test/Transforms/LoopVectorize/tail-folding-argmin-reduction.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -S < %s -p loop-vectorize -force-target-supports-scalable-vectors -force-vector-width='vscale x 2' -force-target-supports-masked-memory-ops -tail-folding-policy=must-fold-tail -force-tail-folding-style=data - | FileCheck %s --check-prefix=DATA
; RUN: opt -S < %s -p loop-vectorize -force-target-supports-scalable-vectors -force-vector-width='vscale x 2' -force-target-supports-masked-memory-ops -tail-folding-policy=must-fold-tail -force-tail-folding-style=data-without-lane-mask - | FileCheck %s --check-prefix=DATA_NO_LANEMASK
; RUN: opt -S < %s -p loop-vectorize -force-target-supports-scalable-vectors -force-vector-width='vscale x 2' -force-target-supports-masked-memory-ops -tail-folding-policy=must-fold-tail -force-tail-folding-style=data-and-control - | FileCheck %s --check-prefix=DATA_AND_CONTROL


define i64 @argmin_first_index(ptr %arr, i64 %n, i64 %start) {
; DATA-LABEL: define i64 @argmin_first_index(
; DATA-SAME: ptr [[ARR:%.*]], i64 [[N:%.*]], i64 [[START:%.*]]) {
; DATA-NEXT:  [[ENTRY:.*:]]
; DATA-NEXT:    br label %[[VECTOR_PH:.*]]
; DATA:       [[VECTOR_PH]]:
; DATA-NEXT:    [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
; DATA-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP4]], 1
; DATA-NEXT:    [[TMP10:%.*]] = sub i64 [[TMP1]], 1
; DATA-NEXT:    [[N_RND_UP:%.*]] = add i64 [[N]], [[TMP10]]
; DATA-NEXT:    [[TMP0:%.*]] = urem i64 [[N_RND_UP]], [[TMP1]]
; DATA-NEXT:    [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[TMP0]]
; DATA-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[START]], i64 0
; DATA-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA-NEXT:    [[TMP3:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
; DATA-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP1]], i64 0
; DATA-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA-NEXT:    br label %[[VECTOR_BODY:.*]]
; DATA:       [[VECTOR_BODY]]:
; DATA-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; DATA-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 2 x i64> [ [[TMP3]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
; DATA-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ [[BROADCAST_SPLAT]], %[[VECTOR_PH]] ], [ [[TMP7:%.*]], %[[VECTOR_BODY]] ]
; DATA-NEXT:    [[VEC_PHI3:%.*]] = phi <vscale x 2 x i64> [ poison, %[[VECTOR_PH]] ], [ [[TMP9:%.*]], %[[VECTOR_BODY]] ]
; DATA-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[INDEX]], i64 [[N]])
; DATA-NEXT:    [[TMP2:%.*]] = getelementptr inbounds nuw [8 x i8], ptr [[ARR]], i64 [[INDEX]]
; DATA-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0(ptr align 4 [[TMP2]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i64> poison)
; DATA-NEXT:    [[TMP5:%.*]] = icmp slt <vscale x 2 x i64> [[WIDE_MASKED_LOAD]], [[VEC_PHI]]
; DATA-NEXT:    [[TMP6:%.*]] = call <vscale x 2 x i64> @llvm.smin.nxv2i64(<vscale x 2 x i64> [[WIDE_MASKED_LOAD]], <vscale x 2 x i64> [[VEC_PHI]])
; DATA-NEXT:    [[TMP7]] = select <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[VEC_PHI]]
; DATA-NEXT:    [[TMP8:%.*]] = select <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i1> [[TMP5]], <vscale x 2 x i1> zeroinitializer
; DATA-NEXT:    [[TMP9]] = select <vscale x 2 x i1> [[TMP8]], <vscale x 2 x i64> [[VEC_IND]], <vscale x 2 x i64> [[VEC_PHI3]]
