1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve.fp -enable-arm-maskedldst %s -o - | FileCheck %s
3
4; i32
5
6; Expand
7define arm_aapcs_vfpcc void @ptr_v2i32(<2 x i32> %v, <2 x i32*>* %offptr) {
8; CHECK-LABEL: ptr_v2i32:
9; CHECK:       @ %bb.0: @ %entry
10; CHECK-NEXT:    vmov r2, s0
11; CHECK-NEXT:    ldrd r1, r0, [r0]
12; CHECK-NEXT:    str r2, [r1]
13; CHECK-NEXT:    vmov r1, s2
14; CHECK-NEXT:    str r1, [r0]
15; CHECK-NEXT:    bx lr
16entry:
17  %offs = load <2 x i32*>, <2 x i32*>* %offptr, align 4
18  call void @llvm.masked.scatter.v2i32.v2p0i32(<2 x i32> %v, <2 x i32*> %offs, i32 4, <2 x i1> <i1 true, i1 true>)
19  ret void
20}
21
22; VSTRW.32 Qd, [offs, 0]
23define arm_aapcs_vfpcc void @ptr_v4i32(<4 x i32> %v, <4 x i32*>* %offptr) {
24; CHECK-LABEL: ptr_v4i32:
25; CHECK:       @ %bb.0: @ %entry
26; CHECK-NEXT:    vldrw.u32 q1, [r0]
27; CHECK-NEXT:    vstrw.32 q0, [q1]
28; CHECK-NEXT:    bx lr
29entry:
30  %offs = load <4 x i32*>, <4 x i32*>* %offptr, align 4
31  call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %v, <4 x i32*> %offs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
32  ret void
33}
34
35; Expand
36define arm_aapcs_vfpcc void @ptr_v8i32(<8 x i32> %v, <8 x i32*>* %offptr) {
37; CHECK-LABEL: ptr_v8i32:
38; CHECK:       @ %bb.0: @ %entry
39; CHECK-NEXT:    vldrw.u32 q3, [r0]
40; CHECK-NEXT:    vldrw.u32 q2, [r0, #16]
41; CHECK-NEXT:    vmov r1, s0
42; CHECK-NEXT:    vmov r0, s12
43; CHECK-NEXT:    str r1, [r0]
44; CHECK-NEXT:    vmov r0, s13
45; CHECK-NEXT:    vmov r1, s1
46; CHECK-NEXT:    str r1, [r0]
47; CHECK-NEXT:    vmov r0, s14
48; CHECK-NEXT:    vmov r1, s2
49; CHECK-NEXT:    str r1, [r0]
50; CHECK-NEXT:    vmov r0, s15
51; CHECK-NEXT:    vmov r1, s3
52; CHECK-NEXT:    str r1, [r0]
53; CHECK-NEXT:    vmov r0, s8
54; CHECK-NEXT:    vmov r1, s4
55; CHECK-NEXT:    str r1, [r0]
56; CHECK-NEXT:    vmov r0, s9
57; CHECK-NEXT:    vmov r1, s5
58; CHECK-NEXT:    str r1, [r0]
59; CHECK-NEXT:    vmov r0, s10
60; CHECK-NEXT:    vmov r1, s6
61; CHECK-NEXT:    str r1, [r0]
62; CHECK-NEXT:    vmov r0, s11
63; CHECK-NEXT:    vmov r1, s7
64; CHECK-NEXT:    str r1, [r0]
65; CHECK-NEXT:    bx lr
66entry:
67  %offs = load <8 x i32*>, <8 x i32*>* %offptr, align 4
68  call void @llvm.masked.scatter.v8i32.v8p0i32(<8 x i32> %v, <8 x i32*> %offs, i32 4, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
69  ret void
70}
71
72; Expand
73define arm_aapcs_vfpcc void @ptr_v16i32(<16 x i32> %v, <16 x i32*>* %offptr) {
74; CHECK-LABEL: ptr_v16i32:
75; CHECK:       @ %bb.0: @ %entry
76; CHECK-NEXT:    .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
77; CHECK-NEXT:    vpush {d8, d9, d10, d11, d12, d13, d14, d15}
78; CHECK-NEXT:    vldrw.u32 q7, [r0]
79; CHECK-NEXT:    vldrw.u32 q4, [r0, #48]
80; CHECK-NEXT:    vldrw.u32 q5, [r0, #32]
