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    default:
 517       unreachable("bad component size");
 518    }
 519
 520    return NULL;
 521 }
 522
 523 static inline nir_ssa_def *
 524 nir_ball_iequal(nir_builder *b, nir_ssa_def *src0, nir_ssa_def *src1)
 525 {
 526    switch (src0->num_components) {
 527    case 1: return nir_ieq(b, src0, src1);
 528    case 2: return nir_ball_iequal2(b, src0, src1);
 529    case 3: return nir_ball_iequal3(b, src0, src1);
 530    case 4: return nir_ball_iequal4(b, src0, src1);
 531    default:
 532       unreachable("bad component size");
 533    }
 534 }
 535
 536 static inline nir_ssa_def *
 537 nir_bany_inequal(nir_builder *b, nir_ssa_def *src0, nir_ssa_def *src1)
 538 {
 539    switch (src0->num_components) {
 540    case 1: return nir_ine(b, src0, src1);
 541    case 2: return nir_bany_inequal2(b, src0, src1);
 542    case 3: return nir_bany_inequal3(b, src0, src1);
 543    case 4: return nir_bany_inequal4(b, src0, src1);
 544    default:
 545       unreachable("bad component size");
 546    }
 547 }
 548
 549 static inline nir_ssa_def *
 550 nir_bany(nir_builder *b, nir_ssa_def *src)
 551 {
 552    return nir_bany_inequal(b, src, nir_imm_false(b));
 553 }
 554
 555 static inline nir_ssa_def *
 556 nir_channel(nir_builder *b, nir_ssa_def *def, unsigned c)
 557 {
 558    return nir_swizzle(b, def, &c, 1);
 559 }
 560
 561 static inline nir_ssa_def *
 562 nir_channels(nir_builder *b, nir_ssa_def *def, nir_component_mask_t mask)
 563 {
 564    unsigned num_channels = 0, swizzle[NIR_MAX_VEC_COMPONENTS] = { 0 };
 565
 566    for (unsigned i = 0; i < NIR_MAX_VEC_COMPONENTS; i++) {
 567       if ((mask & (1 << i)) == 0)
 568          continue;
 569       swizzle[num_channels++] = i;
 570    }
 571
 572    return nir_swizzle(b, def, swizzle, num_channels);
 573 }
 574
 575 static inline nir_ssa_def *
 576 _nir_vector_extract_helper(nir_builder *b, nir_ssa_def *vec, nir_ssa_def *c,
 577                            unsigned start, unsigned end)
 578 {
 579    if (start == end - 1) {
 580       return nir_channel(b, vec, start);
 581    } else {
 582       unsigned mid = start + (end - start) / 2;
 583       return nir_bcsel(b, nir_ilt(b, c, nir_imm_int(b, mid)),
 584                        _nir_vector_extract_helper(b, vec, c, start, mid),
 585                        _nir_vector_extract_helper(b, vec, c, mid, end));
 586    }
 587 }
 588
 589 static inline nir_ssa_def *
 590 nir_vector_extract(nir_builder *b, nir_ssa_def *vec, nir_ssa_def *c)
 591 {
 592    nir_src c_src = nir_src_for_ssa(c);
 593    if (nir_src_is_const(c_src)) {
 594       unsigned c_const = nir_src_as_uint(c_src);
 595       if (c_const < vec->num_components)
 596          return nir_channel(b, vec, c_const);
 597       else
 598          return nir_ssa_undef(b, 1, vec->bit_size);
 599    } else {
 600       return _nir_vector_extract_helper(b, vec, c, 0, vec->num_components);
 601    }
 602 }
 603
 604 static inline nir_ssa_def *
 605 nir_i2i(nir_builder *build, nir_ssa_def *x, unsigned dest_bit_size)
 606 {
 607    if (x->bit_size == dest_bit_size)
 608       return x;
 609
 610    switch (dest_bit_size) {
