src/amd/common/ac_llvm_build.h

   1 /*
   2  * Copyright 2016 Bas Nieuwenhuizen
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the
   6  * "Software"), to deal in the Software without restriction, including
   7  * without limitation the rights to use, copy, modify, merge, publish,
   8  * distribute, sub license, and/or sell copies of the Software, and to
   9  * permit persons to whom the Software is furnished to do so, subject to
  10  * the following conditions:
  11  *
  12  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  13  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  14  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  15  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
  16  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  17  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  18  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  19  *
  20  * The above copyright notice and this permission notice (including the
  21  * next paragraph) shall be included in all copies or substantial portions
  22  * of the Software.
  23  *
  24  */
  25 #ifndef AC_LLVM_BUILD_H
  26 #define AC_LLVM_BUILD_H
  27
  28 #include <stdbool.h>
  29 #include <llvm-c/TargetMachine.h>
  30 #include "compiler/nir/nir.h"
  31 #include "amd_family.h"
  32
  33 #ifdef __cplusplus
  34 extern "C" {
  35 #endif
  36
  37 enum {
  38         AC_ADDR_SPACE_FLAT = 0, /* Slower than global. */
  39         AC_ADDR_SPACE_GLOBAL = 1,
  40         AC_ADDR_SPACE_GDS = 2,
  41         AC_ADDR_SPACE_LDS = 3,
  42         AC_ADDR_SPACE_CONST = 4, /* Global allowing SMEM. */
  43         AC_ADDR_SPACE_CONST_32BIT = 6, /* same as CONST, but the pointer type has 32 bits */
  44 };
  45
  46 /* Combine these with & instead of |. */
  47 #define NOOP_WAITCNT    0xcf7f
  48 #define LGKM_CNT        0xc07f
  49 #define EXP_CNT         0xcf0f
  50 #define VM_CNT          0x0f70 /* On GFX9, vmcnt has 6 bits in [0:3] and [14:15] */
  51
  52 struct ac_llvm_flow;
  53
  54 struct ac_llvm_context {
  55         LLVMContextRef context;
  56         LLVMModuleRef module;
  57         LLVMBuilderRef builder;
  58
  59         LLVMTypeRef voidt;
  60         LLVMTypeRef i1;
  61         LLVMTypeRef i8;
  62         LLVMTypeRef i16;
  63         LLVMTypeRef i32;
  64         LLVMTypeRef i64;
  65         LLVMTypeRef intptr;
  66         LLVMTypeRef f16;
  67         LLVMTypeRef f32;
  68         LLVMTypeRef f64;
  69         LLVMTypeRef v2i16;
  70         LLVMTypeRef v2i32;
  71         LLVMTypeRef v3i32;
  72         LLVMTypeRef v4i32;
  73         LLVMTypeRef v2f32;
  74         LLVMTypeRef v4f32;
  75         LLVMTypeRef v8i32;
  76
  77         LLVMValueRef i16_0;
  78         LLVMValueRef i16_1;
  79         LLVMValueRef i32_0;
  80         LLVMValueRef i32_1;
  81         LLVMValueRef i64_0;
  82         LLVMValueRef i64_1;
  83         LLVMValueRef f32_0;
  84         LLVMValueRef f32_1;
  85         LLVMValueRef f64_0;
  86         LLVMValueRef f64_1;
  87         LLVMValueRef i1true;
  88         LLVMValueRef i1false;
  89
  90         struct ac_llvm_flow *flow;
  91         unsigned flow_depth;
  92         unsigned flow_depth_max;
  93
  94         unsigned range_md_kind;
  95         unsigned invariant_load_md_kind;
  96         unsigned uniform_md_kind;
  97         unsigned fpmath_md_kind;
  98         LLVMValueRef fpmath_md_2p5_ulp;
  99         LLVMValueRef empty_md;
 100
 101         enum chip_class chip_class;
 102         enum radeon_family family;
 103
 104         LLVMValueRef lds;
 105 };
 106
 107 void
 108 ac_llvm_context_init(struct ac_llvm_context *ctx,
 109                      enum chip_class chip_class, enum radeon_family family);
 110
 111 void
 112 ac_llvm_context_dispose(struct ac_llvm_context *ctx);
 113