81; CHECK-NEXT:    vldrw.u32 q6, [r0, #16]
82; CHECK-NEXT:    vmov r0, s28
83; CHECK-NEXT:    vmov r1, s0
84; CHECK-NEXT:    str r1, [r0]
85; CHECK-NEXT:    vmov r0, s29
86; CHECK-NEXT:    vmov r1, s1
87; CHECK-NEXT:    str r1, [r0]
88; CHECK-NEXT:    vmov r0, s30
89; CHECK-NEXT:    vmov r1, s2
90; CHECK-NEXT:    str r1, [r0]
91; CHECK-NEXT:    vmov r0, s31
92; CHECK-NEXT:    vmov r1, s3
93; CHECK-NEXT:    str r1, [r0]
94; CHECK-NEXT:    vmov r0, s24
95; CHECK-NEXT:    vmov r1, s4
96; CHECK-NEXT:    str r1, [r0]
97; CHECK-NEXT:    vmov r0, s25
98; CHECK-NEXT:    vmov r1, s5
99; CHECK-NEXT:    str r1, [r0]
100; CHECK-NEXT:    vmov r0, s26
101; CHECK-NEXT:    vmov r1, s6
102; CHECK-NEXT:    str r1, [r0]
103; CHECK-NEXT:    vmov r0, s27
104; CHECK-NEXT:    vmov r1, s7
105; CHECK-NEXT:    str r1, [r0]
106; CHECK-NEXT:    vmov r0, s20
107; CHECK-NEXT:    vmov r1, s8
108; CHECK-NEXT:    str r1, [r0]
109; CHECK-NEXT:    vmov r0, s21
110; CHECK-NEXT:    vmov r1, s9
111; CHECK-NEXT:    str r1, [r0]
112; CHECK-NEXT:    vmov r0, s22
113; CHECK-NEXT:    vmov r1, s10
114; CHECK-NEXT:    str r1, [r0]
115; CHECK-NEXT:    vmov r0, s23
116; CHECK-NEXT:    vmov r1, s11
117; CHECK-NEXT:    str r1, [r0]
118; CHECK-NEXT:    vmov r0, s16
119; CHECK-NEXT:    vmov r1, s12
120; CHECK-NEXT:    str r1, [r0]
121; CHECK-NEXT:    vmov r0, s17
122; CHECK-NEXT:    vmov r1, s13
123; CHECK-NEXT:    str r1, [r0]
124; CHECK-NEXT:    vmov r0, s18
125; CHECK-NEXT:    vmov r1, s14
126; CHECK-NEXT:    str r1, [r0]
127; CHECK-NEXT:    vmov r0, s19
128; CHECK-NEXT:    vmov r1, s15
129; CHECK-NEXT:    str r1, [r0]
130; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13, d14, d15}
131; CHECK-NEXT:    bx lr
132entry:
133  %offs = load <16 x i32*>, <16 x i32*>* %offptr, align 4
134  call void @llvm.masked.scatter.v16i32.v16p0i32(<16 x i32> %v, <16 x i32*> %offs, i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
135  ret void
136}
137
138; f32
139
140; Expand
141define arm_aapcs_vfpcc void @ptr_v2f32(<2 x float> %v, <2 x float*>* %offptr) {
142; CHECK-LABEL: ptr_v2f32:
143; CHECK:       @ %bb.0: @ %entry
144; CHECK-NEXT:    ldrd r1, r0, [r0]
145; CHECK-NEXT:    vstr s0, [r1]
146; CHECK-NEXT:    vstr s1, [r0]
147; CHECK-NEXT:    bx lr
148entry:
149  %offs = load <2 x float*>, <2 x float*>* %offptr, align 4
150  call void @llvm.masked.scatter.v2f32.v2p0f32(<2 x float> %v, <2 x float*> %offs, i32 4, <2 x i1> <i1 true, i1 true>)
151  ret void
152}
153
154; VSTRW.32 Qd, [offs, 0]
155define arm_aapcs_vfpcc void @ptr_v4f32(<4 x float> %v, <4 x float*>* %offptr) {
156; CHECK-LABEL: ptr_v4f32:
157; CHECK:       @ %bb.0: @ %entry
158; CHECK-NEXT:    vldrw.u32 q1, [r0]
159; CHECK-NEXT:    vstrw.32 q0, [q1]
160; CHECK-NEXT:    bx lr
161entry:
162  %offs = load <4 x float*>, <4 x float*>* %offptr, align 4
163  call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %v, <4 x float*> %offs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
164  ret void
165}
166
167; Expand
168define arm_aapcs_vfpcc void @ptr_v8f32(<8 x float> %v, <8 x float*>* %offptr) {
169; CHECK-LABEL: ptr_v8f32:
170; CHECK:       @ %bb.0: @ %entry
171; CHECK-NEXT:    .save {r4, r5, r7, lr}
172; CHECK-NEXT:    push {r4, r5, r7, lr}
173; CHECK-NEXT:    vldrw.u32 q2, [r0, #16]
174; CHECK-NEXT:    vmov r12, s11
175; CHECK-NEXT:    vmov lr, s10
176; CHECK-NEXT:    vmov r3, s9
177; CHECK-NEXT:    vmov r1, s8
178; CHECK-NEXT:    vldrw.u32 q2, [r0]
179; CHECK-NEXT:    vmov r5, s8
180; CHECK-NEXT:    vmov r0, s11
181; CHECK-NEXT:    vmov r2, s10
182; CHECK-NEXT:    vmov r4, s9
183; CHECK-NEXT:    vstr s0, [r5]
184; CHECK-NEXT:    vstr s1, [r4]
185; CHECK-NEXT:    vstr s2, [r2]
186; CHECK-NEXT:    vstr s3, [r0]
187; CHECK-NEXT:    vstr s4, [r1]
188; CHECK-NEXT:    vstr s5, [r3]
189; CHECK-NEXT:    vstr s6, [lr]
190; CHECK-NEXT:    vstr s7, [r12]
191; CHECK-NEXT:    pop {r4, r5, r7, pc}
192entry:
193  %offs = load <8 x float*>, <8 x float*>* %offptr, align 4
194  call void @llvm.masked.scatter.v8f32.v8p0f32(<8 x float> %v, <8 x float*> %offs, i32 4, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
195  ret void
196}
197
198; i16
199
200; Expand.
201define arm_aapcs_vfpcc void @ptr_i16(<8 x i16> %v, <8 x i16*>* %offptr) {
202; CHECK-LABEL: ptr_i16:
203; CHECK:       @ %bb.0: @ %entry
204; CHECK-NEXT:    vldrw.u32 q2, [r0]
205; CHECK-NEXT:    vldrw.u32 q1, [r0, #16]
206; CHECK-NEXT:    vmov.u16 r1, q0[0]
207; CHECK-NEXT:    vmov r0, s8
208; CHECK-NEXT:    strh r1, [r0]
209; CHECK-NEXT:    vmov r0, s9
210; CHECK-NEXT:    vmov.u16 r1, q0[1]
211; CHECK-NEXT:    strh r1, [r0]
212; CHECK-NEXT:    vmov r0, s10
213; CHECK-NEXT:    vmov.u16 r1, q0[2]
214; CHECK-NEXT:    strh r1, [r0]
215; CHECK-NEXT:    vmov r0, s11
216; CHECK-NEXT:    vmov.u16 r1, q0[3]
217; CHECK-NEXT:    strh r1, [r0]
218; CHECK-NEXT:    vmov r0, s4
219; CHECK-NEXT:    vmov.u16 r1, q0[4]
220; CHECK-NEXT:    strh r1, [r0]
221; CHECK-NEXT:    vmov r0, s5
222; CHECK-NEXT:    vmov.u16 r1, q0[5]
223; CHECK-NEXT:    strh r1, [r0]
224; CHECK-NEXT:    vmov r0, s6
225; CHECK-NEXT:    vmov.u16 r1, q0[6]
226; CHECK-NEXT:    strh r1, [r0]
227; CHECK-NEXT:    vmov r0, s7
228; CHECK-NEXT:    vmov.u16 r1, q0[7]
229; CHECK-NEXT:    strh r1, [r0]
230; CHECK-NEXT:    bx lr
231entry:
232  %offs = load <8 x i16*>, <8 x i16*>* %offptr, align 4
233  call void @llvm.masked.scatter.v8i16.v8p0i16(<8 x i16> %v, <8 x i16*> %offs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
234  ret void
235}
236
237; Expand
238define arm_aapcs_vfpcc void @ptr_v2i16_trunc(<2 x i32> %v, <2 x i16*>* %offptr) {
