diff options
| author |  Rasmus Larsen <rmlarsen@google.com> | 2019-05-20 17:02:39 +0000 |
|---|---|---|
| committer | Rasmus Larsen <rmlarsen@google.com> | 2019-05-20 17:02:39 +0000 |
| commit | e92486b8c34272f6eae563665fc89b7f0abb902f (patch) | |
| tree | 3e5554b3ec06ff8dcee8519ff6c5e824854deb89 /unsupported/Eigen/CXX11 | |
| parent | cc7ecbb1241cdc16b0b0f27631c556d8eda5aa4b (diff) | |
| parent | fd595d42a76437cc9df0fcfbc49c5372e416c8d6 (diff) | |
Merged in rmlarsen/eigen (pull request PR-643)
Make Eigen build with cuda 10 and clang.
Approved-by: Justin Lebar <justin.lebar@gmail.com>
Diffstat (limited to 'unsupported/Eigen/CXX11')
| -rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorReductionGpu.h | 4 |
1 files changed, 0 insertions, 4 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionGpu.h b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionGpu.h index 0718ba2a1..7ee4a6087 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionGpu.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionGpu.h @@ -674,10 +674,6 @@ struct InnerReductionLauncher<Self, Op, Eigen::half, true> { if (num_blocks > 1) { // We initialize the outputs outside the reduction kernel when we can't be sure that there // won't be a race conditions between multiple thread blocks. - const int dyn_blocks = divup<int>(num_preserved_vals, 1024); - const int max_blocks = device.getNumGpuMultiProcessors() * - device.maxGpuThreadsPerMultiProcessor() / 1024; - const int num_blocks = numext::mini<int>(max_blocks, dyn_blocks); LAUNCH_GPU_KERNEL((ReductionInitKernelHalfFloat<Self, Op, Index>), 1, 1, 0, device, reducer, self, num_preserved_vals, output); } |