 114 int
 115 ac_get_llvm_num_components(LLVMValueRef value);
 116
 117 int
 118 ac_get_elem_bits(struct ac_llvm_context *ctx, LLVMTypeRef type);
 119
 120 LLVMValueRef
 121 ac_llvm_extract_elem(struct ac_llvm_context *ac,
 122                      LLVMValueRef value,
 123                      int index);
 124
 125 unsigned ac_get_type_size(LLVMTypeRef type);
 126
 127 LLVMTypeRef ac_to_integer_type(struct ac_llvm_context *ctx, LLVMTypeRef t);
 128 LLVMValueRef ac_to_integer(struct ac_llvm_context *ctx, LLVMValueRef v);
 129 LLVMValueRef ac_to_integer_or_pointer(struct ac_llvm_context *ctx, LLVMValueRef v);
 130 LLVMTypeRef ac_to_float_type(struct ac_llvm_context *ctx, LLVMTypeRef t);
 131 LLVMValueRef ac_to_float(struct ac_llvm_context *ctx, LLVMValueRef v);
 132
 133 LLVMValueRef
 134 ac_build_intrinsic(struct ac_llvm_context *ctx, const char *name,
 135                    LLVMTypeRef return_type, LLVMValueRef *params,
 136                    unsigned param_count, unsigned attrib_mask);
 137
 138 void ac_build_type_name_for_intr(LLVMTypeRef type, char *buf, unsigned bufsize);
 139
 140 LLVMValueRef
 141 ac_build_phi(struct ac_llvm_context *ctx, LLVMTypeRef type,
 142              unsigned count_incoming, LLVMValueRef *values,
 143              LLVMBasicBlockRef *blocks);
 144
 145 void ac_build_s_barrier(struct ac_llvm_context *ctx);
 146 void ac_build_optimization_barrier(struct ac_llvm_context *ctx,
 147                                    LLVMValueRef *pvgpr);
 148
 149 LLVMValueRef ac_build_shader_clock(struct ac_llvm_context *ctx);
 150
 151 LLVMValueRef ac_build_ballot(struct ac_llvm_context *ctx, LLVMValueRef value);
 152
 153 LLVMValueRef ac_build_vote_all(struct ac_llvm_context *ctx, LLVMValueRef value);
 154
 155 LLVMValueRef ac_build_vote_any(struct ac_llvm_context *ctx, LLVMValueRef value);
 156
 157 LLVMValueRef ac_build_vote_eq(struct ac_llvm_context *ctx, LLVMValueRef value);
 158
 159 LLVMValueRef
 160 ac_build_varying_gather_values(struct ac_llvm_context *ctx, LLVMValueRef *values,
 161                                unsigned value_count, unsigned component);
 162
 163 LLVMValueRef
 164 ac_build_gather_values_extended(struct ac_llvm_context *ctx,
 165                                 LLVMValueRef *values,
 166                                 unsigned value_count,
 167                                 unsigned value_stride,
 168                                 bool load,
 169                                 bool always_vector);
 170 LLVMValueRef
 171 ac_build_gather_values(struct ac_llvm_context *ctx,
 172                        LLVMValueRef *values,
 173                        unsigned value_count);
 174 LLVMValueRef ac_build_expand(struct ac_llvm_context *ctx,
 175                              LLVMValueRef value,
 176                              unsigned src_channels, unsigned dst_channels);
 177 LLVMValueRef ac_build_expand_to_vec4(struct ac_llvm_context *ctx,
 178                                      LLVMValueRef value,
 179                                      unsigned num_channels);
 180 LLVMValueRef ac_build_round(struct ac_llvm_context *ctx, LLVMValueRef value);
 181
 182 LLVMValueRef
 183 ac_build_fdiv(struct ac_llvm_context *ctx,
 184               LLVMValueRef num,
 185               LLVMValueRef den);
 186
 187 LLVMValueRef ac_build_fast_udiv(struct ac_llvm_context *ctx,
 188                                 LLVMValueRef num,
 189                                 LLVMValueRef multiplier,
 190                                 LLVMValueRef pre_shift,
 191                                 LLVMValueRef post_shift,
 192                                 LLVMValueRef increment);
 193 LLVMValueRef ac_build_fast_udiv_nuw(struct ac_llvm_context *ctx,
 194                                     LLVMValueRef num,
