Detail publikace

Tuning Parallel Performance of Global Image Processing Operators

Originální název

Tuning Parallel Performance of Global Image Processing Operators

Anglický název

Tuning Parallel Performance of Global Image Processing Operators

Jazyk

en

Originální abstrakt

Parallel performance and power consumption is all that usually matters in custom computing for image processing tasks. Tuning the performance on a particular parallel platform is not easy and can be aided by simulation. This paper investigates performance of two global image processing operators, FFT and FWT, on distributed (shared) memory architecture. As the chip multiprocessors and processor arrays are becoming ever more popular, understanding their parallel performance is essential. Variations in processor count, clock rate, link speed, bus bandwidth, cache line size, cache size, as well as mapping software components to processors influence performance a great deal and are analyzed in the paper.

Anglický abstrakt

Parallel performance and power consumption is all that usually matters in custom computing for image processing tasks. Tuning the performance on a particular parallel platform is not easy and can be aided by simulation. This paper investigates performance of two global image processing operators, FFT and FWT, on distributed (shared) memory architecture. As the chip multiprocessors and processor arrays are becoming ever more popular, understanding their parallel performance is essential. Variations in processor count, clock rate, link speed, bus bandwidth, cache line size, cache size, as well as mapping software components to processors influence performance a great deal and are analyzed in the paper.

BibTex


@inproceedings{BUT13977,
  author="Vladimír {Kutálek} and Václav {Dvořák}",
  title="Tuning Parallel Performance of Global Image Processing Operators",
  annote="Parallel performance and power consumption is all that usually matters
in custom computing for image processing tasks. Tuning the performance
on a particular parallel platform is not easy and can be aided by
simulation. This paper investigates performance of two global image
processing operators, FFT and FWT, on distributed (shared) memory
architecture. As the chip multiprocessors and processor arrays are
becoming ever more popular, understanding their parallel performance is
essential. Variations in processor count, clock rate, link speed, bus
bandwidth, cache line size, cache size, as well as mapping software
components to processors influence performance a great deal and are
analyzed in the paper.",
  booktitle="Proceedings of 37th International Conference MOSIS´03 Modelling and Simulation of Systems",
  chapter="13977",
  year="2003",
  month="may",
  pages="57--64",
  type="conference paper"
}