xref: /llvm-project/llvm/test/CodeGen/AArch64/arm64-abi_align.ll (revision c8e5ad4e123a5fb082355947a896980464689c31)
1; RUN: llc -aarch64-load-store-renaming=true < %s -mtriple=arm64-apple-darwin -mcpu=cyclone -enable-misched=false -frame-pointer=all | FileCheck %s
2; RUN: llc -aarch64-load-store-renaming=true  < %s -mtriple=arm64-apple-darwin -O0 -frame-pointer=all -fast-isel | FileCheck -check-prefix=FAST %s
3
4; rdar://12648441
5; Generated from arm64-arguments.c with -O2.
6; Test passing structs with size < 8, < 16 and > 16
7; with alignment of 16 and without
8
9; Structs with size < 8
10%struct.s38 = type { i32, i16 }
11; With alignment of 16, the size will be padded to multiple of 16 bytes.
12%struct.s39 = type { i32, i16, [10 x i8] }
13; Structs with size < 16
14%struct.s40 = type { i32, i16, i32, i16 }
15%struct.s41 = type { i32, i16, i32, i16 }
16; Structs with size > 16
17%struct.s42 = type { i32, i16, i32, i16, i32, i16 }
18%struct.s43 = type { i32, i16, i32, i16, i32, i16, [10 x i8] }
19
20@g38 = common global %struct.s38 zeroinitializer, align 4
21@g38_2 = common global %struct.s38 zeroinitializer, align 4
22@g39 = common global %struct.s39 zeroinitializer, align 16
23@g39_2 = common global %struct.s39 zeroinitializer, align 16
24@g40 = common global %struct.s40 zeroinitializer, align 4
25@g40_2 = common global %struct.s40 zeroinitializer, align 4
26@g41 = common global %struct.s41 zeroinitializer, align 16
27@g41_2 = common global %struct.s41 zeroinitializer, align 16
28@g42 = common global %struct.s42 zeroinitializer, align 4
29@g42_2 = common global %struct.s42 zeroinitializer, align 4
30@g43 = common global %struct.s43 zeroinitializer, align 16
31@g43_2 = common global %struct.s43 zeroinitializer, align 16
32
33; structs with size < 8 bytes, passed via i64 in x1 and x2
34define i32 @f38(i32 %i, i64 %s1.coerce, i64 %s2.coerce) #0 {
35entry:
36; CHECK-LABEL: f38
37; CHECK: add w[[A:[0-9]+]], w1, w0
38; CHECK: add {{w[0-9]+}}, w[[A]], w2
39  %s1.sroa.0.0.extract.trunc = trunc i64 %s1.coerce to i32
40  %s1.sroa.1.4.extract.shift = lshr i64 %s1.coerce, 32
41  %s2.sroa.0.0.extract.trunc = trunc i64 %s2.coerce to i32
42  %s2.sroa.1.4.extract.shift = lshr i64 %s2.coerce, 32
43  %sext8 = shl nuw nsw i64 %s1.sroa.1.4.extract.shift, 16
44  %sext = trunc i64 %sext8 to i32
45  %conv = ashr exact i32 %sext, 16
46  %sext1011 = shl nuw nsw i64 %s2.sroa.1.4.extract.shift, 16
47  %sext10 = trunc i64 %sext1011 to i32
48  %conv6 = ashr exact i32 %sext10, 16
49  %add = add i32 %s1.sroa.0.0.extract.trunc, %i
50  %add3 = add i32 %add, %s2.sroa.0.0.extract.trunc
51  %add4 = add i32 %add3, %conv
52  %add7 = add i32 %add4, %conv6
53  ret i32 %add7
54}
55
56define i32 @caller38() #1 {
57entry:
58; CHECK-LABEL: caller38
59; CHECK: ldr x1,
60; CHECK: ldr x2,
61  %0 = load i64, ptr @g38, align 4
62  %1 = load i64, ptr @g38_2, align 4
63  %call = tail call i32 @f38(i32 3, i64 %0, i64 %1) #5
64  ret i32 %call
65}
66
67declare i32 @f38_stack(i32 %i, i32 %i2, i32 %i3, i32 %i4, i32 %i5, i32 %i6,