 611    case 64: return nir_i2i64(build, x);
 612    case 32: return nir_i2i32(build, x);
 613    case 16: return nir_i2i16(build, x);
 614    case 8:  return nir_i2i8(build, x);
 615    default: unreachable("Invalid bit size");
 616    }
 617 }
 618
 619 static inline nir_ssa_def *
 620 nir_u2u(nir_builder *build, nir_ssa_def *x, unsigned dest_bit_size)
 621 {
 622    if (x->bit_size == dest_bit_size)
 623       return x;
 624
 625    switch (dest_bit_size) {
 626    case 64: return nir_u2u64(build, x);
 627    case 32: return nir_u2u32(build, x);
 628    case 16: return nir_u2u16(build, x);
 629    case 8:  return nir_u2u8(build, x);
 630    default: unreachable("Invalid bit size");
 631    }
 632 }
 633
 634 static inline nir_ssa_def *
 635 nir_iadd_imm(nir_builder *build, nir_ssa_def *x, uint64_t y)
 636 {
 637    assert(x->bit_size <= 64);
 638    if (x->bit_size < 64)
 639       y &= (1ull << x->bit_size) - 1;
 640
 641    if (y == 0) {
 642       return x;
 643    } else {
 644       return nir_iadd(build, x, nir_imm_intN_t(build, y, x->bit_size));
 645    }
 646 }
 647
 648 static inline nir_ssa_def *
 649 _nir_mul_imm(nir_builder *build, nir_ssa_def *x, uint64_t y, bool amul)
 650 {
 651    assert(x->bit_size <= 64);
 652    if (x->bit_size < 64)
 653       y &= (1ull << x->bit_size) - 1;
 654
 655    if (y == 0) {
 656       return nir_imm_intN_t(build, 0, x->bit_size);
 657    } else if (y == 1) {
 658       return x;
 659    } else if (!build->shader->options->lower_bitops &&
 660               util_is_power_of_two_or_zero64(y)) {
 661       return nir_ishl(build, x, nir_imm_int(build, ffsll(y) - 1));
 662    } else if (amul) {
 663       return nir_amul(build, x, nir_imm_intN_t(build, y, x->bit_size));
 664    } else {
 665       return nir_imul(build, x, nir_imm_intN_t(build, y, x->bit_size));
 666    }
 667 }
 668
 669 static inline nir_ssa_def *
 670 nir_imul_imm(nir_builder *build, nir_ssa_def *x, uint64_t y)
 671 {
 672    return _nir_mul_imm(build, x, y, false);
 673 }
 674
 675 static inline nir_ssa_def *
 676 nir_amul_imm(nir_builder *build, nir_ssa_def *x, uint64_t y)
 677 {
 678    return _nir_mul_imm(build, x, y, true);
 679 }
 680
 681 static inline nir_ssa_def *
 682 nir_fadd_imm(nir_builder *build, nir_ssa_def *x, double y)
 683 {
 684    return nir_fadd(build, x, nir_imm_floatN_t(build, y, x->bit_size));
 685 }
 686
 687 static inline nir_ssa_def *
 688 nir_fmul_imm(nir_builder *build, nir_ssa_def *x, double y)
 689 {
 690    return nir_fmul(build, x, nir_imm_floatN_t(build, y, x->bit_size));
 691 }
 692
 693 static inline nir_ssa_def *
 694 nir_pack_bits(nir_builder *b, nir_ssa_def *src, unsigned dest_bit_size)
 695 {
 696    assert(src->num_components * src->bit_size == dest_bit_size);
 697
 698    switch (dest_bit_size) {
 699    case 64:
 700       switch (src->bit_size) {
 701       case 32: return nir_pack_64_2x32(b, src);
 702       case 16: return nir_pack_64_4x16(b, src);
 703       default: break;
 704       }
 705       break;
 706
 707    case 32:
 708       if (src->bit_size == 16)
 709          return nir_pack_32_2x16(b, src);
 710       break;
 711
 712    default:
 713       break;
 714    }
 715
