previous index next

Now, another trick is to notice that if we have a direct branch that targets a block we've already copied into the cache, rather than context switching back to the interpreter proper, looking up the target, and coming all the way back, we can directly link the two blocks. This gives another dramatic performance improvement, down to about 3X.

Now, looking at indirect branches, we can't link them because their targets vary,

  Copyright © 2004 Derek Bruening