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 float> @vld2lanef(float* %A, <2 x float>* %B) nounwind {
154 ;CHECK-LABEL: vld2lanef:
156 %tmp0 = bitcast float* %A to i8*
157 %tmp1 = load <2 x float>, <2 x float>* %B
158 %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)
159 %tmp3 = extractvalue %struct.__neon_float32x2x2_t %tmp2, 0
160 %tmp4 = extractvalue %struct.__neon_float32x2x2_t %tmp2, 1
161 %tmp5 = fadd <2 x float> %tmp3, %tmp4
162 ret <2 x float> %tmp5
165 define <8 x i16> @vld2laneQi16(i16* %A, <8 x i16>* %B) nounwind {
166 ;CHECK-LABEL: vld2laneQi16:
167 ;Check the (default) alignment.
168 ;CHECK: vld2.16 {d17[1], d19[1]}, [{{r[0-9]+}}]
169 %tmp0 = bitcast i16* %A to i8*
170 %tmp1 = load <8 x i16>, <8 x i16>* %B
171 %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)
172 %tmp3 = extractvalue %struct.__neon_int16x8x2_t %tmp2, 0
173 %tmp4 = extractvalue %struct.__neon_int16x8x2_t %tmp2, 1
174 %tmp5 = add <8 x i16> %tmp3, %tmp4
178 define <4 x i32> @vld2laneQi32(i32* %A, <4 x i32>* %B) nounwind {
179 ;CHECK-LABEL: vld2laneQi32:
180 ;Check the alignment value. Max for this instruction is 64 bits:
181 ;CHECK: vld2.32 {d17[0], d19[0]}, [{{r[0-9]+}}:64]
182 %tmp0 = bitcast i32* %A to i8*
183 %tmp1 = load <4 x i32>, <4 x i32>* %B
184 %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)
185 %tmp3 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 0
186 %tmp4 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 1
187 %tmp5 = add <4 x i32> %tmp3, %tmp4
191 define <4 x float> @vld2laneQf(float* %A, <4 x float>* %B) nounwind {
192 ;CHECK-LABEL: vld2laneQf:
194 %tmp0 = bitcast float* %A to i8*
195 %tmp1 = load <4 x float>, <4 x float>* %B
196 %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)
197 %tmp3 = extractvalue %struct.__neon_float32x4x2_t %tmp2, 0
198 %tmp4 = extractvalue %struct.__neon_float32x4x2_t %tmp2, 1
199 %tmp5 = fadd <4 x float> %tmp3, %tmp4
200 ret <4 x float> %tmp5
203 declare %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2lane.v8i8.p0i8(i8*, <8 x i8>, <8 x i8>, i32, i32) nounwind readonly
204 declare %struct.__neon_int16x4x2_t @llvm.arm.neon.vld2lane.v4i16.p0i8(i8*, <4 x i16>, <4 x i16>, i32, i32) nounwind readonly
205 declare %struct.__neon_int32x2x2_t @llvm.arm.neon.vld2lane.v2i32.p0i8(i8*, <2 x i32>, <2 x i32>, i32, i32) nounwind readonly
206 declare %struct.__neon_float32x2x2_t @llvm.arm.neon.vld2lane.v2f32.p0i8(i8*, <2 x float>, <2 x float>, i32, i32) nounwind readonly
208 declare %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2lane.v8i16.p0i8(i8*, <8 x i16>, <8 x i16>, i32, i32) nounwind readonly
209 declare %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2lane.v4i32.p0i8(i8*, <4 x i32>, <4 x i32>, i32, i32) nounwind readonly
210 declare %struct.__neon_float32x4x2_t @llvm.arm.neon.vld2lane.v4f32.p0i8(i8*, <4 x float>, <4 x float>, i32, i32) nounwind readonly
212 %struct.__neon_int8x8x3_t = type { <8 x i8>, <8 x i8>, <8 x i8> }
213 %struct.__neon_int16x4x3_t = type { <4 x i16>, <4 x i16>, <4 x i16> }
214 %struct.__neon_int32x2x3_t = type { <2 x i32>, <2 x i32>, <2 x i32> }
215 %struct.__neon_float32x2x3_t = type { <2 x float>, <2 x float>, <2 x float> }
217 %struct.__neon_int16x8x3_t = type { <8 x i16>, <8 x i16>, <8 x i16> }
218 %struct.__neon_int32x4x3_t = type { <4 x i32>, <4 x i32>, <4 x i32> }
219 %struct.__neon_float32x4x3_t = type { <4 x float>, <4 x float>, <4 x float> }
221 define <8 x i8> @vld3lanei8(i8* %A, <8 x i8>* %B) nounwind {
222 ;CHECK-LABEL: vld3lanei8:
224 %tmp1 = load <8 x i8>, <8 x i8>* %B
225 %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)
226 %tmp3 = extractvalue %struct.__neon_int8x8x3_t %tmp2, 0
227 %tmp4 = extractvalue %struct.__neon_int8x8x3_t %tmp2, 1
228 %tmp5 = extractvalue %struct.__neon_int8x8x3_t %tmp2, 2
229 %tmp6 = add <8 x i8> %tmp3, %tmp4
230 %tmp7 = add <8 x i8> %tmp5, %tmp6
234 define <4 x i16> @vld3lanei16(i16* %A, <4 x i16>* %B) nounwind {
235 ;CHECK-LABEL: vld3lanei16:
236 ;Check the (default) alignment value. VLD3 does not support alignment.
