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 /**
  67  * Returns true if we're processing a stage whose inputs are arrays indexed
  68  * by a vertex number (such as geometry shader inputs).
  69  */
  70 static bool
  71 stage_uses_per_vertex_inputs(struct lower_io_state *state)
  72 {
  73    gl_shader_stage stage = state->builder.shader->stage;
  74    return stage == MESA_SHADER_GEOMETRY;
  75 }
  76
  77 static unsigned
  78 get_io_offset(nir_deref_var *deref, nir_instr *instr,
  79               nir_ssa_def **vertex_index,
  80               nir_ssa_def **out_indirect,
  81               struct lower_io_state *state)
  82 {
  83    nir_ssa_def *indirect = NULL;
  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
  91    /* For per-vertex input arrays (i.e. geometry shader inputs), keep the
  92     * outermost array index separate.  Process the rest normally.
  93     */
  94    if (vertex_index != NULL) {
  95       tail = tail->child;
  96       assert(tail->deref_type == nir_deref_type_array);
  97       nir_deref_array *deref_array = nir_deref_as_array(tail);
  98
  99       nir_ssa_def *vtx = nir_imm_int(b, deref_array->base_offset);
 100       if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 101          vtx = nir_iadd(b, vtx, nir_ssa_for_src(b, deref_array->indirect, 1));
 102       }
 103       *vertex_index = vtx;
 104    }
 105
 106    while (tail->child != NULL) {
 107       const struct glsl_type *parent_type = tail->type;
 108       tail = tail->child;
 109
 110       if (tail->deref_type == nir_deref_type_array) {
 111          nir_deref_array *deref_array = nir_deref_as_array(tail);
 112          unsigned size = state->type_size(tail->type);
 113
 114          base_offset += size * deref_array->base_offset;
 115
 116          if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 117             nir_ssa_def *mul =
 118                nir_imul(b, nir_imm_int(b, size),
 119                         nir_ssa_for_src(b, deref_array->indirect, 1));
 120
 121             indirect = indirect ? nir_iadd(b, indirect, mul) : mul;
 122          }
 123       } else if (tail->deref_type == nir_deref_type_struct) {
 124          nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
 125
 126          for (unsigned i = 0; i < deref_struct->index; i++) {
 127             base_offset +=
 128                state->type_size(glsl_get_struct_field(parent_type, i));
 129          }
 130       }
 131    }
 132
 133    *out_indirect = indirect;
 134    return base_offset;
 135 }
 136
 137 static nir_intrinsic_op
 138 load_op(struct lower_io_state *state,
 139         nir_variable_mode mode, bool per_vertex, bool has_indirect)
 140 {
 141    nir_intrinsic_op op;
 142    switch (mode) {
 143    case nir_var_shader_in:
 144       if (per_vertex) {
 145          op = has_indirect ? nir_intrinsic_load_per_vertex_input_indirect :
 146                              nir_intrinsic_load_per_vertex_input;
 147       } else {
 148          op = has_indirect ? nir_intrinsic_load_input_indirect :
 149                              nir_intrinsic_load_input;
 150       }
 151       break;
 152    case nir_var_uniform:
 153       op = has_indirect ? nir_intrinsic_load_uniform_indirect :
 154                           nir_intrinsic_load_uniform;
 155       break;
 156    default:
 157       unreachable("Unknown variable mode");
 158    }
 159    return op;
 160 }
 161
 162 static bool
 163 nir_lower_io_block(nir_block *block, void *void_state)
 164 {
 165    struct lower_io_state *state = void_state;
 166
 167    nir_foreach_instr_safe(block, instr) {
 168       if (instr->type != nir_instr_type_intrinsic)
 169          continue;
 170
 171       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 172
 173       if (intrin->intrinsic != nir_intrinsic_load_var &&
 174           intrin->intrinsic != nir_intrinsic_store_var)
 175          continue;
 176
 177       nir_variable_mode mode = intrin->variables[0]->var->data.mode;
 178
 179       if (state->mode != -1 && state->mode != mode)
 180          continue;
 181
 182       switch (intrin->intrinsic) {
 183       case nir_intrinsic_load_var: {
 184          if (mode != nir_var_shader_in && mode != nir_var_uniform)
