xref: /llvm-project/llvm/test/CodeGen/X86/unaligned-spill-folding.ll (revision 2f448bf509432c1a19ec46ab8cbc7353c03c6280)
1; RUN: split-file %s %t
2; RUN: cat %t/main.ll %t/align4.ll > %t/a2.ll
3; RUN: cat %t/main.ll %t/align16.ll > %t/b2.ll
4; RUN: llc -mtriple=i386-unknown-freebsd -mcpu=core2 -relocation-model=pic < %t/a2.ll | FileCheck %s -check-prefix=UNALIGNED
5; RUN: llc -mtriple=i386-unknown-freebsd -mcpu=core2 -relocation-model=pic < %t/b2.ll | FileCheck %s -check-prefix=ALIGNED
6; RUN: llc -mtriple=i386-unknown-freebsd -mcpu=core2 -stackrealign -relocation-model=pic < %t/a2.ll | FileCheck %s -check-prefix=FORCEALIGNED
7
8;--- main.ll
9@arr = internal unnamed_addr global [32 x i32] zeroinitializer, align 16
10
11; PR12250
12define i32 @test1() {
13vector.ph:
14  br label %vector.body
15
16vector.body:
17  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
18  %0 = getelementptr inbounds [32 x i32], ptr @arr, i32 0, i32 %index
19  %wide.load = load <4 x i32>, ptr %0, align 16
20  %1 = add nsw <4 x i32> %wide.load, <i32 10, i32 10, i32 10, i32 10>
21  %2 = xor <4 x i32> %1, <i32 123345, i32 123345, i32 123345, i32 123345>
22  %3 = add nsw <4 x i32> %2, <i32 112, i32 112, i32 112, i32 112>
23  %4 = xor <4 x i32> %3, <i32 543345, i32 543345, i32 543345, i32 543345>
24  %5 = add nsw <4 x i32> %4, <i32 73, i32 73, i32 73, i32 73>
25  %6 = xor <4 x i32> %5, <i32 345987, i32 345987, i32 345987, i32 345987>
26  %7 = add nsw <4 x i32> %6, <i32 48, i32 48, i32 48, i32 48>
27  %8 = xor <4 x i32> %7, <i32 123987, i32 123987, i32 123987, i32 123987>
28  store <4 x i32> %8, ptr %0, align 16
29  %index.next = add i32 %index, 4
30  %9 = icmp eq i32 %index.next, 32
31  br i1 %9, label %middle.block, label %vector.body
32
33middle.block:
34  ret i32 0
35
36; We can't fold the spill into a padd unless the stack is aligned. Just spilling
37; doesn't force stack realignment though
38; UNALIGNED-LABEL: @test1
39; UNALIGNED-NOT: andl $-{{..}}, %esp
40; UNALIGNED: movdqu {{.*}} # 16-byte Spill
41; UNALIGNED-NOT: paddd {{.*}} # 16-byte Folded Reload
42
43; ALIGNED-LABEL: @test1
44; ALIGNED-NOT: andl $-{{..}}, %esp
45; ALIGNED: movdqa {{.*}} # 16-byte Spill
46; ALIGNED: paddd {{.*}} # 16-byte Folded Reload
47
48; FORCEALIGNED-LABEL: @test1
49; FORCEALIGNED: andl $-{{..}}, %esp
50; FORCEALIGNED: movdqa {{.*}} # 16-byte Spill
51; FORCEALIGNED: paddd {{.*}} # 16-byte Folded Reload
52}
53!llvm.module.flags = !{!0}
54;--- align4.ll
55!0 = !{i32 2, !"override-stack-alignment", i32 4}
56;--- align16.ll
57!0 = !{i32 2, !"override-stack-alignment", i32 16}
58