirect->ssa = &mul->dest.dest.ssa;
 174                found_indirect = true;
 175             }
 176          }
 177       } else if (tail->deref_type == nir_deref_type_struct) {
 178          nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
 179
 180          for (unsigned i = 0; i < deref_struct->index; i++)
 181             base_offset += type_size(glsl_get_struct_field(parent_type, i));
 182       }
 183    }
 184
 185    return base_offset;
 186 }
 187
 188 static bool
 189 nir_lower_io_block(nir_block *block, void *void_state)
 190 {
 191    struct lower_io_state *state = void_state;
 192
 193    nir_foreach_instr_safe(block, instr) {
 194       if (instr->type != nir_instr_type_intrinsic)
 195          continue;
 196
 197       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 198
 199       switch (intrin->intrinsic) {
 200       case nir_intrinsic_load_var: {
 201          nir_variable_mode mode = intrin->variables[0]->var->data.mode;
 202          if (mode != nir_var_shader_in && mode != nir_var_uniform)
 203             continue;
 204
 205          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 206
 207          /* Figure out the opcode */
 208          nir_intrinsic_op load_op;
 209          switch (mode) {
 210          case nir_var_shader_in:
 211             load_op = has_indirect ? nir_intrinsic_load_input_indirect :
 212                                      nir_intrinsic_load_input;
 213             break;
 214          case nir_var_uniform:
 215             load_op = has_indirect ? nir_intrinsic_load_uniform_indirect :
 216                                      nir_intrinsic_load_uniform;
 217             break;
 218          default:
 219             unreachable("Unknown variable mode");
 220          }
 221
 222          nir_intrinsic_instr *load = nir_intrinsic_instr_create(state->mem_ctx,
 223                                                                 load_op);
 224          load->num_components = intrin->num_components;
 225
 226          nir_src indirect;
 227          unsigned offset = get_io_offset(intrin->variables[0],
 228                                          &intrin->instr, &indirect, state);
 229          offset += intrin->variables[0]->var->data.driver_location;
 230
 231          load->const_index[0] = offset;
 232          load->const_index[1] = 1;
 233
 234          if (has_indirect)
 235             load->src[0] = indirect;
 236
 237          if (intrin->dest.is_ssa) {
 238             load->dest.is_ssa = true;
 239             nir_ssa_def_init(&load->instr, &load->dest.ssa,
 240                              intrin->num_components, NULL);
 241
 242             nir_src new_src = {
 243                .is_ssa = true,
 244                .ssa = &load->dest.ssa,
 245             };
 246
 247             nir_ssa_def_rewrite_uses(&intrin->dest.ssa, new_src,
 248                                      state->mem_ctx);
 249          } else {
 250             load->dest = nir_dest_copy(intrin->dest, state->mem_ctx);
 251          }
 252
 253          nir_instr_insert_before(&intrin->instr, &load->instr);
 254          nir_instr_remove(&intrin->instr);
 255          break;
 256       }
 257
 258       case nir_intrinsic_store_var: {
 259          if (intrin->variables[0]->var->data.mode != nir_var_shader_out)
 260             continue;
 261
 262          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 263
 264          nir_intrinsic_op store_op;
 265          if (has_indirect) {
 266             store_op = nir_intrinsic_store_output_indirect;
 267          } else {
 268             store_op = nir_intrinsic_store_output;
 269          }
 270
 271          nir_intrinsic_instr *store = nir_intrinsic_instr_create(state->mem_ctx,
 272                                                                  store_op);
 273          store->num_components = intrin->num_components;
 274
 275          nir_src indirect;
 276          unsigned offset = get_io_offset(intrin->variables[0],
 277                                          &intrin->instr, &indirect, state);
 278          offset += intrin->variables[0]->var->data.driver_location;
 279
 280          store->const_index[0] = offset;
 281          store->const_index[1] = 1;
 282
 283          store->src[0] = nir_src_copy(intrin->src[0], state->mem_ctx);
 284
 285          if (has_indirect)
 286             store->src[1] = indirect;
 287
 288          nir_instr_insert_before(&intrin->instr, &store->instr);
 289          nir_instr_remove(&intrin->instr);
 290          break;
 291       }
 292
 293       default:
 294          break;
 295       }
 296    }
 297
 298    return true;
 299 }
 300
 301 static void
 302 nir_lower_io_impl(nir_function_impl *impl)
 303 {
 304    struct lower_io_state state;
 305
 306    state.mem_ctx = ralloc_parent(impl);
 307
 308    nir_foreach_block(impl, nir_lower_io_block, &state);
 309
 310    nir_metadata_preserve(impl, nir_metadata_block_index |
 311                                nir_metadata_dominance);
 312 }
 313
 314 void
 315 nir_lower_io(nir_shader *shader)
 316 {
 317    assign_var_locations_shader(shader);
 318
 319    nir_foreach_overload(shader, overload) {
 320       if (overload->impl)
 321          nir_lower_io_impl(overload->impl);
 322    }
 323 }