; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc < %s -mtriple=arm64-eabi -aarch64-neon-syntax=apple | FileCheck %s

define <8 x i8> @sqsub8b(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: sqsub8b:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr d0, [x0]
; CHECK-NEXT:    ldr d1, [x1]
; CHECK-NEXT:    sqsub.8b v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <8 x i8>, ptr %A
  %tmp2 = load <8 x i8>, ptr %B
  %tmp3 = call <8 x i8> @llvm.aarch64.neon.sqsub.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
  ret <8 x i8> %tmp3
}

define <4 x i16> @sqsub4h(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: sqsub4h:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr d0, [x0]
; CHECK-NEXT:    ldr d1, [x1]
; CHECK-NEXT:    sqsub.4h v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <4 x i16>, ptr %A
  %tmp2 = load <4 x i16>, ptr %B
  %tmp3 = call <4 x i16> @llvm.aarch64.neon.sqsub.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
  ret <4 x i16> %tmp3
}

define <2 x i32> @sqsub2s(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: sqsub2s:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr d0, [x0]
; CHECK-NEXT:    ldr d1, [x1]
; CHECK-NEXT:    sqsub.2s v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <2 x i32>, ptr %A
  %tmp2 = load <2 x i32>, ptr %B
  %tmp3 = call <2 x i32> @llvm.aarch64.neon.sqsub.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
  ret <2 x i32> %tmp3
}

define <1 x i64> @sqsub1d(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: sqsub1d:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr d0, [x0]
; CHECK-NEXT:    ldr d1, [x1]
; CHECK-NEXT:    sqsub d0, d0, d1
; CHECK-NEXT:    ret
  %tmp1 = load <1 x i64>, ptr %A
  %tmp2 = load <1 x i64>, ptr %B
  %tmp3 = call <1 x i64> @llvm.aarch64.neon.sqsub.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
  ret <1 x i64> %tmp3
}


define <8 x i8> @uqsub8b(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: uqsub8b:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr d0, [x0]
; CHECK-NEXT:    ldr d1, [x1]
; CHECK-NEXT:    uqsub.8b v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <8 x i8>, ptr %A
  %tmp2 = load <8 x i8>, ptr %B
  %tmp3 = call <8 x i8> @llvm.aarch64.neon.uqsub.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
  ret <8 x i8> %tmp3
}

define <4 x i16> @uqsub4h(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: uqsub4h:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr d0, [x0]
; CHECK-NEXT:    ldr d1, [x1]
; CHECK-NEXT:    uqsub.4h v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <4 x i16>, ptr %A
  %tmp2 = load <4 x i16>, ptr %B
  %tmp3 = call <4 x i16> @llvm.aarch64.neon.uqsub.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
  ret <4 x i16> %tmp3
}

define <2 x i32> @uqsub2s(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: uqsub2s:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr d0, [x0]
; CHECK-NEXT:    ldr d1, [x1]
; CHECK-NEXT:    uqsub.2s v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <2 x i32>, ptr %A
  %tmp2 = load <2 x i32>, ptr %B
  %tmp3 = call <2 x i32> @llvm.aarch64.neon.uqsub.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
  ret <2 x i32> %tmp3
}

define <1 x i64> @uqsub1d(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: uqsub1d:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr d0, [x0]
; CHECK-NEXT:    ldr d1, [x1]
; CHECK-NEXT:    uqsub d0, d0, d1
; CHECK-NEXT:    ret
  %tmp1 = load <1 x i64>, ptr %A
  %tmp2 = load <1 x i64>, ptr %B
  %tmp3 = call <1 x i64> @llvm.aarch64.neon.uqsub.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
  ret <1 x i64> %tmp3
}


define <16 x i8> @sqsub16b(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: sqsub16b:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr q0, [x0]
; CHECK-NEXT:    ldr q1, [x1]
; CHECK-NEXT:    sqsub.16b v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <16 x i8>, ptr %A
  %tmp2 = load <16 x i8>, ptr %B
  %tmp3 = call <16 x i8> @llvm.aarch64.neon.sqsub.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
  ret <16 x i8> %tmp3
}

define <8 x i16> @sqsub8h(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: sqsub8h:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr q0, [x0]
; CHECK-NEXT:    ldr q1, [x1]
; CHECK-NEXT:    sqsub.8h v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <8 x i16>, ptr %A
  %tmp2 = load <8 x i16>, ptr %B
  %tmp3 = call <8 x i16> @llvm.aarch64.neon.sqsub.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
  ret <8 x i16> %tmp3
}

