1; Test vector insertions of byte-swapped memory values into 0.
2;
3; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z15 | FileCheck %s
4
5declare i16 @llvm.bswap.i16(i16)
6declare i32 @llvm.bswap.i32(i32)
7declare i64 @llvm.bswap.i64(i64)
8declare <8 x i16> @llvm.bswap.v8i16(<8 x i16>)
9declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>)
10declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>)
11
12; Test VLLEBRZH.
13define <8 x i16> @f1(i16 *%ptr) {
14; CHECK-LABEL: f1:
15; CHECK: vllebrzh %v24, 0(%r2)
16; CHECK: br %r14
17  %val = load i16, i16 *%ptr
18  %swap = call i16 @llvm.bswap.i16(i16 %val)
19  %ret = insertelement <8 x i16> zeroinitializer, i16 %swap, i32 3
20  ret <8 x i16> %ret
21}
22
23; Test VLLEBRZH using a vector bswap.
24define <8 x i16> @f2(i16 *%ptr) {
25; CHECK-LABEL: f2:
26; CHECK: vllebrzh %v24, 0(%r2)
27; CHECK: br %r14
28  %val = load i16, i16 *%ptr
29  %insert = insertelement <8 x i16> zeroinitializer, i16 %val, i32 3
30  %ret = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %insert)
31  ret <8 x i16> %ret
32}
33
34; Test VLLEBRZF.
35define <4 x i32> @f3(i32 *%ptr) {
36; CHECK-LABEL: f3:
37; CHECK: vllebrzf %v24, 0(%r2)
38; CHECK: br %r14
39  %val = load i32, i32 *%ptr
40  %swap = call i32 @llvm.bswap.i32(i32 %val)
41  %ret = insertelement <4 x i32> zeroinitializer, i32 %swap, i32 1
42  ret <4 x i32> %ret
43}
44
45; Test VLLEBRZF using a vector bswap.
46define <4 x i32> @f4(i32 *%ptr) {
47; CHECK-LABEL: f4:
48; CHECK: vllebrzf %v24, 0(%r2)
49; CHECK: br %r14
50  %val = load i32, i32 *%ptr
51  %insert = insertelement <4 x i32> zeroinitializer, i32 %val, i32 1
52  %ret = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %insert)
53  ret <4 x i32> %ret
54}
55
56; Test VLLEBRZG.
57define <2 x i64> @f5(i64 *%ptr) {
58; CHECK-LABEL: f5:
59; CHECK: vllebrzg %v24, 0(%r2)
60; CHECK: br %r14
61  %val = load i64, i64 *%ptr
62  %swap = call i64 @llvm.bswap.i64(i64 %val)
63  %ret = insertelement <2 x i64> zeroinitializer, i64 %swap, i32 0
64  ret <2 x i64> %ret
65}
66
67; Test VLLEBRZG using a vector bswap.
68define <2 x i64> @f6(i64 *%ptr) {
69; CHECK-LABEL: f6:
70; CHECK: vllebrzg %v24, 0(%r2)
71; CHECK: br %r14
72  %val = load i64, i64 *%ptr
73  %insert = insertelement <2 x i64> zeroinitializer, i64 %val, i32 0
74  %ret = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %insert)
75  ret <2 x i64> %ret
76}
77
78; Test VLLEBRZE.
79define <4 x i32> @f7(i32 *%ptr) {
80; CHECK-LABEL: f7:
81; CHECK: vllebrze %v24, 0(%r2)
82; CHECK: br %r14
83  %val = load i32, i32 *%ptr
84  %swap = call i32 @llvm.bswap.i32(i32 %val)
85  %ret = insertelement <4 x i32> zeroinitializer, i32 %swap, i32 0
86  ret <4 x i32> %ret
87}
88
89; Test VLLEBRZE using a vector bswap.
90define <4 x i32> @f8(i32 *%ptr) {
91; CHECK-LABEL: f8:
92; CHECK: vllebrze %v24, 0(%r2)
93; CHECK: br %r14
94  %val = load i32, i32 *%ptr
95  %insert = insertelement <4 x i32> zeroinitializer, i32 %val, i32 0
96  %ret = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %insert)
97  ret <4 x i32> %ret
98}
99
100; Test VLLEBRZH with the highest in-range offset.
101define <8 x i16> @f9(i16 *%base) {
102; CHECK-LABEL: f9:
103; CHECK: vllebrzh %v24, 4094(%r2)
104; CHECK: br %r14
105  %ptr = getelementptr i16, i16 *%base, i64 2047
106  %val = load i16, i16 *%ptr
107  %swap = call i16 @llvm.bswap.i16(i16 %val)
108  %ret = insertelement <8 x i16> zeroinitializer, i16 %swap, i32 3
109  ret <8 x i16> %ret
110}
111
112; Test VLLEBRZH with the next highest offset.
113define <8 x i16> @f10(i16 *%base) {
114; CHECK-LABEL: f10:
115; CHECK-NOT: vllebrzh %v24, 4096(%r2)
116; CHECK: br %r14
117  %ptr = getelementptr i16, i16 *%base, i64 2048
118  %val = load i16, i16 *%ptr
119  %swap = call i16 @llvm.bswap.i16(i16 %val)
120  %ret = insertelement <8 x i16> zeroinitializer, i16 %swap, i32 3
121  ret <8 x i16> %ret
122}
123
124; Test that VLLEBRZH allows an index.
125define <8 x i16> @f11(i16 *%base, i64 %index) {
126; CHECK-LABEL: f11:
127; CHECK: sllg [[REG:%r[1-5]]], %r3, 1
128; CHECK: vllebrzh %v24, 0([[REG]],%r2)
129; CHECK: br %r14
130  %ptr = getelementptr i16, i16 *%base, i64 %index
131  %val = load i16, i16 *%ptr
132  %swap = call i16 @llvm.bswap.i16(i16 %val)
133  %ret = insertelement <8 x i16> zeroinitializer, i16 %swap, i32 3
134  ret <8 x i16> %ret
135}
136
137