 195                                     LLVMValueRef multiplier,
 196                                     LLVMValueRef pre_shift,
 197                                     LLVMValueRef post_shift,
 198                                     LLVMValueRef increment);
 199 LLVMValueRef ac_build_fast_udiv_u31_d_not_one(struct ac_llvm_context *ctx,
 200                                               LLVMValueRef num,
 201                                               LLVMValueRef multiplier,
 202                                               LLVMValueRef post_shift);
 203
 204 void
 205 ac_prepare_cube_coords(struct ac_llvm_context *ctx,
 206                        bool is_deriv, bool is_array, bool is_lod,
 207                        LLVMValueRef *coords_arg,
 208                        LLVMValueRef *derivs_arg);
 209
 210
 211 LLVMValueRef
 212 ac_build_fs_interp(struct ac_llvm_context *ctx,
 213                    LLVMValueRef llvm_chan,
 214                    LLVMValueRef attr_number,
 215                    LLVMValueRef params,
 216                    LLVMValueRef i,
 217                    LLVMValueRef j);
 218
 219 LLVMValueRef
 220 ac_build_fs_interp_mov(struct ac_llvm_context *ctx,
 221                        LLVMValueRef parameter,
 222                        LLVMValueRef llvm_chan,
 223                        LLVMValueRef attr_number,
 224                        LLVMValueRef params);
 225
 226 LLVMValueRef
 227 ac_build_gep0(struct ac_llvm_context *ctx,
 228               LLVMValueRef base_ptr,
 229               LLVMValueRef index);
 230 LLVMValueRef ac_build_pointer_add(struct ac_llvm_context *ctx, LLVMValueRef ptr,
 231                                   LLVMValueRef index);
 232
 233 void
 234 ac_build_indexed_store(struct ac_llvm_context *ctx,
 235                        LLVMValueRef base_ptr, LLVMValueRef index,
 236                        LLVMValueRef value);
 237
 238 LLVMValueRef ac_build_load(struct ac_llvm_context *ctx, LLVMValueRef base_ptr,
 239                            LLVMValueRef index);
 240 LLVMValueRef ac_build_load_invariant(struct ac_llvm_context *ctx,
 241                                      LLVMValueRef base_ptr, LLVMValueRef index);
 242 LLVMValueRef ac_build_load_to_sgpr(struct ac_llvm_context *ctx,
 243                                    LLVMValueRef base_ptr, LLVMValueRef index);
 244 LLVMValueRef ac_build_load_to_sgpr_uint_wraparound(struct ac_llvm_context *ctx,
 245                                    LLVMValueRef base_ptr, LLVMValueRef index);
 246
 247 void
 248 ac_build_buffer_store_dword(struct ac_llvm_context *ctx,
 249                             LLVMValueRef rsrc,
 250                             LLVMValueRef vdata,
 251                             unsigned num_channels,
 252                             LLVMValueRef voffset,
 253                             LLVMValueRef soffset,
 254                             unsigned inst_offset,
 255                             bool glc,
 256                             bool slc,
 257                             bool writeonly_memory,
 258                             bool swizzle_enable_hint);
 259 LLVMValueRef
 260 ac_build_buffer_load(struct ac_llvm_context *ctx,
 261                      LLVMValueRef rsrc,
 262                      int num_channels,
 263                      LLVMValueRef vindex,
 264                      LLVMValueRef voffset,
 265                      LLVMValueRef soffset,
 266                      unsigned inst_offset,
 267                      unsigned glc,
 268                      unsigned slc,
 269                      bool can_speculate,
 270                      bool allow_smem);
 271
 272 LLVMValueRef ac_build_buffer_load_format(struct ac_llvm_context *ctx,
 273                                          LLVMValueRef rsrc,
