[dev.simd] cmd/compile: output of simd generator, more ... rewrite rules

Generated by simdgen CL 686378

Change-Id: I876ab91085c266ced59fc82ea12be709dc7eb721
Reviewed-on: https://go-review.googlesource.com/c/go/+/686495
Reviewed-by: Junyang Shao <shaojunyang@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
This commit is contained in:
David Chase 2025-07-08 12:52:30 -04:00
parent aab8b173a9
commit 9ea33ed538
2 changed files with 306 additions and 1650 deletions

View file

@ -377,12 +377,12 @@
(FusedMultiplySubAddMaskedFloat64x2 x y z mask) => (VFMSUBADD213PDMasked128 x y z (VPMOVVec64x2ToM <types.TypeMask> mask))
(FusedMultiplySubAddMaskedFloat64x4 x y z mask) => (VFMSUBADD213PDMasked256 x y z (VPMOVVec64x4ToM <types.TypeMask> mask))
(FusedMultiplySubAddMaskedFloat64x8 x y z mask) => (VFMSUBADD213PDMasked512 x y z (VPMOVVec64x8ToM <types.TypeMask> mask))
(GaloisFieldAffineTransformUint8x16 [a] x y) => (VGF2P8AFFINEQB128 [a] x y)
(GaloisFieldAffineTransformUint8x32 [a] x y) => (VGF2P8AFFINEQB256 [a] x y)
(GaloisFieldAffineTransformUint8x64 [a] x y) => (VGF2P8AFFINEQB512 [a] x y)
(GaloisFieldAffineTransformInversedUint8x16 [a] x y) => (VGF2P8AFFINEINVQB128 [a] x y)
(GaloisFieldAffineTransformInversedUint8x32 [a] x y) => (VGF2P8AFFINEINVQB256 [a] x y)
(GaloisFieldAffineTransformInversedUint8x64 [a] x y) => (VGF2P8AFFINEINVQB512 [a] x y)
(GaloisFieldAffineTransformUint8x16 ...) => (VGF2P8AFFINEQB128 ...)
(GaloisFieldAffineTransformUint8x32 ...) => (VGF2P8AFFINEQB256 ...)
(GaloisFieldAffineTransformUint8x64 ...) => (VGF2P8AFFINEQB512 ...)
(GaloisFieldAffineTransformInversedUint8x16 ...) => (VGF2P8AFFINEINVQB128 ...)
(GaloisFieldAffineTransformInversedUint8x32 ...) => (VGF2P8AFFINEINVQB256 ...)
(GaloisFieldAffineTransformInversedUint8x64 ...) => (VGF2P8AFFINEINVQB512 ...)
(GaloisFieldAffineTransformInversedMaskedUint8x16 [a] x y mask) => (VGF2P8AFFINEINVQBMasked128 [a] x y (VPMOVVec8x16ToM <types.TypeMask> mask))
(GaloisFieldAffineTransformInversedMaskedUint8x32 [a] x y mask) => (VGF2P8AFFINEINVQBMasked256 [a] x y (VPMOVVec8x32ToM <types.TypeMask> mask))
(GaloisFieldAffineTransformInversedMaskedUint8x64 [a] x y mask) => (VGF2P8AFFINEINVQBMasked512 [a] x y (VPMOVVec8x64ToM <types.TypeMask> mask))
@ -395,24 +395,24 @@
(GaloisFieldMulMaskedUint8x16 x y mask) => (VGF2P8MULBMasked128 x y (VPMOVVec8x16ToM <types.TypeMask> mask))
(GaloisFieldMulMaskedUint8x32 x y mask) => (VGF2P8MULBMasked256 x y (VPMOVVec8x32ToM <types.TypeMask> mask))
(GaloisFieldMulMaskedUint8x64 x y mask) => (VGF2P8MULBMasked512 x y (VPMOVVec8x64ToM <types.TypeMask> mask))
(Get128Float32x8 [a] x) => (VEXTRACTF128128 [a] x)
(Get128Float64x4 [a] x) => (VEXTRACTF128128 [a] x)
