4 Replies Latest reply on Oct 21, 2010 8:35 AM by spectral

    Transfer between CPU & GPU... optimization ?

    spectral

      Hi,

      I have the following code :

      int updateDataSet = MustUpdateDataSet(...);

      if (updateDataSet)
          kernel->setArg(i++, *dataSet);

      What I would like is to avoid to send each time the 'dataset' to the GPU because it has already been sent !

      So, my question is... can I do this this ? or I MUST ALWAYS set the argument ?

      But if I must always use 'setArgs' ... how can I avoid to re-transfer each time the huge (a same) set of datas ?