[PyCUDA] PyCUDA and fast user-generated code
bogomips at post.pl
Wed Nov 10 12:55:06 PST 2010
Dnia 2010-11-07, nie o godzinie 22:50 -0500, Andreas Kloeckner pisze:
> > Do you think NVidia would allow me to distribute nvcc with my program
> > (which may be
> > partially close-sourced for now, and could be commercial). This may make
> > PyCUDA less
> > necessary (although the feature-set of PyCUDA, such as the runtime
> > optimization does
> > look enticing in any case). I'm prepared to pay them (and you if I end up
> > using PyCUDA)
> > something, though I'm not rich yet :)
> If you're concerned about compiler availability, why don't you check out
> OpenCL? There, the compiler is part of the API and is guaranteed to be
> available at run time, without further distribution worries and
> licensing charges.
According to licence of NVIDIA CUDA toolkit:
In connection with Section 2.1.2 of this Agreement, the following files
be redistributed with software applications developed by Licensee,
including variations of these files that have version number
embedded in the file name.
Component Windows MacOS Linux
CUDA Runtime cudart.dll libcudart.dylib libcudart.so
CUDA FFT Library cufft.dll libcufft.dylib libcufft.so
CUDA BLAS Library cublas.dll libcublas.dylib libcublas.so
Those are libraries that you can distribute without any problems.
There is also "Linux clause":
2.1.3 Linux/FreeBSD Exception. Notwithstanding the foregoing terms of
Section 2.1.1, SOFTWARE designed exclusively for use on the Linux or
FreeBSD operating systems, or other operating systems derived from the
source code to these operating systems, may be copied and
provided that the binary files thereof are not modified in any way
for unzipping of compressed files).
So if your programs is to be run only on Linux, you should be able
to distribute large parts of CUDA. But:
1. I am not a lawyer - consult one before trying it.
2. This is rather intended for distribution which want to include
NVIDIA CUDA packages in them. I am not sure how it would be looked
at by real lawyers with copyright experience.
3. It limits you to Linux/BSD.
So IMO better follow Andreas' advice and look at OpenCL - especially
as there is PyOpenCL.
Tomasz Rybak <bogomips at post.pl> GPG/PGP key ID: 2AD5 9860
Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 836 bytes
Desc: This is a digitally signed message part
More information about the PyCUDA