Breaking News
Design How-To

Transcoding video with parallel programming on multi-core processors

Page 1 / 5 Next >
More Related Links
View Comments: Newest First | Oldest First | Threaded View
User Rank
re: Transcoding video with parallel programming on multi-core processors
Peleg   12/12/2008 10:10:11 PM
Addressing the Transcoding issues require further understanding of the inner dependency among processing stages. For example, If one want to accelerate motion estimation, then there is an important issue of initial search vector that should be used for each macroblock. this impose specific serial dependency among the processed macroblocks. Therefore one could not do motion estimation in parallel for all of the macroblocks. This is only one example of how things get complicated when parallelizing encoders. Discussing Amdahl's law, it is important to be able to address the problems which are not "embarrassingly parallel". This means that decomposing into fine grain is required to address the "serial code", or hide (overlap) the effect of the serial code if not possible to further decompose. The problem with high overhead machines, such as GPU,PC is the poor support of fine grain computation and fast decision making. Future machines like the HyperCore technology address the mentioned issues, and enable the scaling to 100's of cores on a chip, speeding the computation linearly.

Most Recent Comments
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed