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 #define HAVE_32BIT_POINTERS (HAVE_LLVM >= 0x0700)
  38
  39 enum {
  40         /* CONST is the only address space that selects SMEM loads */
  41         AC_CONST_ADDR_SPACE = HAVE_LLVM >= 0x700 ? 4 : 2,
  42         AC_LOCAL_ADDR_SPACE = 3,
  43         AC_CONST_32BIT_ADDR_SPACE = 6, /* same as CONST, but the pointer type has 32 bits */
  44 };
  45
  46 struct ac_llvm_flow;
  47
  48 struct ac_llvm_context {
  49         LLVMContextRef context;
  50         LLVMModuleRef module;
  51         LLVMBuilderRef builder;
  52
  53         LLVMTypeRef voidt;
  54         LLVMTypeRef i1;
  55         LLVMTypeRef i8;
  56         LLVMTypeRef i16;
  57         LLVMTypeRef i32;
  58         LLVMTypeRef i64;
  59         LLVMTypeRef intptr;
  60         LLVMTypeRef f16;
  61         LLVMTypeRef f32;
  62         LLVMTypeRef f64;
  63         LLVMTypeRef v2i16;
  64         LLVMTypeRef v2i32;
  65         LLVMTypeRef v3i32;
  66         LLVMTypeRef v4i32;
  67         LLVMTypeRef v2f32;
  68         LLVMTypeRef v4f32;
  69         LLVMTypeRef v8i32;
  70
  71         LLVMValueRef i32_0;
  72         LLVMValueRef i32_1;
  73         LLVMValueRef i64_0;
  74         LLVMValueRef i64_1;
  75         LLVMValueRef f32_0;
  76         LLVMValueRef f32_1;
  77         LLVMValueRef f64_0;
  78         LLVMValueRef f64_1;
  79         LLVMValueRef i1true;
  80         LLVMValueRef i1false;
  81
  82         struct ac_llvm_flow *flow;
  83         unsigned flow_depth;
  84         unsigned flow_depth_max;
  85
  86         unsigned range_md_kind;
  87         unsigned invariant_load_md_kind;
  88         unsigned uniform_md_kind;
  89         unsigned fpmath_md_kind;
  90         LLVMValueRef fpmath_md_2p5_ulp;
  91         LLVMValueRef empty_md;
  92
  93         enum chip_class chip_class;
  94         enum radeon_family family;
  95
  96         LLVMValueRef lds;
  97 };
  98
  99 void
 100 ac_llvm_context_init(struct ac_llvm_context *ctx,
 101                      enum chip_class chip_class, enum radeon_family family);
 102
 103 void
 104 ac_llvm_context_dispose(struct ac_llvm_context *ctx);
 105
 106 int
 107 ac_get_llvm_num_components(LLVMValueRef value);
 108
 109 int
 110 ac_get_elem_bits(struct ac_llvm_context *ctx, LLVMTypeRef type);
 111
 112 LLVMValueRef
 113 ac_llvm_extract_elem(struct ac_llvm_context *ac,
 114                      LLVMValueRef value,
 115                      int index);
 116
 117 unsigned ac_get_type_size(LLVMTypeRef type);
 118
 119 LLVMTypeRef ac_to_integer_type(struct ac_llvm_context *ctx, LLVMTypeRef t);
 120 LLVMValueRef ac_to_integer(struct ac_llvm_context *ctx, LLVMValueRef v);
 121 LLVMTypeRef ac_to_float_type(struct ac_llvm_context *ctx, LLVMTypeRef t);
 122 LLVMValueRef ac_to_float(struct ac_llvm_context *ctx, LLVMValueRef v);
 123
 124 LLVMValueRef
 125 ac_build_intrinsic(struct ac_llvm_context *ctx, const char *name,
 126                    LLVMTypeRef return_type, LLVMValueRef *params,
 127                    unsigned param_count, unsigned attrib_mask);
 128
 129 void ac_build_type_name_for_intr(LLVMTypeRef type, char *buf, unsigned bufsize);
 130
 131 LLVMValueRef
 132 ac_build_phi(struct ac_llvm_context *ctx, LLVMTypeRef type,
 133              unsigned count_incoming, LLVMValueRef *values,
 134              LLVMBasicBlockRef *blocks);
 135
 136 void ac_build_s_barrier(struct ac_llvm_context *ctx);
 137 void ac_build_optimization_barrier(struct ac_llvm_context *ctx,
 138                                    LLVMValueRef *pvgpr);
 139
 140 LLVMValueRef ac_build_shader_clock(struct ac_llvm_context *ctx);
 141
 142 LLVMValueRef ac_build_ballot(struct ac_llvm_context *ctx, LLVMValueRef value);
 143
