src/compiler/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 modes;
  42    nir_lower_io_options options;
  43 };
  44
  45 void
  46 nir_assign_var_locations(struct exec_list *var_list, unsigned *size,
  47                          int (*type_size)(const struct glsl_type *))
  48 {
  49    unsigned location = 0;
  50
  51    nir_foreach_variable(var, var_list) {
  52       /*
  53        * UBO's have their own address spaces, so don't count them towards the
  54        * number of global uniforms
  55        */
  56       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
  57           var->interface_type != NULL)
  58          continue;
  59
  60       var->data.driver_location = location;
  61       location += type_size(var->type);
  62    }
  63
  64    *size = location;
  65 }
  66
  67 /**
  68  * Return true if the given variable is a per-vertex input/output array.
  69  * (such as geometry shader inputs).
  70  */
  71 bool
  72 nir_is_per_vertex_io(nir_variable *var, gl_shader_stage stage)
  73 {
  74    if (var->data.patch || !glsl_type_is_array(var->type))
  75       return false;
  76
  77    if (var->data.mode == nir_var_shader_in)
  78       return stage == MESA_SHADER_GEOMETRY ||
  79              stage == MESA_SHADER_TESS_CTRL ||
  80              stage == MESA_SHADER_TESS_EVAL;
  81
  82    if (var->data.mode == nir_var_shader_out)
  83       return stage == MESA_SHADER_TESS_CTRL;
  84
  85    return false;
  86 }
  87
  88 static nir_ssa_def *
  89 get_io_offset(nir_builder *b, nir_deref_var *deref,
  90               nir_ssa_def **vertex_index,
  91               int (*type_size)(const struct glsl_type *))
  92 {
  93    nir_deref *tail = &deref->deref;
  94
  95    /* For per-vertex input arrays (i.e. geometry shader inputs), keep the
  96     * outermost array index separate.  Process the rest normally.
  97     */
  98    if (vertex_index != NULL) {
  99       tail = tail->child;
 100       nir_deref_array *deref_array = nir_deref_as_array(tail);
 101
 102       nir_ssa_def *vtx = nir_imm_int(b, deref_array->base_offset);
 103       if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 104          vtx = nir_iadd(b, vtx, nir_ssa_for_src(b, deref_array->indirect, 1));
 105       }
 106       *vertex_index = vtx;
 107    }
 108
 109    /* Just emit code and let constant-folding go to town */
 110    nir_ssa_def *offset = nir_imm_int(b, 0);
 111
 112    while (tail->child != NULL) {
 113       const struct glsl_type *parent_type = tail->type;
 114       tail = tail->child;
 115
 116       if (tail->deref_type == nir_deref_type_array) {
 117          nir_deref_array *deref_array = nir_deref_as_array(tail);
 118          unsigned size = type_size(tail->type);
 119
 120          offset = nir_iadd(b, offset,
 121                            nir_imm_int(b, size * deref_array->base_offset));
 122
 123          if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 124             nir_ssa_def *mul =
 125                nir_imul(b, nir_imm_int(b, size),
 126                         nir_ssa_for_src(b, deref_array->indirect, 1));
 127
 128             offset = nir_iadd(b, offset, mul);
 129          }
 130       } else if (tail->deref_type == nir_deref_type_struct) {
 131          nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
 132
 133          unsigned field_offset = 0;
 134          for (unsigned i = 0; i < deref_struct->index; i++) {
 135             field_offset += type_size(glsl_get_struct_field(parent_type, i));
 136          }
 137          offset = nir_iadd(b, offset, nir_imm_int(b, field_offset));
 138       }
 139    }