 716    /* If we got here, we have no dedicated unpack opcode. */
 717    nir_ssa_def *dest = nir_imm_intN_t(b, 0, dest_bit_size);
 718    for (unsigned i = 0; i < src->num_components; i++) {
 719       nir_ssa_def *val = nir_u2u(b, nir_channel(b, src, i), dest_bit_size);
 720       val = nir_ishl(b, val, nir_imm_int(b, i * src->bit_size));
 721       dest = nir_ior(b, dest, val);
 722    }
 723    return dest;
 724 }
 725
 726 static inline nir_ssa_def *
 727 nir_unpack_bits(nir_builder *b, nir_ssa_def *src, unsigned dest_bit_size)
 728 {
 729    assert(src->num_components == 1);
 730    assert(src->bit_size > dest_bit_size);
 731    const unsigned dest_num_components = src->bit_size / dest_bit_size;
 732    assert(dest_num_components <= NIR_MAX_VEC_COMPONENTS);
 733
 734    switch (src->bit_size) {
 735    case 64:
 736       switch (dest_bit_size) {
 737       case 32: return nir_unpack_64_2x32(b, src);
 738       case 16: return nir_unpack_64_4x16(b, src);
 739       default: break;
 740       }
 741       break;
 742
 743    case 32:
 744       if (dest_bit_size == 16)
 745          return nir_unpack_32_2x16(b, src);
 746       break;
 747
 748    default:
 749       break;
 750    }
 751
 752    /* If we got here, we have no dedicated unpack opcode. */
 753    nir_ssa_def *dest_comps[NIR_MAX_VEC_COMPONENTS];
 754    for (unsigned i = 0; i < dest_num_components; i++) {
 755       nir_ssa_def *val = nir_ushr(b, src, nir_imm_int(b, i * dest_bit_size));
 756       dest_comps[i] = nir_u2u(b, val, dest_bit_size);
 757    }
 758    return nir_vec(b, dest_comps, dest_num_components);
 759 }
 760
 761 /**
 762  * Treats srcs as if it's one big blob of bits and extracts the range of bits
 763  * given by
 764  *
 765  *       [first_bit, first_bit + dest_num_components * dest_bit_size)
 766  *
 767  * The range can have any alignment or size as long as it's an integer number
 768  * of destination components and fits inside the concatenated sources.
 769  *
 770  * TODO: The one caveat here is that we can't handle byte alignment if 64-bit
 771  * values are involved because that would require pack/unpack to/from a vec8
 772  * which NIR currently does not support.
 773  */
 774 static inline nir_ssa_def *
 775 nir_extract_bits(nir_builder *b, nir_ssa_def **srcs, unsigned num_srcs,
 776                  unsigned first_bit,
 777                  unsigned dest_num_components, unsigned dest_bit_size)
 778 {
 779    const unsigned num_bits = dest_num_components * dest_bit_size;
 780
 781    /* Figure out the common bit size */
 782    unsigned common_bit_size = dest_bit_size;
 783    for (unsigned i = 0; i < num_srcs; i++)
 784       common_bit_size = MIN2(common_bit_size, srcs[i]->bit_size);
 785    if (first_bit > 0)
 786       common_bit_size = MIN2(common_bit_size, (1u << (ffs(first_bit) - 1)));
 787
 788    /* We don't want to have to deal with 1-bit values */
 789    assert(common_bit_size >= 8);
 790
 791    nir_ssa_def *common_comps[NIR_MAX_VEC_COMPONENTS * sizeof(uint64_t)];
 792    assert(num_bits / common_bit_size <= ARRAY_SIZE(common_comps));
 793
 794    /* First, unpack to the common bit size and select the components from the
 795     * source.
