src/simd/archsimd/_gen/simdgen/ops/Moves/go_arm64.yaml

!sum

# GetElem VMOV (UMOV, integer)
- go: GetElem
  asm: "VMOV"
  SSAVariant: "extr"
  in:
  - &immIndex
    class: immediate
    go: Immediate
    name: index
  - &vector
    go: $t
    base: int|uint
  out:
  - &scalar
    class: greg

# GetElem VDUP (duplicate element into scalar float in vreg)
- go: GetElem
  asm: "VDUP"
  SSAVariant: "extr"
  in:
  - *immIndex
  - &fvector
    go: $t
    base: float
  out:
  - &fscalarOut
    class: vreg
    treatLikeAScalarOfSize: 0
    OverwriteClass: greg
    OverwriteBase: float

# SetElem VMOV (INS, integer)
- go: SetElem
  asm: "VMOV"
  SSAVariant: "ins"
  in:
  - *immIndex
  - *vector
  - *scalar
  out:
  - *vector

# SetElem VMOV (float: INS from the lane 0 of the source)
- go: SetElem
  asm: "VMOV"
  SSAVariant: "ins0"
  specialLower: !string "match (%h [0] (VMOVI16B [0]) y:(VDUP%sextr [i] _)) => y"
  in:
  - *immIndex
  - &imm0
    class: immediate
    go: Immediate
    const: 0 # imm+vreg with treatLikeAScalarOfSize are treated as scalar in lane 0
  - *fvector
  - &fscalarIn
    class: vreg
    treatLikeAScalarOfSize: 0
  out:
  - *fvector

# broadcast1To16 VDUP (duplicate element 0 to all 16 byte lanes)
- go: broadcast1To16
  asm: "VDUP"
  SSAVariant: "bcast"
  constImm: 0
  specialLower: !string "match (%h [i] (VMOV%sins [j] _ (MOVDconst [c]))) && i == j && c>=-128 && c<=255 => (VMOVI%a [uint8(c)])"
  in:
  - &immLane0
    name: "@" # tag "fully constant" (no aux field needed)
    const: 0
    class: immediate
    go: Immediate
  - &bcast8
    go: $t
    bits: 128
    elemBits: 8
  out:
  - *bcast8

# broadcast1To8 VDUP (duplicate element 0 to all 8 halfword lanes)
- go: broadcast1To8
  asm: "VDUP"
  SSAVariant: "bcast"
  constImm: 0
  in:
  - *immLane0
  - &bcast16
    go: $t
    bits: 128
    elemBits: 16
  out:
  - *bcast16

# broadcast1To4 VDUP (duplicate element 0 to all 4 word lanes)
- go: broadcast1To4
  asm: "VDUP"
  SSAVariant: "bcast"
  constImm: 0
  in:
  - *immLane0
  - &bcast32
    go: $t
    bits: 128
    elemBits: 32
  out:
  - *bcast32

# broadcast1To2 VDUP (duplicate element 0 to both doubleword lanes)
- go: broadcast1To2
  asm: "VDUP"
  SSAVariant: "bcast"
  constImm: 0
  in:
  - *immLane0
  - &bcast64
    go: $t
    bits: 128
    elemBits: 64
  out:
  - *bcast64

# bitSelect — VBIT (bit insert if true, internal)
# Only Int8x16 is needed since VBIT operates on the full 128-bit register.
# Other types cast to Int8x16 before calling bitSelect.
- go: bitSelect
  asm: "VBIT"
  specialLower: !string "match (%h x y (VNOT16B mask)) => (VBIF16B x y mask)"
  in:
  - &v8
    go: $t
    class: vreg
    base: int
    elemBits: 8
  - *v8
  - &maskReg
    class: vreg
    base: int
    name: mask
  out:
  - *v8

# bitSelectNot — VBIF (bit insert if false, internal)
- go: bitSelectNot
  asm: "VBIF"
  specialLower: !string "match (%h x y (VNOT16B mask)) => (VBIT16B x y mask)"
  in:
  - *v8
  - *v8
  - *maskReg
  out:
  - *v8

# LookupOrZero — TBL (table lookup, zeros for out-of-range indices)
- go: LookupOrZero
  regexpTag: "move"
  asm: "VTBL"
  in:
  - &bytesTable
    go: $t
    base: int|uint
    elemBits: 8
    name: table
  - &byteIndices
    go: $t
    base: int|uint
    elemBits: 8
    name: indices
  out:
  - &bytes
    go: $t
    base: int|uint
    elemBits: 8

# LookupOrKeep — TBX (table lookup extended, keeps original for out-of-range)
- go: lookupOrKeep
  noGenericOps: "true"
  regexpTag: "move"
  asm: "VTBX"
  in:
  - &bytesKeep
    go: $t
    base: int|uint
    elemBits: 8
    name: keep
  - *bytesTable
  - *byteIndices
  out:
  - *bytes

# ConcatShiftBytesRight — EXT (extract bytes from concatenated pair)
- go: ConcatShiftBytesRight
  asm: "VEXT"
  operandOrder: 2I
  in:
  - class: immediate
    immOffset: 0
    name: shift
  - &uint8x128
    go: $t
    base: uint
    bits: 128
    elemBits: 8
  - *uint8x128
  out:
  - *uint8x128

# InterleaveLo — ZIP1 (interleave low halves)
- go: InterleaveLo
  asm: "VZIP1"
  in:
  - &any128
    go: $t
    bits: 128
  - *any128
  out:
  - *any128

# InterleaveHi — ZIP2 (interleave high halves)
- go: InterleaveHi
  asm: "VZIP2"
  in:
  - *any128
  - *any128
  out:
  - *any128

# ConcatEven — UZP1 (extract even-indexed elements)
- go: ConcatEven
  asm: "VUZP1"
  in:
  - *any128
  - *any128
  out:
  - *any128

# ConcatOdd — UZP2 (extract odd-indexed elements)
- go: ConcatOdd
  asm: "VUZP2"
  in:
  - *any128
  - *any128
  out:
  - *any128

# InterleaveEven — TRN1 (transpose even-position pairs)
- go: InterleaveEven
  asm: "VTRN1"
  in:
  - *any128
  - *any128
  out:
  - *any128

# InterleaveOdd — TRN2 (transpose odd-position pairs)
- go: InterleaveOdd
  asm: "VTRN2"
  in:
  - *any128
  - *any128
  out:
  - *any128