Changed in master. Thx! Andreas
On Dienstag 11 August 2009, wtftc wrote:
> I got pycuda working Vista x64 machine with python 2.6, but the
> platform_bits method in tools.py is not working for me. It returns 32 on my
> 64 bit platform, which is causing test_driver.py to fail.
>
> The error I am getting is the following:
> C:\Python26\lib\site-packages\pycuda-0.94beta-py2.6-win-amd64.egg\pycuda\to
>ols.py:343: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
> _ _ _ _ _ _ _ _
>
> dtype = dtype('uint64'), with_fp_tex_hack = False
>
> def dtype_to_ctype(dtype, with_fp_tex_hack=False):
> if dtype is None:
> raise ValueError("dtype may not be None")
>
> import numpy
> dtype = numpy.dtype(dtype)
> if dtype == numpy.int64 and platform_bits() == 64:
> return "long"
> elif dtype == numpy.uint64 and platform_bits() == 64:
> return "unsigned long"
> elif dtype == numpy.int32:
> return "int"
> elif dtype == numpy.uint32:
> return "unsigned int"
> elif dtype == numpy.int16:
> return "short int"
> elif dtype == numpy.uint16:
> return "short unsigned int"
> elif dtype == numpy.int8:
> return "signed char"
> elif dtype == numpy.uint8:
> return "unsigned char"
> elif dtype == numpy.float32:
> if with_fp_tex_hack:
> return "fp_tex_float"
> else:
> return "float"
> elif dtype == numpy.float64:
> if with_fp_tex_hack:
> return "fp_tex_double"
> else:
> return "double"
>
> else:
> > raise ValueError, "unable to map dtype '%s'" % dtype
>
> E ValueError: unable to map dtype 'uint64'
>
> May I suggest an alternate implementation to this (platform_bits1):
> Python 2.6.2 (r262:71605, Apr 14 2009, 22:46:50) [MSC v.1500 64 bit
> (AMD64)] on
> win32
> Type "help", "copyright", "credits" or "license" for more information.
>
> >>> def platform_bits():
>
> ... from struct import calcsize
> ... return calcsize('l') * 8
> ...
>
> >>> print platform_bits()
>
> 32
>
> >>> def platform_bits1():
>
> ... return tuple.__itemsize__ * 8
> ...
>
> >>> print platform_bits1()
>
> 64
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ PyCUDA mailing list [email protected] http://tiker.net/mailman/listinfo/pycuda_tiker.net