239; CHECK-LABEL: ptr_v2i16_trunc:
240; CHECK:       @ %bb.0: @ %entry
241; CHECK-NEXT:    vmov r2, s0
242; CHECK-NEXT:    ldrd r1, r0, [r0]
243; CHECK-NEXT:    strh r2, [r1]
244; CHECK-NEXT:    vmov r1, s2
245; CHECK-NEXT:    strh r1, [r0]
246; CHECK-NEXT:    bx lr
247entry:
248  %offs = load <2 x i16*>, <2 x i16*>* %offptr, align 4
249  %ext = trunc <2 x i32> %v to <2 x i16>
250  call void @llvm.masked.scatter.v2i16.v2p0i16(<2 x i16> %ext, <2 x i16*> %offs, i32 2, <2 x i1> <i1 true, i1 true>)
251  ret void
252}
253
254; Expand
255define arm_aapcs_vfpcc void @ptr_v4i16_trunc(<4 x i32> %v, <4 x i16*>* %offptr) {
256; CHECK-LABEL: ptr_v4i16_trunc:
257; CHECK:       @ %bb.0: @ %entry
258; CHECK-NEXT:    vldrw.u32 q1, [r0]
259; CHECK-NEXT:    vmov r1, s0
260; CHECK-NEXT:    vmov r0, s4
261; CHECK-NEXT:    strh r1, [r0]
262; CHECK-NEXT:    vmov r0, s5
263; CHECK-NEXT:    vmov r1, s1
264; CHECK-NEXT:    strh r1, [r0]
265; CHECK-NEXT:    vmov r0, s6
266; CHECK-NEXT:    vmov r1, s2
267; CHECK-NEXT:    strh r1, [r0]
268; CHECK-NEXT:    vmov r0, s7
269; CHECK-NEXT:    vmov r1, s3
270; CHECK-NEXT:    strh r1, [r0]
271; CHECK-NEXT:    bx lr
272entry:
273  %offs = load <4 x i16*>, <4 x i16*>* %offptr, align 4
274  %ext = trunc <4 x i32> %v to <4 x i16>
275  call void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16> %ext, <4 x i16*> %offs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
276  ret void
277}
278
279; Expand
280define arm_aapcs_vfpcc void @ptr_v8i16_trunc(<8 x i32> %v, <8 x i16*>* %offptr) {
281; CHECK-LABEL: ptr_v8i16_trunc:
282; CHECK:       @ %bb.0: @ %entry
283; CHECK-NEXT:    vldrw.u32 q3, [r0]
284; CHECK-NEXT:    vldrw.u32 q2, [r0, #16]
285; CHECK-NEXT:    vmov r1, s0
286; CHECK-NEXT:    vmov r0, s12
287; CHECK-NEXT:    strh r1, [r0]
288; CHECK-NEXT:    vmov r0, s13
289; CHECK-NEXT:    vmov r1, s1
290; CHECK-NEXT:    strh r1, [r0]
291; CHECK-NEXT:    vmov r0, s14
292; CHECK-NEXT:    vmov r1, s2
293; CHECK-NEXT:    strh r1, [r0]
294; CHECK-NEXT:    vmov r0, s15
295; CHECK-NEXT:    vmov r1, s3
296; CHECK-NEXT:    strh r1, [r0]
297; CHECK-NEXT:    vmov r0, s8
298; CHECK-NEXT:    vmov r1, s4
299; CHECK-NEXT:    strh r1, [r0]
300; CHECK-NEXT:    vmov r0, s9
301; CHECK-NEXT:    vmov r1, s5
302; CHECK-NEXT:    strh r1, [r0]
303; CHECK-NEXT:    vmov r0, s10
304; CHECK-NEXT:    vmov r1, s6
305; CHECK-NEXT:    strh r1, [r0]
306; CHECK-NEXT:    vmov r0, s11
307; CHECK-NEXT:    vmov r1, s7
308; CHECK-NEXT:    strh r1, [r0]
309; CHECK-NEXT:    bx lr
310entry:
311  %offs = load <8 x i16*>, <8 x i16*>* %offptr, align 4
312  %ext = trunc <8 x i32> %v to <8 x i16>
313  call void @llvm.masked.scatter.v8i16.v8p0i16(<8 x i16> %ext, <8 x i16*> %offs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
314  ret void
315}
316
317; f16
318
319; Expand.
