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

declare void @callee(ptr nocapture readonly %p)

define i32 @simple() {
; CHECK-LABEL: @simple(
; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    ret i32 0
;
  %a = alloca i32
  store i32 0, ptr %a
  call void @callee(ptr %a)
  %l1 = load i32, ptr %a
  ret i32 %l1
}

define i32 @smallbig() {
; CHECK-LABEL: @smallbig(
; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
; CHECK-NEXT:    store i8 0, ptr [[A]], align 1
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    [[L1:%.*]] = load i32, ptr [[A]], align 4
; CHECK-NEXT:    ret i32 [[L1]]
;
  %a = alloca i32
  store i8 0, ptr %a
  call void @callee(ptr %a)
  %l1 = load i32, ptr %a
  ret i32 %l1
}

define i32 @twoalloc() {
; CHECK-LABEL: @twoalloc(
; CHECK-NEXT:    [[A:%.*]] = alloca { i32, i32 }, align 8
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    [[B:%.*]] = getelementptr i32, ptr [[A]], i32 1
; CHECK-NEXT:    store i32 1, ptr [[B]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    [[R:%.*]] = add i32 0, 1
; CHECK-NEXT:    ret i32 [[R]]
;
  %a = alloca {i32, i32}
  store i32 0, ptr %a
  %b = getelementptr i32, ptr %a, i32 1
  store i32 1, ptr %b
  call void @callee(ptr %a)
  %l1 = load i32, ptr %a
  %l2 = load i32, ptr %b
  %r = add i32 %l1, %l2
  ret i32 %r
}

define i32 @twostore() {
; CHECK-LABEL: @twostore(
; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
; CHECK-NEXT:    store i32 1, ptr [[A]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    store i32 2, ptr [[A]], align 4
; CHECK-NEXT:    ret i32 2
;
  %a = alloca i32
  store i32 1, ptr %a
  call void @callee(ptr %a)
  store i32 2, ptr %a
  %l = load i32, ptr %a
  ret i32 %l
}

define float @differenttype() {
; CHECK-LABEL: @differenttype(
; CHECK-NEXT:    [[A:%.*]] = alloca { i32, i32 }, align 8
; CHECK-NEXT:    [[B:%.*]] = getelementptr i32, ptr [[A]], i32 1
; CHECK-NEXT:    store i32 1, ptr [[B]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    [[L2:%.*]] = load float, ptr [[B]], align 4
; CHECK-NEXT:    ret float [[L2]]
;
  %a = alloca {i32, i32}
  %b = getelementptr i32, ptr %a, i32 1
  store i32 1, ptr %b
  call void @callee(ptr %a)
  %l2 = load float, ptr %b
  ret float %l2
}

define i32 @twoalloc_store64(i64 %x) {
; CHECK-LABEL: @twoalloc_store64(
; CHECK-NEXT:    [[A:%.*]] = alloca i64, align 8
; CHECK-NEXT:    store i64 [[X:%.*]], ptr [[A]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    [[L1:%.*]] = load i32, ptr [[A]], align 4
; CHECK-NEXT:    [[B:%.*]] = getelementptr i32, ptr [[A]], i32 1
; CHECK-NEXT:    [[L2:%.*]] = load i32, ptr [[B]], align 4
; CHECK-NEXT:    ret i32 [[L2]]
;
  %a = alloca i64
  store i64 %x, ptr %a
  call void @callee(ptr %a)
  %l1 = load i32, ptr %a
  %b = getelementptr i32, ptr %a, i32 1
  %l2 = load i32, ptr %b
  ret i32 %l2
}

define i32 @twocalls() {
; CHECK-LABEL: @twocalls(
; CHECK-NEXT:    [[A:%.*]] = alloca { i32, i32 }, align 8
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    [[B:%.*]] = getelementptr i32, ptr [[A]], i32 1
; CHECK-NEXT:    store i32 1, ptr [[B]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    [[R:%.*]] = add i32 0, 1
; CHECK-NEXT:    ret i32 [[R]]
;
  %a = alloca {i32, i32}
  store i32 0, ptr %a
  %b = getelementptr i32, ptr %a, i32 1
  store i32 1, ptr %b
  call void @callee(ptr %a)
  %l1 = load i32, ptr %a
  call void @callee(ptr %a)
  %l2 = load i32, ptr %b
  %r = add i32 %l1, %l2
  ret i32 %r
}

