|
From: Dirk R. <dre...@ia...> - 2005-11-21 17:13:22
|
On Friday 18 November 2005 21:23, Nicholas Nethercote wrote: > Hi, > > So I'd like to know why people who use Addrcheck do so in preference > to Memcheck. I can think of four possible reasons: > > 1. Addrcheck is faster > 2. Addrcheck uses less memory > 3. Addrcheck does not issue undefined value error messages > 4. Something else I haven't thought of I mainly use Addrcheck because of a combination of 1 and 3. 1 should be obvious, 3 because there are a bunch of warnings that cannot be suppressed for Memcheck when using the nVidia 3D graphics driver. With the latest version that has gotten better, as they only appear during the startup of my program, so right now I could live without Addrcheck. However, that might change as my program and the nVidia drivers develop, so it would be nice if there was a way to do more general suppressions for Memcheck, which would allow me to ignore Addrcheck altogether. Just my .02$ Dirk |
|
From: Nicholas N. <nj...@cs...> - 2005-11-21 17:19:55
|
On Mon, 21 Nov 2005, Dirk Reiners wrote: > I mainly use Addrcheck because of a combination of 1 and 3. 1 should be > obvious, 3 because there are a bunch of warnings that cannot be > suppressed for Memcheck when using the nVidia 3D graphics driver. With > the latest version that has gotten better, as they only appear during > the startup of my program, so right now I could live without Addrcheck. By "latest version" do you mean the latest version of the drivers, or of Valgrind? > However, that might change as my program and the nVidia drivers develop, > so it would be nice if there was a way to do more general suppressions > for Memcheck, which would allow me to ignore Addrcheck altogether. Thanks for the info. Can you remind me why the nVidia driver errors are hard to suppress? Is it because they are lacking debug info and/or symbols? Nick |
|
From: Dirk R. <dre...@ia...> - 2005-11-24 17:49:31
|
Hi Nicholas, On Mon, 2005-11-21 at 11:19 -0600, Nicholas Nethercote wrote: > On Mon, 21 Nov 2005, Dirk Reiners wrote: > > > I mainly use Addrcheck because of a combination of 1 and 3. 1 should be > > obvious, 3 because there are a bunch of warnings that cannot be > > suppressed for Memcheck when using the nVidia 3D graphics driver. With > > the latest version that has gotten better, as they only appear during > > the startup of my program, so right now I could live without Addrcheck. > > By "latest version" do you mean the latest version of the drivers, or of > Valgrind? Both. I got a new box, and reinstalled everything, and things started to look better. > > However, that might change as my program and the nVidia drivers develop, > > so it would be nice if there was a way to do more general suppressions > > for Memcheck, which would allow me to ignore Addrcheck altogether. > > Thanks for the info. Can you remind me why the nVidia driver errors are > hard to suppress? Is it because they are lacking debug info and/or > symbols? I just realized that I hadn't updated lately, and it looks like I can actually suppress everything now. I'm running valgrind 3.0.1 (VEX 1367) on an AMD64 system with NVidia 76.76 drivers. I put the suppressions up at http://www.vrac.iastate.edu/~dreiners/NVidia7676_AMD64_VG301.supp just in case somebody's interested. Most of them are ioctl param problems. I've tried to stay conservative and make the suppressions as specific as I can, which makes the list a little long. Word of warning though: as I extend my program, I keep finding new places that give warnings, so the list is certainly not complete. Now the only reason for running addrcheck is speed, which as I see it is becoming less and less of an issue. So IMHO don't bother reviving it. Thanks Dirk |