Comments on: The Treacherous Optimization http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/ serious code Sat, 05 Jul 2008 18:16:17 +0000 http://wordpress.org/?v=1.5 by: Reinder http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-906 Tue, 30 May 2006 09:20:25 +0000 http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-906 Some possible improvements spring to my mind: - you should be able to offset tp by one, replacing "tp[-1]" by *tp (check first whether your compiler managed to do that, and whether this actually makes a difference on your CPU) - vectorize the search by running searches starting at offsets 0, N, 2N, 3N for haystacks of size N + length( searchString) - 1 in parallel (N would have to be fairly large relative to the search string; measure first whether there is memory bandwidth available) I also wonder whether the grep programmer used actual data to arrive at that 2,3,3,2 pattern. Some possible improvements spring to my mind:
- you should be able to offset tp by one, replacing “tp[-1]” by *tp (check first whether your compiler managed to do that, and whether this actually makes a difference on your CPU)
- vectorize the search by running searches starting at offsets 0, N, 2N, 3N for haystacks of size N + length( searchString) - 1 in parallel (N would have to be fairly large relative to the search string; measure first whether there is memory bandwidth available)

I also wonder whether the grep programmer used actual data to arrive at that 2,3,3,2 pattern.

]]> by: Scott http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-911 Tue, 30 May 2006 10:47:22 +0000 http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-911 I think you mean "prostrate" (with an "r") I think you mean “prostrate” (with an “r”)

]]> by: Vid Boi http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-913 Tue, 30 May 2006 11:20:36 +0000 http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-913 So the question is, how common are these partial matches in the real world (or for the average search in hex fiend)? If rare then perhaps the performance gain by the "treacherous grep" technique is worth it. If partial matches are common, then I would definitely favor the "reformed grep." So the question is, how common are these partial matches in the real world (or for the average search in hex fiend)? If rare then perhaps the performance gain by the “treacherous grep” technique is worth it. If partial matches are common, then I would definitely favor the “reformed grep.”

]]> by: ridiculous_fish http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-914 Tue, 30 May 2006 11:22:34 +0000 http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-914 How embarassing - thank you, Scott. How embarassing - thank you, Scott.

]]> by: Samh http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-920 Tue, 30 May 2006 14:10:20 +0000 http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-920 "clean shaven beer" ? I like my beer fuzz-free “clean shaven beer” ?

I like my beer fuzz-free

]]> by: Ian Holsman http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-925 Tue, 30 May 2006 17:23:21 +0000 http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-925 how does fgrep compare? It uses a different algorithm doesn't it? how does fgrep compare?
It uses a different algorithm doesn’t it?

]]> by: Steve Israelson http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-927 Tue, 30 May 2006 18:45:55 +0000 http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-927 Can't you use something other than character comparisons? Seems to me comparing 8 bytes at a time would be faster if they were aligned. You could make 8 byte shifted versions so you can always compare byte aligned data. Then switch to your method for the remainder bytes at the non aligned start and end of the pattern. Can’t you use something other than character comparisons? Seems to me comparing 8 bytes at a time would be faster if they were aligned. You could make 8 byte shifted versions so you can always compare byte aligned data. Then switch to your method for the remainder bytes at the non aligned start and end of the pattern.

]]> by: Ridiculous Fish takes on grep http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-931 Tue, 30 May 2006 20:20:37 +0000 http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-931 [...] Ridiculous Fish takes on grep Posted by Jon Baer Wed, 31 May 2006 03:20:18 GMT Ridiculous Fish takes on grep: "I love Ridiculous Fish's blog. It makes [...] [...] Ridiculous Fish takes on grep Posted by Jon Baer Wed, 31 May 2006 03:20:18 GMT Ridiculous Fish takes on grep: “I love Ridiculous Fish’s blog. It makes [...]

]]> by: Mark http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-933 Tue, 30 May 2006 20:32:44 +0000 http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-933 I did some snooping around with Google, and I think I've located a picture of your nemesis: http://ducky.net/~kim/web_pictures/jacob_1year/UncleMike_paino_Jacob.jpg (He's the taller one). I did some snooping around with Google, and I think I’ve located a picture of your nemesis:

http://ducky.net/~kim/web_pictures/jacob_1year/UncleMike_paino_Jacob.jpg

(He’s the taller one).

]]> by: user http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-934 Tue, 30 May 2006 21:02:23 +0000 http://ridiculousfish.com/blog/archives/2006/05/30/old-age-and-treachery/#comment-934 nice site design. i like these little squares. very nice. i rarely comment. nice site design. i like these little squares. very nice.
i rarely comment.

]]>