intel-graphics-compiler/IGC/Compiler/CISACodeGen/helper.h

/*========================== begin_copyright_notice ============================

Copyright (C) 2017-2024 Intel Corporation

SPDX-License-Identifier: MIT

============================= end_copyright_notice ===========================*/

#pragma once

#include "common/LLVMWarningsPush.hpp"
#include <llvm/Analysis/AssumptionCache.h>
#include <llvm/Analysis/TargetLibraryInfo.h>
#include <llvm/Analysis/MemorySSAUpdater.h>
#include <llvm/IR/Instruction.h>
#include <llvm/IR/Instructions.h>
#include <llvm/IR/Function.h>
#include <llvm/IR/Intrinsics.h>
#include <llvm/IR/IntrinsicInst.h>
#include <llvm/IR/Constants.h>
#include <llvm/IR/DataLayout.h>
#include <llvm/IR/Metadata.h>
#include <llvm/IR/Operator.h>
#include <llvm/IR/CFG.h>
#include <llvmWrapper/IR/IRBuilder.h>
#include <llvm/IR/PassManager.h>
#include <llvm/ADT/SmallSet.h>
#include <llvm/ADT/SmallVector.h>
#include <llvm/ADT/DenseSet.h>
#include <llvm/Analysis/PostDominators.h>
#include "common/LLVMWarningsPop.hpp"
#include "GenISAIntrinsics/GenIntrinsics.h"
#include "GenISAIntrinsics/GenIntrinsicInst.h"
#include "Compiler/CodeGenPublicEnums.h"
#include "Compiler/CISACodeGen/Platform.hpp"
#include "Compiler/MetaDataApi/MetaDataApi.h"
#include "common/MDFrameWork.h"
#include "common/Types.hpp"
#include "common/igc_resourceDimTypes.h"
#include "LLVM3DBuilder/MetadataBuilder.h"
#include "Probe/Assertion.h"
#include <functional>
#include "common/ResourceAddrSpace.h"

typedef unsigned int uint;

#define SIZE_WORD 2
#define SIZE_DWORD 4
#define SIZE_OWORD 16
#define SIZE_YWORD 32

namespace IGC {
class CodeGenContext;
struct SProgramOutput;

static const char *const INTEL_SYMBOL_TABLE_VOID_PROGRAM = "Intel_Symbol_Table_Void_Program";

#ifdef _DEBUG
template <typename T, size_t N> using smallvector = std::vector<T>;
#else
template <typename T, size_t N> using smallvector = llvm::SmallVector<T, N>;
#endif

// This is used to return true/false/dunno results.
enum class Tristate { Unknown = -1, False = 0, True = 1 };

enum e_llvmType {
  e_Instruction = 0,
  e_Intrinsic = 1,
  e_GenISAIntrinsic = 1,
};
#define LLVMTYPEBYTE 24

#define OPCODE(instName, llvmType) instName | llvmType << LLVMTYPEBYTE

#define DECLARE_OPCODE(instName, llvmType, name, modifiers, sat, pred, condMod, mathIntrinsic, atomicIntrinsic,        \
                       regioning)                                                                                      \
  name = OPCODE(llvm::llvmType::instName, e_##llvmType),
enum EOPCODE {
#include "opCode.h"
};
#undef DECLARE_OPCODE

#define DECLARE_OPCODE(instName, llvmType, name, modifiers, sat, pred, condMod, mathIntrinsic, atomicIntrinsic,        \
                       regioning)                                                                                      \
  static_assert((llvm::llvmType::instName < (1 << LLVMTYPEBYTE)), "Enum bitfield range check");
#include "opCode.h"
#undef DECLARE_OPCODE