define i32 @volatile() {
; CHECK-LABEL: @volatile(
; CHECK-NEXT:    [[A:%.*]] = alloca { i32, i32 }, align 8
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    [[B:%.*]] = getelementptr i32, ptr [[A]], i32 1
; CHECK-NEXT:    store volatile i32 1, ptr [[B]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    [[L1:%.*]] = load volatile i32, ptr [[A]], align 4
; CHECK-NEXT:    [[L2:%.*]] = load i32, ptr [[B]], align 4
; CHECK-NEXT:    [[R:%.*]] = add i32 [[L1]], [[L2]]
; CHECK-NEXT:    ret i32 [[R]]
;
  %a = alloca {i32, i32}
  store i32 0, ptr %a
  %b = getelementptr i32, ptr %a, i32 1
  store volatile i32 1, ptr %b
  call void @callee(ptr %a)
  %l1 = load volatile i32, ptr %a
  %l2 = load i32, ptr %b
  %r = add i32 %l1, %l2
  ret i32 %r
}

define i32 @atomic() {
; CHECK-LABEL: @atomic(
; CHECK-NEXT:    [[A:%.*]] = alloca { i32, i32 }, align 8
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    [[B:%.*]] = getelementptr i32, ptr [[A]], i32 1
; CHECK-NEXT:    store i32 1, ptr [[B]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    [[L1:%.*]] = load atomic i32, ptr [[A]] seq_cst, align 4
; CHECK-NEXT:    [[R:%.*]] = add i32 [[L1]], 1
; CHECK-NEXT:    ret i32 [[R]]
;
  %a = alloca {i32, i32}
  store i32 0, ptr %a
  %b = getelementptr i32, ptr %a, i32 1
  store i32 1, ptr %b
  call void @callee(ptr %a)
  %l1 = load atomic i32, ptr %a seq_cst, align 4
  %l2 = load i32, ptr %b
  %r = add i32 %l1, %l2
  ret i32 %r
}

define i32 @notdominating() {
; CHECK-LABEL: @notdominating(
; CHECK-NEXT:    [[A:%.*]] = alloca { i32, i32 }, align 8
; CHECK-NEXT:    [[B:%.*]] = getelementptr i32, ptr [[A]], i32 1
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    store i32 1, ptr [[B]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    [[R:%.*]] = add i32 undef, undef
; CHECK-NEXT:    ret i32 [[R]]
;
  %a = alloca {i32, i32}
  %b = getelementptr i32, ptr %a, i32 1
  %l1 = load i32, ptr %a
  %l2 = load i32, ptr %b
  store i32 0, ptr %a
  store i32 1, ptr %b
  call void @callee(ptr %a)
  %r = add i32 %l1, %l2
  ret i32 %r
}

declare void @callee_notreadonly(ptr %p)
define i32 @notreadonly() {
; CHECK-LABEL: @notreadonly(
; CHECK-NEXT:    [[A:%.*]] = alloca { i32, i32 }, align 8
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    [[B:%.*]] = getelementptr i32, ptr [[A]], i32 1
; CHECK-NEXT:    store i32 1, ptr [[B]], align 4
; CHECK-NEXT:    call void @callee_notreadonly(ptr [[A]])
; CHECK-NEXT:    [[L1:%.*]] = load i32, ptr [[A]], align 4
; CHECK-NEXT:    [[L2:%.*]] = load i32, ptr [[B]], align 4
; CHECK-NEXT:    [[R:%.*]] = add i32 [[L1]], [[L2]]
; CHECK-NEXT:    ret i32 [[R]]
;
  %a = alloca {i32, i32}
  store i32 0, ptr %a
  %b = getelementptr i32, ptr %a, i32 1
  store i32 1, ptr %b
  call void @callee_notreadonly(ptr %a)
  %l1 = load i32, ptr %a
  %l2 = load i32, ptr %b
  %r = add i32 %l1, %l2
  ret i32 %r
}

