- Marquer comme nouveau
- Marquer
- S'abonner
- Sourdine
- S'abonner au fil RSS
- Surligner
- Imprimer
- Signaler un contenu inapproprié
Hello to everyone,
I have a large program that uses some MKL functions to solve linear systems and eigenproblems. I'm using MKL parallel (/Qmkl:parallel). In almost all the functions calls I can see more that one thread being used. However, GGEV is only using one thread. I'm using the FORTRAN 95 interface of this function.
I'm running FORTRAN Composer XE Update 10 under VS2010 and MKL 10.3 Update 10.
Am I missing something? Thank you in advance for all the help.
Gratefully,
Pedro Santos
I have a large program that uses some MKL functions to solve linear systems and eigenproblems. I'm using MKL parallel (/Qmkl:parallel). In almost all the functions calls I can see more that one thread being used. However, GGEV is only using one thread. I'm using the FORTRAN 95 interface of this function.
I'm running FORTRAN Composer XE Update 10 under VS2010 and MKL 10.3 Update 10.
Am I missing something? Thank you in advance for all the help.
Gratefully,
Pedro Santos
Lien copié
5 Réponses
- Marquer comme nouveau
- Marquer
- S'abonner
- Sourdine
- S'abonner au fil RSS
- Surligner
- Imprimer
- Signaler un contenu inapproprié
Pedro, what are theorder of the input matrices you are trying to solve?
- Marquer comme nouveau
- Marquer
- S'abonner
- Sourdine
- S'abonner au fil RSS
- Surligner
- Imprimer
- Signaler un contenu inapproprié
Pedro,
Which precision of GGEV have being used: z,c,d or s?
Andhow did youverifyonly one thread being used?
Which precision of GGEV have being used: z,c,d or s?
Andhow did youverifyonly one thread being used?
- Marquer comme nouveau
- Marquer
- S'abonner
- Sourdine
- S'abonner au fil RSS
- Surligner
- Imprimer
- Signaler un contenu inapproprié
Hello,
I'm using the double precision real version. At first I've notice this issue in the task manager, simply by observing CPU usage of my program. After discovered this issue, I've made a concurrency test in VTune Amplifier XE to confirm. And I confirmed that only the Main thread is working. There are other three threads (OMP worker thread) that are running, but they are only using residual CPU time. In attachment, I placed a screenshot of the concurrency test. The order of the matrix in this particular test is 500 (usual values are between 160 and 720).
Thank you very much for your help.
Gratefully,
Pedro Santos
I'm using the double precision real version. At first I've notice this issue in the task manager, simply by observing CPU usage of my program. After discovered this issue, I've made a concurrency test in VTune Amplifier XE to confirm. And I confirmed that only the Main thread is working. There are other three threads (OMP worker thread) that are running, but they are only using residual CPU time. In attachment, I placed a screenshot of the concurrency test. The order of the matrix in this particular test is 500 (usual values are between 160 and 720).
Thank you very much for your help.
Gratefully,
Pedro Santos
- Marquer comme nouveau
- Marquer
- S'abonner
- Sourdine
- S'abonner au fil RSS
- Surligner
- Imprimer
- Signaler un contenu inapproprié
Dear Pedro,
Only complex (cggev/zggev) ?GGEV routines support parallelism at the moment. You can see it in Users's Guideshttp://software.intel.com/en-us/articles/intel-math-kernel-library-documentation/,section Threaded Functions and Problems.
- Marquer comme nouveau
- Marquer
- S'abonner
- Sourdine
- S'abonner au fil RSS
- Surligner
- Imprimer
- Signaler un contenu inapproprié
Dear Aleksandr Zotkevich,
Thank you very much. I was completely unaware of it. That explains my issue. Do you know when parallelism will be available in dggev?
Gratefully,
Pedro Santos
Thank you very much. I was completely unaware of it. That explains my issue. Do you know when parallelism will be available in dggev?
Gratefully,
Pedro Santos
Répondre
Options du sujet
- S'abonner au fil RSS
- Marquer le sujet comme nouveau
- Marquer le sujet comme lu
- Placer ce Sujet en tête de liste pour l'utilisateur actuel
- Marquer
- S'abonner
- Page imprimable