 140
 141    return offset;
 142 }
 143
 144 static nir_intrinsic_instr *
 145 lower_load(nir_intrinsic_instr *intrin, struct lower_io_state *state,
 146            nir_ssa_def *vertex_index, nir_ssa_def *offset)
 147 {
 148    const nir_shader *nir = state->builder.shader;
 149    nir_variable *var = intrin->variables[0]->var;
 150    nir_variable_mode mode = var->data.mode;
 151    nir_ssa_def *barycentric = NULL;
 152
 153    nir_intrinsic_op op;
 154    switch (mode) {
 155    case nir_var_shader_in:
 156       if (nir->stage == MESA_SHADER_FRAGMENT &&
 157           nir->options->use_interpolated_input_intrinsics &&
 158           var->data.interpolation != INTERP_MODE_FLAT) {
 159          assert(vertex_index == NULL);
 160
 161          nir_intrinsic_op bary_op;
 162          if (var->data.sample ||
 163              (state->options & nir_lower_io_force_sample_interpolation))
 164             bary_op = nir_intrinsic_load_barycentric_sample;
 165          else if (var->data.centroid)
 166             bary_op = nir_intrinsic_load_barycentric_centroid;
 167          else
 168             bary_op = nir_intrinsic_load_barycentric_pixel;
 169
 170          barycentric = nir_load_barycentric(&state->builder, bary_op,
 171                                             var->data.interpolation);
 172          op = nir_intrinsic_load_interpolated_input;
 173       } else {
 174          op = vertex_index ? nir_intrinsic_load_per_vertex_input :
 175                              nir_intrinsic_load_input;
 176       }
 177       break;
 178    case nir_var_shader_out:
 179       op = vertex_index ? nir_intrinsic_load_per_vertex_output :
 180                           nir_intrinsic_load_output;
 181       break;
 182    case nir_var_uniform:
 183       op = nir_intrinsic_load_uniform;
 184       break;
 185    case nir_var_shared:
 186       op = nir_intrinsic_load_shared;
 187       break;
 188    default:
 189       unreachable("Unknown variable mode");
 190    }
 191
 192    nir_intrinsic_instr *load = nir_intrinsic_instr_create(state->mem_ctx, op);
 193    load->num_components = intrin->num_components;
 194
 195    nir_intrinsic_set_base(load, var->data.driver_location);
 196    if (mode == nir_var_shader_in || mode == nir_var_shader_out)
 197       nir_intrinsic_set_component(load, var->data.location_frac);
 198
 199    if (load->intrinsic == nir_intrinsic_load_uniform)
 200       nir_intrinsic_set_range(load, state->type_size(var->type));
 201
 202    if (vertex_index) {
 203       load->src[0] = nir_src_for_ssa(vertex_index);
 204       load->src[1] = nir_src_for_ssa(offset);
 205    } else if (barycentric) {
 206       load->src[0] = nir_src_for_ssa(barycentric);
 207       load->src[1] = nir_src_for_ssa(offset);
 208    } else {
 209       load->src[0] = nir_src_for_ssa(offset);
 210    }
 211
 212    return load;
 213 }
 214
 215 static nir_intrinsic_instr *
 216 lower_store(nir_intrinsic_instr *intrin, struct lower_io_state *state,
 217             nir_ssa_def *vertex_index, nir_ssa_def *offset)
 218 {
 219    nir_variable *var = intrin->variables[0]->var;
 220    nir_variable_mode mode = var->data.mode;
 221
 222    nir_intrinsic_op op;
 223    if (mode == nir_var_shared) {
 224       op = nir_intrinsic_store_shared;
 225    } else {
 226       assert(mode == nir_var_shader_out);
 227       op = vertex_index ? nir_intrinsic_store_per_vertex_output :
 228                           nir_intrinsic_store_output;
 229    }
 230
 231    nir_intrinsic_instr *store = nir_intrinsic_instr_create(state->mem_ctx, op);
 232    store->num_components = intrin->num_components;
