xref: /llvm-project/llvm/test/CodeGen/SystemZ/vec-bswap-05.ll (revision 0a76f7d9d8c1fc693568ed26420c47d92a6ba0e7) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
2; Test vector insertions of byte-swapped memory values into 0.
3;
4; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z15 | FileCheck %s
5
6declare i16 @llvm.bswap.i16(i16)
7declare i32 @llvm.bswap.i32(i32)
8declare i64 @llvm.bswap.i64(i64)
9declare <8 x i16> @llvm.bswap.v8i16(<8 x i16>)
10declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>)
11declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>)
12
13; Test VLLEBRZH.
14define <8 x i16> @f1(ptr %ptr) {
15; CHECK-LABEL: f1:
16; CHECK:       # %bb.0:
17; CHECK-NEXT:    vllebrzh %v24, 0(%r2)
18; CHECK-NEXT:    br %r14
19  %val = load i16, ptr %ptr
20  %swap = call i16 @llvm.bswap.i16(i16 %val)
21  %ret = insertelement <8 x i16> zeroinitializer, i16 %swap, i32 3
22  ret <8 x i16> %ret
23}
24
25; Test VLLEBRZH using a vector bswap.
26define <8 x i16> @f2(ptr %ptr) {
27; CHECK-LABEL: f2:
28; CHECK:       # %bb.0:
29; CHECK-NEXT:    vllebrzh %v24, 0(%r2)
30; CHECK-NEXT:    br %r14
31  %val = load i16, ptr %ptr
32  %insert = insertelement <8 x i16> zeroinitializer, i16 %val, i32 3
33  %ret = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %insert)
34  ret <8 x i16> %ret
35}
36
37; Test VLLEBRZF.
38define <4 x i32> @f3(ptr %ptr) {
39; CHECK-LABEL: f3:
40; CHECK:       # %bb.0:
41; CHECK-NEXT:    vllebrzf %v24, 0(%r2)
42; CHECK-NEXT:    br %r14
43  %val = load i32, ptr %ptr
44  %swap = call i32 @llvm.bswap.i32(i32 %val)
45  %ret = insertelement <4 x i32> zeroinitializer, i32 %swap, i32 1
46  ret <4 x i32> %ret
47}
48
49; Test VLLEBRZF using a vector bswap.
50define <4 x i32> @f4(ptr %ptr) {
51; CHECK-LABEL: f4:
52; CHECK:       # %bb.0:
53; CHECK-NEXT:    vllebrzf %v24, 0(%r2)
54; CHECK-NEXT:    br %r14
55  %val = load i32, ptr %ptr
56  %insert = insertelement <4 x i32> zeroinitializer, i32 %val, i32 1
57  %ret = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %insert)
58  ret <4 x i32> %ret
59}
60
61; Test VLLEBRZG.
62define <2 x i64> @f5(ptr %ptr) {
63; CHECK-LABEL: f5:
64; CHECK:       # %bb.0:
65; CHECK-NEXT:    vllebrzg %v24, 0(%r2)
66; CHECK-NEXT:    br %r14
67  %val = load i64, ptr %ptr
68  %swap = call i64 @llvm.bswap.i64(i64 %val)
69  %ret = insertelement <2 x i64> zeroinitializer, i64 %swap, i32 0
70  ret <2 x i64> %ret
71}
72
73; Test VLLEBRZG using a vector bswap.
74define <2 x i64> @f6(ptr %ptr) {
75; CHECK-LABEL: f6:
76; CHECK:       # %bb.0:
77; CHECK-NEXT:    vllebrzg %v24, 0(%r2)
78; CHECK-NEXT:    br %r14
79  %val = load i64, ptr %ptr
80  %insert = insertelement <2 x i64> zeroinitializer, i64 %val, i32 0
81  %ret = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %insert)
82  ret <2 x i64> %ret
83}
84
85; Test VLLEBRZE.
86define <4 x i32> @f7(ptr %ptr) {
87; CHECK-LABEL: f7:
88; CHECK:       # %bb.0:
89; CHECK-NEXT:    vllebrze %v24, 0(%r2)
90; CHECK-NEXT:    br %r14
91  %val = load i32, ptr %ptr
92  %swap = call i32 @llvm.bswap.i32(i32 %val)
93  %ret = insertelement <4 x i32> zeroinitializer, i32 %swap, i32 0
94  ret <4 x i32> %ret
95}
96
97; Test VLLEBRZE using a vector bswap.
98define <4 x i32> @f8(ptr %ptr) {
99; CHECK-LABEL: f8:
100; CHECK:       # %bb.0:
101; CHECK-NEXT:    vllebrze %v24, 0(%r2)
102; CHECK-NEXT:    br %r14
103  %val = load i32, ptr %ptr
104  %insert = insertelement <4 x i32> zeroinitializer, i32 %val, i32 0
105  %ret = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %insert)
106  ret <4 x i32> %ret
107}
108
109; Test VLLEBRZH with the highest in-range offset.
110define <8 x i16> @f9(ptr %base) {
111; CHECK-LABEL: f9:
112; CHECK:       # %bb.0:
113; CHECK-NEXT:    vllebrzh %v24, 4094(%r2)
114; CHECK-NEXT:    br %r14
115  %ptr = getelementptr i16, ptr %base, i64 2047
116  %val = load i16, ptr %ptr
117  %swap = call i16 @llvm.bswap.i16(i16 %val)
118  %ret = insertelement <8 x i16> zeroinitializer, i16 %swap, i32 3
119  ret <8 x i16> %ret
120}
121
122; Test VLLEBRZH with the next highest offset.
123define <8 x i16> @f10(ptr %base) {
124; CHECK-LABEL: f10:
125; CHECK:       # %bb.0:
126; CHECK-NEXT:    aghi %r2, 4096
127; CHECK-NEXT:    vllebrzh %v24, 0(%r2)
128; CHECK-NEXT:    br %r14
129  %ptr = getelementptr i16, ptr %base, i64 2048
130  %val = load i16, ptr %ptr
131  %swap = call i16 @llvm.bswap.i16(i16 %val)
132  %ret = insertelement <8 x i16> zeroinitializer, i16 %swap, i32 3
133  ret <8 x i16> %ret
134}
135
136; Test that VLLEBRZH allows an index.
137define <8 x i16> @f11(ptr %base, i64 %index) {
138; CHECK-LABEL: f11:
139; CHECK:       # %bb.0:
140; CHECK-NEXT:    sllg %r1, %r3, 1
141; CHECK-NEXT:    vllebrzh %v24, 0(%r1,%r2)
142; CHECK-NEXT:    br %r14
143  %ptr = getelementptr i16, ptr %base, i64 %index
144  %val = load i16, ptr %ptr
145  %swap = call i16 @llvm.bswap.i16(i16 %val)
146  %ret = insertelement <8 x i16> zeroinitializer, i16 %swap, i32 3
147  ret <8 x i16> %ret
148}
149
150