#include "AMDGPUTargetTransformInfo.h"
#include "AMDGPUSubtarget.h"
#include "AMDGPUTargetMachine.h"
#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
#include "SIModeRegisterDefaults.h"
#include "llvm/ADT/SmallBitVector.h"
#include "llvm/Analysis/InlineCost.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/CodeGen/Analysis.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/IntrinsicsAMDGPU.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/Support/KnownBits.h"
#include <optional>

Macros
#define	DEBUG_TYPE "AMDGPUtti"

Functions
static bool	dependsOnLocalPhi (const Loop L, const Value Cond, unsigned Depth=0)
static bool	canFuseFMulWithFAddSub (const SITargetLowering &TLI, Type Ty, const Instruction FMul, const Instruction *FAddSub)
static bool	intrinsicHasPackedVectorBenefit (Intrinsic::ID ID)
static std::optional< unsigned >	getNumberOfPackedMaskElts (Type *Ty)
static unsigned	adjustInliningThresholdUsingCallee (const CallBase CB, const SITargetLowering TLI, const GCNTTIImpl *TTIImpl)
static unsigned	getCallArgsTotalAllocaSize (const CallBase *CB, const DataLayout &DL)

Variables
static cl::opt< unsigned >	UnrollThresholdPrivate ("amdgpu-unroll-threshold-private", cl::desc("Unroll threshold for AMDGPU if private memory used in a loop"), cl::init(2700), cl::Hidden)
static cl::opt< unsigned >	UnrollThresholdLocal ("amdgpu-unroll-threshold-local", cl::desc("Unroll threshold for AMDGPU if local memory used in a loop"), cl::init(1000), cl::Hidden)
static cl::opt< unsigned >	UnrollThresholdIf ("amdgpu-unroll-threshold-if", cl::desc("Unroll threshold increment for AMDGPU for each if statement inside loop"), cl::init(200), cl::Hidden)
static cl::opt< bool >	UnrollRuntimeLocal ("amdgpu-unroll-runtime-local", cl::desc("Allow runtime unroll for AMDGPU if local memory used in a loop"), cl::init(true), cl::Hidden)
static cl::opt< unsigned >	UnrollMaxBlockToAnalyze ("amdgpu-unroll-max-block-to-analyze", cl::desc("Inner loop block size threshold to analyze in unroll for AMDGPU"), cl::init(32), cl::Hidden)
static cl::opt< unsigned >	ArgAllocaCost ("amdgpu-inline-arg-alloca-cost", cl::Hidden, cl::init(4000), cl::desc("Cost of alloca argument"))
static cl::opt< unsigned >	ArgAllocaCutoff ("amdgpu-inline-arg-alloca-cutoff", cl::Hidden, cl::init(256), cl::desc("Maximum alloca size to use for inline cost"))
static cl::opt< size_t >	InlineMaxBB ("amdgpu-inline-max-bb", cl::Hidden, cl::init(1100), cl::desc("Maximum number of BBs allowed in a function after inlining" " (compile time constraint)"))
static cl::opt< unsigned >	MemcpyLoopUnroll ("amdgpu-memcpy-loop-unroll", cl::desc("Unroll factor (affecting 4x32-bit operations) to use for memory " "operations when lowering statically-sized memcpy, memmove, or" "memset as a loop"), cl::init(16), cl::Hidden)
static constexpr unsigned	MaskPackCostPerElt = 4
static constexpr unsigned	MaskUnpackCostPerElt = 3

Macro Definition Documentation

◆ DEBUG_TYPE

#define DEBUG_TYPE "AMDGPUtti"

Definition at line 36 of file AMDGPUTargetTransformInfo.cpp.

Function Documentation

◆ adjustInliningThresholdUsingCallee()

unsigned adjustInliningThresholdUsingCallee	(	const CallBase *	CB,
		const SITargetLowering *	TLI,
		const GCNTTIImpl *	TTIImpl )

static

Definition at line 1654 of file AMDGPUTargetTransformInfo.cpp.

References A(), llvm::CallBase::args(), llvm::ComputeValueVTs(), DL, llvm::CallBase::getArgOperandNo(), llvm::CallBase::getCallingConv(), llvm::Value::getContext(), llvm::TargetTransformInfoImplBase::getDataLayout(), llvm::InlineConstants::getInstrCost(), llvm::Type::getInt32Ty(), llvm::SITargetLowering::getNumRegistersForCallingConv(), llvm::InstructionCost::getValue(), llvm::AMDGPU::isArgPassedInSGPR(), llvm::AMDGPUAS::PRIVATE_ADDRESS, and llvm::TargetTransformInfo::TCK_SizeAndLatency.

Referenced by llvm::GCNTTIImpl::adjustInliningThreshold().

