1 ; RUN: llc -mtriple=arm-eabi -float-abi=soft -mattr=+neon %s -o - | FileCheck %s
3 ; RUN: llc -mtriple=arm-eabi -float-abi=soft -mattr=+neon -regalloc=basic %s -o - \
6 define <8 x i8> @vld1lanei8(i8* %A, <8 x i8>* %B) nounwind {
7 ;CHECK-LABEL: vld1lanei8:
8 ;Check the (default) alignment value.
9 ;CHECK: vld1.8 {d16[3]}, [r0]
10 %tmp1 = load <8 x i8>, <8 x i8>* %B
11 %tmp2 = load i8, i8* %A, align 8
12 %tmp3 = insertelement <8 x i8> %tmp1, i8 %tmp2, i32 3
16 define <4 x i16> @vld1lanei16(i16* %A, <4 x i16>* %B) nounwind {
17 ;CHECK-LABEL: vld1lanei16:
18 ;Check the alignment value. Max for this instruction is 16 bits:
19 ;CHECK: vld1.16 {d16[2]}, [r0:16]
20 %tmp1 = load <4 x i16>, <4 x i16>* %B
21 %tmp2 = load i16, i16* %A, align 8
22 %tmp3 = insertelement <4 x i16> %tmp1, i16 %tmp2, i32 2
26 define <2 x i32> @vld1lanei32(i32* %A, <2 x i32>* %B) nounwind {
27 ;CHECK-LABEL: vld1lanei32:
28 ;Check the alignment value. Max for this instruction is 32 bits:
29 ;CHECK: vld1.32 {d16[1]}, [r0:32]
30 %tmp1 = load <2 x i32>, <2 x i32>* %B
31 %tmp2 = load i32, i32* %A, align 8
32 %tmp3 = insertelement <2 x i32> %tmp1, i32 %tmp2, i32 1
36 define <2 x i32> @vld1lanei32a32(i32* %A, <2 x i32>* %B) nounwind {
37 ;CHECK-LABEL: vld1lanei32a32:
38 ;Check the alignment value. Legal values are none or :32.
39 ;CHECK: vld1.32 {d16[1]}, [r0:32]
40 %tmp1 = load <2 x i32>, <2 x i32>* %B
41 %tmp2 = load i32, i32* %A, align 4
42 %tmp3 = insertelement <2 x i32> %tmp1, i32 %tmp2, i32 1
46 define <2 x float> @vld1lanef(float* %A, <2 x float>* %B) nounwind {
47 ;CHECK-LABEL: vld1lanef:
48 ;CHECK: vld1.32 {d16[1]}, [r0:32]
49 %tmp1 = load <2 x float>, <2 x float>* %B
50 %tmp2 = load float, float* %A, align 4
51 %tmp3 = insertelement <2 x float> %tmp1, float %tmp2, i32 1
55 define <16 x i8> @vld1laneQi8(i8* %A, <16 x i8>* %B) nounwind {
56 ;CHECK-LABEL: vld1laneQi8:
57 ;CHECK: vld1.8 {d17[1]}, [r0]
58 %tmp1 = load <16 x i8>, <16 x i8>* %B
59 %tmp2 = load i8, i8* %A, align 8
60 %tmp3 = insertelement <16 x i8> %tmp1, i8 %tmp2, i32 9
64 define <8 x i16> @vld1laneQi16(i16* %A, <8 x i16>* %B) nounwind {
65 ;CHECK-LABEL: vld1laneQi16:
66 ;CHECK: vld1.16 {d17[1]}, [r0:16]
67 %tmp1 = load <8 x i16>, <8 x i16>* %B
68 %tmp2 = load i16, i16* %A, align 8
69 %tmp3 = insertelement <8 x i16> %tmp1, i16 %tmp2, i32 5
73 define <4 x i32> @vld1laneQi32(i32* %A, <4 x i32>* %B) nounwind {
74 ;CHECK-LABEL: vld1laneQi32:
75 ;CHECK: vld1.32 {d17[1]}, [r0:32]
76 %tmp1 = load <4 x i32>, <4 x i32>* %B
77 %tmp2 = load i32, i32* %A, align 8
78 %tmp3 = insertelement <4 x i32> %tmp1, i32 %tmp2, i32 3
82 define <4 x float> @vld1laneQf(float* %A, <4 x float>* %B) nounwind {
83 ;CHECK-LABEL: vld1laneQf:
84 ;CHECK: vld1.32 {d16[0]}, [r0:32]
85 %tmp1 = load <4 x float>, <4 x float>* %B
86 %tmp2 = load float, float* %A
87 %tmp3 = insertelement <4 x float> %tmp1, float %tmp2, i32 0
91 %struct.__neon_int8x8x2_t = type { <8 x i8>, <8 x i8> }
92 %struct.__neon_int16x4x2_t = type { <4 x i16>, <4 x i16> }
93 %struct.__neon_int32x2x2_t = type { <2 x i32>, <2 x i32> }
94 %struct.__neon_float32x2x2_t = type { <2 x float>, <2 x float> }
96 %struct.__neon_int16x8x2_t = type { <8 x i16>, <8 x i16> }
97 %struct.__neon_int32x4x2_t = type { <4 x i32>, <4 x i32> }
98 %struct.__neon_float32x4x2_t = type { <4 x float>, <4 x float> }
100 define <8 x i8> @vld2lanei8(i8* %A, <8 x i8>* %B) nounwind {
101 ;CHECK-LABEL: vld2lanei8:
102 ;Check the alignment value. Max for this instruction is 16 bits:
103 ;CHECK: vld2.8 {d16[1], d17[1]}, [r0:16]
