Subject: Pentium gcc patch
To: None <port-i386@NetBSD.ORG>
From: Jeff Thieleke <email@example.com>
Date: 04/20/1996 04:52:59
I am interesting to see if anyone else has tried the Pentium Compiler
Group's <http://www-iss.mach.uni-karlsruhe.de/pcg/> patch (pl9) for gcc
Out of curiosity, I applied the patch to the NetBSD 1.1B's gcc, and was
pleasantly surprised that it applied fairly easily, except for the
configure stuff and the slightly rearranged NetBSD version of gcc (fixed
by a few symlinks).
If you believe the PCG, pgcc will generate much faster code - they
estimate 5-35%. My tests with gzip are a bit less optimistic, though.
Using statically linked versions of gzip, one compiled with gcc and -O2
-m486, and the other compiled with pgcc and -O2 -mpentium, I got a speed
improvement of 2.4% for compressing (-9), and 11.4% for decompressing.
(my data set was the port-i386 mailing list archives 0-150, and the timing
was taking from /usr/bin/time).
The real surprise was the file sizes. I got adventurous and compiled a
kernel... Using the same configuration file and -O2, the gcc -m486 kernel
was 947,986 bytes, while the pgcc -mpentium kernel was only 857,874!
Overall, I've seen about a 5%-10% file size reduction, which is inline
with the PCG's results.
I have been using the pgcc kernel for about a day now, and so far so good.
pgcc is pretty flakely at -O6 (doesn't seem improve speed anyway), but it
built itself to stage 3 at -O2, so I'm fairly confident at that