WIP: Add GPU aware MPI support in cannon algorithm #1472

Job	Run time
build-on-cuda (OPENMP=ON)	7m 58s
build-and-test (MPI=ON, OPENMP=ON, SMM=blas, openmpi)	4m 8s
pre-commit	1m 5s
build-and-test (MPI=ON, OPENMP=ON, SMM=blas, mpich)	4m 1s
build-on-opencl (OPENMP=ON, SMM=libxsmm)	8m 5s
build-and-test (MPI=ON, OPENMP=ON, SMM=libxsmm, openmpi)	4m 0s
build-and-test (MPI=ON, OPENMP=ON, SMM=libxsmm, mpich)	3m 43s
build-on-rocm (OPENMP=ON, G2G=ON)	4m 0s
build-and-test (MPI=ON, OPENMP=OFF, SMM=blas, openmpi)	9m 22s
build-on-rocm (OPENMP=ON, G2G=OFF)	4m 27s
build-and-test (MPI=ON, OPENMP=OFF, SMM=blas, mpich)	9m 34s
build-and-test (MPI=ON, OPENMP=OFF, SMM=libxsmm, openmpi)	8m 52s
build-and-test (MPI=ON, OPENMP=OFF, SMM=libxsmm, mpich)	9m 17s
build-and-test (MPI=OFF, OPENMP=ON, SMM=blas, openmpi)	5m 18s
build-and-test (MPI=OFF, OPENMP=ON, SMM=libxsmm, openmpi)	5m 18s
build-and-test (MPI=OFF, OPENMP=OFF, SMM=blas, openmpi)	3m 28s
build-and-test (MPI=OFF, OPENMP=OFF, SMM=libxsmm, openmpi)	3m 10s
Combine & check coverage.	40s
	1h 36m 26s

Provide feedback