[molpro-user] Molpro 2010.1 with CUDA on Macbook Pro NVIDIA card

[Andy May]

Jacek,

There are no settings in Molpro configure for CUDA on Mac, and it has
never been tested.

It looks like you must have hand-crafted this CONFIG file, and if so
there are some non-obvious that are likely to be missing.

I'll send you a modified configure for you to try, and hopefully we can
get it running.

Best wishes,

Andy

On 11/11/10 16:08, Jacek Antoni Klos wrote:
> Dear Molpro
> 
> I compiled the recent 2010.1 molpro on my macbook pro using ifort 11.1.089 and openmpi 1.4.3 linking
> CUDA code with CUDA OS X SDK 3.2.  Molpro compiles, however, when I try to run
> h2o_vqz_gpu.test I get the error at the first df-ks: 
> 
>  Use fine grid record 1800.2 for energy threshold 1.0D-06
>  Orbital guess generated from atomic densities. Full valence occupancy:    7
> 
> 
>  Coulomb fitting
> 
>  Molecular orbital dump at record        2100.2
> 
>  ITERATION   DDIFF            GRAD            ENERGY        2-EL.EN.            DIPOLE MOMENTS         DIIS   CPU(IT)  CPU(TOT)
>  ? Error
>  ? Attempting to terminate unitialized VFI run
>  ? The problem occurs in vfi3_term
> 
>  GLOBAL ERROR fehler on processor   0
> 
> How to run GPU code to avoid this error. In my CONFIG the CUDA part looks like that: 
> 
> CUDACC=/usr/local/cuda/bin/nvcc
> CUDACCVERSION=
> CUDACDEBUG=-g
> CUDACDEFINE=-D
> CUDACFLAGS="-m 64"
> CUDACOPT=
> CUDACOPT0=-O0
> CUDACOPT1=-O1
> CUDACOPT2=-O2
> CUDACOPT3=-O3
> CUDACPROFILE=-p
> 
> Thanks for suggestions
> 
> Jacek Klos
> UMD Chemistry, College Park, MD 20742 USA
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user