104 %tmp1 = load <8 x i8>, <8 x i8>* %B
105 %tmp2 = call %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2lane.v8i8.p0i8(i8* %A, <8 x i8> %tmp1, <8 x i8> %tmp1, i32 1, i32 4)
106 %tmp3 = extractvalue %struct.__neon_int8x8x2_t %tmp2, 0
107 %tmp4 = extractvalue %struct.__neon_int8x8x2_t %tmp2, 1
108 %tmp5 = add <8 x i8> %tmp3, %tmp4
112 define <4 x i16> @vld2lanei16(i16* %A, <4 x i16>* %B) nounwind {
113 ;CHECK-LABEL: vld2lanei16:
114 ;Check the alignment value. Max for this instruction is 32 bits:
115 ;CHECK: vld2.16 {d16[1], d17[1]}, [r0:32]
116 %tmp0 = bitcast i16* %A to i8*
117 %tmp1 = load <4 x i16>, <4 x i16>* %B
118 %tmp2 = call %struct.__neon_int16x4x2_t @llvm.arm.neon.vld2lane.v4i16.p0i8(i8* %tmp0, <4 x i16> %tmp1, <4 x i16> %tmp1, i32 1, i32 8)
119 %tmp3 = extractvalue %struct.__neon_int16x4x2_t %tmp2, 0
120 %tmp4 = extractvalue %struct.__neon_int16x4x2_t %tmp2, 1
121 %tmp5 = add <4 x i16> %tmp3, %tmp4
125 define <2 x i32> @vld2lanei32(i32* %A, <2 x i32>* %B) nounwind {
126 ;CHECK-LABEL: vld2lanei32:
128 %tmp0 = bitcast i32* %A to i8*
129 %tmp1 = load <2 x i32>, <2 x i32>* %B
130 %tmp2 = call %struct.__neon_int32x2x2_t @llvm.arm.neon.vld2lane.v2i32.p0i8(i8* %tmp0, <2 x i32> %tmp1, <2 x i32> %tmp1, i32 1, i32 1)
131 %tmp3 = extractvalue %struct.__neon_int32x2x2_t %tmp2, 0
132 %tmp4 = extractvalue %struct.__neon_int32x2x2_t %tmp2, 1
133 %tmp5 = add <2 x i32> %tmp3, %tmp4
137 ;Check for a post-increment updating load.
138 define <2 x i32> @vld2lanei32_update(i32** %ptr, <2 x i32>* %B) nounwind {
139 ;CHECK-LABEL: vld2lanei32_update:
140 ;CHECK: vld2.32 {d16[1], d17[1]}, [{{r[0-9]+}}]!
141 %A = load i32*, i32** %ptr
142 %tmp0 = bitcast i32* %A to i8*
143 %tmp1 = load <2 x i32>, <2 x i32>* %B
144 %tmp2 = call %struct.__neon_int32x2x2_t @llvm.arm.neon.vld2lane.v2i32.p0i8(i8* %tmp0, <2 x i32> %tmp1, <2 x i32> %tmp1, i32 1, i32 1)
145 %tmp3 = extractvalue %struct.__neon_int32x2x2_t %tmp2, 0
146 %tmp4 = extractvalue %struct.__neon_int32x2x2_t %tmp2, 1
147 %tmp5 = add <2 x i32> %tmp3, %tmp4
148 %tmp6 = getelementptr i32, i32* %A, i32 2
149 store i32* %tmp6, i32** %ptr
153 define <2 x i32> @vld2lanei32_odd_update(i32** %ptr, <2 x i32>* %B) nounwind {
154 ;CHECK-LABEL: vld2lanei32_odd_update:
155 ;CHECK: mov [[INC:r[0-9]+]], #12
156 ;CHECK: vld2.32 {d16[1], d17[1]}, [{{r[0-9]+}}], [[INC]]
157 %A = load i32*, i32** %ptr
158 %tmp0 = bitcast i32* %A to i8*
159 %tmp1 = load <2 x i32>, <2 x i32>* %B
160 %tmp2 = call %struct.__neon_int32x2x2_t @llvm.arm.neon.vld2lane.v2i32.p0i8(i8* %tmp0, <2 x i32> %tmp1, <2 x i32> %tmp1, i32 1, i32 1)
161 %tmp3 = extractvalue %struct.__neon_int32x2x2_t %tmp2, 0
162 %tmp4 = extractvalue %struct.__neon_int32x2x2_t %tmp2, 1
163 %tmp5 = add <2 x i32> %tmp3, %tmp4
164 %tmp6 = getelementptr i32, i32* %A, i32 3
165 store i32* %tmp6, i32** %ptr
169 define <2 x float> @vld2lanef(float* %A, <2 x float>* %B) nounwind {
170 ;CHECK-LABEL: vld2lanef:
172 %tmp0 = bitcast float* %A to i8*
173 %tmp1 = load <2 x float>, <2 x float>* %B
174 %tmp2 = call %struct.__neon_float32x2x2_t @llvm.arm.neon.vld2lane.v2f32.p0i8(i8* %tmp0, <2 x float> %tmp1, <2 x float> %tmp1, i32 1, i32 1)
175 %tmp3 = extractvalue %struct.__neon_float32x2x2_t %tmp2, 0
176 %tmp4 = extractvalue %struct.__neon_float32x2x2_t %tmp2, 1
177 %tmp5 = fadd <2 x float> %tmp3, %tmp4
178 ret <2 x float> %tmp5
181 define <8 x i16> @vld2laneQi16(i16* %A, <8 x i16>* %B) nounwind {
182 ;CHECK-LABEL: vld2laneQi16:
183 ;Check the (default) alignment.