(Get128Int8x32 [a] x) => (VEXTRACTI128128 [a] x)
(Get128Int16x16 [a] x) => (VEXTRACTI128128 [a] x)
(Get128Int32x8 [a] x) => (VEXTRACTI128128 [a] x)
(Get128Int64x4 [a] x) => (VEXTRACTI128128 [a] x)
(Get128Uint8x32 [a] x) => (VEXTRACTI128128 [a] x)
(Get128Uint16x16 [a] x) => (VEXTRACTI128128 [a] x)
(Get128Uint32x8 [a] x) => (VEXTRACTI128128 [a] x)
(Get128Uint64x4 [a] x) => (VEXTRACTI128128 [a] x)
(GetElemInt8x16 [a] x) => (VPEXTRB128 [a] x)
(GetElemInt16x8 [a] x) => (VPEXTRW128 [a] x)
(GetElemInt32x4 [a] x) => (VPEXTRD128 [a] x)
(GetElemInt64x2 [a] x) => (VPEXTRQ128 [a] x)
(GetElemUint8x16 [a] x) => (VPEXTRB128 [a] x)
(GetElemUint16x8 [a] x) => (VPEXTRW128 [a] x)
(GetElemUint32x4 [a] x) => (VPEXTRD128 [a] x)
(GetElemUint64x2 [a] x) => (VPEXTRQ128 [a] x)
(Get128Float32x8 ...) => (VEXTRACTF128128 ...)
(Get128Float64x4 ...) => (VEXTRACTF128128 ...)
(Get128Int8x32 ...) => (VEXTRACTI128128 ...)
(Get128Int16x16 ...) => (VEXTRACTI128128 ...)
(Get128Int32x8 ...) => (VEXTRACTI128128 ...)
(Get128Int64x4 ...) => (VEXTRACTI128128 ...)
(Get128Uint8x32 ...) => (VEXTRACTI128128 ...)
(Get128Uint16x16 ...) => (VEXTRACTI128128 ...)
(Get128Uint32x8 ...) => (VEXTRACTI128128 ...)
(Get128Uint64x4 ...) => (VEXTRACTI128128 ...)
(GetElemInt8x16 ...) => (VPEXTRB128 ...)
(GetElemInt16x8 ...) => (VPEXTRW128 ...)
(GetElemInt32x4 ...) => (VPEXTRD128 ...)
(GetElemInt64x2 ...) => (VPEXTRQ128 ...)
(GetElemUint8x16 ...) => (VPEXTRB128 ...)
(GetElemUint16x8 ...) => (VPEXTRW128 ...)
(GetElemUint32x4 ...) => (VPEXTRD128 ...)
(GetElemUint64x2 ...) => (VPEXTRQ128 ...)
(GreaterFloat32x4 x y) => (VCMPPS128 [14] x y)
(GreaterFloat32x8 x y) => (VCMPPS256 [14] x y)
(GreaterFloat32x16 x y) => (VPMOVMToVec32x16 (VCMPPS512 [14] x y))
@ -1031,18 +1031,18 @@
(PopCountMaskedUint64x2 x mask) => (VPOPCNTQMasked128 x (VPMOVVec64x2ToM <types.TypeMask> mask))
(PopCountMaskedUint64x4 x mask) => (VPOPCNTQMasked256 x (VPMOVVec64x4ToM <types.TypeMask> mask))
(PopCountMaskedUint64x8 x mask) => (VPOPCNTQMasked512 x (VPMOVVec64x8ToM <types.TypeMask> mask))
(RotateAllLeftInt32x4 [a] x) => (VPROLD128 [a] x)
(RotateAllLeftInt32x8 [a] x) => (VPROLD256 [a] x)
(RotateAllLeftInt32x16 [a] x) => (VPROLD512 [a] x)
(RotateAllLeftInt64x2 [a] x) => (VPROLQ128 [a] x)
(RotateAllLeftInt64x4 [a] x) => (VPROLQ256 [a] x)
(RotateAllLeftInt64x8 [a] x) => (VPROLQ512 [a] x)
(RotateAllLeftUint32x4 [a] x) => (VPROLD128 [a] x)
(RotateAllLeftUint32x8 [a] x) => (VPROLD256 [a] x)
(RotateAllLeftUint32x16 [a] x) => (VPROLD512 [a] x)
(RotateAllLeftUint64x2 [a] x) => (VPROLQ128 [a] x)
(RotateAllLeftUint64x4 [a] x) => (VPROLQ256 [a] x)
(RotateAllLeftUint64x8 [a] x) => (VPROLQ512 [a] x)
(RotateAllLeftInt32x4 ...) => (VPROLD128 ...)
