1; RUN: llc < %s -march=x86 -mattr=+sse4.2 | FileCheck %s
2; RUN: llc < %s -march=x86 -mattr=+sse4.2 -x86-experimental-vector-widening-legalization | FileCheck %s --check-prefix=CHECK-WIDE
3
4define void @update(i64* %dst_i, i64* %src_i, i32 %n) nounwind {
5; CHECK-LABEL: update:
6; CHECK-WIDE-LABEL: update:
7entry:
8	%dst_i.addr = alloca i64*		; <i64**> [#uses=2]
9	%src_i.addr = alloca i64*		; <i64**> [#uses=2]
10	%n.addr = alloca i32		; <i32*> [#uses=2]
11	%i = alloca i32, align 4		; <i32*> [#uses=8]
12	%dst = alloca <8 x i8>*, align 4		; <<8 x i8>**> [#uses=2]
13	%src = alloca <8 x i8>*, align 4		; <<8 x i8>**> [#uses=2]
14	store i64* %dst_i, i64** %dst_i.addr
15	store i64* %src_i, i64** %src_i.addr
16	store i32 %n, i32* %n.addr
17	store i32 0, i32* %i
18	br label %forcond
19
20forcond:		; preds = %forinc, %entry
21	%tmp = load i32, i32* %i		; <i32> [#uses=1]
22	%tmp1 = load i32, i32* %n.addr		; <i32> [#uses=1]
23	%cmp = icmp slt i32 %tmp, %tmp1		; <i1> [#uses=1]
24	br i1 %cmp, label %forbody, label %afterfor
25
26forbody:		; preds = %forcond
27	%tmp2 = load i32, i32* %i		; <i32> [#uses=1]
28	%tmp3 = load i64*, i64** %dst_i.addr		; <i64*> [#uses=1]
29	%arrayidx = getelementptr i64, i64* %tmp3, i32 %tmp2		; <i64*> [#uses=1]
30	%conv = bitcast i64* %arrayidx to <8 x i8>*		; <<8 x i8>*> [#uses=1]
31	store <8 x i8>* %conv, <8 x i8>** %dst
32	%tmp4 = load i32, i32* %i		; <i32> [#uses=1]
33	%tmp5 = load i64*, i64** %src_i.addr		; <i64*> [#uses=1]
34	%arrayidx6 = getelementptr i64, i64* %tmp5, i32 %tmp4		; <i64*> [#uses=1]
35	%conv7 = bitcast i64* %arrayidx6 to <8 x i8>*		; <<8 x i8>*> [#uses=1]
36	store <8 x i8>* %conv7, <8 x i8>** %src
37	%tmp8 = load i32, i32* %i		; <i32> [#uses=1]
38	%tmp9 = load <8 x i8>*, <8 x i8>** %dst		; <<8 x i8>*> [#uses=1]
39	%arrayidx10 = getelementptr <8 x i8>, <8 x i8>* %tmp9, i32 %tmp8		; <<8 x i8>*> [#uses=1]
40	%tmp11 = load i32, i32* %i		; <i32> [#uses=1]
41	%tmp12 = load <8 x i8>*, <8 x i8>** %src		; <<8 x i8>*> [#uses=1]
42	%arrayidx13 = getelementptr <8 x i8>, <8 x i8>* %tmp12, i32 %tmp11		; <<8 x i8>*> [#uses=1]
43	%tmp14 = load <8 x i8>, <8 x i8>* %arrayidx13		; <<8 x i8>> [#uses=1]
44	%add = add <8 x i8> %tmp14, < i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1 >		; <<8 x i8>> [#uses=1]
45	%shr = ashr <8 x i8> %add, < i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2 >		; <<8 x i8>> [#uses=1]
46	store <8 x i8> %shr, <8 x i8>* %arrayidx10
47	br label %forinc
48; CHECK: %forbody
49; CHECK:      pmovzxbw
50; CHECK-NEXT: paddw
51; CHECK-NEXT: psllw $8
52; CHECK-NEXT: psraw $8
53; CHECK-NEXT: psraw $2
54; CHECK-NEXT: pshufb
55; CHECK-NEXT: movq
56;
57; FIXME: We shouldn't require both a movd and an insert.
58; CHECK-WIDE: %forbody
59; CHECK-WIDE:      movd
60; CHECK-WIDE-NEXT: pinsrd
61; CHECK-WIDE-NEXT: paddb
62; CHECK-WIDE-NEXT: psrlw $2
63; CHECK-WIDE-NEXT: pand
64; CHECK-WIDE-NEXT: pxor
65; CHECK-WIDE-NEXT: psubb
66; CHECK-WIDE-NEXT: pextrd
67; CHECK-WIDE-NEXT: movd
68
69forinc:		; preds = %forbody
70	%tmp15 = load i32, i32* %i		; <i32> [#uses=1]
71	%inc = add i32 %tmp15, 1		; <i32> [#uses=1]
72	store i32 %inc, i32* %i
73	br label %forcond
74
75afterfor:		; preds = %forcond
76	ret void
77}
78
79