Dustin Kleckner <[email protected]> writes:
> I’m not familiar with the full details of the code, but naively I
> would expect this could be fixed with a single line of code (diff
> below).  Indeed this does work for me, but perhaps it breaks something
> else?

No, that's actually not correct. Since argument setting is stateful
(i.e. not thread-safe), you need a new copy of the kernel for each
thread from which you'll enqueue kernel runs, so there must be a way to
make multiple kernel instances. CL 2.0 introduces kernel cloning for
this purpose, but until that's widespread, we'll need this way of making
new kernels.

Andreas

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

Reply via email to