So the max_bytes_used metric is more random than the others. It
measures the peak residency (and it's badly named, too) encountered by
the GC. So if you have a program that has a spiky memory profile, as
many programs do, then small changes in the time at which a GC strikes
can have a large impact on max_bytes_used.
To see whether you really have introduced a problem, I suggest running
GHC with +RTS -h -i0.01 (no need to compile for profiling), before and
after, and compare the results.
Cheers,
Simon
On 11/09/2011 09:14, Manuel M T Chakravarty wrote:
In another run on an (almost) identical tree, the test passed. Probably you
are right and it is just not entirely deterministic.
Manuel
Max Bolingbroke:
Hi,
This is a problem that I saw early on in my development of the
no-pred-ty branch but went away ~1 week ago. I have not been able to
reproduce it since then. (I'm on OS X x86_64 as well, FWIW)
Nonetheless, due to the nondeterministic nature of the perf tests it
may be caused by my patch - I did have to bump 1 test's memory limits
up. If this test fails for anyone else we should probably bump the
T3064 limit as well.
I did try to investigate GHC memory usage by following the
instructions at
http://hackage.haskell.org/trac/ghc/wiki/Debugging/ProfilingGhc but
they didn't actually build a -prof GHC for me. Does anyone know if
those instructions are meant to work, or if are they now out of date?
Max
On 10 September 2011 07:50, Manuel M T Chakravarty<[email protected]> wrote:
I am seeing the following performance regression on OS X x86_64:
cd ./perf/compiler&& $MAKE -s --no-print-directory
T4007</dev/null>T4007.run.stdout 2>T4007.run.stderr
max_bytes_used 5428072 is more than maximum allowed 5000000
*** unexpected failure for T3064(normal)
Is this specific to OS X x86_64?
Manuel
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