237 ;CHECK: vld3.16 {d{{.*}}[1], d{{.*}}[1], d{{.*}}[1]}, [{{r[0-9]+}}]
238 %tmp0 = bitcast i16* %A to i8*
239 %tmp1 = load <4 x i16>, <4 x i16>* %B
240 %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)
241 %tmp3 = extractvalue %struct.__neon_int16x4x3_t %tmp2, 0
242 %tmp4 = extractvalue %struct.__neon_int16x4x3_t %tmp2, 1
243 %tmp5 = extractvalue %struct.__neon_int16x4x3_t %tmp2, 2
244 %tmp6 = add <4 x i16> %tmp3, %tmp4
245 %tmp7 = add <4 x i16> %tmp5, %tmp6
249 define <2 x i32> @vld3lanei32(i32* %A, <2 x i32>* %B) nounwind {
250 ;CHECK-LABEL: vld3lanei32:
252 %tmp0 = bitcast i32* %A to i8*
253 %tmp1 = load <2 x i32>, <2 x i32>* %B
254 %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)
255 %tmp3 = extractvalue %struct.__neon_int32x2x3_t %tmp2, 0
256 %tmp4 = extractvalue %struct.__neon_int32x2x3_t %tmp2, 1
257 %tmp5 = extractvalue %struct.__neon_int32x2x3_t %tmp2, 2
258 %tmp6 = add <2 x i32> %tmp3, %tmp4
259 %tmp7 = add <2 x i32> %tmp5, %tmp6
263 define <2 x float> @vld3lanef(float* %A, <2 x float>* %B) nounwind {
264 ;CHECK-LABEL: vld3lanef:
266 %tmp0 = bitcast float* %A to i8*
267 %tmp1 = load <2 x float>, <2 x float>* %B
268 %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)
269 %tmp3 = extractvalue %struct.__neon_float32x2x3_t %tmp2, 0
270 %tmp4 = extractvalue %struct.__neon_float32x2x3_t %tmp2, 1
271 %tmp5 = extractvalue %struct.__neon_float32x2x3_t %tmp2, 2
272 %tmp6 = fadd <2 x float> %tmp3, %tmp4
273 %tmp7 = fadd <2 x float> %tmp5, %tmp6
274 ret <2 x float> %tmp7
277 define <8 x i16> @vld3laneQi16(i16* %A, <8 x i16>* %B) nounwind {
278 ;CHECK-LABEL: vld3laneQi16:
279 ;Check the (default) alignment value. VLD3 does not support alignment.
280 ;CHECK: vld3.16 {d{{.*}}[1], d{{.*}}[1], d{{.*}}[1]}, [{{r[0-9]+}}]
281 %tmp0 = bitcast i16* %A to i8*
282 %tmp1 = load <8 x i16>, <8 x i16>* %B
283 %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)
284 %tmp3 = extractvalue %struct.__neon_int16x8x3_t %tmp2, 0
285 %tmp4 = extractvalue %struct.__neon_int16x8x3_t %tmp2, 1
286 %tmp5 = extractvalue %struct.__neon_int16x8x3_t %tmp2, 2
287 %tmp6 = add <8 x i16> %tmp3, %tmp4
288 %tmp7 = add <8 x i16> %tmp5, %tmp6
292 ;Check for a post-increment updating load with register increment.
