Help needed : Use a custom accelerator device to offload OpenCl Kernels for GROMACS
I want to use a custom accelerator device to offload OpenCl Kernels for GROMACS. My acclerator is using POCL runtime to compute kernels.
Here are the steps I followed
-
I added my vendor name to the GROMACS codebase so the accelerator can be recognised.
-
I am able to run Dummy kernel (defined in ‘gromacs/src/gromacs/hardware/device_management_ocl.cpp’) on my accelerator.
-
When I try to run actual kernels, I get segmentation fault.
I'm attaching some snapshots for the same.
-
Here are the test cases which are failing
-
Here is the logs for one of the kernel 'nbnxn_kernel_prune_opencl' which fails due to segmentation fault
I’m stuck here for long time Help me to fix this.