On 13.02.2018 12:38, Hendrik Leppkes wrote: > On Tue, Feb 13, 2018 at 12:31 PM, Roman Arzumanyan > <rarzuman...@nvidia.com> wrote: >> Please find attached the patch that introduces structure to store CUDA >> function pointers to be used within scale & thumbnail filters. >> > > The entire point of our dynlink_cuda.h was to not include the full > cuda.h, and as such not require having the CUDA SDK installed. > > - Hendrik
agreed. This would only be interesting if it kept the ability to build without cuda.h being present. Possibly optionally this would be ok, if the original features would be kept intact. Then it would only avoid a hard runtime dependency on the cuda libraries. Keep in mind that in the not too distant future our cuda/nv header will move out of the main ffmpeg repository to http://git.videolan.org/?p=ffmpeg/nv-codec-headers.git;a=summary So I'd say it's better to integrate the patch there, possibly duplicating the dynlink_cuda header into one version that works without cuda.h and one that uses it to offer a more complete dynloading.
Description: S/MIME Cryptographic Signature
_______________________________________________ ffmpeg-devel mailing list firstname.lastname@example.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel