Hi,

I've been trying to compile the boinc client for linux (Opteron 64, on FC3). It all works fine up to the module libsched.a, where I get the following error messages:

../db/db_base.C: In member function `int DB_CONN::open(char*, char*, char*, char*)':
../db/db_base.C:44: Fehler: invalid conversion from `char*' to `unsigned int'
../db/mysql.h:239: Fehler: too many arguments to function `MYSQL* mysql_real_connect(MYSQL*, const char*,const char*, const char*, unsigned int, const char*, unsigned int)'
../db/db_base.C:44: Fehler: at this point in file
make[3]: *** [db_base.o] Fehler 1


I'm trying to compile boinc_public-cvs-2005-07-18.tar.gz with following compiler options:

"-march=opteron -mcpu=opteron -O3 -fomit-frame-pointer -funroll-loops -fforce-addr -ftracer"


What am I doing wrong?

Thanks for your help,

Hugo

Have a look at the thrid party site (http://boinc.berkeley.edu/download_other.php), there are very good optimized and stable ports for linux on opteron.

---

Have a look at the thrid party site (http://boinc.berkeley.edu/download_other.php), there are very good optimized and stable ports for linux on opteron.

Very good optimized... yes, if you neglect the fact the results are invalid. I've tried them all.
They are either quite slow (considering I'm using a machine with 2 2.3Ghz AMD 64 CPUs with 4 GB RAM, I would expect a work unit to take less than 8 hours..) or produce invalid results (I ran the reference work unit and compared it with my results as described in http://www.pperry.f2s.com/boinc-compile-seti.htm ; the errors - mostly the best gaussian - are far beyond those accepted for validation, probably because the clients are compiled using ffast-math)

Of course I would like to get as much credit as possible for my work, but there's not much point if that means I'm just producing invalid results.

Anyway... meanwhile I managed to get it to compile, but I get lower Whet and Dhrystones than with the standard client and it takes even longer.

Does anyone know a real 64-Bit compile (without ffast-math)?

Have a look at the thrid party site (http://boinc.berkeley.edu/download_other.php), there are very good optimized and stable ports for linux on opteron.

Very good optimized... yes, if you neglect the fact the results are invalid. I've tried them all.
They are either quite slow (considering I'm using a machine with 2 2.3Ghz AMD 64 CPUs with 4 GB RAM, I would expect a work unit to take less than 8 hours..) or produce invalid results (I ran the reference work unit and compared it with my results as described in http://www.pperry.f2s.com/boinc-compile-seti.htm ; the errors - mostly the best gaussian - are far beyond those accepted for validation, probably because the clients are compiled using ffast-math)

Of course I would like to get as much credit as possible for my work, but there's not much point if that means I'm just producing invalid results.

Anyway... meanwhile I managed to get it to compile, but I get lower Whet and Dhrystones than with the standard client and it takes even longer.

Does anyone know a real 64-Bit compile (without ffast-math)?

Have you informed the porters which build this binaries ? - I think it is not OK if the results of the reference WU is out of the validation ranges.
Btw. on some plattforms and/or with gcc -ffast-math don't break the validation limits. For example gcc 4 on HPUX.
Have you tried "-O3 -funroll-loops" - it gave me the best performance with gcc 4 on HPUX.

How long did the standard client from berkley need for the reference WU or a normal WU ?

Lars

---

Have you tried Stefan Urbats Opteron bianries too ? - He have it well tested, validated, and get the result validated.

---

Have a look at the thrid party site (http://boinc.berkeley.edu/download_other.php), there are very good optimized and stable ports for linux on opteron.

I think i´ve read something that gcc and amd64 is still broken and gives slow code .Not sure where but someone (Ned Slider ???) recommended using the Athlon64 - 32 bit client on 64 bit linux systems.

regards
Crunch3r

---

Join BOINC United now!

Here a sumary of the clients I've tried so far:

Standard client : about 8 hours, valid, 100-120 claimed credit/WU
Stefan Urban : 8 hours, valid , 100-120 claimed credit/WU
pperry : 1 hour, ?, 20-30 claimed credit/WU

Processing the standard WU qith all these clients and comparing it with the reference result I got a big bias with ppery (as in best gaussian power 0.8 instead of 1.5). But somehow, if I use this client with boinc I've been getting mostly valid results... just some invalid now and then.

Anyway... I guess I'm just gonna keep on using ppery and live with the ocasional invalid result. Since Seti is checking the validity anyway, bad results will be filtered out.

Thnaks for the help!

---

back to the original question, you are probably getting MYSQL errors because gcc is trying to compile the server option (nobody uses these so you can disable them)

try:
./configure --disable_server
when you start out, if you havent been doing that. i also think you want the 'k8' flag instead of 'opteron' for your gcc CPU options although i can't recall whether it was that or 'amd64' that worked better.

however, the problem still stands that gcc is not good with 64 bit on the amd platform, even with the -m64 flag. as far as i know, everyone who has tried it (including me) gets crunch times in the 8-10 hour range as opposed to the 1-2 hour range of the optimized 32 bit clients.

if you are still concerned about -ffast-math (which has reliably produced valid results across multiple platforms) email ned (pperry site) to find out the complete procedure he used and just leave that flag out. he also used FC3 i believe.

try those
http://www.pperry.f2s.com/downloads.htm