; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve.fp -verify-machineinstrs %s -o - | FileCheck %s

define void @arm_cmplx_dot_prod_f32(float* %pSrcA, float* %pSrcB, i32 %numSamples, float* nocapture %realResult, float* nocapture %imagResult) {
; CHECK-LABEL: arm_cmplx_dot_prod_f32:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    .save {r4, r5, r7, lr}
; CHECK-NEXT:    push {r4, r5, r7, lr}
; CHECK-NEXT:    .vsave {d8, d9}
; CHECK-NEXT:    vpush {d8, d9}
; CHECK-NEXT:    ldr.w r12, [sp, #32]
; CHECK-NEXT:    cmp r2, #8
; CHECK-NEXT:    blo .LBB0_6
; CHECK-NEXT:  @ %bb.1: @ %while.body.preheader
; CHECK-NEXT:    lsrs r4, r2, #2
; CHECK-NEXT:    mov.w lr, #2
; CHECK-NEXT:    cmp r4, #2
; CHECK-NEXT:    it lt
; CHECK-NEXT:    lsrlt.w lr, r2, #2
; CHECK-NEXT:    rsb r4, lr, r2, lsr #2
; CHECK-NEXT:    vldrw.u32 q2, [r1], #32
; CHECK-NEXT:    add.w lr, r4, #1
; CHECK-NEXT:    vldrw.u32 q1, [r0], #32
; CHECK-NEXT:    vmov.i32 q0, #0x0
; CHECK-NEXT:  .LBB0_2: @ %while.body
; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
; CHECK-NEXT:    vcmla.f32 q0, q1, q2, #0
; CHECK-NEXT:    vldrw.u32 q3, [r1, #-16]
; CHECK-NEXT:    vldrw.u32 q4, [r0, #-16]
; CHECK-NEXT:    vcmla.f32 q0, q1, q2, #90
; CHECK-NEXT:    vldrw.u32 q1, [r0], #32
; CHECK-NEXT:    vldrw.u32 q2, [r1], #32
; CHECK-NEXT:    vcmla.f32 q0, q4, q3, #0
; CHECK-NEXT:    vcmla.f32 q0, q4, q3, #90
; CHECK-NEXT:    le lr, .LBB0_2
; CHECK-NEXT:  @ %bb.3: @ %while.end
; CHECK-NEXT:    vcmla.f32 q0, q1, q2, #0
; CHECK-NEXT:    movs r4, #6
; CHECK-NEXT:    vcmla.f32 q0, q1, q2, #90
; CHECK-NEXT:    vldrw.u32 q1, [r1, #-16]
; CHECK-NEXT:    vldrw.u32 q2, [r0, #-16]
; CHECK-NEXT:    and.w r2, r4, r2, lsl #1
; CHECK-NEXT:    vcmla.f32 q0, q2, q1, #0
; CHECK-NEXT:    vcmla.f32 q0, q2, q1, #90
; CHECK-NEXT:    cbz r2, .LBB0_8
; CHECK-NEXT:  @ %bb.4: @ %while.body9
; CHECK-NEXT:    vctp.32 r2
; CHECK-NEXT:    cmp r2, #4
; CHECK-NEXT:    vpstttt
; CHECK-NEXT:    vldrwt.u32 q1, [r1]
; CHECK-NEXT:    vldrwt.u32 q2, [r0]
; CHECK-NEXT:    vcmlat.f32 q0, q2, q1, #0
; CHECK-NEXT:    vcmlat.f32 q0, q2, q1, #90
; CHECK-NEXT:    bls .LBB0_8
; CHECK-NEXT:  @ %bb.5: @ %while.body9.1
; CHECK-NEXT:    subs r2, #4
; CHECK-NEXT:    vctp.32 r2
; CHECK-NEXT:    vpstttt
; CHECK-NEXT:    vldrwt.u32 q1, [r1, #16]
; CHECK-NEXT:    vldrwt.u32 q2, [r0, #16]
; CHECK-NEXT:    vcmlat.f32 q0, q2, q1, #0
; CHECK-NEXT:    vcmlat.f32 q0, q2, q1, #90
; CHECK-NEXT:    b .LBB0_8
; CHECK-NEXT:  .LBB0_6: @ %if.else
; CHECK-NEXT:    lsls r4, r2, #1
; CHECK-NEXT:    vmov.i32 q0, #0x0
; CHECK-NEXT:    dlstp.32 lr, r4
; CHECK-NEXT:  .LBB0_7: @ %do.body
; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
; CHECK-NEXT:    vldrw.u32 q1, [r1], #16
; CHECK-NEXT:    vldrw.u32 q2, [r0], #16
; CHECK-NEXT:    vcmla.f32 q0, q2, q1, #0
; CHECK-NEXT:    vcmla.f32 q0, q2, q1, #90
; CHECK-NEXT:    letp lr, .LBB0_7
; CHECK-NEXT:  .LBB0_8: @ %if.end
; CHECK-NEXT:    vadd.f32 s0, s0, s2
; CHECK-NEXT:    vadd.f32 s2, s1, s3
; CHECK-NEXT:    vstr s0, [r3]
; CHECK-NEXT:    vstr s2, [r12]
; CHECK-NEXT:    vpop {d8, d9}
; CHECK-NEXT:    pop {r4, r5, r7, pc}
entry:
  %cmp = icmp ugt i32 %numSamples, 7
  br i1 %cmp, label %while.body.preheader, label %if.else

