Re: Perf on T10858

Hi, Am Donnerstag, den 20.10.2016, 10:20 -0400 schrieb Ben Gamari:

...

Simon Peyton Jones via ghc-devs writes:

...

I’m getting this on HEAD on my Linux box (64 bit)

cd "./perf/T10858.run" &&  "/5playpen/simonpj/HEAD-2/inplace/test   spaces/ghc-stage2" -c T10858.hs -dcore-lint -dcmm-lint -no-user-package-db -rtsopts -fno-warn-missed-specialisations -fshow-warning-groups -dno-debug-output  -O +RTS -V0 -tT10858.comp.stats --machine-readable -RTS

bytes allocated value is too low:

(If this is because you have improved GHC, please

update the test so that GHC doesn't regress again)

    Expected    T10858(normal) bytes allocated: 241655120 +/-8%

    Lower bound T10858(normal) bytes allocated: 222322710

    Upper bound T10858(normal) bytes allocated: 260987530

    Actual      T10858(normal) bytes allocated: 221938928

    Deviation   T10858(normal) bytes allocated:      -8.2 %

Does anyone else?  It’s good – but why isn’t Harbormaster complaining?

A very good question. It looks like the result is straddling the edge of acceptable so it's conceivable that Harbormaster is (for some reason) just below the failing threshold. We've seen this sort of small non-determinism in allocations in the past, although I don't have a compelling explanation for why.

this is confirmed here: https://perf.haskell.org/ghc/#graph/tests/alloc/T10858 It is close to the lower edge, and not 100% stable. Greetings, Joachim -- Joachim “nomeata” Breitner   mail@joachim-breitner.de • https://www.joachim-breitner.de/   XMPP: nomeata@joachim-breitner.de • OpenPGP-Key: 0xF0FBF51F   Debian Developer: nomeata@debian.org