; DATA-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP1]]
; DATA-NEXT:    [[VEC_IND_NEXT]] = add nuw nsw <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT2]]
; DATA-NEXT:    [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; DATA-NEXT:    br i1 [[TMP29]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; DATA:       [[MIDDLE_BLOCK]]:
; DATA-NEXT:    [[TMP30:%.*]] = call i64 @llvm.vector.reduce.smin.nxv2i64(<vscale x 2 x i64> [[TMP7]])
; DATA-NEXT:    [[BROADCAST_SPLATINSERT4:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP30]], i64 0
; DATA-NEXT:    [[BROADCAST_SPLAT5:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT4]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA-NEXT:    [[TMP12:%.*]] = icmp eq <vscale x 2 x i64> [[TMP7]], [[BROADCAST_SPLAT5]]
; DATA-NEXT:    [[TMP13:%.*]] = select <vscale x 2 x i1> [[TMP12]], <vscale x 2 x i64> [[TMP9]], <vscale x 2 x i64> splat (i64 -1)
; DATA-NEXT:    [[TMP33:%.*]] = call i64 @llvm.vector.reduce.umin.nxv2i64(<vscale x 2 x i64> [[TMP13]])
; DATA-NEXT:    [[TMP34:%.*]] = icmp eq i64 [[TMP30]], [[START]]
; DATA-NEXT:    [[TMP35:%.*]] = select i1 [[TMP34]], i64 0, i64 [[TMP33]]
; DATA-NEXT:    br label %[[FOR_END:.*]]
; DATA:       [[FOR_END]]:
; DATA-NEXT:    ret i64 [[TMP35]]
;
; DATA_NO_LANEMASK-LABEL: define i64 @argmin_first_index(
; DATA_NO_LANEMASK-SAME: ptr [[ARR:%.*]], i64 [[N:%.*]], i64 [[START:%.*]]) {
; DATA_NO_LANEMASK-NEXT:  [[ENTRY:.*:]]
; DATA_NO_LANEMASK-NEXT:    br label %[[VECTOR_PH:.*]]
; DATA_NO_LANEMASK:       [[VECTOR_PH]]:
; DATA_NO_LANEMASK-NEXT:    [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
; DATA_NO_LANEMASK-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP5]], 1
; DATA_NO_LANEMASK-NEXT:    [[TMP2:%.*]] = sub i64 [[TMP1]], 1
; DATA_NO_LANEMASK-NEXT:    [[N_RND_UP:%.*]] = add i64 [[N]], [[TMP2]]
; DATA_NO_LANEMASK-NEXT:    [[TMP0:%.*]] = urem i64 [[N_RND_UP]], [[TMP1]]
; DATA_NO_LANEMASK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[TMP0]]
; DATA_NO_LANEMASK-NEXT:    [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[N]], 1
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[START]], i64 0
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_NO_LANEMASK-NEXT:    [[TMP3:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP1]], i64 0
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT3]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_NO_LANEMASK-NEXT:    br label %[[VECTOR_BODY:.*]]
; DATA_NO_LANEMASK:       [[VECTOR_BODY]]:
; DATA_NO_LANEMASK-NEXT:    [[TMP20:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; DATA_NO_LANEMASK-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 2 x i64> [ [[TMP3]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
; DATA_NO_LANEMASK-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ [[BROADCAST_SPLAT2]], %[[VECTOR_PH]] ], [ [[TMP8:%.*]], %[[VECTOR_BODY]] ]
; DATA_NO_LANEMASK-NEXT:    [[VEC_PHI5:%.*]] = phi <vscale x 2 x i64> [ poison, %[[VECTOR_PH]] ], [ [[TMP10:%.*]], %[[VECTOR_BODY]] ]
; DATA_NO_LANEMASK-NEXT:    [[TMP4:%.*]] = icmp ule <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
; DATA_NO_LANEMASK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [8 x i8], ptr [[ARR]], i64 [[TMP20]]
; DATA_NO_LANEMASK-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0(ptr align 4 [[TMP21]], <vscale x 2 x i1> [[TMP4]], <vscale x 2 x i64> poison)
