1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926
|
/*========================== begin_copyright_notice ============================
Copyright (C) 2017-2021 Intel Corporation
SPDX-License-Identifier: MIT
============================= end_copyright_notice ===========================*/
#include "Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.hpp"
#include "Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncsAnalysis.hpp"
#include "Compiler/IGCPassSupport.h"
#include "common/LLVMWarningsPush.hpp"
#include <llvmWrapper/IR/IRBuilder.h>
#include <llvm/IR/Function.h>
#include <llvm/IR/Instructions.h>
#include "common/LLVMWarningsPop.hpp"
#include "Probe/Assertion.h"
#include <llvmWrapper/Support/Alignment.h>
#include <llvmWrapper/IR/DerivedTypes.h>
using namespace llvm;
using namespace IGC;
// Register pass to igc-opt
#define PASS_FLAG "igc-wi-func-resolution"
#define PASS_DESCRIPTION "Resolves work item functions"
#define PASS_CFG_ONLY false
#define PASS_ANALYSIS false
IGC_INITIALIZE_PASS_BEGIN(WIFuncResolution, PASS_FLAG, PASS_DESCRIPTION, PASS_CFG_ONLY, PASS_ANALYSIS)
IGC_INITIALIZE_PASS_DEPENDENCY(MetaDataUtilsWrapper)
IGC_INITIALIZE_PASS_DEPENDENCY(CodeGenContextWrapper)
IGC_INITIALIZE_PASS_DEPENDENCY(GenXFunctionGroupAnalysis)
IGC_INITIALIZE_PASS_END(WIFuncResolution, PASS_FLAG, PASS_DESCRIPTION, PASS_CFG_ONLY, PASS_ANALYSIS)
char WIFuncResolution::ID = 0;
WIFuncResolution::WIFuncResolution() : FunctionPass(ID), m_implicitArgs()
{
initializeWIFuncResolutionPass(*PassRegistry::getPassRegistry());
}
void WIFuncResolution::storeImplicitBufferPtrs(llvm::Function& F)
{
CodeGenContext* m_ctx = getAnalysis<CodeGenContextWrapper>().getCodeGenContext();
if (isEntryFunc(m_pMdUtils, &F)
&& !m_ctx->platform.isProductChildOf(IGFX_XE_HP_SDV))
{
if (m_implicitArgs.isImplicitArgExist(ImplicitArg::ArgType::IMPLICIT_ARG_BUFFER_PTR))
{
IGCLLVM::IRBuilder<> Builder(&(*F.getEntryBlock().getFirstInsertionPt()));
auto BufferPtr = m_implicitArgs.getImplicitArgValue(F, ImplicitArg::ArgType::IMPLICIT_ARG_BUFFER_PTR, m_pMdUtils);
// create intrinsic to store implicit arg buffer ptr
auto* M = F.getParent();
llvm::SmallVector<llvm::Type*, 1> Type;
Type.push_back(BufferPtr->getType());
auto* ImplArgFunc = GenISAIntrinsic::getDeclaration(M, GenISAIntrinsic::GenISA_SetImplicitBufferPtr, Type);
llvm::SmallVector<llvm::Value*, 1> Args = { BufferPtr };
auto StoreIntrinsic = Builder.CreateCall(ImplArgFunc, Args);
StoreIntrinsic->setDebugLoc(DebugLoc());
auto& C = F.getParent()->getContext();
auto LocalIdX = m_implicitArgs.getImplicitArgValue(F, ImplicitArg::ArgType::LOCAL_ID_X, m_pMdUtils);
auto LocalIdY = m_implicitArgs.getImplicitArgValue(F, ImplicitArg::ArgType::LOCAL_ID_Y, m_pMdUtils);
auto LocalIdZ = m_implicitArgs.getImplicitArgValue(F, ImplicitArg::ArgType::LOCAL_ID_Z, m_pMdUtils);
auto DataTypeI16 = Type::getInt16Ty(C);
auto AllocaVec = Builder.CreateAlloca(DataTypeI16, ConstantInt::get(DataTypeI16, (uint64_t)3));
auto FirstSlot = Builder.CreatePointerCast(AllocaVec, DataTypeI16->getPointerTo());
Builder.CreateStore(LocalIdX, FirstSlot, true);
auto SecondSlot = Builder.CreatePtrToInt(FirstSlot, Type::getInt64Ty(C));
SecondSlot = Builder.CreateAdd(SecondSlot, ConstantInt::get(SecondSlot->getType(), (uint64_t)2));
SecondSlot = Builder.CreateIntToPtr(SecondSlot, DataTypeI16->getPointerTo());
Builder.CreateStore(LocalIdY, SecondSlot, true);
auto ThirdSlot = Builder.CreatePtrToInt(FirstSlot, Type::getInt64Ty(C));
ThirdSlot = Builder.CreateAdd(ThirdSlot, ConstantInt::get(ThirdSlot->getType(), (uint64_t)4));
ThirdSlot = Builder.CreateIntToPtr(ThirdSlot, DataTypeI16->getPointerTo());
Builder.CreateStore(LocalIdZ, ThirdSlot, true);
auto* LidFunc = GenISAIntrinsic::getDeclaration(M, GenISAIntrinsic::GenISA_SetLocalIdBufferPtr, AllocaVec->getType());
Args = { AllocaVec };
StoreIntrinsic = Builder.CreateCall(LidFunc, Args);
StoreIntrinsic->setDebugLoc(DebugLoc());
}
}
}
