diff options
author | Benoit Steiner <benoit.steiner.goog@gmail.com> | 2016-01-14 09:29:48 -0800 |
---|---|---|
committer | Benoit Steiner <benoit.steiner.goog@gmail.com> | 2016-01-14 09:29:48 -0800 |
commit | 8fe2532e70a8e0261717003d96d4df41ab978756 (patch) | |
tree | 56cee0a781ade05d15fa5d480c30435a9bac461f /unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h | |
parent | 9f013a9d86ad5cf82939bfeab2223652a821c448 (diff) |
Fixed a boundary condition bug in the outer reduction kernel
Diffstat (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h')
-rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h index 89f055134..54ab34ba1 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h @@ -241,7 +241,7 @@ __global__ void OuterReductionKernel(Reducer reducer, const Self input, Index nu } // Do the reduction. - const Index max_iter = divup<Index>(num_coeffs_to_reduce, NumPerThread) * num_preserved_coeffs; + const Index max_iter = num_preserved_coeffs * numext::maxi<Index>(1, (num_coeffs_to_reduce - NumPerThread + 1)); for (Index i = thread_id; i < max_iter; i += num_threads) { const Index input_col = i % num_preserved_coeffs; const Index input_row = (i / num_preserved_coeffs) * NumPerThread; |