◆ canFuseFMulWithFAddSub()

bool canFuseFMulWithFAddSub	(	const SITargetLowering &	TLI,
		Type *	Ty,
		const Instruction *	FMul,
		const Instruction *	FAddSub )

static

Returns: true if FMul and its single fadd/fsub user FAddSub are expected to fuse during instruction selection. Ty is the type the fused operation runs on.

Definition at line 528 of file AMDGPUTargetTransformInfo.cpp.

References assert(), F, llvm::FMul, llvm::Instruction::getFunction(), llvm::Instruction::getOpcode(), llvm::Instruction::hasAllowContract(), llvm::SITargetLowering::isFMADLegal(), and llvm::SITargetLowering::isFMAFasterThanFMulAndFAdd().

Referenced by llvm::GCNTTIImpl::getArithmeticInstrCost(), and llvm::GCNTTIImpl::isProfitableToSinkOperands().

◆ dependsOnLocalPhi()

bool dependsOnLocalPhi	(	const Loop *	L,
		const Value *	Cond,
		unsigned	Depth = 0 )

static

Definition at line 89 of file AMDGPUTargetTransformInfo.cpp.

References Cond, dependsOnLocalPhi(), llvm::Depth, llvm::dyn_cast(), I, llvm::none_of(), and PHI.

Referenced by dependsOnLocalPhi(), and llvm::AMDGPUTTIImpl::getUnrollingPreferences().

◆ getCallArgsTotalAllocaSize()

unsigned getCallArgsTotalAllocaSize	(	const CallBase *	CB,
		const DataLayout &	DL )

static

Definition at line 1700 of file AMDGPUTargetTransformInfo.cpp.

References llvm::CallBase::args(), DL, llvm::dyn_cast(), llvm::AMDGPUAS::FLAT_ADDRESS, llvm::AllocaInst::getAllocationSize(), llvm::getUnderlyingObject(), llvm::SmallPtrSetImpl< PtrType >::insert(), llvm::AllocaInst::isStaticAlloca(), llvm::AMDGPUAS::PRIVATE_ADDRESS, and Size.

Referenced by llvm::GCNTTIImpl::adjustInliningThreshold(), and llvm::GCNTTIImpl::getCallerAllocaCost().

◆ getNumberOfPackedMaskElts()

std::optional< unsigned > getNumberOfPackedMaskElts ( Type * Ty )

static

Definition at line 1000 of file AMDGPUTargetTransformInfo.cpp.

References llvm::dyn_cast().

Referenced by llvm::GCNTTIImpl::getArithmeticReductionCost(), and llvm::GCNTTIImpl::getCastInstrCost().

◆ intrinsicHasPackedVectorBenefit()

bool intrinsicHasPackedVectorBenefit ( Intrinsic::ID ID )

static

Definition at line 704 of file AMDGPUTargetTransformInfo.cpp.

Referenced by llvm::GCNTTIImpl::getIntrinsicInstrCost().

Variable Documentation

◆ ArgAllocaCost

cl::opt< unsigned > ArgAllocaCost("amdgpu-inline-arg-alloca-cost", cl::Hidden, cl::init(4000), cl::desc("Cost of alloca argument"))	(	"amdgpu-inline-arg-alloca-cost"	,
		cl::Hidden	,
		cl::init(4000)	,
		cl::desc("Cost of alloca argument")	)

static

Referenced by llvm::GCNTTIImpl::adjustInliningThreshold(), and llvm::GCNTTIImpl::getCallerAllocaCost().

◆ ArgAllocaCutoff

cl::opt< unsigned > ArgAllocaCutoff("amdgpu-inline-arg-alloca-cutoff", cl::Hidden, cl::init(256), cl::desc("Maximum alloca size to use for inline cost"))	(	"amdgpu-inline-arg-alloca-cutoff"	,
		cl::Hidden	,
		cl::init(256)	,
		cl::desc("Maximum alloca size to use for inline cost")	)

static

Referenced by llvm::GCNTTIImpl::getCallerAllocaCost().

◆ InlineMaxBB

cl::opt< size_t > InlineMaxBB("amdgpu-inline-max-bb", cl::Hidden, cl::init(1100), cl::desc("Maximum number of BBs allowed in a function after inlining" " (compile time constraint)"))	(	"amdgpu-inline-max-bb"	,
		cl::Hidden	,
		cl::init(1100)	,
		cl::desc("Maximum number of BBs allowed in a function after inlining" " (compile time constraint)")	)

static

Referenced by llvm::GCNTTIImpl::areInlineCompatible().

◆ MaskPackCostPerElt

unsigned MaskPackCostPerElt = 4

staticconstexpr

Definition at line 997 of file AMDGPUTargetTransformInfo.cpp.