declare void @callee_multiuse(ptr nocapture readonly %p, ptr nocapture readonly %q)
define i32 @multiuse() {
; CHECK-LABEL: @multiuse(
; CHECK-NEXT:    [[A:%.*]] = alloca { i32, i32 }, align 8
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    [[B:%.*]] = getelementptr i32, ptr [[A]], i32 1
; CHECK-NEXT:    store i32 1, ptr [[B]], align 4
; CHECK-NEXT:    call void @callee_multiuse(ptr [[A]], ptr [[A]])
; CHECK-NEXT:    [[R:%.*]] = add i32 0, 1
; CHECK-NEXT:    ret i32 [[R]]
;
  %a = alloca {i32, i32}
  store i32 0, ptr %a
  %b = getelementptr i32, ptr %a, i32 1
  store i32 1, ptr %b
  call void @callee_multiuse(ptr %a, ptr %a)
  %l1 = load i32, ptr %a
  %l2 = load i32, ptr %b
  %r = add i32 %l1, %l2
  ret i32 %r
}

define i32 @memcpyed(ptr %src) {
; CHECK-LABEL: @memcpyed(
; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[A]], ptr [[SRC:%.*]], i64 4, i1 false)
; CHECK-NEXT:    [[L1:%.*]] = load i32, ptr [[A]], align 4
; CHECK-NEXT:    ret i32 [[L1]]
;
  %a = alloca i32
  store i32 0, ptr %a
  call void @callee(ptr %a)
  call void @llvm.memcpy.p0.p0.i64(ptr %a, ptr %src, i64 4, i1 false)
  %l1 = load i32, ptr %a
  ret i32 %l1
}

define ptr @memcpyedsplit(ptr %src) {
; CHECK-LABEL: @memcpyedsplit(
; CHECK-NEXT:    [[A:%.*]] = alloca { i64, i64 }, align 8
; CHECK-NEXT:    store i8 1, ptr [[A]], align 1
; CHECK-NEXT:    [[B:%.*]] = getelementptr i64, ptr [[A]], i32 1
; CHECK-NEXT:    store ptr null, ptr [[B]], align 8
; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[A]], ptr [[SRC:%.*]], i64 16, i1 false)
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    [[L1:%.*]] = load ptr, ptr [[B]], align 8
; CHECK-NEXT:    ret ptr [[L1]]
;
  %a = alloca { i64, i64 }
  store i8 1, ptr %a
  %b = getelementptr i64, ptr %a, i32 1
  store ptr null, ptr %b
  call void @llvm.memcpy.p0.p0.i64(ptr %a, ptr %src, i64 16, i1 false)
  call void @callee(ptr %a)
  %l1 = load ptr, ptr %b
  ret ptr %l1
}

; This struct contains padding bits. The load should not be replaced by poison.
%struct.LoadImmediateInfo = type { i32 }
define void @incompletestruct(i1 %b, i1 %c) {
; CHECK-LABEL: @incompletestruct(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[LII:%.*]] = alloca [[STRUCT_LOADIMMEDIATEINFO:%.*]], align 4
; CHECK-NEXT:    call void @llvm.lifetime.start.p0(ptr nonnull [[LII]])
; CHECK-NEXT:    [[BF_CLEAR4:%.*]] = and i32 undef, -262144
; CHECK-NEXT:    [[BF_SET5:%.*]] = select i1 [[B:%.*]], i32 196608, i32 131072
; CHECK-NEXT:    [[BF_SET12:%.*]] = or disjoint i32 [[BF_SET5]], [[BF_CLEAR4]]
; CHECK-NEXT:    store i32 [[BF_SET12]], ptr [[LII]], align 4
; CHECK-NEXT:    call void @callee(ptr [[LII]])
; CHECK-NEXT:    call void @llvm.lifetime.end.p0(ptr nonnull [[LII]])
; CHECK-NEXT:    ret void
;
entry:
  %LII = alloca %struct.LoadImmediateInfo, align 4
  call void @llvm.lifetime.start.p0(ptr nonnull %LII)
  %bf.load = load i32, ptr %LII, align 4
  %bf.clear4 = and i32 %bf.load, -262144
  %bf.set5 = select i1 %b, i32 196608, i32 131072
  %bf.set12 = or disjoint i32 %bf.set5, %bf.clear4
  store i32 %bf.set12, ptr %LII, align 4
  call void @callee(ptr %LII)
  call void @llvm.lifetime.end.p0(ptr nonnull %LII)
  ret void
}