293 define <8 x i16> @vld3laneQi16_update(i16** %ptr, <8 x i16>* %B, i32 %inc) nounwind {
294 ;CHECK-LABEL: vld3laneQi16_update:
295 ;CHECK: vld3.16 {d{{.*}}[1], d{{.*}}[1], d{{.*}}[1]}, [{{r[0-9]+}}], {{r[0-9]+}}
296 %A = load i16*, i16** %ptr
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
305 %tmp8 = getelementptr i16, i16* %A, i32 %inc
306 store i16* %tmp8, i16** %ptr
310 define <4 x i32> @vld3laneQi32(i32* %A, <4 x i32>* %B) nounwind {
311 ;CHECK-LABEL: vld3laneQi32:
313 %tmp0 = bitcast i32* %A to i8*
314 %tmp1 = load <4 x i32>, <4 x i32>* %B
315 %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)
316 %tmp3 = extractvalue %struct.__neon_int32x4x3_t %tmp2, 0
317 %tmp4 = extractvalue %struct.__neon_int32x4x3_t %tmp2, 1
318 %tmp5 = extractvalue %struct.__neon_int32x4x3_t %tmp2, 2
319 %tmp6 = add <4 x i32> %tmp3, %tmp4
320 %tmp7 = add <4 x i32> %tmp5, %tmp6
324 define <4 x float> @vld3laneQf(float* %A, <4 x float>* %B) nounwind {
325 ;CHECK-LABEL: vld3laneQf:
327 %tmp0 = bitcast float* %A to i8*
328 %tmp1 = load <4 x float>, <4 x float>* %B
329 %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)
330 %tmp3 = extractvalue %struct.__neon_float32x4x3_t %tmp2, 0
331 %tmp4 = extractvalue %struct.__neon_float32x4x3_t %tmp2, 1
332 %tmp5 = extractvalue %struct.__neon_float32x4x3_t %tmp2, 2
333 %tmp6 = fadd <4 x float> %tmp3, %tmp4
334 %tmp7 = fadd <4 x float> %tmp5, %tmp6
335 ret <4 x float> %tmp7
338 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
339 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
340 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
341 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
343 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
344 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
345 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
347 %struct.__neon_int8x8x4_t = type { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> }
348 %struct.__neon_int16x4x4_t = type { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> }
349 %struct.__neon_int32x2x4_t = type { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> }
350 %struct.__neon_float32x2x4_t = type { <2 x float>, <2 x float>, <2 x float>, <2 x float> }
352 %struct.__neon_int16x8x4_t = type { <8 x i16>, <8 x i16>, <8 x i16>, <8 x i16> }
353 %struct.__neon_int32x4x4_t = type { <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32> }
354 %struct.__neon_float32x4x4_t = type { <4 x float>, <4 x float>, <4 x float>, <4 x float> }
356 define <8 x i8> @vld4lanei8(i8* %A, <8 x i8>* %B) nounwind {
357 ;CHECK-LABEL: vld4lanei8:
358 ;Check the alignment value. Max for this instruction is 32 bits:
359 ;CHECK: vld4.8 {d{{.*}}[1], d{{.*}}[1], d{{.*}}[1], d{{.*}}[1]}, [{{r[0-9]+}}:32]
360 %tmp1 = load <8 x i8>, <8 x i8>* %B
361 %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)
362 %tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 0
363 %tmp4 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 1
364 %tmp5 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 2
365 %tmp6 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 3
366 %tmp7 = add <8 x i8> %tmp3, %tmp4
367 %tmp8 = add <8 x i8> %tmp5, %tmp6
368 %tmp9 = add <8 x i8> %tmp7, %tmp8
372 ;Check for a post-increment updating load.
373 define <8 x i8> @vld4lanei8_update(i8** %ptr, <8 x i8>* %B) nounwind {
374 ;CHECK-LABEL: vld4lanei8_update:
375 ;CHECK: vld4.8 {d16[1], d17[1], d18[1], d19[1]}, [{{r[0-9]+}}:32]!
