Monday, December 8, 2008, 5:35:19 PM, you wrote: > Very interesting. This corresponds to Luiz's finding that MinGW (with > -O2) is twice as fast as VS2008 without -O2.
> So I guess that the conclusion is to explicitly add the -O2 to the > VS2008 build. And then MinGW would be about the same speed as VS2008. > [BTW, I have a special attachment to the dfl001 problem - thanks for > including that in the benchmark!] > -----Original Message----- > From: Andrew Makhorin [mailto:[EMAIL PROTECTED] > Sent: Monday, December 08, 2008 5:26 AM > To: glpk xypron > Cc: Meketon, Marc; [EMAIL PROTECTED]; [email protected] > Subject: Building glpk with vs2008 > Monday, December 8, 2008, 9:57:42 AM, you wrote: >> Hello Mark, >>> May I recommend to Andrew or Xypron that at least -O2 is set on for > the >>> VS2008 makefile? >> thank you for your recommendations >> Flag /O2 is described here >> http://msdn.microsoft.com/en-us/library/8f8h5cxt.aspx >> as a default setting for release builds. >> Flag /GL is described here >> http://msdn.microsoft.com/en-us/library/0zza0de8.aspx >> Some caveats do exist: >> "The format of files produced with /GL in the current version may not >> be readable by subsequent versions of Visual C++." >> I will test with these two flags. >> Best regards >> Xypron > I did not find the difference between /O2 and /Ox (please see some > benchmarks below). However, the code is about twice faster than without > any optimization flags. > (It is interesting to note that the number of simplex iterations is > the same in all cases.) > Andrew Makhorin > ------------------------------------------------------------------------ > Solver: glpsol 4.34 > Computer: Intel Pentium 4 CPU 3GHz, 2GB of RAM > Platform: Windows XP 5.1 Build 2600 Service Pack 3 > Compiler: Microsoft 32-bit C/C++ Optimizing Compiler > Version 15.00.30729.01 for 80x86 > Build: w32/Build_GLPK_with_VC9.bat > Test set: Netlib LP collection <ftp://ftp.netlib.org/lp/data/> > Cmd line: glpsol --mps foobar.mps > ------------------------------------------------------------------------ > CFLAGS = /I. /DHAVE_CONFIG_H /nologo /W3 > Problem Iters Time,s Mem,Mb > d2q06c 5398 11.0 5.8 > dfl001 38894 181.6 11.0 > greenbeb 2595 3.7 5.3 > pilot87 7952 53.8 12.2 > ------------------------------------------------------------------------ > CFLAGS = /I. /DHAVE_CONFIG_H /nologo /W3 /O2 > Problem Iters Time,s Mem,Mb > d2q06c 5398 6.2 5.8 > dfl001 38894 109.6 11.0 > greenbeb 2595 2.0 5.3 > pilot87 7952 31.2 12.2 > ------------------------------------------------------------------------ > CFLAGS = /I. /DHAVE_CONFIG_H /nologo /W3 /Ox > Problem Iters Time,s Mem,Mb > d2q06c 5398 6.2 5.8 > dfl001 38894 109.8 11.0 > greenbeb 2595 2.0 5.3 > pilot87 7952 30.9 12.2 > ------------------------------------------------------------------------ Just for comparison (please see below). (I used a 31-day evaluation version of IC++ 11.0 and the makefile for VS2008, in which I simply replaced cl.exe by icl.exe.) Andrew Makhorin ------------------------------------------------------------------------ Solver: glpsol 4.34 Computer: Intel Pentium 4 CPU 3GHz, 2GB of RAM Platform: Windows XP 5.1 Build 2600 Service Pack 3 Compiler: Intel(R) C++ Compiler Professional for applications running on IA-32, Version 11.0 Build 20081105 Package ID: w_cproc_p_11.0.066 Build: w32/Build_GLPK_with_VC9.bat Test set: Netlib LP collection <ftp://ftp.netlib.org/lp/data/> Cmd line: glpsol --mps foobar.mps ------------------------------------------------------------------------ CFLAGS = /I. /DHAVE_CONFIG_H /nologo /W3 /Ox Problem Iters Time,s Mem,Mb d2q06c 5334 5.5 6.2 dfl001 38762 99.7 11.0 greenbeb 2620 1.8 5.3 pilot87 7780 24.1 12.2 ------------------------------------------------------------------------ _______________________________________________ Help-glpk mailing list [email protected] http://lists.gnu.org/mailman/listinfo/help-glpk