bool WIFuncResolution::runOnFunction(Function& F)
{
m_changed = false;
m_pMdUtils = getAnalysis<MetaDataUtilsWrapper>().getMetaDataUtils();
m_implicitArgs = ImplicitArgs(F, m_pMdUtils);
visit(F);
storeImplicitBufferPtrs(F);
return m_changed;
}
void WIFuncResolution::visitCallInst(CallInst& CI)
{
if (!CI.getCalledFunction())
{
return;
}
Value* wiRes = nullptr;
// Add appropriate sequence and handle out of range where needed
StringRef funcName = CI.getCalledFunction()->getName();
if (funcName.equals(WIFuncsAnalysis::GET_LOCAL_ID_X))
{
wiRes = getLocalId(CI, ImplicitArg::LOCAL_ID_X);
}
else if (funcName.equals(WIFuncsAnalysis::GET_LOCAL_ID_Y))
{
wiRes = getLocalId(CI, ImplicitArg::LOCAL_ID_Y);
}
else if (funcName.equals(WIFuncsAnalysis::GET_LOCAL_ID_Z))
{
wiRes = getLocalId(CI, ImplicitArg::LOCAL_ID_Z);
}
else if (funcName.equals(WIFuncsAnalysis::GET_GROUP_ID))
{
wiRes = getGroupId(CI);
}
else if (funcName.equals(WIFuncsAnalysis::GET_LOCAL_THREAD_ID))
{
wiRes = getLocalThreadId(CI);
}
else if (funcName.equals(WIFuncsAnalysis::GET_GLOBAL_SIZE))
{
wiRes = getGlobalSize(CI);
}
else if (funcName.equals(WIFuncsAnalysis::GET_LOCAL_SIZE))
{
wiRes = getLocalSize(CI);
}
else if (funcName.equals(WIFuncsAnalysis::GET_ENQUEUED_LOCAL_SIZE)) {
wiRes = getEnqueuedLocalSize(CI);
}
else if (funcName.equals(WIFuncsAnalysis::GET_GLOBAL_OFFSET))
{
wiRes = getGlobalOffset(CI);
}
else if (funcName.equals(WIFuncsAnalysis::GET_WORK_DIM))
{
wiRes = getWorkDim(CI);
}
else if (funcName.equals(WIFuncsAnalysis::GET_NUM_GROUPS))
{
wiRes = getNumGroups(CI);
}
else if (funcName.equals(WIFuncsAnalysis::GET_STAGE_IN_GRID_ORIGIN))
{
wiRes = getStageInGridOrigin(CI);
}
else if (funcName.equals(WIFuncsAnalysis::GET_STAGE_IN_GRID_SIZE))
{
wiRes = getStageInGridSize(CI);
}
else if (funcName.equals(WIFuncsAnalysis::GET_SYNC_BUFFER))
{
wiRes = getSyncBufferPtr(CI);
}
else
{
// Non WI function, do nothing
return;
}
// Handle size_t return type for 64 bits
if (wiRes && wiRes->getType()->getScalarSizeInBits() < CI.getType()->getScalarSizeInBits())
{
CastInst* pCast = CastInst::Create(Instruction::ZExt, wiRes, IntegerType::get(CI.getContext(), CI.getType()->getScalarSizeInBits()), wiRes->getName(), &CI);
updateDebugLoc(&CI, pCast);
wiRes = pCast;
}
// Replace original WI call instruction by the result of the appropriate sequence
if (wiRes) { CI.replaceAllUsesWith(wiRes); }
CI.eraseFromParent();
m_changed = true;
}
/************************************************************************************************
R0:
-----------------------------------------------------------------------------------------------
| Local mem | Group | Barrier ID| Sampler | Binding | Scratch | Group | Group |
| mem index/| number | /Interface| state | table | space | number | number |
| URB handle| X | descriptor| pointer | pointer | pointer | Y | Z |
| | 32bit | offset | | | | 32bit | 32bit |
-----------------------------------------------------------------------------------------------
<low> <high>
PayloadHeader:
-----------------------------------------------------------------------------------------------
| Global | Global | Global | Local | Local | Local | Reserved | Num |
| offset | offset | offset | size | size | size | | HW |
| X | Y | Z | X | Y | Z | | Threads |
| 32bit | 32bit | 32bit | 32bit | 32bit | 32bit | | 32bit |
-----------------------------------------------------------------------------------------------
<low> <high>
*************************************************************************************************/
// Structure of side buffer generated by NEO:
//struct implicit_args {
// uint8_t struct_size;
// uint8_t struct_version;
// uint8_t num_work_dim;
// uint8_t simd_width;
// uint32_t local_size_x;
// uint32_t local_size_y;
// uint32_t local_size_z;
// uint64_t global_size_x;
// uint64_t global_size_y;
// uint64_t global_size_z;
// uint64_t printf_buffer_ptr;
// uint64_t global_offset_x;
// uint64_t global_offset_y;
// uint64_t global_offset_z;
// uint64_t local_id_table_ptr;
// uint32_t group_count_x;
// uint32_t group_count_y;
// uint32_t group_count_z;
//};
// For SIMD8:
//struct local_id_s {
// uint16_t lx[8];
// uint16_t reserved[8];
// uint16_t ly[8];
// uint16_t reserved[8];
// uint16_t lz[8];
// uint16_t reserved[8];
//};
// For SIMD16:
//struct local_id_s {
// uint16_t lx[16];
// uint16_t ly[16];
// uint16_t lz[16];
//};
// For SIMD32:
//struct local_id_s {
// uint16_t lx[32];
// uint16_t ly[32];
// uint16_t lz[32];
//};
class GLOBAL_STATE_FIELD_OFFSETS
{
public:
// This class holds offsets of various fields in side buffer
static const uint32_t STRUCT_SIZE = 0;
static const uint32_t VERSION = STRUCT_SIZE + sizeof(uint8_t);
static const uint32_t NUM_WORK_DIM = VERSION + sizeof(uint8_t);
static const uint32_t SIMDSIZE = NUM_WORK_DIM + sizeof(uint8_t);
static const uint32_t LOCAL_SIZES = SIMDSIZE + sizeof(uint8_t);
static const uint32_t LOCAL_SIZE_X = LOCAL_SIZES;
static const uint32_t LOCAL_SIZE_Y = LOCAL_SIZE_X + sizeof(uint32_t);
static const uint32_t LOCAL_SIZE_Z = LOCAL_SIZE_Y + sizeof(uint32_t);
static const uint32_t GLOBAL_SIZES = LOCAL_SIZE_Z + sizeof(uint32_t);
static const uint32_t GLOBAL_SIZE_X = GLOBAL_SIZES;
static const uint32_t GLOBAL_SIZE_Y = GLOBAL_SIZE_X + sizeof(uint64_t);
static const uint32_t GLOBAL_SIZE_Z = GLOBAL_SIZE_Y + sizeof(uint64_t);
static const uint32_t PRINTF_BUFFER = GLOBAL_SIZE_Z + sizeof(uint64_t);
static const uint32_t GLOBAL_OFFSETS = PRINTF_BUFFER + sizeof(uint64_t);
static const uint32_t GLOBAL_OFFSET_X = GLOBAL_OFFSETS;
static const uint32_t GLOBAL_OFFSET_Y = GLOBAL_OFFSET_X + sizeof(uint64_t);
static const uint32_t GLOBAL_OFFSET_Z = GLOBAL_OFFSET_Y + sizeof(uint64_t);
static const uint32_t LOCAL_IDS = GLOBAL_OFFSET_Z + sizeof(uint64_t);
static const uint32_t GROUP_COUNTS = LOCAL_IDS + sizeof(uint64_t);
static const uint32_t GROUP_COUNT_X = GROUP_COUNTS;
static const uint32_t GROUP_COUNT_Y = GROUP_COUNT_X + sizeof(uint32_t);
static const uint32_t GROUP_COUNT_Z = GROUP_COUNT_Y + sizeof(uint32_t);
static const uint32_t TOTAL_SIZE = GROUP_COUNT_Z + sizeof(uint32_t);
};
llvm::Value* LowerImplicitArgIntrinsics::BuildLoadInst(llvm::CallInst& CI, unsigned int Offset, llvm::Type* DataType)
{
// This function computes type aligned address that includes Offset.
// Then it loads DataType number of elements from Offset.
// If Offset is unaligned then it computes aligned offset and loads data.
// If Offset is unaligned then it copies data to new vector of size <i8 x Size>,
// bitcasts it to DataType, and returns it.
// It Offset is aligned, it returns result of LoadInst of type DataType.
auto ElemByteSize = DataType->getScalarSizeInBits() / 8;
auto Size = ElemByteSize;
if (auto DataVecType = dyn_cast<IGCLLVM::FixedVectorType>(DataType))
{
Size *= (unsigned int)DataVecType->getNumElements();
}
unsigned int AlignedOffset = (Offset / ElemByteSize) * ElemByteSize;
unsigned int LoadByteSize = (Offset == AlignedOffset) ? Size : Size * 2;
IGCLLVM::IRBuilder<> Builder(&CI);
unsigned int AddrSpace = ADDRESS_SPACE_GLOBAL;
if (m_ctx->platform.isProductChildOf(IGFX_XE_HP_SDV))
{
AddrSpace = ADDRESS_SPACE_THREAD_ARG;
}
llvm::Value* LoadedData = nullptr;
auto F = CI.getFunction();
auto Int32Ptr = PointerType::get(Type::getInt32Ty(F->getParent()->getContext()), AddrSpace);
auto ElemType = DataType->getScalarType();
auto IntToPtr = Builder.CreateIntToPtr(Builder.getIntN(F->getParent()->getDataLayout().getPointerSizeInBits(AddrSpace), AlignedOffset), Int32Ptr);
if (AddrSpace == ADDRESS_SPACE_GLOBAL)
{
// Add base ptr to AlignedOffset
auto* M = CI.getModule();
SmallVector<Type*, 1> Args = { Int32Ptr };
auto* Func = GenISAIntrinsic::getDeclaration(M, GenISAIntrinsic::GenISA_GetImplicitBufferPtr, Args);
auto* LoadIntrinsic = Builder.CreateCall(Func);
auto PtrBitSize = F->getParent()->getDataLayout().getPointerSizeInBits(AddrSpace);
