xref: /llvm-project/llvm/test/CodeGen/AArch64/arm64-patchpoint.ll (revision 308ed0233a3da5b3a7d646808c6470ac30870603)
1; RUN: llc -mtriple=arm64-apple-darwin -debug-entry-values -enable-misched=0 -mcpu=cyclone                             < %s | FileCheck %s
2; RUN: llc -mtriple=arm64-apple-darwin -debug-entry-values -enable-misched=0 -mcpu=cyclone -fast-isel -fast-isel-abort=1 < %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:       mov  x16, #244834610708480
10; CHECK-NEXT:  movk x16, #48879, lsl #16
11; CHECK-NEXT:  movk x16, #51966
12; CHECK-NEXT:  blr  x16
13; CHECK:       mov  x16, #244834610708480
14; CHECK-NEXT:  movk x16, #48879, lsl #16
15; CHECK-NEXT:  movk x16, #51967
16; CHECK-NEXT:  blr  x16
17; CHECK:       ret
18  %resolveCall2 = inttoptr i64 244837814094590 to ptr
19  %result = tail call i64 (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.i64(i64 2, i32 20, ptr %resolveCall2, i32 4, i64 %p1, i64 %p2, i64 %p3, i64 %p4)
20  %resolveCall3 = inttoptr i64 244837814094591 to ptr
21  tail call void (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.void(i64 3, i32 20, ptr %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:       sub sp, sp, #48
30; CHECK-NEXT:  stp x29, x30, [sp, #32]
31; CHECK-NEXT:  add x29, sp, #32
32; CHECK:       Ltmp
33; CHECK:       add sp, sp, #48
34; CHECK:       ret
35
36define void @caller_meta_leaf() {
37entry:
38  %metadata = alloca i64, i32 3, align 8
39  store i64 11, ptr %metadata
40  store i64 12, ptr %metadata
41  store i64 13, ptr %metadata
42  call void (i64, i32, ...) @llvm.experimental.stackmap(i64 4, i32 0, ptr %metadata)
43  ret void
44}
45
46; Test patchpoints reusing the same TargetConstant.
47; <rdar:15390785> Assertion failed: (CI.getNumArgOperands() >= NumArgs + 4)
48; There is no way to verify this, since it depends on memory allocation.
49; But I think it's useful to include as a working example.
50define i64 @testLowerConstant(i64 %arg, i64 %tmp2, i64 %tmp10, ptr %tmp33, i64 %tmp79) {
51entry:
52  %tmp80 = add i64 %tmp79, -16
53  %tmp81 = inttoptr i64 %tmp80 to ptr
54  %tmp82 = load i64, ptr %tmp81, align 8
55  tail call void (i64, i32, ...) @llvm.experimental.stackmap(i64 14, i32 8, i64 %arg, i64 %tmp2, i64 %tmp10, i64 %tmp82)
56  tail call void (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.void(i64 15, i32 32, ptr null, i32 3, i64 %arg, i64 %tmp10, i64 %tmp82)
57  %tmp83 = load i64, ptr %tmp33, align 8
58  %tmp84 = add i64 %tmp83, -24
59  %tmp85 = inttoptr i64 %tmp84 to ptr
60  %tmp86 = load i64, ptr %tmp85, align 8
61  tail call void (i64, i32, ...) @llvm.experimental.stackmap(i64 17, i32 8, i64 %arg, i64 %tmp10, i64 %tmp86)
62  tail call void (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.void(i64 18, i32 32, ptr null, i32 3, i64 %arg, i64 %tmp10, i64 %tmp86)
63  ret i64 10
64}
65
66; Test small patchpoints that don't emit calls.
67define void @small_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) {
68entry:
69; CHECK-LABEL: small_patchpoint_codegen:
70; CHECK:      Ltmp
71; CHECK:      nop
72; CHECK-NEXT: nop
73; CHECK-NEXT: nop
74; CHECK-NEXT: nop
75; CHECK-NEXT: nop
76; CHECK-NEXT: ldp
77; CHECK-NEXT: ret
78  %result = tail call i64 (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.i64(i64 5, i32 20, ptr null, i32 2, i64 %p1, i64 %p2)
79  ret void
80}
81
82; Test register allocation for an i32 result value of patchpoint.
83define i32 @generic_patchpoint_i32() {
84entry:
85; CHECK-LABEL: generic_patchpoint_i32:
86; CHECK:      Ltmp
87; CHECK-NEXT: nop
88; The return value is already in w0.
89; CHECK-NEXT: ldp
90; CHECK-NEXT: ret
91  %result = tail call i32 (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.i32(i64 5, i32 4, ptr null, i32 0)
92  ret i32 %result
93}
94
95; Test register allocation for an i64 result value of patchpoint.
96define i64 @generic_patchpoint_i64() {
97entry:
98; CHECK-LABEL: generic_patchpoint_i64:
99; CHECK:      Ltmp
100; CHECK-NEXT: nop
101; The return value is already in x0.
102; CHECK-NEXT: ldp
103; CHECK-NEXT: ret
104  %result = tail call i64 (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.i64(i64 5, i32 4, ptr null, i32 0)
105  ret i64 %result
106}
107
108; Test register allocation for a ptr result value of patchpoint.