; DATA_NO_LANEMASK-NEXT:    [[TMP6:%.*]] = icmp slt <vscale x 2 x i64> [[WIDE_MASKED_LOAD]], [[VEC_PHI]]
; DATA_NO_LANEMASK-NEXT:    [[TMP7:%.*]] = call <vscale x 2 x i64> @llvm.smin.nxv2i64(<vscale x 2 x i64> [[WIDE_MASKED_LOAD]], <vscale x 2 x i64> [[VEC_PHI]])
; DATA_NO_LANEMASK-NEXT:    [[TMP8]] = select <vscale x 2 x i1> [[TMP4]], <vscale x 2 x i64> [[TMP7]], <vscale x 2 x i64> [[VEC_PHI]]
; DATA_NO_LANEMASK-NEXT:    [[TMP9:%.*]] = select <vscale x 2 x i1> [[TMP4]], <vscale x 2 x i1> [[TMP6]], <vscale x 2 x i1> zeroinitializer
; DATA_NO_LANEMASK-NEXT:    [[TMP10]] = select <vscale x 2 x i1> [[TMP9]], <vscale x 2 x i64> [[VEC_IND]], <vscale x 2 x i64> [[VEC_PHI5]]
; DATA_NO_LANEMASK-NEXT:    [[INDEX_NEXT]] = add i64 [[TMP20]], [[TMP1]]
; DATA_NO_LANEMASK-NEXT:    [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT4]]
; DATA_NO_LANEMASK-NEXT:    [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; DATA_NO_LANEMASK-NEXT:    br i1 [[TMP30]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; DATA_NO_LANEMASK:       [[MIDDLE_BLOCK]]:
; DATA_NO_LANEMASK-NEXT:    [[TMP31:%.*]] = call i64 @llvm.vector.reduce.smin.nxv2i64(<vscale x 2 x i64> [[TMP8]])
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLATINSERT6:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP31]], i64 0
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLAT7:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT6]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_NO_LANEMASK-NEXT:    [[TMP13:%.*]] = icmp eq <vscale x 2 x i64> [[TMP8]], [[BROADCAST_SPLAT7]]
; DATA_NO_LANEMASK-NEXT:    [[TMP14:%.*]] = select <vscale x 2 x i1> [[TMP13]], <vscale x 2 x i64> [[TMP10]], <vscale x 2 x i64> splat (i64 -1)
; DATA_NO_LANEMASK-NEXT:    [[TMP34:%.*]] = call i64 @llvm.vector.reduce.umin.nxv2i64(<vscale x 2 x i64> [[TMP14]])
; DATA_NO_LANEMASK-NEXT:    [[TMP35:%.*]] = icmp eq i64 [[TMP31]], [[START]]
; DATA_NO_LANEMASK-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i64 0, i64 [[TMP34]]
; DATA_NO_LANEMASK-NEXT:    br label %[[FOR_END:.*]]
; DATA_NO_LANEMASK:       [[FOR_END]]:
; DATA_NO_LANEMASK-NEXT:    ret i64 [[TMP36]]
;
; DATA_AND_CONTROL-LABEL: define i64 @argmin_first_index(
; DATA_AND_CONTROL-SAME: ptr [[ARR:%.*]], i64 [[N:%.*]], i64 [[START:%.*]]) {
; DATA_AND_CONTROL-NEXT:  [[ENTRY:.*:]]
; DATA_AND_CONTROL-NEXT:    br label %[[VECTOR_PH:.*]]
; DATA_AND_CONTROL:       [[VECTOR_PH]]:
; DATA_AND_CONTROL-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
; DATA_AND_CONTROL-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 1
; DATA_AND_CONTROL-NEXT:    [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 0, i64 [[N]])
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[START]], i64 0
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_AND_CONTROL-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP1]], i64 0
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_AND_CONTROL-NEXT:    br label %[[PRED_LOAD_CONTINUE5:.*]]
; DATA_AND_CONTROL:       [[PRED_LOAD_CONTINUE5]]:
; DATA_AND_CONTROL-NEXT:    [[TMP18:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_LOAD_CONTINUE5]] ]
; DATA_AND_CONTROL-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], %[[VECTOR_PH]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], %[[PRED_LOAD_CONTINUE5]] ]
