|
|
|
Hi all,
this is a patch we wrote during the libreoffice hackfest 2011 for
sw/source/core/inc/MarkManager.hxx and sw/source/core/doc/docbm.cxx
which improves speed of mailmerge. Bjoern already knows details so he
might be the one to review the patch... The patch could be applied
under the lgpl.
bye,
Christoph Lutz, Landeshauptstadt München
_______________________________________________
LibreOffice mailing list
[hidden email]
http://lists.freedesktop.org/mailman/listinfo/libreoffice
|
|
|
Hi Björn,
today I managed to improve the patch again. The patch now speeds up
our mailmerge-Szenario enormously and seems to eliminate the
exponential waste of cpu-cycles. Would you please have a look at it?
here some measurings:
MailMerge-Time (in ms) without patch for 10, 50, 100, 150 datasets:
2793, 11821, 28608, 64395
MailMerge-Time (in ms) with first patch for 10, 50, 100, 150 datasets:
2561, 11044, 26226, 55812
MailMerge-Time (in ms) with this patch for 10, 50, 100, 150 datasets:
1838, 7357, 14413, 21558
2011/9/6 Christoph Lutz < [hidden email]>:
> 2011/9/5 Bjoern Michaelsen < [hidden email]>:
>> On Sun, 4 Sep 2011 15:05:32 +0200
>> Christoph Lutz <chrlutz-gM/ [hidden email]> wrote:
>>
>>> this is a patch we wrote during the libreoffice hackfest 2011 for
>>> sw/source/core/inc/MarkManager.hxx and sw/source/core/doc/docbm.cxx
>>> which improves speed of mailmerge. Bjoern already knows details so he
>>> might be the one to review the patch... The patch could be applied
>>> under the lgpl.
>>
>> Patch looks good, Just to clarify before pushing, is it contributed
>> under:
>> MPL 1.1 / GPLv3+ / LGPLv3+
>> as per http://wiki.documentfoundation.org/License_Policy ?
>
That's OK!
Best,
Christoph
_______________________________________________
LibreOffice mailing list
[hidden email]
http://lists.freedesktop.org/mailman/listinfo/libreoffice
|
|
|
On Tue, 6 Sep 2011 17:36:18 +0200
Christoph Lutz < [hidden email]> wrote:
> today I managed to improve the patch again. The patch now speeds up
> our mailmerge-Szenario enormously and seems to eliminate the
> exponential waste of cpu-cycles. Would you please have a look at it?
>
> here some measurings:
>
> MailMerge-Time (in ms) without patch for 10, 50, 100, 150 datasets:
> 2793, 11821, 28608, 64395
> MailMerge-Time (in ms) with first patch for 10, 50, 100, 150 datasets:
> 2561, 11044, 26226, 55812
> MailMerge-Time (in ms) with this patch for 10, 50, 100, 150 datasets:
> 1838, 7357, 14413, 21558
Wow, great win! Pushed as:
http://cgit.freedesktop.org/libreoffice/core/commit/?id=e024f616934bb78fba8c8101264806d507068d7ewith some minor tuning:
- formatting (whitespace mostly)
- constness
Could you maybe try, if a pragmatic:
m_aMarkBasenameMapUniqueOffset.clear();
at:
http://cgit.freedesktop.org/libreoffice/core/tree/sw/source/core/doc/docbm.cxx?id=e024f616934bb78fba8c8101264806d507068d7e#n491and
http://cgit.freedesktop.org/libreoffice/core/tree/sw/source/core/doc/docbm.cxx?id=e024f616934bb78fba8c8101264806d507068d7e#n743kills the performance gain?
If not, that would prevent the behavior change in for example the
scenario:
- Create Mark
- Copy Mark
- Copy Mark
- Delete First Copied Mark
- Copy Mark
Best,
Bjoern
--
https://launchpad.net/~bjoern-michaelsen_______________________________________________
LibreOffice mailing list
[hidden email]
http://lists.freedesktop.org/mailman/listinfo/libreoffice
|
|
|
I am sorry, but I had to revert this patch because it was causing
crashes in Windows build when loading any document into writer or when
trying to save a document from writer.
It would be nice to check on windows to see whether it can be fixed, but
we did not have much time for that now, so I just reverted.
Sorry again
F.
On 06/09/11 19:22, Bjoern Michaelsen wrote:
> On Tue, 6 Sep 2011 17:36:18 +0200
> Christoph Lutz < [hidden email]> wrote:
>
>> today I managed to improve the patch again. The patch now speeds up
>> our mailmerge-Szenario enormously and seems to eliminate the
>> exponential waste of cpu-cycles. Would you please have a look at it?
>>
>> here some measurings:
>>
>> MailMerge-Time (in ms) without patch for 10, 50, 100, 150 datasets:
>> 2793, 11821, 28608, 64395
>> MailMerge-Time (in ms) with first patch for 10, 50, 100, 150 datasets:
>> 2561, 11044, 26226, 55812
>> MailMerge-Time (in ms) with this patch for 10, 50, 100, 150 datasets:
>> 1838, 7357, 14413, 21558
>
> Wow, great win! Pushed as:
> http://cgit.freedesktop.org/libreoffice/core/commit/?id=e024f616934bb78fba8c8101264806d507068d7e> with some minor tuning:
> - formatting (whitespace mostly)
> - constness
>
> Could you maybe try, if a pragmatic:
> m_aMarkBasenameMapUniqueOffset.clear();
> at:
> http://cgit.freedesktop.org/libreoffice/core/tree/sw/source/core/doc/docbm.cxx?id=e024f616934bb78fba8c8101264806d507068d7e#n491> and
> http://cgit.freedesktop.org/libreoffice/core/tree/sw/source/core/doc/docbm.cxx?id=e024f616934bb78fba8c8101264806d507068d7e#n743> kills the performance gain?
>
> If not, that would prevent the behavior change in for example the
> scenario:
> - Create Mark
> - Copy Mark
> - Copy Mark
> - Delete First Copied Mark
> - Copy Mark
>
> Best,
>
> Bjoern
>
_______________________________________________
LibreOffice mailing list
[hidden email]
http://lists.freedesktop.org/mailman/listinfo/libreoffice
|
|
Locked
markmanager.patch (4K) 