1 ; RUN: llc -mtriple=arm-eabi -mattr=+neon %s -o - | 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 {
16 ;Check the alignment value. Max for this instruction is 256 bits:
17 ;CHECK: vld4.8 {d16, d17, d18, d19}, [{{r[0-9]+|lr}}:64]
18 %tmp1 = call %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8.p0i8(i8* %A, i32 8)
19 %tmp2 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 0
20 %tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 2
21 %tmp4 = add <8 x i8> %tmp2, %tmp3
25 ;Check for a post-increment updating load with register increment.
26 define <8 x i8> @vld4i8_update(i8** %ptr, i32 %inc) nounwind {
27 ;CHECK-LABEL: vld4i8_update:
28 ;CHECK: vld4.8 {d16, d17, d18, d19}, [{{r[0-9]+|lr}}:128], r1
29 %A = load i8*, i8** %ptr
30 %tmp1 = call %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8.p0i8(i8* %A, i32 16)
31 %tmp2 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 0
32 %tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 2
33 %tmp4 = add <8 x i8> %tmp2, %tmp3
34 %tmp5 = getelementptr i8, i8* %A, i32 %inc
35 store i8* %tmp5, i8** %ptr
39 define <4 x i16> @vld4i16(i16* %A) nounwind {
40 ;CHECK-LABEL: vld4i16:
41 ;Check the alignment value. Max for this instruction is 256 bits:
42 ;CHECK: vld4.16 {d16, d17, d18, d19}, [{{r[0-9]+|lr}}:128]
43 %tmp0 = bitcast i16* %A to i8*
44 %tmp1 = call %struct.__neon_int16x4x4_t @llvm.arm.neon.vld4.v4i16.p0i8(i8* %tmp0, i32 16)
45 %tmp2 = extractvalue %struct.__neon_int16x4x4_t %tmp1, 0
46 %tmp3 = extractvalue %struct.__neon_int16x4x4_t %tmp1, 2
47 %tmp4 = add <4 x i16> %tmp2, %tmp3
51 define <2 x i32> @vld4i32(i32* %A) nounwind {
52 ;CHECK-LABEL: vld4i32:
53 ;Check the alignment value. Max for this instruction is 256 bits:
54 ;CHECK: vld4.32 {d16, d17, d18, d19}, [{{r[0-9]+|lr}}:256]
55 %tmp0 = bitcast i32* %A to i8*
56 %tmp1 = call %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4.v2i32.p0i8(i8* %tmp0, i32 32)
57 %tmp2 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 0
58 %tmp3 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 2
59 %tmp4 = add <2 x i32> %tmp2, %tmp3
63 define <2 x float> @vld4f(float* %A) nounwind {
66 %tmp0 = bitcast float* %A to i8*
67 %tmp1 = call %struct.__neon_float32x2x4_t @llvm.arm.neon.vld4.v2f32.p0i8(i8* %tmp0, i32 1)
68 %tmp2 = extractvalue %struct.__neon_float32x2x4_t %tmp1, 0
69 %tmp3 = extractvalue %struct.__neon_float32x2x4_t %tmp1, 2
70 %tmp4 = fadd <2 x float> %tmp2, %tmp3
74 define <1 x i64> @vld4i64(i64* %A) nounwind {
75 ;CHECK-LABEL: vld4i64:
76 ;Check the alignment value. Max for this instruction is 256 bits:
77 ;CHECK: vld1.64 {d16, d17, d18, d19}, [{{r[0-9]+|lr}}:256]
78 %tmp0 = bitcast i64* %A to i8*
79 %tmp1 = call %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64.p0i8(i8* %tmp0, i32 64)
80 %tmp2 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 0
81 %tmp3 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 2
82 %tmp4 = add <1 x i64> %tmp2, %tmp3
86 define <1 x i64> @vld4i64_update(i64** %ptr, i64* %A) nounwind {
87 ;CHECK-LABEL: vld4i64_update:
88 ;CHECK: vld1.64 {d16, d17, d18, d19}, [{{r[0-9]+|lr}}:256]!
89 %tmp0 = bitcast i64* %A to i8*
90 %tmp1 = call %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64.p0i8(i8* %tmp0, i32 64)
91 %tmp5 = getelementptr i64, i64* %A, i32 4
92 store i64* %tmp5, i64** %ptr
93 %tmp2 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 0
94 %tmp3 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 2
95 %tmp4 = add <1 x i64> %tmp2, %tmp3
99 define <1 x i64> @vld4i64_reg_update(i64** %ptr, i64* %A) nounwind {
100 ;CHECK-LABEL: vld4i64_reg_update:
101 ;CHECK: vld1.64 {d16, d17, d18, d19}, [{{r[0-9]+|lr}}:256], {{r[0-9]+|lr}}
102 %tmp0 = bitcast i64* %A to i8*
103 %tmp1 = call %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64.p0i8(i8* %tmp0, i32 64)
104 %tmp5 = getelementptr i64, i64* %A, i32 1
105 store i64* %tmp5, i64** %ptr
106 %tmp2 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 0
107 %tmp3 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 2
108 %tmp4 = add <1 x i64> %tmp2, %tmp3
112 define <16 x i8> @vld4Qi8(i8* %A) nounwind {
113 ;CHECK-LABEL: vld4Qi8:
114 ;Check the alignment value. Max for this instruction is 256 bits:
115 ;CHECK: vld4.8 {d16, d18, d20, d22}, [{{r[0-9]+|lr}}:256]!
