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 #include "nir_deref.h"
  37
  38 struct lower_io_state {
  39    void *dead_ctx;
  40    nir_builder builder;
  41    int (*type_size)(const struct glsl_type *type);
  42    nir_variable_mode modes;
  43    nir_lower_io_options options;
  44 };
  45
  46 void
  47 nir_assign_var_locations(struct exec_list *var_list, unsigned *size,
  48                          int (*type_size)(const struct glsl_type *))
  49 {
  50    unsigned location = 0;
  51
  52    nir_foreach_variable(var, var_list) {
  53       /*
  54        * UBOs have their own address spaces, so don't count them towards the
  55        * number of global uniforms
  56        */
  57       if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
  58           var->interface_type != NULL)
  59          continue;
  60
  61       var->data.driver_location = location;
  62       location += type_size(var->type);
  63    }
  64
  65    *size = location;
  66 }
  67
  68 /**
  69  * Return true if the given variable is a per-vertex input/output array.
  70  * (such as geometry shader inputs).
  71  */
  72 bool
  73 nir_is_per_vertex_io(const nir_variable *var, gl_shader_stage stage)
  74 {
  75    if (var->data.patch || !glsl_type_is_array(var->type))
  76       return false;
  77
  78    if (var->data.mode == nir_var_shader_in)
  79       return stage == MESA_SHADER_GEOMETRY ||
  80              stage == MESA_SHADER_TESS_CTRL ||
  81              stage == MESA_SHADER_TESS_EVAL;
  82
  83    if (var->data.mode == nir_var_shader_out)
  84       return stage == MESA_SHADER_TESS_CTRL;
  85
  86    return false;
  87 }
  88
  89 static nir_ssa_def *
  90 get_io_offset(nir_builder *b, nir_deref_instr *deref,
  91               nir_ssa_def **vertex_index,
  92               int (*type_size)(const struct glsl_type *),
  93               unsigned *component)
  94 {
  95    nir_deref_path path;
  96    nir_deref_path_init(&path, deref, NULL);
  97
  98    assert(path.path[0]->deref_type == nir_deref_type_var);
  99    nir_deref_instr **p = &path.path[1];
 100
 101    /* For per-vertex input arrays (i.e. geometry shader inputs), keep the
 102     * outermost array index separate.  Process the rest normally.
 103     */
 104    if (vertex_index != NULL) {
 105       assert((*p)->deref_type == nir_deref_type_array);
 106       *vertex_index = nir_ssa_for_src(b, (*p)->arr.index, 1);
 107       p++;
 108    }
 109
 110    if (path.path[0]->var->data.compact) {
 111       assert((*p)->deref_type == nir_deref_type_array);
 112       assert(glsl_type_is_scalar((*p)->type));
 113
 114       /* We always lower indirect dereferences for "compact" array vars. */
 115       nir_const_value *const_index = nir_src_as_const_value((*p)->arr.index);
 116       assert(const_index);
 117
 118       const unsigned total_offset = *component + const_index->u32[0];
 119       const unsigned slot_offset = total_offset / 4;
 120       *component = total_offset % 4;
 121       return nir_imm_int(b, type_size(glsl_vec4_type()) * slot_offset);
 122    }
 123
 124    /* Just emit code and let constant-folding go to town */
 125    nir_ssa_def *offset = nir_imm_int(b, 0);
 126
 127    for (; *p; p++) {
 128       if ((*p)->deref_type == nir_deref_type_array) {
 129          unsigned size = type_size((*p)->type);
 130
 131          nir_ssa_def *mul =
 132             nir_imul(b, nir_imm_int(b, size),
 133                      nir_ssa_for_src(b, (*p)->arr.index, 1));
 134
 135          offset = nir_iadd(b, offset, mul);
 136       } else if ((*p)->deref_type == nir_deref_type_struct) {
 137          /* p starts at path[1], so this is safe */
 138          nir_deref_instr *parent = *(p - 1);
 139
 140          unsigned field_offset = 0;
 141          for (unsigned i = 0; i < (*p)->strct.index; i++) {
 142             field_offset += type_size(glsl_get_struct_field(parent->type, i));