 274                                          LLVMValueRef vindex,
 275                                          LLVMValueRef voffset,
 276                                          unsigned num_channels,
 277                                          bool glc,
 278                                          bool can_speculate);
 279
 280 /* load_format that handles the stride & element count better if idxen is
 281  * disabled by LLVM. */
 282 LLVMValueRef ac_build_buffer_load_format_gfx9_safe(struct ac_llvm_context *ctx,
 283                                                   LLVMValueRef rsrc,
 284                                                   LLVMValueRef vindex,
 285                                                   LLVMValueRef voffset,
 286                                                   unsigned num_channels,
 287                                                   bool glc,
 288                                                   bool can_speculate);
 289
 290 LLVMValueRef
 291 ac_build_tbuffer_load_short(struct ac_llvm_context *ctx,
 292                             LLVMValueRef rsrc,
 293                             LLVMValueRef vindex,
 294                             LLVMValueRef voffset,
 295                                 LLVMValueRef soffset,
 296                                 LLVMValueRef immoffset,
 297                                 LLVMValueRef glc);
 298
 299 LLVMValueRef
 300 ac_get_thread_id(struct ac_llvm_context *ctx);
 301
 302 #define AC_TID_MASK_TOP_LEFT 0xfffffffc
 303 #define AC_TID_MASK_TOP      0xfffffffd
 304 #define AC_TID_MASK_LEFT     0xfffffffe
 305
 306 LLVMValueRef
 307 ac_build_ddxy(struct ac_llvm_context *ctx,
 308               uint32_t mask,
 309               int idx,
 310               LLVMValueRef val);
 311
 312 #define AC_SENDMSG_GS 2
 313 #define AC_SENDMSG_GS_DONE 3
 314
 315 #define AC_SENDMSG_GS_OP_NOP      (0 << 4)
 316 #define AC_SENDMSG_GS_OP_CUT      (1 << 4)
 317 #define AC_SENDMSG_GS_OP_EMIT     (2 << 4)
 318 #define AC_SENDMSG_GS_OP_EMIT_CUT (3 << 4)
 319
 320 void ac_build_sendmsg(struct ac_llvm_context *ctx,
 321                       uint32_t msg,
 322                       LLVMValueRef wave_id);
 323
 324 LLVMValueRef ac_build_imsb(struct ac_llvm_context *ctx,
 325                            LLVMValueRef arg,
 326                            LLVMTypeRef dst_type);
 327
 328 LLVMValueRef ac_build_umsb(struct ac_llvm_context *ctx,
 329                           LLVMValueRef arg,
 330                           LLVMTypeRef dst_type);
 331 LLVMValueRef ac_build_fmin(struct ac_llvm_context *ctx, LLVMValueRef a,
 332                            LLVMValueRef b);
 333 LLVMValueRef ac_build_fmax(struct ac_llvm_context *ctx, LLVMValueRef a,
 334                            LLVMValueRef b);
 335 LLVMValueRef ac_build_imin(struct ac_llvm_context *ctx, LLVMValueRef a,
 336                            LLVMValueRef b);
 337 LLVMValueRef ac_build_imax(struct ac_llvm_context *ctx, LLVMValueRef a,
 338                            LLVMValueRef b);
 339 LLVMValueRef ac_build_umin(struct ac_llvm_context *ctx, LLVMValueRef a, LLVMValueRef b);
 340 LLVMValueRef ac_build_clamp(struct ac_llvm_context *ctx, LLVMValueRef value);
 341
 342 struct ac_export_args {
 343         LLVMValueRef out[4];
 344         unsigned target;
 345         unsigned enabled_channels;
 346         bool compr;
 347         bool done;
 348         bool valid_mask;
 349 };
 350
 351 void ac_build_export(struct ac_llvm_context *ctx, struct ac_export_args *a);
 352
 353 void ac_build_export_null(struct ac_llvm_context *ctx);
 354
 355 enum ac_image_opcode {
 356         ac_image_sample,
 357         ac_image_gather4,
 358         ac_image_load,
 359         ac_image_load_mip,
 360         ac_image_store,
 361         ac_image_store_mip,
 362         ac_image_get_lod,