376 %A = load i8*, i8** %ptr
377 %tmp1 = load <8 x i8>, <8 x i8>* %B
378 %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)
379 %tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 0
380 %tmp4 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 1
381 %tmp5 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 2
382 %tmp6 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 3
383 %tmp7 = add <8 x i8> %tmp3, %tmp4
384 %tmp8 = add <8 x i8> %tmp5, %tmp6
385 %tmp9 = add <8 x i8> %tmp7, %tmp8
386 %tmp10 = getelementptr i8, i8* %A, i32 4
387 store i8* %tmp10, i8** %ptr
391 define <4 x i16> @vld4lanei16(i16* %A, <4 x i16>* %B) nounwind {
392 ;CHECK-LABEL: vld4lanei16:
393 ;Check that a power-of-two alignment smaller than the total size of the memory
394 ;being loaded is ignored.
395 ;CHECK: vld4.16 {d16[1], d17[1], d18[1], d19[1]}, [{{r[0-9]+}}]
396 %tmp0 = bitcast i16* %A to i8*
397 %tmp1 = load <4 x i16>, <4 x i16>* %B
398 %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)
399 %tmp3 = extractvalue %struct.__neon_int16x4x4_t %tmp2, 0
400 %tmp4 = extractvalue %struct.__neon_int16x4x4_t %tmp2, 1
401 %tmp5 = extractvalue %struct.__neon_int16x4x4_t %tmp2, 2
402 %tmp6 = extractvalue %struct.__neon_int16x4x4_t %tmp2, 3
403 %tmp7 = add <4 x i16> %tmp3, %tmp4
404 %tmp8 = add <4 x i16> %tmp5, %tmp6
405 %tmp9 = add <4 x i16> %tmp7, %tmp8
409 define <2 x i32> @vld4lanei32(i32* %A, <2 x i32>* %B) nounwind {
410 ;CHECK-LABEL: vld4lanei32:
411 ;Check the alignment value. An 8-byte alignment is allowed here even though
412 ;it is smaller than the total size of the memory being loaded.
413 ;CHECK: vld4.32 {d16[1], d17[1], d18[1], d19[1]}, [{{r[0-9]+}}:64]
414 %tmp0 = bitcast i32* %A to i8*
415 %tmp1 = load <2 x i32>, <2 x i32>* %B
416 %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)
417 %tmp3 = extractvalue %struct.__neon_int32x2x4_t %tmp2, 0
418 %tmp4 = extractvalue %struct.__neon_int32x2x4_t %tmp2, 1
419 %tmp5 = extractvalue %struct.__neon_int32x2x4_t %tmp2, 2
420 %tmp6 = extractvalue %struct.__neon_int32x2x4_t %tmp2, 3
421 %tmp7 = add <2 x i32> %tmp3, %tmp4
422 %tmp8 = add <2 x i32> %tmp5, %tmp6
423 %tmp9 = add <2 x i32> %tmp7, %tmp8
427 define <2 x float> @vld4lanef(float* %A, <2 x float>* %B) nounwind {
428 ;CHECK-LABEL: vld4lanef:
430 %tmp0 = bitcast float* %A to i8*
431 %tmp1 = load <2 x float>, <2 x float>* %B
432 %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)
433 %tmp3 = extractvalue %struct.__neon_float32x2x4_t %tmp2, 0
434 %tmp4 = extractvalue %struct.__neon_float32x2x4_t %tmp2, 1
435 %tmp5 = extractvalue %struct.__neon_float32x2x4_t %tmp2, 2
436 %tmp6 = extractvalue %struct.__neon_float32x2x4_t %tmp2, 3
437 %tmp7 = fadd <2 x float> %tmp3, %tmp4
438 %tmp8 = fadd <2 x float> %tmp5, %tmp6
439 %tmp9 = fadd <2 x float> %tmp7, %tmp8
440 ret <2 x float> %tmp9
443 define <8 x i16> @vld4laneQi16(i16* %A, <8 x i16>* %B) nounwind {
444 ;CHECK-LABEL: vld4laneQi16:
445 ;Check the alignment value. Max for this instruction is 64 bits:
446 ;CHECK: vld4.16 {d16[1], d18[1], d20[1], d22[1]}, [{{r[0-9]+}}:64]
447 %tmp0 = bitcast i16* %A to i8*
448 %tmp1 = load <8 x i16>, <8 x i16>* %B
449 %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)
450 %tmp3 = extractvalue %struct.__neon_int16x8x4_t %tmp2, 0
451 %tmp4 = extractvalue %struct.__neon_int16x8x4_t %tmp2, 1
452 %tmp5 = extractvalue %struct.__neon_int16x8x4_t %tmp2, 2
453 %tmp6 = extractvalue %struct.__neon_int16x8x4_t %tmp2, 3
454 %tmp7 = add <8 x i16> %tmp3, %tmp4
455 %tmp8 = add <8 x i16> %tmp5, %tmp6
456 %tmp9 = add <8 x i16> %tmp7, %tmp8
460 define <4 x i32> @vld4laneQi32(i32* %A, <4 x i32>* %B) nounwind {
461 ;CHECK-LABEL: vld4laneQi32:
462 ;Check the (default) alignment.
