Skip to main content.

Archives

This is the archive for 20 July 2011

Wednesday, July 20, 2011

Well, the A register still exists; but now it's completely ruled out from the inner (new) function management, and it doesn't play any role in the call frame.

I took the occasion to get rid of other minor details in the generation of the call frame and in return routine.

As a result, I achieved another 5-10% performance increase in the non-optimized recursive fib test with respect to the old engine. Now it's consistently 20%-33% faster than ever before. And the VM model is now even simpler and more linear.