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

Allow vectorized padding on GPU. This helps speed things up a little.

Browse Source 
Before:
BM_padding/10            5000000        460   217.03 MFlops/s
BM_padding/80            5000000        460 13899.40 MFlops/s
BM_padding/640           5000000        461 888421.17 MFlops/s
BM_padding/4K            5000000        460 54316322.55 MFlops/s
After:
BM_padding/10            5000000        454   220.20 MFlops/s
BM_padding/80            5000000        455 14039.86 MFlops/s
BM_padding/640           5000000        452 904968.83 MFlops/s
BM_padding/4K            5000000        411 60750049.21 MFlops/s
This commit is contained in:
Benoit Steiner
2016-05-17 09:13:27 -07:00
parent a80d875916
commit 8d06c02ffd
4 changed files with 107 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
View File

@@ -328,6 +328,9 @@ 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 InnerReductionKernelHalfFloat(R, const S, I, I, half*, half2*);
#endif
template <int NPT, typename S, typename R, typename I>
@@ -629,11 +632,15 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
#ifdef EIGEN_HAS_CUDA_FP16
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::InnerReductionKernelHalfFloat(R, const S, I, I, half*, half2*);
#endif
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
template <typename S, typename O, typename D> friend struct internal::InnerReducer;
// Returns the Index in the input tensor of the first value that needs to be
// used to compute the reduction at output index "index".
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index firstInput(Index index) const {