; DATA_AND_CONTROL-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 2 x i64> [ [[TMP2]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_LOAD_CONTINUE5]] ]
; DATA_AND_CONTROL-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ [[BROADCAST_SPLAT]], %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[PRED_LOAD_CONTINUE5]] ]
; DATA_AND_CONTROL-NEXT:    [[VEC_PHI3:%.*]] = phi <vscale x 2 x i64> [ poison, %[[VECTOR_PH]] ], [ [[TMP8:%.*]], %[[PRED_LOAD_CONTINUE5]] ]
; DATA_AND_CONTROL-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [8 x i8], ptr [[ARR]], i64 [[TMP18]]
; DATA_AND_CONTROL-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0(ptr align 4 [[TMP19]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i64> poison)
; DATA_AND_CONTROL-NEXT:    [[TMP4:%.*]] = icmp slt <vscale x 2 x i64> [[WIDE_MASKED_LOAD]], [[VEC_PHI]]
; DATA_AND_CONTROL-NEXT:    [[TMP5:%.*]] = call <vscale x 2 x i64> @llvm.smin.nxv2i64(<vscale x 2 x i64> [[WIDE_MASKED_LOAD]], <vscale x 2 x i64> [[VEC_PHI]])
; DATA_AND_CONTROL-NEXT:    [[TMP6]] = select <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[VEC_PHI]]
; DATA_AND_CONTROL-NEXT:    [[TMP7:%.*]] = select <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i1> [[TMP4]], <vscale x 2 x i1> zeroinitializer
; DATA_AND_CONTROL-NEXT:    [[TMP8]] = select <vscale x 2 x i1> [[TMP7]], <vscale x 2 x i64> [[VEC_IND]], <vscale x 2 x i64> [[VEC_PHI3]]
; DATA_AND_CONTROL-NEXT:    [[INDEX_NEXT]] = add i64 [[TMP18]], [[TMP1]]
; DATA_AND_CONTROL-NEXT:    [[ACTIVE_LANE_MASK_NEXT]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[INDEX_NEXT]], i64 [[N]])
; DATA_AND_CONTROL-NEXT:    [[TMP28:%.*]] = extractelement <vscale x 2 x i1> [[ACTIVE_LANE_MASK_NEXT]], i64 0
; DATA_AND_CONTROL-NEXT:    [[TMP29:%.*]] = xor i1 [[TMP28]], true
; DATA_AND_CONTROL-NEXT:    [[VEC_IND_NEXT]] = add nuw nsw <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT2]]
; DATA_AND_CONTROL-NEXT:    br i1 [[TMP29]], label %[[MIDDLE_BLOCK:.*]], label %[[PRED_LOAD_CONTINUE5]], !llvm.loop [[LOOP0:![0-9]+]]
; DATA_AND_CONTROL:       [[MIDDLE_BLOCK]]:
; DATA_AND_CONTROL-NEXT:    [[TMP30:%.*]] = call i64 @llvm.vector.reduce.smin.nxv2i64(<vscale x 2 x i64> [[TMP6]])
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLATINSERT4:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP30]], i64 0
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLAT5:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT4]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_AND_CONTROL-NEXT:    [[TMP12:%.*]] = icmp eq <vscale x 2 x i64> [[TMP6]], [[BROADCAST_SPLAT5]]
; DATA_AND_CONTROL-NEXT:    [[TMP13:%.*]] = select <vscale x 2 x i1> [[TMP12]], <vscale x 2 x i64> [[TMP8]], <vscale x 2 x i64> splat (i64 -1)
; DATA_AND_CONTROL-NEXT:    [[TMP33:%.*]] = call i64 @llvm.vector.reduce.umin.nxv2i64(<vscale x 2 x i64> [[TMP13]])
; DATA_AND_CONTROL-NEXT:    [[TMP34:%.*]] = icmp eq i64 [[TMP30]], [[START]]
; DATA_AND_CONTROL-NEXT:    [[TMP35:%.*]] = select i1 [[TMP34]], i64 0, i64 [[TMP33]]
; DATA_AND_CONTROL-NEXT:    br label %[[FOR_END:.*]]
; DATA_AND_CONTROL:       [[FOR_END]]:
; DATA_AND_CONTROL-NEXT:    ret i64 [[TMP35]]
;
entry:
  br label %for.body

for.body:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
  %rdx = phi i64 [ %start, %entry ], [ %min, %for.body ]
  %rdx2 = phi i64 [ 0, %entry ], [ %select, %for.body ]
