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
It seems that caffe2::TensorShape and other tensor shape constraints are not supposed to be used in CUDA kernels. Does this mean that we can remove the CUDA code generation macros from dim.h?
The text was updated successfully, but these errors were encountered:
dim serve not only as tensor shape but also as element index. In the second case, It may be used in our hand-written CUDA kernels.
Actually, it costs few to implement CUDA supporting in dim, only a few macros and function prefixes. So I tend to keep them for now. If they are turned out to be unnecessary in the future, removing them will also be an easy task.
It seems that
caffe2::TensorShape
and other tensor shape constraints are not supposed to be used in CUDA kernels. Does this mean that we can remove the CUDA code generation macros from dim.h?The text was updated successfully, but these errors were encountered: