|
From: Boris Z.
Attachments:
massif-output.tar.gz
|
I've been running into this problem for a while with various valgrind versions, including 3.2.3. When using massif to track the memory usage of Mozilla, the HTML and PostScript output don't really match. In particular, the top two bands in the PostScript output in the attached file are given as 6.3% and 5.7% of spacetime respectively. The right numbers based on the PostScript and on what I know of the testcase should be closer to 40% and 20% respectively. Is this a known problem, by any chance? -Boris |
|
From: Nicholas N. <nj...@cs...> - 2007-06-14 15:22:25
|
On Wed, 13 Jun 2007, Boris Zbarsky wrote: > I've been running into this problem for a while with various valgrind > versions, including 3.2.3. When using massif to track the memory usage of > Mozilla, the HTML and PostScript output don't really match. In particular, > the top two bands in the PostScript output in the attached file are given as > 6.3% and 5.7% of spacetime respectively. The right numbers based on the > PostScript and on what I know of the testcase should be closer to 40% and 20% > respectively. > > Is this a known problem, by any chance? I've heard something like that before, there might be something in Bugzilla about it. I'm in the middle of redeveloping Massif extensively, hopefully it'll be finished one day and will fix this kind of thing. Nick |
|
From: Boris Z.
|
Nicholas Nethercote wrote: > I've heard something like that before, there might be something in > Bugzilla about it. Ah, looks like http://bugs.kde.org/show_bug.cgi?id=142706 is exactly what I'm seeing. > I'm in the middle of redeveloping Massif extensively, hopefully it'll be > finished one day and will fix this kind of thing. Is it at a point where it's sort of usable? If so, I'd love to give it a spin! -Boris |
|
From: Nicholas N. <nj...@cs...> - 2007-06-14 18:33:51
|
On Thu, 14 Jun 2007, Boris Zbarsky wrote: >> I'm in the middle of redeveloping Massif extensively, hopefully it'll be >> finished one day and will fix this kind of thing. > > Is it at a point where it's sort of usable? If so, I'd love to give it a > spin! There's a repository branch, use: svn co svn://svn.valgrind.org/valgrind/branches/MASSIF2 It's still experimental, but it should work reasonably well. The output is purely textual now, and you have to use ms_print to print the output (like Cachegrind's cg_annotate script). But using a proper file output will allow multiple result viewers to be created (like KCachegrind). I'd be interested to hear feedback about the changes. Nick |
|
From: Boris Z.
|
Nicholas Nethercote wrote: > There's a repository branch, use: > > svn co svn://svn.valgrind.org/valgrind/branches/MASSIF2 So I tried this, but I don't seem to be able to interact with my program while it's running under this version of massif (because massif is using so much of the CPU?) so I can't quit, so I never get any output. Killing the valgrind process (with kill -HUP or anything else) doesn't seem to work either. Is there some way to deal with this? Thanks, Boris |
|
From: Nicholas N. <nj...@cs...> - 2007-07-01 21:41:23
|
On Sun, 1 Jul 2007, Boris Zbarsky wrote: >> There's a repository branch, use: >> >> svn co svn://svn.valgrind.org/valgrind/branches/MASSIF2 > > So I tried this, but I don't seem to be able to interact with my program while > it's running under this version of massif (because massif is using so much of > the CPU?) so I can't quit, so I never get any output. > > Killing the valgrind process (with kill -HUP or anything else) doesn't seem to > work either. Is there some way to deal with this? That sounds like a bug. Does the old Massif work with your program? Because the internals of the new Massif are very similar. Nick |
|
From: Boris Z.
|
Nicholas Nethercote wrote: > That sounds like a bug. Does the old Massif work with your program? It's pretty laggy, but usually if I hit Ctrl-Q or close the window with my windoe manager the program responds within about 5-10 minutes and quits. With the branch version you pointed me to I waited for about two hours with no effect.... Note that the program in this case is Mozilla loading a page that causes it to allocate on the order of hundreds of megabytes to gigabytes of memory, so I wouldn't expect this to be all that fast, necessarily. As I said, ideally I could send a signal to massif itself to kill the program total up things up to that point... -Boris |