1; RUN: llc < %s -mtriple=armv7-apple-ios -O0 | FileCheck %s
2
3; rdar://12713765
4; When realign-stack is set to false, make sure we are not creating stack
5; objects that are assumed to be 64-byte aligned.
6
7define void @test1(<16 x float>* noalias sret(<16 x float>) %agg.result) nounwind ssp "no-realign-stack" {
8; CHECK-LABEL: test1:
9; CHECK: mov r[[PTR:[0-9]+]], r{{[0-9]+}}
10; CHECK: mov r[[NOTALIGNED:[0-9]+]], sp
11; CHECK: add r[[NOTALIGNED]], r[[NOTALIGNED]], #32
12; CHECK: add r[[PTR]], r[[PTR]], #32
13; CHECK: vld1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r[[NOTALIGNED]]:128]
14; CHECK: vld1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r[[PTR]]:128]
15; CHECK: vst1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r[[PTR]]:128]
16; CHECK: vst1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r[[NOTALIGNED]]:128]
17entry:
18 %retval = alloca <16 x float>, align 64
19 %a1 = bitcast <16 x float>* %retval to float*
20 %a2 = getelementptr inbounds float, float* %a1, i64 8
21 %a3 = bitcast float* %a2 to <4 x float>*
22
23 %b1 = bitcast <16 x float>* %agg.result to float*
24 %b2 = getelementptr inbounds float, float* %b1, i64 8
25 %b3 = bitcast float* %b2 to <4 x float>*
26
27 %0 = load <4 x float>, <4 x float>* %a3, align 16
28 %1 = load <4 x float>, <4 x float>* %b3, align 16
29 store <4 x float> %0, <4 x float>* %b3, align 16
30 store <4 x float> %1, <4 x float>* %a3, align 16
31 ret void
32}
33
34define void @test2(<16 x float>* noalias sret(<16 x float>) %agg.result) nounwind ssp {
35; CHECK-LABEL: test2:
36; CHECK: mov r[[PTR:[0-9]+]], r{{[0-9]+}}
37; CHECK: mov r[[ALIGNED:[0-9]+]], sp
38; CHECK: orr r[[ALIGNED]], r[[ALIGNED]], #32
39; CHECK: add r[[PTR]], r[[PTR]], #32
40; CHECK: vld1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r[[ALIGNED]]:128]
41; CHECK: vld1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r[[PTR]]:128]
42; CHECK: vst1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r[[PTR]]:128]
43; CHECK: vst1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r[[ALIGNED]]:128]
44entry:
45 %retval = alloca <16 x float>, align 64
46 %a1 = bitcast <16 x float>* %retval to float*
47 %a2 = getelementptr inbounds float, float* %a1, i64 8
48 %a3 = bitcast float* %a2 to <4 x float>*
49
50 %b1 = bitcast <16 x float>* %agg.result to float*
51 %b2 = getelementptr inbounds float, float* %b1, i64 8
52 %b3 = bitcast float* %b2 to <4 x float>*
53
54 %0 = load <4 x float>, <4 x float>* %a3, align 16
55 %1 = load <4 x float>, <4 x float>* %b3, align 16
56 store <4 x float> %0, <4 x float>* %b3, align 16
57 store <4 x float> %1, <4 x float>* %a3, align 16
58 ret void
59}
60