This is an archive of the discontinued LLVM Phabricator instance.

AMDGPU: Update isFPImmLegal for f16
ClosedPublic

Authored by arsenm on Dec 17 2016, 6:56 PM.

Reviewers

• tstellarAMD

Summary

I don't think this matters because ConstantFP is legal.

arsenm retitled this revision from to AMDGPU: Update isFPImmLegal for f16.

arsenm updated this object.

arsenm added a subscriber: llvm-commits.

LGTM.

This revision is now accepted and ready to land.Dec 21 2016, 6:51 PM

r290299

Path

Size

lib/

Target/

AMDGPU/

3 lines

	Show First 20 Lines • Show All 492 Lines • ▼ Show 20 Lines
	bool AMDGPUTargetLowering::isSelectSupported(SelectSupportKind SelType) const {			bool AMDGPUTargetLowering::isSelectSupported(SelectSupportKind SelType) const {
	return true;			return true;
	}			}

	// The backend supports 32 and 64 bit floating point immediates.			// The backend supports 32 and 64 bit floating point immediates.
	// FIXME: Why are we reporting vectors of FP immediates as legal?			// FIXME: Why are we reporting vectors of FP immediates as legal?
	bool AMDGPUTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {			bool AMDGPUTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
	EVT ScalarVT = VT.getScalarType();			EVT ScalarVT = VT.getScalarType();
	return (ScalarVT == MVT::f32 \|\| ScalarVT == MVT::f64);			return (ScalarVT == MVT::f32 \|\| ScalarVT == MVT::f64 \|\|
				(ScalarVT == MVT::f16 && Subtarget->has16BitInsts()));
	}			}

	// We don't want to shrink f64 / f32 constants.			// We don't want to shrink f64 / f32 constants.
	bool AMDGPUTargetLowering::ShouldShrinkFPConstant(EVT VT) const {			bool AMDGPUTargetLowering::ShouldShrinkFPConstant(EVT VT) const {
	EVT ScalarVT = VT.getScalarType();			EVT ScalarVT = VT.getScalarType();
	return (ScalarVT != MVT::f32 && ScalarVT != MVT::f64);			return (ScalarVT != MVT::f32 && ScalarVT != MVT::f64);
	}			}

	▲ Show 20 Lines • Show All 2,645 Lines • Show Last 20 Lines