GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-05-03 01:04:23 +08:00
Code Issues Packages Projects Releases Wiki Activity

Optimized the configuration of the outer reduction cuda kernel

Browse Source
This commit is contained in:
Benoit Steiner 2015-12-22 16:30:10 -08:00
parent 9c7d96697b
commit a1e08fb2a5
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h
View File

@ -179,7 +179,10 @@ struct OuterReducer<Self, Op, GpuDevice> {
const Index num_coeffs = num_coeffs_to_reduce * num_preserved_vals;
const int block_size = 256;
const int num_per_thread = 16;
const int num_blocks = std::ceil(static_cast<float>(num_coeffs) / (block_size * num_per_thread));
const int dyn_blocks = std::ceil(static_cast<float>(num_coeffs) / (block_size * num_per_thread));
const int max_blocks = device.getNumCudaMultiProcessors() *
device.maxCudaThreadsPerMultiProcessor() / block_size;
const int num_blocks = numext::mini<int>(max_blocks, dyn_blocks);
LAUNCH_CUDA_KERNEL((OuterReductionKernel<num_per_thread>),
num_blocks, block_size, 0, device, reducer, self, num_coeffs_to_reduce, num_preserved_vals, output);