harmony-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "weldon washburn (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HARMONY-3065) [DRLVM][hy_thr] Force inline of p_TLS_vmthread and tmn_suspent_enable/disable.
Date Wed, 28 Feb 2007 00:55:05 GMT

    [ https://issues.apache.org/jira/browse/HARMONY-3065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12476474

weldon washburn commented on HARMONY-3065:

I just now looked at nativecalls.patch.  I have some questions:

Re: hythread_suspend_disable() mods.  Maybe its a typo or maybe its a performance optimization
I don't know about but it looks like the code does a move into AX register  (16-bit) then
a decrement and test of the EAX register (32-bit) .  If the upper 16 bits contain garbage,
I worry that thread_safe_point_impl() will not be called when it really needs to be.  Most
likely this will cause no hard failure.  Maybe just a slight increase in GC latency?? The
code in question:

mov ax, [edx]HyThread.suspend_disable_count
dec eax
test eax, eax

I see EBP get setup as a old fashioned activation record pointer.  But I don't see use of
the frame pointer.  Stuff like "mov ECX, [EBP]-12"....  Can the following lines simply be

push ebp
mov ebp, esp

move esp, ebp
pop ebp


> [DRLVM][hy_thr] Force inline of p_TLS_vmthread and tmn_suspent_enable/disable.
> ------------------------------------------------------------------------------
>                 Key: HARMONY-3065
>                 URL: https://issues.apache.org/jira/browse/HARMONY-3065
>             Project: Harmony
>          Issue Type: Improvement
>          Components: DRLVM
>            Reporter: Pavel Afremov
>         Assigned To: weldon washburn
>         Attachments: 0001-Adapt-TLS-inlining-for-Linux.txt, 0001-Force-inline-of-p_TLS_vmthread-and-tmn_suspent_enable-disable.patch,
0001-Force-inline-of-p_TLS_vmthread-and-tmn_suspent_enable-disable.patch, 0001-Force-inline-of-p_TLS_vmthread-and-tmn_suspent_enable-disable.txt,
JNIExceptionPerf.zip, nativecalls.patch
> Force inline of p_TLS_vmthread and tmn_suspent_enable/disable. It improve performance
up to 3 times.

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

View raw message