These routines compute fluctuationsfp, fm, gp, gm and need to be written as a CUDA kernel(s).
Step2 computes fluctuations by sweeping in x directions and then in y directions. The routine flux2 is called for each horizontal or vertical strip of grid data. The routine flux2 then calls a Riemann solver for each grid interface.
I think it might make sense to combine step2 and flux2 to a single routine which calls a thread for each cell interface.
To start, let's just write a version of the flux computation that does the first order spilt update. That is, code that handles everything through about line 124 in cudaclaw5_flux2.f
We have the basic routines in place for this, so will close the issue. Still need to get limiters, second order corrections, and transverse corrections in place, though.
These routines compute
fluctuations
fp
,fm
,gp
,gm
and need to be written as a CUDA kernel(s).Step2 computes fluctuations by sweeping in x directions and then in y directions. The routine
flux2
is called for each horizontal or vertical strip of grid data. The routine flux2 then calls a Riemann solver for each grid interface.I think it might make sense to combine step2 and flux2 to a single routine which calls a thread for each cell interface.
To start, let's just write a version of the flux computation that does the first order spilt update. That is, code that handles everything through about line 124 in cudaclaw5_flux2.f