; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt %s -passes='function(scalarizer<load-store>,dce)' -S | FileCheck %s
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"

declare <4 x float> @ext(<4 x float>)
@g = global <4 x float> zeroinitializer

define void @f1(<4 x float> %init, ptr %base, i32 %count) {
; CHECK-LABEL: define void @f1(
; CHECK-SAME: <4 x float> [[INIT:%.*]], ptr [[BASE:%.*]], i32 [[COUNT:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    [[INIT_I0:%.*]] = extractelement <4 x float> [[INIT]], i64 0
; CHECK-NEXT:    [[INIT_I1:%.*]] = extractelement <4 x float> [[INIT]], i64 1
; CHECK-NEXT:    [[INIT_I2:%.*]] = extractelement <4 x float> [[INIT]], i64 2
; CHECK-NEXT:    [[INIT_I3:%.*]] = extractelement <4 x float> [[INIT]], i64 3
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[COUNT]], %[[ENTRY]] ], [ [[NEXTI:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[ACC_I0:%.*]] = phi float [ [[INIT_I0]], %[[ENTRY]] ], [ [[SEL_I0:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[ACC_I1:%.*]] = phi float [ [[INIT_I1]], %[[ENTRY]] ], [ [[SEL_I1:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[ACC_I2:%.*]] = phi float [ [[INIT_I2]], %[[ENTRY]] ], [ [[SEL_I2:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[ACC_I3:%.*]] = phi float [ [[INIT_I3]], %[[ENTRY]] ], [ [[SEL_I3:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[NEXTI]] = sub i32 [[I]], 1
; CHECK-NEXT:    [[PTR:%.*]] = getelementptr <4 x float>, ptr [[BASE]], i32 [[I]]
; CHECK-NEXT:    [[VAL_I0:%.*]] = load float, ptr [[PTR]], align 16
; CHECK-NEXT:    [[PTR_I1:%.*]] = getelementptr float, ptr [[PTR]], i32 1
; CHECK-NEXT:    [[VAL_I1:%.*]] = load float, ptr [[PTR_I1]], align 4
; CHECK-NEXT:    [[PTR_I2:%.*]] = getelementptr float, ptr [[PTR]], i32 2
; CHECK-NEXT:    [[VAL_I2:%.*]] = load float, ptr [[PTR_I2]], align 8
; CHECK-NEXT:    [[PTR_I3:%.*]] = getelementptr float, ptr [[PTR]], i32 3
; CHECK-NEXT:    [[VAL_I3:%.*]] = load float, ptr [[PTR_I3]], align 4
; CHECK-NEXT:    [[ADD_I0:%.*]] = fadd float [[VAL_I0]], [[VAL_I2]]
; CHECK-NEXT:    [[ADD_I1:%.*]] = fadd float [[VAL_I1]], [[VAL_I3]]
; CHECK-NEXT:    [[ADD_I2:%.*]] = fadd float [[ACC_I0]], [[ACC_I2]]
; CHECK-NEXT:    [[ADD_I3:%.*]] = fadd float [[ACC_I1]], [[ACC_I3]]
; CHECK-NEXT:    [[ADD_UPTO0:%.*]] = insertelement <4 x float> poison, float [[ADD_I0]], i64 0
; CHECK-NEXT:    [[ADD_UPTO1:%.*]] = insertelement <4 x float> [[ADD_UPTO0]], float [[ADD_I1]], i64 1
; CHECK-NEXT:    [[ADD_UPTO2:%.*]] = insertelement <4 x float> [[ADD_UPTO1]], float [[ADD_I2]], i64 2
; CHECK-NEXT:    [[ADD:%.*]] = insertelement <4 x float> [[ADD_UPTO2]], float [[ADD_I3]], i64 3
; CHECK-NEXT:    [[CALL:%.*]] = call <4 x float> @ext(<4 x float> [[ADD]])
; CHECK-NEXT:    [[CALL_I0:%.*]] = extractelement <4 x float> [[CALL]], i64 0
; CHECK-NEXT:    [[CMP_I0:%.*]] = fcmp ogt float [[CALL_I0]], 1.000000e+00
; CHECK-NEXT:    [[CALL_I1:%.*]] = extractelement <4 x float> [[CALL]], i64 1
; CHECK-NEXT:    [[CMP_I1:%.*]] = fcmp ogt float [[CALL_I1]], 2.000000e+00
; CHECK-NEXT:    [[CALL_I2:%.*]] = extractelement <4 x float> [[CALL]], i64 2
; CHECK-NEXT:    [[CMP_I2:%.*]] = fcmp ogt float [[CALL_I2]], 3.000000e+00
; CHECK-NEXT:    [[CALL_I3:%.*]] = extractelement <4 x float> [[CALL]], i64 3
; CHECK-NEXT:    [[CMP_I3:%.*]] = fcmp ogt float [[CALL_I3]], 4.000000e+00
; CHECK-NEXT:    [[SEL_I0]] = select i1 [[CMP_I0]], float [[CALL_I0]], float 5.000000e+00
; CHECK-NEXT:    [[SEL_I1]] = select i1 [[CMP_I1]], float [[CALL_I1]], float 6.000000e+00
; CHECK-NEXT:    [[SEL_I2]] = select i1 [[CMP_I2]], float [[CALL_I2]], float 7.000000e+00
; CHECK-NEXT:    [[SEL_I3]] = select i1 [[CMP_I3]], float [[CALL_I3]], float 8.000000e+00
; CHECK-NEXT:    store float [[SEL_I0]], ptr [[PTR]], align 16
; CHECK-NEXT:    store float [[SEL_I1]], ptr [[PTR_I1]], align 4
; CHECK-NEXT:    store float [[SEL_I2]], ptr [[PTR_I2]], align 8
; CHECK-NEXT:    store float [[SEL_I3]], ptr [[PTR_I3]], align 4
; CHECK-NEXT:    [[TEST:%.*]] = icmp eq i32 [[NEXTI]], 0
; CHECK-NEXT:    br i1 [[TEST]], label %[[LOOP]], label %[[EXIT:.*]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret void
;
entry:
  br label %loop

loop:
  %i = phi i32 [ %count, %entry ], [ %nexti, %loop ]
  %acc = phi <4 x float> [ %init, %entry ], [ %sel, %loop ]
  %nexti = sub i32 %i, 1

  %ptr = getelementptr <4 x float>, ptr %base, i32 %i
  %val = load <4 x float> , ptr %ptr
  %dval = bitcast <4 x float> %val to <2 x double>
  %dacc = bitcast <4 x float> %acc to <2 x double>
  %shuffle1 = shufflevector <2 x double> %dval, <2 x double> %dacc,
  <2 x i32> <i32 0, i32 2>
  %shuffle2 = shufflevector <2 x double> %dval, <2 x double> %dacc,
  <2 x i32> <i32 1, i32 3>
  %f1 = bitcast <2 x double> %shuffle1 to <4 x float>
  %f2 = bitcast <2 x double> %shuffle2 to <4 x float>
  %add = fadd <4 x float> %f1, %f2
  %call = call <4 x float> @ext(<4 x float> %add)
  %cmp = fcmp ogt <4 x float> %call,
  <float 1.0, float 2.0, float 3.0, float 4.0>
  %sel = select <4 x i1> %cmp, <4 x float> %call,
  <4 x float> <float 5.0, float 6.0, float 7.0, float 8.0>
  store <4 x float> %sel, ptr %ptr

