Bernabé et al. report on how to improve the performance of a method to encode medical video sequences on Intel’s hyper-threading processors. They study several parallelization schemes and conclude that existing automatic techniques do not work well; thus, they focus on data-domain and functional decomposition, and prove that the latter can be implemented at a reasonable cost and results in a significant speedup.
The paper is well organized and its results are easy to follow, even if the reader is not familiar with video encoding techniques or the hyper-threading technology. The method seems solid enough, and the optimizations proposed make it a good choice for encoding medical video sequences. My only complaint is that the experimental results are scarce, since only one video sequence of 64 frames is considered and only a processor, a compiler, and an operating system are used. It would have been useful to study how the optimizations perform on longer sequences, and on sequences in different medical scenarios, and how the results are affected by changes in the compiler or the operating system. Repeating the tests on several hyper-threading processors would also have helped enhance the study.
The reader who is interested in this paper might well be interested in another paper by Tenllado et al. [1], in which the authors report on optimizing the discrete wavelet transform on hyper-threading processors.