// Now extract 8 bytes of data beginning at offset 0
auto* DataTypePtrSize = Type::getIntNTy(F->getParent()->getContext(), PtrBitSize);
auto* PtrToInt = Builder.CreatePtrToInt(LoadIntrinsic, DataTypePtrSize);
auto* AddInst = Builder.CreateAdd(PtrToInt, llvm::ConstantInt::get(DataTypePtrSize, AlignedOffset));
IntToPtr = Builder.CreateIntToPtr(AddInst, Int32Ptr);
}
auto LoadType = IGCLLVM::FixedVectorType::get(ElemType, LoadByteSize / ElemByteSize);
auto PtrType = PointerType::get(LoadType, AddrSpace);
auto BitCast = Builder.CreateBitCast(IntToPtr, PtrType);
auto LoadInst = Builder.CreateLoad(BitCast);
LoadInst->setAlignment(IGCLLVM::getCorrectAlign(ElemByteSize));
LoadedData = LoadInst;
if (Offset != AlignedOffset)
{
auto ByteType = Type::getInt8Ty(Builder.getContext());
auto BitCastToByte = Builder.CreateBitCast(LoadedData, ByteType);
Value* NewVector = UndefValue::get(IGCLLVM::FixedVectorType::get(ByteType, Size));
for (unsigned int I = Offset; I != (Offset + Size); ++I)
{
auto Elem = Builder.CreateExtractElement(BitCastToByte, I - AlignedOffset);
NewVector = Builder.CreateInsertElement(NewVector, Elem, (uint64_t)I - (uint64_t)Offset);
}
auto Result = Builder.CreateBitCast(NewVector, DataType);
return Result;
}
auto Result = Builder.CreateBitCast(LoadedData, DataType);
return Result;
}
Value* WIFuncResolution::getLocalId(CallInst& CI, ImplicitArg::ArgType argType)
{
// Receives:
// call i32 @__builtin_IB_get_local_id_x()
// Creates:
// %localIdX
auto F = CI.getFunction();
Value* V = m_implicitArgs.getImplicitArgValue(*F, argType, m_pMdUtils);
IGC_ASSERT(V);
return V;
}
Value* WIFuncResolution::getGroupId(CallInst& CI)
{
// Receives:
// call i32 @__builtin_IB_get_group_id(i32 %dim)
// Creates:
// %cmpDim = icmp eq i32 %dim, 0
// %tmpOffsetR0 = select i1 %cmpDim, i32 1, i32 5
// %offsetR0 = add i32 %dim, %tmpOffsetR0
// %groupId = extractelement <8 x i32> %r0, i32 %offsetR0
// The cmp select insts are present because:
// if dim = 0 then we need to access R0.1
// if dim = 1 then we need to access R0.6
// if dim = 2 then we need to access R0.7
Value* V = nullptr;
auto F = CI.getParent()->getParent();
V = m_implicitArgs.getImplicitArgValue(*F, ImplicitArg::R0, m_pMdUtils);
Value* dim = CI.getArgOperand(0);
Instruction* cmpDim = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_EQ, dim, ConstantInt::get(Type::getInt32Ty(CI.getContext()), 0), "cmpDim", &CI);
Instruction* offsetR0 = SelectInst::Create(cmpDim, ConstantInt::get(Type::getInt32Ty(CI.getContext()), 1), ConstantInt::get(Type::getInt32Ty(CI.getContext()), 5), "tmpOffsetR0", &CI);
Instruction* index = BinaryOperator::CreateAdd(dim, offsetR0, "offsetR0", &CI);
Instruction* groupId = ExtractElementInst::Create(V, index, "groupId", &CI);
updateDebugLoc(&CI, cmpDim);
updateDebugLoc(&CI, offsetR0);
updateDebugLoc(&CI, index);
updateDebugLoc(&CI, groupId);
return groupId;
}
Value* WIFuncResolution::getLocalThreadId(CallInst &CI)
{
// Receives:
// call spir_func i32 @__builtin_IB_get_local_thread_id()
// Creates:
// %r0second = extractelement <8 x i32> %r0, i32 2
// %localThreadId = trunc i32 %r0second to i8
// we need to access R0.2 bits 0 to 7, which contain HW local thread ID on XeHP_SDV+
Value* V = nullptr;
auto F = CI.getParent()->getParent();
V = m_implicitArgs.getImplicitArgValue(*F, ImplicitArg::R0, m_pMdUtils);
Instruction* r0second = ExtractElementInst::Create(V, ConstantInt::get(Type::getInt32Ty(CI.getContext()), 2), "r0second", &CI);
Instruction* localThreadId = TruncInst::Create(Instruction::CastOps::Trunc, r0second, Type::getInt8Ty(CI.getContext()), "localThreadId", &CI);
updateDebugLoc(&CI, r0second);
updateDebugLoc(&CI, localThreadId);
return localThreadId;
}
Value* WIFuncResolution::getGlobalSize(CallInst& CI)
{
// Receives:
// call i32 @__builtin_IB_get_global_size(i32 %dim)
// Creates:
// %globalSize1 = extractelement <3 x i32> %globalSize, i32 %dim
auto F = CI.getFunction();
