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 void
  80 nir_assign_var_locations_scalar(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 bool
 100 deref_has_indirect(nir_deref_var *deref)
 101 {
 102    for (nir_deref *tail = deref->deref.child; tail; tail = tail->child) {
 103       if (tail->deref_type == nir_deref_type_array) {
 104          nir_deref_array *arr = nir_deref_as_array(tail);
 105          if (arr->deref_array_type == nir_deref_array_type_indirect)
 106             return true;
 107       }
 108    }
 109
 110    return false;
 111 }
 112
 113 static unsigned
 114 get_io_offset(nir_deref_var *deref, nir_instr *instr, nir_src *indirect,
 115               struct lower_io_state *state)
 116 {
 117    bool found_indirect = false;
 118    unsigned base_offset = 0;
 119
 120    nir_deref *tail = &deref->deref;
 121    while (tail->child != NULL) {
 122       const struct glsl_type *parent_type = tail->type;
 123       tail = tail->child;
 124
 125       if (tail->deref_type == nir_deref_type_array) {
 126          nir_deref_array *deref_array = nir_deref_as_array(tail);
 127          unsigned size = type_size(tail->type);
 128
 129          base_offset += size * deref_array->base_offset;
 130
 131          if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 132             nir_load_const_instr *load_const =
 133                nir_load_const_instr_create(state->mem_ctx, 1);
 134             load_const->value.u[0] = size;
 135             nir_instr_insert_before(instr, &load_const->instr);
 136
 137             nir_alu_instr *mul = nir_alu_instr_create(state->mem_ctx,
 138                                                       nir_op_imul);
 139             mul->src[0].src.is_ssa = true;
 140             mul->src[0].src.ssa = &load_const->def;
 141             nir_src_copy(&mul->src[1].src, &deref_array->indirect,
 142                          state->mem_ctx);
 143             mul->dest.write_mask = 1;
 144             nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, NULL);
 145             nir_instr_insert_before(instr, &mul->instr);
 146
 147             if (found_indirect) {
 148                nir_alu_instr *add = nir_alu_instr_create(state->mem_ctx,
 149                                                          nir_op_iadd);
 150                add->src[0].src = *indirect;
 151                add->src[1].src.is_ssa = true;
 152                add->src[1].src.ssa = &mul->dest.dest.ssa;
 153                add->dest.write_mask = 1;
 154                nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, NULL);
 155                nir_instr_insert_before(instr, &add->instr);
 156
 157                indirect->is_ssa = true;
 158                indirect->ssa = &add->dest.dest.ssa;
 159             } else {
 160                indirect->is_ssa = true;
 161                indirect->ssa = &mul->dest.dest.ssa;
 162                found_indirect = true;
 163             }
 164          }
 165       } else if (tail->deref_type == nir_deref_type_struct) {
 166          nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
 167
 168          for (unsigned i = 0; i < deref_struct->index; i++)
 169             base_offset += type_size(glsl_get_struct_field(parent_type, i));
 170       }
 171    }
 172
 173    return base_offset;
 174 }
 175
 176 static bool
 177 nir_lower_io_block(nir_block *block, void *void_state)
 178 {
 179    struct lower_io_state *state = void_state;
 180
 181    nir_foreach_instr_safe(block, instr) {
 182       if (instr->type != nir_instr_type_intrinsic)
 183          continue;
 184
 185       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 186
 187       switch (intrin->intrinsic) {
 188       case nir_intrinsic_load_var: {
 189          nir_variable_mode mode = intrin->variables[0]->var->data.mode;
 190          if (mode != nir_var_shader_in && mode != nir_var_uniform)
 191             continue;
 192
 193          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 194
 195          /* Figure out the opcode */
 196          nir_intrinsic_op load_op;
 197          switch (mode) {
 198          case nir_var_shader_in:
 199             load_op = has_indirect ? nir_intrinsic_load_input_indirect :
 200                                      nir_intrinsic_load_input;
 201             break;
 202          case nir_var_uniform:
 203             load_op = has_indirect ? nir_intrinsic_load_uniform_indirect :
 204                                      nir_intrinsic_load_uniform;
 205             break;
 206          default:
 207             unreachable("Unknown variable mode");
 208          }
 209
 210          nir_intrinsic_instr *load = nir_intrinsic_instr_create(state->mem_ctx,
 211                                                                 load_op);
 212          load->num_components = intrin->num_components;
 213
 214          nir_src indirect;
 215          unsigned offset = get_io_offset(intrin->variables[0],
 216                                          &intrin->instr, &indirect, state);
 217          offset += intrin->variables[0]->var->data.driver_location;
 218
 219          load->const_index[0] = offset;
 220          load->const_index[1] = 1;
 221
 222          if (has_indirect)
 223             load->src[0] = indirect;
 224
 225          if (intrin->dest.is_ssa) {
 226             nir_ssa_dest_init(&load->instr, &load->dest,
 227                               intrin->num_components, NULL);
 228             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 229                                      nir_src_for_ssa(&load->dest.ssa),
 230                                      state->mem_ctx);
 231          } else {
 232             nir_dest_copy(&load->dest, &intrin->dest, state->mem_ctx);
 233          }
 234
 235          nir_instr_insert_before(&intrin->instr, &load->instr);
 236          nir_instr_remove(&intrin->instr);
 237          break;
 238       }
 239
 240       case nir_intrinsic_store_var: {
 241          if (intrin->variables[0]->var->data.mode != nir_var_shader_out)
 242             continue;
 243
 244          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 245
 246          nir_intrinsic_op store_op;
 247          if (has_indirect) {
 248             store_op = nir_intrinsic_store_output_indirect;
 249          } else {
 250             store_op = nir_intrinsic_store_output;
 251          }
 252
 253          nir_intrinsic_instr *store = nir_intrinsic_instr_create(state->mem_ctx,
 254                                                                  store_op);
 255          store->num_components = intrin->num_components;
 256
 257          nir_src indirect;
 258          unsigned offset = get_io_offset(intrin->variables[0],
 259                                          &intrin->instr, &indirect, state);
 260          offset += intrin->variables[0]->var->data.driver_location;
 261
 262          store->const_index[0] = offset;
 263          store->const_index[1] = 1;
 264
 265          nir_src_copy(&store->src[0], &intrin->src[0], state->mem_ctx);
 266
 267          if (has_indirect)
 268             store->src[1] = indirect;
 269
 270          nir_instr_insert_before(&intrin->instr, &store->instr);
 271          nir_instr_remove(&intrin->instr);
 272          break;
 273       }
 274
 275       default:
 276          break;
 277       }
 278    }
 279
 280    return true;
 281 }
 282
 283 static void
 284 nir_lower_io_impl(nir_function_impl *impl)
 285 {
 286    struct lower_io_state state;
 287
 288    state.mem_ctx = ralloc_parent(impl);
 289
 290    nir_foreach_block(impl, nir_lower_io_block, &state);
 291
 292    nir_metadata_preserve(impl, nir_metadata_block_index |
 293                                nir_metadata_dominance);
 294 }
 295
 296 void
 297 nir_lower_io(nir_shader *shader)
 298 {
 299    nir_foreach_overload(shader, overload) {
 300       if (overload->impl)
 301          nir_lower_io_impl(overload->impl);
 302    }
 303 }