116 ;CHECK: vld4.8 {d17, d19, d21, d23}, [{{r[0-9]+|lr}}:256]
117 %tmp1 = call %struct.__neon_int8x16x4_t @llvm.arm.neon.vld4.v16i8.p0i8(i8* %A, i32 64)
118 %tmp2 = extractvalue %struct.__neon_int8x16x4_t %tmp1, 0
119 %tmp3 = extractvalue %struct.__neon_int8x16x4_t %tmp1, 2
120 %tmp4 = add <16 x i8> %tmp2, %tmp3
124 define <8 x i16> @vld4Qi16(i16* %A) nounwind {
125 ;CHECK-LABEL: vld4Qi16:
126 ;Check for no alignment specifier.
127 ;CHECK: vld4.16 {d16, d18, d20, d22}, [{{r[0-9]+|lr}}]!
128 ;CHECK: vld4.16 {d17, d19, d21, d23}, [{{r[0-9]+|lr}}]
129 %tmp0 = bitcast i16* %A to i8*
130 %tmp1 = call %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16.p0i8(i8* %tmp0, i32 1)
131 %tmp2 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 0
132 %tmp3 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 2
133 %tmp4 = add <8 x i16> %tmp2, %tmp3
137 ;Check for a post-increment updating load.
138 define <8 x i16> @vld4Qi16_update(i16** %ptr) nounwind {
139 ;CHECK-LABEL: vld4Qi16_update:
140 ;CHECK: vld4.16 {d16, d18, d20, d22}, [{{r[0-9]+|lr}}:64]!
141 ;CHECK: vld4.16 {d17, d19, d21, d23}, [{{r[0-9]+|lr}}:64]!
142 %A = load i16*, i16** %ptr
143 %tmp0 = bitcast i16* %A to i8*
144 %tmp1 = call %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16.p0i8(i8* %tmp0, i32 8)
145 %tmp2 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 0
146 %tmp3 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 2
147 %tmp4 = add <8 x i16> %tmp2, %tmp3
148 %tmp5 = getelementptr i16, i16* %A, i32 32
149 store i16* %tmp5, i16** %ptr
153 define <4 x i32> @vld4Qi32(i32* %A) nounwind {
154 ;CHECK-LABEL: vld4Qi32:
157 %tmp0 = bitcast i32* %A to i8*
158 %tmp1 = call %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4.v4i32.p0i8(i8* %tmp0, i32 1)
159 %tmp2 = extractvalue %struct.__neon_int32x4x4_t %tmp1, 0
160 %tmp3 = extractvalue %struct.__neon_int32x4x4_t %tmp1, 2
161 %tmp4 = add <4 x i32> %tmp2, %tmp3
165 define <4 x float> @vld4Qf(float* %A) nounwind {
166 ;CHECK-LABEL: vld4Qf:
169 %tmp0 = bitcast float* %A to i8*
170 %tmp1 = call %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4.v4f32.p0i8(i8* %tmp0, i32 1)
171 %tmp2 = extractvalue %struct.__neon_float32x4x4_t %tmp1, 0
172 %tmp3 = extractvalue %struct.__neon_float32x4x4_t %tmp1, 2
173 %tmp4 = fadd <4 x float> %tmp2, %tmp3
174 ret <4 x float> %tmp4
177 declare %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8.p0i8(i8*, i32) nounwind readonly
178 declare %struct.__neon_int16x4x4_t @llvm.arm.neon.vld4.v4i16.p0i8(i8*, i32) nounwind readonly
179 declare %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4.v2i32.p0i8(i8*, i32) nounwind readonly
180 declare %struct.__neon_float32x2x4_t @llvm.arm.neon.vld4.v2f32.p0i8(i8*, i32) nounwind readonly
181 declare %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64.p0i8(i8*, i32) nounwind readonly
183 declare %struct.__neon_int8x16x4_t @llvm.arm.neon.vld4.v16i8.p0i8(i8*, i32) nounwind readonly
184 declare %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16.p0i8(i8*, i32) nounwind readonly
185 declare %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4.v4i32.p0i8(i8*, i32) nounwind readonly
186 declare %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4.v4f32.p0i8(i8*, i32) nounwind readonly