Path: utzoo!utgpu!news-server.csri.toronto.edu!csri.toronto.edu!corkum Newsgroups: comp.sys.sgi From: corkum@csri.toronto.edu (Brent Thomas Corkum) Subject: Still looking for the SPARK in my SPARC!!! Message-ID: <1991Feb22.113611.14143@jarvis.csri.toronto.edu> Organization: Civil Engineering, University of Toronto Date: 22 Feb 91 16:36:11 GMT Lines: 35 This is a followup from my previous posting. First of all I'd like to thank all you people who responded with suggestions. Most people recognized that I didn't use the optimizer, so I tried it on both the Iris and the Sun. I was very pleasantly surprised on the Iris as it sped up my code by 25%, as for the Sun, no SPARK yet. The following are a set of results for the C test benchmark found on boulder.civ.toronto.edu (128.100.14.11) in pub/COMPUTE/compute.tar. The Sun Compiler was the default bsd compiler that came with the SPARCstation 2. machine compile line cpu time (sec) IRIS 4D/25 16MB cc -O2 -float *.c -lm -o compute 10.5 SUN Sparc2 16MB cc *.c -lm -o compute 73.0 SUN Sparc2 16MB cc -O2 *.c -lm -o compute 49.2 SUN Sparc2 16MB cc -O2 -fsingle *.c -lm -o compute 34.7 SUN Sparc2 16MB cc -O3 -fsingle *.c -lm -o compute 37.0 SUN Sparc2 16MB cc -O4 -fsingle *.c -lm -o compute 37.0 SUN Sparc2 16MB cc -O2 -fsingle -dalign -Bstatic *.c -lm -o compute 34.8 As you can see, I can't get any better than a third of the IRIS speed. I also tried the inline library in /usr/lib/libm.il, with the best time being 32.3 sec for -O2 -fsingle directives. So I'm still looking for that SPARK. If anyone can verify my results or try the C1.0 or C1.1 or gnu compiler on my test program it would be much appreciated. For those who are interested you can get my code via anonymous ftp from the above location, there's a README file in a compute.tar file (extract using tar xvof compute.tar). Thanks, Brent corkum@boulder.civ.toronto.edu Brought to you by Super Global Mega Corp .com