  %test = icmp eq i32 %nexti, 0
  br i1 %test, label %loop, label %exit

exit:
  ret void
}

define void @f2(<4 x i32> %init, ptr %base, i32 %count) {
; CHECK-LABEL: define void @f2(
; CHECK-SAME: <4 x i32> [[INIT:%.*]], ptr [[BASE:%.*]], i32 [[COUNT:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    [[INIT_I0:%.*]] = extractelement <4 x i32> [[INIT]], i64 0
; CHECK-NEXT:    [[INIT_I1:%.*]] = extractelement <4 x i32> [[INIT]], i64 1
; CHECK-NEXT:    [[INIT_I2:%.*]] = extractelement <4 x i32> [[INIT]], i64 2
; CHECK-NEXT:    [[INIT_I3:%.*]] = extractelement <4 x i32> [[INIT]], i64 3
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[COUNT]], %[[ENTRY]] ], [ [[NEXTI:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[ACC_I0:%.*]] = phi i32 [ [[INIT_I0]], %[[ENTRY]] ], [ [[SEL_I0:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[ACC_I1:%.*]] = phi i32 [ [[INIT_I1]], %[[ENTRY]] ], [ [[SEL_I1:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[ACC_I2:%.*]] = phi i32 [ [[INIT_I2]], %[[ENTRY]] ], [ [[SEL_I2:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[ACC_I3:%.*]] = phi i32 [ [[INIT_I3]], %[[ENTRY]] ], [ [[SEL_I3:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[NEXTI]] = sub i32 [[I]], 1
; CHECK-NEXT:    [[PTR:%.*]] = getelementptr <4 x i8>, ptr [[BASE]], i32 [[I]]
; CHECK-NEXT:    [[VAL_I0:%.*]] = load i8, ptr [[PTR]], align 4
; CHECK-NEXT:    [[PTR_I1:%.*]] = getelementptr i8, ptr [[PTR]], i32 1
; CHECK-NEXT:    [[VAL_I1:%.*]] = load i8, ptr [[PTR_I1]], align 1
; CHECK-NEXT:    [[PTR_I2:%.*]] = getelementptr i8, ptr [[PTR]], i32 2
; CHECK-NEXT:    [[VAL_I2:%.*]] = load i8, ptr [[PTR_I2]], align 2
; CHECK-NEXT:    [[PTR_I3:%.*]] = getelementptr i8, ptr [[PTR]], i32 3
; CHECK-NEXT:    [[VAL_I3:%.*]] = load i8, ptr [[PTR_I3]], align 1
; CHECK-NEXT:    [[EXT_I0:%.*]] = sext i8 [[VAL_I0]] to i32
; CHECK-NEXT:    [[EXT_I1:%.*]] = sext i8 [[VAL_I1]] to i32
; CHECK-NEXT:    [[EXT_I2:%.*]] = sext i8 [[VAL_I2]] to i32
; CHECK-NEXT:    [[EXT_I3:%.*]] = sext i8 [[VAL_I3]] to i32
; CHECK-NEXT:    [[ADD_I0:%.*]] = add i32 [[EXT_I0]], [[ACC_I0]]
; CHECK-NEXT:    [[ADD_I1:%.*]] = add i32 [[EXT_I1]], [[ACC_I1]]
; CHECK-NEXT:    [[ADD_I2:%.*]] = add i32 [[EXT_I2]], [[ACC_I2]]
; CHECK-NEXT:    [[ADD_I3:%.*]] = add i32 [[EXT_I3]], [[ACC_I3]]
; CHECK-NEXT:    [[CMP_I0:%.*]] = icmp slt i32 [[ADD_I0]], -10
; CHECK-NEXT:    [[CMP_I1:%.*]] = icmp slt i32 [[ADD_I1]], -11
; CHECK-NEXT:    [[CMP_I2:%.*]] = icmp slt i32 [[ADD_I2]], -12
; CHECK-NEXT:    [[CMP_I3:%.*]] = icmp slt i32 [[ADD_I3]], -13
; CHECK-NEXT:    [[SEL_I0]] = select i1 [[CMP_I0]], i32 [[ADD_I0]], i32 [[I]]
; CHECK-NEXT:    [[SEL_I1]] = select i1 [[CMP_I1]], i32 [[ADD_I1]], i32 [[I]]
; CHECK-NEXT:    [[SEL_I2]] = select i1 [[CMP_I2]], i32 [[ADD_I2]], i32 [[I]]
; CHECK-NEXT:    [[SEL_I3]] = select i1 [[CMP_I3]], i32 [[ADD_I3]], i32 [[I]]
; CHECK-NEXT:    [[TRUNC_I0:%.*]] = trunc i32 [[SEL_I0]] to i8
; CHECK-NEXT:    [[TRUNC_I1:%.*]] = trunc i32 [[SEL_I1]] to i8
; CHECK-NEXT:    [[TRUNC_I2:%.*]] = trunc i32 [[SEL_I2]] to i8
; CHECK-NEXT:    [[TRUNC_I3:%.*]] = trunc i32 [[SEL_I3]] to i8
; CHECK-NEXT:    store i8 [[TRUNC_I0]], ptr [[PTR]], align 4
; CHECK-NEXT:    store i8 [[TRUNC_I1]], ptr [[PTR_I1]], align 1
; CHECK-NEXT:    store i8 [[TRUNC_I2]], ptr [[PTR_I2]], align 2
; CHECK-NEXT:    store i8 [[TRUNC_I3]], ptr [[PTR_I3]], align 1
; CHECK-NEXT:    [[TEST:%.*]] = icmp eq i32 [[NEXTI]], 0
; CHECK-NEXT:    br i1 [[TEST]], label %[[LOOP]], label %[[EXIT:.*]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret void
;
entry:
  br label %loop

loop:
  %i = phi i32 [ %count, %entry ], [ %nexti, %loop ]
  %acc = phi <4 x i32> [ %init, %entry ], [ %sel, %loop ]
  %nexti = sub i32 %i, 1

  %ptr = getelementptr <4 x i8>, ptr %base, i32 %i
  %val = load <4 x i8> , ptr %ptr
  %ext = sext <4 x i8> %val to <4 x i32>
  %add = add <4 x i32> %ext, %acc
  %cmp = icmp slt <4 x i32> %add, <i32 -10, i32 -11, i32 -12, i32 -13>
  %single = insertelement <4 x i32> poison, i32 %i, i32 0
  %limit = shufflevector <4 x i32> %single, <4 x i32> poison,
  <4 x i32> zeroinitializer
  %sel = select <4 x i1> %cmp, <4 x i32> %add, <4 x i32> %limit
  %trunc = trunc <4 x i32> %sel to <4 x i8>
  store <4 x i8> %trunc, ptr %ptr

  %test = icmp eq i32 %nexti, 0
  br i1 %test, label %loop, label %exit

exit:
  ret void
}

; Check that !tbaa information is preserved.
define void @f3(ptr %src, ptr %dst) {
; CHECK-LABEL: define void @f3(
; CHECK-SAME: ptr [[SRC:%.*]], ptr [[DST:%.*]]) {
; CHECK-NEXT:    [[DST_I1:%.*]] = getelementptr i32, ptr [[DST]], i32 1
; CHECK-NEXT:    [[DST_I2:%.*]] = getelementptr i32, ptr [[DST]], i32 2
; CHECK-NEXT:    [[DST_I3:%.*]] = getelementptr i32, ptr [[DST]], i32 3
; CHECK-NEXT:    [[VAL_I0:%.*]] = load i32, ptr [[SRC]], align 16, !tbaa [[SET1_TBAA0:![0-9]+]]
; CHECK-NEXT:    [[SRC_I1:%.*]] = getelementptr i32, ptr [[SRC]], i32 1
; CHECK-NEXT:    [[VAL_I1:%.*]] = load i32, ptr [[SRC_I1]], align 4, !tbaa [[SET1_TBAA0]]
; CHECK-NEXT:    [[SRC_I2:%.*]] = getelementptr i32, ptr [[SRC]], i32 2
; CHECK-NEXT:    [[VAL_I2:%.*]] = load i32, ptr [[SRC_I2]], align 8, !tbaa [[SET1_TBAA0]]
; CHECK-NEXT:    [[SRC_I3:%.*]] = getelementptr i32, ptr [[SRC]], i32 3
; CHECK-NEXT:    [[VAL_I3:%.*]] = load i32, ptr [[SRC_I3]], align 4, !tbaa [[SET1_TBAA0]]
; CHECK-NEXT:    [[ADD_I0:%.*]] = add i32 [[VAL_I0]], [[VAL_I0]]
; CHECK-NEXT:    [[ADD_I1:%.*]] = add i32 [[VAL_I1]], [[VAL_I1]]
; CHECK-NEXT:    [[ADD_I2:%.*]] = add i32 [[VAL_I2]], [[VAL_I2]]
; CHECK-NEXT:    [[ADD_I3:%.*]] = add i32 [[VAL_I3]], [[VAL_I3]]
; CHECK-NEXT:    store i32 [[ADD_I0]], ptr [[DST]], align 16, !tbaa [[SET2_TBAA3:![0-9]+]]
; CHECK-NEXT:    store i32 [[ADD_I1]], ptr [[DST_I1]], align 4, !tbaa [[SET2_TBAA3]]
; CHECK-NEXT:    store i32 [[ADD_I2]], ptr [[DST_I2]], align 8, !tbaa [[SET2_TBAA3]]
; CHECK-NEXT:    store i32 [[ADD_I3]], ptr [[DST_I3]], align 4, !tbaa [[SET2_TBAA3]]
; CHECK-NEXT:    ret void
;
  %val = load <4 x i32> , ptr %src, !tbaa !1
  %add = add <4 x i32> %val, %val
  store <4 x i32> %add, ptr %dst, !tbaa !2
  ret void
}

; Check that !tbaa.struct information is preserved.
define void @f4(ptr %src, ptr %dst) {
; CHECK-LABEL: define void @f4(
; CHECK-SAME: ptr [[SRC:%.*]], ptr [[DST:%.*]]) {
; CHECK-NEXT:    [[DST_I1:%.*]] = getelementptr i32, ptr [[DST]], i32 1
; CHECK-NEXT:    [[DST_I2:%.*]] = getelementptr i32, ptr [[DST]], i32 2
; CHECK-NEXT:    [[DST_I3:%.*]] = getelementptr i32, ptr [[DST]], i32 3
; CHECK-NEXT:    [[VAL_I0:%.*]] = load i32, ptr [[SRC]], align 16, !tbaa.struct [[TBAA_STRUCT5:![0-9]+]]
; CHECK-NEXT:    [[SRC_I1:%.*]] = getelementptr i32, ptr [[SRC]], i32 1
; CHECK-NEXT:    [[VAL_I1:%.*]] = load i32, ptr [[SRC_I1]], align 4, !tbaa.struct [[TBAA_STRUCT5]]
; CHECK-NEXT:    [[SRC_I2:%.*]] = getelementptr i32, ptr [[SRC]], i32 2
; CHECK-NEXT:    [[VAL_I2:%.*]] = load i32, ptr [[SRC_I2]], align 8, !tbaa.struct [[TBAA_STRUCT5]]
; CHECK-NEXT:    [[SRC_I3:%.*]] = getelementptr i32, ptr [[SRC]], i32 3
; CHECK-NEXT:    [[VAL_I3:%.*]] = load i32, ptr [[SRC_I3]], align 4, !tbaa.struct [[TBAA_STRUCT5]]
; CHECK-NEXT:    [[ADD_I0:%.*]] = add i32 [[VAL_I0]], [[VAL_I0]]
; CHECK-NEXT:    [[ADD_I1:%.*]] = add i32 [[VAL_I1]], [[VAL_I1]]
; CHECK-NEXT:    [[ADD_I2:%.*]] = add i32 [[VAL_I2]], [[VAL_I2]]
; CHECK-NEXT:    [[ADD_I3:%.*]] = add i32 [[VAL_I3]], [[VAL_I3]]
; CHECK-NEXT:    store i32 [[ADD_I0]], ptr [[DST]], align 16, !tbaa.struct [[TBAA_STRUCT5]]
; CHECK-NEXT:    store i32 [[ADD_I1]], ptr [[DST_I1]], align 4, !tbaa.struct [[TBAA_STRUCT5]]
; CHECK-NEXT:    store i32 [[ADD_I2]], ptr [[DST_I2]], align 8, !tbaa.struct [[TBAA_STRUCT5]]
; CHECK-NEXT:    store i32 [[ADD_I3]], ptr [[DST_I3]], align 4, !tbaa.struct [[TBAA_STRUCT5]]
; CHECK-NEXT:    ret void
;
  %val = load <4 x i32> , ptr %src, !tbaa.struct !5
  %add = add <4 x i32> %val, %val
  store <4 x i32> %add, ptr %dst, !tbaa.struct !5
  ret void
}