 363         ac_image_get_resinfo,
 364         ac_image_atomic,
 365         ac_image_atomic_cmpswap,
 366 };
 367
 368 enum ac_atomic_op {
 369         ac_atomic_swap,
 370         ac_atomic_add,
 371         ac_atomic_sub,
 372         ac_atomic_smin,
 373         ac_atomic_umin,
 374         ac_atomic_smax,
 375         ac_atomic_umax,
 376         ac_atomic_and,
 377         ac_atomic_or,
 378         ac_atomic_xor,
 379 };
 380
 381 enum ac_image_dim {
 382         ac_image_1d,
 383         ac_image_2d,
 384         ac_image_3d,
 385         ac_image_cube, // includes cube arrays
 386         ac_image_1darray,
 387         ac_image_2darray,
 388         ac_image_2dmsaa,
 389         ac_image_2darraymsaa,
 390 };
 391
 392 /* These cache policy bits match the definitions used by the LLVM intrinsics. */
 393 enum ac_image_cache_policy {
 394         ac_glc = 1 << 0,
 395         ac_slc = 1 << 1,
 396 };
 397
 398 struct ac_image_args {
 399         enum ac_image_opcode opcode : 4;
 400         enum ac_atomic_op atomic : 4; /* for the ac_image_atomic opcode */
 401         enum ac_image_dim dim : 3;
 402         unsigned dmask : 4;
 403         unsigned cache_policy : 2;
 404         bool unorm : 1;
 405         bool level_zero : 1;
 406         unsigned attributes; /* additional call-site specific AC_FUNC_ATTRs */
 407
 408         LLVMValueRef resource;
 409         LLVMValueRef sampler;
 410         LLVMValueRef data[2]; /* data[0] is source data (vector); data[1] is cmp for cmpswap */
 411         LLVMValueRef offset;
 412         LLVMValueRef bias;
 413         LLVMValueRef compare;
 414         LLVMValueRef derivs[6];
 415         LLVMValueRef coords[4];
 416         LLVMValueRef lod; // also used by ac_image_get_resinfo
 417 };
 418
 419 LLVMValueRef ac_build_image_opcode(struct ac_llvm_context *ctx,
 420                                    struct ac_image_args *a);
 421 LLVMValueRef ac_build_cvt_pkrtz_f16(struct ac_llvm_context *ctx,
 422                                     LLVMValueRef args[2]);
 423 LLVMValueRef ac_build_cvt_pknorm_i16(struct ac_llvm_context *ctx,
 424                                      LLVMValueRef args[2]);
 425 LLVMValueRef ac_build_cvt_pknorm_u16(struct ac_llvm_context *ctx,
 426                                      LLVMValueRef args[2]);
 427 LLVMValueRef ac_build_cvt_pk_i16(struct ac_llvm_context *ctx,
 428                                  LLVMValueRef args[2], unsigned bits, bool hi);
 429 LLVMValueRef ac_build_cvt_pk_u16(struct ac_llvm_context *ctx,
 430                                  LLVMValueRef args[2], unsigned bits, bool hi);
 431 LLVMValueRef ac_build_wqm_vote(struct ac_llvm_context *ctx, LLVMValueRef i1);
 432 void ac_build_kill_if_false(struct ac_llvm_context *ctx, LLVMValueRef i1);
 433 LLVMValueRef ac_build_bfe(struct ac_llvm_context *ctx, LLVMValueRef input,
 434                           LLVMValueRef offset, LLVMValueRef width,
 435                           bool is_signed);
 436 LLVMValueRef ac_build_imad(struct ac_llvm_context *ctx, LLVMValueRef s0,
 437                            LLVMValueRef s1, LLVMValueRef s2);
 438 LLVMValueRef ac_build_fmad(struct ac_llvm_context *ctx, LLVMValueRef s0,
 439                            LLVMValueRef s1, LLVMValueRef s2);
 440
 441 void ac_build_waitcnt(struct ac_llvm_context *ctx, unsigned simm16);
 442
 443 LLVMValueRef ac_build_fract(struct ac_llvm_context *ctx, LLVMValueRef src0,
 444                            unsigned bitsize);
 445
 446 LLVMValueRef ac_build_isign(struct ac_llvm_context *ctx, LLVMValueRef src0,
 447                             unsigned bitsize);
 448
 449 LLVMValueRef ac_build_fsign(struct ac_llvm_context *ctx, LLVMValueRef src0,
 450                             unsigned bitsize);
 451
