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

Fixed a bug impacting some outer reductions on GPU

Browse Source
This commit is contained in:
Benoit Steiner
2016-09-12 18:36:52 -07:00
parent 5f50f12d2c
commit 028e299577
2 changed files with 110 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
View File

@@ -505,9 +505,14 @@ 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 && num_values_to_reduce > 128) {
data = static_cast<CoeffReturnType*>(m_device.allocate(sizeof(CoeffReturnType) * num_coeffs_to_preserve));
m_result = data;
if (!data) {
if (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;
}
else {
return true;
}
}
Op reducer(m_reducer);
if (internal::InnerReducer<Self, Op, Device>::run(*this, reducer, m_device, data, num_values_to_reduce, num_coeffs_to_preserve)) {
@@ -533,9 +538,14 @@ 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 && num_values_to_reduce > 32) {
data = static_cast<CoeffReturnType*>(m_device.allocate(sizeof(CoeffReturnType) * num_coeffs_to_preserve));
m_result = data;
if (!data) {
if (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;
}
else {
return true;
}
}
Op reducer(m_reducer);
if (internal::OuterReducer<Self, Op, Device>::run(*this, reducer, m_device, data, num_values_to_reduce, num_coeffs_to_preserve)) {
@@ -556,6 +566,7 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
m_impl.cleanup();
if (m_result) {
m_device.deallocate(m_result);
m_result = NULL;
}
}