> Improving the DRC so it covers more PPC instructions and improving the accuracy of
> FPU rounding _is_ real emulation. Not being MAME doesn't mean noting they do is
> worthwhile.
Yeah, although the concept of a fallback to interpreter mode doesn't exist in MAME; if you start up with the DRC enabled, you run every instruction on the DRC.
The major difference is that even though MAME's had a 2-pass DRC forever we don't actually do any optimizations on either the intermediate code or during the x86 backend (nobody on the team is a compiler specialist). The general consensus is that at some point we'll start translating MAME IR to LLVM IR and use LLVM to do optimization passes and final codegen, which should greatly improve the performance of the DRC and significantly boost the headroom of how fast of a system we can emulate.
|
Index 
Flat 
Re: DOLPHIN boost !
Reply
