; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc < %s -start-before=codegenprepare | FileCheck %s

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

define i32 @vmask_reduce_i32_v8i8(<8 x i8> %a, <8 x i8> %b) {
; CHECK-LABEL: vmask_reduce_i32_v8i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.8b, v1.8b, v0.8b
; CHECK-NEXT:    addv b0, v0.8b
; CHECK-NEXT:    smov w0, v0.b[0]
; CHECK-NEXT:    ret
  %mask = icmp slt <8 x i8> %a, %b
  %t1 = sext <8 x i1> %mask to <8 x i8>
  %t2 = call i8 @llvm.vector.reduce.add.v8i8(<8 x i8> %t1)
  %t3 = sext i8 %t2 to i32
  ret i32 %t3
}

define i32 @vmask_reduce_i32_v16i8(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: vmask_reduce_i32_v16i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.16b, v1.16b, v0.16b
; CHECK-NEXT:    addv b0, v0.16b
; CHECK-NEXT:    smov w0, v0.b[0]
; CHECK-NEXT:    ret
  %mask = icmp slt <16 x i8> %a, %b
  %t1 = sext <16 x i1> %mask to <16 x i8>
  %t2 = call i8 @llvm.vector.reduce.add.v16i8(<16 x i8> %t1)
  %t3 = sext i8 %t2 to i32
  ret i32 %t3
}

define i32 @vmask_reduce_i32_v4i16(<4 x i16> %a, <4 x i16> %b) {
; CHECK-LABEL: vmask_reduce_i32_v4i16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.4h, v1.4h, v0.4h
; CHECK-NEXT:    addv h0, v0.4h
; CHECK-NEXT:    smov w0, v0.h[0]
; CHECK-NEXT:    ret
  %mask = icmp slt <4 x i16> %a, %b
  %t1 = sext <4 x i1> %mask to <4 x i16>
  %t2 = call i16 @llvm.vector.reduce.add.v4i16(<4 x i16> %t1)
  %t3 = sext i16 %t2 to i32
  ret i32 %t3
}

define i32 @vmask_reduce_i32_v8i16(<8 x i16> %a, <8 x i16> %b) {
; CHECK-LABEL: vmask_reduce_i32_v8i16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.8h, v1.8h, v0.8h
; CHECK-NEXT:    addv h0, v0.8h
; CHECK-NEXT:    smov w0, v0.h[0]
; CHECK-NEXT:    ret
  %mask = icmp slt <8 x i16> %a, %b
  %t1 = sext <8 x i1> %mask to <8 x i16>
  %t2 = call i16 @llvm.vector.reduce.add.v8i16(<8 x i16> %t1)
  %t3 = sext i16 %t2 to i32
  ret i32 %t3
}

define i64 @vmask_reduce_i32_v2i32(<2 x i32> %a, <2 x i32> %b) {
; CHECK-LABEL: vmask_reduce_i32_v2i32:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.2s, v1.2s, v0.2s
; CHECK-NEXT:    addp v0.2s, v0.2s, v0.2s
; CHECK-NEXT:    smov x0, v0.s[0]
; CHECK-NEXT:    ret
  %mask = icmp slt <2 x i32> %a, %b
  %t1 = sext <2 x i1> %mask to <2 x i32>
  %t2 = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> %t1)
  %t3 = sext i32 %t2 to i64
  ret i64 %t3
}

define i64 @vmask_reduce_i64_v8i8(<8 x i8> %a, <8 x i8> %b) {
; CHECK-LABEL: vmask_reduce_i64_v8i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.8b, v1.8b, v0.8b
; CHECK-NEXT:    addv b0, v0.8b
; CHECK-NEXT:    smov x0, v0.b[0]
; CHECK-NEXT:    ret
  %mask = icmp slt <8 x i8> %a, %b
  %t1 = sext <8 x i1> %mask to <8 x i8>
  %t2 = call i8 @llvm.vector.reduce.add.v8i8(<8 x i8> %t1)
  %t3 = sext i8 %t2 to i64
  ret i64 %t3
}

define i64 @vmask_reduce_i64_v16i8(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: vmask_reduce_i64_v16i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.16b, v1.16b, v0.16b
; CHECK-NEXT:    addv b0, v0.16b
; CHECK-NEXT:    smov x0, v0.b[0]
; CHECK-NEXT:    ret
  %mask = icmp slt <16 x i8> %a, %b
  %t1 = sext <16 x i1> %mask to <16 x i8>
  %t2 = call i8 @llvm.vector.reduce.add.v16i8(<16 x i8> %t1)
  %t3 = sext i8 %t2 to i64
  ret i64 %t3
}

