internal/simdgen/ops/MLOps/categories.yaml

!sum
- go: PairDotProd
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // PairDotProd multiplies the elements and add the pairs together,
    // yielding a vector of half as many elements with twice the input element size.
- go: PairDotProdMasked
  masked: true
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // PairDotProdMasked 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 |-
    // SaturatedUnsignedSignedPairDotProd multiplies the elements and add the pairs together with saturation,
    // yielding a vector of half as many elements with twice the input element size.
- go: SaturatedUnsignedSignedPairDotProdMasked
  masked: true
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // SaturatedUnsignedSignedPairDotProdMasked 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 |-
    // DotProdBroadcast multiplies all elements and broadcasts the sum.
- go: UnsignedSignedQuadDotProdAccumulate
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // UnsignedSignedQuadDotProdAccumulate performs dot products on groups of 4 elements of x and y and then adds z.
- go: UnsignedSignedQuadDotProdAccumulateMasked
  masked: true
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // UnsignedSignedQuadDotProdAccumulateMasked performs dot products on groups of 4 elements of x and y and then adds z.
- go: SaturatedUnsignedSignedQuadDotProdAccumulate
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // SaturatedUnsignedSignedQuadDotProdAccumulate multiplies performs dot products on groups of 4 elements of x and y and then adds z.
- go: SaturatedUnsignedSignedQuadDotProdAccumulateMasked
  masked: true
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // SaturatedUnsignedSignedQuadDotProdAccumulateMasked multiplies performs dot products on groups of 4 elements of x and y and then adds z.
- go: PairDotProdAccumulate
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // PairDotProdAccumulate performs dot products on pairs of elements of x and y and then adds z.
- go: PairDotProdAccumulateMasked
  masked: true
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // PairDotProdAccumulateMasked performs dot products on pairs of elements of x and y and then adds z.
- go: SaturatedPairDotProdAccumulate
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // SaturatedPairDotProdAccumulate performs dot products on pairs of elements of x and y and then adds z.
- go: SaturatedPairDotProdAccumulateMasked
  masked: true
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // SaturatedPairDotProdAccumulateMasked performs dot products on pairs of elements of x and y and then adds z.
- go: FusedMultiplyAddMasked
  masked: true
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // FusedMultiplyAddMasked performs (x * y) + z.
- go: FusedMultiplyAddSubMasked
  masked: true
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // FusedMultiplyAddSubMasked performs (x * y) - z for odd-indexed elements, and (x * y) + z for even-indexed elements.
- go: FusedMultiplySubAddMasked
  masked: true
  commutative: false
  extension: "AVX.*"
  documentation: !string |-
    // FusedMultiplySubAddMasked performs (x * y) + z for odd-indexed elements, and (x * y) - z for even-indexed elements.