OS3.2.2.1 - OxyPatcher, CyberPatcher, HSMathLibs..?

[cloverskull]

Hey friends, in the past (i.e., in the OS3.1 era) I used such tools as OxyPatcher and HSML. However nowadays I haven't but am curious if it is still useful. I'd rather not patch my OS into overoptimization or cause more problems than I solve, so that's why I'm asking the community vs just experimenting.

I have a CD32 with a TF360, an A4000D with a BFG9060, and an A1200 with a BPPC. So, essentially, three different computers all with 060s and presumably similar degrees of FPU functionality. So...this is of interest for me

Thanks for the feedback!

[Thomas Richter]

Oxypatcher replaces FPU instructions the 68060 and 68040 do not support by function calls into FPU emulations. That itself is (sometimes) useful, for 3D rendering and so on. However, Oxypatcher is out of maintainence, no longer supported, and is known to emulate some instructions incorrectly. I would not recommend using it anymore.

If you want to look into something similar, please check with MuRedox, which provides similar features, though uses the stable Motorola FPU emulation that sits in the MuLib based 68040 and 68060 libraries anyhow - and it is still supported.

Just a word of caution: As these tools perform online patching, programs that keep a checksum over their executables (e.g. for copy protection purposes) will fail to work correctly. There is a larger discussion on this in the MuRedox manual.

[cloverskull]

Thanks, Thomas! I'll read the MuRedox manual. I appreciate the advice here.

Do you have any thoughts on the commercial/separate HSMathLibs? I recall those giving me quite a speedup in OS3.1.

[Thomas Richter]

First of all, speed critical code should probably not use the system math libraries in first place but direct FPU instructions in an FPU-enabled version of the software (or an internal dispatcher which runs an FPU-based computational kernel if an FPU is detected).

That said, from 3.9 onwards, the system math libraries are smart enough to detect the presence of a 68040 or 68060 FPU and then do not go through emulator traps, but directly through the emulation code of the CPU libraries, that is, the fpsp.resource. Thus, you already get a speed advantage for them. For 3.1 and below, you can use "FastIEEE" for the same effect.