EOPCODE GetOpCode(const llvm::Instruction *inst);
bool SupportsModifier(llvm::Instruction *inst, const CPlatform &platform);
bool SupportsSaturate(llvm::Instruction *inst);
bool SupportsPredicate(llvm::Instruction *inst);
bool SupportsCondModifier(llvm::Instruction *inst);
bool SupportsRegioning(llvm::Instruction *inst);
bool IsMathIntrinsic(EOPCODE opcode);
bool IsAtomicIntrinsic(EOPCODE opcode);
bool IsGradientIntrinsic(EOPCODE opcode);
bool IsExtendedMathInstruction(llvm::Instruction *Inst);
bool IsSubGroupIntrinsicWithSimd32Implementation(EOPCODE opcode);
bool UsesTypedConstantBuffer(const CodeGenContext *pContext, const BufferType bufType);

bool ComputesGradient(llvm::Instruction *inst);

uint getImmValueU32(const llvm::Value *value);
bool getImmValueBool(const llvm::Value *value);

template <typename EnumT> static inline EnumT getImmValueEnum(const llvm::Value *val) {
  return static_cast<EnumT>(getImmValueU32(val));
}

void VectorToElement(llvm::Value *inst, llvm::Value *elem[], llvm::Type *int32Ty, llvm::Instruction *insert_before,
                     int vsize = 4);
llvm::Value *ElementToVector(llvm::Value *elem[], llvm::Type *int32Ty, llvm::Instruction *insert_before, int vsize = 4);

llvm::Value *ConvertToFloat(llvm::IRBuilder<> &builder, llvm::Value *val);
void ConvertToFloat(llvm::IRBuilder<> &builder, llvm::SmallVectorImpl<llvm::Value *> &instList);
// scalarize aggregate into flattened members
void ScalarizeAggregateMembers(llvm::IRBuilder<> &builder, llvm::Value *val,
                               llvm::SmallVectorImpl<llvm::Value *> &instList);

/// return true if pLLVMInst is load from constant-buffer with immediate constant-buffer index
bool IsLoadFromDirectCB(llvm::Instruction *pLLVMInst, uint &cbId, llvm::Value *&eltPtrVal);
bool IsReadOnlyLoadDirectCB(llvm::Instruction *pLLVMInst, uint &cbId, llvm::Value *&eltPtrVal, BufferType &buftype);

int findSampleInstructionTextureIdx(llvm::Instruction *inst);
llvm::Value *getTextureIndexArgBasedOnOpcode(llvm::Instruction *inst);
llvm::Value *GetBufferOperand(llvm::Instruction *inst);
void GetResourceOperand(llvm::Instruction *inst, llvm::Value *&resValue, llvm::Value *&pairTexValue,
                        llvm::Value *&texValue, llvm::Value *&sampleValue);
void SetResourceOperand(llvm::Instruction *inst, llvm::Value *newResourceOp, llvm::Value *newPairTexureOp,
                        llvm::Value *newTextureOp, llvm::Value *newSamplerOp);
void setBufOperand(llvm::Instruction *inst, llvm::Value *newOp);

llvm::LoadInst *cloneLoad(llvm::LoadInst *Orig, llvm::Type *Ty, llvm::Value *Ptr);
llvm::StoreInst *cloneStore(llvm::StoreInst *Orig, llvm::Value *Val, llvm::Value *Ptr);

llvm::LdRawIntrinsic *CreateLoadRawIntrinsic(llvm::LoadInst *inst, llvm::Value *bufPtr, llvm::Value *offsetVal);
llvm::StoreRawIntrinsic *CreateStoreRawIntrinsic(llvm::StoreInst *inst, llvm::Value *bufPtr, llvm::Value *offsetVal);

void getTextureAndSamplerOperands(llvm::GenIntrinsicInst *pIntr, llvm::Value *&pTextureValue,
                                  llvm::Value *&pSamplerValue);
void getTextureAndSamplerOperands(llvm::GenIntrinsicInst *pIntr, llvm::Value *&pPairedTextureValue,
                                  llvm::Value *&pTextureValue, llvm::Value *&pSamplerValue);
void ChangePtrTypeInIntrinsic(llvm::GenIntrinsicInst *&pIntr, llvm::Value *oldPtr, llvm::Value *newPtr);

