search

DIDSR / VICTRE_MCGPU

Version of MC-GPU used in the VICTRE project
32 stars 18 forks source link

sigint 11 when calculating voxel dose #14

Open davide84 opened 2 years ago

davide84 commented 2 years ago

Hi, I am running the Victre pipeline with minimal changes from git, to get familiar with the parameters. If I set the parameter "tally_voxel_dose" to YES the Python script ends with an exception of "Projection error". The file output_projection.out ends with:

Command terminated by signal 11
316.60user 2.20system 5:19.11elapsed 99%CPU (0avgtext+0avgdata 6978648maxresident)k
0inputs+70328outputs (0major+1731728minor)pagefaults 0swaps

The system is Ubuntu LTS 20.04, kernel 5.4, 64 GB of Ram (10% used during simulations), Intel processor and two GeForce 2080i. My pipeline was compiled with CUDA support and without MPI.

Any idea? Can I provide more data? Thanks in advance!

andreubs commented 2 years ago

Hi Davide. I usually don't activate the voxel dose tally with VICTRE_MCGPU because the VICTRE phantoms are so large that there is no space in the GPU memory to store the dose to each voxel. Defining a smaller ROI should work but I have not tested it lately.

When is the program failing: (a) during code initialization?, (b) during GPU kernel execution?, (c) in the final result reporting after the kernel?

It would help if you could provide the entire simulation output file to see the size of the ROI and all the error messages written by the program.

   Andreu
davide84 commented 2 years ago

Hi! Sorry for the late reply, this part of the project had to briefly be put on hold. I have now resumed the work on this part and investigate again the error, I was able to see that it was a memory issue of the GPU - previously I only checked the system RAM, sorry. With very small ROI regions there is no problem. Thanks again!