src/compiler/nir/nir_builder.h

   1 /*
   2  * Copyright © 2014-2015 Broadcom
   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
  24 #ifndef NIR_BUILDER_H
  25 #define NIR_BUILDER_H
  26
  27 #include "nir_control_flow.h"
  28 #include "util/bitscan.h"
  29 #include "util/half_float.h"
  30
  31 struct exec_list;
  32
  33 typedef struct nir_builder {
  34    nir_cursor cursor;
  35
  36    /* Whether new ALU instructions will be marked "exact" */
  37    bool exact;
  38
  39    nir_shader *shader;
  40    nir_function_impl *impl;
  41 } nir_builder;
  42
  43 static inline void
  44 nir_builder_init(nir_builder *build, nir_function_impl *impl)
  45 {
  46    memset(build, 0, sizeof(*build));
  47    build->exact = false;
  48    build->impl = impl;
  49    build->shader = impl->function->shader;
  50 }
  51
  52 static inline void
  53 nir_builder_init_simple_shader(nir_builder *build, void *mem_ctx,
  54                                gl_shader_stage stage,
  55                                const nir_shader_compiler_options *options)
  56 {
  57    build->shader = nir_shader_create(mem_ctx, stage, options, NULL);
  58    nir_function *func = nir_function_create(build->shader, "main");
  59    func->is_entrypoint = true;
  60    build->exact = false;
  61    build->impl = nir_function_impl_create(func);
  62    build->cursor = nir_after_cf_list(&build->impl->body);
  63 }
  64
  65 static inline void
  66 nir_builder_instr_insert(nir_builder *build, nir_instr *instr)
  67 {
  68    nir_instr_insert(build->cursor, instr);
  69
  70    /* Move the cursor forward. */
  71    build->cursor = nir_after_instr(instr);
  72 }
  73
  74 static inline nir_instr *
  75 nir_builder_last_instr(nir_builder *build)
  76 {
  77    assert(build->cursor.option == nir_cursor_after_instr);
  78    return build->cursor.instr;
  79 }
  80
  81 static inline void
  82 nir_builder_cf_insert(nir_builder *build, nir_cf_node *cf)
  83 {
  84    nir_cf_node_insert(build->cursor, cf);
  85 }
  86
  87 static inline bool
  88 nir_builder_is_inside_cf(nir_builder *build, nir_cf_node *cf_node)
  89 {
  90    nir_block *block = nir_cursor_current_block(build->cursor);
  91    for (nir_cf_node *n = &block->cf_node; n; n = n->parent) {
  92       if (n == cf_node)
  93          return true;
  94    }
  95    return false;
  96 }
  97
  98 static inline nir_if *
  99 nir_push_if(nir_builder *build, nir_ssa_def *condition)
 100 {
 101    nir_if *nif = nir_if_create(build->shader);
 102    nif->condition = nir_src_for_ssa(condition);
 103    nir_builder_cf_insert(build, &nif->cf_node);
 104    build->cursor = nir_before_cf_list(&nif->then_list);
 105    return nif;
 106 }
 107
 108 static inline nir_if *
 109 nir_push_else(nir_builder *build, nir_if *nif)
 110 {
 111    if (nif) {
 112       assert(nir_builder_is_inside_cf(build, &nif->cf_node));
 113    } else {
 114       nir_block *block = nir_cursor_current_block(build->cursor);
 115       nif = nir_cf_node_as_if(block->cf_node.parent);
 116    }
 117    build->cursor = nir_before_cf_list(&nif->else_list);
 118    return nif;
 119 }
 120
 121 static inline void
 122 nir_pop_if(nir_builder *build, nir_if *nif)
 123 {
 124    if (nif) {
 125       assert(nir_builder_is_inside_cf(build, &nif->cf_node));
 126    } else {
 127       nir_block *block = nir_cursor_current_block(build->cursor);
 128       nif = nir_cf_node_as_if(block->cf_node.parent);
 129    }
 130    build->cursor = nir_after_cf_node(&nif->cf_node);
 131 }
 132
 133 static inline nir_ssa_def *
 134 nir_if_phi(nir_builder *build, nir_ssa_def *then_def, nir_ssa_def *else_def)
 135 {
 136    nir_block *block = nir_cursor_current_block(build->cursor);
 137    nir_if *nif = nir_cf_node_as_if(nir_cf_node_prev(&block->cf_node));
 138
 139    nir_phi_instr *phi = nir_phi_instr_create(build->shader);
 140
 141    nir_phi_src *src = ralloc(phi, nir_phi_src);
 142    src->pred = nir_if_last_then_block(nif);
 143    src->src = nir_src_for_ssa(then_def);
 144    exec_list_push_tail(&phi->srcs, &src->node);
 145
 146    src = ralloc(phi, nir_phi_src);
 147    src->pred = nir_if_last_else_block(nif);
 148    src->src = nir_src_for_ssa(else_def);
 149    exec_list_push_tail(&phi->srcs, &src->node);
 150
 151    assert(then_def->num_components == else_def->num_components);
 152    assert(then_def->bit_size == else_def->bit_size);
 153    nir_ssa_dest_init(&phi->instr, &phi->dest,
 154                      then_def->num_components, then_def->bit_size, NULL);
 155
 156    nir_builder_instr_insert(build, &phi->instr);
 157
 158    return &phi->dest.ssa;
 159 }
 160
 161 static inline nir_loop *
 162 nir_push_loop(nir_builder *build)
 163 {
 164    nir_loop *loop = nir_loop_create(build->shader);
 165    nir_builder_cf_insert(build, &loop->cf_node);
 166    build->cursor = nir_before_cf_list(&loop->body);
 167    return loop;
 168 }
 169
 170 static inline void
 171 nir_pop_loop(nir_builder *build, nir_loop *loop)
 172 {
 173    if (loop) {
 174       assert(nir_builder_is_inside_cf(build, &loop->cf_node));
 175    } else {
 176       nir_block *block = nir_cursor_current_block(build->cursor);
 177       loop = nir_cf_node_as_loop(block->cf_node.parent);
 178    }
 179    build->cursor = nir_after_cf_node(&loop->cf_node);
 180 }
 181
 182 static inline nir_ssa_def *
 183 nir_ssa_undef(nir_builder *build, unsigned num_components, unsigned bit_size)
 184 {
 185    nir_ssa_undef_instr *undef =
 186       nir_ssa_undef_instr_create(build->shader, num_components, bit_size);
 187    if (!undef)
 188       return NULL;
 189
 190    nir_instr_insert(nir_before_cf_list(&build->impl->body), &undef->instr);
 191
 192    return &undef->def;
 193 }
 194
 195 static inline nir_ssa_def *
 196 nir_build_imm(nir_builder *build, unsigned num_components,
 197               unsigned bit_size, const nir_const_value *value)
 198 {
 199    nir_load_const_instr *load_const =
 200       nir_load_const_instr_create(build->shader, num_components, bit_size);
 201    if (!load_const)
 202       return NULL;
 203
