Re: Jump Size Optimization Improved...

edmmapi@gmail.com schrieb:


> [Branch optimization was studied extensively in the 1970s by Tom
> Szymanski and others. The general problem if you allow jumps to jumps
> is NP complete, but it's been well known for decades that you can get
> pretty close either by starting with all the jumps short and expanding
> until the code is valid, or starting with all the jumps long and
> shrinking until there's nothing left to shrink. I don't see what's new
> here. -John]


IMO a jump to jump optimization nowadays may badly interfere with
instruction caching. While jump optimization contributes to the locality
of related code blocks, it may increase the number of related pages.


DoDi
[I agree that adding jumps to jumps would be a poor idea in any but an
extremely memory constrained environment. If you just grow or shrink
existing jumps, the problem is O(N) so I still don't see what's
new. -John]

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