negate32

My GCM doesn't take 20% of my compile times last I checked but even so, V8 is gonna be in a special camp because they're comparing against a compiler that doesn't do much code motion. LLVM does a lot of code motion so that "20%" is still being paid by LLVM during their hoisting and local scheduling passes.

One of the big things which makes LLVM very slow is the abundance of passes, I believe I last counted 75 for an unoptimized function? My solution for this is writing more combined passes, due to the SoN design I'm combining a lot of things which are traditionally separate passes. For instance, my equivalent to "SimplifyCFG" "GVNPass", "InstCombine", "EarlyCSEPass" and "JumpThreadingPass" is one combined peephole solver which runs faster than all of these passes separately. This is for two main reasons:

* Less cache churn, I'm doing more work per cacheline loaded in (rather than rescanning the same function over and over again).

* Combining mutually beneficial optimizations can lead to less phase ordering problems and a better solve (this is why SCCP is better than DCE and constant prop separately).

In a few years when TB is mature, I'd wager I'll have maybe 10-20 real passes for the "-O2 competitive" optimizer pipeline because in practice there's no need to have so many passes.

It's a game of knowledge and I have time, if you think I don't know what I'm doing please just call it out so I can politely disagree :P