You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Further to discussion about porting CUFFT.jl to use CUDAdrv (JuliaAttic/CUFFT.jl#12), it would be useful to provide support for CUDA pitched pointers, in addition to standard memory types.
How might such an implementation be best approached, given longer term plans for this package (e.g., is there still a plan to move to the CuArray implementation?).
The text was updated successfully, but these errors were encountered:
is there still a plan to move to the CuArray implementation?
Yes, I'm gradually moving towards deprecating CUDAdrv.CuArray, but am waiting for 0.7/1.0 to be supported by GPUArrays/CuArrays. In that world, using these new alloc methods would be a responsibility of an array type like CuArrays.jl, taking into account the overhead trade offs and effective pitch to index correctly.
Further to discussion about porting
CUFFT.jl
to useCUDAdrv
(JuliaAttic/CUFFT.jl#12), it would be useful to provide support for CUDA pitched pointers, in addition to standard memory types.How might such an implementation be best approached, given longer term plans for this package (e.g., is there still a plan to move to the
CuArray
implementation?).The text was updated successfully, but these errors were encountered: