GPUs are powerful, multi-core processors that provide a combination of task parallelism, thread parallelism, and data parallelism which can be used for high-throughput decoding of LDPC codes [4, 5].
They focus on exploiting the task parallelism of the algorithms, and results prove that this approach provides remarkable performance.
These devices are characterized by a different computation paradigm, focusing on exploiting the task parallelism of the algorithms [14].
All previous works [15 17] to use GPUs for Random Forest classification have relied on coarse-grained task parallelism and have yielded unsatisfactory results.
OmpSs is a programming model that provides a simple and powerful way of annotating sequential programs to exploit heterogeneity and task parallelism based on runtime data dependency analysis, dataflow scheduling and out-of-order task execution; it has greatly influenced Version 4.0 of the OpenMP standard.
Parallelization of multitask application is based on the main idea that most of the parallel applications were built upon a limited number of recurring schemes of parallelization (parallel skeletons) such as SCM for data or static task parallelism, FARM for dynamic data sharing (or dynamic task sharing), and PIPE for task-parallel parallelism, etc.
Nuno Amado et al. [14] presents an overview on the various ideas regarding implementing parallel decision trees using data parallelism, task parallelism and hybrid parallelism in their work.
The first approach is task parallelism using multicore computer and MATLAB® parallel computing toolbox.
Since BB modems require SP accelerators that are performing the same SP tasks on incoming streams of data, there are significant data and task parallelism available, which we exploit in our proposed architecture using the intensive parallel processing capability of a GPU.
Task parallelism involves executing independent tasks directly in parallel while data parallelism requires the identification of tasks whose input can be fragmented in chunks and processed in parallel.
