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 unsigned
  80 get_io_offset(nir_deref_var *deref, nir_instr *instr, nir_src *indirect,
  81               struct lower_io_state *state)
  82 {
  83    bool found_indirect = false;
  84    unsigned base_offset = 0;
  85
  86    nir_builder *b = &state->builder;
  87    b->cursor = nir_before_instr(instr);
  88
  89    nir_deref *tail = &deref->deref;
  90    while (tail->child != NULL) {
  91       const struct glsl_type *parent_type = tail->type;
  92       tail = tail->child;
  93
  94       if (tail->deref_type == nir_deref_type_array) {
  95          nir_deref_array *deref_array = nir_deref_as_array(tail);
  96          unsigned size = state->type_size(tail->type);
  97
  98          base_offset += size * deref_array->base_offset;
  99
 100          if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 101             nir_ssa_def *mul =
 102                nir_imul(b, nir_imm_int(b, size),
 103                         nir_ssa_for_src(b, deref_array->indirect, 1));
 104
 105             if (found_indirect) {
 106                indirect->ssa =
 107                   nir_iadd(b, nir_ssa_for_src(b, *indirect, 1), mul);
 108             } else {
 109                indirect->ssa = mul;
 110             }
 111             indirect->is_ssa = true;
 112             found_indirect = true;
 113          }
 114       } else if (tail->deref_type == nir_deref_type_struct) {
 115          nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
 116
 117          for (unsigned i = 0; i < deref_struct->index; i++) {
 118             base_offset +=
 119                state->type_size(glsl_get_struct_field(parent_type, i));
 120          }
 121       }
 122    }
 123
 124    return base_offset;
 125 }
 126
 127 static nir_intrinsic_op
 128 load_op(nir_variable_mode mode, bool has_indirect)
 129 {
 130    nir_intrinsic_op op;
 131    switch (mode) {
 132    case nir_var_shader_in:
 133       op = has_indirect ? nir_intrinsic_load_input_indirect :
 134                           nir_intrinsic_load_input;
 135       break;
 136    case nir_var_uniform:
 137       op = has_indirect ? nir_intrinsic_load_uniform_indirect :
 138                           nir_intrinsic_load_uniform;
 139       break;
 140    default:
 141       unreachable("Unknown variable mode");
 142    }
 143    return op;
 144 }
 145
 146 static bool
 147 nir_lower_io_block(nir_block *block, void *void_state)
 148 {
 149    struct lower_io_state *state = void_state;
 150
 151    nir_foreach_instr_safe(block, instr) {
 152       if (instr->type != nir_instr_type_intrinsic)
 153          continue;
 154
 155       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 156
 157       switch (intrin->intrinsic) {
 158       case nir_intrinsic_load_var: {
 159          nir_variable_mode mode = intrin->variables[0]->var->data.mode;
 160          if (mode != nir_var_shader_in && mode != nir_var_uniform)
 161             continue;
 162
 163          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 164
 165          nir_intrinsic_instr *load =
 166             nir_intrinsic_instr_create(state->mem_ctx,
 167                                        load_op(mode, has_indirect));
 168          load->num_components = intrin->num_components;
 169
 170          nir_src indirect;
 171          unsigned offset = get_io_offset(intrin->variables[0],
 172                                          &intrin->instr, &indirect, state);
 173
 174          unsigned location = intrin->variables[0]->var->data.driver_location;
 175          if (mode == nir_var_uniform) {
 176             load->const_index[0] = location;
 177             load->const_index[1] = offset;
 178          } else {
 179             load->const_index[0] = location + offset;
 180          }
 181
 182          if (has_indirect)
 183             load->src[0] = indirect;
 184
 185          if (intrin->dest.is_ssa) {
 186             nir_ssa_dest_init(&load->instr, &load->dest,
 187                               intrin->num_components, NULL);
 188             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 189                                      nir_src_for_ssa(&load->dest.ssa));
 190          } else {
 191             nir_dest_copy(&load->dest, &intrin->dest, state->mem_ctx);
 192          }
 193
 194          nir_instr_insert_before(&intrin->instr, &load->instr);
 195          nir_instr_remove(&intrin->instr);
 196          break;
 197       }
 198
 199       case nir_intrinsic_store_var: {
 200          if (intrin->variables[0]->var->data.mode != nir_var_shader_out)
 201             continue;
 202
 203          bool has_indirect = deref_has_indirect(intrin->variables[0]);
 204
 205          nir_intrinsic_op store_op;
 206          if (has_indirect) {
 207             store_op = nir_intrinsic_store_output_indirect;
 208          } else {
 209             store_op = nir_intrinsic_store_output;
 210          }
 211
 212          nir_intrinsic_instr *store = nir_intrinsic_instr_create(state->mem_ctx,
 213                                                                  store_op);
 214          store->num_components = intrin->num_components;
 215
 216          nir_src indirect;
 217          unsigned offset = get_io_offset(intrin->variables[0],
 218                                          &intrin->instr, &indirect, state);
 219          offset += intrin->variables[0]->var->data.driver_location;
 220
 221          store->const_index[0] = offset;
 222
 223          nir_src_copy(&store->src[0], &intrin->src[0], store);
 224
 225          if (has_indirect)
 226             store->src[1] = indirect;
 227
 228          nir_instr_insert_before(&intrin->instr, &store->instr);
 229          nir_instr_remove(&intrin->instr);
 230          break;
 231       }
 232
 233       default:
 234          break;
 235       }
 236    }
 237
 238    return true;
 239 }
 240
 241 static void
 242 nir_lower_io_impl(nir_function_impl *impl, int(*type_size)(const struct glsl_type *))
 243 {
 244    struct lower_io_state state;
 245
 246    nir_builder_init(&state.builder, impl);
 247    state.mem_ctx = ralloc_parent(impl);
 248    state.type_size = type_size;
 249
 250    nir_foreach_block(impl, nir_lower_io_block, &state);
 251
 252    nir_metadata_preserve(impl, nir_metadata_block_index |
 253                                nir_metadata_dominance);
 254 }
 255
 256 void
 257 nir_lower_io(nir_shader *shader, int(*type_size)(const struct glsl_type *))
 258 {
 259    nir_foreach_overload(shader, overload) {
 260       if (overload->impl)
 261          nir_lower_io_impl(overload->impl, type_size);
 262    }
 263 }