On Tue, 5 May 1998, Michael R Zucca wrote:

> > Ditto. I also set CFLAGS to "-O2 -m68040 -pipe" in my environment. This
> > affects the kernel, userland, and all but a few packages. The -pipe option
> > can use a lot of memory, and I sometimes have to tweak the ulimits on my
> > shell to get lib/libc to compile straight through.
> 
> Interesting. How high do you have to set the ulimit to use -pipe?

I have to increase the soft limit form "max locked mem" to compile through
libc. I'm not certain if pipe is even responsible. For compiling kernels
and packages it works just fine at the default of 10M (for my 36M), and
there is, indeed, a noticable speed-up. 

> If you're specifying -m68040 the big speedup is probably comming from gcc
> being smart enough to use the MOVE16 instruction in programs like dt.
> MOVE16 lets you move blocks of memory around. Perfect for blitting
> graphics onto the screen. If you have the disk space you might want to
> try compiling the X server!

Actually, I did compile the X distribution; I had to. The downside of
blazing is your own trail, is that I haven't had very much luck with
distributed binaries. They tend to dump core, and give seg faults. X does
screen scrolls faster than dt, but it didn't compile with any
optimizations!  X's generated makefiles don't seem to respect CFLAGS. 
I'm not sure if there is a simple way to govern the entire build.