  %arrayidx = getelementptr inbounds nuw [8 x i8], ptr %arr, i64 %iv
  %0 = load i64, ptr %arrayidx
  %cmp = icmp slt i64 %0, %rdx
  %select = select i1 %cmp, i64 %iv, i64 %rdx2
  %min = tail call i64 @llvm.smin.i64(i64 %0, i64 %rdx)
  %iv.next = add nuw nsw i64 %iv, 1
  %exitcond.not = icmp eq i64 %iv.next, %n
  br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !0

for.end:
  ret i64 %select
}


define i64 @argmin_last_index(ptr %arr, i64 %n, i64 %start) {
; DATA-LABEL: define i64 @argmin_last_index(
; DATA-SAME: ptr [[ARR:%.*]], i64 [[N:%.*]], i64 [[START:%.*]]) {
; DATA-NEXT:  [[ENTRY:.*:]]
; DATA-NEXT:    br label %[[VECTOR_PH:.*]]
; DATA:       [[VECTOR_PH]]:
; DATA-NEXT:    [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
; DATA-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP4]], 1
; DATA-NEXT:    [[TMP10:%.*]] = sub i64 [[TMP1]], 1
; DATA-NEXT:    [[N_RND_UP:%.*]] = add i64 [[N]], [[TMP10]]
; DATA-NEXT:    [[TMP0:%.*]] = urem i64 [[N_RND_UP]], [[TMP1]]
; DATA-NEXT:    [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[TMP0]]
; DATA-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[START]], i64 0
; DATA-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA-NEXT:    [[TMP3:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
; DATA-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP1]], i64 0
; DATA-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA-NEXT:    br label %[[VECTOR_BODY:.*]]
; DATA:       [[VECTOR_BODY]]:
; DATA-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; DATA-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 2 x i64> [ [[TMP3]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
; DATA-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ [[BROADCAST_SPLAT]], %[[VECTOR_PH]] ], [ [[TMP7:%.*]], %[[VECTOR_BODY]] ]
; DATA-NEXT:    [[VEC_PHI3:%.*]] = phi <vscale x 2 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP9:%.*]], %[[VECTOR_BODY]] ]
; DATA-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[INDEX]], i64 [[N]])
; DATA-NEXT:    [[TMP2:%.*]] = getelementptr inbounds nuw [8 x i8], ptr [[ARR]], i64 [[INDEX]]
; DATA-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0(ptr align 4 [[TMP2]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i64> poison)
; DATA-NEXT:    [[TMP5:%.*]] = icmp sle <vscale x 2 x i64> [[WIDE_MASKED_LOAD]], [[VEC_PHI]]
; DATA-NEXT:    [[TMP6:%.*]] = call <vscale x 2 x i64> @llvm.smin.nxv2i64(<vscale x 2 x i64> [[WIDE_MASKED_LOAD]], <vscale x 2 x i64> [[VEC_PHI]])
; DATA-NEXT:    [[TMP7]] = select <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[VEC_PHI]]
; DATA-NEXT:    [[TMP8:%.*]] = select <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i1> [[TMP5]], <vscale x 2 x i1> zeroinitializer
; DATA-NEXT:    [[TMP9]] = select <vscale x 2 x i1> [[TMP8]], <vscale x 2 x i64> [[VEC_IND]], <vscale x 2 x i64> [[VEC_PHI3]]
; DATA-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP1]]
; DATA-NEXT:    [[VEC_IND_NEXT]] = add nuw nsw <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT2]]
; DATA-NEXT:    [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; DATA-NEXT:    br i1 [[TMP29]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
; DATA:       [[MIDDLE_BLOCK]]:
; DATA-NEXT:    [[TMP11:%.*]] = call i64 @llvm.vector.reduce.smin.nxv2i64(<vscale x 2 x i64> [[TMP7]])
; DATA-NEXT:    [[BROADCAST_SPLATINSERT4:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP11]], i64 0
