Re: operf jit problem. No anon samples?

On 10/16/2014 06:46 PM, Maurice Marks wrote:
> Sorry. AFAIK It worked at one time:
Hi, Maurice,
In future, please don't remove oprofile-list from cc, since we often rely on the community at large to help answer questions.
> 
>  http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20090629/080117.html
I talked to a colleague of mine here at IBM who's involved with LLVM (although, not specifically with this area).  He said that this oprofile support was for the "old" JIT, and the new  MCJIT does not yet have the oprofile support, as far as he knew. He suggests you post a message to the LLVM mailing list -- http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev.
> 
>  and more recently:
> 
> http://rubini.us/2013/03/20/profiling-jitted-ruby-code-with-oprofile/
> 
> Regarding the anonymous mapping. My understanding is that malloc will grab memory from the heap if its relatively small, but do an mmap call to get larger chunks from the kernel  which are marked anonymous. I guess I didn't understand the difference from oprofile's point of view.
> If it regarded the heap as part of lli's address space then it should have attributed its pc samples to lli not ld.so, shouldn't it?
The heap address returned from malloc *is* part of lli's address space.  And if that truly is where LLVM is putting compiled code, then, as I said earlier, a deep dive debugging session would be needed to figure out why samples are being attributed to ld.so.  Go ahead and knock yourself out, but you might want to check in with LLVM developers first to see if it would be worthwhile doing -- i.e., if newer JIT really does not have support for oprofile yet as my colleague stated.

-Maynard
> If it was outside of lli's space, in anonymous memory then it could be "anon".
> 
> I might have to make up a much larger jit example to test this theory.
> 
> 
> 


------------------------------------------------------------------------------
Comprehensive Server Monitoring with Site24x7.
Monitor 10 servers for $9/Month.
Get alerted through email, SMS, voice calls or mobile push notifications.
Take corrective actions from your mobile device.
http://p.sf.net/sfu/Zoho