[FFmpeg-devel] [Hardware Accelerator for Video and Audio]

[Andreas Setterlind]

Another new question; will your video accelerator processor chip be
software programmable? I mean in a way similar to how you can write
custom pixel shaders, OpenGL Shading Language (GLSL, a.k.a. GLslang),
and Cg (C for Graphics) can do many different things on a GPU that it
may not have been hardcoded for, (I am thinking of example running
video post-processing filters, such as sharpen, blurring, denoising,
deringing, requantization, etc. after the normal video decoding
process).


On 2/20/08, kenye ye <kenyeby at gmail.com> wrote:
> define what you expect and want to accelerate.

Video decoding processes that I assume could be hardware accelerated:
- Motion compensation (mocomp)
- Inverse discrete cosine transform (iDCT)
- Inverse telecine conversion (IVTC) 2:3 pull-down
- Inverse modified discrete cosine transform (iMDCT)
- In-loop deblocking filter
- Intra-frame prediction
- Inverse quantization (IQ)
- Context-adaptive variable-length coding (CAVLC) entropy decoding
- Context-adaptive binary arithmetic coding (CABAC) entropy decoding
- Deinterlacing (spatial-temporal and other forms of deinterlacing)
- Bitstream processing
- Video scaling (like 4x4 scaling)
- YUV to RGB conversion, and RGB to YUV conversion
- Video Color Control (brightness, hue, saturation, contrast)

Video encoding processes that I assume could be hardware accelerated:
- Discrete cosine transform (DCT)
- Quantization
- Telecine conversion (3:2 pull-up)

Audio decoding processes that I think could maybe be accelerated:
- SBR (Spectral Band Replication) decoding (used by aacPlus and mp3PRO)
- PS (Parametric Stereo) decoding (used by aacPlus and mp3PRO)
- Sample rate converter (audio re-sampler). Ex; 48 KHz to 44.1 KHz sampling


I think that of all of those CABAC decoding/encoding will get the most benefit