; DATA-NEXT:    [[BROADCAST_SPLAT5:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT4]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA-NEXT:    [[TMP12:%.*]] = icmp eq <vscale x 2 x i64> [[TMP7]], [[BROADCAST_SPLAT5]]
; DATA-NEXT:    [[TMP13:%.*]] = select <vscale x 2 x i1> [[TMP12]], <vscale x 2 x i64> [[TMP9]], <vscale x 2 x i64> splat (i64 -9223372036854775808)
; DATA-NEXT:    [[TMP33:%.*]] = call i64 @llvm.vector.reduce.smax.nxv2i64(<vscale x 2 x i64> [[TMP13]])
; DATA-NEXT:    [[TMP34:%.*]] = icmp ne i64 [[TMP33]], -9223372036854775808
; DATA-NEXT:    [[TMP35:%.*]] = select i1 [[TMP34]], i64 [[TMP33]], i64 0
; DATA-NEXT:    br label %[[FOR_END:.*]]
; DATA:       [[FOR_END]]:
; DATA-NEXT:    ret i64 [[TMP35]]
;
; DATA_NO_LANEMASK-LABEL: define i64 @argmin_last_index(
; DATA_NO_LANEMASK-SAME: ptr [[ARR:%.*]], i64 [[N:%.*]], i64 [[START:%.*]]) {
; DATA_NO_LANEMASK-NEXT:  [[ENTRY:.*:]]
; DATA_NO_LANEMASK-NEXT:    br label %[[VECTOR_PH:.*]]
; DATA_NO_LANEMASK:       [[VECTOR_PH]]:
; DATA_NO_LANEMASK-NEXT:    [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
; DATA_NO_LANEMASK-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP5]], 1
; DATA_NO_LANEMASK-NEXT:    [[TMP2:%.*]] = sub i64 [[TMP1]], 1
; DATA_NO_LANEMASK-NEXT:    [[N_RND_UP:%.*]] = add i64 [[N]], [[TMP2]]
; DATA_NO_LANEMASK-NEXT:    [[TMP0:%.*]] = urem i64 [[N_RND_UP]], [[TMP1]]
; DATA_NO_LANEMASK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[TMP0]]
; DATA_NO_LANEMASK-NEXT:    [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[N]], 1
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[START]], i64 0
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_NO_LANEMASK-NEXT:    [[TMP3:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP1]], i64 0
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT3]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_NO_LANEMASK-NEXT:    br label %[[VECTOR_BODY:.*]]
; DATA_NO_LANEMASK:       [[VECTOR_BODY]]:
; DATA_NO_LANEMASK-NEXT:    [[TMP20:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; DATA_NO_LANEMASK-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 2 x i64> [ [[TMP3]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
; DATA_NO_LANEMASK-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ [[BROADCAST_SPLAT2]], %[[VECTOR_PH]] ], [ [[TMP8:%.*]], %[[VECTOR_BODY]] ]
; DATA_NO_LANEMASK-NEXT:    [[VEC_PHI5:%.*]] = phi <vscale x 2 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP10:%.*]], %[[VECTOR_BODY]] ]
; DATA_NO_LANEMASK-NEXT:    [[TMP4:%.*]] = icmp ule <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
; DATA_NO_LANEMASK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [8 x i8], ptr [[ARR]], i64 [[TMP20]]
; DATA_NO_LANEMASK-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0(ptr align 4 [[TMP21]], <vscale x 2 x i1> [[TMP4]], <vscale x 2 x i64> poison)
; DATA_NO_LANEMASK-NEXT:    [[TMP6:%.*]] = icmp sle <vscale x 2 x i64> [[WIDE_MASKED_LOAD]], [[VEC_PHI]]
; DATA_NO_LANEMASK-NEXT:    [[TMP7:%.*]] = call <vscale x 2 x i64> @llvm.smin.nxv2i64(<vscale x 2 x i64> [[WIDE_MASKED_LOAD]], <vscale x 2 x i64> [[VEC_PHI]])
; DATA_NO_LANEMASK-NEXT:    [[TMP8]] = select <vscale x 2 x i1> [[TMP4]], <vscale x 2 x i64> [[TMP7]], <vscale x 2 x i64> [[VEC_PHI]]
