Is it foolish not to use DirectShow as the basis of a CUDA video processing pipeline for Windows?

When building a CUDA oriented video processing pipeline to run exclusively on windows,there are two approaches.

You either use DirectShow or you开发者_高级运维 don't.

The decision not to use directshow is also the decision to code every portion of the pipeline yourself.

This includes reading, parsing, processing and rendering or retransmission of the stream. if if

With DirectShow, however, you gain access to a fairly large body of existing components that are quite capable of doing some portion of the pipeline's work.

This means you may focus your dev process on coding one or more DirectShow sources, filters, or renderers which use CUDA to encapsulate any novel or otherwise unavailable processing.

And as a result of adopting a component based architecture, you are in a good position to develop alternate implementations i.e. OpenCL. without having to touch the other parts of the pipeline.

Are there any liabilities to DirectShow participation that warrant the extra work involved to develop and maintain the pieces that are not CUDA specific?

If you haven't already, it might be useful to have a look at the "cudaDecodeD3D9" demo in the NVidia GPU SDK (CUDA SDK). It decodes and renders a video without DirectShow (using CUDA and DirectX9). Similarly, there's a "cudaDecodeOpenGL" demo if you prefer.