RFR: 7903461: JMH: perfasm profiler misses some jump edges [v2]

[Aleksey Shipilev]

On Fri, 21 Apr 2023 12:45:06 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

>> 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.
>
> Aleksey Shipilev has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Better trim handling

Thanks for review, I ran it with my daily performance/perfasm chores, and it seems to work reliably.

-------------

PR Comment: https://git.openjdk.org/jmh/pull/100#issuecomment-1518056683