 796     */
 797    int src_idx = -1;
 798    unsigned src_start_bit = 0;
 799    unsigned src_end_bit = 0;
 800    for (unsigned i = 0; i < num_bits / common_bit_size; i++) {
 801       const unsigned bit = first_bit + (i * common_bit_size);
 802       while (bit >= src_end_bit) {
 803          src_idx++;
 804          assert(src_idx < (int) num_srcs);
 805          src_start_bit = src_end_bit;
 806          src_end_bit += srcs[src_idx]->bit_size *
 807                         srcs[src_idx]->num_components;
 808       }
 809       assert(bit >= src_start_bit);
 810       assert(bit + common_bit_size <= src_end_bit);
 811       const unsigned rel_bit = bit - src_start_bit;
 812       const unsigned src_bit_size = srcs[src_idx]->bit_size;
 813
 814       nir_ssa_def *comp = nir_channel(b, srcs[src_idx],
 815                                       rel_bit / src_bit_size);
 816       if (srcs[src_idx]->bit_size > common_bit_size) {
 817          nir_ssa_def *unpacked = nir_unpack_bits(b, comp, common_bit_size);
 818          comp = nir_channel(b, unpacked, (rel_bit % src_bit_size) /
 819                                          common_bit_size);
 820       }
 821       common_comps[i] = comp;
 822    }
 823
 824    /* Now, re-pack the destination if we have to */
 825    if (dest_bit_size > common_bit_size) {
 826       unsigned common_per_dest = dest_bit_size / common_bit_size;
 827       nir_ssa_def *dest_comps[NIR_MAX_VEC_COMPONENTS];
 828       for (unsigned i = 0; i < dest_num_components; i++) {
 829          nir_ssa_def *unpacked = nir_vec(b, common_comps + i * common_per_dest,
 830                                          common_per_dest);
 831          dest_comps[i] = nir_pack_bits(b, unpacked, dest_bit_size);
 832       }
 833       return nir_vec(b, dest_comps, dest_num_components);
 834    } else {
 835       assert(dest_bit_size == common_bit_size);
 836       return nir_vec(b, common_comps, dest_num_components);
 837    }
 838 }
 839
 840 static inline nir_ssa_def *
 841 nir_bitcast_vector(nir_builder *b, nir_ssa_def *src, unsigned dest_bit_size)
 842 {
 843    assert((src->bit_size * src->num_components) % dest_bit_size == 0);
 844    const unsigned dest_num_components =
 845       (src->bit_size * src->num_components) / dest_bit_size;
 846    assert(dest_num_components <= NIR_MAX_VEC_COMPONENTS);
 847
 848    return nir_extract_bits(b, &src, 1, 0, dest_num_components, dest_bit_size);
 849 }
 850
 851 /**
 852  * Turns a nir_src into a nir_ssa_def * so it can be passed to
 853  * nir_build_alu()-based builder calls.
 854  *
 855  * See nir_ssa_for_alu_src() for alu instructions.
 856  */
 857 static inline nir_ssa_def *
 858 nir_ssa_for_src(nir_builder *build, nir_src src, int num_components)
 859 {
 860    if (src.is_ssa && src.ssa->num_components == num_components)
 861       return src.ssa;
 862
 863    nir_alu_src alu = { NIR_SRC_INIT };
 864    alu.src = src;
 865    for (int j = 0; j < 4; j++)
 866       alu.swizzle[j] = j;
 867
 868    return nir_mov_alu(build, alu, num_components);
 869 }
 870
 871 /**
 872  * Similar to nir_ssa_for_src(), but for alu srcs, respecting the
 873  * nir_alu_src's swizzle.
 874  */
 875 static inline nir_ssa_def *
 876 nir_ssa_for_alu_src(nir_builder *build, nir_alu_instr *instr, unsigned srcn)
 877 {
 878    static uint8_t trivial_swizzle[] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 };
 879    STATIC_ASSERT(ARRAY_SIZE(trivial_swizzle) == NIR_MAX_VEC_COMPONENTS);
 880
 881    nir_alu_src *src = &instr->src[srcn];
 882    unsigned num_components = nir_ssa_alu_instr_src_components(instr, srcn);
 883
 884    if (src->src.is_ssa && (src->src.ssa->num_components == num_components) &&
 885        !src->abs && !src->negate &&