(RotateAllLeftInt32x8 ...) => (VPROLD256 ...)
(RotateAllLeftInt32x16 ...) => (VPROLD512 ...)
(RotateAllLeftInt64x2 ...) => (VPROLQ128 ...)
(RotateAllLeftInt64x4 ...) => (VPROLQ256 ...)
(RotateAllLeftInt64x8 ...) => (VPROLQ512 ...)
(RotateAllLeftUint32x4 ...) => (VPROLD128 ...)
(RotateAllLeftUint32x8 ...) => (VPROLD256 ...)
(RotateAllLeftUint32x16 ...) => (VPROLD512 ...)
(RotateAllLeftUint64x2 ...) => (VPROLQ128 ...)
(RotateAllLeftUint64x4 ...) => (VPROLQ256 ...)
(RotateAllLeftUint64x8 ...) => (VPROLQ512 ...)
(RotateAllLeftMaskedInt32x4 [a] x mask) => (VPROLDMasked128 [a] x (VPMOVVec32x4ToM <types.TypeMask> mask))
(RotateAllLeftMaskedInt32x8 [a] x mask) => (VPROLDMasked256 [a] x (VPMOVVec32x8ToM <types.TypeMask> mask))
(RotateAllLeftMaskedInt32x16 [a] x mask) => (VPROLDMasked512 [a] x (VPMOVVec32x16ToM <types.TypeMask> mask))
@ -1055,18 +1055,18 @@
(RotateAllLeftMaskedUint64x2 [a] x mask) => (VPROLQMasked128 [a] x (VPMOVVec64x2ToM <types.TypeMask> mask))
(RotateAllLeftMaskedUint64x4 [a] x mask) => (VPROLQMasked256 [a] x (VPMOVVec64x4ToM <types.TypeMask> mask))
(RotateAllLeftMaskedUint64x8 [a] x mask) => (VPROLQMasked512 [a] x (VPMOVVec64x8ToM <types.TypeMask> mask))
(RotateAllRightInt32x4 [a] x) => (VPRORD128 [a] x)
(RotateAllRightInt32x8 [a] x) => (VPRORD256 [a] x)
(RotateAllRightInt32x16 [a] x) => (VPRORD512 [a] x)
(RotateAllRightInt64x2 [a] x) => (VPRORQ128 [a] x)
(RotateAllRightInt64x4 [a] x) => (VPRORQ256 [a] x)
(RotateAllRightInt64x8 [a] x) => (VPRORQ512 [a] x)
(RotateAllRightUint32x4 [a] x) => (VPRORD128 [a] x)
(RotateAllRightUint32x8 [a] x) => (VPRORD256 [a] x)
(RotateAllRightUint32x16 [a] x) => (VPRORD512 [a] x)
(RotateAllRightUint64x2 [a] x) => (VPRORQ128 [a] x)
(RotateAllRightUint64x4 [a] x) => (VPRORQ256 [a] x)
(RotateAllRightUint64x8 [a] x) => (VPRORQ512 [a] x)
(RotateAllRightInt32x4 ...) => (VPRORD128 ...)