320define arm_aapcs_vfpcc void @ptr_f16(<8 x half> %v, <8 x half*>* %offptr) {
321; CHECK-LABEL: ptr_f16:
322; CHECK:       @ %bb.0: @ %entry
323; CHECK-NEXT:    vldrw.u32 q2, [r0]
324; CHECK-NEXT:    vldrw.u32 q1, [r0, #16]
325; CHECK-NEXT:    vmovx.f16 s12, s0
326; CHECK-NEXT:    vmov r0, s8
327; CHECK-NEXT:    vstr.16 s0, [r0]
328; CHECK-NEXT:    vmov r0, s9
329; CHECK-NEXT:    vstr.16 s12, [r0]
330; CHECK-NEXT:    vmov r0, s10
331; CHECK-NEXT:    vstr.16 s1, [r0]
332; CHECK-NEXT:    vmov r0, s11
333; CHECK-NEXT:    vmovx.f16 s8, s1
334; CHECK-NEXT:    vmovx.f16 s0, s3
335; CHECK-NEXT:    vstr.16 s8, [r0]
336; CHECK-NEXT:    vmov r0, s4
337; CHECK-NEXT:    vstr.16 s2, [r0]
338; CHECK-NEXT:    vmov r0, s5
339; CHECK-NEXT:    vmovx.f16 s8, s2
340; CHECK-NEXT:    vstr.16 s8, [r0]
341; CHECK-NEXT:    vmov r0, s6
342; CHECK-NEXT:    vstr.16 s3, [r0]
343; CHECK-NEXT:    vmov r0, s7
344; CHECK-NEXT:    vstr.16 s0, [r0]
345; CHECK-NEXT:    bx lr
346entry:
347  %offs = load <8 x half*>, <8 x half*>* %offptr, align 4
348  call void @llvm.masked.scatter.v8f16.v8p0f16(<8 x half> %v, <8 x half*> %offs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
349  ret void
350}
351
352; i8
353
354; Expand.
355define arm_aapcs_vfpcc void @ptr_i8(<16 x i8> %v, <16 x i8*>* %offptr) {
356; CHECK-LABEL: ptr_i8:
357; CHECK:       @ %bb.0: @ %entry
358; CHECK-NEXT:    .vsave {d8, d9}
359; CHECK-NEXT:    vpush {d8, d9}
360; CHECK-NEXT:    vldrw.u32 q4, [r0]
361; CHECK-NEXT:    vldrw.u32 q1, [r0, #48]
362; CHECK-NEXT:    vldrw.u32 q2, [r0, #32]
363; CHECK-NEXT:    vldrw.u32 q3, [r0, #16]
364; CHECK-NEXT:    vmov r0, s16
365; CHECK-NEXT:    vmov.u8 r1, q0[0]
366; CHECK-NEXT:    strb r1, [r0]
367; CHECK-NEXT:    vmov r0, s17
368; CHECK-NEXT:    vmov.u8 r1, q0[1]
369; CHECK-NEXT:    strb r1, [r0]
370; CHECK-NEXT:    vmov r0, s18
371; CHECK-NEXT:    vmov.u8 r1, q0[2]
372; CHECK-NEXT:    strb r1, [r0]
373; CHECK-NEXT:    vmov r0, s19
374; CHECK-NEXT:    vmov.u8 r1, q0[3]
375; CHECK-NEXT:    strb r1, [r0]
376; CHECK-NEXT:    vmov r0, s12
377; CHECK-NEXT:    vmov.u8 r1, q0[4]
378; CHECK-NEXT:    strb r1, [r0]
379; CHECK-NEXT:    vmov r0, s13
380; CHECK-NEXT:    vmov.u8 r1, q0[5]
381; CHECK-NEXT:    strb r1, [r0]
382; CHECK-NEXT:    vmov r0, s14
383; CHECK-NEXT:    vmov.u8 r1, q0[6]
384; CHECK-NEXT:    strb r1, [r0]
385; CHECK-NEXT:    vmov r0, s15
386; CHECK-NEXT:    vmov.u8 r1, q0[7]
387; CHECK-NEXT:    strb r1, [r0]
388; CHECK-NEXT:    vmov r0, s8
389; CHECK-NEXT:    vmov.u8 r1, q0[8]
390; CHECK-NEXT:    strb r1, [r0]
391; CHECK-NEXT:    vmov r0, s9
392; CHECK-NEXT:    vmov.u8 r1, q0[9]
393; CHECK-NEXT:    strb r1, [r0]
394; CHECK-NEXT:    vmov r0, s10
395; CHECK-NEXT:    vmov.u8 r1, q0[10]
396; CHECK-NEXT:    strb r1, [r0]
397; CHECK-NEXT:    vmov r0, s11
398; CHECK-NEXT:    vmov.u8 r1, q0[11]
399; CHECK-NEXT:    strb r1, [r0]
400; CHECK-NEXT:    vmov r0, s4
401; CHECK-NEXT:    vmov.u8 r1, q0[12]
402; CHECK-NEXT:    strb r1, [r0]
403; CHECK-NEXT:    vmov r0, s5
404; CHECK-NEXT:    vmov.u8 r1, q0[13]
405; CHECK-NEXT:    strb r1, [r0]
406; CHECK-NEXT:    vmov r0, s6
407; CHECK-NEXT:    vmov.u8 r1, q0[14]
408; CHECK-NEXT:    strb r1, [r0]
409; CHECK-NEXT:    vmov r0, s7