184 ;CHECK: vld2.16 {d17[1], d19[1]}, [{{r[0-9]+}}]
185 %tmp0 = bitcast i16* %A to i8*
186 %tmp1 = load <8 x i16>, <8 x i16>* %B
187 %tmp2 = call %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2lane.v8i16.p0i8(i8* %tmp0, <8 x i16> %tmp1, <8 x i16> %tmp1, i32 5, i32 1)
188 %tmp3 = extractvalue %struct.__neon_int16x8x2_t %tmp2, 0
189 %tmp4 = extractvalue %struct.__neon_int16x8x2_t %tmp2, 1
190 %tmp5 = add <8 x i16> %tmp3, %tmp4
194 define <4 x i32> @vld2laneQi32(i32* %A, <4 x i32>* %B) nounwind {
195 ;CHECK-LABEL: vld2laneQi32:
196 ;Check the alignment value. Max for this instruction is 64 bits:
197 ;CHECK: vld2.32 {d17[0], d19[0]}, [{{r[0-9]+}}:64]
198 %tmp0 = bitcast i32* %A to i8*
199 %tmp1 = load <4 x i32>, <4 x i32>* %B
200 %tmp2 = call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2lane.v4i32.p0i8(i8* %tmp0, <4 x i32> %tmp1, <4 x i32> %tmp1, i32 2, i32 16)
201 %tmp3 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 0
202 %tmp4 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 1
203 %tmp5 = add <4 x i32> %tmp3, %tmp4
207 define <4 x float> @vld2laneQf(float* %A, <4 x float>* %B) nounwind {
208 ;CHECK-LABEL: vld2laneQf:
210 %tmp0 = bitcast float* %A to i8*
211 %tmp1 = load <4 x float>, <4 x float>* %B
212 %tmp2 = call %struct.__neon_float32x4x2_t @llvm.arm.neon.vld2lane.v4f32.p0i8(i8* %tmp0, <4 x float> %tmp1, <4 x float> %tmp1, i32 1, i32 1)
213 %tmp3 = extractvalue %struct.__neon_float32x4x2_t %tmp2, 0
214 %tmp4 = extractvalue %struct.__neon_float32x4x2_t %tmp2, 1
215 %tmp5 = fadd <4 x float> %tmp3, %tmp4
216 ret <4 x float> %tmp5
219 declare %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2lane.v8i8.p0i8(i8*, <8 x i8>, <8 x i8>, i32, i32) nounwind readonly
220 declare %struct.__neon_int16x4x2_t @llvm.arm.neon.vld2lane.v4i16.p0i8(i8*, <4 x i16>, <4 x i16>, i32, i32) nounwind readonly
221 declare %struct.__neon_int32x2x2_t @llvm.arm.neon.vld2lane.v2i32.p0i8(i8*, <2 x i32>, <2 x i32>, i32, i32) nounwind readonly
222 declare %struct.__neon_float32x2x2_t @llvm.arm.neon.vld2lane.v2f32.p0i8(i8*, <2 x float>, <2 x float>, i32, i32) nounwind readonly
224 declare %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2lane.v8i16.p0i8(i8*, <8 x i16>, <8 x i16>, i32, i32) nounwind readonly
225 declare %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2lane.v4i32.p0i8(i8*, <4 x i32>, <4 x i32>, i32, i32) nounwind readonly
226 declare %struct.__neon_float32x4x2_t @llvm.arm.neon.vld2lane.v4f32.p0i8(i8*, <4 x float>, <4 x float>, i32, i32) nounwind readonly
228 %struct.__neon_int8x8x3_t = type { <8 x i8>, <8 x i8>, <8 x i8> }
229 %struct.__neon_int16x4x3_t = type { <4 x i16>, <4 x i16>, <4 x i16> }
230 %struct.__neon_int32x2x3_t = type { <2 x i32>, <2 x i32>, <2 x i32> }
231 %struct.__neon_float32x2x3_t = type { <2 x float>, <2 x float>, <2 x float> }
233 %struct.__neon_int16x8x3_t = type { <8 x i16>, <8 x i16>, <8 x i16> }
