From: Mathieu M. <mat...@gm...> - 2009-10-27 08:28:18
|
On Mon, Oct 26, 2009 at 12:41 PM, Daniele E. Domenichelli <dan...@gm...> wrote: > Hello Mathieu, > Congratulation for the new release! > > Mathieu Malaterre wrote: >> I am still looking into integrating Daniele E. Domenichelli's patch to >> avoid VR:DS being longer than 16bytes, but I thought it was too new to >> be integrated now. I'd like to do so,e benchmark on real dataset and >> see the effect. > > > Have you seen this [1]? > > > I just committed on svn a fix for the only test failing... that was > actually a copy-paste error ;) > In next build it should work correctly! > > Of course using MSVC absolute error on some numbers will be a little > bigger due to the precision lost having 3 digits in exponent instead of 2... thanks for the fix ! I'll prepare a test so that we can compare timing in between the old (buggy) approach and the new one. -- Mathieu |