mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-21 07:19:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

Added a test for multithreaded full reductions

Browse Source
This commit is contained in:
Benoit Steiner 2015-06-30 13:08:12 -07:00
parent a4aa7c6217
commit 109005c6c9
2 changed files with 32 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
unsupported/Eigen/CXX11/src/Tensor/TensorFunctors.h
View File

@ -17,6 +17,7 @@ namespace internal {
template <typename T> struct SumReducer template <typename T> struct SumReducer
{ {
static const bool PacketAccess = true; static const bool PacketAccess = true;
static const bool IsStateful = false;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const {
(*accum) += t; (*accum) += t;
@ -49,6 +50,8 @@ template <typename T> struct SumReducer
template <typename T> struct MeanReducer template <typename T> struct MeanReducer
{ {
static const bool PacketAccess = true; static const bool PacketAccess = true;
static const bool IsStateful = true;
MeanReducer() : scalarCount_(0), packetCount_(0) { } MeanReducer() : scalarCount_(0), packetCount_(0) { }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) {
@ -88,6 +91,7 @@ template <typename T> struct MeanReducer
template <typename T> struct MaxReducer template <typename T> struct MaxReducer
{ {
static const bool PacketAccess = true; static const bool PacketAccess = true;
static const bool IsStateful = false;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const {
if (t > *accum) { *accum = t; } if (t > *accum) { *accum = t; }
@ -120,6 +124,7 @@ template <typename T> struct MaxReducer
template <typename T> struct MinReducer template <typename T> struct MinReducer
{ {
static const bool PacketAccess = true; static const bool PacketAccess = true;
static const bool IsStateful = false;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const {
if (t < *accum) { *accum = t; } if (t < *accum) { *accum = t; }
@ -153,6 +158,7 @@ template <typename T> struct MinReducer
template <typename T> struct ProdReducer template <typename T> struct ProdReducer
{ {
static const bool PacketAccess = true; static const bool PacketAccess = true;
static const bool IsStateful = false;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const {
(*accum) *= t; (*accum) *= t;

26
unsupported/test/cxx11_tensor_thread_pool.cpp
View File

@ -228,6 +228,29 @@ static void test_multithread_contraction_agrees_with_singlethread() {
} }
template<int DataLayout>
static void test_multithreaded_reductions() {
const int num_threads = internal::random<int>(3, 11);
ThreadPool thread_pool(num_threads);
Eigen::ThreadPoolDevice thread_pool_device(&thread_pool, num_threads);
const int num_rows = internal::random<int>(13, 732);
const int num_cols = internal::random<int>(13, 732);
Tensor<float, 2, DataLayout> t1(num_rows, num_cols);
t1.setRandom();
Tensor<float, 1, DataLayout> full_redux(1);
full_redux = t1.sum();
Tensor<float, 1, DataLayout> full_redux_tp(1);
full_redux_tp.device(thread_pool_device) = t1.sum();
// Check that the single threaded and the multi threaded reductions return
// the same result.
VERIFY_IS_APPROX(full_redux(0), full_redux_tp(0));
}
static void test_memcpy() { static void test_memcpy() {
for (int i = 0; i < 5; ++i) { for (int i = 0; i < 5; ++i) {
@ -271,6 +294,9 @@ void test_cxx11_tensor_thread_pool()
CALL_SUBTEST(test_contraction_corner_cases<ColMajor>()); CALL_SUBTEST(test_contraction_corner_cases<ColMajor>());
CALL_SUBTEST(test_contraction_corner_cases<RowMajor>()); CALL_SUBTEST(test_contraction_corner_cases<RowMajor>());
CALL_SUBTEST(test_multithreaded_reductions<ColMajor>());
CALL_SUBTEST(test_multithreaded_reductions<RowMajor>());
CALL_SUBTEST(test_memcpy()); CALL_SUBTEST(test_memcpy());
CALL_SUBTEST(test_multithread_random()); CALL_SUBTEST(test_multithread_random());