Re: [Lightning] Re: Predicting generated code size

[Paolo Bonzini]

> So does an arch-specific JIT_INS_SCALE scaling macro exposing the
> above heuristic sound reasonable for Lightning? Ony downside being,
> the average instruction size heuristic depends a lot on what
> instructions you actually use (divides vs simple moves, etc).

Exactly.  Then, it's as simple IMHO (and more effective IMNSHO...) to 
use a "low water" counter like 20 or 32, so that out of a large 4096 
byte buffer you will use only the first 4064-4076.  When the JIT 
destination pointer says only 20-32 bytes are left in the buffer, 
allocate a new one and jump to it.

Paolo