On Feb 8, 2008 2:40 PM, Nikodemus Siivola <nikodemus@...> wrote:
> 2. Even though originally (maybe in CMUCL antiquity, but still)
> constraint propagation was an optional pass, we now depend on it
> happening in a multitude of ways
Have you looked at older CMUCL releases to see if they rely on this?
It would be an interesting archaeological exercise to see if SBCL's
compiler rewrites/bugfixes/whatever introduced this somehow.
> 4. There are a bunch of IR1 optimizations that cannot be currently
> elided: IR1-OPTIMZE-IF, IR1-OPTIMIZE-COMBINAION, and IR1-OPTIMIZE-CAST
> are required.
Could you give an example of the sort of code that triggers bugs here?