Anonymous platform -- app_info.xml -- developer help needed
10 Nov 2014 14:41:28 UTC
Topic 197791
(moderation:
Hi,
i managed to compile the BRP4G and BRP5 cuda application (same app) on my linux with cuda 6.0.
1) Now I need help setting up a working app_info.xml.
I'd like to run CPU APPS:
- einstein_S5R6_1.01_graphics_i686-pc-linux-gnu
- hsgamma_FGRP4_1.04_x86_64-pc-linux-gnu__FGRP4-SSE2
- einstein_S5R6_1.01_graphics_i686-pc-linux-gnu
and MY GPU APP
- einsteinbinary_x86_64-pc-linux-gnu-cuda
.
2) I also need help about what to do with db.dev and dbhs.dev files? rename?
3) Should I put the NVIDIA v6.0 cudart and cufft files to einstein project directory too? Is it ok if they are in LIBPATH/PATH?
4) Before I try anything with production database I'd like to test my compilation against the standard app to validate results. How to do that off-line (not beta)? Is there a scrip for developers/optimizers?
--
petri33
Anonymous platform -- app_info.xml -- developer help needed
)
Starting with a disclaimer: I'm not associate with the project and I'm not a developer/coder etc. but I might be able to help in some way.
Starting with your first question about app_info.xml.
There is currently only one CPU search running on Einstein and it's called "Gamma-ray pulsar search #4" or FGRP4 for short. The "einstein_S5R6_1.01_graphics..." files you mention are the screen saver apps associated with the now ended Gravity wave search and thus not needed at the moment. (Although they might get reused soon when the new gravity wave search gets under way, who knows!?)
If you have some knowledge about how to write an app_info.xml file you might be able to put something together by studying the client_state.xml file looking at , and tags for tasks assigned to the official app. I don't have enough knowledge and experience to help you with that though, sorry.
Number 2: Haven't got a clue.
Number 3: I'd take the safe alternative and put the files in the einstein project folder and reference them in the app_info.xml.
Number 4: As long as you're the only one running the new app there should be minimal risk of bad data being inserted in the science database as any error in your own app should not match the official app and thus not validate. But to play it safe choose a task assigned to you and the official app and extract the relevant data from the client_state.xml, the tag has the used and also reference all the needed data files, copy the data files to a separate folder along with the app and support files (cudart etc.) then run the app in standalone mode making sure to pass the command line to the app.
I've done something similar when the Intel GPU app was developed and this thread over at the old and now obsolete Albert@home message board might be useful: https://albert.phys.uwm.edu/forum_thread.php?id=8954 especially Message 112610 from Bikeman.
Don't use the files linked in that post to conduct your test but use them to get a grip on how things work with these apps.
Think I'll stop mumbling now and let someone with more knowledge take it from here. Hopefully something of the above was helpful =)
My suggestion would be to
)
My suggestion would be to peek into client_state.xml for the BRP4G/BRP5 CUDA32 App (version) and use the app info there as a template.
All files listed for that app, or better, the respective CUDA 6 counterparts, are needed (with possible exception of the graphics_app if you don't want to see graphics) and should be placed in the project directory. db.dev and dbhs.dev should get unique names to not be confused / overwritten by the CUDA 32 files, but should be copied (tag ) into the slots directory with the original names (db.dev etc.). I would do the same with cudart and cufft (i.e. put these in the project directory and let the client copy these with the correct names). Depending on the client version it might work to set LD_LIBRARY_PATH to the directory where these files are in the environment of the client (which should then be inherited by the app), but I wouldn't bet that the app (or actually the loader) finds and picks the intended versions.
hth,
BM
Addendum: here's our app info for the BRP5 CUDA 32 app version:
einsteinbinary_BRP5 139 7.2.2 einsteinbinary_BRP5_1.39_x86_64-pc-linux-gnu__BRP5-cuda32-nv270 libcudart64_32_16.so libcudart.so.3 libcufft64_32_16.so libcufft.so.3 EULA.txt EULA.txt einsteinbinary_BRP5_1.39_x86_64-pc-linux-gnu__BRP5-cuda32-nv270-db.dev db.dev einsteinbinary_BRP5_1.39_x86_64-pc-linux-gnu__BRP5-cuda32-nv270-dbhs.dev dbhs.devThe ""s for your cudart and cufft will probably be different, depending on the CUDA version and your build. Just use the names the libs have now.
Addendum 2: On E@H every task is ran twice by different participants an then compared. There is no chance you will get the same result as another participants computer with a broken app.
BM
Thank you both, I think
)
Thank you both,
I think this will be an interesting journey to the einstein universe.
Now going to sleep. If I can.
--
petri33
Hi, I have some dev boards
)
Hi,
I have some dev boards with ARM processors (Cortex-A9) running Debian/Ubuntu and I want to run some tests with the EaH client. I manage to compile the client for these platforms but the boinc returns:
Einstein@Home | platform 'arm-unknown-linux-gnueabi' not found.Is there a way to run and validate eah wu in these platforms with eah clients which are compiled by me ?
Thank you,
I manage to create an
)
I manage to create an app_info.xml file and validate same wus with a freescale i.MX6q (Cortex-A9) SoC board (here).
RE: I manage to create an
)
Does the Stock app for Arm work on that host?
If your Boinc client is running a different platform, you could try adding an arm-unknown-linux-gnueabihf option to your cc_config.xml to see if it'll run:
Client_configuration
Claggy
I'm trying to tweak some code
)
I'm trying to tweak some code using anonymous platform.
Are these results different? (7:th meaningful digit or so differs by one)
Different summing order....
Got 8 minutes off from 100 minutes (8%) on the first try.
Will my results be deemed "not validated" ?
I have the same problems
)
I have the same problems (diffs) with the Cortex-A9 client but the results are validated by the project.
Thank You. I'll give a try
)
Thank You. I'll give a try for my version.