| !sum |
| - go: Add |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME adds corresponding elements of two vectors. |
| - go: AddSaturated |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME adds corresponding elements of two vectors with saturation. |
| - go: Sub |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME subtracts corresponding elements of two vectors. |
| - go: SubSaturated |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME subtracts corresponding elements of two vectors with saturation. |
| - go: AddPairs |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME horizontally adds adjacent pairs of elements. |
| // For x = [x0, x1, x2, x3, ...] and y = [y0, y1, y2, y3, ...], the result is [y0+y1, y2+y3, ..., x0+x1, x2+x3, ...]. |
| - go: SubPairs |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME horizontally subtracts adjacent pairs of elements. |
| // For x = [x0, x1, x2, x3, ...] and y = [y0, y1, y2, y3, ...], the result is [y0-y1, y2-y3, ..., x0-x1, x2-x3, ...]. |
| - go: AddPairsSaturated |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME horizontally adds adjacent pairs of elements with saturation. |
| // For x = [x0, x1, x2, x3, ...] and y = [y0, y1, y2, y3, ...], the result is [y0+y1, y2+y3, ..., x0+x1, x2+x3, ...]. |
| - go: SubPairsSaturated |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME horizontally subtracts adjacent pairs of elements with saturation. |
| // For x = [x0, x1, x2, x3, ...] and y = [y0, y1, y2, y3, ...], the result is [y0-y1, y2-y3, ..., x0-x1, x2-x3, ...]. |
| - go: And |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs a bitwise AND operation between two vectors. |
| - go: Or |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs a bitwise OR operation between two vectors. |
| - go: AndNot |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs a bitwise x &^ y. |
| - go: Xor |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs a bitwise XOR operation between two vectors. |
| |
| # We also have PTEST and VPTERNLOG, those should be hidden from the users |
| # and only appear in rewrite rules. |
| # const imm predicate(holds for both float and int|uint): |
| # 0: Equal |
| # 1: Less |
| # 2: LessEqual |
| # 4: NotEqual |
| # 5: GreaterEqual |
| # 6: Greater |
| - go: Equal |
| constImm: 0 |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME compares for equality. |
| - go: Less |
| constImm: 1 |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME compares for less than. |
| - go: LessEqual |
| constImm: 2 |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME compares for less than or equal. |
| - go: IsNan # For float only. |
| constImm: 3 |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME checks if elements are NaN. Use as x.IsNan(x). |
| - go: NotEqual |
| constImm: 4 |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME compares for inequality. |
| - go: GreaterEqual |
| constImm: 13 |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME compares for greater than or equal. |
| - go: Greater |
| constImm: 14 |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME compares for greater than. |
| - go: ConvertToInt32 |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // ConvertToInt32 converts element values to int32. |
| |
| - go: ConvertToUint32 |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // ConvertToUint32Masked converts element values to uint32. |
| - go: Div |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME divides elements of two vectors. |
| - go: Sqrt |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME computes the square root of each element. |
| - go: ApproximateReciprocal |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME computes an approximate reciprocal of each element. |
| - go: ApproximateReciprocalOfSqrt |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME computes an approximate reciprocal of the square root of each element. |
| - go: Scale |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME multiplies elements by a power of 2. |
| - go: Round |
| commutative: false |
| extension: "AVX.*" |
| constImm: 0 |
| documentation: !string |- |
| // NAME rounds elements to the nearest integer. |
| - go: RoundScaled |
| commutative: false |
| extension: "AVX.*" |
| constImm: 0 |
| documentation: !string |- |
| // NAME rounds elements with specified precision. |
| - go: RoundScaledResidue |
| commutative: false |
| extension: "AVX.*" |
| constImm: 0 |
| documentation: !string |- |
| // NAME computes the difference after rounding with specified precision. |
| - go: Floor |
| commutative: false |
| extension: "AVX.*" |
| constImm: 1 |
| documentation: !string |- |
| // NAME rounds elements down to the nearest integer. |
| - go: FloorScaled |
| commutative: false |
| extension: "AVX.*" |
| constImm: 1 |
| documentation: !string |- |
| // NAME rounds elements down with specified precision. |
| - go: FloorScaledResidue |
| commutative: false |
| extension: "AVX.*" |
| constImm: 1 |
| documentation: !string |- |
| // NAME computes the difference after flooring with specified precision. |
| - go: Ceil |
| commutative: false |
| extension: "AVX.*" |
| constImm: 2 |
| documentation: !string |- |
| // NAME rounds elements up to the nearest integer. |
| - go: CeilScaled |
| commutative: false |
| extension: "AVX.*" |
| constImm: 2 |
| documentation: !string |- |
| // NAME rounds elements up with specified precision. |
| - go: CeilScaledResidue |
| commutative: false |
| extension: "AVX.*" |
