devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Improved the performance of full reductions on GPU:

Browse Source 
Before:
BM_fullReduction/10       200000      11751     8.51 MFlops/s
BM_fullReduction/80         5000     523385    12.23 MFlops/s
BM_fullReduction/640          50   36179326    11.32 MFlops/s
BM_fullReduction/4K            1 2173517195    11.50 MFlops/s

After:
BM_fullReduction/10       500000       5987    16.70 MFlops/s
BM_fullReduction/80       200000      10636   601.73 MFlops/s
BM_fullReduction/640       50000      58428  7010.31 MFlops/s
BM_fullReduction/4K         1000    2006106 12461.95 MFlops/s
This commit is contained in:
Benoit Steiner
2016-05-09 17:09:54 -07:00
parent c3859a2b58
commit 4670d7d5ce
2 changed files with 160 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
View File

@@ -322,6 +322,12 @@ struct OuterReducer {
template <int B, int N, typename S, typename R, typename I>
__global__ void FullReductionKernel(R, const S, I, typename S::CoeffReturnType*);
template <typename S, typename R, typename I>
__global__ void ReductionInitKernelHalfFloat(R, const S, I, half2*);
template <int B, int N, typename S, typename R, typename I>
__global__ void FullReductionKernelHalfFloat(R, const S, I, half*, half2*);
template <int NPT, typename S, typename R, typename I>
__global__ void InnerReductionKernel(R, const S, I, I, typename S::CoeffReturnType*);
@@ -618,6 +624,8 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
#endif
#if defined(EIGEN_USE_GPU) && defined(__CUDACC__)
template <int B, int N, typename S, typename R, typename I> friend void internal::FullReductionKernel(R, const S, I, typename S::CoeffReturnType*);
template <typename S, typename R, typename I> friend void internal::ReductionInitKernelHalfFloat(R, const S, I, half2*);
template <int B, int N, typename S, typename R, typename I> friend void internal::FullReductionKernelHalfFloat(R, const S, I, half*, half2*);
template <int NPT, typename S, typename R, typename I> friend void internal::InnerReductionKernel(R, const S, I, I, typename S::CoeffReturnType*);
template <int NPT, typename S, typename R, typename I> friend void internal::OuterReductionKernel(R, const S, I, I, typename S::CoeffReturnType*);
#endif