define void @incompletestruct_bb(i1 %b, i1 %c) {
; CHECK-LABEL: @incompletestruct_bb(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[LII:%.*]] = alloca [[STRUCT_LOADIMMEDIATEINFO:%.*]], align 4
; CHECK-NEXT:    br i1 [[C:%.*]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
; CHECK:       if.then:
; CHECK-NEXT:    call void @llvm.lifetime.start.p0(ptr nonnull [[LII]])
; CHECK-NEXT:    [[BF_CLEAR4:%.*]] = and i32 undef, -262144
; CHECK-NEXT:    [[BF_SET5:%.*]] = select i1 [[B:%.*]], i32 196608, i32 131072
; CHECK-NEXT:    [[BF_SET12:%.*]] = or disjoint i32 [[BF_SET5]], [[BF_CLEAR4]]
; CHECK-NEXT:    store i32 [[BF_SET12]], ptr [[LII]], align 4
; CHECK-NEXT:    call void @callee(ptr [[LII]])
; CHECK-NEXT:    call void @llvm.lifetime.end.p0(ptr nonnull [[LII]])
; CHECK-NEXT:    br label [[IF_END]]
; CHECK:       if.end:
; CHECK-NEXT:    ret void
;
entry:
  %LII = alloca %struct.LoadImmediateInfo, align 4
  br i1 %c, label %if.then, label %if.end

if.then:                                          ; preds = %entry
  call void @llvm.lifetime.start.p0(ptr nonnull %LII)
  %bf.load = load i32, ptr %LII, align 4
  %bf.clear4 = and i32 %bf.load, -262144
  %bf.set5 = select i1 %b, i32 196608, i32 131072
  %bf.set12 = or disjoint i32 %bf.set5, %bf.clear4
  store i32 %bf.set12, ptr %LII, align 4
  call void @callee(ptr %LII)
  call void @llvm.lifetime.end.p0(ptr nonnull %LII)
  br label %if.end

if.end:                                           ; preds = %if.then, %entry
  ret void
}

define i32 @sixteenload() {
; CHECK-LABEL: @sixteenload(
; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    [[A1:%.*]] = add i32 0, 0
; CHECK-NEXT:    [[A2:%.*]] = add i32 [[A1]], 0
; CHECK-NEXT:    ret i32 [[A2]]
;
  %a = alloca i32
  store i32 0, ptr %a
  call void @callee(ptr %a)
  %l1 = load i32, ptr %a
  %l2 = load i32, ptr %a
  %l3 = load i32, ptr %a
  %l4 = load i32, ptr %a
  %l5 = load i32, ptr %a
  %l6 = load i32, ptr %a
  %l7 = load i32, ptr %a
  %l8 = load i32, ptr %a
  %l9 = load i32, ptr %a
  %l10 = load i32, ptr %a
  %l11 = load i32, ptr %a
  %l12 = load i32, ptr %a
  %l13 = load i32, ptr %a
  %l14 = load i32, ptr %a
  %l15 = load i32, ptr %a
  %l16 = load i32, ptr %a
  %a1 = add i32 %l1, %l2
  %a2 = add i32 %a1, %l3
  ret i32 %a2
}

define i32 @testcallalloca() {
; CHECK-LABEL: @testcallalloca(
; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    call void [[A]]()
; CHECK-NEXT:    [[L1:%.*]] = load i32, ptr [[A]], align 4
; CHECK-NEXT:    ret i32 [[L1]]
;
  %a = alloca i32
  store i32 0, ptr %a
  call void %a()
  %l1 = load i32, ptr %a
  ret i32 %l1
}