234 %struct.__neon_int32x4x3_t = type { <4 x i32>, <4 x i32>, <4 x i32> }
235 %struct.__neon_float32x4x3_t = type { <4 x float>, <4 x float>, <4 x float> }
237 define <8 x i8> @vld3lanei8(i8* %A, <8 x i8>* %B) nounwind {
238 ;CHECK-LABEL: vld3lanei8:
240 %tmp1 = load <8 x i8>, <8 x i8>* %B
241 %tmp2 = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3lane.v8i8.p0i8(i8* %A, <8 x i8> %tmp1, <8 x i8> %tmp1, <8 x i8> %tmp1, i32 1, i32 1)
242 %tmp3 = extractvalue %struct.__neon_int8x8x3_t %tmp2, 0
243 %tmp4 = extractvalue %struct.__neon_int8x8x3_t %tmp2, 1
244 %tmp5 = extractvalue %struct.__neon_int8x8x3_t %tmp2, 2
245 %tmp6 = add <8 x i8> %tmp3, %tmp4
246 %tmp7 = add <8 x i8> %tmp5, %tmp6
250 define <4 x i16> @vld3lanei16(i16* %A, <4 x i16>* %B) nounwind {
251 ;CHECK-LABEL: vld3lanei16:
252 ;Check the (default) alignment value. VLD3 does not support alignment.
253 ;CHECK: vld3.16 {d{{.*}}[1], d{{.*}}[1], d{{.*}}[1]}, [{{r[0-9]+}}]
254 %tmp0 = bitcast i16* %A to i8*
255 %tmp1 = load <4 x i16>, <4 x i16>* %B
256 %tmp2 = call %struct.__neon_int16x4x3_t @llvm.arm.neon.vld3lane.v4i16.p0i8(i8* %tmp0, <4 x i16> %tmp1, <4 x i16> %tmp1, <4 x i16> %tmp1, i32 1, i32 8)
257 %tmp3 = extractvalue %struct.__neon_int16x4x3_t %tmp2, 0
258 %tmp4 = extractvalue %struct.__neon_int16x4x3_t %tmp2, 1
259 %tmp5 = extractvalue %struct.__neon_int16x4x3_t %tmp2, 2
260 %tmp6 = add <4 x i16> %tmp3, %tmp4
261 %tmp7 = add <4 x i16> %tmp5, %tmp6
265 define <2 x i32> @vld3lanei32(i32* %A, <2 x i32>* %B) nounwind {
266 ;CHECK-LABEL: vld3lanei32:
268 %tmp0 = bitcast i32* %A to i8*
269 %tmp1 = load <2 x i32>, <2 x i32>* %B
270 %tmp2 = call %struct.__neon_int32x2x3_t @llvm.arm.neon.vld3lane.v2i32.p0i8(i8* %tmp0, <2 x i32> %tmp1, <2 x i32> %tmp1, <2 x i32> %tmp1, i32 1, i32 1)
271 %tmp3 = extractvalue %struct.__neon_int32x2x3_t %tmp2, 0
272 %tmp4 = extractvalue %struct.__neon_int32x2x3_t %tmp2, 1
273 %tmp5 = extractvalue %struct.__neon_int32x2x3_t %tmp2, 2
274 %tmp6 = add <2 x i32> %tmp3, %tmp4
275 %tmp7 = add <2 x i32> %tmp5, %tmp6
279 define <2 x float> @vld3lanef(float* %A, <2 x float>* %B) nounwind {
280 ;CHECK-LABEL: vld3lanef:
282 %tmp0 = bitcast float* %A to i8*
283 %tmp1 = load <2 x float>, <2 x float>* %B
284 %tmp2 = call %struct.__neon_float32x2x3_t @llvm.arm.neon.vld3lane.v2f32.p0i8(i8* %tmp0, <2 x float> %tmp1, <2 x float> %tmp1, <2 x float> %tmp1, i32 1, i32 1)
285 %tmp3 = extractvalue %struct.__neon_float32x2x3_t %tmp2, 0
286 %tmp4 = extractvalue %struct.__neon_float32x2x3_t %tmp2, 1
287 %tmp5 = extractvalue %struct.__neon_float32x2x3_t %tmp2, 2
288 %tmp6 = fadd <2 x float> %tmp3, %tmp4
289 %tmp7 = fadd <2 x float> %tmp5, %tmp6
290 ret <2 x float> %tmp7
293 define <8 x i16> @vld3laneQi16(i16* %A, <8 x i16>* %B) nounwind {
294 ;CHECK-LABEL: vld3laneQi16:
295 ;Check the (default) alignment value. VLD3 does not support alignment.