 144 LLVMValueRef ac_build_vote_all(struct ac_llvm_context *ctx, LLVMValueRef value);
 145
 146 LLVMValueRef ac_build_vote_any(struct ac_llvm_context *ctx, LLVMValueRef value);
 147
 148 LLVMValueRef ac_build_vote_eq(struct ac_llvm_context *ctx, LLVMValueRef value);
 149
 150 LLVMValueRef
 151 ac_build_varying_gather_values(struct ac_llvm_context *ctx, LLVMValueRef *values,
 152                                unsigned value_count, unsigned component);
 153
 154 LLVMValueRef
 155 ac_build_gather_values_extended(struct ac_llvm_context *ctx,
 156                                 LLVMValueRef *values,
 157                                 unsigned value_count,
 158                                 unsigned value_stride,
 159                                 bool load,
 160                                 bool always_vector);
 161 LLVMValueRef
 162 ac_build_gather_values(struct ac_llvm_context *ctx,
 163                        LLVMValueRef *values,
 164                        unsigned value_count);
 165 LLVMValueRef ac_build_expand_to_vec4(struct ac_llvm_context *ctx,
 166                                      LLVMValueRef value,
 167                                      unsigned num_channels);
 168
 169 LLVMValueRef
 170 ac_build_fdiv(struct ac_llvm_context *ctx,
 171               LLVMValueRef num,
 172               LLVMValueRef den);
 173
 174 void
 175 ac_prepare_cube_coords(struct ac_llvm_context *ctx,
 176                        bool is_deriv, bool is_array, bool is_lod,
 177                        LLVMValueRef *coords_arg,
 178                        LLVMValueRef *derivs_arg);
 179
 180
 181 LLVMValueRef
 182 ac_build_fs_interp(struct ac_llvm_context *ctx,
 183                    LLVMValueRef llvm_chan,
 184                    LLVMValueRef attr_number,
 185                    LLVMValueRef params,
 186                    LLVMValueRef i,
 187                    LLVMValueRef j);
 188
 189 LLVMValueRef
 190 ac_build_fs_interp_mov(struct ac_llvm_context *ctx,
 191                        LLVMValueRef parameter,
 192                        LLVMValueRef llvm_chan,
 193                        LLVMValueRef attr_number,
 194                        LLVMValueRef params);
 195
 196 LLVMValueRef
 197 ac_build_gep0(struct ac_llvm_context *ctx,
 198               LLVMValueRef base_ptr,
 199               LLVMValueRef index);
 200
 201 void
 202 ac_build_indexed_store(struct ac_llvm_context *ctx,
 203                        LLVMValueRef base_ptr, LLVMValueRef index,
 204                        LLVMValueRef value);
 205
 206 LLVMValueRef ac_build_load(struct ac_llvm_context *ctx, LLVMValueRef base_ptr,
 207                            LLVMValueRef index);
 208 LLVMValueRef ac_build_load_invariant(struct ac_llvm_context *ctx,
 209                                      LLVMValueRef base_ptr, LLVMValueRef index);
 210 LLVMValueRef ac_build_load_to_sgpr(struct ac_llvm_context *ctx,
 211                                    LLVMValueRef base_ptr, LLVMValueRef index);
 212
 213 void
 214 ac_build_buffer_store_dword(struct ac_llvm_context *ctx,
 215                             LLVMValueRef rsrc,
 216                             LLVMValueRef vdata,
 217                             unsigned num_channels,
 218                             LLVMValueRef voffset,
 219                             LLVMValueRef soffset,
 220                             unsigned inst_offset,
 221                             bool glc,
 222                             bool slc,
 223                             bool writeonly_memory,
 224                             bool swizzle_enable_hint);
 225 LLVMValueRef
 226 ac_build_buffer_load(struct ac_llvm_context *ctx,
 227                      LLVMValueRef rsrc,
 228                      int num_channels,
 229                      LLVMValueRef vindex,