declare void @callee_byval(ptr byval(i32) %p)

define i32 @simple_byval() {
; CHECK-LABEL: @simple_byval(
; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    call void @callee_byval(ptr [[A]])
; CHECK-NEXT:    ret i32 0
;
  %a = alloca i32
  store i32 0, ptr %a
  call void @callee_byval(ptr %a)
  %l1 = load i32, ptr %a
  ret i32 %l1
}

declare void @callee_address_only_capture(ptr readonly captures(address) %p)

define i32 @address_only_capture() {
; CHECK-LABEL: @address_only_capture(
; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    call void @callee_address_only_capture(ptr [[A]])
; CHECK-NEXT:    ret i32 0
;
  %a = alloca i32
  store i32 0, ptr %a
  call void @callee_address_only_capture(ptr %a)
  %l1 = load i32, ptr %a
  ret i32 %l1
}

declare void @callee_read_only_capture(ptr readonly captures(address, read_provenance) %p)

define i32 @read_only_capture() {
; CHECK-LABEL: @read_only_capture(
; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    call void @callee_read_only_capture(ptr [[A]])
; CHECK-NEXT:    ret i32 0
;
  %a = alloca i32
  store i32 0, ptr %a
  call void @callee_read_only_capture(ptr %a)
  %l1 = load i32, ptr %a
  ret i32 %l1
}

declare void @callee_provenance_only_capture(ptr readonly captures(provenance) %p)

; Should not be transformed, as write-provenance is captured.
define i32 @provenance_only_capture() {
; CHECK-LABEL: @provenance_only_capture(
; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    call void @callee_provenance_only_capture(ptr [[A]])
; CHECK-NEXT:    [[L1:%.*]] = load i32, ptr [[A]], align 4
; CHECK-NEXT:    ret i32 [[L1]]
;
  %a = alloca i32
  store i32 0, ptr %a
  call void @callee_provenance_only_capture(ptr %a)
  %l1 = load i32, ptr %a
  ret i32 %l1
}

define i32 @simple_with_lifetimes() {
; CHECK-LABEL: @simple_with_lifetimes(
; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
; CHECK-NEXT:    call void @llvm.lifetime.start.p0(ptr [[A]])
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    call void @llvm.lifetime.end.p0(ptr [[A]])
; CHECK-NEXT:    ret i32 0
;
  %a = alloca i32
  call void @llvm.lifetime.start(ptr %a)
  store i32 0, ptr %a
  call void @callee(ptr %a)
  %l1 = load i32, ptr %a
  call void @llvm.lifetime.end(ptr %a)
  ret i32 %l1
}

define i32 @twoalloc_with_lifetimes() {
; CHECK-LABEL: @twoalloc_with_lifetimes(
; CHECK-NEXT:    [[A:%.*]] = alloca { i32, i32 }, align 8
; CHECK-NEXT:    call void @llvm.lifetime.start.p0(ptr [[A]])
; CHECK-NEXT:    store i32 0, ptr [[A]], align 4
; CHECK-NEXT:    [[B:%.*]] = getelementptr i32, ptr [[A]], i32 1
; CHECK-NEXT:    store i32 1, ptr [[B]], align 4
; CHECK-NEXT:    call void @callee(ptr [[A]])
; CHECK-NEXT:    [[R:%.*]] = add i32 0, 1
; CHECK-NEXT:    call void @llvm.lifetime.end.p0(ptr [[A]])
; CHECK-NEXT:    ret i32 [[R]]
;
  %a = alloca {i32, i32}
  call void @llvm.lifetime.start(ptr %a)
  store i32 0, ptr %a
  %b = getelementptr i32, ptr %a, i32 1
  store i32 1, ptr %b
  call void @callee(ptr %a)
  %l1 = load i32, ptr %a
  %l2 = load i32, ptr %b
  %r = add i32 %l1, %l2
  call void @llvm.lifetime.end(ptr %a)
  ret i32 %r
}

declare void @use.i32(i32)

