Hi,
I have posted a question about the error while using recon-all command with -use–gpu option on a CentOS 6 machine a while back.
CUDA Error in file 'mriconvolve_cuda.cu' on line 945 : unload of CUDA runtime fa
iled.
Abort (core dumped)
ERROR: mri_ca_register with non-zero status 134
The problem apparently came from this command
mri_ca_register_cuda -nobigventricles -T transforms/talairach.lta -align-after
-mask brainmask.mgz norm.mgz /opt/freesurfer/average/RB_all_2008-03-26.gca transforms/talairach.m3z
We are wondering if it was OK to change the recon-all script so that we would enable OpenMP for this mri_ca_register command but use GPU for the rest of recon-all script? That is, we would replace line 2273 in recon-all from
if($UseCuda) set cmd = (mri_ca_register_cuda)
To
if($UseCuda) set cmd = (mri_ca_register)
So, with the –openmp option, this command will run with multiple cores. By not using GPU for this command, we can avoid the CUDA Error message.
However, we are wondering if it's OK that we combined both -use–gnu and –openmp?
Could you help us out?
--
Kevin Tran, Contractor
UNIX System Administrator
The National Institutes of Health
NIMH, Laboratory of Brain and Cognition
10 Center Drive, Room 4C215
Bethesda MD 20892-1366
kevintran(a)mail.nih.gov