while.body.preheader:                             ; preds = %entry
  %vecSrcA.0.in108 = bitcast float* %pSrcA to <4 x float>*
  %vecSrcA.0109 = load <4 x float>, <4 x float>* %vecSrcA.0.in108, align 4
  %vecSrcB.0.in106 = bitcast float* %pSrcB to <4 x float>*
  %vecSrcB.0107 = load <4 x float>, <4 x float>* %vecSrcB.0.in106, align 4
  %pSrcB.addr.0105 = getelementptr inbounds float, float* %pSrcB, i32 4
  %pSrcA.addr.0104 = getelementptr inbounds float, float* %pSrcA, i32 4
  %shr = lshr i32 %numSamples, 2
  br label %while.body

while.body:                                       ; preds = %while.body.preheader, %while.body
  %vecSrcA.0118 = phi <4 x float> [ %vecSrcA.0, %while.body ], [ %vecSrcA.0109, %while.body.preheader ]
  %vecSrcB.0117 = phi <4 x float> [ %vecSrcB.0, %while.body ], [ %vecSrcB.0107, %while.body.preheader ]
  %pSrcB.addr.0116 = phi float* [ %pSrcB.addr.0, %while.body ], [ %pSrcB.addr.0105, %while.body.preheader ]
  %pSrcA.addr.0115 = phi float* [ %pSrcA.addr.0, %while.body ], [ %pSrcA.addr.0104, %while.body.preheader ]
  %vec_acc.0114 = phi <4 x float> [ %7, %while.body ], [ zeroinitializer, %while.body.preheader ]
  %vecSrcB.0.in.in113 = phi float* [ %add.ptr4, %while.body ], [ %pSrcB, %while.body.preheader ]
  %vecSrcA.0.in.in112 = phi float* [ %add.ptr3, %while.body ], [ %pSrcA, %while.body.preheader ]
  %blkCnt.0.in111 = phi i32 [ %blkCnt.0, %while.body ], [ %shr, %while.body.preheader ]
  %blkCnt.0 = add nsw i32 %blkCnt.0.in111, -1
  %0 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.v4f32(i32 0, <4 x float> %vec_acc.0114, <4 x float> %vecSrcA.0118, <4 x float> %vecSrcB.0117)
  %1 = bitcast float* %pSrcA.addr.0115 to <4 x float>*
  %2 = load <4 x float>, <4 x float>* %1, align 4
  %add.ptr3 = getelementptr inbounds float, float* %vecSrcA.0.in.in112, i32 8
  %3 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.v4f32(i32 1, <4 x float> %0, <4 x float> %vecSrcA.0118, <4 x float> %vecSrcB.0117)
  %4 = bitcast float* %pSrcB.addr.0116 to <4 x float>*
  %5 = load <4 x float>, <4 x float>* %4, align 4
  %add.ptr4 = getelementptr inbounds float, float* %vecSrcB.0.in.in113, i32 8
  %6 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.v4f32(i32 0, <4 x float> %3, <4 x float> %2, <4 x float> %5)
  %7 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.v4f32(i32 1, <4 x float> %6, <4 x float> %2, <4 x float> %5)
  %pSrcA.addr.0 = getelementptr inbounds float, float* %vecSrcA.0.in.in112, i32 12
  %pSrcB.addr.0 = getelementptr inbounds float, float* %vecSrcB.0.in.in113, i32 12
  %vecSrcB.0.in = bitcast float* %add.ptr4 to <4 x float>*
  %vecSrcB.0 = load <4 x float>, <4 x float>* %vecSrcB.0.in, align 4
  %vecSrcA.0.in = bitcast float* %add.ptr3 to <4 x float>*
  %vecSrcA.0 = load <4 x float>, <4 x float>* %vecSrcA.0.in, align 4
  %cmp2 = icmp sgt i32 %blkCnt.0.in111, 2
  br i1 %cmp2, label %while.body, label %while.end

