Prebuilt binary questions
15 Aug 2005 3:33:31 UTC
Topic 189697
(moderation:
I have searched a bit and had a few questions about the einstein 4.79 prebuilt binaries. I assume the application is written in Fortran. Is it Fortran 77 or 95? Also may I ask which compiler was used to build it for linux and windoze? Also, is the code optimized to use if available MMX, MMX+, 3dNow!, Enhanced 3dNow!, SSE and SEE2? I am afraid that the binaries are built for F77 using the standard x87 FPU instructions using gcc or something. I only ask because the application seems to give about half the credits on my AMD64 3.2gig that it should (getting credit from server for all workunits). I know credits are affected by a lot of factors but kind of sorry to see my amd desktop getting the same amount of credits as my Pentium M 1.5 laptop. I am now running climate as well on AMD desktop and it's credits seem much more reasonable. I bet if source code was released I could build it with Intel Fortran 9 (works great on AMDs too finally) and destroy the crappy prebuilts. Oh well best boinc project out there as gravity waves and pulsars are very important to future of physics.
Prebuilt binary questions
)
I would think it's because the Pentium M 1.5 GHz is approximately as fast as a P4 with let's say 2.8 GHz
RE: I bet if source code
)
You'll lose.
The major part of the program is written in C, including the main loop which does 99% of the work. The Windows port is compiled with MSC 7.1, which has shown to give the fastest running code (compared to e.g. icc and gcc). Useful suggestions for improving the speed are always welcome.
BM
BM
Cool thanks for the info. I
)
Cool thanks for the info. I am sure you have spent considerable time optimizing and benchmarking the app. A few beers and my programmer ego came out. An interesting site I have came across on the internet discusses how Fortran can be better and more efficient in many ways than C for science apps ( http://www.cts.com.au/compare.html ) but again I am sure this has been analyzed. Finally, for all
those out there that want a good intro into code optimization I have found the following link to Agner Fog's classic document to be excellent but a little out of date ( http://cr.yp.to/2005-590/fog.pdf ) but again I am sure this is kiddie's play for your experience on the subject. I am sure I would have my work cut out for me building as accurate a floating point binary that is efficient even with the latest source code, so I will go back to crunching WU :) .
Interesting references. I
)
Interesting references. I added one as a link to the short article on optimization and am asking for permission to incorporate the other.
Cool, good to spread
)
Cool, good to spread information on scientific computing, as far and wide as possible, as it ultimately benefits mankind. By the way, it is probably better on the optimization link to link directly to Agner Fog's page on it at ( http://www.agner.org/assem/ ). The pdf is found as a link under this page and will probably always be the most up to date link and plus his open source asm libraries can be download here as well.
RE: Cool, good to spread
)
Hmmm,
well, I just looked at that page and I cannot find the fog paper as a link on it at all. There are other interesting links, but, not that one. obviously I missed it somehow.
RE: RE: Cool, good to
)
Its there but hard to see. It is near the top with the link "Download" (hey the guys is great with assembler but his html skills are bit 1995ish :) Here is the direct link to the paper - http://www.agner.org/assem/pentopt.pdf .
Well, that explains the
)
Well, that explains the problem. You sent a link to a PDF named FOG.pdf, and now the link is pentopt.pdf ... same paper, different file names.