devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Alternative way of forcing instantiation of device kernels without

Browse Source 
causing warnings or requiring device to device kernel invocations.

This allows Tensorflow to work on SM 3.0 (ie, Amazon EC2) machines.
This commit is contained in:
Jeremy Barnes
2016-01-10 22:39:13 -05:00
parent b557662e58
commit 403a7cb6c3
3 changed files with 13 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h
View File

@@ -116,7 +116,7 @@ struct FullReducer<Self, Op, GpuDevice, Vectorizable> {
template <typename OutputType>
static void run(const Self& self, Op& reducer, const GpuDevice& device, OutputType* output) {
assert(false && "Should only be called on floats");
eigen_assert(false && "Should only be called on floats");
}
static void run(const Self& self, Op& reducer, const GpuDevice& device, float* output) {
@@ -126,7 +126,7 @@ struct FullReducer<Self, Op, GpuDevice, Vectorizable> {
const int block_size = 256;
const int num_per_thread = 128;
const int num_blocks = std::ceil(static_cast<float>(num_coeffs) / (block_size * num_per_thread));
LAUNCH_CUDA_KERNEL((FullReductionKernel<block_size, num_per_thread>),
LAUNCH_CUDA_KERNEL((FullReductionKernel<block_size, num_per_thread, Self, Op, Index>),
num_blocks, block_size, 0, device, reducer, self, num_coeffs, output);
}
};