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