define i64 @vmask_reduce_i64_v4i16(<4 x i16> %a, <4 x i16> %b) {
; CHECK-LABEL: vmask_reduce_i64_v4i16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.4h, v1.4h, v0.4h
; CHECK-NEXT:    addv h0, v0.4h
; CHECK-NEXT:    smov x0, v0.h[0]
; CHECK-NEXT:    ret
  %mask = icmp slt <4 x i16> %a, %b
  %t1 = sext <4 x i1> %mask to <4 x i16>
  %t2 = call i16 @llvm.vector.reduce.add.v4i16(<4 x i16> %t1)
  %t3 = sext i16 %t2 to i64
  ret i64 %t3
}

define i64 @vmask_reduce_i64_v8i16(<8 x i16> %a, <8 x i16> %b) {
; CHECK-LABEL: vmask_reduce_i64_v8i16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.8h, v1.8h, v0.8h
; CHECK-NEXT:    addv h0, v0.8h
; CHECK-NEXT:    smov x0, v0.h[0]
; CHECK-NEXT:    ret
  %mask = icmp slt <8 x i16> %a, %b
  %t1 = sext <8 x i1> %mask to <8 x i16>
  %t2 = call i16 @llvm.vector.reduce.add.v8i16(<8 x i16> %t1)
  %t3 = sext i16 %t2 to i64
  ret i64 %t3
}

; TODO: We should use a saddlv here to avoid the smov.
define i64 @vmask_reduce_i64_v4i32(<4 x i32> %a, <4 x i32> %b) {
; CHECK-LABEL: vmask_reduce_i64_v4i32:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.4s, v1.4s, v0.4s
; CHECK-NEXT:    addv s0, v0.4s
; CHECK-NEXT:    smov x0, v0.s[0]
; CHECK-NEXT:    ret
  %mask = icmp slt <4 x i32> %a, %b
  %t1 = sext <4 x i1> %mask to <4 x i32>
  %t2 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %t1)
  %t3 = sext i32 %t2 to i64
  ret i64 %t3
}

define i32 @vmask_popcount_i32_v16i8(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: vmask_popcount_i32_v16i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.16b, v1.16b, v0.16b
; CHECK-NEXT:    addv b0, v0.16b
; CHECK-NEXT:    smov w8, v0.b[0]
; CHECK-NEXT:    neg w0, w8
; CHECK-NEXT:    ret
  %mask = icmp slt <16 x i8> %a, %b
  %t1 = sext <16 x i1> %mask to <16 x i8>
  %t2 = call i8 @llvm.vector.reduce.add.v16i8(<16 x i8> %t1)
  %t3 = sext i8 %t2 to i32
  %t4 = sub i32 0, %t3
  ret i32 %t4
}

define i32 @vmask_popcount_i32_v4i16(<4 x i16> %a, <4 x i16> %b) {
; CHECK-LABEL: vmask_popcount_i32_v4i16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.4h, v1.4h, v0.4h
; CHECK-NEXT:    addv h0, v0.4h
; CHECK-NEXT:    smov w8, v0.h[0]
; CHECK-NEXT:    neg w0, w8
; CHECK-NEXT:    ret
  %mask = icmp slt <4 x i16> %a, %b
  %t1 = sext <4 x i1> %mask to <4 x i16>
  %t2 = call i16 @llvm.vector.reduce.add.v4i16(<4 x i16> %t1)
  %t3 = sext i16 %t2 to i32
  %t4 = sub i32 0, %t3
  ret i32 %t4
}

define i64 @vmask_popcount_i64_v16i8(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: vmask_popcount_i64_v16i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.16b, v1.16b, v0.16b
; CHECK-NEXT:    addv b0, v0.16b
; CHECK-NEXT:    smov x8, v0.b[0]
; CHECK-NEXT:    neg x0, x8
; CHECK-NEXT:    ret
  %mask = icmp slt <16 x i8> %a, %b
  %t1 = sext <16 x i1> %mask to <16 x i8>
  %t2 = call i8 @llvm.vector.reduce.add.v16i8(<16 x i8> %t1)
  %t3 = sext i8 %t2 to i64
  %t4 = sub i64 0, %t3
  ret i64 %t4
}

define i64 @vmask_popcount_i64_v4i16(<4 x i16> %a, <4 x i16> %b) {
; CHECK-LABEL: vmask_popcount_i64_v4i16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    cmgt v0.4h, v1.4h, v0.4h
; CHECK-NEXT:    addv h0, v0.4h
; CHECK-NEXT:    smov x8, v0.h[0]
; CHECK-NEXT:    neg x0, x8
; CHECK-NEXT:    ret
  %mask = icmp slt <4 x i16> %a, %b
  %t1 = sext <4 x i1> %mask to <4 x i16>
  %t2 = call i16 @llvm.vector.reduce.add.v4i16(<4 x i16> %t1)
  %t3 = sext i16 %t2 to i64
  %t4 = sub i64 0, %t3
  ret i64 %t4
}