 143          }
 144          offset = nir_iadd(b, offset, nir_imm_int(b, field_offset));
 145       } else {
 146          unreachable("Unsupported deref type");
 147       }
 148    }
 149
 150    nir_deref_path_finish(&path);
 151
 152    return offset;
 153 }
 154
 155 static nir_intrinsic_instr *
 156 lower_load(nir_intrinsic_instr *intrin, struct lower_io_state *state,
 157            nir_ssa_def *vertex_index, nir_variable *var, nir_ssa_def *offset,
 158            unsigned component)
 159 {
 160    const nir_shader *nir = state->builder.shader;
 161    nir_variable_mode mode = var->data.mode;
 162    nir_ssa_def *barycentric = NULL;
 163
 164    nir_intrinsic_op op;
 165    switch (mode) {
 166    case nir_var_shader_in:
 167       if (nir->info.stage == MESA_SHADER_FRAGMENT &&
 168           nir->options->use_interpolated_input_intrinsics &&
 169           var->data.interpolation != INTERP_MODE_FLAT) {
 170          assert(vertex_index == NULL);
 171
 172          nir_intrinsic_op bary_op;
 173          if (var->data.sample ||
 174              (state->options & nir_lower_io_force_sample_interpolation))
 175             bary_op = nir_intrinsic_load_barycentric_sample;
 176          else if (var->data.centroid)
 177             bary_op = nir_intrinsic_load_barycentric_centroid;
 178          else
 179             bary_op = nir_intrinsic_load_barycentric_pixel;
 180
 181          barycentric = nir_load_barycentric(&state->builder, bary_op,
 182                                             var->data.interpolation);
 183          op = nir_intrinsic_load_interpolated_input;
 184       } else {
 185          op = vertex_index ? nir_intrinsic_load_per_vertex_input :
 186                              nir_intrinsic_load_input;
 187       }
 188       break;
 189    case nir_var_shader_out:
 190       op = vertex_index ? nir_intrinsic_load_per_vertex_output :
 191                           nir_intrinsic_load_output;
 192       break;
 193    case nir_var_uniform:
 194       op = nir_intrinsic_load_uniform;
 195       break;
 196    case nir_var_shared:
 197       op = nir_intrinsic_load_shared;
 198       break;
 199    default:
 200       unreachable("Unknown variable mode");
 201    }
 202
 203    nir_intrinsic_instr *load =
 204       nir_intrinsic_instr_create(state->builder.shader, op);
 205    load->num_components = intrin->num_components;
 206
 207    nir_intrinsic_set_base(load, var->data.driver_location);
 208    if (mode == nir_var_shader_in || mode == nir_var_shader_out)
 209       nir_intrinsic_set_component(load, component);
 210
 211    if (load->intrinsic == nir_intrinsic_load_uniform)
 212       nir_intrinsic_set_range(load, state->type_size(var->type));
 213
 214    if (vertex_index) {
 215       load->src[0] = nir_src_for_ssa(vertex_index);
 216       load->src[1] = nir_src_for_ssa(offset);
 217    } else if (barycentric) {
 218       load->src[0] = nir_src_for_ssa(barycentric);
 219       load->src[1] = nir_src_for_ssa(offset);
 220    } else {
 221       load->src[0] = nir_src_for_ssa(offset);
 222    }
 223
 224    return load;
 225 }
 226
 227 static nir_intrinsic_instr *
 228 lower_store(nir_intrinsic_instr *intrin, struct lower_io_state *state,
 229             nir_ssa_def *vertex_index, nir_variable *var, nir_ssa_def *offset,
 230             unsigned component)
 231 {
 232    nir_variable_mode mode = var->data.mode;
 233
 234    nir_intrinsic_op op;
 235    if (mode == nir_var_shared) {
 236       op = nir_intrinsic_store_shared;
 237    } else {
 238       assert(mode == nir_var_shader_out);
 239       op = vertex_index ? nir_intrinsic_store_per_vertex_output :
 240                           nir_intrinsic_store_output;
 241    }
 242
 243    nir_intrinsic_instr *store =
 244       nir_intrinsic_instr_create(state->builder.shader, op);
 245    store->num_components = intrin->num_components;
 246
