ocl: GCN atomics for Mi-GPUs and other improvements #1377

Job	Run time
build-on-cuda (OPENMP=ON)	14m 1s
build-on-opencl (OPENMP=ON, SMM=libxsmm)	11m 12s
build-and-test (MPI=ON, OPENMP=ON, SMM=blas, openmpi)	8m 44s
pre-commit	1m 42s
build-and-test (MPI=ON, OPENMP=ON, SMM=blas, mpich)	6m 31s
build-on-rocm (OPENMP=ON)	12m 27s
build-and-test (MPI=ON, OPENMP=ON, SMM=libxsmm, openmpi)	8m 54s
build-and-test (MPI=ON, OPENMP=ON, SMM=libxsmm, mpich)	6m 43s
build-and-test (MPI=ON, OPENMP=OFF, SMM=blas, openmpi)	5m 20s
build-and-test (MPI=ON, OPENMP=OFF, SMM=blas, mpich)	3m 58s
build-and-test (MPI=ON, OPENMP=OFF, SMM=libxsmm, openmpi)	5m 30s
build-and-test (MPI=ON, OPENMP=OFF, SMM=libxsmm, mpich)	3m 59s
build-and-test (MPI=OFF, OPENMP=ON, SMM=blas, openmpi)	7m 14s
build-and-test (MPI=OFF, OPENMP=ON, SMM=libxsmm, openmpi)	8m 20s
build-and-test (MPI=OFF, OPENMP=OFF, SMM=blas, openmpi)	5m 51s
build-and-test (MPI=OFF, OPENMP=OFF, SMM=libxsmm, openmpi)	4m 28s
Combine & check coverage.	53s
	1h 55m 47s

Provide feedback