68                i32 %i7, i32 %i8, i32 %i9, i64 %s1.coerce, i64 %s2.coerce) #0
69
70; structs with size < 8 bytes, passed on stack at [sp+8] and [sp+16]
71; i9 at [sp]
72define i32 @caller38_stack() #1 {
73entry:
74; CHECK-LABEL: caller38_stack
75; CHECK: stp {{x[0-9]+}}, {{x[0-9]+}}, [sp, #8]
76; CHECK: mov w[[C:[0-9]+]], #9
77; CHECK: str w[[C]], [sp]
78  %0 = load i64, ptr @g38, align 4
79  %1 = load i64, ptr @g38_2, align 4
80  %call = tail call i32 @f38_stack(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6,
81                                   i32 7, i32 8, i32 9, i64 %0, i64 %1) #5
82  ret i32 %call
83}
84
85; structs with size < 8 bytes, alignment of 16
86; passed via i128 in x1 and x3
87define i32 @f39(i32 %i, i128 %s1.coerce, i128 %s2.coerce) #0 {
88entry:
89; CHECK-LABEL: f39
90; CHECK: add w[[A:[0-9]+]], w1, w0
91; CHECK: add {{w[0-9]+}}, w[[A]], w3
92  %s1.sroa.0.0.extract.trunc = trunc i128 %s1.coerce to i32
93  %s1.sroa.1.4.extract.shift = lshr i128 %s1.coerce, 32
94  %s2.sroa.0.0.extract.trunc = trunc i128 %s2.coerce to i32
95  %s2.sroa.1.4.extract.shift = lshr i128 %s2.coerce, 32
96  %sext8 = shl nuw nsw i128 %s1.sroa.1.4.extract.shift, 16
97  %sext = trunc i128 %sext8 to i32
98  %conv = ashr exact i32 %sext, 16
99  %sext1011 = shl nuw nsw i128 %s2.sroa.1.4.extract.shift, 16
100  %sext10 = trunc i128 %sext1011 to i32
101  %conv6 = ashr exact i32 %sext10, 16
102  %add = add i32 %s1.sroa.0.0.extract.trunc, %i
103  %add3 = add i32 %add, %s2.sroa.0.0.extract.trunc
104  %add4 = add i32 %add3, %conv
105  %add7 = add i32 %add4, %conv6
106  ret i32 %add7
107}
108
109define i32 @caller39() #1 {
110entry:
111; CHECK-LABEL: caller39
112; CHECK: ldp x1, x2,
113; CHECK: ldp x3, x4,
114  %0 = load i128, ptr @g39, align 16
115  %1 = load i128, ptr @g39_2, align 16
116  %call = tail call i32 @f39(i32 3, i128 %0, i128 %1) #5
117  ret i32 %call
118}
119
120declare i32 @f39_stack(i32 %i, i32 %i2, i32 %i3, i32 %i4, i32 %i5, i32 %i6,
121                i32 %i7, i32 %i8, i32 %i9, i128 %s1.coerce, i128 %s2.coerce) #0
122
123; structs with size < 8 bytes, alignment 16
124; passed on stack at [sp+16] and [sp+32]
125define i32 @caller39_stack() #1 {
126entry:
127; CHECK-LABEL: caller39_stack
128; CHECK: stp {{x[0-9]+}}, {{x[0-9]+}}, [sp, #32]
129; CHECK: stp {{x[0-9]+}}, {{x[0-9]+}}, [sp, #16]
130; CHECK: mov w[[C:[0-9]+]], #9
131; CHECK: str w[[C]], [sp]
132  %0 = load i128, ptr @g39, align 16
133  %1 = load i128, ptr @g39_2, align 16
134  %call = tail call i32 @f39_stack(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6,
135                                   i32 7, i32 8, i32 9, i128 %0, i128 %1) #5
136  ret i32 %call
137}
138
139; structs with size < 16 bytes
140; passed via i128 in x1 and x3
141define i32 @f40(i32 %i, [2 x i64] %s1.coerce, [2 x i64] %s2.coerce) #0 {
142entry:
143; CHECK-LABEL: f40
144; CHECK: add w[[A:[0-9]+]], w1, w0
145; CHECK: add {{w[0-9]+}}, w[[A]], w3
146  %s1.coerce.fca.0.extract = extractvalue [2 x i64] %s1.coerce, 0
147  %s2.coerce.fca.0.extract = extractvalue [2 x i64] %s2.coerce, 0