 204    memcpy(load_const->value, value, sizeof(nir_const_value) * num_components);
 205
 206    nir_builder_instr_insert(build, &load_const->instr);
 207
 208    return &load_const->def;
 209 }
 210
 211 static inline nir_ssa_def *
 212 nir_imm_zero(nir_builder *build, unsigned num_components, unsigned bit_size)
 213 {
 214    nir_load_const_instr *load_const =
 215       nir_load_const_instr_create(build->shader, num_components, bit_size);
 216
 217    /* nir_load_const_instr_create uses rzalloc so it's already zero */
 218
 219    nir_builder_instr_insert(build, &load_const->instr);
 220
 221    return &load_const->def;
 222 }
 223
 224 static inline nir_ssa_def *
 225 nir_imm_boolN_t(nir_builder *build, bool x, unsigned bit_size)
 226 {
 227    nir_const_value v = nir_const_value_for_bool(x, bit_size);
 228    return nir_build_imm(build, 1, bit_size, &v);
 229 }
 230
 231 static inline nir_ssa_def *
 232 nir_imm_bool(nir_builder *build, bool x)
 233 {
 234    return nir_imm_boolN_t(build, x, 1);
 235 }
 236
 237 static inline nir_ssa_def *
 238 nir_imm_true(nir_builder *build)
 239 {
 240    return nir_imm_bool(build, true);
 241 }
 242
 243 static inline nir_ssa_def *
 244 nir_imm_false(nir_builder *build)
 245 {
 246    return nir_imm_bool(build, false);
 247 }
 248
 249 static inline nir_ssa_def *
 250 nir_imm_floatN_t(nir_builder *build, double x, unsigned bit_size)
 251 {
 252    nir_const_value v = nir_const_value_for_float(x, bit_size);
 253    return nir_build_imm(build, 1, bit_size, &v);
 254 }
 255
 256 static inline nir_ssa_def *
 257 nir_imm_float16(nir_builder *build, float x)
 258 {
 259    return nir_imm_floatN_t(build, x, 16);
 260 }
 261
 262 static inline nir_ssa_def *
 263 nir_imm_float(nir_builder *build, float x)
 264 {
 265    return nir_imm_floatN_t(build, x, 32);
 266 }
 267
 268 static inline nir_ssa_def *
 269 nir_imm_double(nir_builder *build, double x)
 270 {
 271    return nir_imm_floatN_t(build, x, 64);
 272 }
 273
 274 static inline nir_ssa_def *
 275 nir_imm_vec2(nir_builder *build, float x, float y)
 276 {
 277    nir_const_value v[2] = {
 278       nir_const_value_for_float(x, 32),
 279       nir_const_value_for_float(y, 32),
 280    };
 281    return nir_build_imm(build, 2, 32, v);
 282 }
 283
 284 static inline nir_ssa_def *
 285 nir_imm_vec4(nir_builder *build, float x, float y, float z, float w)
 286 {
 287    nir_const_value v[4] = {
 288       nir_const_value_for_float(x, 32),
 289       nir_const_value_for_float(y, 32),
 290       nir_const_value_for_float(z, 32),
 291       nir_const_value_for_float(w, 32),
 292    };
 293
 294    return nir_build_imm(build, 4, 32, v);
 295 }
 296
 297 static inline nir_ssa_def *
 298 nir_imm_vec4_16(nir_builder *build, float x, float y, float z, float w)
 299 {
 300    nir_const_value v[4] = {
 301       nir_const_value_for_float(x, 16),
 302       nir_const_value_for_float(y, 16),
 303       nir_const_value_for_float(z, 16),
 304       nir_const_value_for_float(w, 16),
 305    };
 306
 307    return nir_build_imm(build, 4, 16, v);
 308 }
 309
 310 static inline nir_ssa_def *
 311 nir_imm_intN_t(nir_builder *build, uint64_t x, unsigned bit_size)
 312 {
 313    nir_const_value v = nir_const_value_for_raw_uint(x, bit_size);
 314    return nir_build_imm(build, 1, bit_size, &v);
 315 }
 316
 317 static inline nir_ssa_def *
 318 nir_imm_int(nir_builder *build, int x)
 319 {
 320    return nir_imm_intN_t(build, x, 32);
 321 }
 322
 323 static inline nir_ssa_def *
 324 nir_imm_int64(nir_builder *build, int64_t x)
 325 {
 326    return nir_imm_intN_t(build, x, 64);
 327 }
 328
 329 static inline nir_ssa_def *
 330 nir_imm_ivec2(nir_builder *build, int x, int y)
 331 {
 332    nir_const_value v[2] = {
 333       nir_const_value_for_int(x, 32),
 334       nir_const_value_for_int(y, 32),
 335    };
 336
 337    return nir_build_imm(build, 2, 32, v);
 338 }
 339
 340 static inline nir_ssa_def *
 341 nir_imm_ivec4(nir_builder *build, int x, int y, int z, int w)
 342 {
 343    nir_const_value v[4] = {
 344       nir_const_value_for_int(x, 32),
 345       nir_const_value_for_int(y, 32),
 346       nir_const_value_for_int(z, 32),
 347       nir_const_value_for_int(w, 32),
 348    };
 349
 350    return nir_build_imm(build, 4, 32, v);
 351 }
 352
 353 static inline nir_ssa_def *
 354 nir_builder_alu_instr_finish_and_insert(nir_builder *build, nir_alu_instr *instr)
 355 {
 356    const nir_op_info *op_info = &nir_op_infos[instr->op];
 357
 358    instr->exact = build->exact;
 359
 360    /* Guess the number of components the destination temporary should have
 361     * based on our input sizes, if it's not fixed for the op.
 362     */
 363    unsigned num_components = op_info->output_size;
 364    if (num_components == 0) {
 365       for (unsigned i = 0; i < op_info->num_inputs; i++) {
 366          if (op_info->input_sizes[i] == 0)
 367             num_components = MAX2(num_components,
 368                                   instr->src[i].src.ssa->num_components);
 369       }
 370    }
 371    assert(num_components != 0);
 372
 373    /* Figure out the bitwidth based on the source bitwidth if the instruction
 374     * is variable-width.
 375     */
 376    unsigned bit_size = nir_alu_type_get_type_size(op_info->output_type);
 377    if (bit_size == 0) {
 378       for (unsigned i = 0; i < op_info->num_inputs; i++) {
 379          unsigned src_bit_size = instr->src[i].src.ssa->bit_size;
 380          if (nir_alu_type_get_type_size(op_info->input_types[i]) == 0) {
 381             if (bit_size)
 382                assert(src_bit_size == bit_size);
 383             else
 384                bit_size = src_bit_size;
 385          } else {
 386             assert(src_bit_size ==
 387                nir_alu_type_get_type_size(op_info->input_types[i]));
 388          }
 389       }
 390    }
 391
 392    /* When in doubt, assume 32. */
 393    if (bit_size == 0)
 394       bit_size = 32;
 395
 396    /* Make sure we don't swizzle from outside of our source vector (like if a
 397     * scalar value was passed into a multiply with a vector).