296 ;CHECK: vld3.16 {d{{.*}}[1], d{{.*}}[1], d{{.*}}[1]}, [{{r[0-9]+}}]
297 %tmp0 = bitcast i16* %A to i8*
298 %tmp1 = load <8 x i16>, <8 x i16>* %B
299 %tmp2 = call %struct.__neon_int16x8x3_t @llvm.arm.neon.vld3lane.v8i16.p0i8(i8* %tmp0, <8 x i16> %tmp1, <8 x i16> %tmp1, <8 x i16> %tmp1, i32 1, i32 8)
300 %tmp3 = extractvalue %struct.__neon_int16x8x3_t %tmp2, 0
301 %tmp4 = extractvalue %struct.__neon_int16x8x3_t %tmp2, 1
302 %tmp5 = extractvalue %struct.__neon_int16x8x3_t %tmp2, 2
303 %tmp6 = add <8 x i16> %tmp3, %tmp4
304 %tmp7 = add <8 x i16> %tmp5, %tmp6
308 ;Check for a post-increment updating load with register increment.
309 define <8 x i16> @vld3laneQi16_update(i16** %ptr, <8 x i16>* %B, i32 %inc) nounwind {
310 ;CHECK-LABEL: vld3laneQi16_update:
311 ;CHECK: vld3.16 {d{{.*}}[1], d{{.*}}[1], d{{.*}}[1]}, [{{r[0-9]+|lr}}], {{r[0-9]+}}
312 %A = load i16*, i16** %ptr
313 %tmp0 = bitcast i16* %A to i8*
314 %tmp1 = load <8 x i16>, <8 x i16>* %B
315 %tmp2 = call %struct.__neon_int16x8x3_t @llvm.arm.neon.vld3lane.v8i16.p0i8(i8* %tmp0, <8 x i16> %tmp1, <8 x i16> %tmp1, <8 x i16> %tmp1, i32 1, i32 8)
316 %tmp3 = extractvalue %struct.__neon_int16x8x3_t %tmp2, 0
317 %tmp4 = extractvalue %struct.__neon_int16x8x3_t %tmp2, 1
318 %tmp5 = extractvalue %struct.__neon_int16x8x3_t %tmp2, 2
319 %tmp6 = add <8 x i16> %tmp3, %tmp4
320 %tmp7 = add <8 x i16> %tmp5, %tmp6
321 %tmp8 = getelementptr i16, i16* %A, i32 %inc
322 store i16* %tmp8, i16** %ptr
326 define <4 x i32> @vld3laneQi32(i32* %A, <4 x i32>* %B) nounwind {
327 ;CHECK-LABEL: vld3laneQi32:
329 %tmp0 = bitcast i32* %A to i8*
330 %tmp1 = load <4 x i32>, <4 x i32>* %B
331 %tmp2 = call %struct.__neon_int32x4x3_t @llvm.arm.neon.vld3lane.v4i32.p0i8(i8* %tmp0, <4 x i32> %tmp1, <4 x i32> %tmp1, <4 x i32> %tmp1, i32 3, i32 1)
332 %tmp3 = extractvalue %struct.__neon_int32x4x3_t %tmp2, 0
333 %tmp4 = extractvalue %struct.__neon_int32x4x3_t %tmp2, 1
334 %tmp5 = extractvalue %struct.__neon_int32x4x3_t %tmp2, 2
335 %tmp6 = add <4 x i32> %tmp3, %tmp4
336 %tmp7 = add <4 x i32> %tmp5, %tmp6
340 define <4 x float> @vld3laneQf(float* %A, <4 x float>* %B) nounwind {
341 ;CHECK-LABEL: vld3laneQf:
343 %tmp0 = bitcast float* %A to i8*
344 %tmp1 = load <4 x float>, <4 x float>* %B
345 %tmp2 = call %struct.__neon_float32x4x3_t @llvm.arm.neon.vld3lane.v4f32.p0i8(i8* %tmp0, <4 x float> %tmp1, <4 x float> %tmp1, <4 x float> %tmp1, i32 1, i32 1)
346 %tmp3 = extractvalue %struct.__neon_float32x4x3_t %tmp2, 0
347 %tmp4 = extractvalue %struct.__neon_float32x4x3_t %tmp2, 1
348 %tmp5 = extractvalue %struct.__neon_float32x4x3_t %tmp2, 2
349 %tmp6 = fadd <4 x float> %tmp3, %tmp4
350 %tmp7 = fadd <4 x float> %tmp5, %tmp6
351 ret <4 x float> %tmp7
354 declare %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3lane.v8i8.p0i8(i8*, <8 x i8>, <8 x i8>, <8 x i8>, i32, i32) nounwind readonly
355 declare %struct.__neon_int16x4x3_t @llvm.arm.neon.vld3lane.v4i16.p0i8(i8*, <4 x i16>, <4 x i16>, <4 x i16>, i32, i32) nounwind readonly
356 declare %struct.__neon_int32x2x3_t @llvm.arm.neon.vld3lane.v2i32.p0i8(i8*, <2 x i32>, <2 x i32>, <2 x i32>, i32, i32) nounwind readonly
