loperf: performance testing using callgrind

[Matúš Kukan]

Hi,

Sorry, it's been so long, but finally
I've updated documents in
http://dev-builds.libreoffice.org/callgrind_report/
produced by loperf script running callgrind regularly.

There is an old history file showing a perf. improvement between commits
c3126202c429f512231272fb934e8334777c5231..62a18e8ec17b5ee389a2c82866d69d1130351282
which does not makes sense :-)
Otherwise the numbers are pretty stable modulo changes in the
script/test files.

Hopefully, the new file will be updated regularly and show some nice
performance improvements in the future ;-)

I've changed the script a bit, here is some description of
http://dev-builds.libreoffice.org/callgrind_report/history.fods

offload-first: OOO_EXIT_POST_STARTUP=1 soffice.bin --splash-pipe=0
--headless # after removed instdir/user directory

offload-second: the same with now existing user directory

file-load: OOO_EXIT_POST_STARTUP=1 soffice.bin --splash-pipe=0
--headless <file>

file-convert: soffice.bin --splash-pipe=0 --headless --convert-to
<file-extension> --outdir tmp <file>

The numbers show millions of pseudo-cycles. All with
OOO_DISABLE_RECOVERY=1.
Currently one round (build & test all files) is about 10 hours.

If you have some more ideas / questions and/or test files, please share.
Files are in test-files.git/loperf/

Best,
Matus