1; RUN: llc -mtriple=thumbv6m-eabi -verify-machineinstrs %s -o - | \
2; RUN:    FileCheck %s --check-prefix=CHECK --check-prefix=CHECKV6
3; RUN: llc -mtriple=thumbv6m-eabi -O=0 -verify-machineinstrs %s -o - | \
4; RUN:    FileCheck %s --check-prefix=CHECK --check-prefix=CHECKV6
5; RUN: llc -mtriple=thumbv7a-eabi -mattr=-neon -verify-machineinstrs %s -o - | \
6; RUN:    FileCheck %s --check-prefix=CHECK --check-prefix=CHECKV7
7; RUN: llc -mtriple=armv7a-eabi -mattr=-neon -verify-machineinstrs %s -o - | \
8; RUN:    FileCheck %s --check-prefix=CHECK --check-prefix=CHECKV7
9
10@d = external global [64 x i32]
11@s = external global [64 x i32]
12
13; Function Attrs: nounwind
14define void @t1() #0 {
15entry:
16; CHECK-LABEL: t1:
17; CHECKV6: ldr [[LB:r[0-7]]],
18; CHECKV6-NEXT: ldr [[SB:r[0-7]]],
19; We use '[rl0-9]+' to allow 'r0'..'r12', 'lr'
20; CHECKV7: movt [[LB:[rl0-9]+]], :upper16:d
21; CHECKV7-NEXT: movt [[SB:[rl0-9]+]], :upper16:s
22; CHECK-NEXT: ldm{{(\.w)?}} [[LB]]!,
23; CHECK-NEXT: stm{{(\.w)?}} [[SB]]!,
24; Think of the monstrosity '{{\[}}[[LB]]]' as '[ [[LB]] ]' without the spaces.
25; CHECK-NEXT: ldrb{{(\.w)?}} {{.*}}, {{\[}}[[LB]]]
26; CHECK-NEXT: strb{{(\.w)?}} {{.*}}, {{\[}}[[SB]]]
27    tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 bitcast ([64 x i32]* @s to i8*), i8* align 4 bitcast ([64 x i32]* @d to i8*), i32 17, i1 false)
28    ret void
29}
30
31; Function Attrs: nounwind
32define void @t2() #0 {
33entry:
34; CHECK-LABEL: t2:
35; CHECKV6: ldr [[LB:r[0-7]]],
36; CHECKV6-NEXT: ldr [[SB:r[0-7]]],
37; CHECKV7: movt [[LB:[rl0-9]+]], :upper16:d
38; CHECKV7-NEXT: movt [[SB:[rl0-9]+]], :upper16:s
39; CHECK-NEXT: ldm{{(\.w)?}} [[LB]]!,
40; CHECK-NEXT: stm{{(\.w)?}} [[SB]]!,
41; CHECK-NEXT: ldrh{{(\.w)?}} {{.*}}, {{\[}}[[LB]]]
42; CHECK-NEXT: ldrb{{(\.w)?}} {{.*}}, {{\[}}[[LB]], #2]
43; CHECK-NEXT: strb{{(\.w)?}} {{.*}}, {{\[}}[[SB]], #2]
44; CHECK-NEXT: strh{{(\.w)?}} {{.*}}, {{\[}}[[SB]]]
45    tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 bitcast ([64 x i32]* @s to i8*), i8* align 4 bitcast ([64 x i32]* @d to i8*), i32 15, i1 false)
46    ret void
47}
48
49; PR23768
50%struct.T = type { i8, i64, i8 }
51
52@copy = external global %struct.T, align 8
53@etest = external global %struct.T, align 8
54
55define void @t3() {
56  call void @llvm.memcpy.p0i8.p0i8.i32(
57     i8* align 8 getelementptr inbounds (%struct.T, %struct.T* @copy, i32 0, i32 0),
58     i8* align 8 getelementptr inbounds (%struct.T, %struct.T* @etest, i32 0, i32 0),
59     i32 24, i1 false)
60  call void @llvm.memcpy.p0i8.p0i8.i32(
61     i8* align 8 getelementptr inbounds (%struct.T, %struct.T* @copy, i32 0, i32 0),
62     i8* align 8 getelementptr inbounds (%struct.T, %struct.T* @etest, i32 0, i32 0),
63     i32 24, i1 false)
64  ret void
65}
66
67%struct.S = type { [12 x i32] }
68
69; CHECK-LABEL: test3
70define void @test3(%struct.S* %d, %struct.S* %s) #0 {
71  %1 = bitcast %struct.S* %d to i8*
72  %2 = bitcast %struct.S* %s to i8*
73  tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %1, i8* align 4 %2, i32 48, i1 false)
74; 3 ldm/stm pairs in v6; 2 in v7
75; CHECK: ldm{{(\.w)?}} {{[rl0-9]+!?}}, [[REGLIST1:{.*}]]
76; CHECK: stm{{(\.w)?}} {{[rl0-9]+!?}}, [[REGLIST1]]
77; CHECK: ldm{{(\.w)?}} {{[rl0-9]+!?}}, [[REGLIST2:{.*}]]
78; CHECK: stm{{(\.w)?}} {{[rl0-9]+!?}}, [[REGLIST2]]
79; CHECKV6: ldm {{r[0-7]!?}}, [[REGLIST3:{.*}]]
80; CHECKV6: stm {{r[0-7]!?}}, [[REGLIST3]]
81; CHECKV7-NOT: ldm
82; CHECKV7-NOT: stm
83  %arrayidx = getelementptr inbounds %struct.S, %struct.S* %s, i32 0, i32 0, i32 1
84  tail call void @g(i32* %arrayidx) #3
85  ret void
86}
87
88declare void @g(i32*)
89
90; Set "no-frame-pointer-elim" to increase register pressure
91attributes #0 = { "no-frame-pointer-elim"="true" }
92
93; Function Attrs: nounwind
94declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture readonly, i32, i1) #1
95