llvm::Value *TracePointerSource(llvm::Value *resourcePtr);
llvm::Value *TracePointerSource(llvm::Value *resourcePtr, bool hasBranching, bool enablePhiLoops, bool fillList,
                                std::vector<llvm::Value *> &instList);
llvm::Value *TracePointerSource(llvm::Value *resourcePtr, bool hasBranching, bool enablePhiLoops, bool fillList,
                                std::vector<llvm::Value *> &instList, llvm::SmallSet<llvm::PHINode *, 8> &visitedPHIs);
bool GetResourcePointerInfo(llvm::Value *srcPtr, unsigned &resID, IGC::BufferType &resTy,
                            IGC::BufferAccessType &accessTy, bool &needBufferOffset);
BufferAccessType getDefaultAccessType(BufferType bufTy);
bool GetGRFOffsetFromRTV(llvm::Value *pointerSrc, unsigned &GRFOffset);
bool GetGRFOffsetFromGlobalRootSignatureValue(llvm::Value *pointerSrc, unsigned &GRFOffset);
bool GetStatelessBufferInfo(llvm::Value *pointer, unsigned &bufIdOrGRFOffset, IGC::BufferType &bufferTy,
                            llvm::Value *&bufferSrcPtr, bool &isDirectBuf);
// try to evaluate the address if it is constant.
bool EvalConstantAddress(llvm::Value *address, unsigned int &offset, const llvm::DataLayout *pDL,
                         llvm::Value *ptrSrc = nullptr);
bool getConstantAddress(llvm::Instruction &I, ConstantAddress &cl, CodeGenContext *pContext, bool &directBuf,
                        bool &statelessBuf, bool &bindlessBuf, bool &rootconstantBuf, unsigned int &TableOffset);

bool isSampleLoadGather4InfoInstruction(const llvm::Instruction *inst);
bool isSampleInstruction(const llvm::Instruction *inst);
bool isInfoInstruction(const llvm::Instruction *inst);
bool isLdInstruction(const llvm::Instruction *inst);
bool isGather4Instruction(const llvm::Instruction *inst);

bool IsMediaIOIntrinsic(const llvm::Instruction *inst);
bool IsSIMDBlockIntrinsic(const llvm::Instruction *inst);
bool isSubGroupIntrinsic(const llvm::Instruction *I);
bool isSubGroupIntrinsicPVC(const llvm::Instruction *I);
bool isSubGroupShuffleVariant(const llvm::Instruction *I);
bool subgroupIntrinsicHasHelperLanes(const llvm::Instruction &I);
bool hasSubGroupIntrinsicPVC(llvm::Function &F);

bool isBarrierIntrinsic(const llvm::Instruction *I);

bool isUserFunctionCall(const llvm::Instruction *I);

bool isHidingComplexControlFlow(const llvm::Instruction *I);

bool IsMemLoadIntrinsic(llvm::GenISAIntrinsic::ID id);

bool IsStatelessMemLoadIntrinsic(llvm::GenISAIntrinsic::ID id);
bool IsStatelessMemStoreIntrinsic(llvm::GenISAIntrinsic::ID id);
bool IsStatelessMemAtomicIntrinsic(llvm::GenIntrinsicInst &inst, llvm::GenISAIntrinsic::ID id);
llvm::GenISAIntrinsic::ID GetOutputPSIntrinsic(const CPlatform &platform);

bool isURBWriteIntrinsic(const llvm::Instruction *inst);

llvm::Instruction *AdjustSystemValueCall(llvm::GenIntrinsicInst *inst);

int getConstantBufferLoadOffset(llvm::LoadInst *ld);

unsigned getNumberOfExitBlocks(llvm::Function &function);

bool isDummyBasicBlock(llvm::BasicBlock *BB);

bool isNaNCheck(llvm::FCmpInst &FC);

