[Simh] RE : Simh execution speed improvement - about 45% improvement.

[Boucher, François]

Hi, Marc
 
I am quite impressed by your very high VUPS results!
I did try to reproduce your results, with a sun sunfire 4100 equipped with two single core opteron CPUs running at 2.2GHz, with solaris 10 and sunstudio 11 to compile simh.  Unfortunately, I am a newbie on solaris.
 
First I did not succeed in compiling simh with sunstudio.  I received a huge number of errors of the compiler that complains there are some illegal blanks.  Is there a switch to pass on to the compiler?
 
I then did compile on the same platform with gcc and gmake, simh 3.7-3 unmodified, and it gave me 12.0 VUPS.
with the modified (the memory access functions declared as inline) vax_cpu.c it gave me 15.0 VUPS.
Both speed results are too low for my target setup to be usable.
 
So, i would very very much appreciate if you can give me a few pointers:
1) what is your machine make, model, CPU type and speed?
2) What should i use as arguments in sunstudio to build simh?  Do I use the provided makefile, or should I try to generate a new one?
 
Thanks in advance,
 
François Boucher ing

________________________________

De: Marc Chametzky [mailto:marc at bluevine.net]
Date: ven. 2008-01-11 16:45
À: Boucher, François
Cc: simh at trailing-edge.com
Objet : Re: [Simh] Simh execution speed improvement - about 45% improvement.



> After that, I did compile simh to include the profiling code (by adding
> -pg on the gcc command line).  After studying the profiling, I saw that
> the functions that reads and writes to RAM and ROM are the ones that are
> executed 50% of the time

> The solution is: by declaring those functions as inline!

On my Solaris 10 x64 (AMD Athlon 64 3200+) system, I was already
building the VAX emulator in 64-bit mode using the Sun Studio 11
compiler with the "-m64" switch, and with that, my CALCULATE_VUPS.COM
procedure reported 38.0.

After making the changes to vax_cpu.c, and otherwise leaving everything
else the same, I'm now getting 50.0, a 32% improvement.

Of course, running DCL scripts to approximate computing performance only
gives a rough estimate, but it certainly appears that it's emulating the
VAX more efficiently now.

--Marc


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.trailing-edge.com/pipermail/simh/attachments/20080117/9cba2dac/attachment-0003.html>