Re: [rvm-research] Any way commonly used to collect traces efficiently?

[Filip P. <pi...@pu...>]

Here's an easy solution: insert a call to a tracing function.  Just 
write a static Java method that does what you want (i.e. print stuff, 
appends to a buffer, whatever).  You may even have multiple such methods 
depending on what you want to put into the trace.  Then insert a call to 
that function.

There is no way to do this without significantly perturbing performance 
unless you are just tracing some very-rarely-used code.  Thus, you 
should just make this work in the baseline compiler, which should be 
straight forward.

-Filip


On 3/28/2010 6:15 PM, Kelvin Tian wrote:
> Hi,
>
>      Sorry to bother again, but I want to collect some traces using 
> JikesRVM. The traces are basically some strings generated while 
> programs are running. One way to do that is to instrument client 
> programs in a way that they put the strings in a global buffer, but 
> this is error prone and requires much effort to insert the correct 
> assembly code. The other way I want to try is that if I can create an 
> string appending instruction (similar to instrumented counter, but 
> operands are string, and operator is string append) which can append 
> the trace strings whenever a trace is needed to collect. I didn't find 
> a way to create such an instruction, as there seems no such operators. 
> A third way is that I can use yielding points, but unfortunately they 
> are only executed periodically and I need to collect accurate traces.
>    I suppose collecting traces is common in profiling, so I want to 
> hear some suggestions for similar experiences. Are there any way 
> commonly used to collect traces efficiently? Thanks a lot for any 
> suggestions!
>
>
> --
> Best regards,
> Kelvin
>
>
> ------------------------------------------------------------------------------
> Download Intel® Parallel Studio Eval
> Try the new software tools for yourself. Speed compiling, find bugs
> proactively, and fine-tune applications for parallel performance.
> See why Intel Parallel Studio got high marks during beta.
> http://p.sf.net/sfu/intel-sw-dev
>
>
> _______________________________________________
> Jikesrvm-researchers mailing list
> Jik...@li...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>