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