(RotateAllRightInt32x8 ...) => (VPRORD256 ...)
(RotateAllRightInt32x16 ...) => (VPRORD512 ...)
(RotateAllRightInt64x2 ...) => (VPRORQ128 ...)
(RotateAllRightInt64x4 ...) => (VPRORQ256 ...)
(RotateAllRightInt64x8 ...) => (VPRORQ512 ...)
(RotateAllRightUint32x4 ...) => (VPRORD128 ...)
(RotateAllRightUint32x8 ...) => (VPRORD256 ...)
(RotateAllRightUint32x16 ...) => (VPRORD512 ...)
(RotateAllRightUint64x2 ...) => (VPRORQ128 ...)
(RotateAllRightUint64x4 ...) => (VPRORQ256 ...)
(RotateAllRightUint64x8 ...) => (VPRORQ512 ...)
(RotateAllRightMaskedInt32x4 [a] x mask) => (VPRORDMasked128 [a] x (VPMOVVec32x4ToM <types.TypeMask> mask))
(RotateAllRightMaskedInt32x8 [a] x mask) => (VPRORDMasked256 [a] x (VPMOVVec32x8ToM <types.TypeMask> mask))
(RotateAllRightMaskedInt32x16 [a] x mask) => (VPRORDMasked512 [a] x (VPMOVVec32x16ToM <types.TypeMask> mask))
@ -1219,24 +1219,24 @@
(SaturatedUnsignedSignedQuadDotProdAccumulateMaskedUint32x4 x y z mask) => (VPDPBUSDSMasked128 x y z (VPMOVVec32x4ToM <types.TypeMask> mask))
(SaturatedUnsignedSignedQuadDotProdAccumulateMaskedUint32x8 x y z mask) => (VPDPBUSDSMasked256 x y z (VPMOVVec32x8ToM <types.TypeMask> mask))
(SaturatedUnsignedSignedQuadDotProdAccumulateMaskedUint32x16 x y z mask) => (VPDPBUSDSMasked512 x y z (VPMOVVec32x16ToM <types.TypeMask> mask))
(Set128Float32x8 [a] x y) => (VINSERTF128256 [a] x y)
(Set128Float64x4 [a] x y) => (VINSERTF128256 [a] x y)
(Set128Int8x32 [a] x y) => (VINSERTI128256 [a] x y)
(Set128Int16x16 [a] x y) => (VINSERTI128256 [a] x y)
(Set128Int32x8 [a] x y) => (VINSERTI128256 [a] x y)
(Set128Int64x4 [a] x y) => (VINSERTI128256 [a] x y)
(Set128Uint8x32 [a] x y) => (VINSERTI128256 [a] x y)
(Set128Uint16x16 [a] x y) => (VINSERTI128256 [a] x y)
(Set128Uint32x8 [a] x y) => (VINSERTI128256 [a] x y)
(Set128Uint64x4 [a] x y) => (VINSERTI128256 [a] x y)
(SetElemInt8x16 [a] x y) => (VPINSRB128 [a] x y)
(SetElemInt16x8 [a] x y) => (VPINSRW128 [a] x y)
(SetElemInt32x4 [a] x y) => (VPINSRD128 [a] x y)
(SetElemInt64x2 [a] x y) => (VPINSRQ128 [a] x y)
(SetElemUint8x16 [a] x y) => (VPINSRB128 [a] x y)
(SetElemUint16x8 [a] x y) => (VPINSRW128 [a] x y)
(SetElemUint32x4 [a] x y) => (VPINSRD128 [a] x y)
(SetElemUint64x2 [a] x y) => (VPINSRQ128 [a] x y)
(Set128Float32x8 ...) => (VINSERTF128256 ...)
