1; Test vector merge low.
2;
3; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s
4
5; Test a canonical v16i8 merge low.
6define <16 x i8> @f1(<16 x i8> %val1, <16 x i8> %val2) {
7; CHECK-LABEL: f1:
8; CHECK: vmrlb %v24, %v24, %v26
9; CHECK: br %r14
10  %ret = shufflevector <16 x i8> %val1, <16 x i8> %val2,
11                       <16 x i32> <i32 8, i32 24, i32 9, i32 25,
12                                   i32 10, i32 26, i32 11, i32 27,
13                                   i32 12, i32 28, i32 13, i32 29,
14                                   i32 14, i32 30, i32 15, i32 31>
15  ret <16 x i8> %ret
16}
17
18; Test a reversed v16i8 merge low.
19define <16 x i8> @f2(<16 x i8> %val1, <16 x i8> %val2) {
20; CHECK-LABEL: f2:
21; CHECK: vmrlb %v24, %v26, %v24
22; CHECK: br %r14
23  %ret = shufflevector <16 x i8> %val1, <16 x i8> %val2,
24                       <16 x i32> <i32 24, i32 8, i32 25, i32 9,
25                                   i32 26, i32 10, i32 27, i32 11,
26                                   i32 28, i32 12, i32 29, i32 13,
27                                   i32 30, i32 14, i32 31, i32 15>
28  ret <16 x i8> %ret
29}
30
31; Test a v16i8 merge low with only the first operand being used.
32define <16 x i8> @f3(<16 x i8> %val1, <16 x i8> %val2) {
33; CHECK-LABEL: f3:
34; CHECK: vmrlb %v24, %v24, %v24
35; CHECK: br %r14
36  %ret = shufflevector <16 x i8> %val1, <16 x i8> %val2,
37                       <16 x i32> <i32 8, i32 8, i32 9, i32 9,
38                                   i32 10, i32 10, i32 11, i32 11,
39                                   i32 12, i32 12, i32 13, i32 13,
40                                   i32 14, i32 14, i32 15, i32 15>
41  ret <16 x i8> %ret
42}
43
44; Test a v16i8 merge low with only the second operand being used.
45; This is converted into @f3 by target-independent code.
46define <16 x i8> @f4(<16 x i8> %val1, <16 x i8> %val2) {
47; CHECK-LABEL: f4:
48; CHECK: vmrlb %v24, %v26, %v26
49; CHECK: br %r14
50  %ret = shufflevector <16 x i8> %val1, <16 x i8> %val2,
51                       <16 x i32> <i32 24, i32 24, i32 25, i32 25,
52                                   i32 26, i32 26, i32 27, i32 27,
53                                   i32 28, i32 28, i32 29, i32 29,
54                                   i32 30, i32 30, i32 31, i32 31>
55  ret <16 x i8> %ret
56}
57
58; Test a v16i8 merge with both operands being the same.  This too is
59; converted into @f3 by target-independent code.
60define <16 x i8> @f5(<16 x i8> %val) {
61; CHECK-LABEL: f5:
62; CHECK: vmrlb %v24, %v24, %v24
63; CHECK: br %r14
64  %ret = shufflevector <16 x i8> %val, <16 x i8> %val,
65                       <16 x i32> <i32 8, i32 24, i32 25, i32 25,
66                                   i32 26, i32 10, i32 11, i32 11,
67                                   i32 28, i32 28, i32 13, i32 13,
68                                   i32 14, i32 30, i32 31, i32 15>
69  ret <16 x i8> %ret
70}
71
72; Test a v16i8 merge in which some of the indices are don't care.
73define <16 x i8> @f6(<16 x i8> %val1, <16 x i8> %val2) {
74; CHECK-LABEL: f6:
75; CHECK: vmrlb %v24, %v24, %v26
76; CHECK: br %r14
77  %ret = shufflevector <16 x i8> %val1, <16 x i8> %val2,
78                       <16 x i32> <i32 8, i32 undef, i32 9, i32 25,
79                                   i32 undef, i32 26, i32 undef, i32 undef,
80                                   i32 undef, i32 28, i32 13, i32 29,
81                                   i32 undef, i32 30, i32 15, i32 undef>
82  ret <16 x i8> %ret
83}
84
85; Test a v16i8 merge in which one of the operands is undefined and where
86; indices for that operand are "don't care".  Target-independent code
87; converts the indices themselves into "undef"s.
88define <16 x i8> @f7(<16 x i8> %val) {
89; CHECK-LABEL: f7:
90; CHECK: vmrlb %v24, %v24, %v24
91; CHECK: br %r14
92  %ret = shufflevector <16 x i8> undef, <16 x i8> %val,
93                       <16 x i32> <i32 11, i32 24, i32 25, i32 5,
94                                   i32 26, i32 10, i32 27, i32 27,
95                                   i32 28, i32 28, i32 29, i32 3,
96                                   i32 2, i32 30, i32 9, i32 31>
97  ret <16 x i8> %ret
98}
99
100; Test a canonical v8i16 merge low.