410; CHECK-NEXT:    vmov.u8 r1, q0[15]
411; CHECK-NEXT:    strb r1, [r0]
412; CHECK-NEXT:    vpop {d8, d9}
413; CHECK-NEXT:    bx lr
414entry:
415  %offs = load <16 x i8*>, <16 x i8*>* %offptr, align 4
416  call void @llvm.masked.scatter.v16i8.v16p0i8(<16 x i8> %v, <16 x i8*> %offs, i32 2, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
417  ret void
418}
419
420; Expand
421define arm_aapcs_vfpcc void @ptr_v8i8_trunc16(<8 x i16> %v, <8 x i8*>* %offptr) {
422; CHECK-LABEL: ptr_v8i8_trunc16:
423; CHECK:       @ %bb.0: @ %entry
424; CHECK-NEXT:    vldrw.u32 q2, [r0]
425; CHECK-NEXT:    vldrw.u32 q1, [r0, #16]
426; CHECK-NEXT:    vmov.u16 r1, q0[0]
427; CHECK-NEXT:    vmov r0, s8
428; CHECK-NEXT:    strb r1, [r0]
429; CHECK-NEXT:    vmov r0, s9
430; CHECK-NEXT:    vmov.u16 r1, q0[1]
431; CHECK-NEXT:    strb r1, [r0]
432; CHECK-NEXT:    vmov r0, s10
433; CHECK-NEXT:    vmov.u16 r1, q0[2]
434; CHECK-NEXT:    strb r1, [r0]
435; CHECK-NEXT:    vmov r0, s11
436; CHECK-NEXT:    vmov.u16 r1, q0[3]
437; CHECK-NEXT:    strb r1, [r0]
438; CHECK-NEXT:    vmov r0, s4
439; CHECK-NEXT:    vmov.u16 r1, q0[4]
440; CHECK-NEXT:    strb r1, [r0]
441; CHECK-NEXT:    vmov r0, s5
442; CHECK-NEXT:    vmov.u16 r1, q0[5]
443; CHECK-NEXT:    strb r1, [r0]
444; CHECK-NEXT:    vmov r0, s6
445; CHECK-NEXT:    vmov.u16 r1, q0[6]
446; CHECK-NEXT:    strb r1, [r0]
447; CHECK-NEXT:    vmov r0, s7
448; CHECK-NEXT:    vmov.u16 r1, q0[7]
449; CHECK-NEXT:    strb r1, [r0]
450; CHECK-NEXT:    bx lr
451entry:
452  %offs = load <8 x i8*>, <8 x i8*>* %offptr, align 4
453  %ext = trunc <8 x i16> %v to <8 x i8>
454  call void @llvm.masked.scatter.v8i8.v8p0i8(<8 x i8> %ext, <8 x i8*> %offs, i32 1, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
455  ret void
456}
457
458; Expand
459define arm_aapcs_vfpcc void @ptr_v4i8_trunc32(<4 x i32> %v, <4 x i8*>* %offptr) {
460; CHECK-LABEL: ptr_v4i8_trunc32:
461; CHECK:       @ %bb.0: @ %entry
462; CHECK-NEXT:    vldrw.u32 q1, [r0]
463; CHECK-NEXT:    vmov r1, s0
464; CHECK-NEXT:    vmov r0, s4
465; CHECK-NEXT:    strb r1, [r0]
466; CHECK-NEXT:    vmov r0, s5
467; CHECK-NEXT:    vmov r1, s1
468; CHECK-NEXT:    strb r1, [r0]
469; CHECK-NEXT:    vmov r0, s6
470; CHECK-NEXT:    vmov r1, s2
471; CHECK-NEXT:    strb r1, [r0]
472; CHECK-NEXT:    vmov r0, s7
473; CHECK-NEXT:    vmov r1, s3
474; CHECK-NEXT:    strb r1, [r0]
475; CHECK-NEXT:    bx lr
476entry:
477  %offs = load <4 x i8*>, <4 x i8*>* %offptr, align 4
478  %ext = trunc <4 x i32> %v to <4 x i8>
479  call void @llvm.masked.scatter.v4i8.v4p0i8(<4 x i8> %ext, <4 x i8*> %offs, i32 1, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
480  ret void
481}
482
483; Expand
484define arm_aapcs_vfpcc void @ptr_v8i8_trunc32(<8 x i32> %v, <8 x i8*>* %offptr) {
485; CHECK-LABEL: ptr_v8i8_trunc32:
486; CHECK:       @ %bb.0: @ %entry
487; CHECK-NEXT:    vldrw.u32 q3, [r0]
488; CHECK-NEXT:    vldrw.u32 q2, [r0, #16]
489; CHECK-NEXT:    vmov r1, s0
490; CHECK-NEXT:    vmov r0, s12
491; CHECK-NEXT:    strb r1, [r0]
492; CHECK-NEXT:    vmov r0, s13
493; CHECK-NEXT:    vmov r1, s1
494; CHECK-NEXT:    strb r1, [r0]
495; CHECK-NEXT:    vmov r0, s14
496; CHECK-NEXT:    vmov r1, s2