 247    nir_src_copy(&store->src[0], &intrin->src[1], store);
 248
 249    nir_intrinsic_set_base(store, var->data.driver_location);
 250
 251    if (mode == nir_var_shader_out)
 252       nir_intrinsic_set_component(store, component);
 253
 254    nir_intrinsic_set_write_mask(store, nir_intrinsic_write_mask(intrin));
 255
 256    if (vertex_index)
 257       store->src[1] = nir_src_for_ssa(vertex_index);
 258
 259    store->src[vertex_index ? 2 : 1] = nir_src_for_ssa(offset);
 260
 261    return store;
 262 }
 263
 264 static nir_intrinsic_instr *
 265 lower_atomic(nir_intrinsic_instr *intrin, struct lower_io_state *state,
 266              nir_variable *var, nir_ssa_def *offset)
 267 {
 268    assert(var->data.mode == nir_var_shared);
 269
 270    nir_intrinsic_op op;
 271    switch (intrin->intrinsic) {
 272 #define OP(O) case nir_intrinsic_deref_##O: op = nir_intrinsic_shared_##O; break;
 273    OP(atomic_exchange)
 274    OP(atomic_comp_swap)
 275    OP(atomic_add)
 276    OP(atomic_imin)
 277    OP(atomic_umin)
 278    OP(atomic_imax)
 279    OP(atomic_umax)
 280    OP(atomic_and)
 281    OP(atomic_or)
 282    OP(atomic_xor)
 283 #undef OP
 284    default:
 285       unreachable("Invalid atomic");
 286    }
 287
 288    nir_intrinsic_instr *atomic =
 289       nir_intrinsic_instr_create(state->builder.shader, op);
 290
 291    nir_intrinsic_set_base(atomic, var->data.driver_location);
 292
 293    atomic->src[0] = nir_src_for_ssa(offset);
 294    assert(nir_intrinsic_infos[intrin->intrinsic].num_srcs ==
 295           nir_intrinsic_infos[op].num_srcs);
 296    for (unsigned i = 1; i < nir_intrinsic_infos[op].num_srcs; i++) {
 297       nir_src_copy(&atomic->src[i], &intrin->src[i], atomic);
 298    }
 299
 300    return atomic;
 301 }
 302
 303 static nir_intrinsic_instr *
 304 lower_interpolate_at(nir_intrinsic_instr *intrin, struct lower_io_state *state,
 305                      nir_variable *var, nir_ssa_def *offset, unsigned component)
 306 {
 307    assert(var->data.mode == nir_var_shader_in);
 308
 309    /* Ignore interpolateAt() for flat variables - flat is flat. */
 310    if (var->data.interpolation == INTERP_MODE_FLAT)
 311       return lower_load(intrin, state, NULL, var, offset, component);
 312
 313    nir_intrinsic_op bary_op;
 314    switch (intrin->intrinsic) {
 315    case nir_intrinsic_interp_deref_at_centroid:
 316       bary_op = (state->options & nir_lower_io_force_sample_interpolation) ?
 317                 nir_intrinsic_load_barycentric_sample :
 318                 nir_intrinsic_load_barycentric_centroid;
 319       break;
 320    case nir_intrinsic_interp_deref_at_sample:
 321       bary_op = nir_intrinsic_load_barycentric_at_sample;
 322       break;
 323    case nir_intrinsic_interp_deref_at_offset:
 324       bary_op = nir_intrinsic_load_barycentric_at_offset;
 325       break;
 326    default:
 327       unreachable("Bogus interpolateAt() intrinsic.");
 328    }
 329
 330    nir_intrinsic_instr *bary_setup =
 331       nir_intrinsic_instr_create(state->builder.shader, bary_op);
 332
 333    nir_ssa_dest_init(&bary_setup->instr, &bary_setup->dest, 2, 32, NULL);
 334    nir_intrinsic_set_interp_mode(bary_setup, var->data.interpolation);
 335
 336    if (intrin->intrinsic == nir_intrinsic_interp_deref_at_sample ||
 337        intrin->intrinsic == nir_intrinsic_interp_deref_at_offset)
 338       nir_src_copy(&bary_setup->src[0], &intrin->src[1], bary_setup);
 339
 340    nir_builder_instr_insert(&state->builder, &bary_setup->instr);
 341
 342    nir_intrinsic_instr *load =
 343       nir_intrinsic_instr_create(state->builder.shader,
 344                                  nir_intrinsic_load_interpolated_input);
 345    load->num_components = intrin->num_components;
 346
 347    nir_intrinsic_set_base(load, var->data.driver_location);
 348    nir_intrinsic_set_component(load, component);
 349
 350    load->src[0] = nir_src_for_ssa(&bary_setup->dest.ssa);