 233
 234    nir_src_copy(&store->src[0], &intrin->src[0], store);
 235
 236    nir_intrinsic_set_base(store, var->data.driver_location);
 237
 238    if (mode == nir_var_shader_out)
 239       nir_intrinsic_set_component(store, var->data.location_frac);
 240
 241    nir_intrinsic_set_write_mask(store, nir_intrinsic_write_mask(intrin));
 242
 243    if (vertex_index)
 244       store->src[1] = nir_src_for_ssa(vertex_index);
 245
 246    store->src[vertex_index ? 2 : 1] = nir_src_for_ssa(offset);
 247
 248    return store;
 249 }
 250
 251 static nir_intrinsic_instr *
 252 lower_atomic(nir_intrinsic_instr *intrin, struct lower_io_state *state,
 253              nir_ssa_def *offset)
 254 {
 255    nir_variable *var = intrin->variables[0]->var;
 256
 257    assert(var->data.mode == nir_var_shared);
 258
 259    nir_intrinsic_op op;
 260    switch (intrin->intrinsic) {
 261 #define OP(O) case nir_intrinsic_var_##O: op = nir_intrinsic_shared_##O; break;
 262    OP(atomic_exchange)
 263    OP(atomic_comp_swap)
 264    OP(atomic_add)
 265    OP(atomic_imin)
 266    OP(atomic_umin)
 267    OP(atomic_imax)
 268    OP(atomic_umax)
 269    OP(atomic_and)
 270    OP(atomic_or)
 271    OP(atomic_xor)
 272 #undef OP
 273    default:
 274       unreachable("Invalid atomic");
 275    }
 276
 277    nir_intrinsic_instr *atomic =
 278       nir_intrinsic_instr_create(state->mem_ctx, op);
 279
 280    nir_intrinsic_set_base(atomic, var->data.driver_location);
 281
 282    atomic->src[0] = nir_src_for_ssa(offset);
 283    for (unsigned i = 0; i < nir_intrinsic_infos[intrin->intrinsic].num_srcs; i++) {
 284       nir_src_copy(&atomic->src[i+1], &intrin->src[i], atomic);
 285    }
 286
 287    return atomic;
 288 }
 289
 290 static nir_intrinsic_instr *
 291 lower_interpolate_at(nir_intrinsic_instr *intrin, struct lower_io_state *state,
 292                      nir_ssa_def *offset)
 293 {
 294    nir_variable *var = intrin->variables[0]->var;
 295
 296    assert(var->data.mode == nir_var_shader_in);
 297
 298    /* Ignore interpolateAt() for flat variables - flat is flat. */
 299    if (var->data.interpolation == INTERP_MODE_FLAT)
 300       return lower_load(intrin, state, NULL, offset);
 301
 302    nir_intrinsic_op bary_op;
 303    switch (intrin->intrinsic) {
 304    case nir_intrinsic_interp_var_at_centroid:
 305       bary_op = (state->options & nir_lower_io_force_sample_interpolation) ?
 306                 nir_intrinsic_load_barycentric_sample :
 307                 nir_intrinsic_load_barycentric_centroid;
 308       break;
 309    case nir_intrinsic_interp_var_at_sample:
 310       bary_op = nir_intrinsic_load_barycentric_at_sample;
 311       break;
 312    case nir_intrinsic_interp_var_at_offset:
 313       bary_op = nir_intrinsic_load_barycentric_at_offset;
 314       break;
 315    default:
 316       unreachable("Bogus interpolateAt() intrinsic.");
 317    }
 318
 319    nir_intrinsic_instr *bary_setup =
 320       nir_intrinsic_instr_create(state->mem_ctx, bary_op);
 321
 322    nir_ssa_dest_init(&bary_setup->instr, &bary_setup->dest, 2, 32, NULL);
 323    nir_intrinsic_set_interp_mode(bary_setup, var->data.interpolation);
 324
 325    if (intrin->intrinsic != nir_intrinsic_interp_var_at_centroid)
 326       nir_src_copy(&bary_setup->src[0], &intrin->src[0], bary_setup);
 327
 328    nir_builder_instr_insert(&state->builder, &bary_setup->instr);
 329
 330    nir_intrinsic_instr *load =
 331       nir_intrinsic_instr_create(state->mem_ctx,
 332                                  nir_intrinsic_load_interpolated_input);
 333    load->num_components = intrin->num_components;