inline bool IsBindfull(BufferType t) { return t == UAV || t == CONSTANT_BUFFER || t == RESOURCE; }
inline bool IsBindless(BufferType t) {
  return t == BINDLESS || t == BINDLESS_CONSTANT_BUFFER || t == BINDLESS_READONLY || t == BINDLESS_WRITEONLY ||
         t == BINDLESS_TEXTURE;
}
inline bool IsSSHbindless(BufferType t) {
  return t == SSH_BINDLESS || t == SSH_BINDLESS_CONSTANT_BUFFER || t == SSH_BINDLESS_TEXTURE;
}
inline bool IsStatelessBuffer(BufferType t) { return t == STATELESS || t == STATELESS_READONLY || t == STATELESS_A32; }
inline bool IsTypedBuffer(BufferType t) { return t == RESOURCE || t == BINDLESS_TEXTURE || t == SSH_BINDLESS_TEXTURE; }
inline bool IsUntypedBuffer(BufferType t) {
  return t == UAV || t == CONSTANT_BUFFER || t == BINDLESS || t == BINDLESS_CONSTANT_BUFFER || t == BINDLESS_READONLY ||
         t == BINDLESS_WRITEONLY || t == SSH_BINDLESS || t == SSH_BINDLESS_CONSTANT_BUFFER;
}
inline bool IsWritableBuffer(BufferType t) {
  BufferAccessType accessType = getDefaultAccessType(t);
  return accessType == BufferAccessType::ACCESS_WRITE || accessType == BufferAccessType::ACCESS_READWRITE;
}
inline bool IsReadableBuffer(BufferType t) {
  BufferAccessType accessType = getDefaultAccessType(t);
  return accessType == BufferAccessType::ACCESS_READ || accessType == BufferAccessType::ACCESS_READWRITE;
}

bool IsUnsignedCmp(const llvm::CmpInst::Predicate Pred);
bool IsSignedCmp(const llvm::CmpInst::Predicate Pred);

bool IsBitCastForLifetimeMark(const llvm::Value *V);

ERoundingMode GetRoundingMode_FPCvtInt(ModuleMetaData *modMD, llvm::Instruction *pInst);
ERoundingMode GetRoundingMode_FP(ModuleMetaData *modMD, llvm::Instruction *inst);

// Return true if inst needs specific rounding mode; false otherwise.
bool setsRMExplicitly(llvm::Instruction *inst);

// returns true if the instruction does not care about the rounding mode settings
bool ignoresRoundingMode(llvm::Instruction *inst);

// isA64Ptr - Queries whether given pointer type requires 64-bit representation in vISA
bool isA64Ptr(llvm::PointerType *PT, CodeGenContext *pContext);

// Returns the default dummy kernel to which all symbols are attached
inline llvm::Function *getIntelSymbolTableVoidProgram(llvm::Module *pM, int SimdSz = 0) {
  // Get the default kernel if no Simd Size is specified
  if (SimdSz == 0)
    return pM->getFunction(INTEL_SYMBOL_TABLE_VOID_PROGRAM);
  else {
    IGC_ASSERT(SimdSz == 8 || SimdSz == 16 || SimdSz == 32);
    // SIMD variants of the dummy kernel are created in GenXCodeGenModule, get the variant if it exists
    std::string fName = std::string(INTEL_SYMBOL_TABLE_VOID_PROGRAM) + "_GenXSIMD" + std::to_string(SimdSz);
    return pM->getFunction(fName.c_str());
  }
}

// Check if the current function is a dummy kernel
// Note, the module can contain multiple dummy kernels to support SIMD variants.
// This function returns true if the current function is any of those variant kernels.
inline bool isIntelSymbolTableVoidProgram(llvm::Function *pF) {
  return (pF && pF->getName().startswith(INTEL_SYMBOL_TABLE_VOID_PROGRAM));
}

int getFunctionControl(const CodeGenContext *pContext);

inline bool ForceAlwaysInline(const CodeGenContext *pContext) {
  // return true if FunctionControl is set to INLINE, and SelectiveFunctionControl does not force fcalls.
  return getFunctionControl(pContext) == FLAG_FCALL_FORCE_INLINE && IGC_GET_FLAG_VALUE(SelectiveFunctionControl) == 0;
}