 398     */
 399    for (unsigned i = 0; i < op_info->num_inputs; i++) {
 400       for (unsigned j = instr->src[i].src.ssa->num_components;
 401            j < NIR_MAX_VEC_COMPONENTS; j++) {
 402          instr->src[i].swizzle[j] = instr->src[i].src.ssa->num_components - 1;
 403       }
 404    }
 405
 406    nir_ssa_dest_init(&instr->instr, &instr->dest.dest, num_components,
 407                      bit_size, NULL);
 408    instr->dest.write_mask = (1 << num_components) - 1;
 409
 410    nir_builder_instr_insert(build, &instr->instr);
 411
 412    return &instr->dest.dest.ssa;
 413 }
 414
 415 static inline nir_ssa_def *
 416 nir_build_alu(nir_builder *build, nir_op op, nir_ssa_def *src0,
 417               nir_ssa_def *src1, nir_ssa_def *src2, nir_ssa_def *src3)
 418 {
 419    nir_alu_instr *instr = nir_alu_instr_create(build->shader, op);
 420    if (!instr)
 421       return NULL;
 422
 423    instr->src[0].src = nir_src_for_ssa(src0);
 424    if (src1)
 425       instr->src[1].src = nir_src_for_ssa(src1);
 426    if (src2)
 427       instr->src[2].src = nir_src_for_ssa(src2);
 428    if (src3)
 429       instr->src[3].src = nir_src_for_ssa(src3);
 430
 431    return nir_builder_alu_instr_finish_and_insert(build, instr);
 432 }
 433
 434 /* for the couple special cases with more than 4 src args: */
 435 static inline nir_ssa_def *
 436 nir_build_alu_src_arr(nir_builder *build, nir_op op, nir_ssa_def **srcs)
 437 {
 438    const nir_op_info *op_info = &nir_op_infos[op];
 439    nir_alu_instr *instr = nir_alu_instr_create(build->shader, op);
 440    if (!instr)
 441       return NULL;
 442
 443    for (unsigned i = 0; i < op_info->num_inputs; i++)
 444       instr->src[i].src = nir_src_for_ssa(srcs[i]);
 445
 446    return nir_builder_alu_instr_finish_and_insert(build, instr);
 447 }
 448
 449 #include "nir_builder_opcodes.h"
 450
 451 static inline nir_ssa_def *
 452 nir_vec(nir_builder *build, nir_ssa_def **comp, unsigned num_components)
 453 {
 454    return nir_build_alu_src_arr(build, nir_op_vec(num_components), comp);
 455 }
 456
 457 static inline nir_ssa_def *
 458 nir_mov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
 459 {
 460    assert(!src.abs && !src.negate);
 461    if (src.src.is_ssa && src.src.ssa->num_components == num_components) {
 462       bool any_swizzles = false;
 463       for (unsigned i = 0; i < num_components; i++) {
 464          if (src.swizzle[i] != i)
 465             any_swizzles = true;
 466       }
 467       if (!any_swizzles)
 468          return src.src.ssa;
 469    }
 470
 471    nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_mov);
 472    nir_ssa_dest_init(&mov->instr, &mov->dest.dest, num_components,
 473                      nir_src_bit_size(src.src), NULL);
 474    mov->exact = build->exact;
 475    mov->dest.write_mask = (1 << num_components) - 1;
 476    mov->src[0] = src;
 477    nir_builder_instr_insert(build, &mov->instr);
 478
 479    return &mov->dest.dest.ssa;
 480 }
 481
 482 /**
 483  * Construct an fmov or imov that reswizzles the source's components.
 484  */
 485 static inline nir_ssa_def *
 486 nir_swizzle(nir_builder *build, nir_ssa_def *src, const unsigned *swiz,
 487             unsigned num_components)
 488 {
 489    assert(num_components <= NIR_MAX_VEC_COMPONENTS);
 490    nir_alu_src alu_src = { NIR_SRC_INIT };
 491    alu_src.src = nir_src_for_ssa(src);
 492
 493    bool is_identity_swizzle = true;
 494    for (unsigned i = 0; i < num_components && i < NIR_MAX_VEC_COMPONENTS; i++) {
 495       if (swiz[i] != i)
 496          is_identity_swizzle = false;
 497       alu_src.swizzle[i] = swiz[i];
 498    }
 499
 500    if (num_components == src->num_components && is_identity_swizzle)
 501       return src;
 502
 503    return nir_mov_alu(build, alu_src, num_components);
 504 }
 505
 506 /* Selects the right fdot given the number of components in each source. */
 507 static inline nir_ssa_def *
 508 nir_fdot(nir_builder *build, nir_ssa_def *src0, nir_ssa_def *src1)
 509 {
 510    assert(src0->num_components == src1->num_components);
 511    switch (src0->num_components) {
 512    case 1: return nir_fmul(build, src0, src1);
 513    case 2: return nir_fdot2(build, src0, src1);
 514    case 3: return nir_fdot3(build, src0, src1);
 515    case 4: return nir_fdot4(build, src0, src1);
 516    case 8: return nir_fdot8(build, src0, src1);
 517    case 16: return nir_fdot16(build, src0, src1);
 518    default:
 519       unreachable("bad component size");
 520    }
 521
 522    return NULL;
 523 }
 524
 525 static inline nir_ssa_def *
 526 nir_ball_iequal(nir_builder *b, nir_ssa_def *src0, nir_ssa_def *src1)
 527 {
 528    switch (src0->num_components) {
 529    case 1: return nir_ieq(b, src0, src1);
 530    case 2: return nir_ball_iequal2(b, src0, src1);
 531    case 3: return nir_ball_iequal3(b, src0, src1);
 532    case 4: return nir_ball_iequal4(b, src0, src1);
 533    case 8: return nir_ball_iequal8(b, src0, src1);
 534    case 16: return nir_ball_iequal16(b, src0, src1);
 535    default:
 536       unreachable("bad component size");
 537    }
 538 }
 539
 540 static inline nir_ssa_def *
 541 nir_ball(nir_builder *b, nir_ssa_def *src)
 542 {
 543    return nir_ball_iequal(b, src, nir_imm_true(b));
 544 }
 545
 546 static inline nir_ssa_def *
 547 nir_bany_inequal(nir_builder *b, nir_ssa_def *src0, nir_ssa_def *src1)
 548 {
 549    switch (src0->num_components) {
 550    case 1: return nir_ine(b, src0, src1);
 551    case 2: return nir_bany_inequal2(b, src0, src1);
 552    case 3: return nir_bany_inequal3(b, src0, src1);
 553    case 4: return nir_bany_inequal4(b, src0, src1);
 554    case 8: return nir_bany_inequal8(b, src0, src1);
 555    case 16: return nir_bany_inequal16(b, src0, src1);
 556    default:
 557       unreachable("bad component size");
 558    }
 559 }
 560
 561 static inline nir_ssa_def *
 562 nir_bany(nir_builder *b, nir_ssa_def *src)
 563 {
 564    return nir_bany_inequal(b, src, nir_imm_false(b));
 565 }
 566
 567 static inline nir_ssa_def *
 568 nir_channel(nir_builder *b, nir_ssa_def *def, unsigned c)
 569 {
 570    return nir_swizzle(b, def, &c, 1);
 571 }
 572
 573 static inline nir_ssa_def *
 574 nir_channels(nir_builder *b, nir_ssa_def *def, nir_component_mask_t mask)
 575 {
 576    unsigned num_channels = 0, swizzle[NIR_MAX_VEC_COMPONENTS] = { 0 };
 577