; Check that llvm.access.group information is preserved.
define void @f5(i32 %count, ptr %src, ptr %dst) {
; CHECK-LABEL: define void @f5(
; CHECK-SAME: i32 [[COUNT:%.*]], ptr [[SRC:%.*]], ptr [[DST:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[NEXT_INDEX:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[THIS_SRC:%.*]] = getelementptr <4 x i32>, ptr [[SRC]], i32 [[INDEX]]
; CHECK-NEXT:    [[THIS_SRC_I1:%.*]] = getelementptr i32, ptr [[THIS_SRC]], i32 1
; CHECK-NEXT:    [[THIS_SRC_I2:%.*]] = getelementptr i32, ptr [[THIS_SRC]], i32 2
; CHECK-NEXT:    [[THIS_SRC_I3:%.*]] = getelementptr i32, ptr [[THIS_SRC]], i32 3
; CHECK-NEXT:    [[THIS_DST:%.*]] = getelementptr <4 x i32>, ptr [[DST]], i32 [[INDEX]]
; CHECK-NEXT:    [[THIS_DST_I1:%.*]] = getelementptr i32, ptr [[THIS_DST]], i32 1
; CHECK-NEXT:    [[THIS_DST_I2:%.*]] = getelementptr i32, ptr [[THIS_DST]], i32 2
; CHECK-NEXT:    [[THIS_DST_I3:%.*]] = getelementptr i32, ptr [[THIS_DST]], i32 3
; CHECK-NEXT:    [[VAL_I0:%.*]] = load i32, ptr [[THIS_SRC]], align 16, !llvm.access.group [[ACC_GRP6:![0-9]+]]
; CHECK-NEXT:    [[VAL_I1:%.*]] = load i32, ptr [[THIS_SRC_I1]], align 4, !llvm.access.group [[ACC_GRP6]]
; CHECK-NEXT:    [[VAL_I2:%.*]] = load i32, ptr [[THIS_SRC_I2]], align 8, !llvm.access.group [[ACC_GRP6]]
; CHECK-NEXT:    [[VAL_I3:%.*]] = load i32, ptr [[THIS_SRC_I3]], align 4, !llvm.access.group [[ACC_GRP6]]
; CHECK-NEXT:    [[ADD_I0:%.*]] = add i32 [[VAL_I0]], [[VAL_I0]]
; CHECK-NEXT:    [[ADD_I1:%.*]] = add i32 [[VAL_I1]], [[VAL_I1]]
; CHECK-NEXT:    [[ADD_I2:%.*]] = add i32 [[VAL_I2]], [[VAL_I2]]
; CHECK-NEXT:    [[ADD_I3:%.*]] = add i32 [[VAL_I3]], [[VAL_I3]]
; CHECK-NEXT:    store i32 [[ADD_I0]], ptr [[THIS_DST]], align 16, !llvm.access.group [[ACC_GRP6]]
; CHECK-NEXT:    store i32 [[ADD_I1]], ptr [[THIS_DST_I1]], align 4, !llvm.access.group [[ACC_GRP6]]
; CHECK-NEXT:    store i32 [[ADD_I2]], ptr [[THIS_DST_I2]], align 8, !llvm.access.group [[ACC_GRP6]]
; CHECK-NEXT:    store i32 [[ADD_I3]], ptr [[THIS_DST_I3]], align 4, !llvm.access.group [[ACC_GRP6]]
; CHECK-NEXT:    [[NEXT_INDEX]] = add i32 [[INDEX]], -1
; CHECK-NEXT:    [[CONTINUE:%.*]] = icmp ne i32 [[NEXT_INDEX]], [[COUNT]]
; CHECK-NEXT:    br i1 [[CONTINUE]], label %[[LOOP]], label %[[END:.*]], !llvm.loop [[LOOP7:![0-9]+]]
; CHECK:       [[END]]:
; CHECK-NEXT:    ret void
;
entry:
  br label %loop

loop:
  %index = phi i32 [ 0, %entry ], [ %next_index, %loop ]
  %this_src = getelementptr <4 x i32>, ptr %src, i32 %index
  %this_dst = getelementptr <4 x i32>, ptr %dst, i32 %index
  %val = load <4 x i32> , ptr %this_src, !llvm.access.group !13
  %add = add <4 x i32> %val, %val
  store <4 x i32> %add, ptr %this_dst, !llvm.access.group !13
  %next_index = add i32 %index, -1
  %continue = icmp ne i32 %next_index, %count
  br i1 %continue, label %loop, label %end, !llvm.loop !3

end:
  ret void
}

; Check that fpmath information is preserved.
define <4 x float> @f6(<4 x float> %x) {
; CHECK-LABEL: define <4 x float> @f6(
; CHECK-SAME: <4 x float> [[X:%.*]]) {
; CHECK-NEXT:    [[X_I0:%.*]] = extractelement <4 x float> [[X]], i64 0
; CHECK-NEXT:    [[RES_I0:%.*]] = fadd float [[X_I0]], 1.000000e+00, !fpmath [[META9:![0-9]+]]
; CHECK-NEXT:    [[X_I1:%.*]] = extractelement <4 x float> [[X]], i64 1
; CHECK-NEXT:    [[RES_I1:%.*]] = fadd float [[X_I1]], 2.000000e+00, !fpmath [[META9]]
; CHECK-NEXT:    [[X_I2:%.*]] = extractelement <4 x float> [[X]], i64 2
; CHECK-NEXT:    [[RES_I2:%.*]] = fadd float [[X_I2]], 3.000000e+00, !fpmath [[META9]]
; CHECK-NEXT:    [[X_I3:%.*]] = extractelement <4 x float> [[X]], i64 3
; CHECK-NEXT:    [[RES_I3:%.*]] = fadd float [[X_I3]], 4.000000e+00, !fpmath [[META9]]
; CHECK-NEXT:    [[RES_UPTO0:%.*]] = insertelement <4 x float> poison, float [[RES_I0]], i64 0
; CHECK-NEXT:    [[RES_UPTO1:%.*]] = insertelement <4 x float> [[RES_UPTO0]], float [[RES_I1]], i64 1
; CHECK-NEXT:    [[RES_UPTO2:%.*]] = insertelement <4 x float> [[RES_UPTO1]], float [[RES_I2]], i64 2
; CHECK-NEXT:    [[RES:%.*]] = insertelement <4 x float> [[RES_UPTO2]], float [[RES_I3]], i64 3
; CHECK-NEXT:    ret <4 x float> [[RES]]
;
  %res = fadd <4 x float> %x, <float 1.0, float 2.0, float 3.0, float 4.0>,
  !fpmath !4
  ret <4 x float> %res
}

; Check that random metadata isn't kept.
define void @f7(ptr %src, ptr %dst) {
; CHECK-LABEL: define void @f7(
; CHECK-SAME: ptr [[SRC:%.*]], ptr [[DST:%.*]]) {
; CHECK-NEXT:    [[DST_I1:%.*]] = getelementptr i32, ptr [[DST]], i32 1
; CHECK-NEXT:    [[DST_I2:%.*]] = getelementptr i32, ptr [[DST]], i32 2
; CHECK-NEXT:    [[DST_I3:%.*]] = getelementptr i32, ptr [[DST]], i32 3
; CHECK-NEXT:    [[VAL_I0:%.*]] = load i32, ptr [[SRC]], align 16
; CHECK-NEXT:    [[SRC_I1:%.*]] = getelementptr i32, ptr [[SRC]], i32 1
; CHECK-NEXT:    [[VAL_I1:%.*]] = load i32, ptr [[SRC_I1]], align 4
; CHECK-NEXT:    [[SRC_I2:%.*]] = getelementptr i32, ptr [[SRC]], i32 2
; CHECK-NEXT:    [[VAL_I2:%.*]] = load i32, ptr [[SRC_I2]], align 8
; CHECK-NEXT:    [[SRC_I3:%.*]] = getelementptr i32, ptr [[SRC]], i32 3
; CHECK-NEXT:    [[VAL_I3:%.*]] = load i32, ptr [[SRC_I3]], align 4
; CHECK-NEXT:    [[ADD_I0:%.*]] = add i32 [[VAL_I0]], [[VAL_I0]]
; CHECK-NEXT:    [[ADD_I1:%.*]] = add i32 [[VAL_I1]], [[VAL_I1]]
; CHECK-NEXT:    [[ADD_I2:%.*]] = add i32 [[VAL_I2]], [[VAL_I2]]
; CHECK-NEXT:    [[ADD_I3:%.*]] = add i32 [[VAL_I3]], [[VAL_I3]]
; CHECK-NEXT:    store i32 [[ADD_I0]], ptr [[DST]], align 16
; CHECK-NEXT:    store i32 [[ADD_I1]], ptr [[DST_I1]], align 4
; CHECK-NEXT:    store i32 [[ADD_I2]], ptr [[DST_I2]], align 8
; CHECK-NEXT:    store i32 [[ADD_I3]], ptr [[DST_I3]], align 4
; CHECK-NEXT:    ret void
;
  %val = load <4 x i32> , ptr %src, !foo !5
  %add = add <4 x i32> %val, %val
  store <4 x i32> %add, ptr %dst, !foo !5
  ret void
}

