; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=instcombine -S | FileCheck %s

declare float @llvm.vector.reduce.fadd.f32.v4f32(float, <4 x float>)
declare float @llvm.vector.reduce.fadd.f32.v8f32(float, <8 x float>)
declare float @llvm.vector.reduce.fmul.f32.nxv4f32(float, <vscale x 4 x float>)
declare float @llvm.vector.reduce.fmin.f32.v4f32(float, <4 x float>)
declare float @llvm.vector.reduce.fmax.f32.nxv4f32(float, <vscale x 4 x float>)
declare void @use_f32(float)

declare i32 @llvm.vector.reduce.add.v4i32(<4 x i32>)
declare i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32>)
declare i32 @llvm.vector.reduce.add.v8i32(<8 x i32>)
declare void @use_i32(i32)

declare i32 @llvm.vector.reduce.mul.v4i32(<4 x i32>)
declare i32 @llvm.vector.reduce.mul.nxv4i32(<vscale x 4 x i32>)

declare i32 @llvm.vector.reduce.smin.v4i32(<4 x i32>)
declare i32 @llvm.vector.reduce.smax.nxv4i32(<vscale x 4 x i32>)
declare i32 @llvm.vector.reduce.umin.v4i32(<4 x i32>)
declare i32 @llvm.vector.reduce.umax.nxv4i32(<vscale x 4 x i32>)

declare <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32>)
declare <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float>)
declare <4 x i32> @llvm.vector.reverse.v4i32(<4 x i32>)
declare <4 x float> @llvm.vector.reverse.v4f32(<4 x float>)

define float @diff_of_sums_v4f32(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
; CHECK-LABEL: @diff_of_sums_v4f32(
; CHECK-NEXT:    [[TMP1:%.*]] = fsub reassoc nsz <4 x float> [[V0:%.*]], [[V1:%.*]]
; CHECK-NEXT:    [[TMP2:%.*]] = call reassoc nsz float @llvm.vector.reduce.fadd.v4f32(float [[A0:%.*]], <4 x float> [[TMP1]])
; CHECK-NEXT:    [[R:%.*]] = fsub reassoc nsz float [[TMP2]], [[A1:%.*]]
; CHECK-NEXT:    ret float [[R]]
;
  %r0 = call float @llvm.vector.reduce.fadd.f32.v4f32(float %a0, <4 x float> %v0)
  %r1 = call float @llvm.vector.reduce.fadd.f32.v4f32(float %a1, <4 x float> %v1)
  %r = fsub reassoc nsz float %r0, %r1
  ret float %r
}

define float @reassoc_sum_of_reverse_v4f32(<4 x float> %v0) {
; CHECK-LABEL: @reassoc_sum_of_reverse_v4f32(
; CHECK-NEXT:    [[RED:%.*]] = call reassoc float @llvm.vector.reduce.fadd.v4f32(float 0.000000e+00, <4 x float> [[V0:%.*]])
; CHECK-NEXT:    ret float [[RED]]
;
  %rev = call <4 x float> @llvm.vector.reverse.v4f32(<4 x float> %v0)
  %red = call reassoc float @llvm.vector.reduce.fadd.v4f32(float zeroinitializer, <4 x float> %rev)
  ret float %red
}

define float @reassoc_mul_reduction_of_reverse_nxv4f32(<vscale x 4 x float> %v0) {
; CHECK-LABEL: @reassoc_mul_reduction_of_reverse_nxv4f32(
; CHECK-NEXT:    [[RED:%.*]] = call reassoc float @llvm.vector.reduce.fmul.nxv4f32(float 1.000000e+00, <vscale x 4 x float> [[V0:%.*]])
; CHECK-NEXT:    ret float [[RED]]
;
  %rev = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> %v0)
  %red = call reassoc float @llvm.vector.reduce.fmul.nxv4f32(float 1.0, <vscale x 4 x float> %rev)
  ret float %red
}

define float @fmax_of_reverse_v4f32(<4 x float> %v0) {
; CHECK-LABEL: @fmax_of_reverse_v4f32(
; CHECK-NEXT:    [[RED:%.*]] = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[V0:%.*]])
; CHECK-NEXT:    ret float [[RED]]
;
  %rev = call <4 x float> @llvm.vector.reverse.v4f32(<4 x float> %v0)
  %red = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> %rev)
  ret float %red
}

