Changing mathlibs is it really usefull?

[matthey]

@Gulliver

Often the fpu instructions are put directly in the code. In that case, having fast math libraries will offer no speed increase. There are plenty of cases where the math libraries are used though and the C= versions arer not very optimized.

[matthey]

@SamuraiCrow

There is no external "hardware" fpu. There is a software emulation of the unavailable 68881/68882 instructions that are unavailable in the 68040 and 68060. These are in the 68040.library and 68060.library respectivly. The cpu traps the unavailable older fpu instructions which is slow. They are still generally faster than the 68881/68882 on newer processors even when trapped. Unfortunatly, going through the cpu traps is slow compared to using only the fpu instructions that are built in. There is Oxypatcher type programs to speed up older programs that use 68881/68882 instructions on the 68040 and 68060. Oxypatcher type programs offer a speed increase if...

You have a 68040 or 68060.
You have programs that use the fpu directly instead of the math libraries.
The programs are NOT compiled for 68040 or 68060.

Faster math libraries offer a speed increase if...

The math libraries are used instead of the fpu directly.