148  %s1.sroa.0.0.extract.trunc = trunc i64 %s1.coerce.fca.0.extract to i32
149  %s2.sroa.0.0.extract.trunc = trunc i64 %s2.coerce.fca.0.extract to i32
150  %s1.sroa.0.4.extract.shift = lshr i64 %s1.coerce.fca.0.extract, 32
151  %sext8 = shl nuw nsw i64 %s1.sroa.0.4.extract.shift, 16
152  %sext = trunc i64 %sext8 to i32
153  %conv = ashr exact i32 %sext, 16
154  %s2.sroa.0.4.extract.shift = lshr i64 %s2.coerce.fca.0.extract, 32
155  %sext1011 = shl nuw nsw i64 %s2.sroa.0.4.extract.shift, 16
156  %sext10 = trunc i64 %sext1011 to i32
157  %conv6 = ashr exact i32 %sext10, 16
158  %add = add i32 %s1.sroa.0.0.extract.trunc, %i
159  %add3 = add i32 %add, %s2.sroa.0.0.extract.trunc
160  %add4 = add i32 %add3, %conv
161  %add7 = add i32 %add4, %conv6
162  ret i32 %add7
163}
164
165define i32 @caller40() #1 {
166entry:
167; CHECK-LABEL: caller40
168; CHECK: ldp x1, x2,
169; CHECK: ldp x3, x4,
170  %0 = load [2 x i64], ptr @g40, align 4
171  %1 = load [2 x i64], ptr @g40_2, align 4
172  %call = tail call i32 @f40(i32 3, [2 x i64] %0, [2 x i64] %1) #5
173  ret i32 %call
174}
175
176declare i32 @f40_stack(i32 %i, i32 %i2, i32 %i3, i32 %i4, i32 %i5, i32 %i6,
177                i32 %i7, i32 %i8, i32 %i9, [2 x i64] %s1.coerce, [2 x i64] %s2.coerce) #0
178
179; structs with size < 16 bytes
180; passed on stack at [sp+8] and [sp+24]
181define i32 @caller40_stack() #1 {
182entry:
183; CHECK-LABEL: caller40_stack
184; CHECK: stp {{x[0-9]+}}, {{x[0-9]+}}, [sp, #24]
185; CHECK: stp {{x[0-9]+}}, {{x[0-9]+}}, [sp, #8]
186; CHECK: mov w[[C:[0-9]+]], #9
187; CHECK: str w[[C]], [sp]
188  %0 = load [2 x i64], ptr @g40, align 4
189  %1 = load [2 x i64], ptr @g40_2, align 4
190  %call = tail call i32 @f40_stack(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6,
191                         i32 7, i32 8, i32 9, [2 x i64] %0, [2 x i64] %1) #5
192  ret i32 %call
193}
194
195; structs with size < 16 bytes, alignment of 16
196; passed via i128 in x1 and x3
197define i32 @f41(i32 %i, i128 %s1.coerce, i128 %s2.coerce) #0 {
198entry:
199; CHECK-LABEL: f41
200; CHECK: add w[[A:[0-9]+]], w1, w0
201; CHECK: add {{w[0-9]+}}, w[[A]], w3
202  %s1.sroa.0.0.extract.trunc = trunc i128 %s1.coerce to i32
203  %s1.sroa.1.4.extract.shift = lshr i128 %s1.coerce, 32
204  %s2.sroa.0.0.extract.trunc = trunc i128 %s2.coerce to i32
205  %s2.sroa.1.4.extract.shift = lshr i128 %s2.coerce, 32
206  %sext8 = shl nuw nsw i128 %s1.sroa.1.4.extract.shift, 16
207  %sext = trunc i128 %sext8 to i32
208  %conv = ashr exact i32 %sext, 16
209  %sext1011 = shl nuw nsw i128 %s2.sroa.1.4.extract.shift, 16
210  %sext10 = trunc i128 %sext1011 to i32
211  %conv6 = ashr exact i32 %sext10, 16
212  %add = add i32 %s1.sroa.0.0.extract.trunc, %i
213  %add3 = add i32 %add, %s2.sroa.0.0.extract.trunc
214  %add4 = add i32 %add3, %conv
215  %add7 = add i32 %add4, %conv6
216  ret i32 %add7
217}
218
219define i32 @caller41() #1 {
220entry:
221; CHECK-LABEL: caller41
222; CHECK: ldp x1, x2,
223; CHECK: ldp x3, x4,
224  %0 = load i128, ptr @g41, align 16
225  %1 = load i128, ptr @g41_2, align 16