define float @fmin_of_reverse_nxv4f32(<vscale x 4 x float> %v0) {
; CHECK-LABEL: @fmin_of_reverse_nxv4f32(
; CHECK-NEXT:    [[RED:%.*]] = call float @llvm.vector.reduce.fmin.nxv4f32(<vscale x 4 x float> [[V0:%.*]])
; CHECK-NEXT:    ret float [[RED]]
;
  %rev = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> %v0)
  %red = call float @llvm.vector.reduce.fmin.nxv4f32(<vscale x 4 x float> %rev)
  ret float %red
}

; negative test - fadd cannot be folded with reverse due to lack of reassoc
define float @sum_of_reverse_v4f32(<4 x float> %v0) {
; CHECK-LABEL: @sum_of_reverse_v4f32(
; CHECK-NEXT:    [[REV:%.*]] = call <4 x float> @llvm.vector.reverse.v4f32(<4 x float> [[V0:%.*]])
; CHECK-NEXT:    [[RED:%.*]] = call float @llvm.vector.reduce.fadd.v4f32(float 0.000000e+00, <4 x float> [[REV]])
; CHECK-NEXT:    ret float [[RED]]
;
  %rev = call <4 x float> @llvm.vector.reverse.v4f32(<4 x float> %v0)
  %red = call float @llvm.vector.reduce.fadd.v4f32(float zeroinitializer, <4 x float> %rev)
  ret float %red
}

; negative test - fmul cannot be folded with reverse due to lack of reassoc
define float @mul_reduction_of_reverse_nxv4f32(<vscale x 4 x float> %v0) {
; CHECK-LABEL: @mul_reduction_of_reverse_nxv4f32(
; CHECK-NEXT:    [[REV:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[V0:%.*]])
; CHECK-NEXT:    [[RED:%.*]] = call float @llvm.vector.reduce.fmul.nxv4f32(float 0.000000e+00, <vscale x 4 x float> [[REV]])
; CHECK-NEXT:    ret float [[RED]]
;
  %rev = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> %v0)
  %red = call float @llvm.vector.reduce.fmul.nxv4f32(float zeroinitializer, <vscale x 4 x float> %rev)
  ret float %red
}


; negative test - fsub must allow reassociation

define float @diff_of_sums_v4f32_fmf(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
; CHECK-LABEL: @diff_of_sums_v4f32_fmf(
; CHECK-NEXT:    [[R0:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
; CHECK-NEXT:    [[R1:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A1:%.*]], <4 x float> [[V1:%.*]])
; CHECK-NEXT:    [[R:%.*]] = fsub nnan ninf nsz float [[R0]], [[R1]]
; CHECK-NEXT:    ret float [[R]]
;
  %r0 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a0, <4 x float> %v0)
  %r1 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a1, <4 x float> %v1)
  %r = fsub ninf nnan nsz float %r0, %r1
  ret float %r
}

; negative test - extra uses could create extra instructions

define float @diff_of_sums_extra_use1(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
; CHECK-LABEL: @diff_of_sums_extra_use1(
; CHECK-NEXT:    [[R0:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
; CHECK-NEXT:    call void @use_f32(float [[R0]])
; CHECK-NEXT:    [[R1:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A1:%.*]], <4 x float> [[V1:%.*]])
; CHECK-NEXT:    [[R:%.*]] = fsub fast float [[R0]], [[R1]]
; CHECK-NEXT:    ret float [[R]]
;
  %r0 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a0, <4 x float> %v0)
  call void @use_f32(float %r0)
  %r1 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a1, <4 x float> %v1)
  %r = fsub fast float %r0, %r1
  ret float %r
}

; negative test - extra uses could create extra instructions

define float @diff_of_sums_extra_use2(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
; CHECK-LABEL: @diff_of_sums_extra_use2(
; CHECK-NEXT:    [[R0:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
; CHECK-NEXT:    [[R1:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A1:%.*]], <4 x float> [[V1:%.*]])
; CHECK-NEXT:    call void @use_f32(float [[R1]])
; CHECK-NEXT:    [[R:%.*]] = fsub fast float [[R0]], [[R1]]
; CHECK-NEXT:    ret float [[R]]
;
  %r0 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a0, <4 x float> %v0)
  %r1 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a1, <4 x float> %v1)
  call void @use_f32(float %r1)
  %r = fsub fast float %r0, %r1
  ret float %r
}

