Author: Dann Corbit
Date: 20:35:08 02/23/01
Go up one level in this thread
On February 23, 2001 at 21:30:44, James Swafford wrote: >On February 23, 2001 at 13:46:07, James Swafford wrote: > >Dann emailed me the following results earlier this afternoon: > >"THIS IS THE INTEL C++ 5.0 COMPILER RESULTS:" >mean node rate for benchmark -> 501193 n/s > >"THIS IS THE MS VC++ 6.0 COMPILER RESULTS:" >mean node rate for benchmark -> 378883 n/s > >"THIS IS THE GCC gcc-2.95.2-6 COMPILER RESULTS:" >mean node rate for benchmark -> 322735 n/s > >"THIS IS THE GCC gcc-2.95.2-6 COMPILER RESULTS (Swafford settings):" >mean node rate for benchmark -> 362437 n/s > >So for him, the msvc build came out faster than the gcc, >but not by a very large margin at all: 378883/362437 = 4.5% faster. >Most claimed it would be as much as 30-40% faster. > >I'm curious about two things: >1. The superior (far superior) compilation of the Intel compiler. Depends on the application, but it is usually faster. An exception is Arasan, which is actually faster on the Microsoft compiler. You can also use the Intel compiler to create faster Microsoft builds, by creation of a link order file. >2. Variability in results. In my test, the gcc build was > (slightly) faster, in Dann's, the msvc build (4.5% faster) I would suggest also that the tests should be run multiple times. I saw some variation in single runs of the tests. >Dann stated he thought he could squeeze more speed out of the >thing. I have. One simple gag that has eeked out a tiny bit more is the chummy hack: #include "file1.c" #include "file2.c" ... #include "filen.c" I got about 3-4% better. There are lots more opportunities for better speed.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.