 886        (memcmp(src->swizzle, trivial_swizzle, num_components) == 0))
 887       return src->src.ssa;
 888
 889    return nir_mov_alu(build, *src, num_components);
 890 }
 891
 892 static inline unsigned
 893 nir_get_ptr_bitsize(nir_builder *build)
 894 {
 895    if (build->shader->info.stage == MESA_SHADER_KERNEL)
 896       return build->shader->info.cs.ptr_size;
 897    return 32;
 898 }
 899
 900 static inline nir_deref_instr *
 901 nir_build_deref_var(nir_builder *build, nir_variable *var)
 902 {
 903    nir_deref_instr *deref =
 904       nir_deref_instr_create(build->shader, nir_deref_type_var);
 905
 906    deref->mode = var->data.mode;
 907    deref->type = var->type;
 908    deref->var = var;
 909
 910    nir_ssa_dest_init(&deref->instr, &deref->dest, 1,
 911                      nir_get_ptr_bitsize(build), NULL);
 912
 913    nir_builder_instr_insert(build, &deref->instr);
 914
 915    return deref;
 916 }
 917
 918 static inline nir_deref_instr *
 919 nir_build_deref_array(nir_builder *build, nir_deref_instr *parent,
 920                       nir_ssa_def *index)
 921 {
 922    assert(glsl_type_is_array(parent->type) ||
 923           glsl_type_is_matrix(parent->type) ||
 924           glsl_type_is_vector(parent->type));
 925
 926    assert(index->bit_size == parent->dest.ssa.bit_size);
 927
 928    nir_deref_instr *deref =
 929       nir_deref_instr_create(build->shader, nir_deref_type_array);
 930
 931    deref->mode = parent->mode;
 932    deref->type = glsl_get_array_element(parent->type);
 933    deref->parent = nir_src_for_ssa(&parent->dest.ssa);
 934    deref->arr.index = nir_src_for_ssa(index);
 935
 936    nir_ssa_dest_init(&deref->instr, &deref->dest,
 937                      parent->dest.ssa.num_components,
 938                      parent->dest.ssa.bit_size, NULL);
 939
 940    nir_builder_instr_insert(build, &deref->instr);
 941
 942    return deref;
 943 }
 944
 945 static inline nir_deref_instr *
 946 nir_build_deref_array_imm(nir_builder *build, nir_deref_instr *parent,
 947                           int64_t index)
 948 {
 949    assert(parent->dest.is_ssa);
 950    nir_ssa_def *idx_ssa = nir_imm_intN_t(build, index,
 951                                          parent->dest.ssa.bit_size);
 952
 953    return nir_build_deref_array(build, parent, idx_ssa);
 954 }
 955
 956 static inline nir_deref_instr *
 957 nir_build_deref_ptr_as_array(nir_builder *build, nir_deref_instr *parent,
 958                              nir_ssa_def *index)
 959 {
 960    assert(parent->deref_type == nir_deref_type_array ||
 961           parent->deref_type == nir_deref_type_ptr_as_array ||
 962           parent->deref_type == nir_deref_type_cast);
 963
 964    assert(index->bit_size == parent->dest.ssa.bit_size);
 965
 966    nir_deref_instr *deref =
 967       nir_deref_instr_create(build->shader, nir_deref_type_ptr_as_array);
 968
 969    deref->mode = parent->mode;
 970    deref->type = parent->type;
 971    deref->parent = nir_src_for_ssa(&parent->dest.ssa);
 972    deref->arr.index = nir_src_for_ssa(index);
 973
 974    nir_ssa_dest_init(&deref->instr, &deref->dest,
 975                      parent->dest.ssa.num_components,
 976                      parent->dest.ssa.bit_size, NULL);
 977
 978    nir_builder_instr_insert(build, &deref->instr);
 979
 980    return deref;
 981 }
 982
 983 static inline nir_deref_instr *
 984 nir_build_deref_array_wildcard(nir_builder *build, nir_deref_instr *parent)
 985 {
 986    assert(glsl_type_is_array(parent->type) ||
 987           glsl_type_is_matrix(parent->type));
 988
 989    nir_deref_instr *deref =
 990       nir_deref_instr_create(build->shader, nir_deref_type_array_wildcard);
 991
 992    deref->mode = parent->mode;
