I finally set up some automated tests to see how well pgcc really works, 
especially at different optimization levels.  I have a whole spreadsheet 
full of numbers, but here is a quick summary.  The first percentage is the 
improvement over gcc -O2 -m486 for gzip -9, and the second is the 
improvement for decompression.

Using a pgcc -O2 compiled kernel:

pgcc	-O4		-O3		-O2 	

     5%, 7.85%     3.58%, 12.99%    3.03%, 8.75%  


(I was lazy and used a single pgcc -O2 compiled libgnumalloc, so the -O4 
and -O3 percentages might be a bit low)

I'm not overally impressed with the numbers, and I ran into my first pgcc 
related problem.  After building an entire pgcc -O3 system, I found that 
awk dumped core consistantly.  So I rebuilt using -O2, and everything 
seems fine now (although I am still using a -O3 kernel with seemingly no 
problmes). 

The moral: if you use pgcc, use -O2 and don't expect massive speed 
improvements.  


Jeff Thieleke