Value* V = m_implicitArgs.getImplicitArgValue(*F, ImplicitArg::GLOBAL_SIZE, m_pMdUtils);
IGC_ASSERT(V != nullptr);
Value* dim = CI.getArgOperand(0);
Instruction* globalSize = ExtractElementInst::Create(V, dim, "globalSize", &CI);
updateDebugLoc(&CI, globalSize);
return globalSize;
}
Value* WIFuncResolution::getLocalSize(CallInst& CI)
{
// Receives:
// call i32 @__builtin_IB_get_local_size(i32 %dim)
// Creates:
// %localSize = extractelement <3 x i32> %localSize, i32 %dim
auto F = CI.getFunction();
Value* V = m_implicitArgs.getImplicitArgValue(*F, ImplicitArg::LOCAL_SIZE, m_pMdUtils);
IGC_ASSERT(V != nullptr);
Value* dim = CI.getArgOperand(0);
Instruction* localSize = ExtractElementInst::Create(V, dim, "localSize", &CI);
updateDebugLoc(&CI, localSize);
return localSize;
}
Value* WIFuncResolution::getEnqueuedLocalSize(CallInst& CI) {
// Receives:
// call i32 @__builtin_IB_get_enqueued_local_size(i32 %dim)
// Creates:
// %enqueuedLocalSize1 = extractelement <3 x i32> %enqueuedLocalSize, %dim
auto F = CI.getFunction();
Value* V = m_implicitArgs.getImplicitArgValue(*F, ImplicitArg::ENQUEUED_LOCAL_WORK_SIZE, m_pMdUtils);
IGC_ASSERT(V != nullptr);
Value* dim = CI.getArgOperand(0);
Instruction* enqueuedLocalSize = ExtractElementInst::Create(V, dim, "enqueuedLocalSize", &CI);
updateDebugLoc(&CI, enqueuedLocalSize);
return enqueuedLocalSize;
}
Value* WIFuncResolution::getGlobalOffset(CallInst& CI)
{
// Receives:
// call i32 @__builtin_IB_get_global_offset(i32 %dim)
// Creates:
// %globalOffset = extractelement <8 x i32> %payloadHeader, i32 %dim
auto F = CI.getFunction();
Value* V = m_implicitArgs.getImplicitArgValue(*F, ImplicitArg::PAYLOAD_HEADER, m_pMdUtils);
IGC_ASSERT(V != nullptr);
Value* dim = CI.getArgOperand(0);
auto globalOffset = ExtractElementInst::Create(V, dim, "globalOffset", &CI);
updateDebugLoc(&CI, cast<Instruction>(globalOffset));
return globalOffset;
}
Value* WIFuncResolution::getWorkDim(CallInst& CI)
{
// Receives:
// call i32 @__builtin_IB_get_work_dim()
// Creates:
// %workDim
auto F = CI.getFunction();
Value* V = m_implicitArgs.getImplicitArgValue(*F, ImplicitArg::WORK_DIM, m_pMdUtils);
IGC_ASSERT(V != nullptr);
return V;
}
Value* WIFuncResolution::getNumGroups(CallInst& CI)
{
// Receives:
// call i32 @__builtin_IB_get_num_groups(i32 %dim)
// Creates:
// %numGroups1 = extractelement <3 x i32> %numGroups, i32 %dim
auto F = CI.getFunction();
Value* V = m_implicitArgs.getImplicitArgValue(*F, ImplicitArg::NUM_GROUPS, m_pMdUtils);
IGC_ASSERT(V != nullptr);
Value* dim = CI.getArgOperand(0);
Instruction* numGroups = ExtractElementInst::Create(V, dim, "numGroups", &CI);
updateDebugLoc(&CI, numGroups);
return numGroups;
}
Value* WIFuncResolution::getStageInGridOrigin(CallInst& CI)
{
// Receives:
// call i32 @__builtin_IB_get_grid_origin(i32 %dim)
// Creates:
// %grid_origin1 = extractelement <3 x i32> %globalSize, i32 %dim
auto F = CI.getParent()->getParent();
Value* V = m_implicitArgs.getImplicitArgValue(*F, ImplicitArg::STAGE_IN_GRID_ORIGIN, m_pMdUtils);
IGC_ASSERT(V != nullptr);
Value* dim = CI.getArgOperand(0);
Instruction* globalSize = ExtractElementInst::Create(V, dim, "grid_origin", &CI);
updateDebugLoc(&CI, globalSize);
return globalSize;
}
Value* WIFuncResolution::getStageInGridSize(CallInst& CI)
{
// Receives:
// call i32 @__builtin_IB_get_grid_size(i32 %dim)
// Creates:
// %grid_size1 = extractelement <3 x i32> %globalSize, i32 %dim
auto F = CI.getFunction();
Value* V = m_implicitArgs.getImplicitArgValue(*F, ImplicitArg::STAGE_IN_GRID_SIZE, m_pMdUtils);
IGC_ASSERT(V != nullptr);
Value* dim = CI.getArgOperand(0);
Instruction* globalSize = ExtractElementInst::Create(V, dim, "grid_size", &CI);
updateDebugLoc(&CI, globalSize);
return globalSize;
}
Value* WIFuncResolution::getSyncBufferPtr(CallInst& CI)
{
// Receives:
// call i8 addrspace(1)* @__builtin_IB_get_sync_buffer()
// Creates:
// i8 addrspace(1)* %syncBuffer