226  %call = tail call i32 @f41(i32 3, i128 %0, i128 %1) #5
227  ret i32 %call
228}
229
230declare i32 @f41_stack(i32 %i, i32 %i2, i32 %i3, i32 %i4, i32 %i5, i32 %i6,
231                i32 %i7, i32 %i8, i32 %i9, i128 %s1.coerce, i128 %s2.coerce) #0
232
233; structs with size < 16 bytes, alignment of 16
234; passed on stack at [sp+16] and [sp+32]
235define i32 @caller41_stack() #1 {
236entry:
237; CHECK-LABEL: caller41_stack
238; CHECK: stp {{x[0-9]+}}, {{x[0-9]+}}, [sp, #32]
239; CHECK: stp {{x[0-9]+}}, {{x[0-9]+}}, [sp, #16]
240; CHECK: mov w[[C:[0-9]+]], #9
241; CHECK: str w[[C]], [sp]
242  %0 = load i128, ptr @g41, align 16
243  %1 = load i128, ptr @g41_2, align 16
244  %call = tail call i32 @f41_stack(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6,
245                            i32 7, i32 8, i32 9, i128 %0, i128 %1) #5
246  ret i32 %call
247}
248
249; structs with size of 22 bytes, passed indirectly in x1 and x2
250define i32 @f42(i32 %i, ptr nocapture %s1, ptr nocapture %s2) #2 {
251entry:
252; CHECK-LABEL: f42
253; CHECK: ldr w[[A:[0-9]+]], [x1]
254; CHECK: ldr w[[B:[0-9]+]], [x2]
255; CHECK: add w[[C:[0-9]+]], w[[A]], w0
256; CHECK: add {{w[0-9]+}}, w[[C]], w[[B]]
257; FAST: f42
258; FAST: ldr w[[A:[0-9]+]], [x1]
259; FAST: ldr w[[B:[0-9]+]], [x2]
260; FAST: add w[[C:[0-9]+]], w[[A]], w0
261; FAST: add {{w[0-9]+}}, w[[C]], w[[B]]
262  %0 = load i32, ptr %s1, align 4, !tbaa !0
263  %1 = load i32, ptr %s2, align 4, !tbaa !0
264  %s = getelementptr inbounds %struct.s42, ptr %s1, i64 0, i32 1
265  %2 = load i16, ptr %s, align 2, !tbaa !3
266  %conv = sext i16 %2 to i32
267  %s5 = getelementptr inbounds %struct.s42, ptr %s2, i64 0, i32 1
268  %3 = load i16, ptr %s5, align 2, !tbaa !3
269  %conv6 = sext i16 %3 to i32
270  %add = add i32 %0, %i
271  %add3 = add i32 %add, %1
272  %add4 = add i32 %add3, %conv
273  %add7 = add i32 %add4, %conv6
274  ret i32 %add7
275}
276
277; For s1, we allocate a 22-byte space, pass its address via x1
278define i32 @caller42() #3 {
279entry:
280; CHECK-LABEL: caller42
281; CHECK-DAG: str {{x[0-9]+}}, [sp, #48]
282; CHECK-DAG: str {{q[0-9]+}}, [sp, #32]
283; CHECK-DAG: str {{x[0-9]+}}, [sp, #16]
284; CHECK-DAG: str {{q[0-9]+}}, [sp]
285; CHECK: add x1, sp, #32
286; CHECK: mov x2, sp
287; Space for s1 is allocated at sp+32
288; Space for s2 is allocated at sp
289
290; FAST-LABEL: caller42
291; FAST: sub sp, sp, #64
292; Space for s1 is allocated at fp-24 = sp+24
293; FAST: add x[[A:[0-9]+]], sp, #24
294; Call memcpy with size = 24 (0x18)
295; FAST: mov {{x[0-9]+}}, #24
296; Space for s2 is allocated at sp
297; FAST: mov x[[A:[0-9]+]], sp
298; FAST: bl _memcpy
299  %tmp = alloca %struct.s42, align 4
300  %tmp1 = alloca %struct.s42, align 4
301  call void @llvm.memcpy.p0.p0.i64(ptr align 4 %tmp, ptr align 4 @g42, i64 24, i1 false), !tbaa.struct !4
302  call void @llvm.memcpy.p0.p0.i64(ptr align 4 %tmp1, ptr align 4 @g42_2, i64 24, i1 false), !tbaa.struct !4
303  %call = call i32 @f42(i32 3, ptr %tmp, ptr %tmp1) #5
304  ret i32 %call
305}
306