; We can promote the %i load, even though there is an unknown offset load
; in the loop. It is sufficient that we know all stores.
define void @load_dyn_offset(ptr %ary) {
; CHECK-LABEL: @load_dyn_offset(
; CHECK-NEXT:    [[A:%.*]] = alloca { i64, [4 x i32] }, align 8
; CHECK-NEXT:    store i64 0, ptr [[A]], align 4
; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i8, ptr [[A]], i64 8
; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[GEP]], ptr [[ARY:%.*]], i64 16, i1 false)
; CHECK-NEXT:    br label [[LOOP:%.*]]
; CHECK:       loop:
; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[LOOP]] ], [ 0, [[TMP0:%.*]] ]
; CHECK-NEXT:    [[I_NEXT]] = add i64 [[I]], 1
; CHECK-NEXT:    store i64 [[I_NEXT]], ptr [[A]], align 4
; CHECK-NEXT:    [[GEP_I:%.*]] = getelementptr i32, ptr [[GEP]], i64 [[I]]
; CHECK-NEXT:    [[VAL:%.*]] = load i32, ptr [[GEP_I]], align 4
; CHECK-NEXT:    call void @use.i32(i32 [[VAL]])
; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i64 [[I_NEXT]], 6
; CHECK-NEXT:    br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK:       exit:
; CHECK-NEXT:    ret void
;
  %a = alloca {i64, [4 x i32]}
  store i64 0, ptr %a
  %gep = getelementptr i8, ptr %a, i64 8
  call void @llvm.memcpy(ptr %gep, ptr %ary, i64 16, i1 false)
  br label %loop

loop:
  %i = load i64, ptr %a
  %i.next = add i64 %i, 1
  store i64 %i.next, ptr %a
  %gep.i = getelementptr i32, ptr %gep, i64 %i
  %val = load i32, ptr %gep.i
  call void @use.i32(i32 %val)
  %cmp = icmp eq i64 %i.next, 6
  br i1 %cmp, label %exit, label %loop

exit:
  ret void
}

; Same as previous test, but with an unknown-offset store. We can't promote in
; that case.
define void @store_dyn_offset(ptr %ary) {
; CHECK-LABEL: @store_dyn_offset(
; CHECK-NEXT:    [[A:%.*]] = alloca { i64, [4 x i32] }, align 8
; CHECK-NEXT:    store i64 0, ptr [[A]], align 4
; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i8, ptr [[A]], i64 8
; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[GEP]], ptr [[ARY:%.*]], i64 16, i1 false)
; CHECK-NEXT:    br label [[LOOP:%.*]]
; CHECK:       loop:
; CHECK-NEXT:    [[I:%.*]] = load i64, ptr [[A]], align 4
; CHECK-NEXT:    [[I_NEXT:%.*]] = add i64 [[I]], 1
; CHECK-NEXT:    store i64 [[I_NEXT]], ptr [[A]], align 4
; CHECK-NEXT:    [[GEP_I:%.*]] = getelementptr i32, ptr [[GEP]], i64 [[I]]
; CHECK-NEXT:    [[I_TRUNC:%.*]] = trunc i64 [[I]] to i32
; CHECK-NEXT:    store i32 [[I_TRUNC]], ptr [[GEP_I]], align 4
; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i64 [[I_NEXT]], 6
; CHECK-NEXT:    br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK:       exit:
; CHECK-NEXT:    ret void
;
  %a = alloca {i64, [4 x i32]}
  store i64 0, ptr %a
  %gep = getelementptr i8, ptr %a, i64 8
  call void @llvm.memcpy(ptr %gep, ptr %ary, i64 16, i1 false)
  br label %loop

loop:
  %i = load i64, ptr %a
  %i.next = add i64 %i, 1
  store i64 %i.next, ptr %a
  %gep.i = getelementptr i32, ptr %gep, i64 %i
  %i.trunc = trunc i64 %i to i32
  store i32 %i.trunc, ptr %gep.i
  %cmp = icmp eq i64 %i.next, 6
  br i1 %cmp, label %exit, label %loop

exit:
  ret void
}

declare void @llvm.memcpy.p0.p0.i64(ptr, ptr, i64, i1)