Gregor N. Purdy <[EMAIL PROTECTED]> wrote: > Leo -- > I had posted a program a while ago that generates large > fake programs for testing such things. Did that not help?
I've that program or a successor of it still around somewhere. But unfortunately it doesn't help. The register usage pattern of the simulation is totally different: it creates too much long ranged globals and too much interferences between registers compared to RL programs like Dan's Big Subs. Such a simulation should create much more lines (selectable) almost only temporary register access inside basic blocks and some access to long-ranged globals within reasonable (tweakable) limits i.e. from around 16 to e.g. 64. leo