(Set128Float64x4 ...) => (VINSERTF128256 ...)
(Set128Int8x32 ...) => (VINSERTI128256 ...)
(Set128Int16x16 ...) => (VINSERTI128256 ...)
(Set128Int32x8 ...) => (VINSERTI128256 ...)
(Set128Int64x4 ...) => (VINSERTI128256 ...)
(Set128Uint8x32 ...) => (VINSERTI128256 ...)
(Set128Uint16x16 ...) => (VINSERTI128256 ...)
(Set128Uint32x8 ...) => (VINSERTI128256 ...)
(Set128Uint64x4 ...) => (VINSERTI128256 ...)
(SetElemInt8x16 ...) => (VPINSRB128 ...)
(SetElemInt16x8 ...) => (VPINSRW128 ...)
(SetElemInt32x4 ...) => (VPINSRD128 ...)
(SetElemInt64x2 ...) => (VPINSRQ128 ...)
(SetElemUint8x16 ...) => (VPINSRB128 ...)
(SetElemUint16x8 ...) => (VPINSRW128 ...)
(SetElemUint32x4 ...) => (VPINSRD128 ...)
(SetElemUint64x2 ...) => (VPINSRQ128 ...)
(ShiftAllLeftInt16x8 ...) => (VPSLLW128 ...)
(ShiftAllLeftInt16x16 ...) => (VPSLLW256 ...)
(ShiftAllLeftInt32x4 ...) => (VPSLLD128 ...)
@ -1251,24 +1251,24 @@
(ShiftAllLeftUint64x2 ...) => (VPSLLQ128 ...)
(ShiftAllLeftUint64x4 ...) => (VPSLLQ256 ...)
(ShiftAllLeftUint64x8 ...) => (VPSLLQ512 ...)
(ShiftAllLeftAndFillUpperFromInt16x8 [a] x y) => (VPSHLDW128 [a] x y)
(ShiftAllLeftAndFillUpperFromInt16x16 [a] x y) => (VPSHLDW256 [a] x y)
(ShiftAllLeftAndFillUpperFromInt16x32 [a] x y) => (VPSHLDW512 [a] x y)
(ShiftAllLeftAndFillUpperFromInt32x4 [a] x y) => (VPSHLDD128 [a] x y)
(ShiftAllLeftAndFillUpperFromInt32x8 [a] x y) => (VPSHLDD256 [a] x y)
(ShiftAllLeftAndFillUpperFromInt32x16 [a] x y) => (VPSHLDD512 [a] x y)
(ShiftAllLeftAndFillUpperFromInt64x2 [a] x y) => (VPSHLDQ128 [a] x y)
(ShiftAllLeftAndFillUpperFromInt64x4 [a] x y) => (VPSHLDQ256 [a] x y)
(ShiftAllLeftAndFillUpperFromInt64x8 [a] x y) => (VPSHLDQ512 [a] x y)
(ShiftAllLeftAndFillUpperFromUint16x8 [a] x y) => (VPSHLDW128 [a] x y)
(ShiftAllLeftAndFillUpperFromUint16x16 [a] x y) => (VPSHLDW256 [a] x y)
(ShiftAllLeftAndFillUpperFromUint16x32 [a] x y) => (VPSHLDW512 [a] x y)
(ShiftAllLeftAndFillUpperFromUint32x4 [a] x y) => (VPSHLDD128 [a] x y)
(ShiftAllLeftAndFillUpperFromUint32x8 [a] x y) => (VPSHLDD256 [a] x y)
(ShiftAllLeftAndFillUpperFromUint32x16 [a] x y) => (VPSHLDD512 [a] x y)
(ShiftAllLeftAndFillUpperFromUint64x2 [a] x y) => (VPSHLDQ128 [a] x y)
(ShiftAllLeftAndFillUpperFromUint64x4 [a] x y) => (VPSHLDQ256 [a] x y)
(ShiftAllLeftAndFillUpperFromUint64x8 [a] x y) => (VPSHLDQ512 [a] x y)
(ShiftAllLeftAndFillUpperFromInt16x8 ...) => (VPSHLDW128 ...)
