GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-13 20:26:03 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fixes auto appearance in functor template argument for reduction.

Browse Source
This commit is contained in:
Luke Iwanski 2017-01-04 22:18:44 +00:00
parent 3eda02d78d
commit 90c5bc8d64
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
unsupported/Eigen/CXX11/src/Tensor/TensorReductionSycl.h
View File

@ -74,8 +74,8 @@ struct FullReducer<Self, Op, const Eigen::SyclDevice, Vectorizable> {
static void run(const Self& self, Op& reducer, const Eigen::SyclDevice& dev, CoeffReturnType* output) { static void run(const Self& self, Op& reducer, const Eigen::SyclDevice& dev, CoeffReturnType* output) {
typedef const typename Self::ChildType HostExpr; /// this is the child of reduction typedef const typename Self::ChildType HostExpr; /// this is the child of reduction
auto functors = TensorSycl::internal::extractFunctors(self.impl()); typedef decltype(TensorSycl::internal::extractFunctors(self.impl())) FunctorExpr;
typedef decltype(functors) FunctorExpr; FunctorExpr functors = TensorSycl::internal::extractFunctors(self.impl());
int red_factor =256; /// initial reduction. If the size is less than red_factor we only creates one thread. int red_factor =256; /// initial reduction. If the size is less than red_factor we only creates one thread.
size_t inputSize =self.impl().dimensions().TotalSize(); size_t inputSize =self.impl().dimensions().TotalSize();
size_t rng = inputSize/red_factor; // the total number of thread initially is half the size of the input size_t rng = inputSize/red_factor; // the total number of thread initially is half the size of the input
@ -136,8 +136,8 @@ struct InnerReducer<Self, Op, const Eigen::SyclDevice> {
static bool run(const Self& self, Op& reducer, const Eigen::SyclDevice& dev, CoeffReturnType* output, typename Self::Index , typename Self::Index num_coeffs_to_preserve) { static bool run(const Self& self, Op& reducer, const Eigen::SyclDevice& dev, CoeffReturnType* output, typename Self::Index , typename Self::Index num_coeffs_to_preserve) {
typedef const typename Self::ChildType HostExpr; /// this is the child of reduction typedef const typename Self::ChildType HostExpr; /// this is the child of reduction
auto functors = TensorSycl::internal::extractFunctors(self.impl()); typedef decltype(TensorSycl::internal::extractFunctors(self.impl())) FunctorExpr;
typedef decltype(functors) FunctorExpr; FunctorExpr functors = TensorSycl::internal::extractFunctors(self.impl());
typename Self::Index range, GRange, tileSize; typename Self::Index range, GRange, tileSize;
typedef typename Eigen::internal::remove_all<decltype(self.xprDims())>::type Dims; typedef typename Eigen::internal::remove_all<decltype(self.xprDims())>::type Dims;