; Test GEP with vectors.
define void @f8(ptr %dest, <4 x ptr> %ptr0, <4 x i32> %i0,
; CHECK-LABEL: define void @f8(
; CHECK-SAME: ptr [[DEST:%.*]], <4 x ptr> [[PTR0:%.*]], <4 x i32> [[I0:%.*]], ptr [[OTHER:%.*]]) {
; CHECK-NEXT:    [[DEST_I1:%.*]] = getelementptr ptr, ptr [[DEST]], i32 1
; CHECK-NEXT:    [[DEST_I2:%.*]] = getelementptr ptr, ptr [[DEST]], i32 2
; CHECK-NEXT:    [[DEST_I3:%.*]] = getelementptr ptr, ptr [[DEST]], i32 3
; CHECK-NEXT:    [[PTR0_I0:%.*]] = extractelement <4 x ptr> [[PTR0]], i64 0
; CHECK-NEXT:    [[PTR0_I2:%.*]] = extractelement <4 x ptr> [[PTR0]], i64 2
; CHECK-NEXT:    [[PTR0_I3:%.*]] = extractelement <4 x ptr> [[PTR0]], i64 3
; CHECK-NEXT:    [[I0_I1:%.*]] = extractelement <4 x i32> [[I0]], i64 1
; CHECK-NEXT:    [[I0_I3:%.*]] = extractelement <4 x i32> [[I0]], i64 3
; CHECK-NEXT:    [[VAL_I0:%.*]] = getelementptr float, ptr [[PTR0_I0]], i32 100
; CHECK-NEXT:    [[VAL_I1:%.*]] = getelementptr float, ptr [[OTHER]], i32 [[I0_I1]]
; CHECK-NEXT:    [[VAL_I2:%.*]] = getelementptr float, ptr [[PTR0_I2]], i32 100
; CHECK-NEXT:    [[VAL_I3:%.*]] = getelementptr float, ptr [[PTR0_I3]], i32 [[I0_I3]]
; CHECK-NEXT:    store ptr [[VAL_I0]], ptr [[DEST]], align 32
; CHECK-NEXT:    store ptr [[VAL_I1]], ptr [[DEST_I1]], align 8
; CHECK-NEXT:    store ptr [[VAL_I2]], ptr [[DEST_I2]], align 16
; CHECK-NEXT:    store ptr [[VAL_I3]], ptr [[DEST_I3]], align 8
; CHECK-NEXT:    ret void
;
  ptr %other) {
  %i1 = insertelement <4 x i32> %i0, i32 100, i32 0
  %i2 = insertelement <4 x i32> %i1, i32 100, i32 2
  %ptr1 = insertelement <4 x ptr> %ptr0, ptr %other, i32 1
  %val = getelementptr float, <4 x ptr> %ptr1, <4 x i32> %i2
  store <4 x ptr> %val, ptr %dest
  ret void
}

; Test the handling of unaligned loads.
define void @f9(ptr %dest, ptr %src) {
; CHECK-LABEL: define void @f9(
; CHECK-SAME: ptr [[DEST:%.*]], ptr [[SRC:%.*]]) {
; CHECK-NEXT:    [[DEST_I1:%.*]] = getelementptr float, ptr [[DEST]], i32 1
; CHECK-NEXT:    [[DEST_I2:%.*]] = getelementptr float, ptr [[DEST]], i32 2
; CHECK-NEXT:    [[DEST_I3:%.*]] = getelementptr float, ptr [[DEST]], i32 3
; CHECK-NEXT:    [[VAL_I0:%.*]] = load float, ptr [[SRC]], align 4
; CHECK-NEXT:    [[SRC_I1:%.*]] = getelementptr float, ptr [[SRC]], i32 1
; CHECK-NEXT:    [[VAL_I1:%.*]] = load float, ptr [[SRC_I1]], align 4
; CHECK-NEXT:    [[SRC_I2:%.*]] = getelementptr float, ptr [[SRC]], i32 2
; CHECK-NEXT:    [[VAL_I2:%.*]] = load float, ptr [[SRC_I2]], align 4
; CHECK-NEXT:    [[SRC_I3:%.*]] = getelementptr float, ptr [[SRC]], i32 3
; CHECK-NEXT:    [[VAL_I3:%.*]] = load float, ptr [[SRC_I3]], align 4
; CHECK-NEXT:    store float [[VAL_I0]], ptr [[DEST]], align 8
; CHECK-NEXT:    store float [[VAL_I1]], ptr [[DEST_I1]], align 4
; CHECK-NEXT:    store float [[VAL_I2]], ptr [[DEST_I2]], align 8
; CHECK-NEXT:    store float [[VAL_I3]], ptr [[DEST_I3]], align 4
; CHECK-NEXT:    ret void
;
  %val = load <4 x float> , ptr %src, align 4
  store <4 x float> %val, ptr %dest, align 8
  ret void
}

; ...and again with subelement alignment.
define void @f10(ptr %dest, ptr %src) {
; CHECK-LABEL: define void @f10(
; CHECK-SAME: ptr [[DEST:%.*]], ptr [[SRC:%.*]]) {
; CHECK-NEXT:    [[DEST_I1:%.*]] = getelementptr float, ptr [[DEST]], i32 1
; CHECK-NEXT:    [[DEST_I2:%.*]] = getelementptr float, ptr [[DEST]], i32 2
; CHECK-NEXT:    [[DEST_I3:%.*]] = getelementptr float, ptr [[DEST]], i32 3
; CHECK-NEXT:    [[VAL_I0:%.*]] = load float, ptr [[SRC]], align 1
; CHECK-NEXT:    [[SRC_I1:%.*]] = getelementptr float, ptr [[SRC]], i32 1
; CHECK-NEXT:    [[VAL_I1:%.*]] = load float, ptr [[SRC_I1]], align 1
; CHECK-NEXT:    [[SRC_I2:%.*]] = getelementptr float, ptr [[SRC]], i32 2
; CHECK-NEXT:    [[VAL_I2:%.*]] = load float, ptr [[SRC_I2]], align 1
; CHECK-NEXT:    [[SRC_I3:%.*]] = getelementptr float, ptr [[SRC]], i32 3
; CHECK-NEXT:    [[VAL_I3:%.*]] = load float, ptr [[SRC_I3]], align 1
; CHECK-NEXT:    store float [[VAL_I0]], ptr [[DEST]], align 2
; CHECK-NEXT:    store float [[VAL_I1]], ptr [[DEST_I1]], align 2
; CHECK-NEXT:    store float [[VAL_I2]], ptr [[DEST_I2]], align 2
; CHECK-NEXT:    store float [[VAL_I3]], ptr [[DEST_I3]], align 2
; CHECK-NEXT:    ret void
;
  %val = load <4 x float> , ptr %src, align 1
  store <4 x float> %val, ptr %dest, align 2
  ret void
}