 993    deref->type = glsl_get_array_element(parent->type);
 994    deref->parent = nir_src_for_ssa(&parent->dest.ssa);
 995
 996    nir_ssa_dest_init(&deref->instr, &deref->dest,
 997                      parent->dest.ssa.num_components,
 998                      parent->dest.ssa.bit_size, NULL);
 999
1000    nir_builder_instr_insert(build, &deref->instr);
1001
1002    return deref;
1003 }
1004
1005 static inline nir_deref_instr *
1006 nir_build_deref_struct(nir_builder *build, nir_deref_instr *parent,
1007                        unsigned index)
1008 {
1009    assert(glsl_type_is_struct_or_ifc(parent->type));
1010
1011    nir_deref_instr *deref =
1012       nir_deref_instr_create(build->shader, nir_deref_type_struct);
1013
1014    deref->mode = parent->mode;
1015    deref->type = glsl_get_struct_field(parent->type, index);
1016    deref->parent = nir_src_for_ssa(&parent->dest.ssa);
1017    deref->strct.index = index;
1018
1019    nir_ssa_dest_init(&deref->instr, &deref->dest,
1020                      parent->dest.ssa.num_components,
1021                      parent->dest.ssa.bit_size, NULL);
1022
1023    nir_builder_instr_insert(build, &deref->instr);
1024
1025    return deref;
1026 }
1027
1028 static inline nir_deref_instr *
1029 nir_build_deref_cast(nir_builder *build, nir_ssa_def *parent,
1030                      nir_variable_mode mode, const struct glsl_type *type,
1031                      unsigned ptr_stride)
1032 {
1033    nir_deref_instr *deref =
1034       nir_deref_instr_create(build->shader, nir_deref_type_cast);
1035
1036    deref->mode = mode;
1037    deref->type = type;
1038    deref->parent = nir_src_for_ssa(parent);
1039    deref->cast.ptr_stride = ptr_stride;
1040
1041    nir_ssa_dest_init(&deref->instr, &deref->dest,
1042                      parent->num_components, parent->bit_size, NULL);
1043
1044    nir_builder_instr_insert(build, &deref->instr);
1045
1046    return deref;
1047 }
1048
1049 /** Returns a deref that follows another but starting from the given parent
1050  *
1051  * The new deref will be the same type and take the same array or struct index
1052  * as the leader deref but it may have a different parent.  This is very
1053  * useful for walking deref paths.
1054  */
1055 static inline nir_deref_instr *
1056 nir_build_deref_follower(nir_builder *b, nir_deref_instr *parent,
1057                          nir_deref_instr *leader)
1058 {
1059    /* If the derefs would have the same parent, don't make a new one */
1060    assert(leader->parent.is_ssa);
1061    if (leader->parent.ssa == &parent->dest.ssa)
1062       return leader;
1063
1064    UNUSED nir_deref_instr *leader_parent = nir_src_as_deref(leader->parent);
1065
1066    switch (leader->deref_type) {
1067    case nir_deref_type_var:
1068       unreachable("A var dereference cannot have a parent");
1069       break;
1070
1071    case nir_deref_type_array:
1072    case nir_deref_type_array_wildcard:
1073       assert(glsl_type_is_matrix(parent->type) ||
1074              glsl_type_is_array(parent->type) ||
1075              (leader->deref_type == nir_deref_type_array &&
1076               glsl_type_is_vector(parent->type)));
1077       assert(glsl_get_length(parent->type) ==
1078              glsl_get_length(leader_parent->type));
1079
1080       if (leader->deref_type == nir_deref_type_array) {
1081          assert(leader->arr.index.is_ssa);
1082          nir_ssa_def *index = nir_i2i(b, leader->arr.index.ssa,
1083                                          parent->dest.ssa.bit_size);
1084          return nir_build_deref_array(b, parent, index);
1085       } else {
1086          return nir_build_deref_array_wildcard(b, parent);
1087       }
1088
1089    case nir_deref_type_struct:
1090       assert(glsl_type_is_struct_or_ifc(parent->type));
