On 30/01/14 21:37, Andreas Kloeckner wrote:
> One thing that sort of killed sub-buffers for me as far as usefulness is
> concerned is that they are allowed to have restrictive (and
> platform-dependent!) alignment requirements, based on
> cl.device_info.MEM_BASE_ADDR_ALIGN. On all the platforms where I
> checked, that value is 1024 or 4096.

Yes, I was similarly disappointed by this requirement.  Further, the
spec also disallows creating sub-buffers of sub-buffers.  This (almost
certainly arbitrary) requirement requires extra bookkeeping on the part
of application developers.  (Whenever I make a sub-buffer I also need to
retain a reference to the original buffer and the offset of the sub buffer.)

Interestingly the NVIDIA OpenCL implementation allows one to forgo both
of the above requirements.

> Your offsets appear fine with respect to that, so I'm not sure what
> you're up against here. Looks like an Nvidia bug from here...

More than likely.  My only thought is if pyopencl would be thrown by:

"The implementation may return the same cl_mem object with the reference
count incremented appropriately for multiple calls to clCreateSubBuffer
that use the same values for buffer, flags, buffer_create_type and
buffer_create_info points to the same descriptor or descriptors that
describe values that are exactly the same."

Regards, Freddie.


Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
PyOpenCL mailing list
[email protected]
http://lists.tiker.net/listinfo/pyopencl

Reply via email to