; DATA_NO_LANEMASK-NEXT:    [[TMP9:%.*]] = select <vscale x 2 x i1> [[TMP4]], <vscale x 2 x i1> [[TMP6]], <vscale x 2 x i1> zeroinitializer
; DATA_NO_LANEMASK-NEXT:    [[TMP10]] = select <vscale x 2 x i1> [[TMP9]], <vscale x 2 x i64> [[VEC_IND]], <vscale x 2 x i64> [[VEC_PHI5]]
; DATA_NO_LANEMASK-NEXT:    [[INDEX_NEXT]] = add i64 [[TMP20]], [[TMP1]]
; DATA_NO_LANEMASK-NEXT:    [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT4]]
; DATA_NO_LANEMASK-NEXT:    [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; DATA_NO_LANEMASK-NEXT:    br i1 [[TMP30]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
; DATA_NO_LANEMASK:       [[MIDDLE_BLOCK]]:
; DATA_NO_LANEMASK-NEXT:    [[TMP12:%.*]] = call i64 @llvm.vector.reduce.smin.nxv2i64(<vscale x 2 x i64> [[TMP8]])
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLATINSERT6:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP12]], i64 0
; DATA_NO_LANEMASK-NEXT:    [[BROADCAST_SPLAT7:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT6]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_NO_LANEMASK-NEXT:    [[TMP13:%.*]] = icmp eq <vscale x 2 x i64> [[TMP8]], [[BROADCAST_SPLAT7]]
; DATA_NO_LANEMASK-NEXT:    [[TMP14:%.*]] = select <vscale x 2 x i1> [[TMP13]], <vscale x 2 x i64> [[TMP10]], <vscale x 2 x i64> splat (i64 -9223372036854775808)
; DATA_NO_LANEMASK-NEXT:    [[TMP34:%.*]] = call i64 @llvm.vector.reduce.smax.nxv2i64(<vscale x 2 x i64> [[TMP14]])
; DATA_NO_LANEMASK-NEXT:    [[TMP35:%.*]] = icmp ne i64 [[TMP34]], -9223372036854775808
; DATA_NO_LANEMASK-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i64 [[TMP34]], i64 0
; DATA_NO_LANEMASK-NEXT:    br label %[[FOR_END:.*]]
; DATA_NO_LANEMASK:       [[FOR_END]]:
; DATA_NO_LANEMASK-NEXT:    ret i64 [[TMP36]]
;
; DATA_AND_CONTROL-LABEL: define i64 @argmin_last_index(
; DATA_AND_CONTROL-SAME: ptr [[ARR:%.*]], i64 [[N:%.*]], i64 [[START:%.*]]) {
; DATA_AND_CONTROL-NEXT:  [[ENTRY:.*:]]
; DATA_AND_CONTROL-NEXT:    br label %[[VECTOR_PH:.*]]
; DATA_AND_CONTROL:       [[VECTOR_PH]]:
; DATA_AND_CONTROL-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
; DATA_AND_CONTROL-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 1
; DATA_AND_CONTROL-NEXT:    [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 0, i64 [[N]])
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[START]], i64 0
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_AND_CONTROL-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP1]], i64 0
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_AND_CONTROL-NEXT:    br label %[[PRED_LOAD_CONTINUE5:.*]]
; DATA_AND_CONTROL:       [[PRED_LOAD_CONTINUE5]]:
; DATA_AND_CONTROL-NEXT:    [[TMP18:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_LOAD_CONTINUE5]] ]
; DATA_AND_CONTROL-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], %[[VECTOR_PH]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], %[[PRED_LOAD_CONTINUE5]] ]
; DATA_AND_CONTROL-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 2 x i64> [ [[TMP2]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_LOAD_CONTINUE5]] ]
; DATA_AND_CONTROL-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ [[BROADCAST_SPLAT]], %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[PRED_LOAD_CONTINUE5]] ]
; DATA_AND_CONTROL-NEXT:    [[VEC_PHI3:%.*]] = phi <vscale x 2 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP8:%.*]], %[[PRED_LOAD_CONTINUE5]] ]
