But that’s not fast enough either?
Webkit will compile another version using more optimizations, using code from LLVM.
Is the function still executing some long running loop?
It gets flipped over at the beginning of the next run through the loop.
(see the Hot-Loop Transfer section of the original article)
Just from the description, this sounds stupendously complicated. But the performance improvements are apparently quite considerable.
I’m truly impressed by the effort that goes in and the work that’s been done on this.
Read the original post for the full details: