Oberon Community Platform Forum

Projects => Matrix => Topic started by: Pat on September 09, 2015, 02:09:45 PM

Title: Enhanced linear algebra performance with FoxCompiler doing implicit SSE
Post by: Pat on September 09, 2015, 02:09:45 PM
Hi all
The evolution of the Fox Compiler now including implicit use of SSE for matrix operations has significantly enhanced the speed of the Matrix library(https://www.ocp.inf.ethz.ch/svn/ocp/trunk/Matrix/) , on old hardware as well as on recent multicore machines.
See http://www.ocp.inf.ethz.ch/wiki/MatrixProject/Front
and add your own performance measurements.

- Thanks, Felix Friedrich, Florian Negele, and Alexey Morozov for the compiler enhancements.

Title: Re: Enhanced linear algebra performance - now also for small static arrays
Post by: Pat on September 10, 2015, 01:49:26 PM
The small Matrix operations
Matrix2x2, Matrix3x3,Matrix4x4, MatVec4
multiplications have got a huge performance boost as the small array optimizations in MathArrayBaseOptimized.Mod have been revived by Alexey Morozov today. (Need to recompile the most recent Fox*.Mod and the upgraded LinpackBenchmark.TestMatMul ).