; DATA_AND_CONTROL-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [8 x i8], ptr [[ARR]], i64 [[TMP18]]
; DATA_AND_CONTROL-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0(ptr align 4 [[TMP19]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i64> poison)
; DATA_AND_CONTROL-NEXT:    [[TMP4:%.*]] = icmp sle <vscale x 2 x i64> [[WIDE_MASKED_LOAD]], [[VEC_PHI]]
; DATA_AND_CONTROL-NEXT:    [[TMP5:%.*]] = call <vscale x 2 x i64> @llvm.smin.nxv2i64(<vscale x 2 x i64> [[WIDE_MASKED_LOAD]], <vscale x 2 x i64> [[VEC_PHI]])
; DATA_AND_CONTROL-NEXT:    [[TMP6]] = select <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[VEC_PHI]]
; DATA_AND_CONTROL-NEXT:    [[TMP7:%.*]] = select <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i1> [[TMP4]], <vscale x 2 x i1> zeroinitializer
; DATA_AND_CONTROL-NEXT:    [[TMP8]] = select <vscale x 2 x i1> [[TMP7]], <vscale x 2 x i64> [[VEC_IND]], <vscale x 2 x i64> [[VEC_PHI3]]
; DATA_AND_CONTROL-NEXT:    [[INDEX_NEXT]] = add i64 [[TMP18]], [[TMP1]]
; DATA_AND_CONTROL-NEXT:    [[ACTIVE_LANE_MASK_NEXT]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[INDEX_NEXT]], i64 [[N]])
; DATA_AND_CONTROL-NEXT:    [[TMP28:%.*]] = extractelement <vscale x 2 x i1> [[ACTIVE_LANE_MASK_NEXT]], i64 0
; DATA_AND_CONTROL-NEXT:    [[TMP29:%.*]] = xor i1 [[TMP28]], true
; DATA_AND_CONTROL-NEXT:    [[VEC_IND_NEXT]] = add nuw nsw <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT2]]
; DATA_AND_CONTROL-NEXT:    br i1 [[TMP29]], label %[[MIDDLE_BLOCK:.*]], label %[[PRED_LOAD_CONTINUE5]], !llvm.loop [[LOOP3:![0-9]+]]
; DATA_AND_CONTROL:       [[MIDDLE_BLOCK]]:
; DATA_AND_CONTROL-NEXT:    [[TMP11:%.*]] = call i64 @llvm.vector.reduce.smin.nxv2i64(<vscale x 2 x i64> [[TMP6]])
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLATINSERT4:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP11]], i64 0
; DATA_AND_CONTROL-NEXT:    [[BROADCAST_SPLAT5:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT4]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; DATA_AND_CONTROL-NEXT:    [[TMP12:%.*]] = icmp eq <vscale x 2 x i64> [[TMP6]], [[BROADCAST_SPLAT5]]
; DATA_AND_CONTROL-NEXT:    [[TMP13:%.*]] = select <vscale x 2 x i1> [[TMP12]], <vscale x 2 x i64> [[TMP8]], <vscale x 2 x i64> splat (i64 -9223372036854775808)
; DATA_AND_CONTROL-NEXT:    [[TMP33:%.*]] = call i64 @llvm.vector.reduce.smax.nxv2i64(<vscale x 2 x i64> [[TMP13]])
; DATA_AND_CONTROL-NEXT:    [[TMP34:%.*]] = icmp ne i64 [[TMP33]], -9223372036854775808
; DATA_AND_CONTROL-NEXT:    [[TMP35:%.*]] = select i1 [[TMP34]], i64 [[TMP33]], i64 0
; DATA_AND_CONTROL-NEXT:    br label %[[FOR_END:.*]]
; DATA_AND_CONTROL:       [[FOR_END]]:
; DATA_AND_CONTROL-NEXT:    ret i64 [[TMP35]]
;
entry:
  br label %for.body

for.body:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
  %rdx = phi i64 [ %start, %entry ], [ %min, %for.body ]
  %rdx2 = phi i64 [ 0, %entry ], [ %select, %for.body ]
  %arrayidx = getelementptr inbounds nuw [8 x i8], ptr %arr, i64 %iv
  %0 = load i64, ptr %arrayidx
  %cmp = icmp sle i64 %0, %rdx
  %select = select i1 %cmp, i64 %iv, i64 %rdx2
  %min = tail call i64 @llvm.smin.i64(i64 %0, i64 %rdx)
  %iv.next = add nuw nsw i64 %iv, 1
  %exitcond.not = icmp eq i64 %iv.next, %n
  br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !0

for.end:
  ret i64 %select
}

!0 = distinct !{!0, !1}
!1 = !{!"llvm.loop.vectorize.enable"}