 452 LLVMValueRef ac_build_bit_count(struct ac_llvm_context *ctx, LLVMValueRef src0);
 453
 454 LLVMValueRef ac_build_bitfield_reverse(struct ac_llvm_context *ctx,
 455                                        LLVMValueRef src0);
 456
 457 void ac_optimize_vs_outputs(struct ac_llvm_context *ac,
 458                             LLVMValueRef main_fn,
 459                             uint8_t *vs_output_param_offset,
 460                             uint32_t num_outputs,
 461                             uint8_t *num_param_exports);
 462 void ac_init_exec_full_mask(struct ac_llvm_context *ctx);
 463
 464 void ac_declare_lds_as_pointer(struct ac_llvm_context *ac);
 465 LLVMValueRef ac_lds_load(struct ac_llvm_context *ctx,
 466                          LLVMValueRef dw_addr);
 467 void ac_lds_store(struct ac_llvm_context *ctx,
 468                   LLVMValueRef dw_addr, LLVMValueRef value);
 469
 470 LLVMValueRef ac_find_lsb(struct ac_llvm_context *ctx,
 471                          LLVMTypeRef dst_type,
 472                          LLVMValueRef src0);
 473
 474 LLVMTypeRef ac_array_in_const_addr_space(LLVMTypeRef elem_type);
 475 LLVMTypeRef ac_array_in_const32_addr_space(LLVMTypeRef elem_type);
 476
 477 void ac_build_bgnloop(struct ac_llvm_context *ctx, int lable_id);
 478 void ac_build_break(struct ac_llvm_context *ctx);
 479 void ac_build_continue(struct ac_llvm_context *ctx);
 480 void ac_build_else(struct ac_llvm_context *ctx, int lable_id);
 481 void ac_build_endif(struct ac_llvm_context *ctx, int lable_id);
 482 void ac_build_endloop(struct ac_llvm_context *ctx, int lable_id);
 483 void ac_build_if(struct ac_llvm_context *ctx, LLVMValueRef value,
 484                  int lable_id);
 485 void ac_build_uif(struct ac_llvm_context *ctx, LLVMValueRef value,
 486                   int lable_id);
 487
 488 LLVMValueRef ac_build_alloca(struct ac_llvm_context *ac, LLVMTypeRef type,
 489                              const char *name);
 490 LLVMValueRef ac_build_alloca_undef(struct ac_llvm_context *ac, LLVMTypeRef type,
 491                                    const char *name);
 492
 493 LLVMValueRef ac_cast_ptr(struct ac_llvm_context *ctx, LLVMValueRef ptr,
 494                          LLVMTypeRef type);
 495
 496 LLVMValueRef ac_trim_vector(struct ac_llvm_context *ctx, LLVMValueRef value,
 497                             unsigned count);
 498
 499 LLVMValueRef ac_unpack_param(struct ac_llvm_context *ctx, LLVMValueRef param,
 500                              unsigned rshift, unsigned bitwidth);
 501
 502 void ac_apply_fmask_to_sample(struct ac_llvm_context *ac, LLVMValueRef fmask,
 503                               LLVMValueRef *addr, bool is_array_tex);
 504
 505 LLVMValueRef
 506 ac_build_ds_swizzle(struct ac_llvm_context *ctx, LLVMValueRef src, unsigned mask);
 507
 508 LLVMValueRef
 509 ac_build_readlane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef lane);
 510
 511 LLVMValueRef
 512 ac_build_writelane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef value, LLVMValueRef lane);
 513
 514 LLVMValueRef
 515 ac_build_mbcnt(struct ac_llvm_context *ctx, LLVMValueRef mask);
 516
 517 LLVMValueRef
 518 ac_build_inclusive_scan(struct ac_llvm_context *ctx, LLVMValueRef src, nir_op op);
 519
 520 LLVMValueRef
 521 ac_build_exclusive_scan(struct ac_llvm_context *ctx, LLVMValueRef src, nir_op op);
 522
 523 LLVMValueRef
 524 ac_build_reduce(struct ac_llvm_context *ctx, LLVMValueRef src, nir_op op, unsigned cluster_size);
 525
 526 LLVMValueRef
 527 ac_build_quad_swizzle(struct ac_llvm_context *ctx, LLVMValueRef src,
 528                 unsigned lane0, unsigned lane1, unsigned lane2, unsigned lane3);
 529
 530 LLVMValueRef
 531 ac_build_shuffle(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef index);
 532
 533 #ifdef __cplusplus
 534 }
 535 #endif
 536
 537 #endif