Hello Tomasz,
Sorry for taking so long to check your code. I think I found the error
which cause ScrambledSobol64 to fail. In curandom.py:
class
ScrambledSobol64RandomNumberGenerator(_ScrambledSobolRandomNumberGeneratorBase):
...
def __init__(self, dir_vector=None, scramble_vector=None, offset=0):
...
'curandStateScrambledSobol64',
sobol64_random_source+random_skip_ahead64_source)
The last line should be
'curandStateScrambledSobol64',
scrambledsobol64_random_source+random_skip_ahead64_source)
(artifact of copy-paste, I assume ;)
With this change test_gpuarray.py passes on my Fermi box.
Best regards,
Bogdan
On Fri, Sep 2, 2011 at 12:41 AM, Tomasz Rybak <[email protected]> wrote:
> Dnia 2011-08-26, pią o godzinie 18:09 +0200, Tomasz Rybak pisze:
>> I send new patch version.
>> Bogdan - have you tried to apply patch by
>> "git apply <*.patch" and then git complained about
>> 4 trailing whitespaces? I have fixed it now.
>>
>> I always apply patch using "patch -p1 <*.patch" - I guess
>> it is time to also use git ;-)
>>
>> At the same time I have just checked on GeForce 460,
>> and the test still fails. It looks like error is in the
>> call of cuMemcpyDtoH during copying scramble vector
>> for scrambled Sobol64 - I will look into it in the next week.
>
> New patch.
> I still do not know what is causing Sobol64 to fail,
> so I have disabled it.
> I will enable it as soon as reason for it to fail
> on Fermi is known and fixed.
>
> Best regards.
>
>
>
> --
> Tomasz Rybak <[email protected]> GPG/PGP key ID: 2AD5 9860
> Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860
> http://member.acm.org/~tomaszrybak
>
> _______________________________________________
> PyCUDA mailing list
> [email protected]
> http://lists.tiker.net/listinfo/pycuda
>
>
_______________________________________________
PyCUDA mailing list
[email protected]
http://lists.tiker.net/listinfo/pycuda