; Test that sub-byte loads aren't scalarized.
define void @f11(ptr %dest, ptr %src0) {
; CHECK-LABEL: define void @f11(
; CHECK-SAME: ptr [[DEST:%.*]], ptr [[SRC0:%.*]]) {
; CHECK-NEXT:    [[SRC1:%.*]] = getelementptr <32 x i1>, ptr [[SRC0]], i32 1
; CHECK-NEXT:    [[VAL0:%.*]] = load <32 x i1>, ptr [[SRC0]], align 4
; CHECK-NEXT:    [[VAL0_I0:%.*]] = extractelement <32 x i1> [[VAL0]], i64 0
; CHECK-NEXT:    [[VAL0_I1:%.*]] = extractelement <32 x i1> [[VAL0]], i64 1
; CHECK-NEXT:    [[VAL0_I2:%.*]] = extractelement <32 x i1> [[VAL0]], i64 2
; CHECK-NEXT:    [[VAL0_I3:%.*]] = extractelement <32 x i1> [[VAL0]], i64 3
; CHECK-NEXT:    [[VAL0_I4:%.*]] = extractelement <32 x i1> [[VAL0]], i64 4
; CHECK-NEXT:    [[VAL0_I5:%.*]] = extractelement <32 x i1> [[VAL0]], i64 5
; CHECK-NEXT:    [[VAL0_I6:%.*]] = extractelement <32 x i1> [[VAL0]], i64 6
; CHECK-NEXT:    [[VAL0_I7:%.*]] = extractelement <32 x i1> [[VAL0]], i64 7
; CHECK-NEXT:    [[VAL0_I8:%.*]] = extractelement <32 x i1> [[VAL0]], i64 8
; CHECK-NEXT:    [[VAL0_I9:%.*]] = extractelement <32 x i1> [[VAL0]], i64 9
; CHECK-NEXT:    [[VAL0_I10:%.*]] = extractelement <32 x i1> [[VAL0]], i64 10
; CHECK-NEXT:    [[VAL0_I11:%.*]] = extractelement <32 x i1> [[VAL0]], i64 11
; CHECK-NEXT:    [[VAL0_I12:%.*]] = extractelement <32 x i1> [[VAL0]], i64 12
; CHECK-NEXT:    [[VAL0_I13:%.*]] = extractelement <32 x i1> [[VAL0]], i64 13
; CHECK-NEXT:    [[VAL0_I14:%.*]] = extractelement <32 x i1> [[VAL0]], i64 14
; CHECK-NEXT:    [[VAL0_I15:%.*]] = extractelement <32 x i1> [[VAL0]], i64 15
; CHECK-NEXT:    [[VAL0_I16:%.*]] = extractelement <32 x i1> [[VAL0]], i64 16
; CHECK-NEXT:    [[VAL0_I17:%.*]] = extractelement <32 x i1> [[VAL0]], i64 17
; CHECK-NEXT:    [[VAL0_I18:%.*]] = extractelement <32 x i1> [[VAL0]], i64 18
; CHECK-NEXT:    [[VAL0_I19:%.*]] = extractelement <32 x i1> [[VAL0]], i64 19
; CHECK-NEXT:    [[VAL0_I20:%.*]] = extractelement <32 x i1> [[VAL0]], i64 20
; CHECK-NEXT:    [[VAL0_I21:%.*]] = extractelement <32 x i1> [[VAL0]], i64 21
; CHECK-NEXT:    [[VAL0_I22:%.*]] = extractelement <32 x i1> [[VAL0]], i64 22
; CHECK-NEXT:    [[VAL0_I23:%.*]] = extractelement <32 x i1> [[VAL0]], i64 23
; CHECK-NEXT:    [[VAL0_I24:%.*]] = extractelement <32 x i1> [[VAL0]], i64 24
; CHECK-NEXT:    [[VAL0_I25:%.*]] = extractelement <32 x i1> [[VAL0]], i64 25
; CHECK-NEXT:    [[VAL0_I26:%.*]] = extractelement <32 x i1> [[VAL0]], i64 26
; CHECK-NEXT:    [[VAL0_I27:%.*]] = extractelement <32 x i1> [[VAL0]], i64 27
; CHECK-NEXT:    [[VAL0_I28:%.*]] = extractelement <32 x i1> [[VAL0]], i64 28
; CHECK-NEXT:    [[VAL0_I29:%.*]] = extractelement <32 x i1> [[VAL0]], i64 29
; CHECK-NEXT:    [[VAL0_I30:%.*]] = extractelement <32 x i1> [[VAL0]], i64 30
; CHECK-NEXT:    [[VAL0_I31:%.*]] = extractelement <32 x i1> [[VAL0]], i64 31
; CHECK-NEXT:    [[VAL1:%.*]] = load <32 x i1>, ptr [[SRC1]], align 4
; CHECK-NEXT:    [[VAL1_I0:%.*]] = extractelement <32 x i1> [[VAL1]], i64 0
; CHECK-NEXT:    [[AND_I0:%.*]] = and i1 [[VAL0_I0]], [[VAL1_I0]]
; CHECK-NEXT:    [[VAL1_I1:%.*]] = extractelement <32 x i1> [[VAL1]], i64 1
; CHECK-NEXT:    [[AND_I1:%.*]] = and i1 [[VAL0_I1]], [[VAL1_I1]]
; CHECK-NEXT:    [[VAL1_I2:%.*]] = extractelement <32 x i1> [[VAL1]], i64 2
; CHECK-NEXT:    [[AND_I2:%.*]] = and i1 [[VAL0_I2]], [[VAL1_I2]]
; CHECK-NEXT:    [[VAL1_I3:%.*]] = extractelement <32 x i1> [[VAL1]], i64 3
; CHECK-NEXT:    [[AND_I3:%.*]] = and i1 [[VAL0_I3]], [[VAL1_I3]]
; CHECK-NEXT:    [[VAL1_I4:%.*]] = extractelement <32 x i1> [[VAL1]], i64 4
; CHECK-NEXT:    [[AND_I4:%.*]] = and i1 [[VAL0_I4]], [[VAL1_I4]]
; CHECK-NEXT:    [[VAL1_I5:%.*]] = extractelement <32 x i1> [[VAL1]], i64 5
; CHECK-NEXT:    [[AND_I5:%.*]] = and i1 [[VAL0_I5]], [[VAL1_I5]]
; CHECK-NEXT:    [[VAL1_I6:%.*]] = extractelement <32 x i1> [[VAL1]], i64 6
; CHECK-NEXT:    [[AND_I6:%.*]] = and i1 [[VAL0_I6]], [[VAL1_I6]]
; CHECK-NEXT:    [[VAL1_I7:%.*]] = extractelement <32 x i1> [[VAL1]], i64 7
; CHECK-NEXT:    [[AND_I7:%.*]] = and i1 [[VAL0_I7]], [[VAL1_I7]]
; CHECK-NEXT:    [[VAL1_I8:%.*]] = extractelement <32 x i1> [[VAL1]], i64 8
; CHECK-NEXT:    [[AND_I8:%.*]] = and i1 [[VAL0_I8]], [[VAL1_I8]]
; CHECK-NEXT:    [[VAL1_I9:%.*]] = extractelement <32 x i1> [[VAL1]], i64 9
; CHECK-NEXT:    [[AND_I9:%.*]] = and i1 [[VAL0_I9]], [[VAL1_I9]]
; CHECK-NEXT:    [[VAL1_I10:%.*]] = extractelement <32 x i1> [[VAL1]], i64 10
; CHECK-NEXT:    [[AND_I10:%.*]] = and i1 [[VAL0_I10]], [[VAL1_I10]]
; CHECK-NEXT:    [[VAL1_I11:%.*]] = extractelement <32 x i1> [[VAL1]], i64 11
; CHECK-NEXT:    [[AND_I11:%.*]] = and i1 [[VAL0_I11]], [[VAL1_I11]]
; CHECK-NEXT:    [[VAL1_I12:%.*]] = extractelement <32 x i1> [[VAL1]], i64 12
; CHECK-NEXT:    [[AND_I12:%.*]] = and i1 [[VAL0_I12]], [[VAL1_I12]]
; CHECK-NEXT:    [[VAL1_I13:%.*]] = extractelement <32 x i1> [[VAL1]], i64 13
; CHECK-NEXT:    [[AND_I13:%.*]] = and i1 [[VAL0_I13]], [[VAL1_I13]]
; CHECK-NEXT:    [[VAL1_I14:%.*]] = extractelement <32 x i1> [[VAL1]], i64 14
; CHECK-NEXT:    [[AND_I14:%.*]] = and i1 [[VAL0_I14]], [[VAL1_I14]]
; CHECK-NEXT:    [[VAL1_I15:%.*]] = extractelement <32 x i1> [[VAL1]], i64 15
; CHECK-NEXT:    [[AND_I15:%.*]] = and i1 [[VAL0_I15]], [[VAL1_I15]]
; CHECK-NEXT:    [[VAL1_I16:%.*]] = extractelement <32 x i1> [[VAL1]], i64 16
; CHECK-NEXT:    [[AND_I16:%.*]] = and i1 [[VAL0_I16]], [[VAL1_I16]]
; CHECK-NEXT:    [[VAL1_I17:%.*]] = extractelement <32 x i1> [[VAL1]], i64 17
; CHECK-NEXT:    [[AND_I17:%.*]] = and i1 [[VAL0_I17]], [[VAL1_I17]]
; CHECK-NEXT:    [[VAL1_I18:%.*]] = extractelement <32 x i1> [[VAL1]], i64 18
; CHECK-NEXT:    [[AND_I18:%.*]] = and i1 [[VAL0_I18]], [[VAL1_I18]]
; CHECK-NEXT:    [[VAL1_I19:%.*]] = extractelement <32 x i1> [[VAL1]], i64 19
; CHECK-NEXT:    [[AND_I19:%.*]] = and i1 [[VAL0_I19]], [[VAL1_I19]]
; CHECK-NEXT:    [[VAL1_I20:%.*]] = extractelement <32 x i1> [[VAL1]], i64 20
; CHECK-NEXT:    [[AND_I20:%.*]] = and i1 [[VAL0_I20]], [[VAL1_I20]]
; CHECK-NEXT:    [[VAL1_I21:%.*]] = extractelement <32 x i1> [[VAL1]], i64 21
; CHECK-NEXT:    [[AND_I21:%.*]] = and i1 [[VAL0_I21]], [[VAL1_I21]]
; CHECK-NEXT:    [[VAL1_I22:%.*]] = extractelement <32 x i1> [[VAL1]], i64 22
; CHECK-NEXT:    [[AND_I22:%.*]] = and i1 [[VAL0_I22]], [[VAL1_I22]]
; CHECK-NEXT:    [[VAL1_I23:%.*]] = extractelement <32 x i1> [[VAL1]], i64 23
; CHECK-NEXT:    [[AND_I23:%.*]] = and i1 [[VAL0_I23]], [[VAL1_I23]]
; CHECK-NEXT:    [[VAL1_I24:%.*]] = extractelement <32 x i1> [[VAL1]], i64 24
; CHECK-NEXT:    [[AND_I24:%.*]] = and i1 [[VAL0_I24]], [[VAL1_I24]]
; CHECK-NEXT:    [[VAL1_I25:%.*]] = extractelement <32 x i1> [[VAL1]], i64 25
; CHECK-NEXT:    [[AND_I25:%.*]] = and i1 [[VAL0_I25]], [[VAL1_I25]]
; CHECK-NEXT:    [[VAL1_I26:%.*]] = extractelement <32 x i1> [[VAL1]], i64 26
; CHECK-NEXT:    [[AND_I26:%.*]] = and i1 [[VAL0_I26]], [[VAL1_I26]]
; CHECK-NEXT:    [[VAL1_I27:%.*]] = extractelement <32 x i1> [[VAL1]], i64 27
; CHECK-NEXT:    [[AND_I27:%.*]] = and i1 [[VAL0_I27]], [[VAL1_I27]]
; CHECK-NEXT:    [[VAL1_I28:%.*]] = extractelement <32 x i1> [[VAL1]], i64 28
; CHECK-NEXT:    [[AND_I28:%.*]] = and i1 [[VAL0_I28]], [[VAL1_I28]]
; CHECK-NEXT:    [[VAL1_I29:%.*]] = extractelement <32 x i1> [[VAL1]], i64 29
; CHECK-NEXT:    [[AND_I29:%.*]] = and i1 [[VAL0_I29]], [[VAL1_I29]]
; CHECK-NEXT:    [[VAL1_I30:%.*]] = extractelement <32 x i1> [[VAL1]], i64 30
; CHECK-NEXT:    [[AND_I30:%.*]] = and i1 [[VAL0_I30]], [[VAL1_I30]]
; CHECK-NEXT:    [[VAL1_I31:%.*]] = extractelement <32 x i1> [[VAL1]], i64 31
; CHECK-NEXT:    [[AND_I31:%.*]] = and i1 [[VAL0_I31]], [[VAL1_I31]]
; CHECK-NEXT:    [[AND_UPTO0:%.*]] = insertelement <32 x i1> poison, i1 [[AND_I0]], i64 0
; CHECK-NEXT:    [[AND_UPTO1:%.*]] = insertelement <32 x i1> [[AND_UPTO0]], i1 [[AND_I1]], i64 1
; CHECK-NEXT:    [[AND_UPTO2:%.*]] = insertelement <32 x i1> [[AND_UPTO1]], i1 [[AND_I2]], i64 2
; CHECK-NEXT:    [[AND_UPTO3:%.*]] = insertelement <32 x i1> [[AND_UPTO2]], i1 [[AND_I3]], i64 3
; CHECK-NEXT:    [[AND_UPTO4:%.*]] = insertelement <32 x i1> [[AND_UPTO3]], i1 [[AND_I4]], i64 4
; CHECK-NEXT:    [[AND_UPTO5:%.*]] = insertelement <32 x i1> [[AND_UPTO4]], i1 [[AND_I5]], i64 5
; CHECK-NEXT:    [[AND_UPTO6:%.*]] = insertelement <32 x i1> [[AND_UPTO5]], i1 [[AND_I6]], i64 6
; CHECK-NEXT:    [[AND_UPTO7:%.*]] = insertelement <32 x i1> [[AND_UPTO6]], i1 [[AND_I7]], i64 7
; CHECK-NEXT:    [[AND_UPTO8:%.*]] = insertelement <32 x i1> [[AND_UPTO7]], i1 [[AND_I8]], i64 8
; CHECK-NEXT:    [[AND_UPTO9:%.*]] = insertelement <32 x i1> [[AND_UPTO8]], i1 [[AND_I9]], i64 9
; CHECK-NEXT:    [[AND_UPTO10:%.*]] = insertelement <32 x i1> [[AND_UPTO9]], i1 [[AND_I10]], i64 10
; CHECK-NEXT:    [[AND_UPTO11:%.*]] = insertelement <32 x i1> [[AND_UPTO10]], i1 [[AND_I11]], i64 11
; CHECK-NEXT:    [[AND_UPTO12:%.*]] = insertelement <32 x i1> [[AND_UPTO11]], i1 [[AND_I12]], i64 12
; CHECK-NEXT:    [[AND_UPTO13:%.*]] = insertelement <32 x i1> [[AND_UPTO12]], i1 [[AND_I13]], i64 13
; CHECK-NEXT:    [[AND_UPTO14:%.*]] = insertelement <32 x i1> [[AND_UPTO13]], i1 [[AND_I14]], i64 14
; CHECK-NEXT:    [[AND_UPTO15:%.*]] = insertelement <32 x i1> [[AND_UPTO14]], i1 [[AND_I15]], i64 15
; CHECK-NEXT:    [[AND_UPTO16:%.*]] = insertelement <32 x i1> [[AND_UPTO15]], i1 [[AND_I16]], i64 16
; CHECK-NEXT:    [[AND_UPTO17:%.*]] = insertelement <32 x i1> [[AND_UPTO16]], i1 [[AND_I17]], i64 17
; CHECK-NEXT:    [[AND_UPTO18:%.*]] = insertelement <32 x i1> [[AND_UPTO17]], i1 [[AND_I18]], i64 18
; CHECK-NEXT:    [[AND_UPTO19:%.*]] = insertelement <32 x i1> [[AND_UPTO18]], i1 [[AND_I19]], i64 19
; CHECK-NEXT:    [[AND_UPTO20:%.*]] = insertelement <32 x i1> [[AND_UPTO19]], i1 [[AND_I20]], i64 20
; CHECK-NEXT:    [[AND_UPTO21:%.*]] = insertelement <32 x i1> [[AND_UPTO20]], i1 [[AND_I21]], i64 21
; CHECK-NEXT:    [[AND_UPTO22:%.*]] = insertelement <32 x i1> [[AND_UPTO21]], i1 [[AND_I22]], i64 22
; CHECK-NEXT:    [[AND_UPTO23:%.*]] = insertelement <32 x i1> [[AND_UPTO22]], i1 [[AND_I23]], i64 23
; CHECK-NEXT:    [[AND_UPTO24:%.*]] = insertelement <32 x i1> [[AND_UPTO23]], i1 [[AND_I24]], i64 24
; CHECK-NEXT:    [[AND_UPTO25:%.*]] = insertelement <32 x i1> [[AND_UPTO24]], i1 [[AND_I25]], i64 25
; CHECK-NEXT:    [[AND_UPTO26:%.*]] = insertelement <32 x i1> [[AND_UPTO25]], i1 [[AND_I26]], i64 26
; CHECK-NEXT:    [[AND_UPTO27:%.*]] = insertelement <32 x i1> [[AND_UPTO26]], i1 [[AND_I27]], i64 27
; CHECK-NEXT:    [[AND_UPTO28:%.*]] = insertelement <32 x i1> [[AND_UPTO27]], i1 [[AND_I28]], i64 28
; CHECK-NEXT:    [[AND_UPTO29:%.*]] = insertelement <32 x i1> [[AND_UPTO28]], i1 [[AND_I29]], i64 29
; CHECK-NEXT:    [[AND_UPTO30:%.*]] = insertelement <32 x i1> [[AND_UPTO29]], i1 [[AND_I30]], i64 30
; CHECK-NEXT:    [[AND:%.*]] = insertelement <32 x i1> [[AND_UPTO30]], i1 [[AND_I31]], i64 31
; CHECK-NEXT:    store <32 x i1> [[AND]], ptr [[DEST]], align 4
; CHECK-NEXT:    ret void
;
  %src1 = getelementptr <32 x i1>, ptr %src0, i32 1
  %val0 = load <32 x i1> , ptr %src0
  %val1 = load <32 x i1> , ptr %src1
  %and = and <32 x i1> %val0, %val1
  store <32 x i1> %and, ptr %dest
  ret void
}