define <4 x i32> @sqsub4s(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: sqsub4s:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr q0, [x0]
; CHECK-NEXT:    ldr q1, [x1]
; CHECK-NEXT:    sqsub.4s v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <4 x i32>, ptr %A
  %tmp2 = load <4 x i32>, ptr %B
  %tmp3 = call <4 x i32> @llvm.aarch64.neon.sqsub.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
  ret <4 x i32> %tmp3
}

define <2 x i64> @sqsub2d(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: sqsub2d:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr q0, [x0]
; CHECK-NEXT:    ldr q1, [x1]
; CHECK-NEXT:    sqsub.2d v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <2 x i64>, ptr %A
  %tmp2 = load <2 x i64>, ptr %B
  %tmp3 = call <2 x i64> @llvm.aarch64.neon.sqsub.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
  ret <2 x i64> %tmp3
}


define <16 x i8> @uqsub16b(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: uqsub16b:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr q0, [x0]
; CHECK-NEXT:    ldr q1, [x1]
; CHECK-NEXT:    uqsub.16b v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <16 x i8>, ptr %A
  %tmp2 = load <16 x i8>, ptr %B
  %tmp3 = call <16 x i8> @llvm.aarch64.neon.uqsub.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
  ret <16 x i8> %tmp3
}

define <8 x i16> @uqsub8h(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: uqsub8h:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr q0, [x0]
; CHECK-NEXT:    ldr q1, [x1]
; CHECK-NEXT:    uqsub.8h v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <8 x i16>, ptr %A
  %tmp2 = load <8 x i16>, ptr %B
  %tmp3 = call <8 x i16> @llvm.aarch64.neon.uqsub.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
  ret <8 x i16> %tmp3
}

define <4 x i32> @uqsub4s(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: uqsub4s:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr q0, [x0]
; CHECK-NEXT:    ldr q1, [x1]
; CHECK-NEXT:    uqsub.4s v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <4 x i32>, ptr %A
  %tmp2 = load <4 x i32>, ptr %B
  %tmp3 = call <4 x i32> @llvm.aarch64.neon.uqsub.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
  ret <4 x i32> %tmp3
}

define <2 x i64> @uqsub2d(ptr %A, ptr %B) nounwind {
; CHECK-LABEL: uqsub2d:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr q0, [x0]
; CHECK-NEXT:    ldr q1, [x1]
; CHECK-NEXT:    uqsub.2d v0, v0, v1
; CHECK-NEXT:    ret
  %tmp1 = load <2 x i64>, ptr %A
  %tmp2 = load <2 x i64>, ptr %B
  %tmp3 = call <2 x i64> @llvm.aarch64.neon.uqsub.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
  ret <2 x i64> %tmp3
}

declare <8 x i8>  @llvm.aarch64.neon.sqsub.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
declare <4 x i16> @llvm.aarch64.neon.sqsub.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
declare <2 x i32> @llvm.aarch64.neon.sqsub.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
declare <1 x i64> @llvm.aarch64.neon.sqsub.v1i64(<1 x i64>, <1 x i64>) nounwind readnone

declare <8 x i8>  @llvm.aarch64.neon.uqsub.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
declare <4 x i16> @llvm.aarch64.neon.uqsub.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
declare <2 x i32> @llvm.aarch64.neon.uqsub.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
declare <1 x i64> @llvm.aarch64.neon.uqsub.v1i64(<1 x i64>, <1 x i64>) nounwind readnone

declare <16 x i8> @llvm.aarch64.neon.sqsub.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
declare <8 x i16> @llvm.aarch64.neon.sqsub.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
declare <4 x i32> @llvm.aarch64.neon.sqsub.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
declare <2 x i64> @llvm.aarch64.neon.sqsub.v2i64(<2 x i64>, <2 x i64>) nounwind readnone

declare <16 x i8> @llvm.aarch64.neon.uqsub.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
declare <8 x i16> @llvm.aarch64.neon.uqsub.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
declare <4 x i32> @llvm.aarch64.neon.uqsub.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
declare <2 x i64> @llvm.aarch64.neon.uqsub.v2i64(<2 x i64>, <2 x i64>) nounwind readnone