Untitled

Bouganim, Luc; Florescu, Daniela; Valduriez, Patrick

doi:10.1023/a:1008642513285

Cited by 11 publications

(1 citation statement)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Synchronous computations were carried out on each group of four cores with a shared memory, which matched the distribution of CPU cores across NUMA nodes and offers the quickest memory access for threads operating in synchronous mode. 13 The combination of synchronous and asynchronous parallelizations is schematically shown in Fig. 4.…”

Section: Combination Of Synchronous and Asynchronous Parallelizationsmentioning

confidence: 99%

Virtual prototyping of complex optical systems on multiprocessor workstations

Zhdanov

Sorokin

et al. 2022

Opt. Eng.

View full text Add to dashboard Cite

Nowadays, optical designers often use multiprocessor workstations for the virtual prototyping of complex optical systems. Modern workstations may have several CPUs with up to 128 virtual cores each and a non-uniform access speed to different memory areas. Effective implementation of virtual prototyping methods and algorithms requires the development of special methods for efficient algorithm parallelization and shared memory access. As a basis for the virtual prototyping, we proposed a progressive backward photon mapping method that allows for reducing the amount of data used by photon maps, speeding up the luminance calculation process, and estimating the luminance errors for the resulting image. The main algorithmic complexity of this method is the need to synchronize data when calculating and accumulating the luminance of indirect and caustic illumination. The authors propose the three-level semi-synchronous parallelization method, which consists of fully synchronous, semi-synchronous, and asynchronous levels with illumination processing effectively distributed among the computation threads. The main benefit of the developed method is that it does not require additional synchronization when accumulating luminance, thus increasing the image synthesis speed. The designed three-level method can also be used in distributed systems with good scalability. The results obtained with virtual prototypes of complex optical systems are presented.

show abstract

Section: Combination Of Synchronous and Asynchronous Parallelizationsmentioning

confidence: 99%