Citation Context

...ls search area (SA) on a commodity desktop platform equipped with a multi-core CPU and two GPUs. To the best of the authors’ knowledge, this is one of the first approaches that applies adaptive load balancing with dynamically built partial estimations of the FPMs to tackle efficient collaborative execution of complex multi-module problems, such as video encoding, in heterogenous environments. II. Related Work There are only few state-of-the-art approaches that deal with the efficient parallel implementation of the entire video encoder (or its main functional parts), namely, for multi-core CPU [21], GPU [22], or CPU+GPU [14, 15] environments. In CPU+GPU systems, these approaches either i) simply offload a single inter-loop module in its entirety (mainly the ME) to the GPU, while performing the rest of the encoder on the CPU [9,19], or ii) exploit simultaneous CPU+GPU processing at the level of a single inter-loop module [15, 23]. These approaches have a limited scalability (only one GPU can be employed) and cannot fully exploit the capabilities of CPU+GPU systems (since the CPU is idle, while the GPU processes the entire offloaded module) [19]. In [9] the pipelining granularity is decid...