307declare void @llvm.memcpy.p0.p0.i64(ptr nocapture, ptr nocapture, i64, i1) #4
308
309declare i32 @f42_stack(i32 %i, i32 %i2, i32 %i3, i32 %i4, i32 %i5, i32 %i6,
310                       i32 %i7, i32 %i8, i32 %i9, ptr nocapture %s1,
311                       ptr nocapture %s2) #2
312
313define i32 @caller42_stack() #3 {
314entry:
315; CHECK-LABEL: caller42_stack
316; CHECK: sub sp, sp, #112
317; CHECK: add x29, sp, #96
318; CHECK-DAG: stur {{x[0-9]+}}, [x29, #-16]
319; CHECK-DAG: stur {{q[0-9]+}}, [x29, #-32]
320; CHECK-DAG: str {{x[0-9]+}}, [sp, #48]
321; CHECK-DAG: str {{q[0-9]+}}, [sp, #32]
322; Space for s1 is allocated at x29-32 = sp+64
323; Space for s2 is allocated at sp+32
324; CHECK: add x[[B:[0-9]+]], sp, #32
325; CHECK: str x[[B]], [sp, #16]
326; CHECK: sub x[[A:[0-9]+]], x29, #32
327; Address of s1 is passed on stack at sp+8
328; CHECK: str x[[A]], [sp, #8]
329; CHECK: mov w[[C:[0-9]+]], #9
330; CHECK: str w[[C]], [sp]
331
332; FAST-LABEL: caller42_stack
333; Space for s1 is allocated at fp-24
334; FAST: sub x[[A:[0-9]+]], x29, #24
335; Call memcpy with size = 24 (0x18)
336; FAST: mov {{x[0-9]+}}, #24
337; FAST: bl _memcpy
338; Space for s2 is allocated at sp+32
339; FAST: add x[[B:[0-9]+]], sp, #32
340; Call memcpy again
341; FAST: bl _memcpy
342; Address of s1 is passed on stack at sp+8
343; FAST: str {{w[0-9]+}}, [sp]
344; FAST: str {{x[0-9]+}}, [sp, #8]
345; FAST: str {{x[0-9]+}}, [sp, #16]
346  %tmp = alloca %struct.s42, align 4
347  %tmp1 = alloca %struct.s42, align 4
348  call void @llvm.memcpy.p0.p0.i64(ptr align 4 %tmp, ptr align 4 @g42, i64 24, i1 false), !tbaa.struct !4
349  call void @llvm.memcpy.p0.p0.i64(ptr align 4 %tmp1, ptr align 4 @g42_2, i64 24, i1 false), !tbaa.struct !4
350  %call = call i32 @f42_stack(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7,
351                       i32 8, i32 9, ptr %tmp, ptr %tmp1) #5
352  ret i32 %call
353}
354
355; structs with size of 22 bytes, alignment of 16
356; passed indirectly in x1 and x2
357define i32 @f43(i32 %i, ptr nocapture %s1, ptr nocapture %s2) #2 {
358entry:
359; CHECK-LABEL: f43
360; CHECK: ldr w[[A:[0-9]+]], [x1]
361; CHECK: ldr w[[B:[0-9]+]], [x2]
362; CHECK: add w[[C:[0-9]+]], w[[A]], w0
363; CHECK: add {{w[0-9]+}}, w[[C]], w[[B]]
364; FAST-LABEL: f43
365; FAST: ldr w[[A:[0-9]+]], [x1]
366; FAST: ldr w[[B:[0-9]+]], [x2]
367; FAST: add w[[C:[0-9]+]], w[[A]], w0
368; FAST: add {{w[0-9]+}}, w[[C]], w[[B]]
369  %0 = load i32, ptr %s1, align 4, !tbaa !0
370  %1 = load i32, ptr %s2, align 4, !tbaa !0
371  %s = getelementptr inbounds %struct.s43, ptr %s1, i64 0, i32 1
372  %2 = load i16, ptr %s, align 2, !tbaa !3
373  %conv = sext i16 %2 to i32
374  %s5 = getelementptr inbounds %struct.s43, ptr %s2, i64 0, i32 1
375  %3 = load i16, ptr %s5, align 2, !tbaa !3
376  %conv6 = sext i16 %3 to i32
377  %add = add i32 %0, %i
378  %add3 = add i32 %add, %1
379  %add4 = add i32 %add3, %conv
380  %add7 = add i32 %add4, %conv6
381  ret i32 %add7
382}
383
384define i32 @caller43() #3 {
385entry:
386; CHECK-LABEL: caller43
387; CHECK-DAG: stp q1, q0, [sp, #32]
