Hi Brian, I used trunk. Is that the correct place to get the code from?
I am off to a talk now, but when I get back I'll look into replicating Jason's timings. This is all on sage.math, which is an uber fast machine. So I hope we are not just looking at architecture differences between your core2 and that machine here. Bill. 2009/3/3 Cactus <rieman...@googlemail.com>: > - Show quoted text - > > > On Mar 3, 2:07 pm, Bill Hart <goodwillh...@googlemail.com> wrote: >> Here are figures for a 2.66 GHz Xeon (Core 2) under linux: >> >> ***** MPIRbench version 0.1 ***** >> Using default CFLAGS = "-O2 -fomit-frame-pointer -I/home/wbhart/mpir-core2/" >> Using default CC = "gcc" >> Using default LIBS = "-static -L/home/wbhart/mpir-core2/.libs/ -lmpir" >> Using compilation command: gcc -O2 -fomit-frame-pointer >> -I/home/wbhart/mpir-core2/ foo.c -o foo -static >> -L/home/wbhart/mpir-core2/.libs/ -lmpir >> You may want to override CC, CFLAGS, and LIBS >> Using MPIR version: 0.9.0 >> Compiling benchmarks >> Running benchmarks >> Category base >> Program multiply >> multiply 128 128 >> MPIRbench.base.multiply.128,128 result: 36290755 >> multiply 512 512 >> MPIRbench.base.multiply.512,512 result: 6974234 >> multiply 8192 8192 >> MPIRbench.base.multiply.8192,8192 result: 64000 >> multiply 131072 131072 >> MPIRbench.base.multiply.131072,131072 result: 928 >> multiply 2097152 2097152 >> MPIRbench.base.multiply.2097152,2097152 result: 36.4 >> MPIRbench.base.multiply result: 55927 >> Program divide >> divide 8192 32 >> MPIRbench.base.divide.8192,32 result: 603601 >> divide 8192 64 >> MPIRbench.base.divide.8192,64 result: 620812 >> divide 8192 128 >> MPIRbench.base.divide.8192,128 result: 336842 >> divide 8192 4096 >> MPIRbench.base.divide.8192,4096 result: 109596 >> divide 8192 8064 >> MPIRbench.base.divide.8192,8064 result: 1404355 >> divide 131072 8192 >> MPIRbench.base.divide.131072,8192 result: 2292 >> divide 131072 65536 >> MPIRbench.base.divide.131072,65536 result: 1212 >> divide 8388608 4194304 >> MPIRbench.base.divide.8388608,4194304 result: 3.34 >> MPIRbench.base.divide result: 25526 >> MPIRbench.base result: 37784 >> Category app >> Program rsa >> rsa 512 >> MPIRbench.app.rsa.512 result: 15277 >> rsa 1024 >> MPIRbench.app.rsa.1024 result: 3134 >> rsa 2048 >> MPIRbench.app.rsa.2048 result: 476 >> MPIRbench.app.rsa result: 2835.2 >> MPIRbench.app result: 2835.2 >> MPIRbench result: 10350 >> >> Bill. >> >> 2009/3/3 Bill Hart <goodwillh...@googlemail.com>: >> >> > As the Sage project is currently investing a lot into Windows porting, >> > I asked on the sage-devel list if anyone has a suitable Core 2 machine >> > with Windows. >> >> > Bill. >> >> > 2009/3/3 Bill Hart <goodwillh...@googlemail.com>: >> >> I should add that I don't have access to Windows on Core 2 anywhere. >> >> In fact I do not know anyone who has this. >> >> >> Bill. >> >> >> 2009/3/3 Bill Hart <goodwillh...@googlemail.com>: >> >> - Show quoted text - >> > - Show quoted text - >> >>> I finished some work I was doing on my other project FLINT 5 days >> >>> early!! So I will spend a little time on mpir again. I'll work up some >> >>> timings for you on a core 2. >> >> >>> Bill. >> >> >>> 2009/3/3 Cactus <rieman...@googlemail.com>: >> >> >>>> On Mar 3, 12:42 pm, Jeff Gilchrist <jeff.gilchr...@gmail.com> wrote: >> >>>>> On Mon, Mar 2, 2009 at 3:10 PM, Cactus <rieman...@googlemail.com> >> >>>>> wrote: >> >>>>> > I have added batch files in the vc9.build directory - to_gmp.bat and >> >>>>> > to_mpir.bat - for name conversion. >> >> >>>>> Great, thanks, it seems to work well for the build.vc9 directory. >> >>>>> When you eventually convert your support for gmp-ecm and others to >> >>>>> MPIR, you will notice that it looks for "gmp.h" in the root of the >> >>>>> gmp/ directory so the batch file to convert from mpir to gmp misses >> >>>>> changing the root mpir.h to gmp.h. >> >> >>>> Thanks, I'll fix that. >> >> >>>>> > I have also added the new assembler code. >> >> >>>>> I will be testing it soon. >> >> >>>> I am confident in the AMD64 stuff. But I am NOT yet confident in the >> >>>> performance of the Core2 code. >> >> >>>> I need x86_64 results for the Core2 code on both Windows and Linux/gcc/ >> >>>> gas to know where I am with this architecture in performance terms. >> >> >>>> Brian >> >>>> - Show quoted text - >> >>> - Show quoted text - > > Thanks Bill, this is just what I needed to understand whether the > WIndows Core2 stuff is ok - and sadly it isn't :-( > > Scaling up my 2.13 Ghz Core2 result for the overall benchmark to 2.66 > Ghz gives 9150 - 13% slower. But more importantly scaling my multiply > figure to 2.66 GHz gives 50251 - 11% slower. Since my K8 multiply > figues are essentially the same as the Linux figures, it seems that > the code has not moved from the K8 to the Core2 unscathed. This is a > pain since the chaanges are very minor and shouls have moved over > without too much trouble. > > All very frustrating, I'm afraid. Jason gave figures earleir for the > cycle time of the assembler code for each routine. It would help me to > identify problems if you or another kind soul could duplicaate these > figures for Core2. > > Brian > - Show quoted text - > > > --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "mpir-devel" group. To post to this group, send email to mpir-devel@googlegroups.com To unsubscribe from this group, send email to mpir-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/mpir-devel?hl=en -~----------~----~----~----~------~----~------~--~---