 578    for (unsigned i = 0; i < NIR_MAX_VEC_COMPONENTS; i++) {
 579       if ((mask & (1 << i)) == 0)
 580          continue;
 581       swizzle[num_channels++] = i;
 582    }
 583
 584    return nir_swizzle(b, def, swizzle, num_channels);
 585 }
 586
 587 static inline nir_ssa_def *
 588 _nir_vector_extract_helper(nir_builder *b, nir_ssa_def *vec, nir_ssa_def *c,
 589                            unsigned start, unsigned end)
 590 {
 591    if (start == end - 1) {
 592       return nir_channel(b, vec, start);
 593    } else {
 594       unsigned mid = start + (end - start) / 2;
 595       return nir_bcsel(b, nir_ilt(b, c, nir_imm_intN_t(b, mid, c->bit_size)),
 596                        _nir_vector_extract_helper(b, vec, c, start, mid),
 597                        _nir_vector_extract_helper(b, vec, c, mid, end));
 598    }
 599 }
 600
 601 static inline nir_ssa_def *
 602 nir_vector_extract(nir_builder *b, nir_ssa_def *vec, nir_ssa_def *c)
 603 {
 604    nir_src c_src = nir_src_for_ssa(c);
 605    if (nir_src_is_const(c_src)) {
 606       uint64_t c_const = nir_src_as_uint(c_src);
 607       if (c_const < vec->num_components)
 608          return nir_channel(b, vec, c_const);
 609       else
 610          return nir_ssa_undef(b, 1, vec->bit_size);
 611    } else {
 612       return _nir_vector_extract_helper(b, vec, c, 0, vec->num_components);
 613    }
 614 }
 615
 616 /** Replaces the component of `vec` specified by `c` with `scalar` */
 617 static inline nir_ssa_def *
 618 nir_vector_insert_imm(nir_builder *b, nir_ssa_def *vec,
 619                       nir_ssa_def *scalar, unsigned c)
 620 {
 621    assert(scalar->num_components == 1);
 622    assert(c < vec->num_components);
 623
 624    nir_op vec_op = nir_op_vec(vec->num_components);
 625    nir_alu_instr *vec_instr = nir_alu_instr_create(b->shader, vec_op);
 626
 627    for (unsigned i = 0; i < vec->num_components; i++) {
 628       if (i == c) {
 629          vec_instr->src[i].src = nir_src_for_ssa(scalar);
 630          vec_instr->src[i].swizzle[0] = 0;
 631       } else {
 632          vec_instr->src[i].src = nir_src_for_ssa(vec);
 633          vec_instr->src[i].swizzle[0] = i;
 634       }
 635    }
 636
 637    return nir_builder_alu_instr_finish_and_insert(b, vec_instr);
 638 }
 639
 640 /** Replaces the component of `vec` specified by `c` with `scalar` */
 641 static inline nir_ssa_def *
 642 nir_vector_insert(nir_builder *b, nir_ssa_def *vec, nir_ssa_def *scalar,
 643                   nir_ssa_def *c)
 644 {
 645    assert(scalar->num_components == 1);
 646    assert(c->num_components == 1);
 647
 648    nir_src c_src = nir_src_for_ssa(c);
 649    if (nir_src_is_const(c_src)) {
 650       uint64_t c_const = nir_src_as_uint(c_src);
 651       if (c_const < vec->num_components)
 652          return nir_vector_insert_imm(b, vec, scalar, c_const);
 653       else
 654          return vec;
 655    } else {
 656       nir_const_value per_comp_idx_const[NIR_MAX_VEC_COMPONENTS];
 657       for (unsigned i = 0; i < NIR_MAX_VEC_COMPONENTS; i++)
 658          per_comp_idx_const[i] = nir_const_value_for_int(i, c->bit_size);
 659       nir_ssa_def *per_comp_idx =
 660          nir_build_imm(b, vec->num_components,
 661                        c->bit_size, per_comp_idx_const);
 662
 663       /* nir_builder will automatically splat out scalars to vectors so an
 664        * insert is as simple as "if I'm the channel, replace me with the
 665        * scalar."
 666        */
 667       return nir_bcsel(b, nir_ieq(b, c, per_comp_idx), scalar, vec);
 668    }
 669 }
 670
 671 static inline nir_ssa_def *
 672 nir_i2i(nir_builder *build, nir_ssa_def *x, unsigned dest_bit_size)
 673 {
 674    if (x->bit_size == dest_bit_size)
 675       return x;
 676
 677    switch (dest_bit_size) {
 678    case 64: return nir_i2i64(build, x);
 679    case 32: return nir_i2i32(build, x);
 680    case 16: return nir_i2i16(build, x);
 681    case 8:  return nir_i2i8(build, x);
 682    default: unreachable("Invalid bit size");
 683    }
 684 }
 685
 686 static inline nir_ssa_def *
 687 nir_u2u(nir_builder *build, nir_ssa_def *x, unsigned dest_bit_size)
 688 {
 689    if (x->bit_size == dest_bit_size)
 690       return x;
 691
 692    switch (dest_bit_size) {
 693    case 64: return nir_u2u64(build, x);
 694    case 32: return nir_u2u32(build, x);
 695    case 16: return nir_u2u16(build, x);
 696    case 8:  return nir_u2u8(build, x);
 697    default: unreachable("Invalid bit size");
 698    }
 699 }
 700
 701 static inline nir_ssa_def *
 702 nir_iadd_imm(nir_builder *build, nir_ssa_def *x, uint64_t y)
 703 {
 704    assert(x->bit_size <= 64);
 705    if (x->bit_size < 64)
 706       y &= (1ull << x->bit_size) - 1;
 707
 708    if (y == 0) {
 709       return x;
 710    } else {
 711       return nir_iadd(build, x, nir_imm_intN_t(build, y, x->bit_size));
 712    }
 713 }
 714
 715 static inline nir_ssa_def *
 716 _nir_mul_imm(nir_builder *build, nir_ssa_def *x, uint64_t y, bool amul)
 717 {
 718    assert(x->bit_size <= 64);
 719    if (x->bit_size < 64)
 720       y &= (1ull << x->bit_size) - 1;
 721
 722    if (y == 0) {
 723       return nir_imm_intN_t(build, 0, x->bit_size);
 724    } else if (y == 1) {
 725       return x;
 726    } else if (!build->shader->options->lower_bitops &&
 727               util_is_power_of_two_or_zero64(y)) {
 728       return nir_ishl(build, x, nir_imm_int(build, ffsll(y) - 1));
 729    } else if (amul) {
 730       return nir_amul(build, x, nir_imm_intN_t(build, y, x->bit_size));
 731    } else {
 732       return nir_imul(build, x, nir_imm_intN_t(build, y, x->bit_size));
 733    }
 734 }
 735
 736 static inline nir_ssa_def *
 737 nir_imul_imm(nir_builder *build, nir_ssa_def *x, uint64_t y)
 738 {
 739    return _nir_mul_imm(build, x, y, false);
 740 }
 741
 742 static inline nir_ssa_def *
 743 nir_amul_imm(nir_builder *build, nir_ssa_def *x, uint64_t y)
 744 {
 745    return _nir_mul_imm(build, x, y, true);
 746 }
 747
 748 static inline nir_ssa_def *
 749 nir_fadd_imm(nir_builder *build, nir_ssa_def *x, double y)
 750 {
 751    return nir_fadd(build, x, nir_imm_floatN_t(build, y, x->bit_size));
 752 }
 753
 754 static inline nir_ssa_def *
 755 nir_fmul_imm(nir_builder *build, nir_ssa_def *x, double y)
 756 {
 757    return nir_fmul(build, x, nir_imm_floatN_t(build, y, x->bit_size));
 758 }
 759