1091       assert(glsl_get_length(parent->type) ==
1092              glsl_get_length(leader_parent->type));
1093
1094       return nir_build_deref_struct(b, parent, leader->strct.index);
1095
1096    default:
1097       unreachable("Invalid deref instruction type");
1098    }
1099 }
1100
1101 static inline nir_ssa_def *
1102 nir_load_reg(nir_builder *build, nir_register *reg)
1103 {
1104    return nir_ssa_for_src(build, nir_src_for_reg(reg), reg->num_components);
1105 }
1106
1107 static inline nir_ssa_def *
1108 nir_load_deref_with_access(nir_builder *build, nir_deref_instr *deref,
1109                            enum gl_access_qualifier access)
1110 {
1111    nir_intrinsic_instr *load =
1112       nir_intrinsic_instr_create(build->shader, nir_intrinsic_load_deref);
1113    load->num_components = glsl_get_vector_elements(deref->type);
1114    load->src[0] = nir_src_for_ssa(&deref->dest.ssa);
1115    nir_ssa_dest_init(&load->instr, &load->dest, load->num_components,
1116                      glsl_get_bit_size(deref->type), NULL);
1117    nir_intrinsic_set_access(load, access);
1118    nir_builder_instr_insert(build, &load->instr);
1119    return &load->dest.ssa;
1120 }
1121
1122 static inline nir_ssa_def *
1123 nir_load_deref(nir_builder *build, nir_deref_instr *deref)
1124 {
1125    return nir_load_deref_with_access(build, deref, (enum gl_access_qualifier)0);
1126 }
1127
1128 static inline void
1129 nir_store_deref_with_access(nir_builder *build, nir_deref_instr *deref,
1130                             nir_ssa_def *value, unsigned writemask,
1131                             enum gl_access_qualifier access)
1132 {
1133    nir_intrinsic_instr *store =
1134       nir_intrinsic_instr_create(build->shader, nir_intrinsic_store_deref);
1135    store->num_components = glsl_get_vector_elements(deref->type);
1136    store->src[0] = nir_src_for_ssa(&deref->dest.ssa);
1137    store->src[1] = nir_src_for_ssa(value);
1138    nir_intrinsic_set_write_mask(store,
1139                                 writemask & ((1 << store->num_components) - 1));
1140    nir_intrinsic_set_access(store, access);
1141    nir_builder_instr_insert(build, &store->instr);
1142 }
1143
1144 static inline void
1145 nir_store_deref(nir_builder *build, nir_deref_instr *deref,
1146                 nir_ssa_def *value, unsigned writemask)
1147 {
1148    nir_store_deref_with_access(build, deref, value, writemask,
1149                                (enum gl_access_qualifier)0);
1150 }
1151
1152 static inline void
1153 nir_copy_deref_with_access(nir_builder *build, nir_deref_instr *dest,
1154                            nir_deref_instr *src,
1155                            enum gl_access_qualifier dest_access,
1156                            enum gl_access_qualifier src_access)
1157 {
1158    nir_intrinsic_instr *copy =
1159       nir_intrinsic_instr_create(build->shader, nir_intrinsic_copy_deref);
1160    copy->src[0] = nir_src_for_ssa(&dest->dest.ssa);
1161    copy->src[1] = nir_src_for_ssa(&src->dest.ssa);
1162    nir_intrinsic_set_dst_access(copy, dest_access);
1163    nir_intrinsic_set_src_access(copy, src_access);
1164    nir_builder_instr_insert(build, &copy->instr);
1165 }
1166
1167 static inline void
1168 nir_copy_deref(nir_builder *build, nir_deref_instr *dest, nir_deref_instr *src)
1169 {
1170    nir_copy_deref_with_access(build, dest, src,
1171                               (enum gl_access_qualifier) 0,
1172                               (enum gl_access_qualifier) 0);
1173 }
1174
1175 static inline nir_ssa_def *
1176 nir_load_var(nir_builder *build, nir_variable *var)
1177 {
1178    return nir_load_deref(build, nir_build_deref_var(build, var));
1179 }
1180
1181 static inline void
1182 nir_store_var(nir_builder *build, nir_variable *var, nir_ssa_def *value,
