From: Axel Kohlmeyer (akohlmey_at_cmm.chem.upenn.edu)
Date: Mon Dec 08 2008 - 03:36:50 CST
On 12/8/08, Christopher Hartshorn <cmhartshorn_at_wsu.edu> wrote:
> Hello all. I have successfully compiled two seperate 64bit versions
> of NAMD (actually one is a pseudo-64bit) for Intel Xeon Macs. I did
> two builds with the following pieces:
> Pseudo 64 bit=
> -Latest source dump of charm++ compiled twice (once as 32bit and the
> other as 64 bit)
> -Bleeding edge NAMD
> -Precompiled 32bit Intel Mac versions of TCL and FFTW (8.4 and 2.1,
> -I did a "make" on the NAMD using the 32 pieces as a 32bit compile
> using the CFLAGS=-O3 -arch x86 -fexpensive-optimizations -fomit-frame-
> pointer -ffast-math -fno-strict-aliasing
> -Then I took the 64bit charmrun and replaced the 32bit charmrun that
> was in the folder with namd2, etc.
this is nonsense. charmrun is a shell script and thus
64-bit or 32-bin compile doesn't make a difference.
> Full 64 bit=
> -Latest source dump of charm++ compiled only as 64bit Intel
> -Bleeding edge NAMD
> -I compiled 64 bit TCL8.5 using ./configure --disable-shared --
> enable-64bit --disable-corefoundation CFLAGS="-O3 -D_NO_STRING_INLINES
> -I compiled 64 bit FFTW2.1 using ./configure --enable-float --enable-
> type-prefix CFLAGS="-O3 -arch x86_64"
> -I did a "make" on the NAMD using the 64 pieces as a 64bit compile
> using the CFLAGS=-O3 -arch x86_64 -fexpensive-optimizations -fomit-
> frame-pointer -ffast-math -fno-strict-aliasing
> After running the two to check performance, I noticed the following on
> the same initial simulation and same computer over 300 total steps each:
> Original precompiled NAMD download for Mac Intel 32bit---1.05/step
> Pseudo 64 bit (half 32bit/half64bit)---0.73/step
> Full 64 bit--------0.77/step
> The pseudo 64bit is faster then my fully 64bit. Which is certainly
> possible, but I think it is probably something else. The only
yes. this can easily be the impact of OS jitter. 300 MD steps is not
much and the reliability of timings depend a lot on what else is going on
on the specific machine or nodes, as well as the system size and many
other little details. have a look at papers describing benchmarks. it is
almost a science in itself...
> difference between the builds are the following:
> Pseudo uses precompiled binaries with CFLAGS that are potentially
> different then mine because I have know idea which were used when
> whomever made them, made them.
> Full is using TCL8.5 not 8.4.
> My questions are the following:
> 1) Is there any reason why TCL8.5 would be a problem versus 8.4 that
> anyone knows of?
> 2) Can anyone tell me what the configure and/or the CFLAGS were on
> those precompiled binaries for Mac Intel of FFTW and TCL8.4? I would
> like to repeat with the same flags (with the exception of -arch
> x86_64) in order to see if that is the main problem.
please note that the "speed" and type of compilation of TCL has no impact
unless you actually use an input that uses the TCL interpreter.
> 3) Can anyone tell me if there is any flag above that would cause any
> problems for either the NAMD or the other makes?
> Finally, I have to note that when running the make on the full 64bit
> version during the compiling of most files that had to do with TCL I
> got the following warning (usually many times over):
> src/<X>tcl.C Warning: deprecated conversion from string constant to
> char* where <X> was a different name each time (eg. X=globalmaster).
that is a "feature" of g++. the GNU developers have some strange
habit of converting people to the one true style of programming.
for programs that are supposed to be highly portable, that is not
possible. just ignore it.
> Thanks for all the input thus far and for any more to come.
> Chris Hartshorn
-- ======================================================================= Axel Kohlmeyer akohlmey_at_cmm.chem.upenn.edu http://www.cmm.upenn.edu Center for Molecular Modeling -- University of Pennsylvania Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323 tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425 ======================================================================= If you make something idiot-proof, the universe creates a better idiot.
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:50:13 CST