Fix segmentation fault for models exceeding 40B on AMD GPUs & optimize mul_mat_axpy operation (#217)

* fix segementation fault when model exceeds 40B on ROCm platform

* optimize axpy kernel

* optimize op: mulmat_axpy_sparse

* fix bug when model exceeds 40B on AMD GPU

* optimize op: mulmat_axpy_sparse

---------

Co-authored-by: tworan <tworan.csu@gmail.com>