RFR: 7903461: JMH: perfasm profiler misses some jump edges
Aleksey Shipilev
shade at openjdk.org
Fri Apr 21 12:24:19 UTC 2023
On some platforms, JMH perfasm does not render the jump edges properly, because it fails to parse the address lines correctly in the presence of oddly formatted disassembly output. This happens almost reliably on AArch64 hosts:
....[Hottest Region 1]..............................................................................
c2, level 4, org.openjdk.jmh.samples.jmh_generated.EnumBench_testMethod_jmhTest::testMethod_avgt_jmhStub, version 6, compile id 778
0x0000ffff9c5c7fd4: nop ; {other}
0x0000ffff9c5c7fd8: movk xzr, #0x2c4
0x0000ffff9c5c7fdc: movk xzr, #0x0
0x0000ffff9c5c7fe0: mov x19, x29
0x0000ffff9c5c7fe4: ldr x14, [sp, #64]
0x0000ffff9c5c7fe8: ldp x13, x12, [sp]
0x0000ffff9c5c7fec: ldp x20, x15, [sp, #16] ;*invokespecial hashCode {reexecute=0 rethrow=0 return_oop=0}
; - java.lang.Enum::hashCode at 1 (line 175)
; - org.openjdk.jmh.samples.EnumBench::testMethod at 3 (line 52)
; - org.openjdk.jmh.samples.jmh_generated.EnumBench_testMethod_jmhTest::testMethod_avgt_jmhStub at 17 (line 186)
3.10% 0x0000ffff9c5c7ff0: ldarb w11, [x15] ;*getfield isDone {reexecute=0 rethrow=0 return_oop=0}
; - org.openjdk.jmh.samples.jmh_generated.EnumBench_testMethod_jmhTest::testMethod_avgt_jmhStub at 30 (line 188)
0x0000ffff9c5c7ff4: ldr x10, [x28, #896]
0x0000ffff9c5c7ff8: add x20, x20, #0x1 ; ImmutableOopMap {r12=Oop r13=Oop r15=Derived_oop_r13 r14=Oop r19=Oop }
;*ifeq {reexecute=1 rethrow=0 return_oop=0}
; - (reexecute) org.openjdk.jmh.samples.jmh_generated.EnumBench_testMethod_jmhTest::testMethod_avgt_jmhStub at 33 (line 188)
0x0000ffff9c5c7ffc: ldr wzr, [x10] ; {poll}
1.47% ╭ 0x0000ffff9c5c8000: cbnz w11, 0x0000ffff9c5c8034 ;*aload_1 {reexecute=0 rethrow=0 return_oop=0}
│ ; - org.openjdk.jmh.samples.jmh_generated.EnumBench_testMethod_jmhTest::testMethod_avgt_jmhStub at 36 (line 189)
10.53% │ 0x0000ffff9c5c8004: mov x10, #0xcce0 // #52448
│ ; {oop(a 'org/openjdk/jmh/samples/EnumBench$E'{0x0000000465dfcce0})}
│ 0x0000ffff9c5c8008: movk x10, #0x65df, lsl #16
12.19% │ 0x0000ffff9c5c800c: movk x10, #0x4, lsl #32
│ 0x0000ffff9c5c8010: ldr x10, [x10]
6.97% │ 0x0000ffff9c5c8014: and x11, x10, #0x3
│ 0x0000ffff9c5c8018: cmp x11, #0x1
│ 0x0000ffff9c5c801c: b.ne 0x0000ffff9c5c7fb4 // b.any
38.49% │ 0x0000ffff9c5c8020: lsr x10, x10, #8
│ 0x0000ffff9c5c8024: and w0, w10, #0x7fffffff
│ 0x0000ffff9c5c8028: cbz w0, 0x0000ffff9c5c7fb4 ;*invokespecial hashCode {reexecute=0 rethrow=0 return_oop=0}
│ ; - java.lang.Enum::hashCode at 1 (line 175)
│ ; - org.openjdk.jmh.samples.EnumBench::testMethod at 3 (line 52)
│ ; - org.openjdk.jmh.samples.jmh_generated.EnumBench_testMethod_jmhTest::testMethod_avgt_jmhStub at 17 (line 186)
23.96% │ 0x0000ffff9c5c802c: b 0x0000ffff9c5c7ff0
│ 0x0000ffff9c5c8030: orr x20, xzr, #0x1 ;*aload_1 {reexecute=0 rethrow=0 return_oop=0}
│ ; - org.openjdk.jmh.samples.jmh_generated.EnumBench_testMethod_jmhTest::testMethod_avgt_jmhStub at 36 (line 189)
↘ 0x0000ffff9c5c8034: adr x9, 0x0000ffff9c5c804c
0x0000ffff9c5c8038: mov x8, #0xb4c8 // #46280
; {runtime_call os::javaTimeNanos()}
0x0000ffff9c5c803c: movk x8, #0xb487, lsl #16
0x0000ffff9c5c8040: movk x8, #0xffff, lsl #32
....................................................................................................
Back branch to `0x0000ffff9c5c7ff0` is missing, for example.
-------------
Commit messages:
- Fixes
Changes: https://git.openjdk.org/jmh/pull/100/files
Webrev: https://webrevs.openjdk.org/?repo=jmh&pr=100&range=00
Issue: https://bugs.openjdk.org/browse/CODETOOLS-7903461
Stats: 167 lines in 2 files changed: 124 ins; 20 del; 23 mod
Patch: https://git.openjdk.org/jmh/pull/100.diff
Fetch: git fetch https://git.openjdk.org/jmh.git pull/100/head:pull/100
PR: https://git.openjdk.org/jmh/pull/100
More information about the jmh-dev
mailing list