1183               unsigned writemask)
1184 {
1185    nir_store_deref(build, nir_build_deref_var(build, var), value, writemask);
1186 }
1187
1188 static inline void
1189 nir_copy_var(nir_builder *build, nir_variable *dest, nir_variable *src)
1190 {
1191    nir_copy_deref(build, nir_build_deref_var(build, dest),
1192                          nir_build_deref_var(build, src));
1193 }
1194
1195 static inline nir_ssa_def *
1196 nir_load_param(nir_builder *build, uint32_t param_idx)
1197 {
1198    assert(param_idx < build->impl->function->num_params);
1199    nir_parameter *param = &build->impl->function->params[param_idx];
1200
1201    nir_intrinsic_instr *load =
1202       nir_intrinsic_instr_create(build->shader, nir_intrinsic_load_param);
1203    nir_intrinsic_set_param_idx(load, param_idx);
1204    load->num_components = param->num_components;
1205    nir_ssa_dest_init(&load->instr, &load->dest,
1206                      param->num_components, param->bit_size, NULL);
1207    nir_builder_instr_insert(build, &load->instr);
1208    return &load->dest.ssa;
1209 }
1210
1211 #include "nir_builder_opcodes.h"
1212
1213 static inline nir_ssa_def *
1214 nir_f2b(nir_builder *build, nir_ssa_def *f)
1215 {
1216    return nir_f2b1(build, f);
1217 }
1218
1219 static inline nir_ssa_def *
1220 nir_i2b(nir_builder *build, nir_ssa_def *i)
1221 {
1222    return nir_i2b1(build, i);
1223 }
1224
1225 static inline nir_ssa_def *
1226 nir_b2f(nir_builder *build, nir_ssa_def *b, uint32_t bit_size)
1227 {
1228    switch (bit_size) {
1229    case 64: return nir_b2f64(build, b);
1230    case 32: return nir_b2f32(build, b);
1231    case 16: return nir_b2f16(build, b);
1232    default:
1233       unreachable("Invalid bit-size");
1234    };
1235 }
1236
1237 static inline nir_ssa_def *
1238 nir_b2i(nir_builder *build, nir_ssa_def *b, uint32_t bit_size)
1239 {
1240    switch (bit_size) {
1241    case 64: return nir_b2i64(build, b);
1242    case 32: return nir_b2i32(build, b);
1243    case 16: return nir_b2i16(build, b);
1244    case 8:  return nir_b2i8(build, b);
1245    default:
1246       unreachable("Invalid bit-size");
1247    };
1248 }
1249 static inline nir_ssa_def *
1250 nir_load_barycentric(nir_builder *build, nir_intrinsic_op op,
1251                      unsigned interp_mode)
1252 {
1253    nir_intrinsic_instr *bary = nir_intrinsic_instr_create(build->shader, op);
1254    nir_ssa_dest_init(&bary->instr, &bary->dest, 2, 32, NULL);
1255    nir_intrinsic_set_interp_mode(bary, interp_mode);
1256    nir_builder_instr_insert(build, &bary->instr);
1257    return &bary->dest.ssa;
1258 }
1259
1260 static inline void
1261 nir_jump(nir_builder *build, nir_jump_type jump_type)
1262 {
1263    nir_jump_instr *jump = nir_jump_instr_create(build->shader, jump_type);
1264    nir_builder_instr_insert(build, &jump->instr);
1265 }
1266
1267 static inline nir_ssa_def *
1268 nir_compare_func(nir_builder *b, enum compare_func func,
1269                  nir_ssa_def *src0, nir_ssa_def *src1)
1270 {
1271    switch (func) {
1272    case COMPARE_FUNC_NEVER:
1273       return nir_imm_int(b, 0);
1274    case COMPARE_FUNC_ALWAYS:
1275       return nir_imm_int(b, ~0);
1276    case COMPARE_FUNC_EQUAL:
1277       return nir_feq(b, src0, src1);
1278    case COMPARE_FUNC_NOTEQUAL:
1279       return nir_fne(b, src0, src1);
1280    case COMPARE_FUNC_GREATER:
1281       return nir_flt(b, src1, src0);
1282    case COMPARE_FUNC_GEQUAL:
1283       return nir_fge(b, src0, src1);
1284    case COMPARE_FUNC_LESS:
1285       return nir_flt(b, src0, src1);
1286    case COMPARE_FUNC_LEQUAL:
1287       return nir_fge(b, src1, src0);
1288    }
1289    unreachable("bad compare func");
1290 }
1291
1292 #endif /* NIR_BUILDER_H */