 334
 335    nir_intrinsic_set_base(load, var->data.driver_location);
 336    nir_intrinsic_set_component(load, var->data.location_frac);
 337
 338    load->src[0] = nir_src_for_ssa(&bary_setup->dest.ssa);
 339    load->src[1] = nir_src_for_ssa(offset);
 340
 341    return load;
 342 }
 343
 344 static bool
 345 nir_lower_io_block(nir_block *block,
 346                    struct lower_io_state *state)
 347 {
 348    nir_builder *b = &state->builder;
 349    const nir_shader_compiler_options *options = b->shader->options;
 350
 351    nir_foreach_instr_safe(instr, block) {
 352       if (instr->type != nir_instr_type_intrinsic)
 353          continue;
 354
 355       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 356
 357       switch (intrin->intrinsic) {
 358       case nir_intrinsic_load_var:
 359       case nir_intrinsic_store_var:
 360       case nir_intrinsic_var_atomic_add:
 361       case nir_intrinsic_var_atomic_imin:
 362       case nir_intrinsic_var_atomic_umin:
 363       case nir_intrinsic_var_atomic_imax:
 364       case nir_intrinsic_var_atomic_umax:
 365       case nir_intrinsic_var_atomic_and:
 366       case nir_intrinsic_var_atomic_or:
 367       case nir_intrinsic_var_atomic_xor:
 368       case nir_intrinsic_var_atomic_exchange:
 369       case nir_intrinsic_var_atomic_comp_swap:
 370          /* We can lower the io for this nir instrinsic */
 371          break;
 372       case nir_intrinsic_interp_var_at_centroid:
 373       case nir_intrinsic_interp_var_at_sample:
 374       case nir_intrinsic_interp_var_at_offset:
 375          /* We can optionally lower these to load_interpolated_input */
 376          if (options->use_interpolated_input_intrinsics)
 377             break;
 378       default:
 379          /* We can't lower the io for this nir instrinsic, so skip it */
 380          continue;
 381       }
 382
 383       nir_variable *var = intrin->variables[0]->var;
 384       nir_variable_mode mode = var->data.mode;
 385
 386       if ((state->modes & mode) == 0)
 387          continue;
 388
 389       if (mode != nir_var_shader_in &&
 390           mode != nir_var_shader_out &&
 391           mode != nir_var_shared &&
 392           mode != nir_var_uniform)
 393          continue;
 394
 395       b->cursor = nir_before_instr(instr);
 396
 397       const bool per_vertex = nir_is_per_vertex_io(var, b->shader->stage);
 398
 399       nir_ssa_def *offset;
 400       nir_ssa_def *vertex_index = NULL;
 401
 402       offset = get_io_offset(b, intrin->variables[0],
 403                              per_vertex ? &vertex_index : NULL,
 404                              state->type_size);
 405
 406       nir_intrinsic_instr *replacement;
 407
 408       switch (intrin->intrinsic) {
 409       case nir_intrinsic_load_var:
 410          replacement = lower_load(intrin, state, vertex_index, offset);
 411          break;
 412
 413       case nir_intrinsic_store_var:
 414          replacement = lower_store(intrin, state, vertex_index, offset);
 415          break;
 416
 417       case nir_intrinsic_var_atomic_add:
 418       case nir_intrinsic_var_atomic_imin:
 419       case nir_intrinsic_var_atomic_umin:
 420       case nir_intrinsic_var_atomic_imax:
 421       case nir_intrinsic_var_atomic_umax:
 422       case nir_intrinsic_var_atomic_and:
 423       case nir_intrinsic_var_atomic_or:
 424       case nir_intrinsic_var_atomic_xor:
 425       case nir_intrinsic_var_atomic_exchange:
 426       case nir_intrinsic_var_atomic_comp_swap:
 427          assert(vertex_index == NULL);
 428          replacement = lower_atomic(intrin, state, offset);