463 ;CHECK: vld4.32 {d17[0], d19[0], d21[0], d23[0]}, [{{r[0-9]+}}]
464 %tmp0 = bitcast i32* %A to i8*
465 %tmp1 = load <4 x i32>, <4 x i32>* %B
466 %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)
467 %tmp3 = extractvalue %struct.__neon_int32x4x4_t %tmp2, 0
468 %tmp4 = extractvalue %struct.__neon_int32x4x4_t %tmp2, 1
469 %tmp5 = extractvalue %struct.__neon_int32x4x4_t %tmp2, 2
470 %tmp6 = extractvalue %struct.__neon_int32x4x4_t %tmp2, 3
471 %tmp7 = add <4 x i32> %tmp3, %tmp4
472 %tmp8 = add <4 x i32> %tmp5, %tmp6
473 %tmp9 = add <4 x i32> %tmp7, %tmp8
477 define <4 x float> @vld4laneQf(float* %A, <4 x float>* %B) nounwind {
478 ;CHECK-LABEL: vld4laneQf:
480 %tmp0 = bitcast float* %A to i8*
481 %tmp1 = load <4 x float>, <4 x float>* %B
482 %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)
483 %tmp3 = extractvalue %struct.__neon_float32x4x4_t %tmp2, 0
484 %tmp4 = extractvalue %struct.__neon_float32x4x4_t %tmp2, 1
485 %tmp5 = extractvalue %struct.__neon_float32x4x4_t %tmp2, 2
486 %tmp6 = extractvalue %struct.__neon_float32x4x4_t %tmp2, 3
487 %tmp7 = fadd <4 x float> %tmp3, %tmp4
488 %tmp8 = fadd <4 x float> %tmp5, %tmp6
489 %tmp9 = fadd <4 x float> %tmp7, %tmp8
490 ret <4 x float> %tmp9
493 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
494 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
495 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
496 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
498 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
499 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
500 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
502 ; Radar 8776599: If one of the operands to a QQQQ REG_SEQUENCE is a register
503 ; in the QPR_VFP2 regclass, it needs to be copied to a QPR regclass because
504 ; we don't currently have a QQQQ_VFP2 super-regclass. (The "0" for the low
505 ; part of %ins67 is supposed to be loaded by a VLDRS instruction in this test.)
506 define <8 x i16> @test_qqqq_regsequence_subreg([6 x i64] %b) nounwind {
507 ;CHECK-LABEL: test_qqqq_regsequence_subreg:
509 %tmp63 = extractvalue [6 x i64] %b, 5
510 %tmp64 = zext i64 %tmp63 to i128
511 %tmp65 = shl i128 %tmp64, 64
512 %ins67 = or i128 %tmp65, 0
513 %tmp78 = bitcast i128 %ins67 to <8 x i16>
514 %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)
515 %tmp3 = extractvalue %struct.__neon_int16x8x3_t %vld3_lane, 0
516 %tmp4 = extractvalue %struct.__neon_int16x8x3_t %vld3_lane, 1
517 %tmp5 = extractvalue %struct.__neon_int16x8x3_t %vld3_lane, 2
518 %tmp6 = add <8 x i16> %tmp3, %tmp4
519 %tmp7 = add <8 x i16> %tmp5, %tmp6
523 declare void @llvm.trap() nounwind