xref: /llvm-project/llvm/test/CodeGen/SystemZ/vec-bswap-03.ll (revision 0a76f7d9d8c1fc693568ed26420c47d92a6ba0e7) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
2; Test vector insertion of byte-swapped memory values.
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 v8i16 insertion into the first element.
14define <8 x i16> @f1(<8 x i16> %val, ptr %ptr) {
15; CHECK-LABEL: f1:
16; CHECK:       # %bb.0:
17; CHECK-NEXT:    vlebrh %v24, 0(%r2), 0
18; CHECK-NEXT:    br %r14
19  %element = load i16, ptr %ptr
20  %swap = call i16 @llvm.bswap.i16(i16 %element)
21  %ret = insertelement <8 x i16> %val, i16 %swap, i32 0
22  ret <8 x i16> %ret
23}
24
25; Test v8i16 insertion into the last element.
26define <8 x i16> @f2(<8 x i16> %val, ptr %ptr) {
27; CHECK-LABEL: f2:
28; CHECK:       # %bb.0:
29; CHECK-NEXT:    vlebrh %v24, 0(%r2), 7
30; CHECK-NEXT:    br %r14
31  %element = load i16, ptr %ptr
32  %swap = call i16 @llvm.bswap.i16(i16 %element)
33  %ret = insertelement <8 x i16> %val, i16 %swap, i32 7
34  ret <8 x i16> %ret
35}
36
37; Test v8i16 insertion with the highest in-range offset.
38define <8 x i16> @f3(<8 x i16> %val, ptr %base) {
39; CHECK-LABEL: f3:
40; CHECK:       # %bb.0:
41; CHECK-NEXT:    vlebrh %v24, 4094(%r2), 5
42; CHECK-NEXT:    br %r14
43  %ptr = getelementptr i16, ptr %base, i32 2047
44  %element = load i16, ptr %ptr
45  %swap = call i16 @llvm.bswap.i16(i16 %element)
46  %ret = insertelement <8 x i16> %val, i16 %swap, i32 5
47  ret <8 x i16> %ret
48}
49
50; Test v8i16 insertion with the first ouf-of-range offset.
51define <8 x i16> @f4(<8 x i16> %val, ptr %base) {
52; CHECK-LABEL: f4:
53; CHECK:       # %bb.0:
54; CHECK-NEXT:    aghi %r2, 4096
55; CHECK-NEXT:    vlebrh %v24, 0(%r2), 1
56; CHECK-NEXT:    br %r14
57  %ptr = getelementptr i16, ptr %base, i32 2048
58  %element = load i16, ptr %ptr
59  %swap = call i16 @llvm.bswap.i16(i16 %element)
60  %ret = insertelement <8 x i16> %val, i16 %swap, i32 1
61  ret <8 x i16> %ret
62}
63
64; Test v8i16 insertion into a variable element.
65define <8 x i16> @f5(<8 x i16> %val, ptr %ptr, i32 %index) {
66; CHECK-LABEL: f5:
67; CHECK:       # %bb.0:
68; CHECK-NEXT:    lrvh %r0, 0(%r2)
69; CHECK-NEXT:    vlvgh %v24, %r0, 0(%r3)
70; CHECK-NEXT:    br %r14
71  %element = load i16, ptr %ptr
72  %swap = call i16 @llvm.bswap.i16(i16 %element)
73  %ret = insertelement <8 x i16> %val, i16 %swap, i32 %index
74  ret <8 x i16> %ret
75}
76
77; Test v8i16 insertion using a pair of vector bswaps.
78define <8 x i16> @f6(<8 x i16> %val, ptr %ptr) {
79; CHECK-LABEL: f6:
80; CHECK:       # %bb.0:
81; CHECK-NEXT:    vlebrh %v24, 0(%r2), 0
82; CHECK-NEXT:    br %r14
83  %element = load i16, ptr %ptr
84  %swapval = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %val)
85  %insert = insertelement <8 x i16> %swapval, i16 %element, i32 0
86  %ret = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %insert)
87  ret <8 x i16> %ret
88}
89
90; Test v4i32 insertion into the first element.