; negative test - can't reassociate different vector types

define float @diff_of_sums_type_mismatch(float %a0, <4 x float> %v0, float %a1, <8 x float> %v1) {
; CHECK-LABEL: @diff_of_sums_type_mismatch(
; CHECK-NEXT:    [[R0:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
; CHECK-NEXT:    [[R1:%.*]] = call fast float @llvm.vector.reduce.fadd.v8f32(float [[A1:%.*]], <8 x float> [[V1:%.*]])
; CHECK-NEXT:    [[R:%.*]] = fsub fast float [[R0]], [[R1]]
; CHECK-NEXT:    ret float [[R]]
;
  %r0 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a0, <4 x float> %v0)
  %r1 = call fast float @llvm.vector.reduce.fadd.f32.v8f32(float %a1, <8 x float> %v1)
  %r = fsub fast float %r0, %r1
  ret float %r
}

define i32 @diff_of_sums_v4i32(<4 x i32> %v0, <4 x i32> %v1) {
; CHECK-LABEL: @diff_of_sums_v4i32(
; CHECK-NEXT:    [[TMP1:%.*]] = sub <4 x i32> [[V0:%.*]], [[V1:%.*]]
; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP1]])
; CHECK-NEXT:    ret i32 [[R]]
;
  %r0 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v0)
  %r1 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v1)
  %r = sub i32 %r0, %r1
  ret i32 %r
}

define i32 @sum_of_reverse_v4i32(<4 x i32> %v0) {
; CHECK-LABEL: @sum_of_reverse_v4i32(
; CHECK-NEXT:    [[RED:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[V0:%.*]])
; CHECK-NEXT:    ret i32 [[RED]]
;
  %rev = call <4 x i32> @llvm.vector.reverse.v4i32(<4 x i32> %v0)
  %red = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %rev)
  ret i32 %red
}

define i32 @sum_of_reverse_nxv4i32(<vscale x 4 x i32> %v0) {
; CHECK-LABEL: @sum_of_reverse_nxv4i32(
; CHECK-NEXT:    [[RED:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[V0:%.*]])
; CHECK-NEXT:    ret i32 [[RED]]
;
  %rev = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> %v0)
  %red = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> %rev)
  ret i32 %red
}

define i32 @mul_reduce_of_reverse_v4i32(<4 x i32> %v0) {
; CHECK-LABEL: @mul_reduce_of_reverse_v4i32(
; CHECK-NEXT:    [[RED:%.*]] = call i32 @llvm.vector.reduce.mul.v4i32(<4 x i32> [[V0:%.*]])
; CHECK-NEXT:    ret i32 [[RED]]
;
  %rev = call <4 x i32> @llvm.vector.reverse.v4i32(<4 x i32> %v0)
  %red = call i32 @llvm.vector.reduce.mul.v4i32(<4 x i32> %rev)
  ret i32 %red
}

define i32 @mul_reduce_of_reverse_nxv4i32(<vscale x 4 x i32> %v0) {
; CHECK-LABEL: @mul_reduce_of_reverse_nxv4i32(
; CHECK-NEXT:    [[RED:%.*]] = call i32 @llvm.vector.reduce.mul.nxv4i32(<vscale x 4 x i32> [[V0:%.*]])
; CHECK-NEXT:    ret i32 [[RED]]
;
  %rev = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> %v0)
  %red = call i32 @llvm.vector.reduce.mul.nxv4i32(<vscale x 4 x i32> %rev)
  ret i32 %red
}

define i32 @smin_reduce_of_reverse_v4i32(<4 x i32> %v0) {
; CHECK-LABEL: @smin_reduce_of_reverse_v4i32(
; CHECK-NEXT:    [[RED:%.*]] = call i32 @llvm.vector.reduce.smin.v4i32(<4 x i32> [[V0:%.*]])
; CHECK-NEXT:    ret i32 [[RED]]
;
  %rev = call <4 x i32> @llvm.vector.reverse.v4i32(<4 x i32> %v0)
  %red = call i32 @llvm.vector.reduce.smin.v4i32(<4 x i32> %rev)
  ret i32 %red
}