 230                      LLVMValueRef voffset,
 231                      LLVMValueRef soffset,
 232                      unsigned inst_offset,
 233                      unsigned glc,
 234                      unsigned slc,
 235                      bool can_speculate,
 236                      bool allow_smem);
 237
 238 LLVMValueRef ac_build_buffer_load_format(struct ac_llvm_context *ctx,
 239                                          LLVMValueRef rsrc,
 240                                          LLVMValueRef vindex,
 241                                          LLVMValueRef voffset,
 242                                          unsigned num_channels,
 243                                          bool glc,
 244                                          bool can_speculate);
 245
 246 /* load_format that handles the stride & element count better if idxen is
 247  * disabled by LLVM. */
 248 LLVMValueRef ac_build_buffer_load_format_gfx9_safe(struct ac_llvm_context *ctx,
 249                                                   LLVMValueRef rsrc,
 250                                                   LLVMValueRef vindex,
 251                                                   LLVMValueRef voffset,
 252                                                   unsigned num_channels,
 253                                                   bool glc,
 254                                                   bool can_speculate);
 255
 256 LLVMValueRef
 257 ac_build_tbuffer_load_short(struct ac_llvm_context *ctx,
 258                             LLVMValueRef rsrc,
 259                             LLVMValueRef vindex,
 260                             LLVMValueRef voffset,
 261                                 LLVMValueRef soffset,
 262                                 LLVMValueRef immoffset);
 263
 264 LLVMValueRef
 265 ac_get_thread_id(struct ac_llvm_context *ctx);
 266
 267 #define AC_TID_MASK_TOP_LEFT 0xfffffffc
 268 #define AC_TID_MASK_TOP      0xfffffffd
 269 #define AC_TID_MASK_LEFT     0xfffffffe
 270
 271 LLVMValueRef
 272 ac_build_ddxy(struct ac_llvm_context *ctx,
 273               uint32_t mask,
 274               int idx,
 275               LLVMValueRef val);
 276
 277 #define AC_SENDMSG_GS 2
 278 #define AC_SENDMSG_GS_DONE 3
 279
 280 #define AC_SENDMSG_GS_OP_NOP      (0 << 4)
 281 #define AC_SENDMSG_GS_OP_CUT      (1 << 4)
 282 #define AC_SENDMSG_GS_OP_EMIT     (2 << 4)
 283 #define AC_SENDMSG_GS_OP_EMIT_CUT (3 << 4)
 284
 285 void ac_build_sendmsg(struct ac_llvm_context *ctx,
 286                       uint32_t msg,
 287                       LLVMValueRef wave_id);
 288
 289 LLVMValueRef ac_build_imsb(struct ac_llvm_context *ctx,
 290                            LLVMValueRef arg,
 291                            LLVMTypeRef dst_type);
 292
 293 LLVMValueRef ac_build_umsb(struct ac_llvm_context *ctx,
 294                           LLVMValueRef arg,
 295                           LLVMTypeRef dst_type);
 296 LLVMValueRef ac_build_fmin(struct ac_llvm_context *ctx, LLVMValueRef a,
 297                            LLVMValueRef b);
 298 LLVMValueRef ac_build_fmax(struct ac_llvm_context *ctx, LLVMValueRef a,
 299                            LLVMValueRef b);
 300 LLVMValueRef ac_build_imin(struct ac_llvm_context *ctx, LLVMValueRef a,
 301                            LLVMValueRef b);
 302 LLVMValueRef ac_build_imax(struct ac_llvm_context *ctx, LLVMValueRef a,
 303                            LLVMValueRef b);
 304 LLVMValueRef ac_build_umin(struct ac_llvm_context *ctx, LLVMValueRef a, LLVMValueRef b);
 305 LLVMValueRef ac_build_clamp(struct ac_llvm_context *ctx, LLVMValueRef value);
 306
 307 struct ac_export_args {
 308         LLVMValueRef out[4];
 309         unsigned target;
 310         unsigned enabled_channels;
 311         bool compr;
