1; RUN: llc                             < %s | FileCheck %s
2; RUN: llc -fast-isel -fast-isel-abort < %s | FileCheck %s
3target datalayout = "E-m:e-i64:64-n32:64"
4target triple = "powerpc64-unknown-linux-gnu"
5
6; Trivial patchpoint codegen
7;
8define i64 @trivial_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) {
9entry:
10; CHECK-LABEL: trivial_patchpoint_codegen:
11
12; CHECK: li 11, -8531
13; CHECK-NEXT: rldic 11, 11, 32, 16
14; CHECK-NEXT: oris 11, 11, 48879
15; CHECK-NEXT: ori 11, 11, 51966
16; CHECK-NEXT: mtctr 11
17; CHECK-NEXT: bctrl
18
19; CHECK: li 11, -8531
20; CHECK-NEXT: rldic 11, 11, 32, 16
21; CHECK-NEXT: oris 11, 11, 48879
22; CHECK-NEXT: ori 11, 11, 51967
23; CHECK-NEXT: mtctr 11
24; CHECK-NEXT: bctrl
25
26; CHECK: blr
27
28  %resolveCall2 = inttoptr i64 244837814094590 to i8*
29  %result = tail call i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 2, i32 24, i8* %resolveCall2, i32 4, i64 %p1, i64 %p2, i64 %p3, i64 %p4)
30  %resolveCall3 = inttoptr i64 244837814094591 to i8*
31  tail call void (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i64 3, i32 24, i8* %resolveCall3, i32 2, i64 %p1, i64 %result)
32  ret i64 %result
33}
34
35; Caller frame metadata with stackmaps. This should not be optimized
36; as a leaf function.
37;
38; CHECK-LABEL: caller_meta_leaf
39; CHECK: stdu 1, -80(1)
40; CHECK: Ltmp
41; CHECK: addi 1, 1, 80
42; CHECK: blr
43
44define void @caller_meta_leaf() {
45entry:
46  %metadata = alloca i64, i32 3, align 8
47  store i64 11, i64* %metadata
48  store i64 12, i64* %metadata
49  store i64 13, i64* %metadata
50  call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 4, i32 0, i64* %metadata)
51  ret void
52}
53
54; Test patchpoints reusing the same TargetConstant.
55; <rdar:15390785> Assertion failed: (CI.getNumArgOperands() >= NumArgs + 4)
56; There is no way to verify this, since it depends on memory allocation.
57; But I think it's useful to include as a working example.
58define i64 @testLowerConstant(i64 %arg, i64 %tmp2, i64 %tmp10, i64* %tmp33, i64 %tmp79) {
59entry:
60  %tmp80 = add i64 %tmp79, -16
61  %tmp81 = inttoptr i64 %tmp80 to i64*
62  %tmp82 = load i64* %tmp81, align 8
63  tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 14, i32 8, i64 %arg, i64 %tmp2, i64 %tmp10, i64 %tmp82)
64  tail call void (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i64 15, i32 32, i8* null, i32 3, i64 %arg, i64 %tmp10, i64 %tmp82)
65  %tmp83 = load i64* %tmp33, align 8
66  %tmp84 = add i64 %tmp83, -24
67  %tmp85 = inttoptr i64 %tmp84 to i64*
68  %tmp86 = load i64* %tmp85, align 8
69  tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 17, i32 8, i64 %arg, i64 %tmp10, i64 %tmp86)
70  tail call void (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i64 18, i32 32, i8* null, i32 3, i64 %arg, i64 %tmp10, i64 %tmp86)
71  ret i64 10
72}
73
74; Test small patchpoints that don't emit calls.
75define void @small_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) {
76entry:
77; CHECK-LABEL: small_patchpoint_codegen:
78; CHECK:      Ltmp
79; CHECK:      nop
80; CHECK-NEXT: nop
81; CHECK-NEXT: nop
82; CHECK-NEXT: nop
83; CHECK-NEXT: nop
84; CHECK-NOT:  nop
85; CHECK: blr
86  %result = tail call i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 5, i32 20, i8* null, i32 2, i64 %p1, i64 %p2)
87  ret void
88}
89
90declare void @llvm.experimental.stackmap(i64, i32, ...)
91declare void @llvm.experimental.patchpoint.void(i64, i32, i8*, i32, ...)
92declare i64 @llvm.experimental.patchpoint.i64(i64, i32, i8*, i32, ...)
93
94