Compiling for anonymous platform


Message boards : Number crunching : Compiling for anonymous platform

Message board moderation

To post messages, you must log in.
AuthorMessage
EXT64

Send message
Joined: 13 May 18
Posts: 2
Credit: 25,210,080
RAC: 0
Message 6157 - Posted: 22 Jan 2019, 19:39:59 UTC
Hi,
I am interested in possibly compiling the asteroids source for an anonymous platform (which seems to have been done before based on some discussion in this forum). I downloaded the source code linked by the following thread:
https://asteroidsathome.net/boinc/forum_thread.php?id=572#5206

Source code link:
http://astro.troja.mff.cuni.cz/projects/asteroids3D/web.php?page=download_software

Unfortunately, I'm a bit confused by the fact that there are several executables that can be built using the linked sources and makefiles. I tried using objdump to disassemble the "period_search_10210..." applications running on my machine, and based on the internal functions it appears the asteroids period_search apps are built from the "convexinv" source. Is this the right track, or is there more involved in building the app?

I also wanted to be sure the test cases included with the above source would be sufficient for verifying a compiled app. Is it better to "capture" real work units for testing?

Any advice from those who have compiled the asteroids app before (or forum admins) would be greatly appreciated--thanks!
ID: 6157 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
EXT64

Send message
Joined: 13 May 18
Posts: 2
Credit: 25,210,080
RAC: 0
Message 6158 - Posted: 27 Jan 2019, 17:57:33 UTC - in response to Message 6157.  
Just in case anyone else comes across this post with the same confusion I had:

It appears the asteroids@home period search app is based around the above open-source code, however the deployed application uses a Boinc-interface library (functions like "boinc_finish", "boinc_resolve_filename", "do_checkpoint", etc that you will find in other project apps). Since none of this is present in the base source it can't be used to simply compile the complete period search app.

Additionally, it looks like (based on the disassembly of the "period_search_10210_x86_64-pc-linux-gnu__avx" executable) significant effort has been put into optimizing the deployed app--complex loops in some functions have been fully vectorized (intrinsics?). Nothing like this is present in the baseline source code.
ID: 6158 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile JStateson
Avatar

Send message
Joined: 16 Jan 14
Posts: 17
Credit: 26,944,347
RAC: 24,030
Message 6428 - Posted: 19 Jan 2020, 22:44:18 UTC

Last modified: 19 Jan 2020, 22:46:04 UTC
I don't see the CUDA source anywhere. I have had success linking SETI with later (10.2) CUDA libs but need "something" other than that 0.2.1 database whatever.
ID: 6428 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Compiling for anonymous platform