Gregory Perry <[email protected]> writes:
>>Gregory Perry <[email protected]> writes:
>>> Quick question: based on Python's innate capability for arbitrary
>>> precision math, are there any ways to extend those capabilities with
>>> PyCUDA?  For example if I wanted to implement the RSA encryption
>>> algorithm using only PyCUDA, is there any ability there to leverage
>>> Python's multi precision math capabilities to generate keys and
>>> perform modulus arithmetic on a single CUDA hardware thread/core?
>>
>>There's no pre-made facility for this, you would have to write your
>>own. If the data is genuinely variable-size for each thread, then that
>>presents somewhat of a performance/coding issue. (But I feel like for
>>RSA it might not.)
>>
>>HTH,
>>Andreas
>
> This is somewhat new, haven't seen any talk of this yet:
>
> http://www.hpcs.cs.tsukuba.ac.jp/~nakayama/cump/
>
> I am surprised they were able to port GNUMP to CUDA, especially given
> the amount of hardware-specific voodoo GNUMP requires for 
> arbitrary precision math.

I suspected (but didn't know) that something like this might
exist. Thanks for the pointer.

Andreas

_______________________________________________
PyCUDA mailing list
[email protected]
http://lists.tiker.net/listinfo/pycuda

Reply via email to