1
; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s
3
%struct.__neon_int8x8x4_t = type { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> }
4
%struct.__neon_int16x4x4_t = type { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> }
5
%struct.__neon_int32x2x4_t = type { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> }
6
%struct.__neon_float32x2x4_t = type { <2 x float>, <2 x float>, <2 x float>, <2 x float> }
7
%struct.__neon_int64x1x4_t = type { <1 x i64>, <1 x i64>, <1 x i64>, <1 x i64> }
9
%struct.__neon_int8x16x4_t = type { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> }
10
%struct.__neon_int16x8x4_t = type { <8 x i16>, <8 x i16>, <8 x i16>, <8 x i16> }
11
%struct.__neon_int32x4x4_t = type { <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32> }
12
%struct.__neon_float32x4x4_t = type { <4 x float>, <4 x float>, <4 x float>, <4 x float> }
14
define <8 x i8> @vld4i8(i8* %A) nounwind {
17
%tmp1 = call %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8(i8* %A)
18
%tmp2 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 0
19
%tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 2
20
%tmp4 = add <8 x i8> %tmp2, %tmp3
24
define <4 x i16> @vld4i16(i16* %A) nounwind {
27
%tmp1 = call %struct.__neon_int16x4x4_t @llvm.arm.neon.vld4.v4i16(i16* %A)
28
%tmp2 = extractvalue %struct.__neon_int16x4x4_t %tmp1, 0
29
%tmp3 = extractvalue %struct.__neon_int16x4x4_t %tmp1, 2
30
%tmp4 = add <4 x i16> %tmp2, %tmp3
34
define <2 x i32> @vld4i32(i32* %A) nounwind {
37
%tmp1 = call %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4.v2i32(i32* %A)
38
%tmp2 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 0
39
%tmp3 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 2
40
%tmp4 = add <2 x i32> %tmp2, %tmp3
44
define <2 x float> @vld4f(float* %A) nounwind {
47
%tmp1 = call %struct.__neon_float32x2x4_t @llvm.arm.neon.vld4.v2f32(float* %A)
48
%tmp2 = extractvalue %struct.__neon_float32x2x4_t %tmp1, 0
49
%tmp3 = extractvalue %struct.__neon_float32x2x4_t %tmp1, 2
50
%tmp4 = add <2 x float> %tmp2, %tmp3
54
define <1 x i64> @vld4i64(i64* %A) nounwind {
57
%tmp1 = call %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64(i64* %A)
58
%tmp2 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 0
59
%tmp3 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 2
60
%tmp4 = add <1 x i64> %tmp2, %tmp3
64
define <16 x i8> @vld4Qi8(i8* %A) nounwind {
68
%tmp1 = call %struct.__neon_int8x16x4_t @llvm.arm.neon.vld4.v16i8(i8* %A)
69
%tmp2 = extractvalue %struct.__neon_int8x16x4_t %tmp1, 0
70
%tmp3 = extractvalue %struct.__neon_int8x16x4_t %tmp1, 2
71
%tmp4 = add <16 x i8> %tmp2, %tmp3
75
define <8 x i16> @vld4Qi16(i16* %A) nounwind {
79
%tmp1 = call %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16(i16* %A)
80
%tmp2 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 0
81
%tmp3 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 2
82
%tmp4 = add <8 x i16> %tmp2, %tmp3
86
define <4 x i32> @vld4Qi32(i32* %A) nounwind {
90
%tmp1 = call %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4.v4i32(i32* %A)
91
%tmp2 = extractvalue %struct.__neon_int32x4x4_t %tmp1, 0
92
%tmp3 = extractvalue %struct.__neon_int32x4x4_t %tmp1, 2
93
%tmp4 = add <4 x i32> %tmp2, %tmp3
97
define <4 x float> @vld4Qf(float* %A) nounwind {
101
%tmp1 = call %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4.v4f32(float* %A)
102
%tmp2 = extractvalue %struct.__neon_float32x4x4_t %tmp1, 0
103
%tmp3 = extractvalue %struct.__neon_float32x4x4_t %tmp1, 2
104
%tmp4 = add <4 x float> %tmp2, %tmp3
105
ret <4 x float> %tmp4
108
declare %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8(i8*) nounwind readonly
109
declare %struct.__neon_int16x4x4_t @llvm.arm.neon.vld4.v4i16(i8*) nounwind readonly
110
declare %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4.v2i32(i8*) nounwind readonly
111
declare %struct.__neon_float32x2x4_t @llvm.arm.neon.vld4.v2f32(i8*) nounwind readonly
112
declare %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64(i8*) nounwind readonly
114
declare %struct.__neon_int8x16x4_t @llvm.arm.neon.vld4.v16i8(i8*) nounwind readonly
115
declare %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16(i8*) nounwind readonly
116
declare %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4.v4i32(i8*) nounwind readonly
117
declare %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4.v4f32(i8*) nounwind readonly