 351    load->src[1] = nir_src_for_ssa(offset);
 352
 353    return load;
 354 }
 355
 356 static bool
 357 nir_lower_io_block(nir_block *block,
 358                    struct lower_io_state *state)
 359 {
 360    nir_builder *b = &state->builder;
 361    const nir_shader_compiler_options *options = b->shader->options;
 362    bool progress = false;
 363
 364    nir_foreach_instr_safe(instr, block) {
 365       if (instr->type != nir_instr_type_intrinsic)
 366          continue;
 367
 368       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 369
 370       switch (intrin->intrinsic) {
 371       case nir_intrinsic_load_deref:
 372       case nir_intrinsic_store_deref:
 373       case nir_intrinsic_deref_atomic_add:
 374       case nir_intrinsic_deref_atomic_imin:
 375       case nir_intrinsic_deref_atomic_umin:
 376       case nir_intrinsic_deref_atomic_imax:
 377       case nir_intrinsic_deref_atomic_umax:
 378       case nir_intrinsic_deref_atomic_and:
 379       case nir_intrinsic_deref_atomic_or:
 380       case nir_intrinsic_deref_atomic_xor:
 381       case nir_intrinsic_deref_atomic_exchange:
 382       case nir_intrinsic_deref_atomic_comp_swap:
 383          /* We can lower the io for this nir instrinsic */
 384          break;
 385       case nir_intrinsic_interp_deref_at_centroid:
 386       case nir_intrinsic_interp_deref_at_sample:
 387       case nir_intrinsic_interp_deref_at_offset:
 388          /* We can optionally lower these to load_interpolated_input */
 389          if (options->use_interpolated_input_intrinsics)
 390             break;
 391       default:
 392          /* We can't lower the io for this nir instrinsic, so skip it */
 393          continue;
 394       }
 395
 396       nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
 397
 398       nir_variable *var = nir_deref_instr_get_variable(deref);
 399       nir_variable_mode mode = var->data.mode;
 400
 401       if ((state->modes & mode) == 0)
 402          continue;
 403
 404       if (mode != nir_var_shader_in &&
 405           mode != nir_var_shader_out &&
 406           mode != nir_var_shared &&
 407           mode != nir_var_uniform)
 408          continue;
 409
 410       b->cursor = nir_before_instr(instr);
 411
 412       const bool per_vertex = nir_is_per_vertex_io(var, b->shader->info.stage);
 413
 414       nir_ssa_def *offset;
 415       nir_ssa_def *vertex_index = NULL;
 416       unsigned component_offset = var->data.location_frac;
 417
 418       offset = get_io_offset(b, deref, per_vertex ? &vertex_index : NULL,
 419                              state->type_size, &component_offset);
 420
 421       nir_intrinsic_instr *replacement;
 422
 423       switch (intrin->intrinsic) {
 424       case nir_intrinsic_load_deref:
 425          replacement = lower_load(intrin, state, vertex_index, var, offset,
 426                                   component_offset);
 427          break;
 428
 429       case nir_intrinsic_store_deref:
 430          replacement = lower_store(intrin, state, vertex_index, var, offset,
 431                                    component_offset);
 432          break;
 433
 434       case nir_intrinsic_deref_atomic_add:
 435       case nir_intrinsic_deref_atomic_imin:
 436       case nir_intrinsic_deref_atomic_umin:
 437       case nir_intrinsic_deref_atomic_imax:
 438       case nir_intrinsic_deref_atomic_umax:
 439       case nir_intrinsic_deref_atomic_and:
 440       case nir_intrinsic_deref_atomic_or:
 441       case nir_intrinsic_deref_atomic_xor:
 442       case nir_intrinsic_deref_atomic_exchange:
 443       case nir_intrinsic_deref_atomic_comp_swap:
 444          assert(vertex_index == NULL);
 445          replacement = lower_atomic(intrin, state, var, offset);
 446          break;
 447
 448       case nir_intrinsic_interp_deref_at_centroid:
 449       case nir_intrinsic_interp_deref_at_sample:
