On Tue, 2005-03-01 at 23:51 +1100, Benno wrote: >This sounds interesting but I would have thought that it was as much >an operating systems thing as a languages compilers thing.
IO-Lite is a related work for this. >One thing to watch out for is that this is likely to be OS dependant; >how the IO system, buffering and paging is implemented will have a >large affect on results. > >From the systems I've tested on (MacOSX/PPC, Linux2.6.10/ia32), mmap >is a much faster than alternatives, the question is however, how much >impact that would have on the speed of the lexer. > >I think the main thing you would need to show first is that in current >lexers I/O is the dominating factor, and not say, the actual lexing of >the data. I wouldn't have expected I/O costs to be dominant, but >a quick use of gprof would definately show one way or the other. gprof won't show the impact of seek times, nor blocking / async time. oprofile is a better choice AIUI. Also, trying a zero copy IO approach would be useful for comparison. Cheers, Rob -- GPG key available at: <http://www.robertcollins.net/keys.txt>.
signature.asc
Description: This is a digitally signed message part
-- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
