Btw, you can set the -fmax-inline-alloc-size flag to a really high value to prevent inline allocation, if you want to test if that's the culprit. Inline allocation shouldn't change the amount being allocated though, just how it's allocated.

T4267  doesn't use byte arrays at all so that should definitely not be related.


On Sun, Apr 6, 2014 at 1:22 PM, Johan Tibell <johan.tibell@gmail.com> wrote:
I don't have a 32-bit machine to test on, but I thought I set the InlineByteArrayAlloc test to only run on 64-bit. Feel free to set the expected value on 32-bit to whatever you're getting.


On Sun, Apr 6, 2014 at 11:49 AM, Simon Peyton Jones <simonpj@microsoft.com> wrote:

Several perf tests have started failing on 32-bit windows.  (See below).  Some of them have very small allocation anyway, so it may be a library issue.  I’m wondering whether it might be a consequence of the new inline byte array stuff.  I don’t know which commit made this start happening, but it’s pretty recent.

Johan, did you test on 32 bit?

Thanks

Simon

 

   perf/should_run                    InlineByteArrayAlloc [stat not good enough] (normal)

   perf/should_run                    T4267 [stat not good enough] (normal)

   perf/should_run                    T5949 [stat not good enough] (normal)

   perf/should_run                    T7619 [stat not good enough] (normal)


_______________________________________________
ghc-devs mailing list
ghc-devs@haskell.org
http://www.haskell.org/mailman/listinfo/ghc-devs