Parallel Processing in supportive formats
From PDI
Many of the current and upcoming standards of video have a potential for parallel processing. This feature can be used in multi processor paradigms like multi core processors or massively parallel processing on GPU using NVIDIA's CUDA platform or OpenCL. A lot of the ideas are related to the upcoming H.265 standard.
Contents |
Usual steps in decoding
Entropy decoding is the first stage of image reconstruction. The next step in reconstruction of a frame is prediction done which is intra / inter frame reconstruction. This is followed by inverse DCT transform to get the residual field, which when added to the reconstructed image, gives the decoded frame.
Possibility of Parallellization
Inter frame prediction can be parallelized trivially, especially on a GPU. This does not have dependence on currently decoded frame, and hence each of the inter predicted blocks can be decoded parallelly.
Intra frame prediction depends on the current frame. This cannot be parallelly. However many of the frames have only about 10-20% of the blocks intra predicted. This implies that several chains of independent intra blocks. These chains can be processed parallelly.
Inverse transform stage can be processed parallelly trivially, because each block is independent of the other.
Entropy decoding in parallel
Displaying directly
Frame reconstruction done on the GPU, gives rise to the possibility of displaying the frame directly on the screen. This is of great advantage because transferring data via the PCI bus could have been very costly otherwise.
