cython scan: use int64 to don't limit the size of inputs
Currently cython scan use int32 for all data in C. In particular, if we scan over each element of a vector, we limit the size of the vector to 2G. This could happen for tensor with more dimensions but the probability get lower.
If we decide to continus in this type of scan implementation(see James proposal) We should remove this problem at some point.
If we decide to continus in this type of scan implementation(see James proposal) We should remove this problem at some point.
Leave a comment