BLAZE_DEVICE_AVAILABLE annotations for views

STEllAR-GROUP / blaze_cuda

WIP · CUDA compatibility for Blaze · https://bitbucket.org/blaze-lib/blaze

17 stars 3 forks source link

BLAZE_DEVICE_AVAILABLE annotations for views #9

Open JPenuchot opened 5 years ago

JPenuchot commented 5 years ago

Referring to https://github.com/STEllAR-GROUP/blaze/issues/1

hkaiser commented 5 years ago

@JPenuchot should we add similar annotations to the types in https://github.com/STEllAR-GROUP/blaze_tensor?

JPenuchot commented 5 years ago

@JPenuchot should we add similar annotations to the types in https://github.com/STEllAR-GROUP/blaze_tensor?

At some point it will be necessary, however I might rename the BLAZE_HOST_DEVICE macro to BLAZE_DEVICE_AVAILABLE or something like that to make its purpose more clear, I'll confirm that later

JPenuchot commented 5 years ago

See: https://github.com/harrism/cpp11-range/blob/70f844968c5f669ce85f8ce4cbd24a3584c57f4b/range.hpp#L9

JPenuchot commented 5 years ago

@hkaiser The macro has been changed to BLAZE_DEVICE_CALLABLE, the macros can be added to iterators now.

Here's a list of what should be checked before adding the macro:

The iterator has to carry a pointer to the underlying data at the end, not a reference to the CUDA vector/matrix/tensor
load() and other SIMD-related members should obviously not be device-callable

As you told me we're not rushing for that, getting core functionality working properly is the priority right now