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

Use scalar_sum_op and scalar_quotient_op instead of operator+ and operator/ in MeanReducer.

Browse Source 
Improves support for std::complex types when compiling for CUDA.

Expands on e2e9cdd169
 and 2bda1b0d93
.
This commit is contained in:
RJ Ryan
2017-04-14 13:23:35 -07:00
parent d9084ac8e1
commit 949a2da38c
2 changed files with 42 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -166,7 +166,8 @@ template <typename T> struct MeanReducer
return pset1<Packet>(initialize());
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalize(const T accum) const {
return accum / scalarCount_;
internal::scalar_quotient_op<T> quotient_op;
return quotient_op(accum, T(scalarCount_));
}
template <typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet finalizePacket(const Packet& vaccum) const {
@@ -175,7 +176,10 @@ template <typename T> struct MeanReducer
template <typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalizeBoth(const T saccum, const Packet& vaccum) const {
internal::scalar_sum_op<T> sum_op;
return sum_op(saccum, predux(vaccum)) / (scalarCount_ + packetCount_ * unpacket_traits<Packet>::size);
internal::scalar_quotient_op<T> quotient_op;
return quotient_op(
sum_op(saccum, predux(vaccum)),
T(scalarCount_ + packetCount_ * unpacket_traits<Packet>::size));
}
protected: