1 ; RUN: llc -mtriple=arm-eabi -mattr=+neon %s -o -| FileCheck %s
2 ; RUN: llc -mtriple=arm-eabi -mattr=+neon -regalloc=basic %s -o - | FileCheck %s
4 %struct.__neon_int8x8x3_t = type { <8 x i8>, <8 x i8>, <8 x i8> }
5 %struct.__neon_int16x4x3_t = type { <4 x i16>, <4 x i16>, <4 x i16> }
6 %struct.__neon_int32x2x3_t = type { <2 x i32>, <2 x i32>, <2 x i32> }
7 %struct.__neon_float32x2x3_t = type { <2 x float>, <2 x float>, <2 x float> }
8 %struct.__neon_int64x1x3_t = type { <1 x i64>, <1 x i64>, <1 x i64> }
10 %struct.__neon_int8x16x3_t = type { <16 x i8>, <16 x i8>, <16 x i8> }
11 %struct.__neon_int16x8x3_t = type { <8 x i16>, <8 x i16>, <8 x i16> }
12 %struct.__neon_int32x4x3_t = type { <4 x i32>, <4 x i32>, <4 x i32> }
13 %struct.__neon_float32x4x3_t = type { <4 x float>, <4 x float>, <4 x float> }
15 define <8 x i8> @vld3i8(i8* %A) nounwind {
17 ;Check the alignment value. Max for this instruction is 64 bits:
18 ;CHECK: vld3.8 {d16, d17, d18}, [{{r[0-9]+|lr}}:64]
19 %tmp1 = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8.p0i8(i8* %A, i32 32)
20 %tmp2 = extractvalue %struct.__neon_int8x8x3_t %tmp1, 0
21 %tmp3 = extractvalue %struct.__neon_int8x8x3_t %tmp1, 2
22 %tmp4 = add <8 x i8> %tmp2, %tmp3
26 define <4 x i16> @vld3i16(i16* %A) nounwind {
27 ;CHECK-LABEL: vld3i16:
29 %tmp0 = bitcast i16* %A to i8*
30 %tmp1 = call %struct.__neon_int16x4x3_t @llvm.arm.neon.vld3.v4i16.p0i8(i8* %tmp0, i32 1)
31 %tmp2 = extractvalue %struct.__neon_int16x4x3_t %tmp1, 0
32 %tmp3 = extractvalue %struct.__neon_int16x4x3_t %tmp1, 2
33 %tmp4 = add <4 x i16> %tmp2, %tmp3
37 ;Check for a post-increment updating load with register increment.
38 define <4 x i16> @vld3i16_update(i16** %ptr, i32 %inc) nounwind {
39 ;CHECK-LABEL: vld3i16_update:
40 ;CHECK: vld3.16 {d16, d17, d18}, [{{r[0-9]+|lr}}], {{r[0-9]+|lr}}
41 %A = load i16*, i16** %ptr
42 %tmp0 = bitcast i16* %A to i8*
43 %tmp1 = call %struct.__neon_int16x4x3_t @llvm.arm.neon.vld3.v4i16.p0i8(i8* %tmp0, i32 1)
44 %tmp2 = extractvalue %struct.__neon_int16x4x3_t %tmp1, 0
45 %tmp3 = extractvalue %struct.__neon_int16x4x3_t %tmp1, 2
46 %tmp4 = add <4 x i16> %tmp2, %tmp3
47 %tmp5 = getelementptr i16, i16* %A, i32 %inc
48 store i16* %tmp5, i16** %ptr
52 define <2 x i32> @vld3i32(i32* %A) nounwind {
53 ;CHECK-LABEL: vld3i32:
55 %tmp0 = bitcast i32* %A to i8*
56 %tmp1 = call %struct.__neon_int32x2x3_t @llvm.arm.neon.vld3.v2i32.p0i8(i8* %tmp0, i32 1)
57 %tmp2 = extractvalue %struct.__neon_int32x2x3_t %tmp1, 0
58 %tmp3 = extractvalue %struct.__neon_int32x2x3_t %tmp1, 2
59 %tmp4 = add <2 x i32> %tmp2, %tmp3
63 define <2 x float> @vld3f(float* %A) nounwind {
66 %tmp0 = bitcast float* %A to i8*
67 %tmp1 = call %struct.__neon_float32x2x3_t @llvm.arm.neon.vld3.v2f32.p0i8(i8* %tmp0, i32 1)
68 %tmp2 = extractvalue %struct.__neon_float32x2x3_t %tmp1, 0
69 %tmp3 = extractvalue %struct.__neon_float32x2x3_t %tmp1, 2
70 %tmp4 = fadd <2 x float> %tmp2, %tmp3
74 define <1 x i64> @vld3i64(i64* %A) nounwind {
75 ;CHECK-LABEL: vld3i64:
76 ;Check the alignment value. Max for this instruction is 64 bits:
77 ;CHECK: vld1.64 {d16, d17, d18}, [{{r[0-9]+|lr}}:64]
78 %tmp0 = bitcast i64* %A to i8*
79 %tmp1 = call %struct.__neon_int64x1x3_t @llvm.arm.neon.vld3.v1i64.p0i8(i8* %tmp0, i32 16)
80 %tmp2 = extractvalue %struct.__neon_int64x1x3_t %tmp1, 0
81 %tmp3 = extractvalue %struct.__neon_int64x1x3_t %tmp1, 2
82 %tmp4 = add <1 x i64> %tmp2, %tmp3
86 define <1 x i64> @vld3i64_update(i64** %ptr, i64* %A) nounwind {
87 ;CHECK-LABEL: vld3i64_update:
88 ;CHECK: vld1.64 {d16, d17, d18}, [{{r[0-9]+|lr}}:64]!
89 %tmp0 = bitcast i64* %A to i8*
90 %tmp1 = call %struct.__neon_int64x1x3_t @llvm.arm.neon.vld3.v1i64.p0i8(i8* %tmp0, i32 16)
91 %tmp5 = getelementptr i64, i64* %A, i32 3
92 store i64* %tmp5, i64** %ptr
93 %tmp2 = extractvalue %struct.__neon_int64x1x3_t %tmp1, 0
94 %tmp3 = extractvalue %struct.__neon_int64x1x3_t %tmp1, 2
95 %tmp4 = add <1 x i64> %tmp2, %tmp3
99 define <1 x i64> @vld3i64_reg_update(i64** %ptr, i64* %A) nounwind {
100 ;CHECK-LABEL: vld3i64_reg_update:
101 ;CHECK: vld1.64 {d16, d17, d18}, [{{r[0-9]+|lr}}:64], {{r[0-9]+|lr}}
102 %tmp0 = bitcast i64* %A to i8*
103 %tmp1 = call %struct.__neon_int64x1x3_t @llvm.arm.neon.vld3.v1i64.p0i8(i8* %tmp0, i32 16)
104 %tmp5 = getelementptr i64, i64* %A, i32 1
105 store i64* %tmp5, i64** %ptr
106 %tmp2 = extractvalue %struct.__neon_int64x1x3_t %tmp1, 0
107 %tmp3 = extractvalue %struct.__neon_int64x1x3_t %tmp1, 2
108 %tmp4 = add <1 x i64> %tmp2, %tmp3
112 define <16 x i8> @vld3Qi8(i8* %A) nounwind {
113 ;CHECK-LABEL: vld3Qi8:
114 ;Check the alignment value. Max for this instruction is 64 bits:
115 ;CHECK: vld3.8 {d16, d18, d20}, [{{r[0-9]+|lr}}:64]!
