1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
|
; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s
%struct.__neon_int8x8x4_t = type { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> }
%struct.__neon_int16x4x4_t = type { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> }
%struct.__neon_int32x2x4_t = type { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> }
%struct.__neon_float32x2x4_t = type { <2 x float>, <2 x float>, <2 x float>, <2 x float> }
%struct.__neon_int64x1x4_t = type { <1 x i64>, <1 x i64>, <1 x i64>, <1 x i64> }
%struct.__neon_int8x16x4_t = type { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> }
%struct.__neon_int16x8x4_t = type { <8 x i16>, <8 x i16>, <8 x i16>, <8 x i16> }
%struct.__neon_int32x4x4_t = type { <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32> }
%struct.__neon_float32x4x4_t = type { <4 x float>, <4 x float>, <4 x float>, <4 x float> }
define <8 x i8> @vld4i8(i8* %A) nounwind {
;CHECK-LABEL: vld4i8:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld4.8 {d16, d17, d18, d19}, [r0:64]
%tmp1 = call %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8(i8* %A, i32 8)
%tmp2 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 2
%tmp4 = add <8 x i8> %tmp2, %tmp3
ret <8 x i8> %tmp4
}
;Check for a post-increment updating load with register increment.
define <8 x i8> @vld4i8_update(i8** %ptr, i32 %inc) nounwind {
;CHECK-LABEL: vld4i8_update:
;CHECK: vld4.8 {d16, d17, d18, d19}, [r2:128], r1
%A = load i8** %ptr
%tmp1 = call %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8(i8* %A, i32 16)
%tmp2 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 2
%tmp4 = add <8 x i8> %tmp2, %tmp3
%tmp5 = getelementptr i8* %A, i32 %inc
store i8* %tmp5, i8** %ptr
ret <8 x i8> %tmp4
}
define <4 x i16> @vld4i16(i16* %A) nounwind {
;CHECK-LABEL: vld4i16:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld4.16 {d16, d17, d18, d19}, [r0:128]
%tmp0 = bitcast i16* %A to i8*
%tmp1 = call %struct.__neon_int16x4x4_t @llvm.arm.neon.vld4.v4i16(i8* %tmp0, i32 16)
%tmp2 = extractvalue %struct.__neon_int16x4x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int16x4x4_t %tmp1, 2
%tmp4 = add <4 x i16> %tmp2, %tmp3
ret <4 x i16> %tmp4
}
define <2 x i32> @vld4i32(i32* %A) nounwind {
;CHECK-LABEL: vld4i32:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld4.32 {d16, d17, d18, d19}, [r0:256]
%tmp0 = bitcast i32* %A to i8*
%tmp1 = call %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4.v2i32(i8* %tmp0, i32 32)
%tmp2 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 2
%tmp4 = add <2 x i32> %tmp2, %tmp3
ret <2 x i32> %tmp4
}
define <2 x float> @vld4f(float* %A) nounwind {
;CHECK-LABEL: vld4f:
;CHECK: vld4.32
%tmp0 = bitcast float* %A to i8*
%tmp1 = call %struct.__neon_float32x2x4_t @llvm.arm.neon.vld4.v2f32(i8* %tmp0, i32 1)
%tmp2 = extractvalue %struct.__neon_float32x2x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_float32x2x4_t %tmp1, 2
%tmp4 = fadd <2 x float> %tmp2, %tmp3
ret <2 x float> %tmp4
}
define <1 x i64> @vld4i64(i64* %A) nounwind {
;CHECK-LABEL: vld4i64:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld1.64 {d16, d17, d18, d19}, [r0:256]
%tmp0 = bitcast i64* %A to i8*
%tmp1 = call %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64(i8* %tmp0, i32 64)
%tmp2 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 2
%tmp4 = add <1 x i64> %tmp2, %tmp3
ret <1 x i64> %tmp4
}
define <1 x i64> @vld4i64_update(i64** %ptr, i64* %A) nounwind {
;CHECK-LABEL: vld4i64_update:
;CHECK: vld1.64 {d16, d17, d18, d19}, [r1:256]!
