Andreas Klöckner
Andreas Klöckner
I can believe that it would behave the same way. I've disabled it, too, pending a fix.
There's not currently code to do that, but I'd be happy to consider a patch if it's sufficiently clean. A particular problem is that images are often 2D.
Yes, I can see not-too-invasive ways in which the current code could be generalized to that setting. In terms of "the underlying buffer" of an image, I'm not sure there...
TBH, I don't see the point of this. Why not keep an array of integers and maintain information about the enum semantics at a higher level in the software stack?
I tried reproducing this, but I discovered it appears I need `silx` installed. Is `pip install silx` enough?
#447 is a very similar problem that can probably reuse the solution for this.
* `eneuque_copy` unifies all the copy interfaces, and it adopts C `memcpy`'s (target, source) ordering consistently. That's admittedly different from `clEnqueueCopyBuffer`, but at this point I'd say that's unlikely to...
#452 has deprecations to standardize on `dst_offset`.
I'd like to do that, and I'd be happy to have help. Could you send a PR?
TBH, changing the existing Boost.Python wrapper to use pybind11 should be much easier than starting from scratch. BPL and pybind11 are extremely similar.