101define <8 x i16> @f8(<8 x i16> %val1, <8 x i16> %val2) {
102; CHECK-LABEL: f8:
103; CHECK: vmrlh %v24, %v24, %v26
104; CHECK: br %r14
105  %ret = shufflevector <8 x i16> %val1, <8 x i16> %val2,
106                       <8 x i32> <i32 4, i32 12, i32 5, i32 13,
107                                  i32 6, i32 14, i32 7, i32 15>
108  ret <8 x i16> %ret
109}
110
111; Test a reversed v8i16 merge low.
112define <8 x i16> @f9(<8 x i16> %val1, <8 x i16> %val2) {
113; CHECK-LABEL: f9:
114; CHECK: vmrlh %v24, %v26, %v24
115; CHECK: br %r14
116  %ret = shufflevector <8 x i16> %val1, <8 x i16> %val2,
117                       <8 x i32> <i32 12, i32 4, i32 13, i32 5,
118                                  i32 14, i32 6, i32 15, i32 7>
119  ret <8 x i16> %ret
120}
121
122; Test a canonical v4i32 merge low.
123define <4 x i32> @f10(<4 x i32> %val1, <4 x i32> %val2) {
124; CHECK-LABEL: f10:
125; CHECK: vmrlf %v24, %v24, %v26
126; CHECK: br %r14
127  %ret = shufflevector <4 x i32> %val1, <4 x i32> %val2,
128                       <4 x i32> <i32 2, i32 6, i32 3, i32 7>
129  ret <4 x i32> %ret
130}
131
132; Test a reversed v4i32 merge low.
133define <4 x i32> @f11(<4 x i32> %val1, <4 x i32> %val2) {
134; CHECK-LABEL: f11:
135; CHECK: vmrlf %v24, %v26, %v24
136; CHECK: br %r14
137  %ret = shufflevector <4 x i32> %val1, <4 x i32> %val2,
138                       <4 x i32> <i32 6, i32 2, i32 7, i32 3>
139  ret <4 x i32> %ret
140}
141
142; Test a canonical v2i64 merge low.
143define <2 x i64> @f12(<2 x i64> %val1, <2 x i64> %val2) {
144; CHECK-LABEL: f12:
145; CHECK: vmrlg %v24, %v24, %v26
146; CHECK: br %r14
147  %ret = shufflevector <2 x i64> %val1, <2 x i64> %val2,
148                       <2 x i32> <i32 1, i32 3>
149  ret <2 x i64> %ret
150}
151
152; Test a reversed v2i64 merge low.
153define <2 x i64> @f13(<2 x i64> %val1, <2 x i64> %val2) {
154; CHECK-LABEL: f13:
155; CHECK: vmrlg %v24, %v26, %v24
156; CHECK: br %r14
157  %ret = shufflevector <2 x i64> %val1, <2 x i64> %val2,
158                       <2 x i32> <i32 3, i32 1>
159  ret <2 x i64> %ret
160}
161
162; Test a canonical v4f32 merge low.
163define <4 x float> @f14(<4 x float> %val1, <4 x float> %val2) {
164; CHECK-LABEL: f14:
165; CHECK: vmrlf %v24, %v24, %v26
166; CHECK: br %r14
167  %ret = shufflevector <4 x float> %val1, <4 x float> %val2,
168                       <4 x i32> <i32 2, i32 6, i32 3, i32 7>
169  ret <4 x float> %ret
170}
171
172; Test a reversed v4f32 merge low.
173define <4 x float> @f15(<4 x float> %val1, <4 x float> %val2) {
174; CHECK-LABEL: f15:
175; CHECK: vmrlf %v24, %v26, %v24
176; CHECK: br %r14
177  %ret = shufflevector <4 x float> %val1, <4 x float> %val2,
178                       <4 x i32> <i32 6, i32 2, i32 7, i32 3>
179  ret <4 x float> %ret
180}
181
182; Test a canonical v2f64 merge low.
183define <2 x double> @f16(<2 x double> %val1, <2 x double> %val2) {
184; CHECK-LABEL: f16:
185; CHECK: vmrlg %v24, %v24, %v26
186; CHECK: br %r14
187  %ret = shufflevector <2 x double> %val1, <2 x double> %val2,
188                       <2 x i32> <i32 1, i32 3>
189  ret <2 x double> %ret
190}
191
192; Test a reversed v2f64 merge low.
193define <2 x double> @f17(<2 x double> %val1, <2 x double> %val2) {
194; CHECK-LABEL: f17:
195; CHECK: vmrlg %v24, %v26, %v24
196; CHECK: br %r14
197  %ret = shufflevector <2 x double> %val1, <2 x double> %val2,
198                       <2 x i32> <i32 3, i32 1>
199  ret <2 x double> %ret
200}
201