Skip to content

Latest commit

 

History

History
70 lines (51 loc) · 3.37 KB

README.md

File metadata and controls

70 lines (51 loc) · 3.37 KB
Raw

FastDifferentialCoding

Build Status

These functions compute fast successive differences, and recover the original values from the fast successive differences (i.e., they compute a prefix sum) using fast SIMD instructions.

They can be useful as part of compressed index.

Rationale

Array of sorted integers, such as 1, 50, 100, 110, 120... are more often compressible when we first compute the successive differences (49, 50, 10, 10,....). The downside of this improved compressibility is the need to recover the original integers by computing the prefix sum.

Programmers are often tempted to computer a prefix sum in the straight-forward manner:

for (size_t i=1;i<size;i++) {
  data[i] += data[i-1];
}

Unfortunately, this approach can be slow. Thus we prefer to use SIMD instructions available on recent Intel processors (SSE, available since the Pentium 4).

This library provide such fast functions.

Usage

compute_deltas_inplace(datain, length, 0);// compute the deltas in-place
compute_prefix_sum_inplace(datain, length, 0);// under the delta computations (datain is back)

See example.c for a complete example.

Reference

Daniel Lemire, Nathan Kurz, Leonid Boytsov, SIMD Compression and the Intersection of Sorted Integers, Software: Practice and Experience 46 (6), 2016. http://arxiv.org/abs/1401.6399

There are other relevant references. For example, this code was used in Crane et al. (2017):

Matt Crane, J.Shane Culpepper, Jimmy Lin, Joel Mackenzie and Andrew Trotman, A Comparison of Document-at-a-Time and Score-at-a-Time Query Evaluation, Proceedings of the Tenth ACM International Conference on Web Search and Data Mining (WSDM 2017), 2017.

Other libraries