On Tue, 3 May 2011 17:06:22 -0400, Frédéric Bastien <no...@nouiz.org> wrote: > I find those 2 names good. > > I have 2 questions about .contig_bytes: > 1) Why do we need it? As we changed the interface, I don't see a need. > User can use .flags.forc to check if it is contiguous.
I actually like the semantics of asserting contiguity by choosing to
access the data through a certain attribute.
But having .gpudata and .bytes seems asymmetric. So I'd remove .gpudata
From documentation, then deprecate it (in a few months), and finally
remove it a few releases down the road, only keeping .bytes and
.contig_bytes.
> 2) What should .contig_bytes do when it is not contiguous? Don't
> exist? Copy it automatically?
No, just be a synonym for what .gpudata is now.
> Do have the copy automatic, I think we should use the same
> interface as numpy.{ascontiguousarray(c contiguous),asfortranarray}.
Agreed--if it costs time, it shouldn't be an attribute.
Andreas
pgpQdVNQOGIxY.pgp
Description: PGP signature
_______________________________________________ PyCUDA mailing list PyCUDA@tiker.net http://lists.tiker.net/listinfo/pycuda