388; CHECK-DAG: stp q1, q0, [sp]
389; CHECK: add x1, sp, #32
390; CHECK: mov x2, sp
391; Space for s1 is allocated at sp+32
392; Space for s2 is allocated at sp
393
394; FAST-LABEL: caller43
395; FAST: add x29, sp, #64
396; Space for s1 is allocated at sp+32
397; Space for s2 is allocated at sp
398; FAST: str {{x[0-9]+}}, [sp, #32]
399; FAST: str {{x[0-9]+}}, [sp, #40]
400; FAST: str {{x[0-9]+}}, [sp, #48]
401; FAST: str {{x[0-9]+}}, [sp, #56]
402; FAST: str {{x[0-9]+}}, [sp]
403; FAST: str {{x[0-9]+}}, [sp, #8]
404; FAST: str {{x[0-9]+}}, [sp, #16]
405; FAST: str {{x[0-9]+}}, [sp, #24]
406; FAST: add x1, sp, #32
407; FAST: mov x2, sp
408  %tmp = alloca %struct.s43, align 16
409  %tmp1 = alloca %struct.s43, align 16
410  call void @llvm.memcpy.p0.p0.i64(ptr align 16 %tmp, ptr align 16 @g43, i64 32, i1 false), !tbaa.struct !4
411  call void @llvm.memcpy.p0.p0.i64(ptr align 16 %tmp1, ptr align 16 @g43_2, i64 32, i1 false), !tbaa.struct !4
412  %call = call i32 @f43(i32 3, ptr %tmp, ptr %tmp1) #5
413  ret i32 %call
414}
415
416declare i32 @f43_stack(i32 %i, i32 %i2, i32 %i3, i32 %i4, i32 %i5, i32 %i6,
417                       i32 %i7, i32 %i8, i32 %i9, ptr nocapture %s1,
418                       ptr nocapture %s2) #2
419
420define i32 @caller43_stack() #3 {
421entry:
422; CHECK-LABEL: caller43_stack
423; CHECK: sub sp, sp, #112
424; CHECK: add x29, sp, #96
425; CHECK-DAG: stp q1, q0, [x29, #-32]
426; CHECK-DAG: stp q1, q0, [sp, #32]
427; Space for s1 is allocated at x29-32 = sp+64
428; Space for s2 is allocated at sp+32
429; CHECK: add x[[B:[0-9]+]], sp, #32
430; CHECK: str x[[B]], [sp, #16]
431; CHECK: sub x[[A:[0-9]+]], x29, #32
432; Address of s1 is passed on stack at sp+8
433; CHECK: str x[[A]], [sp, #8]
434; CHECK: mov w[[C:[0-9]+]], #9
435; CHECK: str w[[C]], [sp]
436
437; FAST-LABEL: caller43_stack
438; FAST: sub sp, sp, #112
439; Space for s1 is allocated at fp-32 = sp+64
440; Space for s2 is allocated at sp+32
441; FAST: stur {{x[0-9]+}}, [x29, #-32]
442; FAST: stur {{x[0-9]+}}, [x29, #-24]
443; FAST: stur {{x[0-9]+}}, [x29, #-16]
444; FAST: stur {{x[0-9]+}}, [x29, #-8]
445; FAST: str {{x[0-9]+}}, [sp, #32]
446; FAST: str {{x[0-9]+}}, [sp, #40]
447; FAST: str {{x[0-9]+}}, [sp, #48]
448; FAST: str {{x[0-9]+}}, [sp, #56]
449; Address of s1 is passed on stack at sp+8
450; FAST: sub x[[A:[0-9]+]], x29, #32
451; FAST: add x[[B:[0-9]+]], sp, #32
452; FAST: str {{w[0-9]+}}, [sp]
453; FAST: str x[[A]], [sp, #8]
454; FAST: str x[[B]], [sp, #16]
455  %tmp = alloca %struct.s43, align 16
456  %tmp1 = alloca %struct.s43, align 16
457  call void @llvm.memcpy.p0.p0.i64(ptr align 16 %tmp, ptr align 16 @g43, i64 32, i1 false), !tbaa.struct !4
458  call void @llvm.memcpy.p0.p0.i64(ptr align 16 %tmp1, ptr align 16 @g43_2, i64 32, i1 false), !tbaa.struct !4
459  %call = call i32 @f43_stack(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7,
460                       i32 8, i32 9, ptr %tmp, ptr %tmp1) #5
461  ret i32 %call
462}
463
464; rdar://13668927
465; Check that we don't split an i128.