 760 static inline nir_ssa_def *
 761 nir_pack_bits(nir_builder *b, nir_ssa_def *src, unsigned dest_bit_size)
 762 {
 763    assert(src->num_components * src->bit_size == dest_bit_size);
 764
 765    switch (dest_bit_size) {
 766    case 64:
 767       switch (src->bit_size) {
 768       case 32: return nir_pack_64_2x32(b, src);
 769       case 16: return nir_pack_64_4x16(b, src);
 770       default: break;
 771       }
 772       break;
 773
 774    case 32:
 775       if (src->bit_size == 16)
 776          return nir_pack_32_2x16(b, src);
 777       break;
 778
 779    default:
 780       break;
 781    }
 782
 783    /* If we got here, we have no dedicated unpack opcode. */
 784    nir_ssa_def *dest = nir_imm_intN_t(b, 0, dest_bit_size);
 785    for (unsigned i = 0; i < src->num_components; i++) {
 786       nir_ssa_def *val = nir_u2u(b, nir_channel(b, src, i), dest_bit_size);
 787       val = nir_ishl(b, val, nir_imm_int(b, i * src->bit_size));
 788       dest = nir_ior(b, dest, val);
 789    }
 790    return dest;
 791 }
 792
 793 static inline nir_ssa_def *
 794 nir_unpack_bits(nir_builder *b, nir_ssa_def *src, unsigned dest_bit_size)
 795 {
 796    assert(src->num_components == 1);
 797    assert(src->bit_size > dest_bit_size);
 798    const unsigned dest_num_components = src->bit_size / dest_bit_size;
 799    assert(dest_num_components <= NIR_MAX_VEC_COMPONENTS);
 800
 801    switch (src->bit_size) {
 802    case 64:
 803       switch (dest_bit_size) {
 804       case 32: return nir_unpack_64_2x32(b, src);
 805       case 16: return nir_unpack_64_4x16(b, src);
 806       default: break;
 807       }
 808       break;
 809
 810    case 32:
 811       if (dest_bit_size == 16)
 812          return nir_unpack_32_2x16(b, src);
 813       break;
 814
 815    default:
 816       break;
 817    }
 818
 819    /* If we got here, we have no dedicated unpack opcode. */
 820    nir_ssa_def *dest_comps[NIR_MAX_VEC_COMPONENTS];
 821    for (unsigned i = 0; i < dest_num_components; i++) {
 822       nir_ssa_def *val = nir_ushr(b, src, nir_imm_int(b, i * dest_bit_size));
 823       dest_comps[i] = nir_u2u(b, val, dest_bit_size);
 824    }
 825    return nir_vec(b, dest_comps, dest_num_components);
 826 }
 827
 828 /**
 829  * Treats srcs as if it's one big blob of bits and extracts the range of bits
 830  * given by
 831  *
 832  *       [first_bit, first_bit + dest_num_components * dest_bit_size)
 833  *
 834  * The range can have any alignment or size as long as it's an integer number
 835  * of destination components and fits inside the concatenated sources.
 836  *
 837  * TODO: The one caveat here is that we can't handle byte alignment if 64-bit
 838  * values are involved because that would require pack/unpack to/from a vec8
 839  * which NIR currently does not support.
 840  */
 841 static inline nir_ssa_def *
 842 nir_extract_bits(nir_builder *b, nir_ssa_def **srcs, unsigned num_srcs,
 843                  unsigned first_bit,
 844                  unsigned dest_num_components, unsigned dest_bit_size)
 845 {
 846    const unsigned num_bits = dest_num_components * dest_bit_size;
 847
 848    /* Figure out the common bit size */
 849    unsigned common_bit_size = dest_bit_size;
 850    for (unsigned i = 0; i < num_srcs; i++)
 851       common_bit_size = MIN2(common_bit_size, srcs[i]->bit_size);
 852    if (first_bit > 0)
 853       common_bit_size = MIN2(common_bit_size, (1u << (ffs(first_bit) - 1)));
 854
 855    /* We don't want to have to deal with 1-bit values */
 856    assert(common_bit_size >= 8);
 857
 858    nir_ssa_def *common_comps[NIR_MAX_VEC_COMPONENTS * sizeof(uint64_t)];
 859    assert(num_bits / common_bit_size <= ARRAY_SIZE(common_comps));
 860
 861    /* First, unpack to the common bit size and select the components from the
 862     * source.
 863     */
 864    int src_idx = -1;
 865    unsigned src_start_bit = 0;
 866    unsigned src_end_bit = 0;
 867    for (unsigned i = 0; i < num_bits / common_bit_size; i++) {
 868       const unsigned bit = first_bit + (i * common_bit_size);
 869       while (bit >= src_end_bit) {
 870          src_idx++;
 871          assert(src_idx < (int) num_srcs);
 872          src_start_bit = src_end_bit;
 873          src_end_bit += srcs[src_idx]->bit_size *
 874                         srcs[src_idx]->num_components;
 875       }
 876       assert(bit >= src_start_bit);
 877       assert(bit + common_bit_size <= src_end_bit);
 878       const unsigned rel_bit = bit - src_start_bit;
 879       const unsigned src_bit_size = srcs[src_idx]->bit_size;
 880
 881       nir_ssa_def *comp = nir_channel(b, srcs[src_idx],
 882                                       rel_bit / src_bit_size);
 883       if (srcs[src_idx]->bit_size > common_bit_size) {
 884          nir_ssa_def *unpacked = nir_unpack_bits(b, comp, common_bit_size);
 885          comp = nir_channel(b, unpacked, (rel_bit % src_bit_size) /
 886                                          common_bit_size);
 887       }
 888       common_comps[i] = comp;
 889    }
 890
 891    /* Now, re-pack the destination if we have to */
 892    if (dest_bit_size > common_bit_size) {
 893       unsigned common_per_dest = dest_bit_size / common_bit_size;
 894       nir_ssa_def *dest_comps[NIR_MAX_VEC_COMPONENTS];
 895       for (unsigned i = 0; i < dest_num_components; i++) {
 896          nir_ssa_def *unpacked = nir_vec(b, common_comps + i * common_per_dest,
 897                                          common_per_dest);
 898          dest_comps[i] = nir_pack_bits(b, unpacked, dest_bit_size);
 899       }
 900       return nir_vec(b, dest_comps, dest_num_components);
 901    } else {
 902       assert(dest_bit_size == common_bit_size);
 903       return nir_vec(b, common_comps, dest_num_components);
 904    }
 905 }
 906
 907 static inline nir_ssa_def *
 908 nir_bitcast_vector(nir_builder *b, nir_ssa_def *src, unsigned dest_bit_size)
 909 {
 910    assert((src->bit_size * src->num_components) % dest_bit_size == 0);
 911    const unsigned dest_num_components =
 912       (src->bit_size * src->num_components) / dest_bit_size;
 913    assert(dest_num_components <= NIR_MAX_VEC_COMPONENTS);
 914
 915    return nir_extract_bits(b, &src, 1, 0, dest_num_components, dest_bit_size);
 916 }
 917
 918 /**
 919  * Turns a nir_src into a nir_ssa_def * so it can be passed to
 920  * nir_build_alu()-based builder calls.