 185             continue;
 186
 187          bool per_vertex = stage_uses_per_vertex_inputs(state) &&
 188                            mode == nir_var_shader_in;
 189
 190          nir_ssa_def *indirect;
 191          nir_ssa_def *vertex_index;
 192
 193          unsigned offset = get_io_offset(intrin->variables[0], &intrin->instr,
 194                                          per_vertex ? &vertex_index : NULL,
 195                                          &indirect, state);
 196
 197          nir_intrinsic_instr *load =
 198             nir_intrinsic_instr_create(state->mem_ctx,
 199                                        load_op(state, mode, per_vertex,
 200                                                indirect));
 201          load->num_components = intrin->num_components;
 202
 203          unsigned location = intrin->variables[0]->var->data.driver_location;
 204          if (mode == nir_var_uniform) {
 205             load->const_index[0] = location;
 206             load->const_index[1] = offset;
 207          } else {
 208             load->const_index[0] = location + offset;
 209          }
 210
 211          if (per_vertex)
 212             load->src[0] = nir_src_for_ssa(vertex_index);
 213
 214          if (indirect)
 215             load->src[per_vertex ? 1 : 0] = nir_src_for_ssa(indirect);
 216
 217          if (intrin->dest.is_ssa) {
 218             nir_ssa_dest_init(&load->instr, &load->dest,
 219                               intrin->num_components, NULL);
 220             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 221                                      nir_src_for_ssa(&load->dest.ssa));
 222          } else {
 223             nir_dest_copy(&load->dest, &intrin->dest, state->mem_ctx);
 224          }
 225
 226          nir_instr_insert_before(&intrin->instr, &load->instr);
 227          nir_instr_remove(&intrin->instr);
 228          break;
 229       }
 230
 231       case nir_intrinsic_store_var: {
 232          if (intrin->variables[0]->var->data.mode != nir_var_shader_out)
 233             continue;
 234
 235          nir_ssa_def *indirect;
 236
 237          unsigned offset = get_io_offset(intrin->variables[0], &intrin->instr,
 238                                          NULL, &indirect, state);
 239          offset += intrin->variables[0]->var->data.driver_location;
 240
 241          nir_intrinsic_op store_op;
 242          if (indirect) {
 243             store_op = nir_intrinsic_store_output_indirect;
 244          } else {
 245             store_op = nir_intrinsic_store_output;
 246          }
 247
 248          nir_intrinsic_instr *store = nir_intrinsic_instr_create(state->mem_ctx,
 249                                                                  store_op);
 250          store->num_components = intrin->num_components;
 251          store->const_index[0] = offset;
 252
 253          nir_src_copy(&store->src[0], &intrin->src[0], store);
 254
 255          if (indirect)
 256             store->src[1] = nir_src_for_ssa(indirect);
 257
 258          nir_instr_insert_before(&intrin->instr, &store->instr);
 259          nir_instr_remove(&intrin->instr);
 260          break;
 261       }
 262
 263       default:
 264          break;
 265       }
 266    }
 267
 268    return true;
 269 }
 270
 271 static void
 272 nir_lower_io_impl(nir_function_impl *impl,
 273                   nir_variable_mode mode,
 274                   int (*type_size)(const struct glsl_type *))
 275 {
 276    struct lower_io_state state;
 277
 278    nir_builder_init(&state.builder, impl);
 279    state.mem_ctx = ralloc_parent(impl);
 280    state.mode = mode;
 281    state.type_size = type_size;
 282
 283    nir_foreach_block(impl, nir_lower_io_block, &state);
 284
 285    nir_metadata_preserve(impl, nir_metadata_block_index |
 286                                nir_metadata_dominance);
 287 }
 288
 289 void
 290 nir_lower_io(nir_shader *shader, nir_variable_mode mode,
 291              int (*type_size)(const struct glsl_type *))
 292 {
 293    nir_foreach_overload(shader, overload) {
 294       if (overload->impl)
 295          nir_lower_io_impl(overload->impl, mode, type_size);
 296    }
 297 }