01.10.2015 Views

Histogram calculation in CUDA

Whitepaper - nVIDIA CUDA Toolkit Documentation

Whitepaper - nVIDIA CUDA Toolkit Documentation

SHOW MORE
SHOW LESS
  • No tags were found...

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Abstract<br />

<strong>Histogram</strong>s are a commonly used analysis tool <strong>in</strong> image process<strong>in</strong>g and data m<strong>in</strong><strong>in</strong>g<br />

applications. They show the frequency of occurrence of each data element.<br />

Although trivial to compute on the CPU, histograms are traditionally quite difficult to<br />

compute efficiently on the GPU. Previously proposed methods <strong>in</strong>clude us<strong>in</strong>g the occlusion<br />

query mechanism (which requires a render<strong>in</strong>g pass for each histogram bucket), or sort<strong>in</strong>g the<br />

pixels of the image and then search<strong>in</strong>g for the start of each bucket, both of which are quite<br />

expensive.<br />

We can use <strong>CUDA</strong> and the shared memory to efficiently produce histograms, which can<br />

then either be read back to the host or kept on the GPU for later use. The two <strong>CUDA</strong> SDK<br />

samples: histogram64 and histogram256 demonstrate different approaches to efficient<br />

histogram computation on GPU us<strong>in</strong>g <strong>CUDA</strong>.<br />

NVIDIA Corporation<br />

2701 San Tomas Expressway<br />

Santa Clara, CA 95050<br />

www.nvidia.com

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!