Out of shared memory for J-engine

[wxj6000]

Some of CUDA kernel functions for MD J engine use shared memory more than 64 KB. Some Nvidia GPUs (even modern ones, such as CC 7.5) still only have 64 KB shared memory per SM. When J engine is used, it will raise invalid argument in the kernel launch.

https://docs.nvidia.com/cuda/cuda-c-programming-guide/#features-and-technical-specifications-technical-specifications-per-compute-capability