[CostModel] FRINT/LRINT/LLRINT cost is not modeled on X86

[phoebewang]

Found during discussing in #126217

[llvmbot]

@llvm/issue-subscribers-backend-x86

Author: Phoebe Wang (phoebewang)

Found during discussing in https://github.com//pull/126217

[RKSimon]

@phoebewang We do this in X86TTIImpl.getIntrinsicInstrCost, what costs do we need?

llvm-project/llvm/lib/Target/X86/X86TargetTransformInfo.cpp

Lines 4507 to 4514 in 1435c8e

	case Intrinsic::lrint:
	case Intrinsic::llrint: {
	// X86 can use the CVTP2SI instructions to lower lrint/llrint calls, which
	// have the same costs as the CVTTP2SI (fptosi) instructions
	const SmallVectorImpl<Type *> &ArgTys = ICA.getArgTypes();
	return getCastInstrCost(Instruction::FPToSI, RetTy, ArgTys[0],
	TTI::CastContextHint::None, CostKind);
	}

[phoebewang]

@phoebewang We do this in X86TTIImpl.getIntrinsicInstrCost, what costs do we need?

llvm-project/llvm/lib/Target/X86/X86TargetTransformInfo.cpp

Lines 4507 to 4514 in 1435c8e

case Intrinsic::lrint:
case Intrinsic::llrint: {
// X86 can use the CVTP2SI instructions to lower lrint/llrint calls, which
// have the same costs as the CVTTP2SI (fptosi) instructions
const SmallVectorImpl<Type *> &ArgTys = ICA.getArgTypes();
return getCastInstrCost(Instruction::FPToSI, RetTy, ArgTys[0],
TTI::CastContextHint::None, CostKind);
}

Thanks! How about the vXi64? I don't see we model them before AVX512DQ. And we still missing FRINT.