// Strips the clone postfix added by GenXCodeGenModule in the function name
inline std::string StripCloneName(std::string name) {
  auto found = name.rfind("_GenXClone");
  if (found != std::string::npos) {
    return name.substr(0, found);
  }
  return name;
}

inline bool isOptDisabledForModule(ModuleMetaData *modMD, llvm::StringRef optStr) {
  if (modMD->m_OptsToDisable.count(optStr.str()) != 0) {
    return true;
  }
  return false;
}

inline bool isOptDisabledForFunction(ModuleMetaData *modMD, llvm::StringRef optStr, llvm::Function *F) {
  // Search function metadata to check if pass needs to be disabled
  auto funcIt = modMD->FuncMD.find(F);
  if (funcIt != modMD->FuncMD.end()) {
    if (funcIt->second.m_OptsToDisablePerFunc.count(optStr.str()) != 0) {
      return true;
    }
  }
  return false;
}

void InsertOptsMetadata(CodeGenContext *pCtx, llvm::Function *F = nullptr);

/// Return true if F is an entry function of a kernel or a shader.
///    A entry function must have an entry in FunctionInfoMetaData
///       with type KernelFunction;
///    A non-entry function may have an entry, if so, that entry in
///       FunctionInfoMetaData must have type UserFunction.
inline bool isEntryFunc(const IGCMD::MetaDataUtils *pM, const llvm::Function *CF) {
  llvm::Function *F = const_cast<llvm::Function *>(CF);
  if (F == nullptr || F->empty() || pM->findFunctionsInfoItem(F) == pM->end_FunctionsInfo())
    return false;

  IGCMD::FunctionInfoMetaDataHandle Info = pM->getFunctionsInfoItem(F);
  IGC_ASSERT_MESSAGE(Info->isTypeHasValue(), "FunctionInfoMetaData missing type!");
  return Info->getType() == FunctionTypeMD::KernelFunction;
}

inline bool isPixelShaderPhaseFunction(const llvm::Function *CF) {
  const llvm::Module *M = CF->getParent();
  static const char *const phases[] = {NAMED_METADATA_COARSE_PHASE, NAMED_METADATA_PIXEL_PHASE};
  for (auto phase : phases) {
    if (auto MD = M->getNamedMetadata(phase)) {
      if (MD->getOperand(0) && MD->getOperand(0)->getOperand(0)) {
        auto Func = llvm::mdconst::dyn_extract<llvm::Function>(MD->getOperand(0)->getOperand(0));
        if (Func == CF)
          return true;
      }
    }
  }
  return false;
}

inline bool isCoarsePhaseFunction(const llvm::Function *CF) {
  const llvm::Module *M = CF->getParent();
  if (auto MD = M->getNamedMetadata(NAMED_METADATA_COARSE_PHASE)) {
    if (MD->getOperand(0) && MD->getOperand(0)->getOperand(0)) {
      auto Func = llvm::mdconst::dyn_extract<llvm::Function>(MD->getOperand(0)->getOperand(0));
      return Func == CF;
    }
  }
  return false;
}

inline bool isPixelPhaseFunction(const llvm::Function *CF) {
  const llvm::Module *M = CF->getParent();
  if (auto MD = M->getNamedMetadata(NAMED_METADATA_PIXEL_PHASE)) {
    if (MD->getOperand(0) && MD->getOperand(0)->getOperand(0)) {
      auto Func = llvm::mdconst::dyn_extract<llvm::Function>(MD->getOperand(0)->getOperand(0));
      return Func == CF;
    }
  }
  return false;
}

inline bool isNonEntryMultirateShader(const llvm::Function *CF) {
  if (isPixelPhaseFunction(CF)) {
    const llvm::Module *CM = CF->getParent();
    if (auto MD = CM->getNamedMetadata(NAMED_METADATA_COARSE_PHASE)) {
      if (MD->getOperand(0) && MD->getOperand(0)->getOperand(0)) {
        auto Func = llvm::mdconst::dyn_extract<llvm::Function>(MD->getOperand(0)->getOperand(0));
        return Func != nullptr;
      }
    }
  }
  return false;
}

