On 10/07/2011 15:01, Karel Gardas wrote:
Hello,
while working on ARM port, I've found bug in WSDeque.c which might
possibly also hurt other architectures -- depending on how weak is their
memory model (i.e. how much load/store reordering it permits).
I'm certainly not sure if it affects x86, but I'm not sure if PowerPC is
safe here, hence I'm writing this.
For fix preview and explanation see:
https://github.com/kgardas/ghc/commit/ac7c59eba9806fc8216f44342c12b784e6680483
Yes - and I'm pretty certain you'll run into other issues like this if
ARM doesn't provide total store ordering. Well done for tracking this
down anyway (was it the testwsdeque.c test case that failed?).
Will you submit this patch, or a pull request?
Have you tried running some heavy -N2 tests? e.g. try the benchmarks in
nofib/parallel.
Cheers,
Simon
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
