|any performance profiling tools ?? email@example.com (1997-09-23)|
|Re: any performance profiling tools ?? firstname.lastname@example.org (1997-09-28)|
|Re: any performance profiling tools ?? email@example.com (Glenn Everhart) (1997-09-30)|
|Re: any performance profiling tools ?? firstname.lastname@example.org (1997-10-01)|
|(fwd) Re: any performance profiling tools ?? email@example.com (Philip Koopman) (1997-10-02)|
|From:||Philip Koopman <firstname.lastname@example.org>|
|Date:||2 Oct 1997 14:38:34 -0400|
|Organization:||Carnegie Mellon University, ECE & ICES|
|References:||97-09-084 97-09-119 97-09-126 97-10-010|
>>[Do any of these address the original question about profiling at a low
>>enough level to count cache misses and the like? -John]
I am using Atom in a computer architecture course for exactly that. A
slight modification of the canned Atom instrumentation package gives you
instruction-by-instruction address trace generation (both I addresses
and D addresses), which we feed through a modified version of dinero
that can accept 64-bit addresses for cache simulation. Works great.
AFAIK, in order to get Atom to work you need access to the source code
so you can re-compile with some uncommonly used compiler flags. But
other than that you don't touch the source code at all. I've done this
with some multimedia code and it worked out just fine once I had hacked
the CCFLAGS variable in the make file.
The bad news is that it only runs on Alphas.
Phil Koopman -- email@example.com -- http://www.ece.cmu.edu/koopman
Return to the
Search the comp.compilers archives again.