// Return a unique entry function.
// If more than one entry exists, return the first and and set it as unique.
// All subsequent calls to this function will get the entry set by the first call.
llvm::Function *getUniqueEntryFunc(const IGCMD::MetaDataUtils *pM, IGC::ModuleMetaData *pModMD);

// Returns a SIMD size for given function from metadata.
// Returns 0 if function is not in metadata or function has not defined SIMD size.
int getSIMDSize(const IGCMD::MetaDataUtils *M, llvm::Function *F);

template <typename T> inline bool RTWriteHasSource0Alpha(const T *rtWrite, ModuleMetaData *md) {
  return (nullptr != rtWrite->getSource0Alpha()) && !llvm::isa<llvm::UndefValue>(rtWrite->getSource0Alpha());
}

template <typename T> inline bool DoesRTWriteSrc0AlphaBelongToHomogeneousPart(const T *rtWrite, ModuleMetaData *md) {
  return !rtWrite->hasMask() && RTWriteHasSource0Alpha(rtWrite, md);
}

inline bool
VectorUsedByConstExtractOnly(llvm::Value *val,
                             llvm::SmallVector<llvm::SmallVector<llvm::ExtractElementInst *, 1>, 4> &extracts) {
  for (auto UI = val->user_begin(), UE = val->user_end(); UI != UE; ++UI) {
    llvm::ExtractElementInst *ei = llvm::dyn_cast<llvm::ExtractElementInst>(*UI);
    if (!ei) {
      return false;
    } else {
      llvm::ConstantInt *idxv = llvm::dyn_cast<llvm::ConstantInt>(ei->getIndexOperand());
      if (!idxv) {
        return false;
      }
      uint idx = (uint)idxv->getZExtValue();
      extracts[idx].push_back(ei);
    }
  }
  return true;
}

inline bool
LoadUsedByConstExtractOnly(llvm::LoadInst *ld,
                           llvm::SmallVector<llvm::SmallVector<llvm::ExtractElementInst *, 1>, 4> &extracts) {
  return VectorUsedByConstExtractOnly(ld, extracts);
}

llvm::Value *mutatePtrType(llvm::Value *ptrv, llvm::PointerType *newType, llvm::IRBuilder<> &builder,
                           const llvm::Twine &name = "");

unsigned int AppendConservativeRastWAHeader(IGC::SProgramOutput *program, SIMDMode simdmode);
unsigned int AppendConservativeRastWAHeader(void *&pBinary, unsigned int &binarySize, SIMDMode simdmode);


/// \brief Check whether inst precedes given position in one basic block
inline bool isInstPrecede(const llvm::Instruction *inst, const llvm::Instruction *pos) {
  // must within same basic block
  IGC_ASSERT(inst->getParent() == pos->getParent());
  if (inst == pos) {
    return true;
  }

  auto II = inst->getParent()->begin();
  for (; &*II != inst && &*II != pos; ++II)
    ;
  return &*II == inst;
}

// If true, the codegen will not emit any code for this instruction
// (So dst and src are aliased to each other.)
bool isNoOpInst(llvm::Instruction *I, CodeGenContext *Ctx);