116 ;CHECK: vld3.8 {d17, d19, d21}, [{{r[0-9]+|lr}}:64]
117 %tmp1 = call %struct.__neon_int8x16x3_t @llvm.arm.neon.vld3.v16i8.p0i8(i8* %A, i32 32)
118 %tmp2 = extractvalue %struct.__neon_int8x16x3_t %tmp1, 0
119 %tmp3 = extractvalue %struct.__neon_int8x16x3_t %tmp1, 2
120 %tmp4 = add <16 x i8> %tmp2, %tmp3
124 define <8 x i16> @vld3Qi16(i16* %A) nounwind {
125 ;CHECK-LABEL: vld3Qi16:
128 %tmp0 = bitcast i16* %A to i8*
129 %tmp1 = call %struct.__neon_int16x8x3_t @llvm.arm.neon.vld3.v8i16.p0i8(i8* %tmp0, i32 1)
130 %tmp2 = extractvalue %struct.__neon_int16x8x3_t %tmp1, 0
131 %tmp3 = extractvalue %struct.__neon_int16x8x3_t %tmp1, 2
132 %tmp4 = add <8 x i16> %tmp2, %tmp3
136 define <4 x i32> @vld3Qi32(i32* %A) nounwind {
137 ;CHECK-LABEL: vld3Qi32:
140 %tmp0 = bitcast i32* %A to i8*
141 %tmp1 = call %struct.__neon_int32x4x3_t @llvm.arm.neon.vld3.v4i32.p0i8(i8* %tmp0, i32 1)
142 %tmp2 = extractvalue %struct.__neon_int32x4x3_t %tmp1, 0
143 %tmp3 = extractvalue %struct.__neon_int32x4x3_t %tmp1, 2
144 %tmp4 = add <4 x i32> %tmp2, %tmp3
148 ;Check for a post-increment updating load.
149 define <4 x i32> @vld3Qi32_update(i32** %ptr) nounwind {
150 ;CHECK-LABEL: vld3Qi32_update:
151 ;CHECK: vld3.32 {d16, d18, d20}, {{\[}}[[R:r[0-9]+|lr]]]!
152 ;CHECK: vld3.32 {d17, d19, d21}, {{\[}}[[R]]]!
153 %A = load i32*, i32** %ptr
154 %tmp0 = bitcast i32* %A to i8*
155 %tmp1 = call %struct.__neon_int32x4x3_t @llvm.arm.neon.vld3.v4i32.p0i8(i8* %tmp0, i32 1)
156 %tmp2 = extractvalue %struct.__neon_int32x4x3_t %tmp1, 0
157 %tmp3 = extractvalue %struct.__neon_int32x4x3_t %tmp1, 2
158 %tmp4 = add <4 x i32> %tmp2, %tmp3
159 %tmp5 = getelementptr i32, i32* %A, i32 12
160 store i32* %tmp5, i32** %ptr
164 define <4 x float> @vld3Qf(float* %A) nounwind {
165 ;CHECK-LABEL: vld3Qf:
168 %tmp0 = bitcast float* %A to i8*
169 %tmp1 = call %struct.__neon_float32x4x3_t @llvm.arm.neon.vld3.v4f32.p0i8(i8* %tmp0, i32 1)
170 %tmp2 = extractvalue %struct.__neon_float32x4x3_t %tmp1, 0
171 %tmp3 = extractvalue %struct.__neon_float32x4x3_t %tmp1, 2
172 %tmp4 = fadd <4 x float> %tmp2, %tmp3
173 ret <4 x float> %tmp4
176 declare %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8.p0i8(i8*, i32) nounwind readonly
177 declare %struct.__neon_int16x4x3_t @llvm.arm.neon.vld3.v4i16.p0i8(i8*, i32) nounwind readonly
178 declare %struct.__neon_int32x2x3_t @llvm.arm.neon.vld3.v2i32.p0i8(i8*, i32) nounwind readonly
179 declare %struct.__neon_float32x2x3_t @llvm.arm.neon.vld3.v2f32.p0i8(i8*, i32) nounwind readonly
180 declare %struct.__neon_int64x1x3_t @llvm.arm.neon.vld3.v1i64.p0i8(i8*, i32) nounwind readonly
182 declare %struct.__neon_int8x16x3_t @llvm.arm.neon.vld3.v16i8.p0i8(i8*, i32) nounwind readonly
183 declare %struct.__neon_int16x8x3_t @llvm.arm.neon.vld3.v8i16.p0i8(i8*, i32) nounwind readonly
184 declare %struct.__neon_int32x4x3_t @llvm.arm.neon.vld3.v4i32.p0i8(i8*, i32) nounwind readonly
185 declare %struct.__neon_float32x4x3_t @llvm.arm.neon.vld3.v4f32.p0i8(i8*, i32) nounwind readonly