while.end:                                        ; preds = %while.body
  %8 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.v4f32(i32 0, <4 x float> %7, <4 x float> %vecSrcA.0, <4 x float> %vecSrcB.0)
  %9 = bitcast float* %pSrcA.addr.0 to <4 x float>*
  %10 = load <4 x float>, <4 x float>* %9, align 4
  %11 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.v4f32(i32 1, <4 x float> %8, <4 x float> %vecSrcA.0, <4 x float> %vecSrcB.0)
  %12 = bitcast float* %pSrcB.addr.0 to <4 x float>*
  %13 = load <4 x float>, <4 x float>* %12, align 4
  %14 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.v4f32(i32 0, <4 x float> %11, <4 x float> %10, <4 x float> %13)
  %15 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.v4f32(i32 1, <4 x float> %14, <4 x float> %10, <4 x float> %13)
  %and = shl i32 %numSamples, 1
  %mul = and i32 %and, 6
  %cmp8123.not = icmp eq i32 %mul, 0
  br i1 %cmp8123.not, label %if.end, label %while.body9

while.body9:                                      ; preds = %while.end
  %16 = tail call <4 x i1> @llvm.arm.mve.vctp32(i32 %mul)
  %add.ptr10 = getelementptr inbounds float, float* %vecSrcA.0.in.in112, i32 16
  %add.ptr11 = getelementptr inbounds float, float* %vecSrcB.0.in.in113, i32 16
  %17 = bitcast float* %add.ptr10 to <4 x float>*
  %18 = tail call fast <4 x float> @llvm.masked.load.v4f32.p0v4f32(<4 x float>* nonnull %17, i32 4, <4 x i1> %16, <4 x float> zeroinitializer)
  %19 = bitcast float* %add.ptr11 to <4 x float>*
  %20 = tail call fast <4 x float> @llvm.masked.load.v4f32.p0v4f32(<4 x float>* nonnull %19, i32 4, <4 x i1> %16, <4 x float> zeroinitializer)
  %21 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.predicated.v4f32.v4i1(i32 0, <4 x float> %15, <4 x float> %18, <4 x float> %20, <4 x i1> %16)
  %22 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.predicated.v4f32.v4i1(i32 1, <4 x float> %21, <4 x float> %18, <4 x float> %20, <4 x i1> %16)
  %cmp8 = icmp ugt i32 %mul, 4
  br i1 %cmp8, label %while.body9.1, label %if.end

if.else:                                          ; preds = %entry
  %mul14 = shl nuw nsw i32 %numSamples, 1
  br label %do.body