357 declare %struct.__neon_float32x2x3_t @llvm.arm.neon.vld3lane.v2f32.p0i8(i8*, <2 x float>, <2 x float>, <2 x float>, i32, i32) nounwind readonly
359 declare %struct.__neon_int16x8x3_t @llvm.arm.neon.vld3lane.v8i16.p0i8(i8*, <8 x i16>, <8 x i16>, <8 x i16>, i32, i32) nounwind readonly
360 declare %struct.__neon_int32x4x3_t @llvm.arm.neon.vld3lane.v4i32.p0i8(i8*, <4 x i32>, <4 x i32>, <4 x i32>, i32, i32) nounwind readonly
361 declare %struct.__neon_float32x4x3_t @llvm.arm.neon.vld3lane.v4f32.p0i8(i8*, <4 x float>, <4 x float>, <4 x float>, i32, i32) nounwind readonly
363 %struct.__neon_int8x8x4_t = type { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> }
364 %struct.__neon_int16x4x4_t = type { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> }
365 %struct.__neon_int32x2x4_t = type { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> }
366 %struct.__neon_float32x2x4_t = type { <2 x float>, <2 x float>, <2 x float>, <2 x float> }
368 %struct.__neon_int16x8x4_t = type { <8 x i16>, <8 x i16>, <8 x i16>, <8 x i16> }
369 %struct.__neon_int32x4x4_t = type { <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32> }
370 %struct.__neon_float32x4x4_t = type { <4 x float>, <4 x float>, <4 x float>, <4 x float> }
372 define <8 x i8> @vld4lanei8(i8* %A, <8 x i8>* %B) nounwind {
373 ;CHECK-LABEL: vld4lanei8:
374 ;Check the alignment value. Max for this instruction is 32 bits:
375 ;CHECK: vld4.8 {d{{.*}}[1], d{{.*}}[1], d{{.*}}[1], d{{.*}}[1]}, [{{r[0-9]+}}:32]
376 %tmp1 = load <8 x i8>, <8 x i8>* %B
377 %tmp2 = call %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4lane.v8i8.p0i8(i8* %A, <8 x i8> %tmp1, <8 x i8> %tmp1, <8 x i8> %tmp1, <8 x i8> %tmp1, i32 1, i32 8)
378 %tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 0
379 %tmp4 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 1
380 %tmp5 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 2
381 %tmp6 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 3
382 %tmp7 = add <8 x i8> %tmp3, %tmp4
383 %tmp8 = add <8 x i8> %tmp5, %tmp6
384 %tmp9 = add <8 x i8> %tmp7, %tmp8
388 ;Check for a post-increment updating load.
389 define <8 x i8> @vld4lanei8_update(i8** %ptr, <8 x i8>* %B) nounwind {
390 ;CHECK-LABEL: vld4lanei8_update:
391 ;CHECK: vld4.8 {d16[1], d17[1], d18[1], d19[1]}, [{{r[0-9]+}}:32]!
392 %A = load i8*, i8** %ptr
393 %tmp1 = load <8 x i8>, <8 x i8>* %B
394 %tmp2 = call %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4lane.v8i8.p0i8(i8* %A, <8 x i8> %tmp1, <8 x i8> %tmp1, <8 x i8> %tmp1, <8 x i8> %tmp1, i32 1, i32 8)
395 %tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 0
396 %tmp4 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 1
397 %tmp5 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 2
398 %tmp6 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 3
399 %tmp7 = add <8 x i8> %tmp3, %tmp4
400 %tmp8 = add <8 x i8> %tmp5, %tmp6
401 %tmp9 = add <8 x i8> %tmp7, %tmp8
402 %tmp10 = getelementptr i8, i8* %A, i32 4
403 store i8* %tmp10, i8** %ptr
407 define <4 x i16> @vld4lanei16(i16* %A, <4 x i16>* %B) nounwind {
408 ;CHECK-LABEL: vld4lanei16:
409 ;Check that a power-of-two alignment smaller than the total size of the memory
410 ;being loaded is ignored.
