On Tue, 4 Nov 2008, Bob Friesenhahn wrote: > > I shall run a gruelling test to prove/disprove your hypothesis while I > am off voting for the leader of our great American nation. We shall > see if your hypothesis is correct.
Ok, now I am back and my mini-benchmark is completed. Out of 78 tests completed these are the final statistics: GCC: 39 Sun C: 30 Tie: 9 These tests were executed on a 4-core Opteron system (Sun Ultra-40 M2) with all possible optimizations enabled. The application is GraphicsMagick compiled as 32-bit with OpenMP enabled so that all 4 cores are active. The test runs the application 20 times to assure accuracy. In the scoring, if GCC was ahead by a ratio of 1.01 then GCC was declared the winner. To be fair to Sun C, it seems that several algorithms where it fell behind are using a particular underlying math-intensive summation algorithm which perhaps can be reformulated to be more pleasing to Sun C without annoying GCC. Sun C seems to prefer FORTRAN style array syntax whereas GCC does a bit better with raw pointer access. Compilers/flags used: gcc (GCC) 4.2.3 -fPIC -march=opteron -O -Wall -Winline -W -Wextra -Wno-unknown-pragmas -D_REENTRANT -pthreads cc: Sun C 5.9 SunOS_i386 Patch 124868-06 2008/06/24 -v -errtags=yes -erroff=E_END_OF_LOOP_CODE_NOT_REACHED,E_STATEMENT_NOT_REACHED -xbuiltin=%all -xalias_level=std -m32 -xO5 -xtarget=native' Bob ====================================== Bob Friesenhahn bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/ GraphicsMagick Maintainer, http://www.GraphicsMagick.org/