Referenced by llvm::GCNTTIImpl::getArithmeticReductionCost(), and llvm::GCNTTIImpl::getCastInstrCost().

◆ MaskUnpackCostPerElt

unsigned MaskUnpackCostPerElt = 3

staticconstexpr

Definition at line 998 of file AMDGPUTargetTransformInfo.cpp.

Referenced by llvm::GCNTTIImpl::getCastInstrCost().

◆ MemcpyLoopUnroll

cl::opt< unsigned > MemcpyLoopUnroll("amdgpu-memcpy-loop-unroll", cl::desc("Unroll factor (affecting 4x32-bit operations) to use for memory " "operations when lowering statically-sized memcpy, memmove, or" "memset as a loop"), cl::init(16), cl::Hidden)	(	"amdgpu-memcpy-loop-unroll"	,
		cl::desc("Unroll factor (affecting 4x32-bit operations) to use for memory " "operations when lowering statically-sized memcpy, memmove, or" "memset as a loop")	,
		cl::init(16)	,
		cl::Hidden	)

static

Referenced by llvm::GCNTTIImpl::getMemcpyLoopLoweringType().

◆ UnrollMaxBlockToAnalyze

cl::opt< unsigned > UnrollMaxBlockToAnalyze("amdgpu-unroll-max-block-to-analyze", cl::desc("Inner loop block size threshold to analyze in unroll for AMDGPU"), cl::init(32), cl::Hidden)	(	"amdgpu-unroll-max-block-to-analyze"	,
		cl::desc("Inner loop block size threshold to analyze in unroll for AMDGPU")	,
		cl::init(32)	,
		cl::Hidden	)

static

Referenced by llvm::AMDGPUTTIImpl::getUnrollingPreferences().

◆ UnrollRuntimeLocal

cl::opt< bool > UnrollRuntimeLocal("amdgpu-unroll-runtime-local", cl::desc("Allow runtime unroll for AMDGPU if local memory used in a loop"), cl::init(true), cl::Hidden)	(	"amdgpu-unroll-runtime-local"	,
		cl::desc("Allow runtime unroll for AMDGPU if local memory used in a loop")	,
		cl::init(true)	,
		cl::Hidden	)

static

Referenced by llvm::AMDGPUTTIImpl::getUnrollingPreferences().

◆ UnrollThresholdIf

cl::opt< unsigned > UnrollThresholdIf("amdgpu-unroll-threshold-if", cl::desc("Unroll threshold increment for AMDGPU for each if statement inside loop"), cl::init(200), cl::Hidden)	(	"amdgpu-unroll-threshold-if"	,
		cl::desc("Unroll threshold increment for AMDGPU for each if statement inside loop")	,
		cl::init(200)	,
		cl::Hidden	)

static

Referenced by llvm::AMDGPUTTIImpl::getUnrollingPreferences().

◆ UnrollThresholdLocal

cl::opt< unsigned > UnrollThresholdLocal("amdgpu-unroll-threshold-local", cl::desc("Unroll threshold for AMDGPU if local memory used in a loop"), cl::init(1000), cl::Hidden)	(	"amdgpu-unroll-threshold-local"	,
		cl::desc("Unroll threshold for AMDGPU if local memory used in a loop")	,
		cl::init(1000)	,
		cl::Hidden	)

static

Referenced by llvm::AMDGPUTTIImpl::getUnrollingPreferences().

◆ UnrollThresholdPrivate

cl::opt< unsigned > UnrollThresholdPrivate("amdgpu-unroll-threshold-private", cl::desc("Unroll threshold for AMDGPU if private memory used in a loop"), cl::init(2700), cl::Hidden)	(	"amdgpu-unroll-threshold-private"	,
		cl::desc("Unroll threshold for AMDGPU if private memory used in a loop")	,
		cl::init(2700)	,
		cl::Hidden	)

static

Referenced by llvm::AMDGPUTTIImpl::getUnrollingPreferences().

Macros

Functions

Variables

Macro Definition Documentation

◆ DEBUG_TYPE

Function Documentation

◆ adjustInliningThresholdUsingCallee()

◆ canFuseFMulWithFAddSub()

◆ dependsOnLocalPhi()

◆ getCallArgsTotalAllocaSize()

◆ getNumberOfPackedMaskElts()

◆ intrinsicHasPackedVectorBenefit()

Variable Documentation

◆ ArgAllocaCost

◆ ArgAllocaCutoff

◆ InlineMaxBB

◆ MaskPackCostPerElt

◆ MaskUnpackCostPerElt

◆ MemcpyLoopUnroll

◆ UnrollMaxBlockToAnalyze

◆ UnrollRuntimeLocal

◆ UnrollThresholdIf

◆ UnrollThresholdLocal

◆ UnrollThresholdPrivate