CUDASparseContext#

class MCEq.solvers.CUDASparseContext(int_m, dec_m, device_id=0)[source]#

Bases: object

This class handles the transfer between CPU and GPU memory, and the calling of GPU kernels. Initialized by MCEq.core.MCEqRun and used by solv_CUDA_sparse().

Methods Summary

`alloc_grid_sol`(dim, nsols)	Allocates memory for intermediate if grid solution requested.
`dump_sol`()	Saves current solution to a new index in grid solution memory.
`get_gridsol`()	Downloads grid solution to main memory.
`get_phi`()	Downloads current solution from GPU memory.
`set_matrices`(int_m, dec_m)	Upload sparce matrices to GPU memory
`set_phi`(phi)	Uploads initial condition to GPU memory.
`solve_step`(rho_inv, dX)	Makes one solver step on GPU using cuSparse (BLAS)

Methods Documentation

alloc_grid_sol(dim, nsols)[source]#: Allocates memory for intermediate if grid solution requested.

dump_sol()[source]#: Saves current solution to a new index in grid solution memory.

get_gridsol()[source]#: Downloads grid solution to main memory.

get_phi()[source]#: Downloads current solution from GPU memory.

set_matrices(int_m, dec_m)[source]#: Upload sparce matrices to GPU memory

set_phi(phi)[source]#: Uploads initial condition to GPU memory.

solve_step(rho_inv, dX)[source]#: Makes one solver step on GPU using cuSparse (BLAS)