(ShiftAllLeftAndFillUpperFromInt16x16 ...) => (VPSHLDW256 ...)
(ShiftAllLeftAndFillUpperFromInt16x32 ...) => (VPSHLDW512 ...)
(ShiftAllLeftAndFillUpperFromInt32x4 ...) => (VPSHLDD128 ...)
(ShiftAllLeftAndFillUpperFromInt32x8 ...) => (VPSHLDD256 ...)
(ShiftAllLeftAndFillUpperFromInt32x16 ...) => (VPSHLDD512 ...)
(ShiftAllLeftAndFillUpperFromInt64x2 ...) => (VPSHLDQ128 ...)
(ShiftAllLeftAndFillUpperFromInt64x4 ...) => (VPSHLDQ256 ...)
(ShiftAllLeftAndFillUpperFromInt64x8 ...) => (VPSHLDQ512 ...)
(ShiftAllLeftAndFillUpperFromUint16x8 ...) => (VPSHLDW128 ...)
(ShiftAllLeftAndFillUpperFromUint16x16 ...) => (VPSHLDW256 ...)
(ShiftAllLeftAndFillUpperFromUint16x32 ...) => (VPSHLDW512 ...)
(ShiftAllLeftAndFillUpperFromUint32x4 ...) => (VPSHLDD128 ...)
(ShiftAllLeftAndFillUpperFromUint32x8 ...) => (VPSHLDD256 ...)
(ShiftAllLeftAndFillUpperFromUint32x16 ...) => (VPSHLDD512 ...)
(ShiftAllLeftAndFillUpperFromUint64x2 ...) => (VPSHLDQ128 ...)
(ShiftAllLeftAndFillUpperFromUint64x4 ...) => (VPSHLDQ256 ...)
(ShiftAllLeftAndFillUpperFromUint64x8 ...) => (VPSHLDQ512 ...)
(ShiftAllLeftAndFillUpperFromMaskedInt16x8 [a] x y mask) => (VPSHLDWMasked128 [a] x y (VPMOVVec16x8ToM <types.TypeMask> mask))
(ShiftAllLeftAndFillUpperFromMaskedInt16x16 [a] x y mask) => (VPSHLDWMasked256 [a] x y (VPMOVVec16x16ToM <types.TypeMask> mask))
(ShiftAllLeftAndFillUpperFromMaskedInt16x32 [a] x y mask) => (VPSHLDWMasked512 [a] x y (VPMOVVec16x32ToM <types.TypeMask> mask))
@ -1307,24 +1307,24 @@
(ShiftAllRightUint64x2 ...) => (VPSRLQ128 ...)
(ShiftAllRightUint64x4 ...) => (VPSRLQ256 ...)
(ShiftAllRightUint64x8 ...) => (VPSRLQ512 ...)
