Slava Pestov wrote: > Hi Phil, > > Looks like my latest coalescing changes broke a few things in the > build reports, but I'm working through those... however I noticed that > performance on benchmark.dawes improved even more. It went from 229 ms > in the May 31st build, to 160 ms two weeks ago, and now its running at > 105ms. This is on my Core 2 Duo Mac Book Pro. >
Excellent work - that's a v impressive improvement. > [ ... ] > If I recall correctly, I based this benchmark on the inner loop of > some bloom filter code you posted at one point. It would be > interesting to see if the original code got faster too. > I'm really looking forward to digging into the new compiler. The problem I was trying to solve was basically doing indices for a set of inputs: http://www.phildawes.net/blog/2008/09/10/searching-arrays-in-x86-assembler-with-a-bloom-filter/ " I have a set of 32bit integers, and I want to search a large unsorted 32bit array accumulating the array indices of values matching those from the set. To solve this in factor I've been putting the input set in a hashtable and then sequentially looking up each element from the array, which algorithmically is pretty optimal O(n+m) but the ever present constant is pretty high. This takes about 4 seconds on a processor pegged at 800mhz for a 48MB array." I ended up using a simple bloom filter and some assembler to find approximate matches and then filtering in factor afterwards, for a total of 100ms or so at 800mhz. http://phildawes.net/blog/2008/09/12/searching-arrays-in-x86-assembler-with-a-bloom-filter-pt-3/ It'd be cool to try a pure factor implementation of the same approach so I'll try this again soon. Cheers, Phil ------------------------------------------------------------------------------ Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ Factor-talk mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/factor-talk
