|
From: | Aaron Bentley |
Subject: | Re: [Gnu-arch-users] Slow inventories on large source trees |
Date: | Thu, 22 Apr 2004 12:14:53 -0400 |
User-agent: | Mozilla Thunderbird 0.5 (X11/20040309) |
Tom Lord wrote:
I assume that you are passing a regmatch_t array with just one element, right?
Right.
The cut operator can work with just a simple DFA translation of aregexp -- and Rx supports that internally -- but it appears that the test for whether or not that optimization applies within the posixentry points doesn't know about `cut'. It's as if you had asked for a full set of `regmatch_t' values --- Rx is doing lots of backtracking work on every match.
Note also that the it's filename_matches (which is untouched) that's consuming most of the runtime, not filename_matches_cut (which is a variant of the above that returns the cut matched).
I suppose it's possible that the demands of filename_matches_cut leave a cache in disarray, causing filename_matches to work much harder.
Aaron -- Aaron Bentley Director of Technology Panometrics, Inc.
[Prev in Thread] | Current Thread | [Next in Thread] |