109define ptr @generic_patchpoint_p0() {
110entry:
111; CHECK-LABEL: generic_patchpoint_p0:
112; CHECK:      Ltmp
113; CHECK-NEXT: nop
114; The return value is already in x0.
115; CHECK-NEXT: ldp
116; CHECK-NEXT: ret
117  %result = tail call ptr (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.p0(i64 5, i32 4, ptr null, i32 0)
118  ret ptr %result
119}
120
121; Test register allocation for a half result value of patchpoint.
122define half @generic_patchpoint_f16() {
123entry:
124; CHECK-LABEL: generic_patchpoint_f16:
125; CHECK:      Ltmp
126; CHECK-NEXT: nop
127; The return value is already in h0.
128; CHECK-NEXT: ldp
129; CHECK-NEXT: ret
130  %result = tail call half (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.f16(i64 5, i32 4, ptr null, i32 0)
131  ret half %result
132}
133
134; Test register allocation for a float result value of patchpoint.
135define float @generic_patchpoint_f32() {
136entry:
137; CHECK-LABEL: generic_patchpoint_f32:
138; CHECK:      Ltmp
139; CHECK-NEXT: nop
140; The return value is already in s0.
141; CHECK-NEXT: ldp
142; CHECK-NEXT: ret
143  %result = tail call float (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.f32(i64 5, i32 4, ptr null, i32 0)
144  ret float %result
145}
146
147; Test register allocation for a double result value of patchpoint.
148define double @generic_patchpoint_f64() {
149entry:
150; CHECK-LABEL: generic_patchpoint_f64:
151; CHECK:      Ltmp
152; CHECK-NEXT: nop
153; The return value is already in d0.
154; CHECK-NEXT: ldp
155; CHECK-NEXT: ret
156  %result = tail call double (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.f64(i64 5, i32 4, ptr null, i32 0)
157  ret double %result
158}
159
160; Test register allocation for a <16 x i8> result value of patchpoint.
161define <16 x i8> @generic_patchpoint_v16i8() {
162entry:
163; CHECK-LABEL: generic_patchpoint_v16i8:
164; CHECK:      Ltmp
165; CHECK-NEXT: nop
166; The return value is already in v0.16b.
167; CHECK-NEXT: ldp
168; CHECK-NEXT: ret
169  %result = tail call <16 x i8> (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.v16i8(i64 5, i32 4, ptr null, i32 0)
170  ret <16 x i8> %result
171}
172
173; Test register allocation for a <4 x i32> result value of patchpoint.
174define <4 x i32> @generic_patchpoint_v4i32() {
175entry:
176; CHECK-LABEL: generic_patchpoint_v4i32:
177; CHECK:      Ltmp
178; CHECK-NEXT: nop
179; The return value is already in v0.4s.
180; CHECK-NEXT: ldp
181; CHECK-NEXT: ret
182  %result = tail call <4 x i32> (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.v4i32(i64 5, i32 4, ptr null, i32 0)
183  ret <4 x i32> %result
184}
185
186; Test register allocation for a <4 x float> result value of patchpoint.
187define <4 x float> @generic_patchpoint_v4f32() {
188entry:
189; CHECK-LABEL: generic_patchpoint_v4f32:
190; CHECK:      Ltmp
191; CHECK-NEXT: nop
192; The return value is already in v0.4s.
193; CHECK-NEXT: ldp
194; CHECK-NEXT: ret
195  %result = tail call <4 x float> (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.v4f32(i64 5, i32 4, ptr null, i32 0)
196  ret <4 x float> %result
197}
198
199; Test register allocation for a <2 x double> result value of patchpoint.
200define <2 x double> @generic_patchpoint_v2f64() {
201entry:
202; CHECK-LABEL: generic_patchpoint_v2f64:
203; CHECK:      Ltmp
204; CHECK-NEXT: nop
205; The return value is already in v0.2d.
206; CHECK-NEXT: ldp
207; CHECK-NEXT: ret
208  %result = tail call <2 x double> (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.v2f64(i64 5, i32 4, ptr null, i32 0)
209  ret <2 x double> %result
210}
211
212declare void @llvm.experimental.stackmap(i64, i32, ...)
213declare void @llvm.experimental.patchpoint.void(i64, i32, ptr, i32, ...)
214declare i32 @llvm.experimental.patchpoint.i32(i64, i32, ptr, i32, ...)
215declare i64 @llvm.experimental.patchpoint.i64(i64, i32, ptr, i32, ...)
216declare ptr @llvm.experimental.patchpoint.p0(i64, i32, ptr, i32, ...)
217declare half @llvm.experimental.patchpoint.f16(i64, i32, ptr, i32, ...)
218declare float @llvm.experimental.patchpoint.f32(i64, i32, ptr, i32, ...)
219declare double @llvm.experimental.patchpoint.f64(i64, i32, ptr, i32, ...)
220declare <16 x i8> @llvm.experimental.patchpoint.v16i8(i64, i32, ptr, i32, ...)
221declare <4 x i32> @llvm.experimental.patchpoint.v4i32(i64, i32, ptr, i32, ...)
222declare <4 x float> @llvm.experimental.patchpoint.v4f32(i64, i32, ptr, i32, ...)
223declare <2 x double> @llvm.experimental.patchpoint.v2f64(i64, i32, ptr, i32, ...)
224