%tmp0 = bitcast i64* %A to i8*
%tmp1 = call %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64(i8* %tmp0, i32 64)
%tmp5 = getelementptr i64* %A, i32 4
store i64* %tmp5, i64** %ptr
%tmp2 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 2
%tmp4 = add <1 x i64> %tmp2, %tmp3
ret <1 x i64> %tmp4
}
define <16 x i8> @vld4Qi8(i8* %A) nounwind {
;CHECK-LABEL: vld4Qi8:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld4.8 {d16, d18, d20, d22}, [r0:256]!
;CHECK: vld4.8 {d17, d19, d21, d23}, [r0:256]
%tmp1 = call %struct.__neon_int8x16x4_t @llvm.arm.neon.vld4.v16i8(i8* %A, i32 64)
%tmp2 = extractvalue %struct.__neon_int8x16x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int8x16x4_t %tmp1, 2
%tmp4 = add <16 x i8> %tmp2, %tmp3
ret <16 x i8> %tmp4
}
define <8 x i16> @vld4Qi16(i16* %A) nounwind {
;CHECK-LABEL: vld4Qi16:
;Check for no alignment specifier.
;CHECK: vld4.16 {d16, d18, d20, d22}, [r0]!
;CHECK: vld4.16 {d17, d19, d21, d23}, [r0]
%tmp0 = bitcast i16* %A to i8*
%tmp1 = call %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16(i8* %tmp0, i32 1)
%tmp2 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 2
%tmp4 = add <8 x i16> %tmp2, %tmp3
ret <8 x i16> %tmp4
}
;Check for a post-increment updating load.
define <8 x i16> @vld4Qi16_update(i16** %ptr) nounwind {
;CHECK-LABEL: vld4Qi16_update:
;CHECK: vld4.16 {d16, d18, d20, d22}, [r1:64]!
;CHECK: vld4.16 {d17, d19, d21, d23}, [r1:64]!
%A = load i16** %ptr
%tmp0 = bitcast i16* %A to i8*
%tmp1 = call %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16(i8* %tmp0, i32 8)
%tmp2 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 2
%tmp4 = add <8 x i16> %tmp2, %tmp3
%tmp5 = getelementptr i16* %A, i32 32
store i16* %tmp5, i16** %ptr
ret <8 x i16> %tmp4
}
define <4 x i32> @vld4Qi32(i32* %A) nounwind {
;CHECK-LABEL: vld4Qi32:
;CHECK: vld4.32
;CHECK: vld4.32
%tmp0 = bitcast i32* %A to i8*
%tmp1 = call %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4.v4i32(i8* %tmp0, i32 1)
%tmp2 = extractvalue %struct.__neon_int32x4x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int32x4x4_t %tmp1, 2
%tmp4 = add <4 x i32> %tmp2, %tmp3
ret <4 x i32> %tmp4
}
define <4 x float> @vld4Qf(float* %A) nounwind {
;CHECK-LABEL: vld4Qf:
;CHECK: vld4.32
;CHECK: vld4.32
%tmp0 = bitcast float* %A to i8*
%tmp1 = call %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4.v4f32(i8* %tmp0, i32 1)
%tmp2 = extractvalue %struct.__neon_float32x4x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_float32x4x4_t %tmp1, 2
%tmp4 = fadd <4 x float> %tmp2, %tmp3
ret <4 x float> %tmp4
}
declare %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8(i8*, i32) nounwind readonly
declare %struct.__neon_int16x4x4_t @llvm.arm.neon.vld4.v4i16(i8*, i32) nounwind readonly
declare %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4.v2i32(i8*, i32) nounwind readonly
declare %struct.__neon_float32x2x4_t @llvm.arm.neon.vld4.v2f32(i8*, i32) nounwind readonly
declare %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64(i8*, i32) nounwind readonly
declare %struct.__neon_int8x16x4_t @llvm.arm.neon.vld4.v16i8(i8*, i32) nounwind readonly
declare %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16(i8*, i32) nounwind readonly
declare %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4.v4i32(i8*, i32) nounwind readonly
declare %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4.v4f32(i8*, i32) nounwind readonly
|
