[Rtk-users] FDK and SART with GPU support is not terminating

[Simon Rit]

Hi,
Thank you very much for your report. Unfortunately, we can't reproduce 
it on our Windows and Linux machines and I have no idea what is the 
issue. Can you try to run it from MSVC and pause it when it's stucked to 
find our where it is stucked? You might have to compile RTK with the 
debug info (RelWithDebInfo).
I hope that we can find out what is the issue. Thanks for your help,
Simon

On 04/02/2016 13:51, Jonas Westberg Jahn wrote:
> We have curerntly testing RTK with GPU support on a Lenovo P500 Workstation (Windows 7 64 bit, Xeon processor, 8GB RAM, and nVidia Quadro K620)
>
> - Microsoft Visual Studio (Express) 2013
> - CMake 3.4
> - CUDA dev 7.5
> - FFTW 3.3.4 (precompiled 64bit)
> - ITK 4.8.2
> - RTK (from Git 2016-01-20)
>
> Copied "C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\bin\x86_amd64" to "amd64" (in bin folder) to make nvcc to work in MS VS Express, and created vcvars64.bat with the contents "CALL setenv /x64"
>   
> The following test works
>
> rtksimulatedgeometry -n 180 -o geometry.xml
> rtkprojectgeometricphantom -g geometry.xml -o projections.mha --spacing 2 --dimension 256 --phantomfile SheppLogan.txt
> rtkadmmtotalvariation -p . -r projections.mha -o rtkadmmtotalvariation.mha -g geometry.xml --spacing 2 --dimension 256 --fp CudaRayCast --bp CudaVoxelBased
> rtkadmmwavelets -p . -r projections.mha -o rtkadmmwavelets.mha -g geometry.xml --spacing 2 --dimension 256 --fp CudaRayCast --bp CudaVoxelBased
> rtkconjugategradient -p . -r projections.mha -o rtkconjugategradient.mha -g geometry.xml --spacing 2 --dimension 256 --fp CudaRayCast --bp CudaVoxelBased
>
> Fdk and sart does execute and create the output file when done processing. The GPU and CPU usage drops to zero, memory is still allocated but the process does not terminate!
>
> rtkfdk -p . -r projections.mha -o fdk.mha -g geometry.xml --spacing 2 --dimension 256 --hardware cuda
> rtksart -p . -r projections.mha -o sart.mha -g geometry.xml --spacing 2 --dimension 256 --fp CudaRayCast --bp CudaVoxelBased
>
> You could kill the process manually but this would require monitoring the output file, GPU and CPU usage - not an ideal solution.
>
> PS: I work with Rune Think at Odense University Hospital that also have collaborated to rtk by providing data for testing on cone beam from Elekta. I configure and compile rtk and he is using it in his Pd.D.
>
> Regards,
>
> Jonas Westberg
> MS in Computer Science ; Software Developer
> Department of Oncology
> Laboratory of Radiation Physics
> jonas.westberg at rsyd.dk - www.ouh.dk
>
> OUH
> Odense University Hospital
>
> Sdr. Boulevard 29, Entrance 97, 5000 Odense C
> Denmark
>
>
>
> _______________________________________________
> Rtk-users mailing list
> Rtk-users at public.kitware.com
> http://public.kitware.com/mailman/listinfo/rtk-users