1; RUN: llc -verify-machineinstrs -O3 -mcpu=pwr8 \
2; RUN:   -mtriple=powerpc64le-unknown-linux-gnu < %s | FileCheck %s
3
4; RUN: llc -verify-machineinstrs -O3 -mcpu=pwr8 -disable-ppc-vsx-swap-removal \
5; RUN:   -mtriple=powerpc64le-unknown-linux-gnu < %s | FileCheck \
6; RUN:   -check-prefix=NOOPTSWAP %s
7
8; RUN: llc -O3 -mcpu=pwr9 -mtriple=powerpc64le-unknown-linux-gnu \
9; RUN:  -verify-machineinstrs -ppc-vsr-nums-as-vr < %s | FileCheck \
10; RUN:  -check-prefix=CHECK-P9 --implicit-check-not xxswapd %s
11
12; RUN: llc -O3 -mcpu=pwr9 -disable-ppc-vsx-swap-removal -mattr=-power9-vector \
13; RUN:  -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu < %s \
14; RUN:  | FileCheck -check-prefix=NOOPTSWAP %s
15
16; This test was generated from the following source:
17;
18; #define N 4096
19; int ca[N] __attribute__((aligned(16)));
20; int cb[N] __attribute__((aligned(16)));
21; int cc[N] __attribute__((aligned(16)));
22; int cd[N] __attribute__((aligned(16)));
23;
24; void foo ()
25; {
26;   int i;
27;   for (i = 0; i < N; i++) {
28;     ca[i] = (cb[i] + cc[i]) * cd[i];
29;   }
30; }
31
32@cb = common global [4096 x i32] zeroinitializer, align 16
33@cc = common global [4096 x i32] zeroinitializer, align 16
34@cd = common global [4096 x i32] zeroinitializer, align 16
35@ca = common global [4096 x i32] zeroinitializer, align 16
36
37define void @foo() {
38entry:
39  br label %vector.body
40
41vector.body:
42  %index = phi i64 [ 0, %entry ], [ %index.next.3, %vector.body ]
43  %0 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cb, i64 0, i64 %index
44  %1 = bitcast i32* %0 to <4 x i32>*
45  %wide.load = load <4 x i32>, <4 x i32>* %1, align 16
46  %2 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cc, i64 0, i64 %index
47  %3 = bitcast i32* %2 to <4 x i32>*
48  %wide.load13 = load <4 x i32>, <4 x i32>* %3, align 16
49  %4 = add nsw <4 x i32> %wide.load13, %wide.load
50  %5 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cd, i64 0, i64 %index
51  %6 = bitcast i32* %5 to <4 x i32>*
52  %wide.load14 = load <4 x i32>, <4 x i32>* %6, align 16
53  %7 = mul nsw <4 x i32> %4, %wide.load14
54  %8 = getelementptr inbounds [4096 x i32], [4096 x i32]* @ca, i64 0, i64 %index
55  %9 = bitcast i32* %8 to <4 x i32>*
56  store <4 x i32> %7, <4 x i32>* %9, align 16
57  %index.next = add nuw nsw i64 %index, 4
58  %10 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cb, i64 0, i64 %index.next
59  %11 = bitcast i32* %10 to <4 x i32>*
60  %wide.load.1 = load <4 x i32>, <4 x i32>* %11, align 16
61  %12 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cc, i64 0, i64 %index.next
62  %13 = bitcast i32* %12 to <4 x i32>*
63  %wide.load13.1 = load <4 x i32>, <4 x i32>* %13, align 16
64  %14 = add nsw <4 x i32> %wide.load13.1, %wide.load.1
65  %15 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cd, i64 0, i64 %index.next
66  %16 = bitcast i32* %15 to <4 x i32>*
67  %wide.load14.1 = load <4 x i32>, <4 x i32>* %16, align 16
68  %17 = mul nsw <4 x i32> %14, %wide.load14.1
69  %18 = getelementptr inbounds [4096 x i32], [4096 x i32]* @ca, i64 0, i64 %index.next
70  %19 = bitcast i32* %18 to <4 x i32>*
71  store <4 x i32> %17, <4 x i32>* %19, align 16
72  %index.next.1 = add nuw nsw i64 %index.next, 4
73  %20 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cb, i64 0, i64 %index.next.1
74  %21 = bitcast i32* %20 to <4 x i32>*
75  %wide.load.2 = load <4 x i32>, <4 x i32>* %21, align 16
76  %22 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cc, i64 0, i64 %index.next.1
77  %23 = bitcast i32* %22 to <4 x i32>*
78  %wide.load13.2 = load <4 x i32>, <4 x i32>* %23, align 16
79  %24 = add nsw <4 x i32> %wide.load13.2, %wide.load.2
80  %25 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cd, i64 0, i64 %index.next.1
81  %26 = bitcast i32* %25 to <4 x i32>*
82  %wide.load14.2 = load <4 x i32>, <4 x i32>* %26, align 16
83  %27 = mul nsw <4 x i32> %24, %wide.load14.2
84  %28 = getelementptr inbounds [4096 x i32], [4096 x i32]* @ca, i64 0, i64 %index.next.1
85  %29 = bitcast i32* %28 to <4 x i32>*
86  store <4 x i32> %27, <4 x i32>* %29, align 16
87  %index.next.2 = add nuw nsw i64 %index.next.1, 4
88  %30 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cb, i64 0, i64 %index.next.2
89  %31 = bitcast i32* %30 to <4 x i32>*
90  %wide.load.3 = load <4 x i32>, <4 x i32>* %31, align 16
91  %32 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cc, i64 0, i64 %index.next.2
92  %33 = bitcast i32* %32 to <4 x i32>*
93  %wide.load13.3 = load <4 x i32>, <4 x i32>* %33, align 16
94  %34 = add nsw <4 x i32> %wide.load13.3, %wide.load.3
95  %35 = getelementptr inbounds [4096 x i32], [4096 x i32]* @cd, i64 0, i64 %index.next.2
96  %36 = bitcast i32* %35 to <4 x i32>*
97  %wide.load14.3 = load <4 x i32>, <4 x i32>* %36, align 16
98  %37 = mul nsw <4 x i32> %34, %wide.load14.3
99  %38 = getelementptr inbounds [4096 x i32], [4096 x i32]* @ca, i64 0, i64 %index.next.2
100  %39 = bitcast i32* %38 to <4 x i32>*
101  store <4 x i32> %37, <4 x i32>* %39, align 16
102  %index.next.3 = add nuw nsw i64 %index.next.2, 4
103  %40 = icmp eq i64 %index.next.3, 4096
104  br i1 %40, label %for.end, label %vector.body
105
106for.end:
107  ret void
108}
109
110; CHECK-LABEL: @foo
111; CHECK-NOT: xxpermdi
112; CHECK-NOT: xxswapd
113; CHECK-P9-NOT: xxpermdi
114
115; CHECK: lxvd2x
116; CHECK: lxvd2x
117; CHECK-DAG: lxvd2x
118; CHECK-DAG: vadduwm
119; CHECK: vmuluwm
120; CHECK: stxvd2x
121
122; CHECK: lxvd2x
123; CHECK: lxvd2x
124; CHECK-DAG: lxvd2x
125; CHECK-DAG: vadduwm
126; CHECK: vmuluwm
127; CHECK: stxvd2x
128
129; CHECK: lxvd2x
130; CHECK: lxvd2x
131; CHECK-DAG: lxvd2x
132; CHECK-DAG: vadduwm
133; CHECK: vmuluwm
134; CHECK: stxvd2x
135
136; CHECK: lxvd2x
137; CHECK: lxvd2x
138; CHECK-DAG: lxvd2x
139; CHECK-DAG: vadduwm
140; CHECK: vmuluwm
141; CHECK: stxvd2x
142
143; NOOPTSWAP-LABEL: @foo
144
145; NOOPTSWAP: lxvd2x
146; NOOPTSWAP-DAG: lxvd2x
147; NOOPTSWAP-DAG: lxvd2x
148; NOOPTSWAP-DAG: xxswapd
149; NOOPTSWAP-DAG: xxswapd
150; NOOPTSWAP-DAG: xxswapd
151; NOOPTSWAP-DAG: vadduwm
152; NOOPTSWAP: vmuluwm
153; NOOPTSWAP: xxswapd
154; NOOPTSWAP-DAG: xxswapd
155; NOOPTSWAP-DAG: xxswapd
156; NOOPTSWAP-DAG: stxvd2x
157; NOOPTSWAP-DAG: stxvd2x
158; NOOPTSWAP: stxvd2x
159
160; CHECK-P9-LABEL: @foo
161; CHECK-P9-DAG: lxvx
162; CHECK-P9-DAG: lxvx
163; CHECK-P9-DAG: lxvx
164; CHECK-P9-DAG: lxvx
165; CHECK-P9-DAG: lxvx
166; CHECK-P9-DAG: lxvx
167; CHECK-P9-DAG: lxvx
168; CHECK-P9-DAG: lxvx
169; CHECK-P9-DAG: lxvx
170; CHECK-P9-DAG: lxvx
171; CHECK-P9-DAG: lxvx
172; CHECK-P9-DAG: lxvx
173; CHECK-P9-DAG: vadduwm
174; CHECK-P9-DAG: vadduwm
175; CHECK-P9-DAG: vadduwm
176; CHECK-P9-DAG: vadduwm
177; CHECK-P9-DAG: vmuluwm
178; CHECK-P9-DAG: vmuluwm
179; CHECK-P9-DAG: vmuluwm
180; CHECK-P9-DAG: vmuluwm
181; CHECK-P9-DAG: stxvx
182; CHECK-P9-DAG: stxvx
183; CHECK-P9-DAG: stxvx
184; CHECK-P9-DAG: stxvx
185
186