1; RUN: llc -mtriple=arm-eabi -mattr=+neon %s -o - | FileCheck %s
2
3; PR12281
4; Test generataion of code for vmull instruction when multiplying 128-bit
5; vectors that were created by sign-extending smaller vector sizes.
6;
7; The vmull operation requires 64-bit vectors, so we must extend the original
8; vector size to 64 bits for vmull operation.
9; Previously failed with an assertion because the <4 x i8> vector was too small
10; for vmull.
11
12; Vector x Constant
13; v4i8
14;
15define void @sextload_v4i8_c(<4 x i8>* %v) nounwind {
16;CHECK-LABEL: sextload_v4i8_c:
17entry:
18  %0 = load <4 x i8>, <4 x i8>* %v, align 8
19  %v0  = sext <4 x i8> %0 to <4 x i32>
20;CHECK: vmull
21  %v1 = mul <4 x i32>  %v0, <i32 3, i32 3, i32 3, i32 3>
22  store <4 x i32> %v1, <4 x i32>* undef, align 8
23  ret void;
24}
25
26; v2i8
27;
28define void @sextload_v2i8_c(<2 x i8>* %v) nounwind {
29;CHECK-LABEL: sextload_v2i8_c:
30entry:
31  %0   = load <2 x i8>, <2 x i8>* %v, align 8
32  %v0  = sext <2 x i8>  %0 to <2 x i64>
33;CHECK: vmull
34  %v1  = mul <2 x i64>  %v0, <i64 3, i64 3>
35  store <2 x i64> %v1, <2 x i64>* undef, align 8
36  ret void;
37}
38
39; v2i16
40;
41define void @sextload_v2i16_c(<2 x i16>* %v) nounwind {
42;CHECK-LABEL: sextload_v2i16_c:
43entry:
44  %0   = load <2 x i16>, <2 x i16>* %v, align 8
45  %v0  = sext <2 x i16>  %0 to <2 x i64>
46;CHECK: vmull
47  %v1  = mul <2 x i64>  %v0, <i64 3, i64 3>
48  store <2 x i64> %v1, <2 x i64>* undef, align 8
49  ret void;
50}
51
52
53; Vector x Vector
54; v4i8
55;
56define void @sextload_v4i8_v(<4 x i8>* %v, <4 x i8>* %p) nounwind {
57;CHECK-LABEL: sextload_v4i8_v:
58entry:
59  %0 = load <4 x i8>, <4 x i8>* %v, align 8
60  %v0  = sext <4 x i8> %0 to <4 x i32>
61
62  %1  = load <4 x i8>, <4 x i8>* %p, align 8
63  %v2 = sext <4 x i8> %1 to <4 x i32>
64;CHECK: vmull
65  %v1 = mul <4 x i32>  %v0, %v2
66  store <4 x i32> %v1, <4 x i32>* undef, align 8
67  ret void;
68}
69
70; v2i8
71;
72define void @sextload_v2i8_v(<2 x i8>* %v, <2 x i8>* %p) nounwind {
73;CHECK-LABEL: sextload_v2i8_v:
74entry:
75  %0 = load <2 x i8>, <2 x i8>* %v, align 8
76  %v0  = sext <2 x i8> %0 to <2 x i64>
77
78  %1  = load <2 x i8>, <2 x i8>* %p, align 8
79  %v2 = sext <2 x i8> %1 to <2 x i64>
80;CHECK: vmull
81  %v1 = mul <2 x i64>  %v0, %v2
82  store <2 x i64> %v1, <2 x i64>* undef, align 8
83  ret void;
84}
85
86; v2i16
87;
88define void @sextload_v2i16_v(<2 x i16>* %v, <2 x i16>* %p) nounwind {
89;CHECK-LABEL: sextload_v2i16_v:
90entry:
91  %0 = load <2 x i16>, <2 x i16>* %v, align 8
92  %v0  = sext <2 x i16> %0 to <2 x i64>
93
94  %1  = load <2 x i16>, <2 x i16>* %p, align 8
95  %v2 = sext <2 x i16> %1 to <2 x i64>
96;CHECK: vmull
97  %v1 = mul <2 x i64>  %v0, %v2
98  store <2 x i64> %v1, <2 x i64>* undef, align 8
99  ret void;
100}
101
102
103; Vector(small) x Vector(big)
104; v4i8 x v4i16
105;
106define void @sextload_v4i8_vs(<4 x i8>* %v, <4 x i16>* %p) nounwind {
107;CHECK-LABEL: sextload_v4i8_vs:
108entry:
109  %0 = load <4 x i8>, <4 x i8>* %v, align 8
110  %v0  = sext <4 x i8> %0 to <4 x i32>
111
112  %1  = load <4 x i16>, <4 x i16>* %p, align 8
113  %v2 = sext <4 x i16> %1 to <4 x i32>
114;CHECK: vmull
115  %v1 = mul <4 x i32>  %v0, %v2
116  store <4 x i32> %v1, <4 x i32>* undef, align 8
117  ret void;
118}
119
120; v2i8
121; v2i8 x v2i16
122define void @sextload_v2i8_vs(<2 x i8>* %v, <2 x i16>* %p) nounwind {
123;CHECK-LABEL: sextload_v2i8_vs:
124entry:
125  %0 = load <2 x i8>, <2 x i8>* %v, align 8
126  %v0  = sext <2 x i8> %0 to <2 x i64>
127
128  %1  = load <2 x i16>, <2 x i16>* %p, align 8
129  %v2 = sext <2 x i16> %1 to <2 x i64>
130;CHECK: vmull
131  %v1 = mul <2 x i64>  %v0, %v2
132  store <2 x i64> %v1, <2 x i64>* undef, align 8
133  ret void;
134}
135
136; v2i16
137; v2i16 x v2i32
138define void @sextload_v2i16_vs(<2 x i16>* %v, <2 x i32>* %p) nounwind {
139;CHECK-LABEL: sextload_v2i16_vs:
140entry:
141  %0 = load <2 x i16>, <2 x i16>* %v, align 8
142  %v0  = sext <2 x i16> %0 to <2 x i64>
143
144  %1  = load <2 x i32>, <2 x i32>* %p, align 8
145  %v2 = sext <2 x i32> %1 to <2 x i64>
146;CHECK: vmull
147  %v1 = mul <2 x i64>  %v0, %v2
148  store <2 x i64> %v1, <2 x i64>* undef, align 8
149  ret void;
150}
151