Re: Detailed algorithms on inline optimization

>> You can inline recursive calls if you limit the depth.
>> [True. Does anyone do that, inline the first few times though and
>> then punt to the normal routine? -John]
> You can turn the recursion into a loop (if possible) and then inline
> the loop code.
> Offhand I can't point to any examples, but I'd look at Scheme and ML
> family compilers. Scheme and ML require tail recursion - at least
> simple tail recursion - to be implemented as a loop, so I'd guess that
> optimizing compilers for these languages would be able to demonstrate
> some inlining of loop-transformed recursion.


Yes, for such functional languages, all loops appear as recursive calls,
so any "loop optimisation" will apply to recursive functions as well.


In this context, John's question:


> [True. Does anyone do that, inline the first few times though and
> then punt to the normal routine? -John]


is the same as asking "does any do loop peeling?".
I can reply for some versions of SML/NJ: it tries not to do
loop-peeling, tho it can be tricked into doing it if the recursion is
not syntactically obvious; and it usually tries to inline the whole
loops. Last I heard it doesn't do loop-unrolling either.




                Stefan

Return to the comp.compilers page.
Search the comp.compilers archives again.