497; CHECK-NEXT:    strb r1, [r0]
498; CHECK-NEXT:    vmov r0, s15
499; CHECK-NEXT:    vmov r1, s3
500; CHECK-NEXT:    strb r1, [r0]
501; CHECK-NEXT:    vmov r0, s8
502; CHECK-NEXT:    vmov r1, s4
503; CHECK-NEXT:    strb r1, [r0]
504; CHECK-NEXT:    vmov r0, s9
505; CHECK-NEXT:    vmov r1, s5
506; CHECK-NEXT:    strb r1, [r0]
507; CHECK-NEXT:    vmov r0, s10
508; CHECK-NEXT:    vmov r1, s6
509; CHECK-NEXT:    strb r1, [r0]
510; CHECK-NEXT:    vmov r0, s11
511; CHECK-NEXT:    vmov r1, s7
512; CHECK-NEXT:    strb r1, [r0]
513; CHECK-NEXT:    bx lr
514entry:
515  %offs = load <8 x i8*>, <8 x i8*>* %offptr, align 4
516  %ext = trunc <8 x i32> %v to <8 x i8>
517  call void @llvm.masked.scatter.v8i8.v8p0i8(<8 x i8> %ext, <8 x i8*> %offs, i32 1, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
518  ret void
519}
520
521; loops
522
523define void @foo_ptr_p_int32_t(i32* %dest, i32** %src, i32 %n) {
524; CHECK-LABEL: foo_ptr_p_int32_t:
525; CHECK:       @ %bb.0: @ %entry
526; CHECK-NEXT:    bic r3, r2, #15
527; CHECK-NEXT:    cmp r3, #1
528; CHECK-NEXT:    it lt
529; CHECK-NEXT:    bxlt lr
530; CHECK-NEXT:  .LBB16_1: @ %vector.body
531; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
532; CHECK-NEXT:    vldrw.u32 q0, [r1], #16
533; CHECK-NEXT:    subs r2, #4
534; CHECK-NEXT:    vptt.i32 ne, q0, zr
535; CHECK-NEXT:    vldrwt.u32 q1, [r0], #16
536; CHECK-NEXT:    vstrwt.32 q1, [q0]
537; CHECK-NEXT:    bne .LBB16_1
538; CHECK-NEXT:  @ %bb.2: @ %for.end
539; CHECK-NEXT:    bx lr
540entry:
541  %and = and i32 %n, -16
542  %cmp11 = icmp sgt i32 %and, 0
543  br i1 %cmp11, label %vector.body, label %for.end
544
545vector.body:                                      ; preds = %entry, %vector.body
546  %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
547  %0 = getelementptr inbounds i32*, i32** %src, i32 %index
548  %1 = bitcast i32** %0 to <4 x i32*>*
549  %wide.load = load <4 x i32*>, <4 x i32*>* %1, align 4
550  %2 = icmp ne <4 x i32*> %wide.load, zeroinitializer
551  %3 = getelementptr inbounds i32, i32* %dest, i32 %index
552  %4 = bitcast i32* %3 to <4 x i32>*
553  %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.v4p0i32(<4 x i32>* %4, i32 4, <4 x i1> %2, <4 x i32> undef)
554  call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %wide.masked.load, <4 x i32*> %wide.load, i32 4, <4 x i1> %2)
555  %index.next = add i32 %index, 4
556  %5 = icmp eq i32 %index.next, %n
557  br i1 %5, label %for.end, label %vector.body
558
559for.end:                                          ; preds = %vector.body, %entry
560  ret void
561}
562
563define void @foo_ptr_p_float(float* %dest, float** %src, i32 %n) {
564; CHECK-LABEL: foo_ptr_p_float:
565; CHECK:       @ %bb.0: @ %entry
566; CHECK-NEXT:    bic r3, r2, #15
567; CHECK-NEXT:    cmp r3, #1
568; CHECK-NEXT:    it lt
569; CHECK-NEXT:    bxlt lr
570; CHECK-NEXT:  .LBB17_1: @ %vector.body
571; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
572; CHECK-NEXT:    vldrw.u32 q0, [r1], #16
573; CHECK-NEXT:    subs r2, #4
574; CHECK-NEXT:    vptt.i32 ne, q0, zr
575; CHECK-NEXT:    vldrwt.u32 q1, [r0], #16
576; CHECK-NEXT:    vstrwt.32 q1, [q0]
577; CHECK-NEXT:    bne .LBB17_1