 312         bool done;
 313         bool valid_mask;
 314 };
 315
 316 void ac_build_export(struct ac_llvm_context *ctx, struct ac_export_args *a);
 317
 318 void ac_build_export_null(struct ac_llvm_context *ctx);
 319
 320 enum ac_image_opcode {
 321         ac_image_sample,
 322         ac_image_gather4,
 323         ac_image_load,
 324         ac_image_load_mip,
 325         ac_image_store,
 326         ac_image_store_mip,
 327         ac_image_get_lod,
 328         ac_image_get_resinfo,
 329         ac_image_atomic,
 330         ac_image_atomic_cmpswap,
 331 };
 332
 333 enum ac_atomic_op {
 334         ac_atomic_swap,
 335         ac_atomic_add,
 336         ac_atomic_sub,
 337         ac_atomic_smin,
 338         ac_atomic_umin,
 339         ac_atomic_smax,
 340         ac_atomic_umax,
 341         ac_atomic_and,
 342         ac_atomic_or,
 343         ac_atomic_xor,
 344 };
 345
 346 enum ac_image_dim {
 347         ac_image_1d,
 348         ac_image_2d,
 349         ac_image_3d,
 350         ac_image_cube, // includes cube arrays
 351         ac_image_1darray,
 352         ac_image_2darray,
 353         ac_image_2dmsaa,
 354         ac_image_2darraymsaa,
 355 };
 356
 357 /* These cache policy bits match the definitions used by the LLVM intrinsics. */
 358 enum ac_image_cache_policy {
 359         ac_glc = 1 << 0,
 360         ac_slc = 1 << 1,
 361 };
 362
 363 struct ac_image_args {
 364         enum ac_image_opcode opcode : 4;
 365         enum ac_atomic_op atomic : 4; /* for the ac_image_atomic opcode */
 366         enum ac_image_dim dim : 3;
 367         unsigned dmask : 4;
 368         unsigned cache_policy : 2;
 369         bool unorm : 1;
 370         bool level_zero : 1;
 371         unsigned attributes; /* additional call-site specific AC_FUNC_ATTRs */
 372
 373         LLVMValueRef resource;
 374         LLVMValueRef sampler;
 375         LLVMValueRef data[2]; /* data[0] is source data (vector); data[1] is cmp for cmpswap */
 376         LLVMValueRef offset;
 377         LLVMValueRef bias;
 378         LLVMValueRef compare;
 379         LLVMValueRef derivs[6];
 380         LLVMValueRef coords[4];
 381         LLVMValueRef lod; // also used by ac_image_get_resinfo
 382 };
 383
 384 LLVMValueRef ac_build_image_opcode(struct ac_llvm_context *ctx,
 385                                    struct ac_image_args *a);
 386 LLVMValueRef ac_build_cvt_pkrtz_f16(struct ac_llvm_context *ctx,
 387                                     LLVMValueRef args[2]);
 388 LLVMValueRef ac_build_cvt_pknorm_i16(struct ac_llvm_context *ctx,
 389                                      LLVMValueRef args[2]);
 390 LLVMValueRef ac_build_cvt_pknorm_u16(struct ac_llvm_context *ctx,
 391                                      LLVMValueRef args[2]);
 392 LLVMValueRef ac_build_cvt_pk_i16(struct ac_llvm_context *ctx,
 393                                  LLVMValueRef args[2], unsigned bits, bool hi);
 394 LLVMValueRef ac_build_cvt_pk_u16(struct ac_llvm_context *ctx,
 395                                  LLVMValueRef args[2], unsigned bits, bool hi);
 396 LLVMValueRef ac_build_wqm_vote(struct ac_llvm_context *ctx, LLVMValueRef i1);
 397 void ac_build_kill_if_false(struct ac_llvm_context *ctx, LLVMValueRef i1);
 398 LLVMValueRef ac_build_bfe(struct ac_llvm_context *ctx, LLVMValueRef input,
 399                           LLVMValueRef offset, LLVMValueRef width,
 400                           bool is_signed);
 401
 402 void ac_build_waitcnt(struct ac_llvm_context *ctx, unsigned simm16);
 403
 404 LLVMValueRef ac_build_fract(struct ac_llvm_context *ctx, LLVMValueRef src0,
 405                            unsigned bitsize);
 406
 407 LLVMValueRef ac_build_isign(struct ac_llvm_context *ctx, LLVMValueRef src0,
 408                             unsigned bitsize);
