diff options
author | Benoit Steiner <benoit.steiner.goog@gmail.com> | 2016-04-20 18:08:20 -0700 |
---|---|---|
committer | Benoit Steiner <benoit.steiner.goog@gmail.com> | 2016-04-20 18:08:20 -0700 |
commit | 2dde1b102866e1928e925678951463f2a7051af1 (patch) | |
tree | d6405558903617aa9f5fabe4be8b5601fe6d406a /unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h | |
parent | a792cd357d31f0a4fce62ed1fa4cc0334cf2f143 (diff) |
Don't crash when attempting to reduce empty tensors.
Diffstat (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h')
-rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h b/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h index bf6e10a7b..c3edae477 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h @@ -238,7 +238,7 @@ inline void TensorExecutor<Expression, GpuDevice, Vectorizable>::run( device.maxCudaThreadsPerMultiProcessor() / block_size; const Index size = array_prod(evaluator.dimensions()); // Create a least one block to ensure we won't crash when tensorflow calls with tensors of size 0. - const int num_blocks = numext::maxi<int>(numext::mini<int>(max_blocks, (size + block_size - 1) / block_size), 1); + const int num_blocks = numext::maxi<int>(numext::mini<int>(max_blocks, divup<int>(size, block_size)), 1); LAUNCH_CUDA_KERNEL( (EigenMetaKernel<TensorEvaluator<Expression, GpuDevice>, Index>), |