Hiroshi Inoue           

"Fast Interpolation of Grid Data at a Non-Grid Point"
Hiroshi Inoue
IEEE International Conference on Big Data (IEEE BigData 2017), Boston, MA, USA, December 11-14, 2017.

Full text [PDF]: IEEEBigdata2017_RabbitCT.pdf
Slide [PDF]: IEEEBigData2017_RabbitCT_slides.pdf

Abstract

Defining data at a non-grid point by interpolating grid data is a common operation in many workloads including scientific applications and imaging applications. This paper describes our technique to accelerate this interpolation operation and show its performance benefit using 3D computed tomography reconstruction. The 3D CT is one of the compute-intensive medical imaging applications that frequently interpolates grid data (2D images) at a non-grid point. To efficiently execute this operation with SIMD instructions, we create an in-memory pre-computed table from the input 2D image at runtime before projecting voxels onto each image to 1) reduce the amount of computation and 2) avoid non-contiguous memory accesses that attenuate the benefits of SIMD instructions. We implemented and evaluated our pre-computation technique using a bilinear interpolation and a 3rd-degree Lagrange interpolation on POWER8 processors; it yields up to 75% and 57% performance improvements in the RabbitCT benchmark for the two interpolation algorithms respectively.

Copyright (c) 2017 by IEEE. Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit. To copy otherwise, to republish, to post on servers, or to redistribute to lists, requires prior specific permission and/or a fee.