Zigfried Hampel-Arias <zig.hampel-ar...@icecube.wisc.edu> writes: > I have a large (~500 MB) lookup table loaded as a numpy array > for an embarrassingly parallel transport calculation. This is > too large for constant memory, but I can load it into global > memory, permitting all threads access to any element > required. However each thread must grab a set of 4 float values > at every transport step. I eventually want to move to accessing > 9 such sets (36 floats) at each step to do some table > interpolation. Are there any good PyOpenCL examples putting such > a read-only table into image memory? I’m not sure how much > performance gain moving to image mem there would be > anyways. Anyone have a suggestion? Thanks,
You could use one of the tests as an example: https://github.com/inducer/pyopencl/blob/master/test/test_wrapper.py#L357-L425 Andreas _______________________________________________ PyOpenCL mailing list PyOpenCL@tiker.net https://lists.tiker.net/listinfo/pyopencl