; Test vector GEPs with more than one index.
define void @f13(ptr %dest, <4 x ptr> %ptr, <4 x i32> %i,
; CHECK-LABEL: define void @f13(
; CHECK-SAME: ptr [[DEST:%.*]], <4 x ptr> [[PTR:%.*]], <4 x i32> [[I:%.*]], ptr [[OTHER:%.*]]) {
; CHECK-NEXT:    [[DEST_I1:%.*]] = getelementptr ptr, ptr [[DEST]], i32 1
; CHECK-NEXT:    [[DEST_I2:%.*]] = getelementptr ptr, ptr [[DEST]], i32 2
; CHECK-NEXT:    [[DEST_I3:%.*]] = getelementptr ptr, ptr [[DEST]], i32 3
; CHECK-NEXT:    [[PTR_I0:%.*]] = extractelement <4 x ptr> [[PTR]], i64 0
; CHECK-NEXT:    [[I_I0:%.*]] = extractelement <4 x i32> [[I]], i64 0
; CHECK-NEXT:    [[VAL_I0:%.*]] = getelementptr inbounds [4 x float], ptr [[PTR_I0]], i32 0, i32 [[I_I0]]
; CHECK-NEXT:    [[PTR_I1:%.*]] = extractelement <4 x ptr> [[PTR]], i64 1
; CHECK-NEXT:    [[I_I1:%.*]] = extractelement <4 x i32> [[I]], i64 1
; CHECK-NEXT:    [[VAL_I1:%.*]] = getelementptr inbounds [4 x float], ptr [[PTR_I1]], i32 1, i32 [[I_I1]]
; CHECK-NEXT:    [[PTR_I2:%.*]] = extractelement <4 x ptr> [[PTR]], i64 2
; CHECK-NEXT:    [[I_I2:%.*]] = extractelement <4 x i32> [[I]], i64 2
; CHECK-NEXT:    [[VAL_I2:%.*]] = getelementptr inbounds [4 x float], ptr [[PTR_I2]], i32 2, i32 [[I_I2]]
; CHECK-NEXT:    [[PTR_I3:%.*]] = extractelement <4 x ptr> [[PTR]], i64 3
; CHECK-NEXT:    [[I_I3:%.*]] = extractelement <4 x i32> [[I]], i64 3
; CHECK-NEXT:    [[VAL_I3:%.*]] = getelementptr inbounds [4 x float], ptr [[PTR_I3]], i32 3, i32 [[I_I3]]
; CHECK-NEXT:    store ptr [[VAL_I0]], ptr [[DEST]], align 32
; CHECK-NEXT:    store ptr [[VAL_I1]], ptr [[DEST_I1]], align 8
; CHECK-NEXT:    store ptr [[VAL_I2]], ptr [[DEST_I2]], align 16
; CHECK-NEXT:    store ptr [[VAL_I3]], ptr [[DEST_I3]], align 8
; CHECK-NEXT:    ret void
;
  ptr %other) {
  %val = getelementptr inbounds [4 x float], <4 x ptr> %ptr,
  <4 x i32> <i32 0, i32 1, i32 2, i32 3>,
  <4 x i32> %i
  store <4 x ptr> %val, ptr %dest
  ret void
}

; Test combinations of vector and non-vector PHIs.
define <4 x float> @f14(<4 x float> %acc, i32 %count) {
; CHECK-LABEL: define <4 x float> @f14(
; CHECK-SAME: <4 x float> [[ACC:%.*]], i32 [[COUNT:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    [[ACC_I0:%.*]] = extractelement <4 x float> [[ACC]], i64 0
; CHECK-NEXT:    [[ACC_I1:%.*]] = extractelement <4 x float> [[ACC]], i64 1
; CHECK-NEXT:    [[ACC_I2:%.*]] = extractelement <4 x float> [[ACC]], i64 2
; CHECK-NEXT:    [[ACC_I3:%.*]] = extractelement <4 x float> [[ACC]], i64 3
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[THIS_ACC_I0:%.*]] = phi float [ [[ACC_I0]], %[[ENTRY]] ], [ [[NEXT_ACC_I0:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[THIS_ACC_I1:%.*]] = phi float [ [[ACC_I1]], %[[ENTRY]] ], [ [[NEXT_ACC_I1:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[THIS_ACC_I2:%.*]] = phi float [ [[ACC_I2]], %[[ENTRY]] ], [ [[NEXT_ACC_I2:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[THIS_ACC_I3:%.*]] = phi float [ [[ACC_I3]], %[[ENTRY]] ], [ [[NEXT_ACC_I3:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[THIS_COUNT:%.*]] = phi i32 [ [[COUNT]], %[[ENTRY]] ], [ [[NEXT_COUNT:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[THIS_ACC_UPTO0:%.*]] = insertelement <4 x float> poison, float [[THIS_ACC_I0]], i64 0
; CHECK-NEXT:    [[THIS_ACC_UPTO1:%.*]] = insertelement <4 x float> [[THIS_ACC_UPTO0]], float [[THIS_ACC_I1]], i64 1
; CHECK-NEXT:    [[THIS_ACC_UPTO2:%.*]] = insertelement <4 x float> [[THIS_ACC_UPTO1]], float [[THIS_ACC_I2]], i64 2
; CHECK-NEXT:    [[THIS_ACC:%.*]] = insertelement <4 x float> [[THIS_ACC_UPTO2]], float [[THIS_ACC_I3]], i64 3
; CHECK-NEXT:    [[FOO:%.*]] = call <4 x float> @ext(<4 x float> [[THIS_ACC]])
; CHECK-NEXT:    [[FOO_I0:%.*]] = extractelement <4 x float> [[FOO]], i64 0
; CHECK-NEXT:    [[NEXT_ACC_I0]] = fadd float [[THIS_ACC_I0]], [[FOO_I0]]
; CHECK-NEXT:    [[FOO_I1:%.*]] = extractelement <4 x float> [[FOO]], i64 1
; CHECK-NEXT:    [[NEXT_ACC_I1]] = fadd float [[THIS_ACC_I1]], [[FOO_I1]]
; CHECK-NEXT:    [[FOO_I2:%.*]] = extractelement <4 x float> [[FOO]], i64 2
; CHECK-NEXT:    [[NEXT_ACC_I2]] = fadd float [[THIS_ACC_I2]], [[FOO_I2]]
; CHECK-NEXT:    [[FOO_I3:%.*]] = extractelement <4 x float> [[FOO]], i64 3
; CHECK-NEXT:    [[NEXT_ACC_I3]] = fadd float [[THIS_ACC_I3]], [[FOO_I3]]
; CHECK-NEXT:    [[NEXT_ACC_UPTO0:%.*]] = insertelement <4 x float> poison, float [[NEXT_ACC_I0]], i64 0
; CHECK-NEXT:    [[NEXT_ACC_UPTO1:%.*]] = insertelement <4 x float> [[NEXT_ACC_UPTO0]], float [[NEXT_ACC_I1]], i64 1
; CHECK-NEXT:    [[NEXT_ACC_UPTO2:%.*]] = insertelement <4 x float> [[NEXT_ACC_UPTO1]], float [[NEXT_ACC_I2]], i64 2
; CHECK-NEXT:    [[NEXT_ACC:%.*]] = insertelement <4 x float> [[NEXT_ACC_UPTO2]], float [[NEXT_ACC_I3]], i64 3
; CHECK-NEXT:    [[NEXT_COUNT]] = sub i32 [[THIS_COUNT]], 1
; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[NEXT_COUNT]], 0
; CHECK-NEXT:    br i1 [[CMP]], label %[[LOOP]], label %[[EXIT:.*]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret <4 x float> [[NEXT_ACC]]
;
entry:
  br label %loop

loop:
  %this_acc = phi <4 x float> [ %acc, %entry ], [ %next_acc, %loop ]
  %this_count = phi i32 [ %count, %entry ], [ %next_count, %loop ]
  %foo = call <4 x float> @ext(<4 x float> %this_acc)
  %next_acc = fadd <4 x float> %this_acc, %foo
  %next_count = sub i32 %this_count, 1
  %cmp = icmp eq i32 %next_count, 0
  br i1 %cmp, label %loop, label %exit

exit:
  ret <4 x float> %next_acc
}

; Test unary operator scalarization.
define void @f15(<4 x float> %init, ptr %base, i32 %count) {
; CHECK-LABEL: define void @f15(
; CHECK-SAME: <4 x float> [[INIT:%.*]], ptr [[BASE:%.*]], i32 [[COUNT:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[COUNT]], %[[ENTRY]] ], [ [[NEXTI:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[NEXTI]] = sub i32 [[I]], 1
; CHECK-NEXT:    [[PTR:%.*]] = getelementptr <4 x float>, ptr [[BASE]], i32 [[I]]
; CHECK-NEXT:    [[VAL_I0:%.*]] = load float, ptr [[PTR]], align 16
; CHECK-NEXT:    [[PTR_I1:%.*]] = getelementptr float, ptr [[PTR]], i32 1
; CHECK-NEXT:    [[VAL_I1:%.*]] = load float, ptr [[PTR_I1]], align 4
; CHECK-NEXT:    [[PTR_I2:%.*]] = getelementptr float, ptr [[PTR]], i32 2
; CHECK-NEXT:    [[VAL_I2:%.*]] = load float, ptr [[PTR_I2]], align 8
; CHECK-NEXT:    [[PTR_I3:%.*]] = getelementptr float, ptr [[PTR]], i32 3
; CHECK-NEXT:    [[VAL_I3:%.*]] = load float, ptr [[PTR_I3]], align 4
; CHECK-NEXT:    [[NEG_I0:%.*]] = fneg float [[VAL_I0]]
; CHECK-NEXT:    [[NEG_I1:%.*]] = fneg float [[VAL_I1]]
; CHECK-NEXT:    [[NEG_I2:%.*]] = fneg float [[VAL_I2]]
; CHECK-NEXT:    [[NEG_I3:%.*]] = fneg float [[VAL_I3]]
; CHECK-NEXT:    [[NEG_UPTO0:%.*]] = insertelement <4 x float> poison, float [[NEG_I0]], i64 0
; CHECK-NEXT:    [[NEG_UPTO1:%.*]] = insertelement <4 x float> [[NEG_UPTO0]], float [[NEG_I1]], i64 1
; CHECK-NEXT:    [[NEG_UPTO2:%.*]] = insertelement <4 x float> [[NEG_UPTO1]], float [[NEG_I2]], i64 2
; CHECK-NEXT:    [[NEG:%.*]] = insertelement <4 x float> [[NEG_UPTO2]], float [[NEG_I3]], i64 3
; CHECK-NEXT:    [[CALL:%.*]] = call <4 x float> @ext(<4 x float> [[NEG]])
; CHECK-NEXT:    [[CALL_I0:%.*]] = extractelement <4 x float> [[CALL]], i64 0
; CHECK-NEXT:    [[CMP_I0:%.*]] = fcmp ogt float [[CALL_I0]], 1.000000e+00
; CHECK-NEXT:    [[CALL_I1:%.*]] = extractelement <4 x float> [[CALL]], i64 1
; CHECK-NEXT:    [[CMP_I1:%.*]] = fcmp ogt float [[CALL_I1]], 2.000000e+00
; CHECK-NEXT:    [[CALL_I2:%.*]] = extractelement <4 x float> [[CALL]], i64 2
; CHECK-NEXT:    [[CMP_I2:%.*]] = fcmp ogt float [[CALL_I2]], 3.000000e+00
; CHECK-NEXT:    [[CALL_I3:%.*]] = extractelement <4 x float> [[CALL]], i64 3
; CHECK-NEXT:    [[CMP_I3:%.*]] = fcmp ogt float [[CALL_I3]], 4.000000e+00
; CHECK-NEXT:    [[SEL_I0:%.*]] = select i1 [[CMP_I0]], float [[CALL_I0]], float 5.000000e+00
; CHECK-NEXT:    [[SEL_I1:%.*]] = select i1 [[CMP_I1]], float [[CALL_I1]], float 6.000000e+00
; CHECK-NEXT:    [[SEL_I2:%.*]] = select i1 [[CMP_I2]], float [[CALL_I2]], float 7.000000e+00
; CHECK-NEXT:    [[SEL_I3:%.*]] = select i1 [[CMP_I3]], float [[CALL_I3]], float 8.000000e+00
; CHECK-NEXT:    store float [[SEL_I0]], ptr [[PTR]], align 16
; CHECK-NEXT:    store float [[SEL_I1]], ptr [[PTR_I1]], align 4
; CHECK-NEXT:    store float [[SEL_I2]], ptr [[PTR_I2]], align 8
; CHECK-NEXT:    store float [[SEL_I3]], ptr [[PTR_I3]], align 4
; CHECK-NEXT:    [[TEST:%.*]] = icmp eq i32 [[NEXTI]], 0
; CHECK-NEXT:    br i1 [[TEST]], label %[[LOOP]], label %[[EXIT:.*]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret void
;
entry:
  br label %loop

loop:
  %i = phi i32 [ %count, %entry ], [ %nexti, %loop ]
  %acc = phi <4 x float> [ %init, %entry ], [ %sel, %loop ]
  %nexti = sub i32 %i, 1

  %ptr = getelementptr <4 x float>, ptr %base, i32 %i
  %val = load <4 x float> , ptr %ptr
  %neg = fneg <4 x float> %val
  %call = call <4 x float> @ext(<4 x float> %neg)
  %cmp = fcmp ogt <4 x float> %call,
  <float 1.0, float 2.0, float 3.0, float 4.0>
  %sel = select <4 x i1> %cmp, <4 x float> %call,
  <4 x float> <float 5.0, float 6.0, float 7.0, float 8.0>
  store <4 x float> %sel, ptr %ptr

