[molpro-user] Compiling molpro2009.1 for AMD

[Kirk Peterson]

I'm using ifort 11.1.076 on my Intel MacBook Pro and these test jobs all work fine with 2009.2.  I've linked to the mkl libraries however.  I'm also not sure I guess
how the ifort for Mac differs from the ifort for Linux. I would guess at least in terms of optimization strategy they should be identical.

-Kirk

On Dec 9, 2009, at 10:56 AM, Andy May wrote:

> Holger (& Jo)
> 
> I have just tested with ifort 11.0.074, currently the most recent
> version we have installed, and all of the testjobs run fine with this
> version.
> 
> Problems with testjobs h2o_cc2 and h2odimer_lccsdt1b have been seen on
> other systems and are already subject to bug reports.
> 
> The hang up failures have never been seen before and this probably means
> that ifort 11.1 either exposes bugs not seen with earlier versions of
> the compiler, or contains a bug itself (most likely overly aggressive
> optimization), or a combination of both.
> 
> I will get a copy of ifort 11.1 and then see if I can find the problem
> for you.
> 
> Best wishes,
> 
> Andy
> 
> On 08/12/09 15:44, Holger Naundorf wrote:
>> Hello,
>> I have built molpro 2009.1 (p26) on an AMD Opteron machine using the
>> Intel 11.1 compilers. This Version was serial and not using any external
>> BLAS/Lapack. The compiling went without any hitch, but I get several
>> tests which fail, namely:
>> 
>> cpp.test        
>> hybrid_dft_ci.test    
>> tl_so_i.test
>> h2f_merge.test  
>> tlbr_mp2_dft_so.test
>> 
>> do not give any error message, simply continue running - right now I
>> have not noted where exactely they stop, but they stop doing anything
>> and just stay at 100% CPU and do not terminate. 
>> 
>> h2o_cc2.test 
>> has some I/O trouble:
>> 
>> h2o_cc2.errout
>> READM:  RECORD=    6100.7  EXTENSION= 0  OFFSET=          0.  ADDRESS=
>> 16130.  LEN=          47  IMPLEMENTATION=df
>> CURRENT WRITTEN FILE LENGTH:    15519.TRYING TO READ UP TO ADDRESS:
>> 16177.
>> ? Error
>> ? Trying to read beyond end of file
>> ? The problem occurs in readm
>> 
>> and h2odimer_lccsdt1b.test has a too large error:
>> 
>> Results for job h2odimer_lccsdt1b.test
>> ERRORS DETECTED in h2odimer_lccsdt1b.test Max error for
>> energy=1.91403464D-06                                                                              
>> PROG              E                E-EOLD
>> LCCSDT-1b0   -152.4901737   -1.68045020D-06
>> LCCSD(T0)    -152.4900905   -3.77241349D-10
>> LCCSDT-1b1   -152.4903618   -1.91403464D-06
>> LCCSD(T1)    -152.4902679   -3.51633389D-09
>> LCCSDT-1b    -152.4903951    8.09356806D-07
>> LCCSD(T)     -152.4903014   -3.28208216D-09
>> 
>> 
>>        LCCSD(T)       LCCSDT-1B        LCCSD(T)       LCCSDT-1B
>> LCCSD(T)       LCCSDT-1B        HF-SCF
>>   -152.49030144   -152.49039507   -152.49026787   -152.49036177
>> -152.49009050   -152.49017366   -152.06213916
>> 
>> Are these bugs other people have already seen and can give some hints
>> what might cause them (compiler versions/switches etc)? 
>> 
>> Regards,
>> Holger Naundorf
>> 
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user