1; RUN: llc -mtriple=arm64-apple-darwin -enable-misched=0 -mcpu=cyclone                             < %s | FileCheck %s
2; RUN: llc -mtriple=arm64-apple-darwin -enable-misched=0 -mcpu=cyclone -fast-isel -fast-isel-abort < %s | FileCheck %s
3
4; Trivial patchpoint codegen
5;
6define i64 @trivial_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) {
7entry:
8; CHECK-LABEL: trivial_patchpoint_codegen:
9; CHECK:       movz x16, #0xdead, lsl #32
10; CHECK-NEXT:  movk x16, #0xbeef, lsl #16
11; CHECK-NEXT:  movk x16, #0xcafe
12; CHECK-NEXT:  blr  x16
13; CHECK:       movz x16, #0xdead, lsl #32
14; CHECK-NEXT:  movk x16, #0xbeef, lsl #16
15; CHECK-NEXT:  movk x16, #0xcaff
16; CHECK-NEXT:  blr  x16
17; CHECK:       ret
18  %resolveCall2 = inttoptr i64 244837814094590 to i8*
19  %result = tail call i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 2, i32 20, i8* %resolveCall2, i32 4, i64 %p1, i64 %p2, i64 %p3, i64 %p4)
20  %resolveCall3 = inttoptr i64 244837814094591 to i8*
21  tail call void (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i64 3, i32 20, i8* %resolveCall3, i32 2, i64 %p1, i64 %result)
22  ret i64 %result
23}
24
25; Caller frame metadata with stackmaps. This should not be optimized
26; as a leaf function.
27;
28; CHECK-LABEL: caller_meta_leaf
29; CHECK:       mov x29, sp
30; CHECK-NEXT:  sub sp, sp, #32
31; CHECK:       Ltmp
32; CHECK:       mov sp, x29
33; CHECK:       ret
34
35define void @caller_meta_leaf() {
36entry:
37  %metadata = alloca i64, i32 3, align 8
38  store i64 11, i64* %metadata
39  store i64 12, i64* %metadata
40  store i64 13, i64* %metadata
41  call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 4, i32 0, i64* %metadata)
42  ret void
43}
44
45; Test patchpoints reusing the same TargetConstant.
46; <rdar:15390785> Assertion failed: (CI.getNumArgOperands() >= NumArgs + 4)
47; There is no way to verify this, since it depends on memory allocation.
48; But I think it's useful to include as a working example.
49define i64 @testLowerConstant(i64 %arg, i64 %tmp2, i64 %tmp10, i64* %tmp33, i64 %tmp79) {
50entry:
51  %tmp80 = add i64 %tmp79, -16
52  %tmp81 = inttoptr i64 %tmp80 to i64*
53  %tmp82 = load i64* %tmp81, align 8
54  tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 14, i32 8, i64 %arg, i64 %tmp2, i64 %tmp10, i64 %tmp82)
55  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)
56  %tmp83 = load i64* %tmp33, align 8
57  %tmp84 = add i64 %tmp83, -24
58  %tmp85 = inttoptr i64 %tmp84 to i64*
59  %tmp86 = load i64* %tmp85, align 8
60  tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 17, i32 8, i64 %arg, i64 %tmp10, i64 %tmp86)
61  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)
62  ret i64 10
63}
64
65; Test small patchpoints that don't emit calls.
66define void @small_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) {
67entry:
68; CHECK-LABEL: small_patchpoint_codegen:
69; CHECK:      Ltmp
70; CHECK:      nop
71; CHECK-NEXT: nop
72; CHECK-NEXT: nop
73; CHECK-NEXT: nop
74; CHECK-NEXT: nop
75; CHECK-NEXT: ldp
76; CHECK-NEXT: ret
77  %result = tail call i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 5, i32 20, i8* null, i32 2, i64 %p1, i64 %p2)
78  ret void
79}
80
81declare void @llvm.experimental.stackmap(i64, i32, ...)
82declare void @llvm.experimental.patchpoint.void(i64, i32, i8*, i32, ...)
83declare i64 @llvm.experimental.patchpoint.i64(i64, i32, i8*, i32, ...)
84
85