Commit af6d1d6b authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Added default stream synchronizations after kernel launches in CudaPrefixSumKernel.h

parent 8d0d2638

src/TNL/Containers/Algorithms/CudaPrefixSumKernel.h

+9 −8
Original line number Diff line number Diff line
@@ -231,6 +231,7 @@ struct CudaPrefixSumKernelLauncher 
           output,

           auxArray1.getData(),

           gridShift );

      cudaStreamSynchronize(0);

      TNL_CHECK_CUDA_DEVICE;
@@ -260,6 +261,7 @@ struct CudaPrefixSumKernelLauncher 
           gridShift,

           auxArray2.getData(),

           output );

      cudaStreamSynchronize(0);

      TNL_CHECK_CUDA_DEVICE;



      cudaMemcpy( &gridShift,
@@ -323,11 +325,10 @@ struct CudaPrefixSumKernelLauncher 
            gridShift,

            &deviceInput[ gridOffset ],

            &deviceOutput[ gridOffset ] );

         TNL_CHECK_CUDA_DEVICE;

      }



      /***

       * Store the number of CUDA grids for a purpose of unit testing, i.e.

       * Store the number of CUDA grids for the purpose of unit testing, i.e.

       * to check if we test the algorithm with more than one CUDA grid.

       */

      gridsCount = numberOfGrids;