Closed UmarAbubacar closed 12 months ago
The GPU code is not tested in our CI, it's not unlikely that things break every now and then. I personally do not have access to a GPU machine so I have never used it.
It seems though as if the real_t
broke it judging from your errors. So you might want to try commit cebe047f81398dacf3e687aa06dee96c654650da
which is right before the introduction of real_t
but, unfortunately, ubuntu22 was not yet supported at that time. You may have to somehow add commit d1dc64eed1835bfbcc8d23bd93d638a0fe14455a
to add ubuntu 22.04 support.
Other than that, debugging the build and submitting a PR for it would be great.
The GPU code was done by @Senui , maybe he can help with the issue. :)
@UmarAbubacar did you manage to get that running?
@TobiasDuswald I managed to get it running with some changes but I quickly stopped using it as my simulations required non-spherical agents. The version I was testing with was master at the time. I think this code being experimental naturally requires some more effort to get it working.
I am trying to use the GPU acceleration code however when I try to build with the Dcuda=ON flag I first get errors for the path to "/samples/common/inc/helper_math.h". After I update this path in the CMakeLists.txt (line:634) to where the file is located on my computer, I get errors for missing files from the file: src/core/gpu/mechanical_forces_op_cuda_kernel.cu on line:
#include "core/gpu/helper_math_real_t.h"
When I replace the file with what I think it should be
#include "core/gpu/helper_math_double.h"
I then get errors saying real_t not defined.I am not sure if I need a specific version of the source code or a specific version of cuda. I am using the master branch on Ubuntu 22.04 and I have cuda version:
I tried to use the version of biodynamo mentioned in the paper https://arxiv.org/pdf/2105.00039.pdf but that also resulted in errors.