411 ;CHECK: vld4.16 {d16[1], d17[1], d18[1], d19[1]}, [{{r[0-9]+}}]
412 %tmp0 = bitcast i16* %A to i8*
413 %tmp1 = load <4 x i16>, <4 x i16>* %B
414 %tmp2 = call %struct.__neon_int16x4x4_t @llvm.arm.neon.vld4lane.v4i16.p0i8(i8* %tmp0, <4 x i16> %tmp1, <4 x i16> %tmp1, <4 x i16> %tmp1, <4 x i16> %tmp1, i32 1, i32 4)
415 %tmp3 = extractvalue %struct.__neon_int16x4x4_t %tmp2, 0
416 %tmp4 = extractvalue %struct.__neon_int16x4x4_t %tmp2, 1
417 %tmp5 = extractvalue %struct.__neon_int16x4x4_t %tmp2, 2
418 %tmp6 = extractvalue %struct.__neon_int16x4x4_t %tmp2, 3
419 %tmp7 = add <4 x i16> %tmp3, %tmp4
420 %tmp8 = add <4 x i16> %tmp5, %tmp6
421 %tmp9 = add <4 x i16> %tmp7, %tmp8
425 define <2 x i32> @vld4lanei32(i32* %A, <2 x i32>* %B) nounwind {
426 ;CHECK-LABEL: vld4lanei32:
427 ;Check the alignment value. An 8-byte alignment is allowed here even though
428 ;it is smaller than the total size of the memory being loaded.
429 ;CHECK: vld4.32 {d16[1], d17[1], d18[1], d19[1]}, [{{r[0-9]+}}:64]
430 %tmp0 = bitcast i32* %A to i8*
431 %tmp1 = load <2 x i32>, <2 x i32>* %B
432 %tmp2 = call %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4lane.v2i32.p0i8(i8* %tmp0, <2 x i32> %tmp1, <2 x i32> %tmp1, <2 x i32> %tmp1, <2 x i32> %tmp1, i32 1, i32 8)
433 %tmp3 = extractvalue %struct.__neon_int32x2x4_t %tmp2, 0
434 %tmp4 = extractvalue %struct.__neon_int32x2x4_t %tmp2, 1
435 %tmp5 = extractvalue %struct.__neon_int32x2x4_t %tmp2, 2
436 %tmp6 = extractvalue %struct.__neon_int32x2x4_t %tmp2, 3
437 %tmp7 = add <2 x i32> %tmp3, %tmp4
438 %tmp8 = add <2 x i32> %tmp5, %tmp6
439 %tmp9 = add <2 x i32> %tmp7, %tmp8
443 define <2 x float> @vld4lanef(float* %A, <2 x float>* %B) nounwind {
444 ;CHECK-LABEL: vld4lanef:
446 %tmp0 = bitcast float* %A to i8*
447 %tmp1 = load <2 x float>, <2 x float>* %B
448 %tmp2 = call %struct.__neon_float32x2x4_t @llvm.arm.neon.vld4lane.v2f32.p0i8(i8* %tmp0, <2 x float> %tmp1, <2 x float> %tmp1, <2 x float> %tmp1, <2 x float> %tmp1, i32 1, i32 1)
449 %tmp3 = extractvalue %struct.__neon_float32x2x4_t %tmp2, 0
450 %tmp4 = extractvalue %struct.__neon_float32x2x4_t %tmp2, 1
451 %tmp5 = extractvalue %struct.__neon_float32x2x4_t %tmp2, 2
452 %tmp6 = extractvalue %struct.__neon_float32x2x4_t %tmp2, 3
453 %tmp7 = fadd <2 x float> %tmp3, %tmp4
454 %tmp8 = fadd <2 x float> %tmp5, %tmp6
455 %tmp9 = fadd <2 x float> %tmp7, %tmp8
456 ret <2 x float> %tmp9
459 define <8 x i16> @vld4laneQi16(i16* %A, <8 x i16>* %B) nounwind {
460 ;CHECK-LABEL: vld4laneQi16:
461 ;Check the alignment value. Max for this instruction is 64 bits:
462 ;CHECK: vld4.16 {d16[1], d18[1], d20[1], d22[1]}, [{{r[0-9]+}}:64]
463 %tmp0 = bitcast i16* %A to i8*
464 %tmp1 = load <8 x i16>, <8 x i16>* %B
465 %tmp2 = call %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4lane.v8i16.p0i8(i8* %tmp0, <8 x i16> %tmp1, <8 x i16> %tmp1, <8 x i16> %tmp1, <8 x i16> %tmp1, i32 1, i32 16)
466 %tmp3 = extractvalue %struct.__neon_int16x8x4_t %tmp2, 0
467 %tmp4 = extractvalue %struct.__neon_int16x8x4_t %tmp2, 1
468 %tmp5 = extractvalue %struct.__neon_int16x8x4_t %tmp2, 2
469 %tmp6 = extractvalue %struct.__neon_int16x8x4_t %tmp2, 3
470 %tmp7 = add <8 x i16> %tmp3, %tmp4
471 %tmp8 = add <8 x i16> %tmp5, %tmp6
472 %tmp9 = add <8 x i16> %tmp7, %tmp8
476 define <4 x i32> @vld4laneQi32(i32* %A, <4 x i32>* %B) nounwind {
477 ;CHECK-LABEL: vld4laneQi32:
478 ;Check the (default) alignment.
