31 Dec
2012
31 Dec
'12
4:06 p.m.
On 31 Dec 2012, at 17:01, Clem Cole <clemc at ccc.com> wrote:
Interesting indeed. Lots of variables here so something is different. Assume
same HW and not running a VM under the covers etc... which compiler did you use and what
memory model when you built SIMH. Are all the same optimizations turned on? If
different HW, could be a SW emulation of HW instruction optimizations not available on the
other CPU implementation.
For this reason, besides making sure the HW is that I would standardize the compiler if
you can. FYI: The Intel C/C++ compiler for both Linux and OS X are the same
compiler, although sometimes slightly different libraries. Tends to be a better Apples
to Apples test than GCC because of some of the variations introduced into gcc by different
players (plus Apple is moving away from gcc to LLVM due the silliness caused the GPL
restrictions).
[Non-Commerical download can be obtained at:
http://software.intel.com/en-us/non-commercial-software-development These are the
product bits, but has a free license key - nothing is restricted - just be honest when you
sign up].
I'm not sure if we have a free download.about that Mac OS X compiler - I'll
try to ask the product manager when she/he gets back after the new year.
Anyway - with identical HW, a 64 bit model, using the same compiler with same
optimizations for the CPU, I would expect very similar performance with some places were
OS X to be slightly and Linux in a few others.
In the case of SIMH on a laptop system, my expectation would be that OS X and its I/O
system would have a same advantage over Linux on a same HW. Something sounds very messy
if its off by a factor of 2.
Clem
The point is that it's on the same HW, just in a VM running Linux. Go figure, just
thought it was interesting, I don't really have time to properly investigate this
right now :)
It has to be either the OS or the compiler, my guess is the compiler.
Sampsa