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
4 ; Trivial patchpoint codegen
6 define i64 @trivial_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) {
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
13 ; CHECK: mov x16, #244834610708480
14 ; CHECK-NEXT: movk x16, #48879, lsl #16
15 ; CHECK-NEXT: movk x16, #51967
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)
25 ; Caller frame metadata with stackmaps. This should not be optimized
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
33 ; CHECK: add sp, sp, #48
36 define void @caller_meta_leaf() {
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)
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.
50 define i64 @testLowerConstant(i64 %arg, i64 %tmp2, i64 %tmp10, ptr %tmp33, i64 %tmp79) {
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)
66 ; Test small patchpoints that don't emit calls.
67 define void @small_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) {
69 ; CHECK-LABEL: small_patchpoint_codegen:
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)
82 ; Test register allocation for an i32 result value of patchpoint.
83 define i32 @generic_patchpoint_i32() {
85 ; CHECK-LABEL: generic_patchpoint_i32:
88 ; The return value is already in w0.
91 %result = tail call i32 (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.i32(i64 5, i32 4, ptr null, i32 0)
95 ; Test register allocation for an i64 result value of patchpoint.
96 define i64 @generic_patchpoint_i64() {
98 ; CHECK-LABEL: generic_patchpoint_i64:
101 ; The return value is already in x0.
104 %result = tail call i64 (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.i64(i64 5, i32 4, ptr null, i32 0)
108 ; Test register allocation for a ptr result value of patchpoint.
109 define ptr @generic_patchpoint_p0() {
111 ; CHECK-LABEL: generic_patchpoint_p0:
114 ; The return value is already in x0.
117 %result = tail call ptr (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.p0(i64 5, i32 4, ptr null, i32 0)
121 ; Test register allocation for a half result value of patchpoint.
122 define half @generic_patchpoint_f16() {
124 ; CHECK-LABEL: generic_patchpoint_f16:
127 ; The return value is already in h0.
130 %result = tail call half (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.f16(i64 5, i32 4, ptr null, i32 0)
134 ; Test register allocation for a float result value of patchpoint.
135 define float @generic_patchpoint_f32() {
137 ; CHECK-LABEL: generic_patchpoint_f32:
140 ; The return value is already in s0.
143 %result = tail call float (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.f32(i64 5, i32 4, ptr null, i32 0)
147 ; Test register allocation for a double result value of patchpoint.
148 define double @generic_patchpoint_f64() {
150 ; CHECK-LABEL: generic_patchpoint_f64:
153 ; The return value is already in d0.
156 %result = tail call double (i64, i32, ptr, i32, ...) @llvm.experimental.patchpoint.f64(i64 5, i32 4, ptr null, i32 0)
160 ; Test register allocation for a <16 x i8> result value of patchpoint.
161 define <16 x i8> @generic_patchpoint_v16i8() {
163 ; CHECK-LABEL: generic_patchpoint_v16i8:
166 ; The return value is already in v0.16b.
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
173 ; Test register allocation for a <4 x i32> result value of patchpoint.
174 define <4 x i32> @generic_patchpoint_v4i32() {
176 ; CHECK-LABEL: generic_patchpoint_v4i32:
179 ; The return value is already in v0.4s.
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
186 ; Test register allocation for a <4 x float> result value of patchpoint.
187 define <4 x float> @generic_patchpoint_v4f32() {
189 ; CHECK-LABEL: generic_patchpoint_v4f32:
192 ; The return value is already in v0.4s.
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
199 ; Test register allocation for a <2 x double> result value of patchpoint.
200 define <2 x double> @generic_patchpoint_v2f64() {
202 ; CHECK-LABEL: generic_patchpoint_v2f64:
205 ; The return value is already in v0.2d.
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
212 declare void @llvm.experimental.stackmap(i64, i32, ...)
213 declare void @llvm.experimental.patchpoint.void(i64, i32, ptr, i32, ...)
214 declare i32 @llvm.experimental.patchpoint.i32(i64, i32, ptr, i32, ...)
215 declare i64 @llvm.experimental.patchpoint.i64(i64, i32, ptr, i32, ...)
216 declare ptr @llvm.experimental.patchpoint.p0(i64, i32, ptr, i32, ...)
217 declare half @llvm.experimental.patchpoint.f16(i64, i32, ptr, i32, ...)
218 declare float @llvm.experimental.patchpoint.f32(i64, i32, ptr, i32, ...)
219 declare double @llvm.experimental.patchpoint.f64(i64, i32, ptr, i32, ...)
220 declare <16 x i8> @llvm.experimental.patchpoint.v16i8(i64, i32, ptr, i32, ...)
221 declare <4 x i32> @llvm.experimental.patchpoint.v4i32(i64, i32, ptr, i32, ...)
222 declare <4 x float> @llvm.experimental.patchpoint.v4f32(i64, i32, ptr, i32, ...)
223 declare <2 x double> @llvm.experimental.patchpoint.v2f64(i64, i32, ptr, i32, ...)