// CxtI is the instruction at which V is checked whether
// it is positive or not.
bool valueIsPositive(llvm::Value *V, const llvm::DataLayout *DL, llvm::AssumptionCache *AC = nullptr,
                     llvm::Instruction *CxtI = nullptr);

inline float GetThreadOccupancyPerSubslice(SIMDMode simdMode, unsigned threadGroupSize, unsigned hwThreadPerSubslice,
                                           unsigned slmSize, unsigned slmSizePerSubSlice) {
  unsigned simdWidth = 8;

  switch (simdMode) {
  case SIMDMode::SIMD8:
    simdWidth = 8;
    break;
  case SIMDMode::SIMD16:
    simdWidth = 16;
    break;
  case SIMDMode::SIMD32:
    simdWidth = 32;
    break;
  default:
    IGC_ASSERT_MESSAGE(0, "Invalid SIMD mode");
    break;
  }

  IGC_ASSERT(simdWidth);
  const unsigned nThreadsPerTG = (threadGroupSize + simdWidth - 1) / simdWidth;
  IGC_ASSERT(nThreadsPerTG);
  const unsigned TGPerSubsliceNoSLM = hwThreadPerSubslice / nThreadsPerTG;
  const unsigned nTGDispatch =
      (slmSize == 0) ? TGPerSubsliceNoSLM : std::min(TGPerSubsliceNoSLM, slmSizePerSubSlice / slmSize);
  IGC_ASSERT(float(hwThreadPerSubslice));
  const float occupancy = float(nTGDispatch * nThreadsPerTG) / float(hwThreadPerSubslice);
  return occupancy;
}

// Duplicate of the LLVM function in llvm/Transforms/Utils/ModuleUtils.h
// Global can now be any pointer type that uses addrspace
void appendToUsed(llvm::Module &M, llvm::ArrayRef<llvm::GlobalValue *> Values);

void setupTriple(CodeGenContext &Ctx, llvm::StringRef OS = "");

bool safeScheduleUp(llvm::BasicBlock *BB, llvm::Value *V, llvm::Instruction *&InsertPos,
                    llvm::DenseSet<llvm::Instruction *> Scheduled);

inline unsigned GetHwThreadsPerWG(const IGC::CPlatform &platform) {
  unsigned hwThreadPerWorkgroup = platform.getMaxNumberHWThreadForEachWG();
  if (platform.supportPooledEU()) {
    hwThreadPerWorkgroup = std::min(platform.getMaxNumberThreadPerWorkgroupPooledMax(), (unsigned)64);
  }
  return hwThreadPerWorkgroup ? hwThreadPerWorkgroup : 1;
}

inline SIMDMode getLeastSIMDAllowed(unsigned int threadGroupSize, unsigned int hwThreadPerWorkgroup) {
  if (hwThreadPerWorkgroup == 0) {
    hwThreadPerWorkgroup = 42; // On GT1 HW, there are 7 threads/EU and 6 EU/subslice, 42 is the minimum
                               // threads/workgroup any HW can support
  }
  if (threadGroupSize <= hwThreadPerWorkgroup * 8) {
    return SIMDMode::SIMD8;
  } else if (threadGroupSize <= hwThreadPerWorkgroup * 16) {
    return SIMDMode::SIMD16;
  } else {
    return SIMDMode::SIMD32;
  }
}

enum dim { ThreadGroupSize_X, ThreadGroupSize_Y, ThreadGroupSize_Z };

unsigned int GetthreadGroupSize(const llvm::Module &M, dim dimension);
void SetthreadGroupSize(llvm::Module &M, llvm::Constant *size, dim dimension);

// Debug line info helper function
inline void updateDebugLoc(llvm::Instruction *pOrigin, llvm::Instruction *pNew) {
  IGC_ASSERT_MESSAGE(nullptr != pOrigin, "Expect valid instructions");
  IGC_ASSERT_MESSAGE(nullptr != pNew, "Expect valid instructions");
  pNew->setDebugLoc(pOrigin->getDebugLoc());
}

inline bool isDbgIntrinsic(const llvm::Instruction *I) {
  if (auto *GXI = llvm::dyn_cast<llvm::GenIntrinsicInst>(I))
    return GXI->getIntrinsicID() == llvm::GenISAIntrinsic::GenISA_CatchAllDebugLine;
  return llvm::isa<llvm::DbgInfoIntrinsic>(I);
}