do.body:                                          ; preds = %do.body, %if.else
  %blkCnt.2 = phi i32 [ %mul14, %if.else ], [ %sub18, %do.body ]
  %vec_acc.2 = phi <4 x float> [ zeroinitializer, %if.else ], [ %29, %do.body ]
  %pSrcB.addr.2 = phi float* [ %pSrcB, %if.else ], [ %add.ptr17, %do.body ]
  %pSrcA.addr.2 = phi float* [ %pSrcA, %if.else ], [ %add.ptr16, %do.body ]
  %23 = tail call <4 x i1> @llvm.arm.mve.vctp32(i32 %blkCnt.2)
  %24 = bitcast float* %pSrcA.addr.2 to <4 x float>*
  %25 = tail call fast <4 x float> @llvm.masked.load.v4f32.p0v4f32(<4 x float>* %24, i32 4, <4 x i1> %23, <4 x float> zeroinitializer)
  %26 = bitcast float* %pSrcB.addr.2 to <4 x float>*
  %27 = tail call fast <4 x float> @llvm.masked.load.v4f32.p0v4f32(<4 x float>* %26, i32 4, <4 x i1> %23, <4 x float> zeroinitializer)
  %28 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.predicated.v4f32.v4i1(i32 0, <4 x float> %vec_acc.2, <4 x float> %25, <4 x float> %27, <4 x i1> %23)
  %29 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.predicated.v4f32.v4i1(i32 1, <4 x float> %28, <4 x float> %25, <4 x float> %27, <4 x i1> %23)
  %add.ptr16 = getelementptr inbounds float, float* %pSrcA.addr.2, i32 4
  %add.ptr17 = getelementptr inbounds float, float* %pSrcB.addr.2, i32 4
  %sub18 = add nsw i32 %blkCnt.2, -4
  %cmp19 = icmp sgt i32 %blkCnt.2, 4
  br i1 %cmp19, label %do.body, label %if.end

if.end:                                           ; preds = %do.body, %while.body9, %while.body9.1, %while.end
  %vec_acc.3 = phi <4 x float> [ %15, %while.end ], [ %22, %while.body9 ], [ %40, %while.body9.1 ], [ %29, %do.body ]
  %30 = extractelement <4 x float> %vec_acc.3, i32 0
  %31 = extractelement <4 x float> %vec_acc.3, i32 2
  %add = fadd fast float %30, %31
  %32 = extractelement <4 x float> %vec_acc.3, i32 1
  %33 = extractelement <4 x float> %vec_acc.3, i32 3
  %add20 = fadd fast float %32, %33
  store float %add, float* %realResult, align 4
  store float %add20, float* %imagResult, align 4
  ret void

while.body9.1:                                    ; preds = %while.body9
  %sub12 = add nsw i32 %mul, -4
  %34 = tail call <4 x i1> @llvm.arm.mve.vctp32(i32 %sub12)
  %add.ptr10.1 = getelementptr inbounds float, float* %vecSrcA.0.in.in112, i32 20
  %add.ptr11.1 = getelementptr inbounds float, float* %vecSrcB.0.in.in113, i32 20
  %35 = bitcast float* %add.ptr10.1 to <4 x float>*
  %36 = tail call fast <4 x float> @llvm.masked.load.v4f32.p0v4f32(<4 x float>* nonnull %35, i32 4, <4 x i1> %34, <4 x float> zeroinitializer)
  %37 = bitcast float* %add.ptr11.1 to <4 x float>*
  %38 = tail call fast <4 x float> @llvm.masked.load.v4f32.p0v4f32(<4 x float>* nonnull %37, i32 4, <4 x i1> %34, <4 x float> zeroinitializer)
  %39 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.predicated.v4f32.v4i1(i32 0, <4 x float> %22, <4 x float> %36, <4 x float> %38, <4 x i1> %34)
  %40 = tail call fast <4 x float> @llvm.arm.mve.vcmlaq.predicated.v4f32.v4i1(i32 1, <4 x float> %39, <4 x float> %36, <4 x float> %38, <4 x i1> %34)
  br label %if.end
}

declare <4 x float> @llvm.arm.mve.vcmlaq.v4f32(i32, <4 x float>, <4 x float>, <4 x float>) #1
declare <4 x i1> @llvm.arm.mve.vctp32(i32) #1
declare <4 x float> @llvm.masked.load.v4f32.p0v4f32(<4 x float>*, i32 immarg, <4 x i1>, <4 x float>) #2
declare <4 x float> @llvm.arm.mve.vcmlaq.predicated.v4f32.v4i1(i32, <4 x float>, <4 x float>, <4 x float>, <4 x i1>) #1