[ https://issues.apache.org/jira/browse/HARMONY-4714?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Aleksey Shipilev updated HARMONY-4714:
--------------------------------------

    Description: 
During JNI transition the exception flags are checked twice.

1. One check is generated in JNI lil stub (vm/vmcore/src/jit/compile.cpp:675)

    //***** Part 11: Rethrow exception
    cs = lil_parse_onto_end(cs,
                            "l0=ts;"
                            "ld l2,[l0+%0i:ref];"
                            "jc l2!=0,_exn_raised;"
                            "ld l2,[l0+%1i:ref];"
                            "jc l2=0,_no_exn;"
                            ":_exn_raised;"
                            "m2n_save_all;"
                            "out platform::void;"
                            "call.noret %2i;"
                            ":_no_exn;",
                            eoo, eco, exn_rethrow);
    assert(cs);

 2. Second check is done during pop_m2n frame generation: (ex.: vm/port/src/lil/ia32/pim/m2n_ia32.cpp:259)

 static void m2n_free_local_handles() {
    assert(!hythread_is_suspend_enabled());

    // AGAIN!
    if (exn_raised()) {
        exn_rethrow();
    }

    M2nFrame * m2n = m2n_get_last_frame();
    free_local_object_handles3(m2n->local_object_handles);
 }

So, we might throw away the exception handling from JNI lil stub, as far push_m2n is made always during JNI transition. Much more effective is to throw away blocks from m2n_free_local_handles and m2n_pop_local_handles, but that could affect compatibility across the m2n calls.

Simple JNI test (executing several millions of JNI transitions) shows:
$ ../../Builds/Harmony-clean/bin/java -cp . -Xmx128m -Xms128m nalog.nalog
iteration: 0 millis:8763
iteration: 1 millis:8722
iteration: 2 millis:8833

$ ../../Builds/Harmony-noEXP/bin/java -cp . -Xmx128m -Xms128m nalog.nalog
iteration: 0 millis:7824
iteration: 1 millis:7845
iteration: 2 millis:7836

$ ../../Builds/Harmony-noEXP-clean/bin/java -cp . -Xmx128m -Xms128m nalog.nalog
iteration: 0 millis:8241
iteration: 1 millis:8255
iteration: 2 millis:8214


  was:
During JNI transition the exception flags are checked twice.

1. One check is generated in JNI lil stub (vm/vmcore/src/jit/compile.cpp:675)

    //***** Part 11: Rethrow exception
    cs = lil_parse_onto_end(cs,
                            "l0=ts;"
                            "ld l2,[l0+%0i:ref];"
                            "jc l2!=0,_exn_raised;"
                            "ld l2,[l0+%1i:ref];"
                            "jc l2=0,_no_exn;"
                            ":_exn_raised;"
                            "m2n_save_all;"
                            "out platform::void;"
                            "call.noret %2i;"
                            ":_no_exn;",
                            eoo, eco, exn_rethrow);
    assert(cs);

 2. Second check is done during pop_m2n frame generation: (ex.: vm/port/src/lil/ia32/pim/m2n_ia32.cpp:259)

 static void m2n_free_local_handles() {
    assert(!hythread_is_suspend_enabled());

    // AGAIN!
    if (exn_raised()) {
        exn_rethrow();
    }

    M2nFrame * m2n = m2n_get_last_frame();
    free_local_object_handles3(m2n->local_object_handles);
 }

So, we might throw away the exception handling from JNI lil stub, as far push_m2n is made always during JNI transition. Much more effective is to throw away blocks from m2n_free_local_handles and m2n_pop_local_handles, but that could affect compatibility across the m2n calls.


> [drlvm][jni] JNI transition check exceptions twice
> --------------------------------------------------
>
>                 Key: HARMONY-4714
>                 URL: https://issues.apache.org/jira/browse/HARMONY-4714
>             Project: Harmony
>          Issue Type: Improvement
>            Reporter: Aleksey Shipilev
>         Attachments: HARMONY-4714-dirty.patch, HARMONY-4714.patch
>
>
> During JNI transition the exception flags are checked twice.
> 1. One check is generated in JNI lil stub (vm/vmcore/src/jit/compile.cpp:675)
>     //***** Part 11: Rethrow exception
>     cs = lil_parse_onto_end(cs,
>                             "l0=ts;"
>                             "ld l2,[l0+%0i:ref];"
>                             "jc l2!=0,_exn_raised;"
>                             "ld l2,[l0+%1i:ref];"
>                             "jc l2=0,_no_exn;"
>                             ":_exn_raised;"
>                             "m2n_save_all;"
>                             "out platform::void;"
>                             "call.noret %2i;"
>                             ":_no_exn;",
>                             eoo, eco, exn_rethrow);
>     assert(cs);
>  2. Second check is done during pop_m2n frame generation: (ex.: vm/port/src/lil/ia32/pim/m2n_ia32.cpp:259)
>  static void m2n_free_local_handles() {
>     assert(!hythread_is_suspend_enabled());
>     // AGAIN!
>     if (exn_raised()) {
>         exn_rethrow();
>     }
>     M2nFrame * m2n = m2n_get_last_frame();
>     free_local_object_handles3(m2n->local_object_handles);
>  }
> So, we might throw away the exception handling from JNI lil stub, as far push_m2n is made always during JNI transition. Much more effective is to throw away blocks from m2n_free_local_handles and m2n_pop_local_handles, but that could affect compatibility across the m2n calls.
> Simple JNI test (executing several millions of JNI transitions) shows:
> $ ../../Builds/Harmony-clean/bin/java -cp . -Xmx128m -Xms128m nalog.nalog
> iteration: 0 millis:8763
> iteration: 1 millis:8722
> iteration: 2 millis:8833
> $ ../../Builds/Harmony-noEXP/bin/java -cp . -Xmx128m -Xms128m nalog.nalog
> iteration: 0 millis:7824
> iteration: 1 millis:7845
> iteration: 2 millis:7836
> $ ../../Builds/Harmony-noEXP-clean/bin/java -cp . -Xmx128m -Xms128m nalog.nalog
> iteration: 0 millis:8241
> iteration: 1 millis:8255
> iteration: 2 millis:8214

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.