  %test = icmp eq i32 %nexti, 0
  br i1 %test, label %loop, label %exit

exit:
  ret void
}

; Check that IR flags are preserved.
define <2 x i32> @f16(<2 x i32> %i, <2 x i32> %j) {
; CHECK-LABEL: define <2 x i32> @f16(
; CHECK-SAME: <2 x i32> [[I:%.*]], <2 x i32> [[J:%.*]]) {
; CHECK-NEXT:    [[I_I0:%.*]] = extractelement <2 x i32> [[I]], i64 0
; CHECK-NEXT:    [[J_I0:%.*]] = extractelement <2 x i32> [[J]], i64 0
; CHECK-NEXT:    [[RES_I0:%.*]] = add nuw nsw i32 [[I_I0]], [[J_I0]]
; CHECK-NEXT:    [[I_I1:%.*]] = extractelement <2 x i32> [[I]], i64 1
; CHECK-NEXT:    [[J_I1:%.*]] = extractelement <2 x i32> [[J]], i64 1
; CHECK-NEXT:    [[RES_I1:%.*]] = add nuw nsw i32 [[I_I1]], [[J_I1]]
; CHECK-NEXT:    [[RES_UPTO0:%.*]] = insertelement <2 x i32> poison, i32 [[RES_I0]], i64 0
; CHECK-NEXT:    [[RES:%.*]] = insertelement <2 x i32> [[RES_UPTO0]], i32 [[RES_I1]], i64 1
; CHECK-NEXT:    ret <2 x i32> [[RES]]
;
  %res = add nuw nsw <2 x i32> %i, %j
  ret <2 x i32> %res
}
define <2 x i32> @f17(<2 x i32> %i, <2 x i32> %j) {
; CHECK-LABEL: define <2 x i32> @f17(
; CHECK-SAME: <2 x i32> [[I:%.*]], <2 x i32> [[J:%.*]]) {
; CHECK-NEXT:    [[I_I0:%.*]] = extractelement <2 x i32> [[I]], i64 0
; CHECK-NEXT:    [[J_I0:%.*]] = extractelement <2 x i32> [[J]], i64 0
; CHECK-NEXT:    [[RES_I0:%.*]] = sdiv exact i32 [[I_I0]], [[J_I0]]
; CHECK-NEXT:    [[I_I1:%.*]] = extractelement <2 x i32> [[I]], i64 1
; CHECK-NEXT:    [[J_I1:%.*]] = extractelement <2 x i32> [[J]], i64 1
; CHECK-NEXT:    [[RES_I1:%.*]] = sdiv exact i32 [[I_I1]], [[J_I1]]
; CHECK-NEXT:    [[RES_UPTO0:%.*]] = insertelement <2 x i32> poison, i32 [[RES_I0]], i64 0
; CHECK-NEXT:    [[RES:%.*]] = insertelement <2 x i32> [[RES_UPTO0]], i32 [[RES_I1]], i64 1
; CHECK-NEXT:    ret <2 x i32> [[RES]]
;
  %res = sdiv exact <2 x i32> %i, %j
  ret <2 x i32> %res
}
define <2 x float> @f18(<2 x float> %x, <2 x float> %y) {
; CHECK-LABEL: define <2 x float> @f18(
; CHECK-SAME: <2 x float> [[X:%.*]], <2 x float> [[Y:%.*]]) {
; CHECK-NEXT:    [[X_I0:%.*]] = extractelement <2 x float> [[X]], i64 0
; CHECK-NEXT:    [[Y_I0:%.*]] = extractelement <2 x float> [[Y]], i64 0
; CHECK-NEXT:    [[RES_I0:%.*]] = fadd fast float [[X_I0]], [[Y_I0]]
; CHECK-NEXT:    [[X_I1:%.*]] = extractelement <2 x float> [[X]], i64 1
; CHECK-NEXT:    [[Y_I1:%.*]] = extractelement <2 x float> [[Y]], i64 1
; CHECK-NEXT:    [[RES_I1:%.*]] = fadd fast float [[X_I1]], [[Y_I1]]
; CHECK-NEXT:    [[RES_UPTO0:%.*]] = insertelement <2 x float> poison, float [[RES_I0]], i64 0
; CHECK-NEXT:    [[RES:%.*]] = insertelement <2 x float> [[RES_UPTO0]], float [[RES_I1]], i64 1
; CHECK-NEXT:    ret <2 x float> [[RES]]
;
  %res = fadd fast <2 x float> %x, %y
  ret <2 x float> %res
}
define <2 x float> @f19(<2 x float> %x) {
; CHECK-LABEL: define <2 x float> @f19(
; CHECK-SAME: <2 x float> [[X:%.*]]) {
; CHECK-NEXT:    [[X_I0:%.*]] = extractelement <2 x float> [[X]], i64 0
; CHECK-NEXT:    [[RES_I0:%.*]] = fneg fast float [[X_I0]]
; CHECK-NEXT:    [[X_I1:%.*]] = extractelement <2 x float> [[X]], i64 1
; CHECK-NEXT:    [[RES_I1:%.*]] = fneg fast float [[X_I1]]
; CHECK-NEXT:    [[RES_UPTO0:%.*]] = insertelement <2 x float> poison, float [[RES_I0]], i64 0
; CHECK-NEXT:    [[RES:%.*]] = insertelement <2 x float> [[RES_UPTO0]], float [[RES_I1]], i64 1
; CHECK-NEXT:    ret <2 x float> [[RES]]
;
  %res = fneg fast <2 x float> %x
  ret <2 x float> %res
}
define <2 x i1> @f20(<2 x float> %x, <2 x float> %y) {
; CHECK-LABEL: define <2 x i1> @f20(
; CHECK-SAME: <2 x float> [[X:%.*]], <2 x float> [[Y:%.*]]) {
; CHECK-NEXT:    [[X_I0:%.*]] = extractelement <2 x float> [[X]], i64 0
; CHECK-NEXT:    [[Y_I0:%.*]] = extractelement <2 x float> [[Y]], i64 0
; CHECK-NEXT:    [[RES_I0:%.*]] = fcmp fast ogt float [[X_I0]], [[Y_I0]]
; CHECK-NEXT:    [[X_I1:%.*]] = extractelement <2 x float> [[X]], i64 1
; CHECK-NEXT:    [[Y_I1:%.*]] = extractelement <2 x float> [[Y]], i64 1
; CHECK-NEXT:    [[RES_I1:%.*]] = fcmp fast ogt float [[X_I1]], [[Y_I1]]
; CHECK-NEXT:    [[RES_UPTO0:%.*]] = insertelement <2 x i1> poison, i1 [[RES_I0]], i64 0
; CHECK-NEXT:    [[RES:%.*]] = insertelement <2 x i1> [[RES_UPTO0]], i1 [[RES_I1]], i64 1
; CHECK-NEXT:    ret <2 x i1> [[RES]]
;
  %res = fcmp fast ogt <2 x float> %x, %y
  ret <2 x i1> %res
}
declare <2 x float> @llvm.sqrt.v2f32(<2 x float>)
define <2 x float> @f21(<2 x float> %x) {
; CHECK-LABEL: define <2 x float> @f21(
; CHECK-SAME: <2 x float> [[X:%.*]]) {
; CHECK-NEXT:    [[X_I0:%.*]] = extractelement <2 x float> [[X]], i64 0
; CHECK-NEXT:    [[RES_I0:%.*]] = call fast float @llvm.sqrt.f32(float [[X_I0]])
; CHECK-NEXT:    [[X_I1:%.*]] = extractelement <2 x float> [[X]], i64 1
; CHECK-NEXT:    [[RES_I1:%.*]] = call fast float @llvm.sqrt.f32(float [[X_I1]])
; CHECK-NEXT:    [[RES_UPTO0:%.*]] = insertelement <2 x float> poison, float [[RES_I0]], i64 0
; CHECK-NEXT:    [[RES:%.*]] = insertelement <2 x float> [[RES_UPTO0]], float [[RES_I1]], i64 1
; CHECK-NEXT:    ret <2 x float> [[RES]]
;
  %res = call fast <2 x float> @llvm.sqrt.v2f32(<2 x float> %x)
  ret <2 x float> %res
}
declare <2 x float> @llvm.fma.v2f32(<2 x float>, <2 x float>, <2 x float>)
define <2 x float> @f22(<2 x float> %x, <2 x float> %y, <2 x float> %z) {
; CHECK-LABEL: define <2 x float> @f22(
; CHECK-SAME: <2 x float> [[X:%.*]], <2 x float> [[Y:%.*]], <2 x float> [[Z:%.*]]) {
; CHECK-NEXT:    [[X_I0:%.*]] = extractelement <2 x float> [[X]], i64 0
; CHECK-NEXT:    [[Y_I0:%.*]] = extractelement <2 x float> [[Y]], i64 0
; CHECK-NEXT:    [[Z_I0:%.*]] = extractelement <2 x float> [[Z]], i64 0
; CHECK-NEXT:    [[RES_I0:%.*]] = call fast float @llvm.fma.f32(float [[X_I0]], float [[Y_I0]], float [[Z_I0]])
; CHECK-NEXT:    [[X_I1:%.*]] = extractelement <2 x float> [[X]], i64 1
; CHECK-NEXT:    [[Y_I1:%.*]] = extractelement <2 x float> [[Y]], i64 1
; CHECK-NEXT:    [[Z_I1:%.*]] = extractelement <2 x float> [[Z]], i64 1
; CHECK-NEXT:    [[RES_I1:%.*]] = call fast float @llvm.fma.f32(float [[X_I1]], float [[Y_I1]], float [[Z_I1]])
; CHECK-NEXT:    [[RES_UPTO0:%.*]] = insertelement <2 x float> poison, float [[RES_I0]], i64 0
; CHECK-NEXT:    [[RES:%.*]] = insertelement <2 x float> [[RES_UPTO0]], float [[RES_I1]], i64 1
; CHECK-NEXT:    ret <2 x float> [[RES]]
;
  %res = call fast <2 x float> @llvm.fma.v2f32(<2 x float> %x, <2 x float> %y, <2 x float> %z)
  ret <2 x float> %res
}

; See https://reviews.llvm.org/D83101#2133062
define <2 x i32> @f23_crash(<2 x i32> %srcvec, i32 %v1) {
; CHECK-LABEL: define <2 x i32> @f23_crash(
; CHECK-SAME: <2 x i32> [[SRCVEC:%.*]], i32 [[V1:%.*]]) {
; CHECK-NEXT:    [[SRCVEC_I0:%.*]] = extractelement <2 x i32> [[SRCVEC]], i64 0
; CHECK-NEXT:    [[T1_UPTO0:%.*]] = insertelement <2 x i32> poison, i32 [[SRCVEC_I0]], i64 0
; CHECK-NEXT:    [[T1:%.*]] = insertelement <2 x i32> [[T1_UPTO0]], i32 [[V1]], i64 1
; CHECK-NEXT:    ret <2 x i32> [[T1]]
;
  %v0 = extractelement <2 x i32> %srcvec, i32 0
  %t0 = insertelement <2 x i32> poison, i32 %v0, i32 0
  %t1 = insertelement <2 x i32> %t0, i32 %v1, i32 1
  ret <2 x i32> %t1
}

!0 = !{ !"root" }
!1 = !{ !"set1", !0 }
!2 = !{ !"set2", !0 }
!3 = !{ !3, !{!"llvm.loop.parallel_accesses", !13} }
!4 = !{ float 4.0 }
!5 = !{ i64 0, i64 8, null }
!13 = distinct !{}
;.
; CHECK: [[SET1_TBAA0]] = !{[[META1:![0-9]+]], [[META1]], i64 0}
; CHECK: [[META1]] = !{!"set1", [[META2:![0-9]+]]}
; CHECK: [[META2]] = !{!"root"}
; CHECK: [[SET2_TBAA3]] = !{[[META4:![0-9]+]], [[META4]], i64 0}
; CHECK: [[META4]] = !{!"set2", [[META2]]}
; CHECK: [[TBAA_STRUCT5]] = !{i64 0, i64 8, null}
; CHECK: [[ACC_GRP6]] = distinct !{}
; CHECK: [[LOOP7]] = distinct !{[[LOOP7]], [[META8:![0-9]+]]}
; CHECK: [[META8]] = !{!"llvm.loop.parallel_accesses", [[ACC_GRP6]]}
; CHECK: [[META9]] = !{float 4.000000e+00}
;.