Increasing SpeedUp, reducing overheads, Intel Fortran,OpenMP

General OpenMP discussion

Re: Increasing SpeedUp, reducing overheads, Intel Fortran,Op

Postby ftinetti » Wed Nov 30, 2011 6:14 am

Hi again,

I was trying to compile and run your application but it is almost impossible for me in Linux - gfortran. I do not have ifort and I do not have time to install ifort. So, if you send me the instructions for compile-run the program with Linux-gfortran maybe I would be able to see something (not guaranteed, of course).

My "work":

$ gfortran -c -ffree-line-length-none Funkce.f90
$ gfortran -c -ffree-line-length-none ModelyPodobnosti.f90
ModelyPodobnosti.f90:241.150:

slTestRadkyPrut(RokTest, KrokPredTestRadek, KrokPred, 2) + PrutokyNormData(Pozi
1
Aviso: Extensión: índice de matriz REAL en (1)
ModelyPodobnosti.f90:123.114:

KrokPred, 2) = VyslMrizkaBodPrut (RokPred, KrokPred, 2) + PrutokyNormData(Pozi
1
Aviso: Extensión: índice de matriz REAL en (1)
$ gfortran -c -ffree-line-length-none ZpracovaniDatModul.f90
$ gfortran -c -ffree-line-length-none ZpracovaniVstupuModul.f90
$ gfortran -ffree-line-length-none -fopenmp KalibraceMP.f90
/tmp/cckzoUTo.o: In function `MAIN__':
KalibraceMP.f90:(.text+0xc52): undefined reference to `__zpracovanidatmodul_MOD_zpracovanivstupnichdat'
KalibraceMP.f90:(.text+0xcae): undefined reference to `__zpracovanidatmodul_MOD_zpracovanivstupnichdat'
KalibraceMP.f90:(.text+0xd0a): undefined reference to `__zpracovanidatmodul_MOD_zpracovanivstupnichdat'
KalibraceMP.f90:(.text+0xd50): undefined reference to `__zpracovanidatmodul_MOD_generatorvah'
/tmp/cckzoUTo.o: In function `MAIN__.omp_fn.0':
KalibraceMP.f90:(.text+0x22ab): undefined reference to `__funkce_MOD_kd'
KalibraceMP.f90:(.text+0x22f2): undefined reference to `__funkce_MOD_kda'
KalibraceMP.f90:(.text+0x2a67): undefined reference to `__modelypodobnosti_MOD_modelodchylekpred'
KalibraceMP.f90:(.text+0x2d7f): undefined reference to `__funkce_MOD_kd'
KalibraceMP.f90:(.text+0x2def): undefined reference to `__funkce_MOD_kda'
KalibraceMP.f90:(.text+0x3385): undefined reference to `__modelypodobnosti_MOD_modelodchylekkal'
collect2: ld devolvió el estado de salida 1

I think my non-existent experience with projects including modules is not helping here...
ftinetti
 
Posts: 580
Joined: Wed Feb 10, 2010 2:44 pm

Re: Increasing SpeedUp, reducing overheads, Intel Fortran,Op

Postby ftinetti » Wed Nov 30, 2011 7:19 am

Finally, I was able to compile-run the program, and I did not see anything too strange regarding performance:

KolikVstupuVahovat = 6
---------------------------

$ export OMP_NUM_THREADS=1
$ a.out
Time of operation was 486.85001 seconds

$ export OMP_NUM_THREADS=2
$ a.out
Time of operation was 266.52899 seconds

$ export OMP_NUM_THREADS=4
$ a.out
Time of operation was 140.87000 seconds

$ export OMP_NUM_THREADS=6
$ a.out
Time of operation was 97.436996 seconds

i.e. efficiency starts at 0.91 with 2 CPUs and drops to 0.83 with 6 CPUs.


KolikVstupuVahovat = 8
---------------------------

$ export OMP_NUM_THREADS=1
$ a.out
Time of operation was 824.03198 seconds

$ export OMP_NUM_THREADS=2
$ a.out
Time of operation was 449.42499 seconds

$ export OMP_NUM_THREADS=4
$ a.out
Time of operation was 231.33099 seconds

$ export OMP_NUM_THREADS=6
$ a.out
Time of operation was 158.25700 seconds

i.e. from 0.91 0.86 efficiency.

If 8 cores are used, the runtime is almost identical to that of 6 cores, which is in some way expected, given that the total number of iterations that are distributed among threads are 12.

Basically, I cannot reproduce your performance results, everything is going fine in a dual-quad core Xeon (8 cores in total) with Linux-gfortran.
ftinetti
 
Posts: 580
Joined: Wed Feb 10, 2010 2:44 pm

Re: Increasing SpeedUp, reducing overheads, Intel Fortran,Op

Postby ftinetti » Wed Nov 30, 2011 10:17 am

Uuups! I've seen that Core i5 CPU 760 "Intel® Turbo Boost Technology" according to
http://ark.intel.com/products/48496/Int ... e-2_80-GHz)
and this may explain the strange/poor results in your case. Please take a look at the thread
viewtopic.php?f=3&t=1289
which has some explanations about the problem of performance evaluation & "Intel® Turbo Boost Technology"

HTH.
ftinetti
 
Posts: 580
Joined: Wed Feb 10, 2010 2:44 pm

Previous

Return to Using OpenMP

Who is online

Users browsing this forum: Google [Bot], Yahoo [Bot] and 6 guests