Meh--I was thinking more something that would adapt to the memory available. :)
But if nothing better comes out of the woodwork, I'll merge this. Andreas On Donnerstag 15 Oktober 2009, Nicolas Pinto wrote: > If you'd like to prepare a fix, I'd certainly merge it. > > Is the following patch enough ? > > --- transpose_original.py 2009-10-15 17:55:54.000000000 -0400 > +++ transpose.py 2009-10-15 17:54:15.000000000 -0400 > @@ -133,9 +133,8 @@ > def check_transpose(): > from pycuda.curandom import rand > > - for i in numpy.arange(10, 13, 0.125): > + for i in numpy.arange(10, 11.5, 0.125): > size = int(((2**i) // 16) * 16) > - print size > > source = rand((size, size), dtype=numpy.float32) > > @@ -158,9 +157,11 @@ > sizes = [] > bandwidths = [] > times = [] > - for i in numpy.arange(10, 13, 2**(-6)): > + for i in numpy.arange(10, 11.5, 2**(-6)): > size = int(((2**i) // 16) * 16) > > + print size, i > + > source = rand((size, size), dtype=numpy.float32) > target = gpuarray.empty((size, size), dtype=source.dtype) >
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ PyCUDA mailing list [email protected] http://tiker.net/mailman/listinfo/pycuda_tiker.net