define i32 @smax_reduce_of_reverse_nxv4i32(<vscale x 4 x i32> %v0) {
; CHECK-LABEL: @smax_reduce_of_reverse_nxv4i32(
; CHECK-NEXT:    [[RED:%.*]] = call i32 @llvm.vector.reduce.smax.nxv4i32(<vscale x 4 x i32> [[V0:%.*]])
; CHECK-NEXT:    ret i32 [[RED]]
;
  %rev = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> %v0)
  %red = call i32 @llvm.vector.reduce.smax.nxv4i32(<vscale x 4 x i32> %rev)
  ret i32 %red
}

define i32 @umin_reduce_of_reverse_v4i32(<4 x i32> %v0) {
; CHECK-LABEL: @umin_reduce_of_reverse_v4i32(
; CHECK-NEXT:    [[RED:%.*]] = call i32 @llvm.vector.reduce.umin.v4i32(<4 x i32> [[V0:%.*]])
; CHECK-NEXT:    ret i32 [[RED]]
;
  %rev = call <4 x i32> @llvm.vector.reverse.v4i32(<4 x i32> %v0)
  %red = call i32 @llvm.vector.reduce.umin.v4i32(<4 x i32> %rev)
  ret i32 %red
}

define i32 @umax_reduce_of_reverse_nxv4i32(<vscale x 4 x i32> %v0) {
; CHECK-LABEL: @umax_reduce_of_reverse_nxv4i32(
; CHECK-NEXT:    [[RED:%.*]] = call i32 @llvm.vector.reduce.umax.nxv4i32(<vscale x 4 x i32> [[V0:%.*]])
; CHECK-NEXT:    ret i32 [[RED]]
;
  %rev = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> %v0)
  %red = call i32 @llvm.vector.reduce.umax.nxv4i32(<vscale x 4 x i32> %rev)
  ret i32 %red
}

; negative test - extra uses could create extra instructions

define i32 @diff_of_sums_v4i32_extra_use1(<4 x i32> %v0, <4 x i32> %v1) {
; CHECK-LABEL: @diff_of_sums_v4i32_extra_use1(
; CHECK-NEXT:    [[R0:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[V0:%.*]])
; CHECK-NEXT:    call void @use_i32(i32 [[R0]])
; CHECK-NEXT:    [[R1:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[V1:%.*]])
; CHECK-NEXT:    [[R:%.*]] = sub i32 [[R0]], [[R1]]
; CHECK-NEXT:    ret i32 [[R]]
;
  %r0 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v0)
  call void @use_i32(i32 %r0)
  %r1 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v1)
  %r = sub i32 %r0, %r1
  ret i32 %r
}

; negative test - extra uses could create extra instructions

define i32 @diff_of_sums_v4i32_extra_use2(<4 x i32> %v0, <4 x i32> %v1) {
; CHECK-LABEL: @diff_of_sums_v4i32_extra_use2(
; CHECK-NEXT:    [[R0:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[V0:%.*]])
; CHECK-NEXT:    [[R1:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[V1:%.*]])
; CHECK-NEXT:    call void @use_i32(i32 [[R1]])
; CHECK-NEXT:    [[R:%.*]] = sub i32 [[R0]], [[R1]]
; CHECK-NEXT:    ret i32 [[R]]
;
  %r0 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v0)
  %r1 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v1)
  call void @use_i32(i32 %r1)
  %r = sub i32 %r0, %r1
  ret i32 %r
}

; negative test - can't reassociate different vector types

define i32 @diff_of_sums_type_mismatch2(<8 x i32> %v0, <4 x i32> %v1) {
; CHECK-LABEL: @diff_of_sums_type_mismatch2(
; CHECK-NEXT:    [[R0:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[V0:%.*]])
; CHECK-NEXT:    [[R1:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[V1:%.*]])
; CHECK-NEXT:    [[R:%.*]] = sub i32 [[R0]], [[R1]]
; CHECK-NEXT:    ret i32 [[R]]
;
  %r0 = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> %v0)
  %r1 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v1)
  %r = sub i32 %r0, %r1
  ret i32 %r
}