On , 22 Mar 2002, Sven Neumann wrote: > yes, it's kinda cool. It didn't like the MMX check in gimp-1.3 and > mumbled something like "Insane instruction". It works fine however > with the latest gimp from CVS when passing --no-mmx on the gimp-1.3 > command-line. I'm not sure if I prefer it over memprof but it's > definitely helpful. I don't know much about memprof, but there's some cutting edge research in valgrind: the author was giving a talk on it at our lab this afternoon. One of his stated aims was to reduce the number of false positives, so that (as far as possible) every warning it issues is a real bug. > We are getting a decent amount of warnings about > "Use of uninitialised CPU condition code" This indicates that program control flow is non-deterministic: ie you've done a "switch" or "if" on a variable which contains junk. The actually loading of junk from memory into the variable may happen earlier than the test, so you'll need to look back at the program logic a bit. > I'm not certain if we can trust valgrind here, but it would > definitely be a good idea to investigate this to make sure. I would trust valgrind :) Of course, it may have found a bug in glib, gtk, or the X libraries. Apparently the author found a number of libc bugs, and gcc code generation errors(!) Austin