llvm::ConstantInt *getConstantSInt(llvm::IRBuilder<> &Builder, const int bitSize, int64_t val);
llvm::ConstantInt *getConstantUInt(llvm::IRBuilder<> &Builder, const int bitSize, uint64_t val);
llvm::Value *CreateMulhS64(llvm::IRBuilder<> &B, llvm::Value *const u, llvm::Value *const v);
llvm::Value *CreateMulhU64(llvm::IRBuilder<> &B, llvm::Value *const u, llvm::Value *const v);
llvm::Value *CreateMulh(llvm::Function &F, llvm::IRBuilder<> &B, const bool isSigned, llvm::Value *const u,
                        llvm::Value *const v);

// Ported from PostDominators.cpp of llvm10 or later
// replace this with PDT.dominates(I1, I2) once we upgrade
bool PDT_dominates(llvm::PostDominatorTree &PTD, const llvm::Instruction *I1, const llvm::Instruction *I2);

// Returns true if a function has an inline asm call instruction
bool hasInlineAsmInFunc(llvm::Function &F);

std::tuple<std::string, std::string, unsigned> ParseVectorVariantFunctionString(llvm::StringRef varStr);

// Return base type of complex type or nullptr if it cannot be processed
llvm::Type *GetBaseType(llvm::Type *ProcessedType, bool StructAsBaseOk = false);
bool isSimpleStructTy(llvm::StructType *STy, uint32_t EltBytes);

// Function modifies address space in selected uses of given input value
void FixAddressSpaceInAllUses(llvm::Value *ptr, uint newAS, uint oldAS);

llvm::Value *CombineSampleOrGather4Params(llvm::IRBuilder<> &builder, llvm::Value *param1, llvm::Value *param2,
                                          uint numBits, const std::string &param1Name, const std::string &param2Name);

// Returns the dynamic URB base offset and an immediate const offset
// from the dynamic base. The function calculates the result by walking
// the use-def chain of pUrbOffset.
// If pUrbOffset is an immediate constant (==offset) then
// <nullptr, offset> is returned.
// In all other cases <pUrbOffset, 0> is returned.
std::pair<llvm::Value *, unsigned int> GetURBBaseAndOffset(llvm::Value *pUrbOffset);

std::vector<std::pair<unsigned int, std::string>> GetPrintfStrings(llvm::Module &M);

template <typename Fn> struct Defer {
  Defer(Fn F) : F(F) {}
  Defer(const Defer &) = delete;
  Defer &operator=(const Defer &) = delete;
  ~Defer() {
    if (Do)
      F();
  }
  void operator()() {
    if (Do)
      F();
    Do = false;
  }

private:
  bool Do = true;
  Fn F;
};

// Mimic LLVM functions:
//   RecursivelyDeleteTriviallyDeadInstructions()
// The difference is that the input here are dead instructions and
// are not necessarily trivially dead. For example, store instruction.
void RecursivelyDeleteDeadInstructions(
    llvm::Instruction *I, const llvm::TargetLibraryInfo *TLI = nullptr, llvm::MemorySSAUpdater *MSSAU = nullptr,
    const std::function<void(llvm::Value *)> &AboutToDeleteCallback = std::function<void(llvm::Value *)>());

void RecursivelyDeleteDeadInstructions(
    const llvm::SmallVectorImpl<llvm::Instruction *> &DeadInsts, const llvm::TargetLibraryInfo *TLI = nullptr,
    llvm::MemorySSAUpdater *MSSAU = nullptr,
    const std::function<void(llvm::Value *)> &AboutToDeleteCallback = std::function<void(llvm::Value *)>());

bool SeparateSpillAndScratch(const CodeGenContext *ctx);
bool UsedWithoutImmInMemInst(llvm::Value *v);

bool AllowShortImplicitPayloadHeader(const CodeGenContext *ctx);
bool AllowRemovingUnusedImplicitArguments(const CodeGenContext *ctx);
} // namespace IGC