Re: fast math lib?
I think we can separate the vectorized library from other codes.
Then we can use libcpml with vectorized sqrt etc.
The code is independent from other libffm codes, then it is easy to do.
In article <Pine.SOL.3.96.1000628162550.22260Afirstname.lastname@example.org.Uni-Koeln.DE>
>> On Mon, 26 Jun 2000, Adam C Powell IV wrote:
>> > > Both do well, but I'd personally prefer the Compaq CPML library. It's
>> > > faster in most cases, and provides for more functions (and
>> > > exceptions) than libffm. I maintain the libffm debs, so those are
>> > > available for use, in case you wanted to keep it "free" though.
>> > Indeed, CPML is faster for most things, but the "vector" routines dsqrtv =
>> > dsqrtiv in libffm can be several times faster than a loop doing the same.
>> > Try installing libffm-dev, prototypes are in /usr/include/libffm.h .
>> I guess, this is at the moment a little bit too tricky for me to try. I
>> am a biologist, not a programmer (got stuck after 2/3 of "(dirty) C for
>> Dummies" ... ;-) ). Thank you for that tip anyway, perhaps I will be
>> able one day to roam around C-source codes by myself and optimise. Since
>> my first alien attempts didn=B4t work out too well (the Compaq compiler
>> and math lib were my first try ever to debianise something), I just
>> decided to have at first a working gcc with libffm and give it a little
>> more time to find out about the ccc things to consider.
>> To UNSUBSCRIBE, email to email@example.com
>> with a subject of "unsubscribe". Trouble? Contact firstname.lastname@example.org
Dept. Communication Engr./Univ. TOKAI
1117 Kitakaname Hiratsuka 259-12 Japan
TEL.+81-463-58-1211(ext. 4084) FAX.+81-463-58-8320