nico

Forum Replies Created

Viewing 15 posts - 1 through 15 (of 62 total)
  • Author
    Posts
  • in reply to: BiCGStab with DIA diverging for large problem #1509

    nico
    Participant

    Hi Deborah,

    sorry for the delay. Due to different rounding, small variations could happen between CSR and DIA format. However, this should not lead to divergence. Is it possible to get access to your linear system where this is happening?

    Best
    Nico

    in reply to: Linking error related to Local Vector? #1508

    nico
    Participant

    Hey Klaus,

    unfortunately, we support only OpenFOAM 2.3.

    Best
    Nico


    nico
    Participant

    Hey Klaus,

    unfortunately, we support only OpenFOAM 2.3.

    Best
    Nico


    nico
    Participant

    Hello,

    in order to overcome the memory transfers and latencies, your test case has to be sufficiently large. You can use our OpenFOAM laplace example for some tests.

    Best
    Nico

    in reply to: OpenMP trouble in Visual Studio compilation #1501

    nico
    Participant

    Hi Matthaios,

    I did some very simple tests without PARALUTION and could not get OpenMP working either. It looks to me, that there is no native OpenMP support in VS2010.

    Best
    Nico

    in reply to: GPU memory for 40m cells OpenFoam case? #1500

    nico
    Participant

    Hi Klaus,

    this depends a lot on your specific case. There is no general rule on memory consumption for 40M cells.

    Best
    Nico

    in reply to: Order of the columns of a CSR matrix #1485

    nico
    Participant

    Hi Jeremie,

    Most of our algorithms require sorted rows in CSR format.

    Best
    Nico

    in reply to: Reading data #1484

    nico
    Participant

    Hi,

    you can use SetDataPtr functions. Please have a look at the user manual for details.

    Best
    Nico

    in reply to: Laplace example not working for foam-extend-3.1 #1477

    nico
    Participant

    Hi Savee,

    unfortunately, we currently have no support for OpenFOAM 3.1. However, you can adjust the plugin files supplied with PARALUTION to your OpenFOAM version. Just have a look at the OpenFOAM 3.1 solver implementation and change the OpenFOAM class, function and variable names to the 3.1 version.

    Best
    Nico

    • This reply was modified 9 months, 1 week ago by  nico.
    in reply to: Info on solver implementation? #1476

    nico
    Participant

    Hello Deborah,

    since PARALUTION is released under the terms of the GPLv3 license, you can find all implementation details directly in the source code! The BiCGStab method for example is implemented in the solvers/krylov/bicgstab.cpp file.

    Best
    Nico

    in reply to: CUDA OpenCL and openmp #1473

    nico
    Participant

    Hi Najib,

    looks like there is something wrong with your OpenCL SDK installation. Please make sure that everything related to OpenCL works, before moving on to PARALUTION.

    Regarding the OpenMP test you did, please try again with a larger case. You can check the UF Sparse Matrix Collection for example to find a larger test case.

    Best
    Nico

    in reply to: CUDA OpenCL and openmp #1470

    nico
    Participant

    Hi Najib,

    Regarding the CPU scalability: Is your test case sufficiently large? What kind of CPU are you using, is it possibly, that your memory bandwidth is already fully utilized by very few cores?
    You are trying to run the CUDA backend on AMD hardware, this is not possible since CUDA is only supported by NVIDIA hardware.
    Regarding OpenCL, can you make sure that the OpenCL runtime environment is properly configured and working? You can test this by running some opencl samples supplied with the AMD opencl sdk or by checking clinfo.

    Best
    Nico

    in reply to: Low performance of SpMV on Xeon Phi #1465

    nico
    Participant

    Hi Daniel,

    our current MIC implementation is not very optimized and therefore not optimal. You should make sure that MKL is enabled during the compilation process. Unfortunately, due to very few MIC users, we have currently no plans to further improve the MIC kernels.

    Best
    Nico

    in reply to: Matrix arrays not accessible after moving to accelerator #1464

    nico
    Participant

    Hi Isaac,

    if you use the Set/LeaveDataPtr functions, raw pointers to the structures are passed from paralution to you. You can do with the data what ever you like. You don’t have to reallocate anything, just update the data with your own functions (keep in mind, that the pointers are only valid on the device where you extracted them from).

    Best
    Nico

    in reply to: Matrix arrays not accessible after moving to accelerator #1461

    nico
    Participant

    Hi Isaac,

    you can use the Set/LeaveDataPtr functions to access the raw data pointers without copying.

    Best
    Nico

Viewing 15 posts - 1 through 15 (of 62 total)