| constImm: 2 |
| documentation: !string |- |
| // NAME computes the difference after ceiling with specified precision. |
| - go: Trunc |
| commutative: false |
| extension: "AVX.*" |
| constImm: 3 |
| documentation: !string |- |
| // NAME truncates elements towards zero. |
| - go: TruncScaled |
| commutative: false |
| extension: "AVX.*" |
| constImm: 3 |
| documentation: !string |- |
| // NAME truncates elements with specified precision. |
| - go: TruncScaledResidue |
| commutative: false |
| extension: "AVX.*" |
| constImm: 3 |
| documentation: !string |- |
| // NAME computes the difference after truncating with specified precision. |
| - go: AddSub |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME subtracts even elements and adds odd elements of two vectors. |
| - go: GaloisFieldAffineTransform |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME computes an affine transformation in GF(2^8): |
| // x is a vector of 8-bit vectors, with each adjacent 8 as a group; y is a vector of 8x8 1-bit matrixes; |
| // b is an 8-bit vector. The affine transformation is y * x + b, with each element of y |
| // corresponding to a group of 8 elements in x. |
| - go: GaloisFieldAffineTransformInverse |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME computes an affine transformation in GF(2^8), |
| // with x inverted with respect to reduction polynomial x^8 + x^4 + x^3 + x + 1: |
| // x is a vector of 8-bit vectors, with each adjacent 8 as a group; y is a vector of 8x8 1-bit matrixes; |
| // b is an 8-bit vector. The affine transformation is y * x + b, with each element of y |
| // corresponding to a group of 8 elements in x. |
| - go: GaloisFieldMul |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME computes element-wise GF(2^8) multiplication with |
| // reduction polynomial x^8 + x^4 + x^3 + x + 1. |
| - go: Average |
| commutative: true |
| extension: "AVX.*" # VPAVGB/W are available across various AVX versions |
| documentation: !string |- |
| // NAME computes the rounded average of corresponding elements. |
| - go: Absolute |
| commutative: false |
| # Unary operation, not commutative |
| extension: "AVX.*" # VPABSB/W/D are AVX, VPABSQ is AVX512 |
| documentation: !string |- |
| // NAME computes the absolute value of each element. |
| - go: Sign |
| # Applies sign of second operand to first: sign(val, sign_src) |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME returns the product of the first operand with -1, 0, or 1, |
| // whichever constant is nearest to the value of the second operand. |
| # Sign does not have masked version |
| - go: PopCount |
| commutative: false |
| extension: "AVX512.*" |
| documentation: !string |- |
| // NAME counts the number of set bits in each element. |
| - go: PairDotProd |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME multiplies the elements and add the pairs together, |
| // yielding a vector of half as many elements with twice the input element size. |
| # TODO: maybe simplify this name within the receiver-type + method-naming scheme we use. |
| - go: SaturatedUnsignedSignedPairDotProd |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME multiplies the elements and add the pairs together with saturation, |
| // yielding a vector of half as many elements with twice the input element size. |
| # QuadDotProd, i.e. VPDPBUSD(S) are operations with src/dst on the same register, we are not supporting this as of now. |
| - go: DotProdBroadcast |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME multiplies all elements and broadcasts the sum. |
| - go: UnsignedSignedQuadDotProdAccumulate |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs dot products on groups of 4 elements of x and y and then adds z. |
| - go: SaturatedUnsignedSignedQuadDotProdAccumulate |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME multiplies performs dot products on groups of 4 elements of x and y and then adds z. |
| - go: AddDotProd |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs dot products on pairs of elements of y and z and then adds x. |
| - go: SaturatedAddDotProd |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs dot products on pairs of elements of y and z and then adds x. |
| - go: FusedMultiplyAdd |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs (x * y) + z. |
| - go: FusedMultiplyAddSub |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs (x * y) - z for odd-indexed elements, and (x * y) + z for even-indexed elements. |
| - go: FusedMultiplySubAdd |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs (x * y) + z for odd-indexed elements, and (x * y) - z for even-indexed elements. |
| - go: Max |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME computes the maximum of corresponding elements. |
| - go: Min |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME computes the minimum of corresponding elements. |
| - go: SetElem |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME sets a single constant-indexed element's value. |
| - go: GetElem |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME retrieves a single constant-indexed element's value. |
| - go: Set128 |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME combines a 128-bit vector with a 256-bit vector, where the constant operand specifies whether the low (0) or high (1) half is receives the smaller vector. |
| - go: Get128 |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME retrieves the upper (1) or lower (0) half of a 256-bit vector, depending on the constant operand. |