479 ;CHECK: vld4.32 {d17[0], d19[0], d21[0], d23[0]}, [{{r[0-9]+}}]
480 %tmp0 = bitcast i32* %A to i8*
481 %tmp1 = load <4 x i32>, <4 x i32>* %B
482 %tmp2 = call %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4lane.v4i32.p0i8(i8* %tmp0, <4 x i32> %tmp1, <4 x i32> %tmp1, <4 x i32> %tmp1, <4 x i32> %tmp1, i32 2, i32 1)
483 %tmp3 = extractvalue %struct.__neon_int32x4x4_t %tmp2, 0
484 %tmp4 = extractvalue %struct.__neon_int32x4x4_t %tmp2, 1
485 %tmp5 = extractvalue %struct.__neon_int32x4x4_t %tmp2, 2
486 %tmp6 = extractvalue %struct.__neon_int32x4x4_t %tmp2, 3
487 %tmp7 = add <4 x i32> %tmp3, %tmp4
488 %tmp8 = add <4 x i32> %tmp5, %tmp6
489 %tmp9 = add <4 x i32> %tmp7, %tmp8
493 define <4 x float> @vld4laneQf(float* %A, <4 x float>* %B) nounwind {
494 ;CHECK-LABEL: vld4laneQf:
496 %tmp0 = bitcast float* %A to i8*
497 %tmp1 = load <4 x float>, <4 x float>* %B
498 %tmp2 = call %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4lane.v4f32.p0i8(i8* %tmp0, <4 x float> %tmp1, <4 x float> %tmp1, <4 x float> %tmp1, <4 x float> %tmp1, i32 1, i32 1)
499 %tmp3 = extractvalue %struct.__neon_float32x4x4_t %tmp2, 0
500 %tmp4 = extractvalue %struct.__neon_float32x4x4_t %tmp2, 1
501 %tmp5 = extractvalue %struct.__neon_float32x4x4_t %tmp2, 2
502 %tmp6 = extractvalue %struct.__neon_float32x4x4_t %tmp2, 3
503 %tmp7 = fadd <4 x float> %tmp3, %tmp4
504 %tmp8 = fadd <4 x float> %tmp5, %tmp6
505 %tmp9 = fadd <4 x float> %tmp7, %tmp8
506 ret <4 x float> %tmp9
509 declare %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4lane.v8i8.p0i8(i8*, <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8>, i32, i32) nounwind readonly
510 declare %struct.__neon_int16x4x4_t @llvm.arm.neon.vld4lane.v4i16.p0i8(i8*, <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16>, i32, i32) nounwind readonly
511 declare %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4lane.v2i32.p0i8(i8*, <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32>, i32, i32) nounwind readonly
512 declare %struct.__neon_float32x2x4_t @llvm.arm.neon.vld4lane.v2f32.p0i8(i8*, <2 x float>, <2 x float>, <2 x float>, <2 x float>, i32, i32) nounwind readonly
514 declare %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4lane.v8i16.p0i8(i8*, <8 x i16>, <8 x i16>, <8 x i16>, <8 x i16>, i32, i32) nounwind readonly
515 declare %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4lane.v4i32.p0i8(i8*, <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, i32, i32) nounwind readonly
516 declare %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4lane.v4f32.p0i8(i8*, <4 x float>, <4 x float>, <4 x float>, <4 x float>, i32, i32) nounwind readonly
518 ; Radar 8776599: If one of the operands to a QQQQ REG_SEQUENCE is a register
519 ; in the QPR_VFP2 regclass, it needs to be copied to a QPR regclass because
520 ; we don't currently have a QQQQ_VFP2 super-regclass. (The "0" for the low
521 ; part of %ins67 is supposed to be loaded by a VLDRS instruction in this test.)
522 define <8 x i16> @test_qqqq_regsequence_subreg([6 x i64] %b) nounwind {
523 ;CHECK-LABEL: test_qqqq_regsequence_subreg:
525 %tmp63 = extractvalue [6 x i64] %b, 5
526 %tmp64 = zext i64 %tmp63 to i128
527 %tmp65 = shl i128 %tmp64, 64
528 %ins67 = or i128 %tmp65, 0
529 %tmp78 = bitcast i128 %ins67 to <8 x i16>
530 %vld3_lane = tail call %struct.__neon_int16x8x3_t @llvm.arm.neon.vld3lane.v8i16.p0i8(i8* undef, <8 x i16> undef, <8 x i16> undef, <8 x i16> %tmp78, i32 1, i32 2)
531 %tmp3 = extractvalue %struct.__neon_int16x8x3_t %vld3_lane, 0
532 %tmp4 = extractvalue %struct.__neon_int16x8x3_t %vld3_lane, 1
533 %tmp5 = extractvalue %struct.__neon_int16x8x3_t %vld3_lane, 2
534 %tmp6 = add <8 x i16> %tmp3, %tmp4
535 %tmp7 = add <8 x i16> %tmp5, %tmp6
539 declare void @llvm.trap() nounwind