 450       case nir_intrinsic_interp_deref_at_offset:
 451          assert(vertex_index == NULL);
 452          replacement = lower_interpolate_at(intrin, state, var, offset,
 453                                             component_offset);
 454          break;
 455
 456       default:
 457          continue;
 458       }
 459
 460       if (nir_intrinsic_infos[intrin->intrinsic].has_dest) {
 461          if (intrin->dest.is_ssa) {
 462             nir_ssa_dest_init(&replacement->instr, &replacement->dest,
 463                               intrin->dest.ssa.num_components,
 464                               intrin->dest.ssa.bit_size, NULL);
 465             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 466                                      nir_src_for_ssa(&replacement->dest.ssa));
 467          } else {
 468             nir_dest_copy(&replacement->dest, &intrin->dest, &intrin->instr);
 469          }
 470       }
 471
 472       nir_instr_insert_before(&intrin->instr, &replacement->instr);
 473       nir_instr_remove(&intrin->instr);
 474       progress = true;
 475    }
 476
 477    return progress;
 478 }
 479
 480 static bool
 481 nir_lower_io_impl(nir_function_impl *impl,
 482                   nir_variable_mode modes,
 483                   int (*type_size)(const struct glsl_type *),
 484                   nir_lower_io_options options)
 485 {
 486    struct lower_io_state state;
 487    bool progress = false;
 488
 489    nir_builder_init(&state.builder, impl);
 490    state.dead_ctx = ralloc_context(NULL);
 491    state.modes = modes;
 492    state.type_size = type_size;
 493    state.options = options;
 494
 495    nir_foreach_block(block, impl) {
 496       progress |= nir_lower_io_block(block, &state);
 497    }
 498
 499    ralloc_free(state.dead_ctx);
 500
 501    nir_metadata_preserve(impl, nir_metadata_block_index |
 502                                nir_metadata_dominance);
 503    return progress;
 504 }
 505
 506 bool
 507 nir_lower_io(nir_shader *shader, nir_variable_mode modes,
 508              int (*type_size)(const struct glsl_type *),
 509              nir_lower_io_options options)
 510 {
 511    bool progress = false;
 512
 513    nir_assert_unlowered_derefs(shader, nir_lower_load_store_derefs | nir_lower_interp_derefs | nir_lower_atomic_derefs);
 514
 515    nir_foreach_function(function, shader) {
 516       if (function->impl) {
 517          progress |= nir_lower_io_impl(function->impl, modes,
 518                                        type_size, options);
 519       }
 520    }
 521
 522    return progress;
 523 }
 524
 525 /**
 526  * Return the offset source for a load/store intrinsic.
 527  */
 528 nir_src *
 529 nir_get_io_offset_src(nir_intrinsic_instr *instr)
 530 {
 531    switch (instr->intrinsic) {
 532    case nir_intrinsic_load_input:
 533    case nir_intrinsic_load_output:
 534    case nir_intrinsic_load_uniform:
 535       return &instr->src[0];
 536    case nir_intrinsic_load_ubo:
 537    case nir_intrinsic_load_ssbo:
 538    case nir_intrinsic_load_per_vertex_input:
 539    case nir_intrinsic_load_per_vertex_output:
 540    case nir_intrinsic_load_interpolated_input:
 541    case nir_intrinsic_store_output:
 542       return &instr->src[1];
 543    case nir_intrinsic_store_ssbo:
 544    case nir_intrinsic_store_per_vertex_output:
 545       return &instr->src[2];
 546    default:
 547       return NULL;
 548    }
 549 }
 550
 551 /**
 552  * Return the vertex index source for a load/store per_vertex intrinsic.
 553  */
 554 nir_src *
 555 nir_get_io_vertex_index_src(nir_intrinsic_instr *instr)
 556 {
 557    switch (instr->intrinsic) {
 558    case nir_intrinsic_load_per_vertex_input:
 559    case nir_intrinsic_load_per_vertex_output:
 560       return &instr->src[0];
 561    case nir_intrinsic_store_per_vertex_output:
 562       return &instr->src[1];
 563    default:
 564       return NULL;
 565    }
 566 }