 409
 410 LLVMValueRef ac_build_fsign(struct ac_llvm_context *ctx, LLVMValueRef src0,
 411                             unsigned bitsize);
 412
 413 void ac_optimize_vs_outputs(struct ac_llvm_context *ac,
 414                             LLVMValueRef main_fn,
 415                             uint8_t *vs_output_param_offset,
 416                             uint32_t num_outputs,
 417                             uint8_t *num_param_exports);
 418 void ac_init_exec_full_mask(struct ac_llvm_context *ctx);
 419
 420 void ac_declare_lds_as_pointer(struct ac_llvm_context *ac);
 421 LLVMValueRef ac_lds_load(struct ac_llvm_context *ctx,
 422                          LLVMValueRef dw_addr);
 423 void ac_lds_store(struct ac_llvm_context *ctx,
 424                   LLVMValueRef dw_addr, LLVMValueRef value);
 425
 426 LLVMValueRef ac_find_lsb(struct ac_llvm_context *ctx,
 427                          LLVMTypeRef dst_type,
 428                          LLVMValueRef src0);
 429
 430 LLVMTypeRef ac_array_in_const_addr_space(LLVMTypeRef elem_type);
 431 LLVMTypeRef ac_array_in_const32_addr_space(LLVMTypeRef elem_type);
 432
 433 void ac_build_bgnloop(struct ac_llvm_context *ctx, int lable_id);
 434 void ac_build_break(struct ac_llvm_context *ctx);
 435 void ac_build_continue(struct ac_llvm_context *ctx);
 436 void ac_build_else(struct ac_llvm_context *ctx, int lable_id);
 437 void ac_build_endif(struct ac_llvm_context *ctx, int lable_id);
 438 void ac_build_endloop(struct ac_llvm_context *ctx, int lable_id);
 439 void ac_build_if(struct ac_llvm_context *ctx, LLVMValueRef value,
 440                  int lable_id);
 441 void ac_build_uif(struct ac_llvm_context *ctx, LLVMValueRef value,
 442                   int lable_id);
 443
 444 LLVMValueRef ac_build_alloca(struct ac_llvm_context *ac, LLVMTypeRef type,
 445                              const char *name);
 446 LLVMValueRef ac_build_alloca_undef(struct ac_llvm_context *ac, LLVMTypeRef type,
 447                                    const char *name);
 448
 449 LLVMValueRef ac_cast_ptr(struct ac_llvm_context *ctx, LLVMValueRef ptr,
 450                          LLVMTypeRef type);
 451
 452 LLVMValueRef ac_trim_vector(struct ac_llvm_context *ctx, LLVMValueRef value,
 453                             unsigned count);
 454
 455 LLVMValueRef ac_unpack_param(struct ac_llvm_context *ctx, LLVMValueRef param,
 456                              unsigned rshift, unsigned bitwidth);
 457
 458 void ac_apply_fmask_to_sample(struct ac_llvm_context *ac, LLVMValueRef fmask,
 459                               LLVMValueRef *addr, bool is_array_tex);
 460
 461 LLVMValueRef
 462 ac_build_ds_swizzle(struct ac_llvm_context *ctx, LLVMValueRef src, unsigned mask);
 463
 464 LLVMValueRef
 465 ac_build_readlane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef lane);
 466
 467 LLVMValueRef
 468 ac_build_writelane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef value, LLVMValueRef lane);
 469
 470 LLVMValueRef
 471 ac_build_mbcnt(struct ac_llvm_context *ctx, LLVMValueRef mask);
 472
 473 LLVMValueRef
 474 ac_build_inclusive_scan(struct ac_llvm_context *ctx, LLVMValueRef src, nir_op op);
 475
 476 LLVMValueRef
 477 ac_build_exclusive_scan(struct ac_llvm_context *ctx, LLVMValueRef src, nir_op op);
 478
 479 LLVMValueRef
 480 ac_build_reduce(struct ac_llvm_context *ctx, LLVMValueRef src, nir_op op, unsigned cluster_size);
 481
 482 LLVMValueRef
 483 ac_build_quad_swizzle(struct ac_llvm_context *ctx, LLVMValueRef src,
 484                 unsigned lane0, unsigned lane1, unsigned lane2, unsigned lane3);
 485
 486 LLVMValueRef
 487 ac_build_shuffle(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef index);
 488
 489 #ifdef __cplusplus
 490 }
 491 #endif
 492
 493 #endif