GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-06-04 18:54:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

Don't attempt to optimize partial reductions when the optimized implementation doesn't buy anything.

Browse Source
This commit is contained in:
Benoit Steiner 2016-08-08 19:29:59 -07:00
parent 5157ce8cbf
commit 64e68cbe87
1 changed files with 20 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
View File

@ -505,12 +505,20 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
(reducing_inner_dims || ReducingInnerMostDims)) {
const Index num_values_to_reduce = internal::array_prod(m_reducedDims);
const Index num_coeffs_to_preserve = internal::array_prod(m_dimensions);
if (!data && num_coeffs_to_preserve < 1024 && num_values_to_reduce > num_coeffs_to_preserve) {
if (!data && num_coeffs_to_preserve < 1024 && num_values_to_reduce > num_coeffs_to_preserve && num_values_to_reduce > 128) {
data = static_cast<CoeffReturnType*>(m_device.allocate(sizeof(CoeffReturnType) * num_coeffs_to_preserve));
m_result = data;
}
Op reducer(m_reducer);
return internal::InnerReducer<Self, Op, Device>::run(*this, reducer, m_device, data, num_values_to_reduce, num_coeffs_to_preserve) || (m_result != NULL);
if (internal::InnerReducer<Self, Op, Device>::run(*this, reducer, m_device, data, num_values_to_reduce, num_coeffs_to_preserve)) {
if (m_result) {
m_device.deallocate(m_result);
m_result = NULL;
}
return true;
} else {
return (m_result != NULL);
}
}
bool preserving_inner_dims = true;
@ -525,12 +533,20 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
preserving_inner_dims) {
const Index num_values_to_reduce = internal::array_prod(m_reducedDims);
const Index num_coeffs_to_preserve = internal::array_prod(m_dimensions);
if (!data && num_coeffs_to_preserve < 1024 && num_values_to_reduce > num_coeffs_to_preserve) {
if (!data && num_coeffs_to_preserve < 1024 && num_values_to_reduce > num_coeffs_to_preserve && num_values_to_reduce > 32) {
data = static_cast<CoeffReturnType*>(m_device.allocate(sizeof(CoeffReturnType) * num_coeffs_to_preserve));
m_result = data;
}
Op reducer(m_reducer);
return internal::OuterReducer<Self, Op, Device>::run(*this, reducer, m_device, data, num_values_to_reduce, num_coeffs_to_preserve) || (m_result != NULL);
if (internal::OuterReducer<Self, Op, Device>::run(*this, reducer, m_device, data, num_values_to_reduce, num_coeffs_to_preserve)) {
if (m_result) {
m_device.deallocate(m_result);
m_result = NULL;
}
return true;
} else {
return (m_result != NULL);
}
}
}
return true;