auto F = CI.getParent()->getParent();
Value* syncBuffer = m_implicitArgs.getImplicitArgValue(*F, ImplicitArg::SYNC_BUFFER, m_pMdUtils);
return syncBuffer;
}
// Register pass to igc-opt
#define PASS_FLAG2 "igc-lower-implicit-arg-intrinsic"
#define PASS_DESCRIPTION2 "igc-lower-implicit-arg-intrinsic"
#define PASS_CFG_ONLY2 false
#define PASS_ANALYSIS2 false
IGC_INITIALIZE_PASS_BEGIN(LowerImplicitArgIntrinsics, PASS_FLAG2, PASS_DESCRIPTION2, PASS_CFG_ONLY2, PASS_ANALYSIS2)
IGC_INITIALIZE_PASS_DEPENDENCY(MetaDataUtilsWrapper)
IGC_INITIALIZE_PASS_DEPENDENCY(CodeGenContextWrapper)
IGC_INITIALIZE_PASS_END(LowerImplicitArgIntrinsics, PASS_FLAG2, PASS_DESCRIPTION2, PASS_CFG_ONLY2, PASS_ANALYSIS2)
char LowerImplicitArgIntrinsics::ID = 0;
LowerImplicitArgIntrinsics::LowerImplicitArgIntrinsics() : FunctionPass(ID)
{
initializeLowerImplicitArgIntrinsicsPass(*PassRegistry::getPassRegistry());
}
Constant* getKnownWorkGroupSize(IGCMD::MetaDataUtils* MDUtils, llvm::Function& F)
{
auto Dims = IGCMD::IGCMetaDataHelper::getThreadGroupDims(*MDUtils, &F);
if (!Dims)
return nullptr;
return ConstantDataVector::get(F.getContext(), *Dims);
}
bool LowerImplicitArgIntrinsics::runOnFunction(Function& F)
{
m_FGA = getAnalysisIfAvailable<GenXFunctionGroupAnalysis>();
m_ctx = getAnalysis<CodeGenContextWrapper>().getCodeGenContext();
usedIntrinsicsMap.clear();
visit(F);
/// If the work group size is known at compile time, emit it as a
/// literal rather than reading from the payload.
auto MDUtils = getAnalysis<MetaDataUtilsWrapper>().getMetaDataUtils();
if (Constant* KnownWorkGroupSize = getKnownWorkGroupSize(MDUtils, F))
{
ImplicitArgs IAS(F, MDUtils);
if (auto* V = IAS.getImplicitArgValue(F, ImplicitArg::ENQUEUED_LOCAL_WORK_SIZE, MDUtils))
V->replaceAllUsesWith(KnownWorkGroupSize);
}
return false;
}
llvm::Value* LowerImplicitArgIntrinsics::getIntrinsicCall(llvm::Function* F, llvm::GenISAIntrinsic::ID IntrinsicID, llvm::Type* DataType)
{
auto entry = usedIntrinsicsMap.find(IntrinsicID);
if (entry != usedIntrinsicsMap.end())
return entry->second;
for (auto II = inst_begin(F), IE = inst_end(F); II != IE; II++)
{
if (GenIntrinsicInst* inst = dyn_cast<GenIntrinsicInst>(&*II))
{
// if intrinsic call exists - use it
if (inst->getIntrinsicID() == IntrinsicID)
{
// Make sure that intrinsic that we use is called before we use it's result
auto parentFunction = inst->getParent()->getParent();
auto firstFunc = parentFunction->getEntryBlock().getFirstNonPHI();
inst->moveBefore(firstFunc);
return inst;
}
}
}
// if intrinsic call doesn't exist - create it
auto getFunctionDeclaration = GenISAIntrinsic::getDeclaration(F->getParent(), IntrinsicID, DataType);
llvm::IRBuilder<> Builder(&*F->getEntryBlock().begin());
auto callValue = Builder.CreateCall(getFunctionDeclaration);
usedIntrinsicsMap.insert(std::pair(IntrinsicID, callValue));
return callValue;
}
void LowerImplicitArgIntrinsics::visitCallInst(CallInst& CI)
{
Function* F = CI.getParent()->getParent();
auto MDUtils = getAnalysis<MetaDataUtilsWrapper>().getMetaDataUtils();
// Not a GenISAIntrinsic
GenIntrinsicInst* inst = dyn_cast<GenIntrinsicInst>(&CI);
if (!inst) return;
// Not a valid implicit arg intrinsic
auto ID = inst->getIntrinsicID();
ImplicitArg::ArgType argTy = ImplicitArgs::getArgType(ID);
if (argTy == ImplicitArg::ArgType::NUM_IMPLICIT_ARGS) return;
// If the intrinsic no longer have a use, just remove it
if (inst->use_empty())
{
CI.eraseFromParent();
return;
}
// Lower intrinsic usage in the kernel to kernel args
if (isEntryFunc(MDUtils, F))
{
ImplicitArgs IAS(*F, MDUtils);
Argument* Arg = IAS.getImplicitArg(*F, argTy);
if (Arg)
{
CI.replaceAllUsesWith(Arg);
CI.eraseFromParent();
}
return;
}
// Load from implicit arg buffer for intrinsic usage in stackcall
bool LoadFromImplicitArgBuffer = F->hasFnAttribute("visaStackCall");
// If the current function is a subroutine, but the caller is a stackcall, we
// still need to use the implicit arg buffer.
