; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc -mattr=+sve2,+dotprod < %s | FileCheck %s

target triple = "aarch64-unknown-linux-gnu"

define <vscale x 16 x i8> @sabs_nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
; CHECK-LABEL: sabs_nxv16i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ptrue p0.b
; CHECK-NEXT:    sabd z0.b, p0/m, z0.b, z1.b
; CHECK-NEXT:    ret
  %smax = tail call <vscale x 16 x i8> @llvm.smax.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
  %smin = tail call <vscale x 16 x i8> @llvm.smin.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
  %sabs = sub <vscale x 16 x i8> %smax, %smin
  ret <vscale x 16 x i8> %sabs
}

define <vscale x 16 x i8> @uabs_nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
; CHECK-LABEL: uabs_nxv16i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ptrue p0.b
; CHECK-NEXT:    uabd z0.b, p0/m, z0.b, z1.b
; CHECK-NEXT:    ret
  %umax = tail call <vscale x 16 x i8> @llvm.umax.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
  %umin = tail call <vscale x 16 x i8> @llvm.umin.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
  %uabs = sub <vscale x 16 x i8> %umax, %umin
  ret <vscale x 16 x i8> %uabs
}

; TODO: This case could be lowered to a sabal[bt] pair.
define <vscale x 8 x i16> @sabs_nxv16i8_wide_add(<vscale x 8 x i16> %acc, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
; CHECK-LABEL: sabs_nxv16i8_wide_add:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ptrue p0.b
; CHECK-NEXT:    sabd z1.b, p0/m, z1.b, z2.b
; CHECK-NEXT:    uaddwb z0.h, z0.h, z1.b
; CHECK-NEXT:    uaddwt z0.h, z0.h, z1.b
; CHECK-NEXT:    ret
  %smax = tail call <vscale x 16 x i8> @llvm.smax.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
  %smin = tail call <vscale x 16 x i8> @llvm.smin.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
  %sabs = sub <vscale x 16 x i8> %smax, %smin
  %ext = zext <vscale x 16 x i8> %sabs to <vscale x 16 x i16>
  %reduce = call <vscale x 8 x i16> @llvm.vector.partial.reduce.add.v4i32.v16i32(<vscale x 8 x i16> %acc, <vscale x 16 x i16> %ext)
  ret <vscale x 8 x i16>  %reduce
}

define <vscale x 4 x i32> @sabs_nxv16i8_dot(<vscale x 4 x i32> %acc, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
; CHECK-LABEL: sabs_nxv16i8_dot:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ptrue p0.b
; CHECK-NEXT:    sabd z1.b, p0/m, z1.b, z2.b
; CHECK-NEXT:    mov z2.b, #1 // =0x1
; CHECK-NEXT:    udot z0.s, z1.b, z2.b
; CHECK-NEXT:    ret
  %smax = tail call <vscale x 16 x i8> @llvm.smax.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
  %smin = tail call <vscale x 16 x i8> @llvm.smin.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
  %sabs = sub <vscale x 16 x i8> %smax, %smin
  %ext = zext <vscale x 16 x i8> %sabs to <vscale x 16 x i32>
  %dot = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<vscale x 4 x i32> %acc, <vscale x 16 x i32> %ext)
  ret <vscale x 4 x i32> %dot
}

define <16 x i8> @sabs_v16i8(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: sabs_v16i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    sabd v0.16b, v0.16b, v1.16b
; CHECK-NEXT:    ret
  %smax = tail call <16 x i8> @llvm.smax.v16i8(<16 x i8> %a, <16 x i8> %b)
  %smin = tail call <16 x i8> @llvm.smin.v16i8(<16 x i8> %a, <16 x i8> %b)
  %sabs = sub <16 x i8> %smax, %smin
  ret <16 x i8> %sabs
}

define <16 x i8> @uabs_v16i8(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: uabs_v16i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    uabd v0.16b, v0.16b, v1.16b
; CHECK-NEXT:    ret
  %umax = tail call <16 x i8> @llvm.umax.v16i8(<16 x i8> %a, <16 x i8> %b)
  %umin = tail call <16 x i8> @llvm.umin.v16i8(<16 x i8> %a, <16 x i8> %b)
  %uabs = sub <16 x i8> %umax, %umin
  ret <16 x i8> %uabs
}

; TODO: This case could be lowered to a uabal[bt] pair.
define <vscale x 4 x i32> @uabs_nxv16i8_wide_add(<vscale x 4 x i32> %acc, <vscale x 8 x i16> %a, <vscale x 8 x i16> %b) {
; CHECK-LABEL: uabs_nxv16i8_wide_add:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ptrue p0.h
; CHECK-NEXT:    uabd z1.h, p0/m, z1.h, z2.h
; CHECK-NEXT:    uaddwb z0.s, z0.s, z1.h
; CHECK-NEXT:    uaddwt z0.s, z0.s, z1.h
; CHECK-NEXT:    ret
  %umax = tail call <vscale x 8 x i16> @llvm.umax.nxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b)
  %umin = tail call <vscale x 8 x i16> @llvm.umin.nxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b)
  %uabs = sub <vscale x 8 x i16> %umax, %umin
  %ext = zext <vscale x 8 x i16> %uabs to <vscale x 8 x i32>
  %reduce = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<vscale x 4 x i32> %acc, <vscale x 8 x i32> %ext)
  ret <vscale x 4 x i32> %reduce
}

define <4 x i32> @uabs_v16i8_dot(<4 x i32> %acc, <16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: uabs_v16i8_dot:
; CHECK:       // %bb.0:
; CHECK-NEXT:    mov z3.b, #1 // =0x1
; CHECK-NEXT:    uabd v1.16b, v1.16b, v2.16b
; CHECK-NEXT:    // kill: def $q0 killed $q0 def $z0
; CHECK-NEXT:    udot z0.s, z1.b, z3.b
; CHECK-NEXT:    // kill: def $q0 killed $q0 killed $z0
; CHECK-NEXT:    ret
  %umax = tail call <16 x i8> @llvm.umax.v16i8(<16 x i8> %a, <16 x i8> %b)
  %umin = tail call <16 x i8> @llvm.umin.v16i8(<16 x i8> %a, <16 x i8> %b)
  %uabs = sub <16 x i8> %umax, %umin
  %ext = zext <16 x i8> %uabs to <16 x i32>
  %dot = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> %acc, <16 x i32> %ext)
  ret <4 x i32> %dot
}