Skip to content
forked from NVIDIA/cub

CUB is a flexible library of cooperative threadblock primitives and other utilities for CUDA kernel programming.

License

Notifications You must be signed in to change notification settings

maddyscientist/cub

 
 

Repository files navigation


About CUB

Current release: v1.1.1 (December 11, 2013)

We recommend the CUB Project Website and the cub-users discussion forum for further information and examples.

CUB provides state-of-the-art, reusable software components for every layer of the CUDA programming model:

Orientation of collective primitives within the CUDA software stack



A Simple Example

#include <cub/cub.cuh>
 
// Block-sorting CUDA kernel
__global__ void BlockSortKernel(int *d_in, int *d_out)
{
     using namespace cub;

     // Specialize BlockRadixSort, BlockLoad, and BlockStore for 128 threads 
     // owning 16 integer items each
     typedef BlockRadixSort<int, 128, 16>                     BlockRadixSort;
     typedef BlockLoad<int*, 128, 16, BLOCK_LOAD_TRANSPOSE>   BlockLoad;
     typedef BlockStore<int*, 128, 16, BLOCK_STORE_TRANSPOSE> BlockStore;
 
     // Allocate shared memory
     __shared__ union {
         typename BlockRadixSort::TempStorage  sort;
         typename BlockLoad::TempStorage       load; 
         typename BlockStore::TempStorage      store; 
     } temp_storage; 

     int block_offset = blockIdx.x * (128 * 16);	  // Offset for this block's ment

     // Obtain a segment of 2048 consecutive keys that are blocked across threads
     int thread_keys[16];
     BlockLoad(temp_storage.load).Load(d_in + block_offset, thread_keys);
     __syncthreads();

     // Collectively sort the keys
     BlockRadixSort(temp_storage.sort).Sort(thread_keys);
     __syncthreads();

     // Store the sorted segment 
     BlockStore(temp_storage.store).Store(d_out + block_offset, thread_keys);
}

Each thread block uses cub::BlockRadixSort to collectively sort its own input segment. The class is specialized by the data type being sorted, by the number of threads per block, by the number of keys per thread, and implicitly by the targeted compilation architecture.

The cub::BlockLoad and cub::BlockStore classes are similarly specialized.
Furthermore, to provide coalesced accesses to device memory, these primitives are configured to access memory using a striped access pattern (where consecutive threads simultaneously access consecutive items) and then transpose the keys into a blocked arrangement of elements across threads.

Once specialized, these classes expose opaque \p TempStorage member types.
The thread block uses these storage types to statically allocate the union of shared memory needed by the thread block. (Alternatively these storage types could be aliased to global memory allocations).



Releases

See CUB Project Website for more information.

Date Version
12/10/2013 CUB v1.1.1 Primary Release
08/08/2013 CUB v1.0.1 Primary Release
05/07/2013 CUB v0.9.4 Update Release
04/04/2013 CUB v0.9.2 Update Release
03/09/2013 CUB v0.9.1 Update Release
03/07/2013 CUB v0.9.0 Preview Release



Contributors

CUB is developed as an open-source project by NVIDIA Research. The primary contributor is Duane Merrill.



Open Source License

CUB is available under the "New BSD" open-source license:

Copyright (c) 2010-2011, Duane Merrill.  All rights reserved.
Copyright (c) 2011-2013, NVIDIA CORPORATION.  All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
   *  Redistributions of source code must retain the above copyright
      notice, this list of conditions and the following disclaimer.
   *  Redistributions in binary form must reproduce the above copyright
      notice, this list of conditions and the following disclaimer in the
      documentation and/or other materials provided with the distribution.
   *  Neither the name of the NVIDIA CORPORATION nor the
      names of its contributors may be used to endorse or promote products
      derived from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE FOR ANY
DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

About

CUB is a flexible library of cooperative threadblock primitives and other utilities for CUDA kernel programming.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published