91define <4 x i32> @f7(<4 x i32> %val, ptr %ptr) {
92; CHECK-LABEL: f7:
93; CHECK:       # %bb.0:
94; CHECK-NEXT:    vlebrf %v24, 0(%r2), 0
95; CHECK-NEXT:    br %r14
96  %element = load i32, ptr %ptr
97  %swap = call i32 @llvm.bswap.i32(i32 %element)
98  %ret = insertelement <4 x i32> %val, i32 %swap, i32 0
99  ret <4 x i32> %ret
100}
101
102; Test v4i32 insertion into the last element.
103define <4 x i32> @f8(<4 x i32> %val, ptr %ptr) {
104; CHECK-LABEL: f8:
105; CHECK:       # %bb.0:
106; CHECK-NEXT:    vlebrf %v24, 0(%r2), 3
107; CHECK-NEXT:    br %r14
108  %element = load i32, ptr %ptr
109  %swap = call i32 @llvm.bswap.i32(i32 %element)
110  %ret = insertelement <4 x i32> %val, i32 %swap, i32 3
111  ret <4 x i32> %ret
112}
113
114; Test v4i32 insertion with the highest in-range offset.
115define <4 x i32> @f9(<4 x i32> %val, ptr %base) {
116; CHECK-LABEL: f9:
117; CHECK:       # %bb.0:
118; CHECK-NEXT:    vlebrf %v24, 4092(%r2), 2
119; CHECK-NEXT:    br %r14
120  %ptr = getelementptr i32, ptr %base, i32 1023
121  %element = load i32, ptr %ptr
122  %swap = call i32 @llvm.bswap.i32(i32 %element)
123  %ret = insertelement <4 x i32> %val, i32 %swap, i32 2
124  ret <4 x i32> %ret
125}
126
127; Test v4i32 insertion with the first ouf-of-range offset.
128define <4 x i32> @f10(<4 x i32> %val, ptr %base) {
129; CHECK-LABEL: f10:
130; CHECK:       # %bb.0:
131; CHECK-NEXT:    aghi %r2, 4096
132; CHECK-NEXT:    vlebrf %v24, 0(%r2), 1
133; CHECK-NEXT:    br %r14
134  %ptr = getelementptr i32, ptr %base, i32 1024
135  %element = load i32, ptr %ptr
136  %swap = call i32 @llvm.bswap.i32(i32 %element)
137  %ret = insertelement <4 x i32> %val, i32 %swap, i32 1
138  ret <4 x i32> %ret
139}
140
141; Test v4i32 insertion into a variable element.
142define <4 x i32> @f11(<4 x i32> %val, ptr %ptr, i32 %index) {
143; CHECK-LABEL: f11:
144; CHECK:       # %bb.0:
145; CHECK-NEXT:    lrv %r0, 0(%r2)
146; CHECK-NEXT:    vlvgf %v24, %r0, 0(%r3)
147; CHECK-NEXT:    br %r14
148  %element = load i32, ptr %ptr
149  %swap = call i32 @llvm.bswap.i32(i32 %element)
150  %ret = insertelement <4 x i32> %val, i32 %swap, i32 %index
151  ret <4 x i32> %ret
152}
153
154; Test v4i32 insertion using a pair of vector bswaps.
155define <4 x i32> @f12(<4 x i32> %val, ptr %ptr) {
156; CHECK-LABEL: f12:
157; CHECK:       # %bb.0:
158; CHECK-NEXT:    vlebrf %v24, 0(%r2), 0
159; CHECK-NEXT:    br %r14
160  %element = load i32, ptr %ptr
161  %swapval = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %val)
162  %insert = insertelement <4 x i32> %swapval, i32 %element, i32 0
163  %ret = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %insert)
164  ret <4 x i32> %ret
165}
166
167; Test v2i64 insertion into the first element.
168define <2 x i64> @f13(<2 x i64> %val, ptr %ptr) {
169; CHECK-LABEL: f13:
170; CHECK:       # %bb.0:
171; CHECK-NEXT:    vlebrg %v24, 0(%r2), 0
172; CHECK-NEXT:    br %r14
173  %element = load i64, ptr %ptr
174  %swap = call i64 @llvm.bswap.i64(i64 %element)
175  %ret = insertelement <2 x i64> %val, i64 %swap, i32 0
176  ret <2 x i64> %ret
177}
178
179; Test v2i64 insertion into the last element.
180define <2 x i64> @f14(<2 x i64> %val, ptr %ptr) {
181; CHECK-LABEL: f14:
182; CHECK:       # %bb.0:
183; CHECK-NEXT:    vlebrg %v24, 0(%r2), 1
184; CHECK-NEXT:    br %r14
185  %element = load i64, ptr %ptr
186  %swap = call i64 @llvm.bswap.i64(i64 %element)
187  %ret = insertelement <2 x i64> %val, i64 %swap, i32 1
188  ret <2 x i64> %ret
189}
190
191; Test v2i64 insertion with the highest in-range offset.
192define <2 x i64> @f15(<2 x i64> %val, ptr %base) {
193; CHECK-LABEL: f15:
194; CHECK:       # %bb.0:
195; CHECK-NEXT:    vlebrg %v24, 4088(%r2), 1
196; CHECK-NEXT:    br %r14
197  %ptr = getelementptr i64, ptr %base, i32 511
198  %element = load i64, ptr %ptr
199  %swap = call i64 @llvm.bswap.i64(i64 %element)
200  %ret = insertelement <2 x i64> %val, i64 %swap, i32 1
201  ret <2 x i64> %ret
202}
203
204; Test v2i64 insertion with the first ouf-of-range offset.
205define <2 x i64> @f16(<2 x i64> %val, ptr %base) {
206; CHECK-LABEL: f16:
207; CHECK:       # %bb.0:
208; CHECK-NEXT:    aghi %r2, 4096
209; CHECK-NEXT:    vlebrg %v24, 0(%r2), 0
210; CHECK-NEXT:    br %r14
211  %ptr = getelementptr i64, ptr %base, i32 512
212  %element = load i64, ptr %ptr
213  %swap = call i64 @llvm.bswap.i64(i64 %element)
214  %ret = insertelement <2 x i64> %val, i64 %swap, i32 0
215  ret <2 x i64> %ret
216}
217
218; Test v2i64 insertion into a variable element.
219define <2 x i64> @f17(<2 x i64> %val, ptr %ptr, i32 %index) {
220; CHECK-LABEL: f17:
221; CHECK:       # %bb.0:
222; CHECK-NEXT:    lrvg %r0, 0(%r2)
223; CHECK-NEXT:    vlvgg %v24, %r0, 0(%r3)
224; CHECK-NEXT:    br %r14
225  %element = load i64, ptr %ptr
226  %swap = call i64 @llvm.bswap.i64(i64 %element)
227  %ret = insertelement <2 x i64> %val, i64 %swap, i32 %index
228  ret <2 x i64> %ret
229}
230
231; Test v2i64 insertion using a pair of vector bswaps.
232define <2 x i64> @f18(<2 x i64> %val, ptr %ptr) {
233; CHECK-LABEL: f18:
234; CHECK:       # %bb.0:
235; CHECK-NEXT:    vlebrg %v24, 0(%r2), 0
236; CHECK-NEXT:    br %r14
237  %element = load i64, ptr %ptr
238  %swapval = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %val)
239  %insert = insertelement <2 x i64> %swapval, i64 %element, i32 0
240  %ret = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %insert)
241  ret <2 x i64> %ret
242}
243