|
From: Rick G. <rcg...@ve...> - 2013-03-25 17:42:21
|
Hi - I'm interested in adding support for AMD's LWP (LightWeight
Profiling) extension, but generally not sure of general conventions used
here.
Some questions I have:
- is there a FAQ describing general policies re: enhancements
- Are monolithic diffs preferred over smaller sub-setted diffs?
The context/purpose of adding this is to enable per-thread/self
monitoring code, such as might be used by a managed runtime, etc.
without having to make system calls to/from the OS. One of the really
interesting features of the LWP extension is that it supports value
profiling.
I've got the XSAVE/XRSTOR support (in the form of a patch to linux-3.1)
to support LWP, and a low-level library to make using it a bit easier.
Unfortunately, my sample/demonstrator executable is faulting; I am of
the belief that it is due to a non-obvious memory leak/access somewhere
in my code. Which gets me back to needing to add support in valgrind
for the LWP instructions....
Regards,
Richard Gorton
|
|
From: Julian S. <js...@ac...> - 2013-03-29 09:12:21
|
> Some questions I have: > - is there a FAQ describing general policies re: enhancements Not really. > - Are monolithic diffs preferred over smaller sub-setted diffs? I think it depends a lot on the problem. Is there any documentation of LWP ? Do you have a working patch for V ? How big is the patch? Generally it's best to create a bug report and attach patches to that so that they can be reviewed. It can take a while but usually does get done. J |