I have written a DirectCompte program that sends some data to the GPU, processes the data and then retrieves the resultant grid. i.e. this is a GPGPU application. Is anyone familiar with how to do the below without retrieving ALL the Grid data between iterations and sending ALL grid data back to the GPU for every dispatch. There seems to be no easy way to do this sort of thing with DIrectX. Any reasonable direction much appreciated.
i.e Normal process is:Grid data Resource allocation+view, Dispatch x 100 iterations, Retrieve Grid data from GPU
Now I want to run it as:
Resouce allocation + view
Dispatch 1st iteration
Retrieve only a small part of the GRID. i.e. the first 10 grid points
Update these first 10 grid points with new values from elsewhere
Send the updated 10 grid points back to GPU
Dispatch 2nd iteration
...... as before ....... 10 grid points only
Dispatch last iteration
Retrieve all Grid data
end of process.
moving some data off and on GPU between dispatches
1 reply to this topic