[m-rev.] diff/for review: last call modulo construction

Peter Ross pro at missioncriticalit.com
Mon Sep 12 15:28:01 AEST 2005


On Mon, Sep 12, 2005 at 02:50:08PM +1000, Zoltan Somogyi wrote:
> We have been missing this optimization for a long time. Now we ought to
> get significantly better results on some standard benchmarks such as
> naive reverse. However, the impact on the compiler performance in general
> is not as good. My preliminary benchmarking shows that the new optimization
> gives a 2% speedup at -O2, but slowdowns at higher optimization levels.
> This must be due to an unfavourable interaction with some other optimization.
> I will look into this later.
> 
When dmo and I were benchmarking this optimization on the alias branch,
we found that it was actually a slow-down for benchmarks, but a speed up
in the compiler.

The problem was simple, the variable holding the address of the
construction cell wasn't being zero'd so the garbage collector was
thinking it was a pointer and tracing it, leading to significantly
longer GC_mallocs.

Maybe you're having the same problem.
--------------------------------------------------------------------------
mercury-reviews mailing list
post:  mercury-reviews at cs.mu.oz.au
administrative address: owner-mercury-reviews at cs.mu.oz.au
unsubscribe: Address: mercury-reviews-request at cs.mu.oz.au Message: unsubscribe
subscribe:   Address: mercury-reviews-request at cs.mu.oz.au Message: subscribe
--------------------------------------------------------------------------



More information about the reviews mailing list