if (!LoadFromImplicitArgBuffer)
{
if (m_FGA)
{
Function* subGroupMapHead = m_FGA->getSubGroupMap(F);
if (subGroupMapHead && subGroupMapHead->hasFnAttribute("visaStackCall"))
LoadFromImplicitArgBuffer = true;
}
}
if (LoadFromImplicitArgBuffer)
{
Value* V = nullptr;
IGCLLVM::IRBuilder<> Builder(&CI);
switch (ID)
{
case GenISAIntrinsic::GenISA_getLocalID_X:
case GenISAIntrinsic::GenISA_getLocalID_Y:
case GenISAIntrinsic::GenISA_getLocalID_Z:
{
// Get SIMD lane id
llvm::Value* SimdLaneId = getIntrinsicCall(F, GenISAIntrinsic::ID::GenISA_simdLaneId, Type::getInt16Ty(F->getParent()->getContext()));
llvm::Value* Result = nullptr;
if (!m_ctx->platform.isProductChildOf(IGFX_XE_HP_SDV))
{
// Get local id buffer base ptr
auto Int32Ptr = PointerType::get(Type::getInt32Ty(F->getParent()->getContext()), ADDRESS_SPACE_GLOBAL);
auto* M = CI.getModule();
SmallVector<Type*, 1> Args = { Int32Ptr };
auto* Func = GenISAIntrinsic::getDeclaration(M, GenISAIntrinsic::GenISA_GetLocalIdBufferPtr, Args);
auto* LocalIdPtr = Builder.CreateCall(Func);
auto PtrBitSize = F->getParent()->getDataLayout().getPointerSizeInBits(ADDRESS_SPACE_GLOBAL);
auto* DataTypePtrSize = Type::getIntNTy(F->getParent()->getContext(), PtrBitSize);
auto* LocalIdBase = Builder.CreatePtrToInt(LocalIdPtr, DataTypePtrSize);
SimdLaneId = Builder.CreateMul(SimdLaneId, ConstantInt::get(SimdLaneId->getType(), (uint64_t)6));
if (argTy == ImplicitArg::ArgType::LOCAL_ID_X)
{
}
else if (argTy == ImplicitArg::ArgType::LOCAL_ID_Y)
{
SimdLaneId = Builder.CreateAdd(SimdLaneId, ConstantInt::get(SimdLaneId->getType(), (uint64_t)2));
}
else if (argTy == ImplicitArg::ArgType::LOCAL_ID_Z)
{
SimdLaneId = Builder.CreateAdd(SimdLaneId, ConstantInt::get(SimdLaneId->getType(), (uint64_t)4));
}
SimdLaneId = Builder.CreateZExt(SimdLaneId, LocalIdBase->getType());
Result = Builder.CreateAdd(SimdLaneId, LocalIdBase);
}
else
{
// LocalIDBase = oword_ld
// LocalThreadId = r0.2
// ThreadBaseOffset = LocalIDBase + LocalThreadId * (SimdSize * 3 * 2)
// BaseOffset_X = ThreadBaseOffset + 0 * (SimdSize * 2) + (SimdLaneId * 2) OR
// BaseOffset_Y = ThreadBaseOffset + 1 * (SimdSize * 2) + (SimdLaneId * 2) OR
// BaseOffset_Z = ThreadBaseOffset + 2 * (SimdSize * 2) + (SimdLaneId * 2)
// Load from BaseOffset_[X|Y|Z]
// Get SIMD Size
llvm::Value* SimdSize = getIntrinsicCall(F, GenISAIntrinsic::ID::GenISA_simdSize, Type::getInt32Ty(F->getParent()->getContext()));
// SimdSize = max(SimdSize, 16)
auto CmpInst = Builder.CreateICmpSGT(SimdSize, ConstantInt::get(SimdSize->getType(), (uint64_t)16));
SimdSize = Builder.CreateSelect(CmpInst, SimdSize, ConstantInt::get(SimdSize->getType(), (uint64_t)16));
// Get Local ID Base Ptr
auto DataTypeI64 = Type::getInt64Ty(F->getParent()->getContext());
unsigned int Offset = GLOBAL_STATE_FIELD_OFFSETS::LOCAL_IDS;
auto LocalIDBase = BuildLoadInst(CI, Offset, DataTypeI64);
// Get local thread id
ImplicitArgs IAS(*F, MDUtils);
auto R0Val = IAS.getImplicitArgValue(*F, ImplicitArg::R0, MDUtils);
auto LocalThreadId = Builder.CreateExtractElement(R0Val, ConstantInt::get(Type::getInt32Ty(CI.getContext()), 2));
LocalThreadId = Builder.CreateAnd(LocalThreadId, (uint16_t)255);
// Compute thread base offset where local ids for current thread are stored
// ThreadBaseOffset = LocalIDBasePtr + LocalThreadId * (simd size * 3 * 2)
auto ThreadBaseOffset = Builder.CreateMul(SimdSize, ConstantInt::get(SimdSize->getType(), (uint64_t)6));
ThreadBaseOffset = Builder.CreateMul(Builder.CreateZExt(ThreadBaseOffset, LocalThreadId->getType()), LocalThreadId);
ThreadBaseOffset = Builder.CreateAdd(Builder.CreateZExt(ThreadBaseOffset, LocalIDBase->getType()), LocalIDBase);
// Compute offset per lane
uint8_t Factor = 0;
if (argTy == ImplicitArg::ArgType::LOCAL_ID_Y)
{
Factor = 2;
}
else if (argTy == ImplicitArg::ArgType::LOCAL_ID_Z)
