[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [MiNT] Some mintlib patches
Hello1
Yes, normal instructions like movep or 64bit division aren't used. However,
since m68020-60 means "compile for 68020 and FPU" it uses all 68881
instructions. And here it comes -- m68060 doesn't use emulated floating
point instrucions what m68020-60 does.
Are you sure the full 68881 instruction set is used? I would think only
the reduced instruction set is used (or is this not possible?).
Except the above things I assume it uses much wisely instruction and data
caches, one can't optimize for 030 and 060 in the same time... For example,
on 030 is unrolling always good option (as long as it fits into 256B cache),
on 060 there's branch prediction so it doesn't matter very much. I think
there's good chance gcc guys know about this :) But sure, except FPU stuff,
the speed difference might be 10-20% max but -- in programs like zip or bash
every speedup counts.
10-20% is just your suggestion? I can't believe that some loop unrolling
or not make such a big difference.
include the package). But as I say, for me it's more matter of "good
feeling" everything runs at max. possible speed than any serious reason.
Look at gentoo guys -- everyone of them will tell you their system is "much,
much faster", it's quite popular and even despite the fact there are many
i686 distros where the performance loss is really minimal.
I know this feeling "that everything is compiled specially for my system".
But I'm on the realistic side. If all this optimization make no big
difference I think it's not worth all the required work and time to
establish a system with tons of specially optimized libraries (you not
only need to compile the mintlib; all other libraries need to be
compiled in several special optimized version too; otherwise you have no
big win).
Really like to see some concrete numbers about the speed advantage ...
Regards,
Frank
--
ATARI FALCON 060 // MILAN 060
-----------------------------
http://sparemint.org/
e-Mail: fnaumann@boerde.de