On Thu, 20 May 1999 kim@pvv.ntnu.no wrote:

> > You know I feel this is a lot unfair, espically on Mark.
> 
> I disagree. Mark has done a lot of work, which is good and impressive.
> However, he has also left a lot of work unfinished and in a state where
> others have not been able to complete it.

Let me first state that I'm quite new to this group... but, I've examined
the kernel code because 1.4 doesn't run on my machine. OK, so not
everything is well documented and completly clean at the moment; I would
gladly help Mark with kernel stuff, but I'm quite a newby on virtual
memory; so helping him is quite difficult :-(. Why don't you take a task
on you ? 

> especially connected with the buggy K S-ARM, and how other people
> have complained, problems have been plenty and persistent.
That's a Intel/Digital problem; not a NetBSD's.

> It seems to me that you have not seen how marvelously fast the StrongARM
> can be when programmed properly. 
I agree, but ever tried to program a compiler as smart as you are ? Ever
noticed the difference between optimising for ARM2 v.s. ARM3 ? How would
you explain that to a (generic) compiler...

> My point is that they outperform NetBSD much more than they
> theoretically should. They do not outperform well written code by me,
> compiled by Acorn C/C++ that much.

OK, OK, gcc isn't the best compiler for the ARM, but have you tried to get
a free distribution licence from ARM for their next generation Norcroft
compilers just for the benefits of NetBSD ?

> Oh, but it is. I wrote earlier an example of exactly how integer arithmetic
> code was as bad as it could be considering the sequence of instructions.
> I then resequenced the instructions to get approximately twice the speed.

Have you looked at the peephole optimalisations yourselve ? 

> It is very typical of this maillist that people make thoughtful and
> realistic observations of problems, and how to fix them, and then get
> ignored, even when supplying those fixes. A year later it all happens
> again, with the same problems.

Maybe... but I thought you should report bugs/comments etc. at the NetBSD
bug report list... then they aren't forgotten..

> > As for the "barely working" nature of NetBSD/arm32, I can assure you that
> > a lot of us do NOT feel that way, and those that might have, have worked
> > long and hard to help fix it.
> 
> A lot of people have worked long and hard just messing around with
> uncompilable and wrong source code.

Never noticed... on my NetBSD 1.3.2, even a bloated X works (OK, the VM
roaches are still there, but Hey, they are allready fixed now). And I've
compiled a lot of diffent versions of the kernel to :-) (1, 2, 3, 4, 5,
..)

> A lot of people have worked at fixes wich have been ignored. I have
> personally messed with GCC in NetBSD ARM32, without getting anywhere.
> However, doing the same in Linux is much more straightforward. 

Well, is the Linux GCC that different then?

Reinoud