Cuda-6.0 video decoding out of a pipe on Linux

Hi comunity,
I have compiled the cudaDecodeGL sample code, on an Ubuntu12.04 box. And it works fine on MPEG-2 ts file format. However, I would like to run it on live video, that is, on a continuous stream of MPEG-2 format.
Is there any API like ‘cuvidCreateVideoSource’ for named pipes (instead of for files)?
Thanks for helping?


I have found out that the cuda ‘cuvidCreateVideoSource()’ API inquires (fstat) very early on, about the size of the input file, then does a ‘lseek’ towards the last few bytes and reads them before anything else. And obviously ‘lseek’ fails when the input is a FIFO pipe.
So at this point, what options are there to decode a live video stream (on Linux)?
I have read in this forum a comment talking about avoiding ‘cuvidCreateVideoSource()’ API and using ‘cuvidCreateVideoParser()’ instead. Can anyone suggest a sample code or some explaining?
Thank you for any hint.


Does anyone have an answer for this? I have the same use case, trying to decode an arbitrary stream (pipe, socket, etc) rather than a file.