Don Stewart wrote:
marlowsd:
Duncan Coutts wrote:
I was about to bump and release the parallel library because it has
changed since the last release, and needs to be released with
ghc-6.10.1.
I was looking at what changed and thus what the new version number
should be. What has changed is that par and pseq have changed from being
invixl 9 to infixr 0. This is theoretically an API change (code that
compiled before may no longer) so the version number of the parallel
package should by bumped from 1.0.0.1 to 1.1.0.0.
I'd like someone else to double check this and apply the patches if
appropriate. If not, some other version number is required, we cannot
use the same as the previous release as there are real changes.
The changes are correct: par and pseq should have been infixr 0, but have
been bogusly infixl 9 since 6.6. This is a bad mistake, I can't believe it
went unnoticed for so long. Various code snippets from papers and suchlike
will have poor parallel performance due to this.
Patches pushed, thanks Duncan.
Clearly we need parallel performance benchmarking with decent coverage
of the parallel lib.
As a (small) step towards this, I recently checked in a few programs into
nofib/parallel that I'm using for benchmarking the parallel RTS. The
programs are mostly microbenchmarks, and some are from the old GUM
benchmark suite, and some hardly parallelise at all (but I don't want to
fix the programs at this stage, because that risks working around bugs in
the RTS, I'd rather use them as test cases first).
Cheers,
Simon
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
