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    nir_variable_mode mode;
  42 };
  43
  44 void
  45 nir_assign_var_locations(struct exec_list *var_list, unsigned *size,
  46                          int (*type_size)(const struct glsl_type *))
  47 {
  48    unsigned location = 0;
  49
  50    nir_foreach_variable(var, var_list) {
  51       /*
  52        * UBO's have their own address spaces, so don't count them towards the
  53        * number of global uniforms
  54        */
  55       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
  56           var->interface_type != NULL)
  57          continue;
  58
  59       var->data.driver_location = location;
  60       location += type_size(var->type);
  61    }
  62
  63    *size = location;
  64 }
  65
  66 static unsigned
  67 get_io_offset(nir_deref_var *deref, nir_instr *instr,
  68               nir_ssa_def **out_indirect,
  69               struct lower_io_state *state)
  70 {
  71    nir_ssa_def *indirect = NULL;
  72    unsigned base_offset = 0;
  73
  74    nir_builder *b = &state->builder;
  75    b->cursor = nir_before_instr(instr);
  76
  77    nir_deref *tail = &deref->deref;
  78    while (tail->child != NULL) {
  79       const struct glsl_type *parent_type = tail->type;
  80       tail = tail->child;
  81
  82       if (tail->deref_type == nir_deref_type_array) {
  83          nir_deref_array *deref_array = nir_deref_as_array(tail);
  84          unsigned size = state->type_size(tail->type);
  85
  86          base_offset += size * deref_array->base_offset;
  87
  88          if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
  89             nir_ssa_def *mul =
  90                nir_imul(b, nir_imm_int(b, size),
  91                         nir_ssa_for_src(b, deref_array->indirect, 1));
  92
  93             indirect = indirect ? nir_iadd(b, indirect, mul) : mul;
  94          }
  95       } else if (tail->deref_type == nir_deref_type_struct) {
  96          nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
  97
  98          for (unsigned i = 0; i < deref_struct->index; i++) {
  99             base_offset +=
 100                state->type_size(glsl_get_struct_field(parent_type, i));
 101          }
 102       }
 103    }
 104
 105    *out_indirect = indirect;
 106    return base_offset;
 107 }
 108
 109 static nir_intrinsic_op
 110 load_op(nir_variable_mode mode, bool has_indirect)
 111 {
 112    nir_intrinsic_op op;
 113    switch (mode) {
 114    case nir_var_shader_in:
 115       op = has_indirect ? nir_intrinsic_load_input_indirect :
 116                           nir_intrinsic_load_input;
 117       break;
 118    case nir_var_uniform:
 119       op = has_indirect ? nir_intrinsic_load_uniform_indirect :
 120                           nir_intrinsic_load_uniform;
 121       break;
 122    default:
 123       unreachable("Unknown variable mode");
 124    }
 125    return op;
 126 }
 127
 128 static bool
 129 nir_lower_io_block(nir_block *block, void *void_state)
 130 {
 131    struct lower_io_state *state = void_state;
 132
 133    nir_foreach_instr_safe(block, instr) {
 134       if (instr->type != nir_instr_type_intrinsic)
 135          continue;
 136
 137       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 138
 139       if (intrin->intrinsic != nir_intrinsic_load_var &&
 140           intrin->intrinsic != nir_intrinsic_store_var)
 141          continue;
 142
 143       nir_variable_mode mode = intrin->variables[0]->var->data.mode;
 144
 145       if (state->mode != -1 && state->mode != mode)
 146          continue;
 147
 148       switch (intrin->intrinsic) {
 149       case nir_intrinsic_load_var: {
 150          if (mode != nir_var_shader_in && mode != nir_var_uniform)
 151             continue;
 152
 153          nir_ssa_def *indirect;
 154
 155          unsigned offset = get_io_offset(intrin->variables[0], &intrin->instr,
 156                                          &indirect, state);
 157
 158          nir_intrinsic_instr *load =
 159             nir_intrinsic_instr_create(state->mem_ctx,
 160                                        load_op(mode, indirect));
 161          load->num_components = intrin->num_components;
 162
 163          unsigned location = intrin->variables[0]->var->data.driver_location;
 164          if (mode == nir_var_uniform) {
 165             load->const_index[0] = location;
 166             load->const_index[1] = offset;
 167          } else {
 168             load->const_index[0] = location + offset;
 169          }
 170
 171          if (indirect)
 172             load->src[0] = nir_src_for_ssa(indirect);
 173
 174          if (intrin->dest.is_ssa) {
 175             nir_ssa_dest_init(&load->instr, &load->dest,
 176                               intrin->num_components, NULL);
 177             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 178                                      nir_src_for_ssa(&load->dest.ssa));
 179          } else {
 180             nir_dest_copy(&load->dest, &intrin->dest, state->mem_ctx);
 181          }
 182
 183          nir_instr_insert_before(&intrin->instr, &load->instr);
 184          nir_instr_remove(&intrin->instr);
 185          break;
 186       }
 187
 188       case nir_intrinsic_store_var: {
 189          if (intrin->variables[0]->var->data.mode != nir_var_shader_out)
 190             continue;
 191
 192          nir_ssa_def *indirect;
 193
 194          unsigned offset = get_io_offset(intrin->variables[0], &intrin->instr,
 195                                          &indirect, state);
 196          offset += intrin->variables[0]->var->data.driver_location;
 197
 198          nir_intrinsic_op store_op;
 199          if (indirect) {
 200             store_op = nir_intrinsic_store_output_indirect;
 201          } else {
 202             store_op = nir_intrinsic_store_output;
 203          }
 204
 205          nir_intrinsic_instr *store = nir_intrinsic_instr_create(state->mem_ctx,
 206                                                                  store_op);
 207          store->num_components = intrin->num_components;
 208          store->const_index[0] = offset;
 209
 210          nir_src_copy(&store->src[0], &intrin->src[0], store);
 211
 212          if (indirect)
 213             store->src[1] = nir_src_for_ssa(indirect);
 214
 215          nir_instr_insert_before(&intrin->instr, &store->instr);
 216          nir_instr_remove(&intrin->instr);
 217          break;
 218       }
 219
 220       default:
 221          break;
 222       }
 223    }
 224
 225    return true;
 226 }
 227
 228 static void
 229 nir_lower_io_impl(nir_function_impl *impl,
 230                   nir_variable_mode mode,
 231                   int (*type_size)(const struct glsl_type *))
 232 {
 233    struct lower_io_state state;
 234
 235    nir_builder_init(&state.builder, impl);
 236    state.mem_ctx = ralloc_parent(impl);
 237    state.mode = mode;
 238    state.type_size = type_size;
 239
 240    nir_foreach_block(impl, nir_lower_io_block, &state);
 241
 242    nir_metadata_preserve(impl, nir_metadata_block_index |
 243                                nir_metadata_dominance);
 244 }
 245
 246 void
 247 nir_lower_io(nir_shader *shader, nir_variable_mode mode,
 248              int (*type_size)(const struct glsl_type *))
 249 {
 250    nir_foreach_overload(shader, overload) {
 251       if (overload->impl)
 252          nir_lower_io_impl(overload->impl, mode, type_size);
 253    }
 254 }