466declare i32 @callee_i128_split(i32 %i, i32 %i2, i32 %i3, i32 %i4, i32 %i5,
467                               i32 %i6, i32 %i7, i128 %s1, i32 %i8)
468
469define i32 @i128_split() {
470entry:
471; CHECK-LABEL: i128_split
472; "i128 %0" should be on stack at [sp].
473; "i32 8" should be on stack at [sp, #16].
474; CHECK: str {{w[0-9]+}}, [sp, #16]
475; CHECK: stp {{x[0-9]+}}, {{x[0-9]+}}, [sp]
476; FAST-LABEL: i128_split
477; FAST: sub sp, sp
478; FAST: mov x[[ADDR:[0-9]+]], sp
479; FAST: str {{w[0-9]+}}, [x[[ADDR]], #16]
480; Load/Store opt is disabled with -O0, so the i128 is split.
481; FAST: str {{x[0-9]+}}, [x[[ADDR]], #8]
482; FAST: str {{x[0-9]+}}, [x[[ADDR]]]
483  %0 = load i128, ptr @g41, align 16
484  %call = tail call i32 @callee_i128_split(i32 1, i32 2, i32 3, i32 4, i32 5,
485                                           i32 6, i32 7, i128 %0, i32 8) #5
486  ret i32 %call
487}
488
489declare i32 @callee_i64(i32 %i, i32 %i2, i32 %i3, i32 %i4, i32 %i5,
490                               i32 %i6, i32 %i7, i64 %s1, i32 %i8)
491
492define i32 @i64_split() {
493entry:
494; CHECK-LABEL: i64_split
495; "i64 %0" should be in register x7.
496; "i32 8" should be on stack at [sp].
497; CHECK: ldr x7, [{{x[0-9]+}}]
498; CHECK: str {{w[0-9]+}}, [sp]
499; FAST-LABEL: i64_split
500; FAST: ldr x7, [{{x[0-9]+}}]
501; FAST: mov x[[R0:[0-9]+]], sp
502; FAST: mov w[[R1:[0-9]+]], #8
503; FAST: str w[[R1]], [x[[R0]]]
504  %0 = load i64, ptr @g41, align 16
505  %call = tail call i32 @callee_i64(i32 1, i32 2, i32 3, i32 4, i32 5,
506                                    i32 6, i32 7, i64 %0, i32 8) #5
507  ret i32 %call
508}
509
510attributes #0 = { noinline nounwind readnone "fp-contract-model"="standard" "relocation-model"="pic" "ssp-buffers-size"="8" }
511attributes #1 = { nounwind readonly "fp-contract-model"="standard" "relocation-model"="pic" "ssp-buffers-size"="8" }
512attributes #2 = { noinline nounwind readonly "fp-contract-model"="standard" "relocation-model"="pic" "ssp-buffers-size"="8" }
513attributes #3 = { nounwind "fp-contract-model"="standard" "relocation-model"="pic" "ssp-buffers-size"="8" }
514attributes #4 = { nounwind }
515attributes #5 = { nobuiltin }
516
517!0 = !{!"int", !1}
518!1 = !{!"omnipotent char", !2}
519!2 = !{!"Simple C/C++ TBAA"}
520!3 = !{!"short", !1}
521!4 = !{i64 0, i64 4, !0, i64 4, i64 2, !3, i64 8, i64 4, !0, i64 12, i64 2, !3, i64 16, i64 4, !0, i64 20, i64 2, !3}
522