 921  *
 922  * See nir_ssa_for_alu_src() for alu instructions.
 923  */
 924 static inline nir_ssa_def *
 925 nir_ssa_for_src(nir_builder *build, nir_src src, int num_components)
 926 {
 927    if (src.is_ssa && src.ssa->num_components == num_components)
 928       return src.ssa;
 929
 930    nir_alu_src alu = { NIR_SRC_INIT };
 931    alu.src = src;
 932    for (int j = 0; j < 4; j++)
 933       alu.swizzle[j] = j;
 934
 935    return nir_mov_alu(build, alu, num_components);
 936 }
 937
 938 /**
 939  * Similar to nir_ssa_for_src(), but for alu srcs, respecting the
 940  * nir_alu_src's swizzle.
 941  */
 942 static inline nir_ssa_def *
 943 nir_ssa_for_alu_src(nir_builder *build, nir_alu_instr *instr, unsigned srcn)
 944 {
 945    static uint8_t trivial_swizzle[] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 };
 946    STATIC_ASSERT(ARRAY_SIZE(trivial_swizzle) == NIR_MAX_VEC_COMPONENTS);
 947
 948    nir_alu_src *src = &instr->src[srcn];
 949    unsigned num_components = nir_ssa_alu_instr_src_components(instr, srcn);
 950
 951    if (src->src.is_ssa && (src->src.ssa->num_components == num_components) &&
 952        !src->abs && !src->negate &&
 953        (memcmp(src->swizzle, trivial_swizzle, num_components) == 0))
 954       return src->src.ssa;
 955
 956    return nir_mov_alu(build, *src, num_components);
 957 }
 958
 959 static inline unsigned
 960 nir_get_ptr_bitsize(nir_builder *build)
 961 {
 962    if (build->shader->info.stage == MESA_SHADER_KERNEL)
 963       return build->shader->info.cs.ptr_size;
 964    return 32;
 965 }
 966
 967 static inline nir_deref_instr *
 968 nir_build_deref_var(nir_builder *build, nir_variable *var)
 969 {
 970    nir_deref_instr *deref =
 971       nir_deref_instr_create(build->shader, nir_deref_type_var);
 972
 973    deref->mode = var->data.mode;
 974    deref->type = var->type;
 975    deref->var = var;
 976
 977    nir_ssa_dest_init(&deref->instr, &deref->dest, 1,
 978                      nir_get_ptr_bitsize(build), NULL);
 979
 980    nir_builder_instr_insert(build, &deref->instr);
 981
 982    return deref;
 983 }
 984
 985 static inline nir_deref_instr *
 986 nir_build_deref_array(nir_builder *build, nir_deref_instr *parent,
 987                       nir_ssa_def *index)
 988 {
 989    assert(glsl_type_is_array(parent->type) ||
 990           glsl_type_is_matrix(parent->type) ||
 991           glsl_type_is_vector(parent->type));
 992
 993    assert(index->bit_size == parent->dest.ssa.bit_size);
 994
 995    nir_deref_instr *deref =
 996       nir_deref_instr_create(build->shader, nir_deref_type_array);
 997
 998    deref->mode = parent->mode;
 999    deref->type = glsl_get_array_element(parent->type);
1000    deref->parent = nir_src_for_ssa(&parent->dest.ssa);
1001    deref->arr.index = nir_src_for_ssa(index);
1002
1003    nir_ssa_dest_init(&deref->instr, &deref->dest,
1004                      parent->dest.ssa.num_components,
1005                      parent->dest.ssa.bit_size, NULL);
1006
1007    nir_builder_instr_insert(build, &deref->instr);
1008
1009    return deref;
1010 }
1011
1012 static inline nir_deref_instr *
1013 nir_build_deref_array_imm(nir_builder *build, nir_deref_instr *parent,
1014                           int64_t index)
1015 {
1016    assert(parent->dest.is_ssa);
1017    nir_ssa_def *idx_ssa = nir_imm_intN_t(build, index,
1018                                          parent->dest.ssa.bit_size);
1019
1020    return nir_build_deref_array(build, parent, idx_ssa);
1021 }
1022
1023 static inline nir_deref_instr *
1024 nir_build_deref_ptr_as_array(nir_builder *build, nir_deref_instr *parent,
1025                              nir_ssa_def *index)
1026 {
1027    assert(parent->deref_type == nir_deref_type_array ||
1028           parent->deref_type == nir_deref_type_ptr_as_array ||
1029           parent->deref_type == nir_deref_type_cast);
1030
1031    assert(index->bit_size == parent->dest.ssa.bit_size);
1032
1033    nir_deref_instr *deref =
1034       nir_deref_instr_create(build->shader, nir_deref_type_ptr_as_array);
1035
1036    deref->mode = parent->mode;
1037    deref->type = parent->type;
1038    deref->parent = nir_src_for_ssa(&parent->dest.ssa);
1039    deref->arr.index = nir_src_for_ssa(index);
1040
1041    nir_ssa_dest_init(&deref->instr, &deref->dest,
1042                      parent->dest.ssa.num_components,
1043                      parent->dest.ssa.bit_size, NULL);
1044
1045    nir_builder_instr_insert(build, &deref->instr);
1046
1047    return deref;
1048 }
1049
1050 static inline nir_deref_instr *
1051 nir_build_deref_array_wildcard(nir_builder *build, nir_deref_instr *parent)
1052 {
1053    assert(glsl_type_is_array(parent->type) ||
1054           glsl_type_is_matrix(parent->type));
1055
1056    nir_deref_instr *deref =
1057       nir_deref_instr_create(build->shader, nir_deref_type_array_wildcard);
1058
1059    deref->mode = parent->mode;
1060    deref->type = glsl_get_array_element(parent->type);
1061    deref->parent = nir_src_for_ssa(&parent->dest.ssa);
1062
1063    nir_ssa_dest_init(&deref->instr, &deref->dest,
1064                      parent->dest.ssa.num_components,
1065                      parent->dest.ssa.bit_size, NULL);
1066
1067    nir_builder_instr_insert(build, &deref->instr);
1068
1069    return deref;
1070 }
1071
1072 static inline nir_deref_instr *
1073 nir_build_deref_struct(nir_builder *build, nir_deref_instr *parent,
1074                        unsigned index)
1075 {
1076    assert(glsl_type_is_struct_or_ifc(parent->type));
1077
1078    nir_deref_instr *deref =
1079       nir_deref_instr_create(build->shader, nir_deref_type_struct);
1080
1081    deref->mode = parent->mode;
1082    deref->type = glsl_get_struct_field(parent->type, index);
1083    deref->parent = nir_src_for_ssa(&parent->dest.ssa);
1084    deref->strct.index = index;
1085
1086    nir_ssa_dest_init(&deref->instr, &deref->dest,
1087                      parent->dest.ssa.num_components,
1088                      parent->dest.ssa.bit_size, NULL);
1089
1090    nir_builder_instr_insert(build, &deref->instr);
1091
1092    return deref;
1093 }
1094
1095 static inline nir_deref_instr *
1096 nir_build_deref_cast(nir_builder *build, nir_ssa_def *parent,
1097                      nir_variable_mode mode, const struct glsl_type *type,
1098                      unsigned ptr_stride)
1099 {
1100    nir_deref_instr *deref =
1101       nir_deref_instr_create(build->shader, nir_deref_type_cast);
1102
1103    deref->mode = mode;
1104    deref->type = type;
1105    deref->parent = nir_src_for_ssa(parent);
1106    deref->cast.ptr_stride = ptr_stride;
1107
1108    nir_ssa_dest_init(&deref->instr, &deref->dest,
1109                      parent->num_components, parent->bit_size, NULL);
1110
1111    nir_builder_instr_insert(build, &deref->instr);
1112
1113    return deref;
1114 }
1115
1116 /** Returns a deref that follows another but starting from the given parent
1117  *
1118  * The new deref will be the same type and take the same array or struct index
1119  * as the leader deref but it may have a different parent.  This is very
1120  * useful for walking deref paths.
