summaryrefslogtreecommitdiff
ModeNameSize
-rw-r--r--CHANGELOG369logplain
-rw-r--r--README1027logplain
-rwxr-xr-xhib-dlagent4710logplain
hings are more understandableEric Anderson Most things are always "compiled" now, but testing did show that gcc's dead code elimination successfully removes them from the executable. Testing did not show any performance impact. 2010-12-27lcs_upper_bound's second sum is required all the time and try to handle endiansEric Anderson 2010-12-26Only do second sum if it can make a differenceEric Anderson 2010-12-26Auto-choose at runtime to do second sum in lcs_upper_boundEric Anderson This has no impact for non-DO_SECOND_SUM, but puts DO_SECOND_SUM less than a percent behind. 2010-12-26Simple movement of return calculation in lcs_upper_boundEric Anderson This increased performance of the non-DO_SECOND_SUM code a _lot_ for something so small. 2010-12-26Only add one to lcs_upper_bound or don't do the second sum.Eric Anderson The DO_SECOND_SUM define has reduced false positives, but without it has a reasonable amount less work. Thus, you must empirically test to find which has the best boost. Right now, DO_SECOND_SUM is faster. 2010-12-26Add lcs_upper_bound check before lcs_32_rough for added speedEric Anderson