578; CHECK-NEXT:  @ %bb.2: @ %for.end
579; CHECK-NEXT:    bx lr
580entry:
581  %and = and i32 %n, -16
582  %cmp11 = icmp sgt i32 %and, 0
583  br i1 %cmp11, label %vector.body, label %for.end
584
585vector.body:                                      ; preds = %entry, %vector.body
586  %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
587  %0 = getelementptr inbounds float*, float** %src, i32 %index
588  %1 = bitcast float** %0 to <4 x float*>*
589  %wide.load = load <4 x float*>, <4 x float*>* %1, align 4
590  %2 = icmp ne <4 x float*> %wide.load, zeroinitializer
591  %3 = getelementptr inbounds float, float* %dest, i32 %index
592  %4 = bitcast float* %3 to <4 x i32>*
593  %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.v4p0i32(<4 x i32>* %4, i32 4, <4 x i1> %2, <4 x i32> undef)
594  %5 = bitcast <4 x float*> %wide.load to <4 x i32*>
595  call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %wide.masked.load, <4 x i32*> %5, i32 4, <4 x i1> %2)
596  %index.next = add i32 %index, 4
597  %6 = icmp eq i32 %index.next, %n
598  br i1 %6, label %for.end, label %vector.body
599
600for.end:                                          ; preds = %vector.body, %entry
601  ret void
602}
603
604; VLSTW.u32 Qd, [P, 4]
605define arm_aapcs_vfpcc void @qi4(<4 x i32> %v, <4 x i32*> %p) {
606; CHECK-LABEL: qi4:
607; CHECK:       @ %bb.0: @ %entry
608; CHECK-NEXT:    vmov.i32 q2, #0x10
609; CHECK-NEXT:    vadd.i32 q1, q1, q2
610; CHECK-NEXT:    vstrw.32 q0, [q1]
611; CHECK-NEXT:    bx lr
612entry:
613  %g = getelementptr inbounds i32, <4 x i32*> %p, i32 4
614  call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %v, <4 x i32*> %g, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
615  ret void
616}
617
618declare void @llvm.masked.scatter.v2i16.v2p0i16(<2 x i16>, <2 x i16*>, i32, <2 x i1>)
619declare void @llvm.masked.scatter.v2i32.v2p0i32(<2 x i32>, <2 x i32*>, i32, <2 x i1>)
620declare void @llvm.masked.scatter.v2f32.v2p0f32(<2 x float>, <2 x float*>, i32, <2 x i1>)
621declare void @llvm.masked.scatter.v4i8.v4p0i8(<4 x i8>, <4 x i8*>, i32, <4 x i1>)
622declare void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16>, <4 x i16*>, i32, <4 x i1>)
623declare void @llvm.masked.scatter.v4f16.v4p0f16(<4 x half>, <4 x half*>, i32, <4 x i1>)
624declare void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32>, <4 x i32*>, i32, <4 x i1>)
625declare void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float>, <4 x float*>, i32, <4 x i1>)
626declare void @llvm.masked.scatter.v8i8.v8p0i8(<8 x i8>, <8 x i8*>, i32, <8 x i1>)
627declare void @llvm.masked.scatter.v8i16.v8p0i16(<8 x i16>, <8 x i16*>, i32, <8 x i1>)
628declare void @llvm.masked.scatter.v8f16.v8p0f16(<8 x half>, <8 x half*>, i32, <8 x i1>)
629declare void @llvm.masked.scatter.v8i32.v8p0i32(<8 x i32>, <8 x i32*>, i32, <8 x i1>)
630declare void @llvm.masked.scatter.v8f32.v8p0f32(<8 x float>, <8 x float*>, i32, <8 x i1>)
631declare void @llvm.masked.scatter.v16i8.v16p0i8(<16 x i8>, <16 x i8*>, i32, <16 x i1>)
632declare void @llvm.masked.scatter.v16i32.v16p0i32(<16 x i32>, <16 x i32*>, i32, <16 x i1>)
633declare <4 x i32> @llvm.masked.load.v4i32.v4p0i32(<4 x i32>*, i32, <4 x i1>, <4 x i32>)
634