Issue 8 in opencurrent: 7 of 17 unit tests failed

[openc...@googlecode.com]

Status: New
Owner: ----
Labels: Type-Defect Priority-Medium

New issue 8 by a...@ddsw.nl: 7 of 17 unit tests failed
http://code.google.com/p/opencurrent/issues/detail?id=8

What steps will reproduce the problem?
1. build for sm_10 (using netcdf4 and a GeForce 8800 GTX)
2. make test

What is the expected output?
All tests should be ok.

What do you see instead?
see attached unittest.txt

What version of the product are you using? On what operating system?
1.0.0 on Ubuntu 8.04 LTS

Please provide any additional information below.
I ran utest MultigridTest under a debugger and found that at line 388 of
sol_mgpressure3ddev.cu the value of _num_level is not 6 as it should be
but a very large value. Something is obviously overwritten. I also ran
utest MultigridTest under valgrind and the output is in the attached file
memcheck.txt.

Attachments:
memcheck.txt 6.0 KB
unittest.txt 6.9 KB

--
You received this message because you are listed in the owner
or CC fields of this issue, or because you starred this issue.
You may adjust your issue notification preferences at:
http://code.google.com/hosting/settings

[openc...@googlecode.com]

Comment #1 on issue 8 by jcohen.personal: 7 of 17 unit tests failed
http://code.google.com/p/opencurrent/issues/detail?id=8

Could you try checking out the latest version from hg and seeing if you can
reproduce?

Instructions here:
http://code.google.com/p/opencurrent/source/checkout

Also, are you on a 32-bit or 64-bit OS?

[openc...@googlecode.com]

Comment #2 on issue 8 by a...@ddsw.nl: 7 of 17 unit tests failed
http://code.google.com/p/opencurrent/issues/detail?id=8

Dear Jonathan,

I downloaded your latest versions per your instructions and I got the
attached output from the unit tests. My system is 32-bit and has a
GeForce 8800 GTX which only has floats, no doubles. I use NetCdf 4.0.

Running utest MultigridTest in a debugger now revealed that the
segmentation fault was caused by 'this' in the code below having the
value 0xad0 which is obviously not a valid address of an object in
Linux.

Grid2DDevice<T>::~Grid2DDevice()
{
cudaFree(this->_buffer);
}

This destructor was called from delete _diag_grid[l]; (line 335
sol_mgpressure3ddev.cu).

Hope this helps!

Attachments:
tests.txt 20.1 KB

[openc...@googlecode.com]

Comment #3 on issue 8 by a...@ddsw.nl: 7 of 17 unit tests failed
http://code.google.com/p/opencurrent/issues/detail?id=8

Dear Jonathan,
Last week I got a new graphics card. I now have a GTX-285. This made things
even worse.
I recompiled for sm_13 and got the attached output for make test. This is
the version
I downloaded on March 19. I also tried the official version 1.0.0 again and
here too
things got worse. I attach both make test outputs.
I hope you can figure out what is wrong.
Sincerely, Anneke

Attachments:
unittests-1.0.0.txt 19.0 KB
unittest-recent.txt 37.8 KB

[openc...@googlecode.com]

Comment #4 on issue 8 by jcohen.personal: 7 of 17 unit tests failed
http://code.google.com/p/opencurrent/issues/detail?id=8

hmm, it looks like this has nothing to do with the graphics card, and maybe
nothing
to do with cuda. Might be an issue with how netcdf was built?

My guess is a libc incompatibility or something like that.
Could you run 'ldd -v tests/utest' and attach the results?

[openc...@googlecode.com]

Comment #5 on issue 8 by a...@ddsw.nl: 7 of 17 unit tests failed
http://code.google.com/p/opencurrent/issues/detail?id=8

Here are the results. btw, if I compile for netcdf 3, which is included in
Ubuntu
8.04, so not built by me, the problems are only slightly different, see the
other
attachment which was made with the recent opencurrent snapshot, not with
1.0.0.

Attachments:
ldd_netcdf4.txt 7.4 KB
unittest-recent-netcdf3.txt 42.9 KB

[openc...@googlecode.com]

Comment #6 on issue 8 by jcohen.personal: 7 of 17 unit tests failed
http://code.google.com/p/opencurrent/issues/detail?id=8

are you able to compile and run the cuda sdk samples? or any cuda progam?

[anneke]

Dear Jonathan,
Sorry that it took so long to reply...
Yes, all CUDA SDK sample programs and one of my own work fine!
Anneke

[openc...@googlecode.com]

Comment #7 on issue 8 by a...@ddsw.nl: 7 of 17 unit tests failed
http://code.google.com/p/opencurrent/issues/detail?id=8

Yes I am able to compile and run all CUDA example programs plus Iterative
CUDA and
OpenNL. I hope this message gets throgh now at last, there seems to be
something
wrong with this bulletin board system.

[openc...@googlecode.com]

Comment #8 on issue 8 by a...@ddsw.nl: 7 of 17 unit tests failed

http://code.google.com/p/opencurrent/issues/detail?id=8

Yes I am able to compile and run all CUDA example programs plus Iterative
CUDA and

OpenNL with CUDA support.
I hope this message gets through now at last, I tried it over a few weeks
repeatedly. There seems to be some problem with me contacting this bulletin
board
system.