[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [MiNT] Some mintlib patches
Hello!
10-20% is just your suggestion? I can't believe that some loop unrolling
or not make such a big difference.
It's hard to say, I can say for sure 10-20% in asm code with loop
optimizations is very realistic but I have no idea how efficient gcc is, it
was only assumption..
You really mean 10-20% only between the two compiler options -m68020-60
and -m68060? That's very optimistic I think (I speak about the compiler,
not hand optimized assembler).
well, that's the case. maybe you noticed me and some other guys (Mark for
example) would like to see complete 060 distro for ct60 falcons. I really
don't like to use all software compiled for 68000 with no FPU. And when we
are doing >68000, why not to use the possibility to target 060 instead of
generic 020+ with emulated FPU stuff?
I know this. But please take a look on the realistic side. What speed
advantage have most of the tools under /usr/bin from an FPU? Nothing,
except some special tools like tex for example. They simply don't
use the FPU or only for some very few calculations. The time needed to
type the command into the command line and the time needed to draw the
result on the screen is much, much higher than any performance gain you
get from using the FPU in this case.
Or loading the tool from hard disc into memory. You know the speed
difference between the harddisc and the main memory (also on a Falcon is a
big factor :-)). So the few microseconds you save CPU you never notice
over the milliseconds the harddisc need to deliver the data.
So in result, to see any noticeable effect of a special 68060 optimization
you need tools that run a) a longer time and b) only use the CPU (e.g. no
I/O wait for harddisc, network, serial, ...). I'm sure you don't find much
tools that fulfill these criteria.
Btw. bash *scripts* are so slow because fork() isn't handled very
efficiently inside kernel (without virtual memory). If you really want to
speedup the bash scripts (they use a lot of fork()'s) by a significant
factor implement virtual memory into the kernel ...
Really like to see some concrete numbers about the speed advantage ...
question is not only how to measure but also what to measure... math stuff?
compression?
Ah, you get the point ...
Btw. the gentoo guys that say their system feel much faster, this is just
an illusion (even if you take the CPU power of modern intel/amd cpus into
account).
And, please don't get me wrong. I'm not against special optimized
software. But it must have a noticeable effect. Otherwise you spent lot of
time for nothing. So optimizing the handful of tools there it make sense
is enough.
Regards,
Frank
--
ATARI FALCON 060 // MILAN 060
-----------------------------
http://sparemint.org/
e-Mail: fnaumann@boerde.de