Well, heres round one of patches extracted from GIFE. applying these ten patches makes the feature extractor hold a copy of the value plane, in double precision. it prevents a conversion from double->int->double, and therefore improves the speed of the feature extractor by 5%.
the whole "reverse ordering thing" is part of future proofing for my next set of optimizations. Feel free to apply, comment, whatever. 5% is good enough to start, right? ;) Julia Longtin <[EMAIL PROTECTED]>
patches_round_1.tar.gz
Description: GNU Zip compressed data
_______________________________________________ help-GIFT mailing list [email protected] http://lists.gnu.org/mailman/listinfo/help-gift