 429          break;
 430
 431       case nir_intrinsic_interp_var_at_centroid:
 432       case nir_intrinsic_interp_var_at_sample:
 433       case nir_intrinsic_interp_var_at_offset:
 434          assert(vertex_index == NULL);
 435          replacement = lower_interpolate_at(intrin, state, offset);
 436          break;
 437
 438       default:
 439          continue;
 440       }
 441
 442       if (nir_intrinsic_infos[intrin->intrinsic].has_dest) {
 443          if (intrin->dest.is_ssa) {
 444             nir_ssa_dest_init(&replacement->instr, &replacement->dest,
 445                               intrin->dest.ssa.num_components,
 446                               intrin->dest.ssa.bit_size, NULL);
 447             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 448                                      nir_src_for_ssa(&replacement->dest.ssa));
 449          } else {
 450             nir_dest_copy(&replacement->dest, &intrin->dest, state->mem_ctx);
 451          }
 452       }
 453
 454       nir_instr_insert_before(&intrin->instr, &replacement->instr);
 455       nir_instr_remove(&intrin->instr);
 456    }
 457
 458    return true;
 459 }
 460
 461 static void
 462 nir_lower_io_impl(nir_function_impl *impl,
 463                   nir_variable_mode modes,
 464                   int (*type_size)(const struct glsl_type *),
 465                   nir_lower_io_options options)
 466 {
 467    struct lower_io_state state;
 468
 469    nir_builder_init(&state.builder, impl);
 470    state.mem_ctx = ralloc_parent(impl);
 471    state.modes = modes;
 472    state.type_size = type_size;
 473    state.options = options;
 474
 475    nir_foreach_block(block, impl) {
 476       nir_lower_io_block(block, &state);
 477    }
 478
 479    nir_metadata_preserve(impl, nir_metadata_block_index |
 480                                nir_metadata_dominance);
 481 }
 482
 483 void
 484 nir_lower_io(nir_shader *shader, nir_variable_mode modes,
 485              int (*type_size)(const struct glsl_type *),
 486              nir_lower_io_options options)
 487 {
 488    nir_foreach_function(function, shader) {
 489       if (function->impl) {
 490          nir_lower_io_impl(function->impl, modes, type_size, options);
 491       }
 492    }
 493 }
 494
 495 /**
 496  * Return the offset soruce for a load/store intrinsic.
 497  */
 498 nir_src *
 499 nir_get_io_offset_src(nir_intrinsic_instr *instr)
 500 {
 501    switch (instr->intrinsic) {
 502    case nir_intrinsic_load_input:
 503    case nir_intrinsic_load_output:
 504    case nir_intrinsic_load_uniform:
 505       return &instr->src[0];
 506    case nir_intrinsic_load_ubo:
 507    case nir_intrinsic_load_ssbo:
 508    case nir_intrinsic_load_per_vertex_input:
 509    case nir_intrinsic_load_per_vertex_output:
 510    case nir_intrinsic_load_interpolated_input:
 511    case nir_intrinsic_store_output:
 512       return &instr->src[1];
 513    case nir_intrinsic_store_ssbo:
 514    case nir_intrinsic_store_per_vertex_output:
 515       return &instr->src[2];
 516    default:
 517       return NULL;
 518    }
 519 }
 520
 521 /**
 522  * Return the vertex index source for a load/store per_vertex intrinsic.
 523  */
 524 nir_src *
 525 nir_get_io_vertex_index_src(nir_intrinsic_instr *instr)
 526 {
 527    switch (instr->intrinsic) {
 528    case nir_intrinsic_load_per_vertex_input:
 529    case nir_intrinsic_load_per_vertex_output:
 530       return &instr->src[0];
 531    case nir_intrinsic_store_per_vertex_output:
 532       return &instr->src[1];
 533    default:
 534       return NULL;
 535    }
 536 }