1121  */
1122 static inline nir_deref_instr *
1123 nir_build_deref_follower(nir_builder *b, nir_deref_instr *parent,
1124                          nir_deref_instr *leader)
1125 {
1126    /* If the derefs would have the same parent, don't make a new one */
1127    assert(leader->parent.is_ssa);
1128    if (leader->parent.ssa == &parent->dest.ssa)
1129       return leader;
1130
1131    UNUSED nir_deref_instr *leader_parent = nir_src_as_deref(leader->parent);
1132
1133    switch (leader->deref_type) {
1134    case nir_deref_type_var:
1135       unreachable("A var dereference cannot have a parent");
1136       break;
1137
1138    case nir_deref_type_array:
1139    case nir_deref_type_array_wildcard:
1140       assert(glsl_type_is_matrix(parent->type) ||
1141              glsl_type_is_array(parent->type) ||
1142              (leader->deref_type == nir_deref_type_array &&
1143               glsl_type_is_vector(parent->type)));
1144       assert(glsl_get_length(parent->type) ==
1145              glsl_get_length(leader_parent->type));
1146
1147       if (leader->deref_type == nir_deref_type_array) {
1148          assert(leader->arr.index.is_ssa);
1149          nir_ssa_def *index = nir_i2i(b, leader->arr.index.ssa,
1150                                          parent->dest.ssa.bit_size);
1151          return nir_build_deref_array(b, parent, index);
1152       } else {
1153          return nir_build_deref_array_wildcard(b, parent);
1154       }
1155
1156    case nir_deref_type_struct:
1157       assert(glsl_type_is_struct_or_ifc(parent->type));
1158       assert(glsl_get_length(parent->type) ==
1159              glsl_get_length(leader_parent->type));
1160
1161       return nir_build_deref_struct(b, parent, leader->strct.index);
1162
1163    default:
1164       unreachable("Invalid deref instruction type");
1165    }
1166 }
1167
1168 static inline nir_ssa_def *
1169 nir_load_reg(nir_builder *build, nir_register *reg)
1170 {
1171    return nir_ssa_for_src(build, nir_src_for_reg(reg), reg->num_components);
1172 }
1173
1174 static inline void
1175 nir_store_reg(nir_builder *build, nir_register *reg,
1176               nir_ssa_def *def, nir_component_mask_t write_mask)
1177 {
1178    assert(reg->num_components == def->num_components);
1179    assert(reg->bit_size == def->bit_size);
1180
1181    nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_mov);
1182    mov->src[0].src = nir_src_for_ssa(def);
1183    mov->dest.dest = nir_dest_for_reg(reg);
1184    mov->dest.write_mask = write_mask & BITFIELD_MASK(reg->num_components);
1185    nir_builder_instr_insert(build, &mov->instr);
1186 }
1187
1188 static inline nir_ssa_def *
1189 nir_load_deref_with_access(nir_builder *build, nir_deref_instr *deref,
1190                            enum gl_access_qualifier access)
1191 {
1192    nir_intrinsic_instr *load =
1193       nir_intrinsic_instr_create(build->shader, nir_intrinsic_load_deref);
1194    load->num_components = glsl_get_vector_elements(deref->type);
1195    load->src[0] = nir_src_for_ssa(&deref->dest.ssa);
1196    nir_ssa_dest_init(&load->instr, &load->dest, load->num_components,
1197                      glsl_get_bit_size(deref->type), NULL);
1198    nir_intrinsic_set_access(load, access);
1199    nir_builder_instr_insert(build, &load->instr);
1200    return &load->dest.ssa;
1201 }
1202
1203 static inline nir_ssa_def *
1204 nir_load_deref(nir_builder *build, nir_deref_instr *deref)
1205 {
1206    return nir_load_deref_with_access(build, deref, (enum gl_access_qualifier)0);
1207 }
1208
1209 static inline void
1210 nir_store_deref_with_access(nir_builder *build, nir_deref_instr *deref,
1211                             nir_ssa_def *value, unsigned writemask,
1212                             enum gl_access_qualifier access)
1213 {
1214    nir_intrinsic_instr *store =
1215       nir_intrinsic_instr_create(build->shader, nir_intrinsic_store_deref);
1216    store->num_components = glsl_get_vector_elements(deref->type);
1217    store->src[0] = nir_src_for_ssa(&deref->dest.ssa);
1218    store->src[1] = nir_src_for_ssa(value);
1219    nir_intrinsic_set_write_mask(store,
1220                                 writemask & ((1 << store->num_components) - 1));
1221    nir_intrinsic_set_access(store, access);
1222    nir_builder_instr_insert(build, &store->instr);
1223 }
1224
1225 static inline void
1226 nir_store_deref(nir_builder *build, nir_deref_instr *deref,
1227                 nir_ssa_def *value, unsigned writemask)
1228 {
1229    nir_store_deref_with_access(build, deref, value, writemask,
1230                                (enum gl_access_qualifier)0);
1231 }
1232
1233 static inline void
1234 nir_copy_deref_with_access(nir_builder *build, nir_deref_instr *dest,
1235                            nir_deref_instr *src,
1236                            enum gl_access_qualifier dest_access,
1237                            enum gl_access_qualifier src_access)
1238 {
1239    nir_intrinsic_instr *copy =
1240       nir_intrinsic_instr_create(build->shader, nir_intrinsic_copy_deref);
1241    copy->src[0] = nir_src_for_ssa(&dest->dest.ssa);
1242    copy->src[1] = nir_src_for_ssa(&src->dest.ssa);
1243    nir_intrinsic_set_dst_access(copy, dest_access);
1244    nir_intrinsic_set_src_access(copy, src_access);
1245    nir_builder_instr_insert(build, &copy->instr);
1246 }
1247
1248 static inline void
1249 nir_copy_deref(nir_builder *build, nir_deref_instr *dest, nir_deref_instr *src)
1250 {
1251    nir_copy_deref_with_access(build, dest, src,
1252                               (enum gl_access_qualifier) 0,
1253                               (enum gl_access_qualifier) 0);
1254 }
1255
1256 static inline nir_ssa_def *
1257 nir_load_var(nir_builder *build, nir_variable *var)
1258 {
1259    return nir_load_deref(build, nir_build_deref_var(build, var));
1260 }
1261
1262 static inline void
1263 nir_store_var(nir_builder *build, nir_variable *var, nir_ssa_def *value,
1264               unsigned writemask)
1265 {
1266    nir_store_deref(build, nir_build_deref_var(build, var), value, writemask);