{
Factor = 4;
}
// Compute Factor*(simd size) * 2 to arrive at base of local id for current thread
auto Expr1 = Builder.CreateMul(SimdSize, ConstantInt::get(SimdSize->getType(), Factor));
// Compute offset to current lane
auto Expr2 = Builder.CreateMul(SimdLaneId, ConstantInt::get(SimdLaneId->getType(), 2));
Result = Builder.CreateAdd(Builder.CreateZExt(Expr1, LocalIDBase->getType()),
Builder.CreateZExt(Expr2, LocalIDBase->getType()));
Result = Builder.CreateAdd(Result, ThreadBaseOffset);
}
// Load data
auto Int16Ptr = Type::getInt16PtrTy(F->getContext(), ADDRESS_SPACE_GLOBAL);
auto Addr = Builder.CreateIntToPtr(Result, Int16Ptr);
auto LoadInst = Builder.CreateLoad(Addr);
auto Trunc = Builder.CreateZExtOrBitCast(LoadInst, CI.getType());
V = Trunc;
break;
}
case GenISAIntrinsic::GenISA_getLocalSize:
case GenISAIntrinsic::GenISA_getEnqueuedLocalSize:
{
// Assume local size and enqueued local size are the same
auto ElemTypeD = Type::getInt32Ty(F->getParent()->getContext());
auto VecTyD = IGCLLVM::FixedVectorType::get(ElemTypeD, 3);
unsigned int Offset = GLOBAL_STATE_FIELD_OFFSETS::LOCAL_SIZE_X;
auto LoadInst = BuildLoadInst(CI, Offset, VecTyD);
V = LoadInst;
break;
}
case GenISAIntrinsic::GenISA_getPayloadHeader:
{
// global_offset is loaded from PayloadHeader[0:2]
// currently there are no other uses for payload header.
unsigned int Offset = GLOBAL_STATE_FIELD_OFFSETS::GLOBAL_OFFSET_X;
auto ElemTypeD = Type::getInt32Ty(F->getParent()->getContext());
auto VecTyQ = IGCLLVM::FixedVectorType::get(Type::getInt64Ty(F->getParent()->getContext()), 3);
auto LoadInst = BuildLoadInst(CI, Offset, VecTyQ);
Value* Undef = UndefValue::get(CI.getType());
for (auto I = 0; I != 3; I++)
{
auto Elem = Builder.CreateExtractElement(LoadInst, (uint64_t)I);
auto TruncElem = Builder.CreateTrunc(Elem, ElemTypeD);
Undef = Builder.CreateInsertElement(Undef, TruncElem, (uint64_t)I);
}
V = Undef;
break;
}
case GenISAIntrinsic::GenISA_getGlobalSize:
case GenISAIntrinsic::GenISA_getStageInGridSize:
{
unsigned int Offset = GLOBAL_STATE_FIELD_OFFSETS::GLOBAL_SIZE_X;
auto VecTyQ = IGCLLVM::FixedVectorType::get(Type::getInt64Ty(F->getParent()->getContext()), 3);
auto ElemTypeD = Type::getInt32Ty(F->getParent()->getContext());
auto LoadInst = BuildLoadInst(CI, Offset, VecTyQ);
Value* Undef = UndefValue::get(CI.getType());
for (auto I = 0; I != 3; I++)
{
auto Elem = Builder.CreateExtractElement(LoadInst, (uint64_t)I);
auto TruncElem = Builder.CreateTrunc(Elem, ElemTypeD);
Undef = Builder.CreateInsertElement(Undef, TruncElem, (uint64_t)I);
}
V = Undef;
break;
}
case GenISAIntrinsic::GenISA_getNumWorkGroups:
{
auto ElemTypeUD = Type::getInt32Ty(F->getParent()->getContext());
auto VecTyUD = IGCLLVM::FixedVectorType::get(ElemTypeUD, 3);
unsigned int Offset = GLOBAL_STATE_FIELD_OFFSETS::GROUP_COUNT_X;
auto LoadInst = BuildLoadInst(CI, Offset, VecTyUD);
V = LoadInst;
break;
}
case GenISAIntrinsic::GenISA_getWorkDim:
{
unsigned int Size = 4;
unsigned int Offset = GLOBAL_STATE_FIELD_OFFSETS::NUM_WORK_DIM / Size;
auto TypeUD = Type::getInt32Ty(F->getParent()->getContext());
auto LoadInst = BuildLoadInst(CI, Offset, TypeUD);
auto LShr = Builder.CreateLShr(LoadInst, (uint64_t)16);
auto And = Builder.CreateAnd(LShr, (uint16_t)255);
V = And;
break;
}
case GenISAIntrinsic::GenISA_getPrintfBuffer:
{
// This function is invoked when expanding printf call to retrieve printf buffer ptr.
auto DataTypeI64 = Type::getInt64Ty(CI.getFunction()->getParent()->getContext());
unsigned int Offset = GLOBAL_STATE_FIELD_OFFSETS::PRINTF_BUFFER;
auto Result = BuildLoadInst(CI, Offset, DataTypeI64);
Result = Builder.CreateIntToPtr(Result, CI.getType());
V = Result;
break;
}
default:
break;
}
if (V != nullptr)
{
CI.replaceAllUsesWith(V);
CI.eraseFromParent();
}
}
}
|