Commit 4bbf495c authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Added specialization of CudaBlockReduce using __shfl instructions

parent 05903a8f

src/TNL/Algorithms/detail/CudaReductionKernel.h

+211 −72

File changed.

Preview size limit exceeded, changes collapsed.

src/TNL/Algorithms/detail/CudaScanKernel.h

+1 −1
Original line number Diff line number Diff line
@@ -312,7 +312,7 @@ CudaScanKernelUpsweep( const InputView input, 
   __syncthreads();



   // Perform the parallel reduction.

   value = BlockReduce::reduce( reduction, value, threadIdx.x, storage.blockReduceStorage );

   value = BlockReduce::reduce( reduction, identity, value, threadIdx.x, storage.blockReduceStorage );



   // Store the block result in the global memory.

   if( threadIdx.x == 0 )