On 19/05/2012 04:42, Finn Thain wrote:
Has anyone benchmarked those binaries (running on the hybrid platform) against native 68020 and Coldfire binaries (running on their respective native platforms)?
No, not yet.My remarks were only intuitions, an in this profiling area, intuitions are often wrong.
It's the slower-clocked 680x0 chips which need that optimisation more than the faster CF chips need it. So the concept of a generic binary is flawed IMHO. (But I'd be happy to be proven wrong by suitable benchmarks.)
I totally agree. Personally, I don't want such hybrid binaries in my own projects. Except in very specific situations such as benchmarking, etc.
Are there more recent Coldfire devices (newer than V4e) that might help with the ABI issues?
There is the V4m core, used in MCF5445x chips. It features a few new instructions for bit manipulation (ISA_C instruction set), but nothing useful for 680x0 compatibility. Furthermore, there is no more FPU, so IMHO those V4m chips have few interest.
The ColdFire is similar to 680x0, but from the beginning it was designed as incompatible with existing 680x0 software. Every "complicated" instruction or addressing mode has been intentionally removed to decrease the processor complexity. Also, the supervisor instructions (exceptions, cache management, MMU...) are very different. So the 680x0 and ColdFire are really different processors (while they share a lot of common instructions), IMHO it is a nonsense to try to be compatible with both, except if the goal is to get the slowest compatible binaries.
-- Vincent Rivière