You need to take the extra step, before compiling with Intel compiler, to execute the following:
. /Library/Frameworks/Intel_MKL.framework/Versions/Current/tools/environment/mklvarsem64t.sh
. /opt/intel/Compiler/11.1/076/bin/ifortvars.sh intel64
You should probably add this to your ~/.bash_profile as well, so that when you start a new shell, it will add all of the necessary paths for the Intel compiler.
On Apr 27, 2010, at 5:49 AM, Alan wrote:
> Hi There,
>
> So I downloaded and compiled AT 1.4 for OSX 10.6.3, 64 bits with:
>
> 1) Intel compilers with
>
> export MKL_HOME=/opt/intel/Compiler/11.1/076/Frameworks/mkl
>
> ./configure intel
>
> make
>
> cd ../test; make test
>
>
> 2) With GNU (gfortran from Fink), by modifying gcc/g++ to gcc-4/g++-4 in
> ./configure, then
>
>
> unset MKL_HOME
> make clean; find ../../ -name "*.a" -delete # need to rebuild *.a lib
>
> ./configure -macAccelerate gnu
>
> make
>
> cd ../test; make test
>
> Using 'time', test 1) gave:
>
> 717.37s user 33.30s system 93% cpu 13:21.99 total
>
> and test 2):
>
> 1074.62s user 30.04s system 96% cpu 19:04.75 total
>
> Which would lead to use Intel version, but a detailed look at antechamber
> and sqm tests tells me that the difference are small
> (e.g. ./Run.sustiva 1m59.591s x 2m18.981s)
>
> I use my own patch to set 6 decimals instead of 3
> in src/sqm/qm2_print_charges.f.
>
> Despite of this, tests for 1) showed me only one FAILURE worth noting:
>
> possible FAILURE: check pbdmp.out.dif
> /Users/alan/Programmes/amber11/AmberTools/test/nab
> 11,21c11,21
> < ff: 0 604.48 730.88 0.69 -37.39 0.55 -90.25
> 2.41E+2
> < ff: 1 594.54 720.93 0.69 -37.39 0.55 -90.24
> 2.38E+2
> < ff: 2 510.12 636.36 0.65 -37.37 0.48 -90.00
> 2.15E+2
> < ff: 3 122.80 248.65 0.44 -37.28 -0.09 -88.92
> 6.82E+1
> < ff: 4 54.53 180.02 1.27 -37.15 -0.08 -89.53
> 5.13E+1
> < ff: 5 -26.19 91.51 11.33 -36.89 -0.12 -92.02
> 4.83E+1
> < ff: 6 -22.11 85.54 21.27 -39.05 -0.37 -89.51
> 9.01E+1
> < ff: 7 -56.15 58.09 14.69 -37.81 -0.31 -90.80
> 3.36E+1
> < ff: 8 -77.16 43.24 8.97 -38.41 -0.26 -90.71
> 2.27E+1
> < ff: 9 -79.04 43.81 6.98 -38.77 -0.24 -90.81
> 3.74E+1
> < ff: 10 -78.92 45.76 6.73 -38.95 0.13 -92.59
> 4.56E+1
> ---
>> ff: 0 -139.71 730.88 0.69 -37.39 -833.89 0.
> 2.37E+2
>> ff: 1 -150.35 720.77 0.68 -37.38 -834.42 0.
> 2.34E+2
>> ff: 2 -185.60 687.17 0.67 -37.37 -836.07 0.
> 2.25E+2
>> ff: 3 -281.05 589.90 0.61 -37.31 -834.24 0.
> 1.97E+2
>> ff: 4 -612.59 250.85 0.32 -37.01 -826.74 0.
> 6.75E+1
>> ff: 5 -677.39 190.01 0.86 -36.71 -831.56 0.
> 5.25E+1
>> ff: 6 -724.24 113.28 5.71 -36.01 -807.23 0.
> 4.57E+1
>> ff: 7 -758.84 88.02 10.36 -36.77 -820.45 0.
> 7.35E+1
>> ff: 8 -766.58 75.59 8.53 -36.52 -814.18 0.
> 4.55E+1
>> ff: 9 -786.45 50.15 5.64 -36.85 -805.39 0.
> 3.64E+1
>> ff: 10 -787.24 53.96 5.22 -36.93 -809.50 0.
> 4.61E+1
>
> Note that nonpolar and EPB columns differs hugely while others seems fine
> (except 1st col for Total of course).
>
> Tests for 2) went all fine. Any idea why Intel Compilers would give the
> results above for pbdmp in nab?
>
> Many thanks in advance,
>
> Alan
>
> --
> Alan Wilter S. da Silva, D.Sc. - CCPN Research Associate
> Department of Biochemistry, University of Cambridge.
> 80 Tennis Court Road, Cambridge CB2 1GA, UK.
>>> http://www.bio.cam.ac.uk/~awd28<<
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Apr 27 2010 - 07:00:04 PDT