(ShiftAllRightAndFillUpperFromInt16x8 [a] x y) => (VPSHRDW128 [a] x y)
(ShiftAllRightAndFillUpperFromInt16x16 [a] x y) => (VPSHRDW256 [a] x y)
(ShiftAllRightAndFillUpperFromInt16x32 [a] x y) => (VPSHRDW512 [a] x y)
(ShiftAllRightAndFillUpperFromInt32x4 [a] x y) => (VPSHRDD128 [a] x y)
(ShiftAllRightAndFillUpperFromInt32x8 [a] x y) => (VPSHRDD256 [a] x y)
(ShiftAllRightAndFillUpperFromInt32x16 [a] x y) => (VPSHRDD512 [a] x y)
(ShiftAllRightAndFillUpperFromInt64x2 [a] x y) => (VPSHRDQ128 [a] x y)
(ShiftAllRightAndFillUpperFromInt64x4 [a] x y) => (VPSHRDQ256 [a] x y)
(ShiftAllRightAndFillUpperFromInt64x8 [a] x y) => (VPSHRDQ512 [a] x y)
(ShiftAllRightAndFillUpperFromUint16x8 [a] x y) => (VPSHRDW128 [a] x y)
(ShiftAllRightAndFillUpperFromUint16x16 [a] x y) => (VPSHRDW256 [a] x y)
(ShiftAllRightAndFillUpperFromUint16x32 [a] x y) => (VPSHRDW512 [a] x y)
(ShiftAllRightAndFillUpperFromUint32x4 [a] x y) => (VPSHRDD128 [a] x y)
(ShiftAllRightAndFillUpperFromUint32x8 [a] x y) => (VPSHRDD256 [a] x y)
(ShiftAllRightAndFillUpperFromUint32x16 [a] x y) => (VPSHRDD512 [a] x y)
(ShiftAllRightAndFillUpperFromUint64x2 [a] x y) => (VPSHRDQ128 [a] x y)
(ShiftAllRightAndFillUpperFromUint64x4 [a] x y) => (VPSHRDQ256 [a] x y)
(ShiftAllRightAndFillUpperFromUint64x8 [a] x y) => (VPSHRDQ512 [a] x y)
(ShiftAllRightAndFillUpperFromInt16x8 ...) => (VPSHRDW128 ...)
(ShiftAllRightAndFillUpperFromInt16x16 ...) => (VPSHRDW256 ...)
(ShiftAllRightAndFillUpperFromInt16x32 ...) => (VPSHRDW512 ...)
(ShiftAllRightAndFillUpperFromInt32x4 ...) => (VPSHRDD128 ...)
(ShiftAllRightAndFillUpperFromInt32x8 ...) => (VPSHRDD256 ...)
(ShiftAllRightAndFillUpperFromInt32x16 ...) => (VPSHRDD512 ...)
(ShiftAllRightAndFillUpperFromInt64x2 ...) => (VPSHRDQ128 ...)
(ShiftAllRightAndFillUpperFromInt64x4 ...) => (VPSHRDQ256 ...)
(ShiftAllRightAndFillUpperFromInt64x8 ...) => (VPSHRDQ512 ...)
(ShiftAllRightAndFillUpperFromUint16x8 ...) => (VPSHRDW128 ...)
(ShiftAllRightAndFillUpperFromUint16x16 ...) => (VPSHRDW256 ...)
(ShiftAllRightAndFillUpperFromUint16x32 ...) => (VPSHRDW512 ...)
(ShiftAllRightAndFillUpperFromUint32x4 ...) => (VPSHRDD128 ...)
(ShiftAllRightAndFillUpperFromUint32x8 ...) => (VPSHRDD256 ...)
(ShiftAllRightAndFillUpperFromUint32x16 ...) => (VPSHRDD512 ...)
(ShiftAllRightAndFillUpperFromUint64x2 ...) => (VPSHRDQ128 ...)
(ShiftAllRightAndFillUpperFromUint64x4 ...) => (VPSHRDQ256 ...)
(ShiftAllRightAndFillUpperFromUint64x8 ...) => (VPSHRDQ512 ...)
(ShiftAllRightAndFillUpperFromMaskedInt16x8 [a] x y mask) => (VPSHRDWMasked128 [a] x y (VPMOVVec16x8ToM <types.TypeMask> mask))
(ShiftAllRightAndFillUpperFromMaskedInt16x16 [a] x y mask) => (VPSHRDWMasked256 [a] x y (VPMOVVec16x16ToM <types.TypeMask> mask))
(ShiftAllRightAndFillUpperFromMaskedInt16x32 [a] x y mask) => (VPSHRDWMasked512 [a] x y (VPMOVVec16x32ToM <types.TypeMask> mask))

File diff suppressed because it is too large Load diff