(From the company announcement http://www.codex.gr/article/libfreevec_103_released)
Please feel free to report any bugs, performance gains (even losses :-)
Thanks!

really really a great news!!
I'll go to test it on my iBook G4 :-D


Acrux

_________________
acrux _at_ linuxmail _dot_ org

And here is the software that I used to create all the benchmarks.

http://www.codex.gr/article/benchsuite_102_released

Every function has its own testing/benchmarking software and you can configure the number of threads (yes, you can actually run the same test in parallel and measure the actual overhead), the number of loops, min/max sizes, test aligned/unaligned access, same/random data (random data access is much more realistic than an artificial benchmark clearing the cache each time) and other features.

The benchmark data are written on a SQLite database, so you can work on them afterwards any way you like!

Though developed for libfreevec, benchsuite is easily adapted to work with any function.
(perhaps I will include a small howto if there is demand).

Anyway, enjoy!

I hereby demand it :)

_________________
Matt Sealey

Hi. Unfortunately I can't download this package :(Web-site www.codex.gr can not be found from my computer (but Google has cached pages from it) and also I can't find download section on freevec.org ...

Please advice where can I get these sources? Thank you!

P.S. It looks like codex.gr was expired yesterday:

Domain Name:codex.gr
Domain Handle:d4a2db3c7bf034c10ad90e07cada7a7df-gr
Protocol Number:979899
Creation Date:19-4-2004
Expiration Date:19-4-2008
Updated Date:24-4-2007
Comment:For a period of 15 days following a domain name expiration, a new registration application for the domain name and its bundle names forms may only be filed by the owner of the expired domain
Registrar:FORTHNET A.E
Registrar Referral URL:http://www.forthnet.gr
Registrar Referral Email:sme-admin@forthnet.gr
Whois Server:
Bundle Name:codex.gr
Name Server:dns1.name-services.com
Name Server:dns2.name-services.com
Name Server:dns3.name-services.com
Name Server:dns4.name-services.com
Name Server:dns5.name-services.com

_________________
Alexander Shabarshin

Hi,

Yes there was a problem with my registrar, though I had paid for the domain's renewal, they didn't actually do it, so the domain expired. It should be working ok now, so please try again in the same URL :)

Thanks! Link is working now :)

_________________
Alexander Shabarshin

Cool, I would really really appreciate any feedback on the library :)

I can't build benchsuit on my PowerBook G4...

cd libfreevec-1.0.3
configure
make
make install
ldconfig
cd ..

cd benchsuite-1.0.2
configure
make
What do I do wrong?

Update: put /usr/local/lib/freevec to /etc/ld.so.conf (and ldconfig after that) doesn't help, only copy libfreevec.* from /usr/local/lib/freevec to /usr/local/lib helps...

Update 2: I have standard Debian 4.0 etch and I can't run sh-scripts from distribution, because of unknown option -executable that used for find, also with number of threads more than 1 sqlite3 can't write to database - so I edited script to avoid -executable and with threads number equals 1...

_________________
Alexander Shabarshin

Another question: Should random.dat consist of random data? On my system I have random.dat with zeros after running prepare_data.sh...

_________________
Alexander Shabarshin

LOL,

I should have used /dev/random instead of /dev/zero in prepare_data.sh!!

Sorry about that, fixed in svn :)

Could you please tell me SVN path to checkout this source code?
On svn://svn.codex.gr/trunk/benchsuite it's still /dev/zero

P.S. With random data (that was generated by simple C-program) test-strncpy is failed...

_________________
Alexander Shabarshin

Same repo, could you try again now? I removed the shell script (was taking ages to populate ~6MB of random data) and replaced it with a C file that is producing real random data that includes NO zero bytes (which were causing the test-strncpy test program to fail for obvious reasons, it was mistakingly copying data up to the wrong position). Now it reports success.

Also, I've added some -work in progress- test programs for some altivec math replacement functions that I'm working on (sinf, cosf, etc, replacements). These are still not committed in the libfreevec repo so the test will fail :-/

Thanks for the feedback, it's really appreciated :)

Also I noticed that sometimes you use random() that is not ANSI standard instead of rand() and INT_MAX and CHAR_MAX instead of RAND_MAX. And why in strcmp reported alighnments are -1 and 0?

P.S. Now all tests are passed ;)

_________________
Alexander Shabarshin

Fixed thanks again :)
Where exactly do you mean?