| - go: Permute |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs a full permutation of vector x using indices: |
| // result := {x[indices[0]], x[indices[1]], ..., x[indices[n]]} |
| // Only the needed bits to represent x's index are used in indices' elements. |
| - go: Permute2 # Permute2 is only available on or after AVX512 |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs a full permutation of vector x, y using indices: |
| // result := {xy[indices[0]], xy[indices[1]], ..., xy[indices[n]]} |
| // where xy is x appending y. |
| // Only the needed bits to represent xy's index are used in indices' elements. |
| - go: Compress |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME performs a compression on vector x using mask by |
| // selecting elements as indicated by mask, and pack them to lower indexed elements. |
| - go: blend |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME blends two vectors based on mask values, choosing either |
| // the first or the second based on whether the third is false or true |
| - go: Mul |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME multiplies corresponding elements of two vectors. |
| - go: MulEvenWiden |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME multiplies even-indexed elements, widening the result. |
| // Result[i] = v1.Even[i] * v2.Even[i]. |
| - go: MulHigh |
| commutative: true |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME multiplies elements and stores the high part of the result. |
| - go: ShiftAllLeft |
| nameAndSizeCheck: true |
| specialLower: sftimm |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME shifts each element to the left by the specified number of bits. Emptied lower bits are zeroed. |
| - go: ShiftAllRight |
| signed: false |
| nameAndSizeCheck: true |
| specialLower: sftimm |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME shifts each element to the right by the specified number of bits. Emptied upper bits are zeroed. |
| - go: ShiftAllRight |
| signed: true |
| specialLower: sftimm |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME shifts each element to the right by the specified number of bits. Emptied upper bits are filled with the sign bit. |
| - go: shiftAllLeftConst # no APIs, only ssa ops. |
| noTypes: "true" |
| noGenericOps: "true" |
| SSAVariant: "const" # to avoid its name colliding with reg version of this instruction, amend this to its ssa op name. |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| - go: shiftAllRightConst # no APIs, only ssa ops. |
| noTypes: "true" |
| noGenericOps: "true" |
| SSAVariant: "const" |
| signed: false |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| - go: shiftAllRightConst # no APIs, only ssa ops. |
| noTypes: "true" |
| noGenericOps: "true" |
| SSAVariant: "const" |
| signed: true |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| |
| - go: ShiftLeft |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME shifts each element in x to the left by the number of bits specified in y's corresponding elements. Emptied lower bits are zeroed. |
| - go: ShiftRight |
| signed: false |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME shifts each element in x to the right by the number of bits specified in y's corresponding elements. Emptied upper bits are zeroed. |
| - go: ShiftRight |
| signed: true |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME shifts each element in x to the right by the number of bits specified in y's corresponding elements. Emptied upper bits are filled with the sign bit. |
| - go: RotateAllLeft |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME rotates each element to the left by the number of bits specified by the immediate. |
| - go: RotateLeft |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME rotates each element in x to the left by the number of bits specified by y's corresponding elements. |
| - go: RotateAllRight |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME rotates each element to the right by the number of bits specified by the immediate. |
| - go: RotateRight |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME rotates each element in x to the right by the number of bits specified by y's corresponding elements. |
| - go: ShiftAllLeftConcat |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME shifts each element of x to the left by the number of bits specified by the |
| // immediate(only the lower 5 bits are used), and then copies the upper bits of y to the emptied lower bits of the shifted x. |
| - go: ShiftAllRightConcat |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME shifts each element of x to the right by the number of bits specified by the |
| // immediate(only the lower 5 bits are used), and then copies the lower bits of y to the emptied upper bits of the shifted x. |
| - go: ShiftLeftConcat |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME shifts each element of x to the left by the number of bits specified by the |
| // corresponding elements in y(only the lower 5 bits are used), and then copies the upper bits of z to the emptied lower bits of the shifted x. |
| - go: ShiftRightConcat |
| nameAndSizeCheck: true |
| commutative: false |
| extension: "AVX.*" |
| documentation: !string |- |
| // NAME shifts each element of x to the right by the number of bits specified by the |
| // corresponding elements in y(only the lower 5 bits are used), and then copies the lower bits of z to the emptied upper bits of the shifted x. |