1267 }
1268
1269 static inline void
1270 nir_copy_var(nir_builder *build, nir_variable *dest, nir_variable *src)
1271 {
1272    nir_copy_deref(build, nir_build_deref_var(build, dest),
1273                          nir_build_deref_var(build, src));
1274 }
1275
1276 static inline nir_ssa_def *
1277 nir_load_param(nir_builder *build, uint32_t param_idx)
1278 {
1279    assert(param_idx < build->impl->function->num_params);
1280    nir_parameter *param = &build->impl->function->params[param_idx];
1281
1282    nir_intrinsic_instr *load =
1283       nir_intrinsic_instr_create(build->shader, nir_intrinsic_load_param);
1284    nir_intrinsic_set_param_idx(load, param_idx);
1285    load->num_components = param->num_components;
1286    nir_ssa_dest_init(&load->instr, &load->dest,
1287                      param->num_components, param->bit_size, NULL);
1288    nir_builder_instr_insert(build, &load->instr);
1289    return &load->dest.ssa;
1290 }
1291
1292 #include "nir_builder_opcodes.h"
1293
1294 static inline nir_ssa_def *
1295 nir_f2b(nir_builder *build, nir_ssa_def *f)
1296 {
1297    return nir_f2b1(build, f);
1298 }
1299
1300 static inline nir_ssa_def *
1301 nir_i2b(nir_builder *build, nir_ssa_def *i)
1302 {
1303    return nir_i2b1(build, i);
1304 }
1305
1306 static inline nir_ssa_def *
1307 nir_b2f(nir_builder *build, nir_ssa_def *b, uint32_t bit_size)
1308 {
1309    switch (bit_size) {
1310    case 64: return nir_b2f64(build, b);
1311    case 32: return nir_b2f32(build, b);
1312    case 16: return nir_b2f16(build, b);
1313    default:
1314       unreachable("Invalid bit-size");
1315    };
1316 }
1317
1318 static inline nir_ssa_def *
1319 nir_b2i(nir_builder *build, nir_ssa_def *b, uint32_t bit_size)
1320 {
1321    switch (bit_size) {
1322    case 64: return nir_b2i64(build, b);
1323    case 32: return nir_b2i32(build, b);
1324    case 16: return nir_b2i16(build, b);
1325    case 8:  return nir_b2i8(build, b);
1326    default:
1327       unreachable("Invalid bit-size");
1328    };
1329 }
1330 static inline nir_ssa_def *
1331 nir_load_barycentric(nir_builder *build, nir_intrinsic_op op,
1332                      unsigned interp_mode)
1333 {
1334    unsigned num_components = op == nir_intrinsic_load_barycentric_model ? 3 : 2;
1335    nir_intrinsic_instr *bary = nir_intrinsic_instr_create(build->shader, op);
1336    nir_ssa_dest_init(&bary->instr, &bary->dest, num_components, 32, NULL);
1337    nir_intrinsic_set_interp_mode(bary, interp_mode);
1338    nir_builder_instr_insert(build, &bary->instr);
1339    return &bary->dest.ssa;
1340 }
1341
1342 static inline void
1343 nir_jump(nir_builder *build, nir_jump_type jump_type)
1344 {
1345    nir_jump_instr *jump = nir_jump_instr_create(build->shader, jump_type);
1346    nir_builder_instr_insert(build, &jump->instr);
1347 }
1348
1349 static inline nir_ssa_def *
1350 nir_compare_func(nir_builder *b, enum compare_func func,
1351                  nir_ssa_def *src0, nir_ssa_def *src1)
1352 {
1353    switch (func) {
1354    case COMPARE_FUNC_NEVER:
1355       return nir_imm_int(b, 0);
1356    case COMPARE_FUNC_ALWAYS:
1357       return nir_imm_int(b, ~0);
1358    case COMPARE_FUNC_EQUAL:
1359       return nir_feq(b, src0, src1);
1360    case COMPARE_FUNC_NOTEQUAL:
1361       return nir_fne(b, src0, src1);
1362    case COMPARE_FUNC_GREATER:
1363       return nir_flt(b, src1, src0);
1364    case COMPARE_FUNC_GEQUAL:
1365       return nir_fge(b, src0, src1);
1366    case COMPARE_FUNC_LESS:
1367       return nir_flt(b, src0, src1);
1368    case COMPARE_FUNC_LEQUAL:
1369       return nir_fge(b, src1, src0);
1370    }
1371    unreachable("bad compare func");
1372 }
1373
1374 static inline void
1375 nir_scoped_barrier(nir_builder *b,
1376                    nir_scope exec_scope,
1377                    nir_scope mem_scope,
1378                    nir_memory_semantics mem_semantics,
1379                    nir_variable_mode mem_modes)
1380 {
1381    nir_intrinsic_instr *intrin =
1382       nir_intrinsic_instr_create(b->shader, nir_intrinsic_scoped_barrier);
1383    nir_intrinsic_set_execution_scope(intrin, exec_scope);
1384    nir_intrinsic_set_memory_scope(intrin, mem_scope);
1385    nir_intrinsic_set_memory_semantics(intrin, mem_semantics);
1386    nir_intrinsic_set_memory_modes(intrin, mem_modes);
1387    nir_builder_instr_insert(b, &intrin->instr);
1388 }
1389
1390 static inline void
1391 nir_scoped_memory_barrier(nir_builder *b,
1392                           nir_scope scope,
1393                           nir_memory_semantics semantics,
1394                           nir_variable_mode modes)
1395 {
1396    nir_scoped_barrier(b, NIR_SCOPE_NONE, scope, semantics, modes);
1397 }
1398
1399 static inline nir_ssa_def *
1400 nir_convert_to_bit_size(nir_builder *b,
1401                     nir_ssa_def *src,
1402                     nir_alu_type type,
1403                     unsigned bit_size)
1404 {
1405    nir_alu_type base_type = nir_alu_type_get_base_type(type);
1406    nir_alu_type dst_type = (nir_alu_type)(bit_size | base_type);
1407
1408    nir_op opcode =
1409       nir_type_conversion_op(type, dst_type, nir_rounding_mode_undef);
1410
1411    return nir_build_alu(b, opcode, src, NULL, NULL, NULL);
1412 }
1413
1414 static inline nir_ssa_def *
1415 nir_i2iN(nir_builder *b, nir_ssa_def *src, unsigned bit_size)
1416 {
1417    return nir_convert_to_bit_size(b, src, nir_type_int, bit_size);
1418 }
1419
1420 static inline nir_ssa_def *
1421 nir_u2uN(nir_builder *b, nir_ssa_def *src, unsigned bit_size)
1422 {
1423    return nir_convert_to_bit_size(b, src, nir_type_uint, bit_size);
1424 }
1425
1426 static inline nir_ssa_def *
1427 nir_b2bN(nir_builder *b, nir_ssa_def *src, unsigned bit_size)
1428 {
1429    return nir_convert_to_bit_size(b, src, nir_type_bool, bit_size);
1430 }
1431
1432 static inline nir_ssa_def *
1433 nir_f2fN(nir_builder *b, nir_ssa_def *src, unsigned bit_size)
1434 {
1435    return nir_convert_to_bit_size(b, src, nir_type_float, bit_size);
1436 }
1437
1438 #endif /* NIR_BUILDER_H */