From ab0a30e4292f109326b444dabb1e9e0c1dc29881 Mon Sep 17 00:00:00 2001 From: Rasmus Munk Larsen Date: Wed, 15 May 2019 13:32:15 -0700 Subject: Make Eigen build with cuda 10 and clang. --- unsupported/Eigen/CXX11/src/Tensor/TensorReductionGpu.h | 4 ---- 1 file changed, 4 deletions(-) (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorReductionGpu.h') 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 { 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(num_preserved_vals, 1024); - const int max_blocks = device.getNumGpuMultiProcessors() * - device.maxGpuThreadsPerMultiProcessor() / 1024; - const int num_blocks = numext::mini(max_blocks, dyn_blocks); LAUNCH_GPU_KERNEL((ReductionInitKernelHalfFloat), 1, 1, 0, device, reducer, self, num_preserved_vals, output); } -- cgit v1.2.3