|
From: <mcr...@nc...> - 2006-08-15 15:51:22
|
All- I'm interested in investigating where Valgrind incurs most of it's overhead. My first thought was to use Oprofile and the P4 performance monitoring hardware to collect PC samples, but I'm not sure if this is possible given the nature of Valgrind (i.e., it's not your usual user application). Any suggestions? I haven't tried Oprofile yet as I'm still in the process of learning the tool, but hope to have things figured out by the end of the day. Best, Chad Rosier |
|
From: Nicholas N. <nj...@cs...> - 2006-08-15 16:01:35
|
On Tue, 15 Aug 2006 mcr...@nc... wrote: > All- > I'm interested in investigating where Valgrind incurs most of it's > overhead. My first thought was to use Oprofile and the P4 performance > monitoring hardware to collect PC samples, but I'm not sure if this is > possible given the nature of Valgrind (i.e., it's not your usual user > application). Any suggestions? I haven't tried Oprofile yet as I'm still > in the process of learning the tool, but hope to have things figured out > by the end of the day. We have successfully used Cachegrind to analyse Nulgrind and Memcheck -- that's how we got some of the speed-ups in 3.2.0. See the info about self-hosting in README_DEVELOPERS for how. Note that you'll see big differences for the different tools -- are you interested in profiling Memcheck, or Nulgrind, or Cachegrind...? If you can get Oprofile or any other profiling tool working on Valgrind, that would be really useful. Nick |