05.08.2012 Views

GPU Sparse Graph Traversal

GPU Sparse Graph Traversal

GPU Sparse Graph Traversal

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Prefix sum for allocation<br />

Input (e.g., adjacency list sizes)<br />

Result of prefix sum<br />

Frontier output<br />

Duane Merrill, University of Virginia<br />

A C D G<br />

2 1 0 3 2<br />

A0 2C 3 D3 G6<br />

Thread<br />

Thread Thread Thread Thread<br />

0 1 2 3 4 5 6 7<br />

Each output is computed to be the sum<br />

of the previous inputs<br />

O(n) work<br />

Use results as a scatter offsets<br />

Fits the <strong>GPU</strong> machine model well 1<br />

Proceed at copy-bandwidth<br />

Only ~8 thread-instructions per input<br />

1. Merrill et al. Parallel Scan for Stream Architectures. Technical<br />

Report CS2009-14, University of Virginia. 2009<br />

13

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

Saved successfully!

Ooh no, something went wrong!