Implement generic j-reduction in the nbnxm SYCL kernels
authorSzilárd Páll <pall.szilard@gmail.com>
Tue, 30 Mar 2021 13:24:08 +0000 (15:24 +0200)
committerMark Abraham <mark.j.abraham@gmail.com>
Wed, 26 May 2021 08:08:26 +0000 (08:08 +0000)
commit2c750f8e218bf8ace3e3395668102259e7337886
tree0973fa1804555de2ab1489a5c2c246dd4065113b
parentb5af08d11f7219d605f28975691b60c6c91ff595
Implement generic j-reduction in the nbnxm SYCL kernels

This commit implements the generic j-reduction identical to the OpenCL
version of the same.

Also added a subGroupBarrier() helper which is needed for
correctness on the CUDA backend when targetting NVIDIA
architectures.

Refs #3934
src/gromacs/gpu_utils/sycl_kernel_utils.h
src/gromacs/nbnxm/sycl/nbnxm_sycl_kernel.cpp
src/gromacs/nbnxm/sycl/nbnxm_sycl_kernel_utils.h