mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-03-07 18:27:40 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix broken tensor executor test, allow tensor packets of size 1.

Browse Source
This commit is contained in:
Antonio Sánchez 2022-03-07 20:30:37 +00:00 committed by Rasmus Munk Larsen
parent 28c7c1a629
commit 008ff3483a
13 changed files with 28 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
unsupported/Eigen/CXX11/src/Tensor/README.md
View File

@ -1847,8 +1847,3 @@ product of 2 1d tensors (through contractions) returns a 0d tensor.
* The IndexList class requires a cxx11 compliant compiler. You can use an
array of indices instead if you don't have access to a modern compiler.
* On GPUs only floating point values are properly tested and optimized for.
* Complex and integer values are known to be broken on GPUs. If you try to use
them you'll most likely end up triggering a static assertion failure such as
EIGEN_STATIC_ASSERT(packetSize > 1, YOU_MADE_A_PROGRAMMING_MISTAKE)

5
unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h
View File

@ -370,7 +370,6 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packetOneByNByOne
(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
EIGEN_ALIGN_MAX typename internal::remove_const<CoeffReturnType>::type values[PacketSize];
@ -416,7 +415,6 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
// Concatenates m_broadcast[dim] copies,
// [v0, ..., vN, v0, ..., vN, ... ]
// with dim == NumDims - 1 for col-major, dim == 0 for row-major.
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
// Size of flattened tensor.
@ -445,7 +443,6 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
// Interleaves m_broadcast[dim] copies,
// [v0, v0, ..., v1, v1, ..., vN, vN, ... ]
// with dim == 0 for col-major, dim == NumDims - 1 for row-major.
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index + PacketSize-1 < dimensions().TotalSize());
const Index M = (static_cast<int>(Layout) == static_cast<int>(ColMajor)) ?
@ -476,7 +473,6 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packetColMajor(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
const Index originalIndex = index;
@ -534,7 +530,6 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packetRowMajor(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
const Index originalIndex = index;

3
unsupported/Eigen/CXX11/src/Tensor/TensorChipping.h
View File

@ -219,7 +219,6 @@ struct TensorEvaluator<const TensorChippingOp<DimId, ArgType>, Device>
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
if (isInnerChipping()) {
@ -447,8 +446,6 @@ struct TensorEvaluator<TensorChippingOp<DimId, ArgType>, Device>
template <int StoreMode> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketReturnType& x)
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
if (this->isInnerChipping()) {
// m_stride is equal to 1, so let's avoid the integer division.
eigen_assert(this->m_stride == 1);

1
unsupported/Eigen/CXX11/src/Tensor/TensorGenerator.h
View File

@ -155,7 +155,6 @@ struct TensorEvaluator<const TensorGeneratorOp<Generator, ArgType>, Device>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
{
const int packetSize = PacketType<CoeffReturnType, Device>::size;
EIGEN_STATIC_ASSERT((packetSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+packetSize-1 < dimensions().TotalSize());
EIGEN_ALIGN_MAX typename internal::remove_const<CoeffReturnType>::type values[packetSize];

1
unsupported/Eigen/CXX11/src/Tensor/TensorImagePatch.h
View File

@ -449,7 +449,6 @@ struct TensorEvaluator<const TensorImagePatchOp<Rows, Cols, ArgType>, Device>
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
if (m_in_row_strides != 1 || m_in_col_strides != 1 || m_row_inflate_strides != 1 || m_col_inflate_strides != 1) {

2
unsupported/Eigen/CXX11/src/Tensor/TensorPadding.h
View File

@ -551,7 +551,6 @@ struct TensorEvaluator<const TensorPaddingOp<PaddingDimensions, ArgType>, Device
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packetColMajor(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
const Index initialIndex = index;
@ -609,7 +608,6 @@ struct TensorEvaluator<const TensorPaddingOp<PaddingDimensions, ArgType>, Device
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packetRowMajor(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
const Index initialIndex = index;

1
unsupported/Eigen/CXX11/src/Tensor/TensorPatch.h
View File

@ -197,7 +197,6 @@ struct TensorEvaluator<const TensorPatchOp<PatchDim, ArgType>, Device>
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
Index output_stride_index = (static_cast<int>(Layout) == static_cast<int>(ColMajor)) ? NumDims - 1 : 0;

1
unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
View File

@ -837,7 +837,6 @@ static const bool RunningOnGPU = false;
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index + PacketSize - 1 < Index(internal::array_prod(dimensions())));
if (RunningOnGPU && m_result) {

2
unsupported/Eigen/CXX11/src/Tensor/TensorReverse.h
View File

@ -215,7 +215,6 @@ struct TensorEvaluator<const TensorReverseOp<ReverseDimensions, ArgType>, Device
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
PacketReturnType packet(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
// TODO(ndjaitly): write a better packing routine that uses
@ -448,7 +447,6 @@ struct TensorEvaluator<TensorReverseOp<ReverseDimensions, ArgType>, Device>
template <int StoreMode> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketReturnType& x) {
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
// This code is pilfered from TensorMorphing.h

5
unsupported/Eigen/CXX11/src/Tensor/TensorShuffling.h
View File

@ -227,8 +227,7 @@ struct TensorEvaluator<const TensorShufflingOp<Shuffle, ArgType>, Device>
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index + PacketSize - 1 < dimensions().TotalSize());
eigen_assert(index + PacketSize - 1 < dimensions().TotalSize());
return PacketLoader<LoadMode, Self, TensorEvaluator<ArgType, Device>::PacketAccess>::Run(*this, index);
}
@ -399,8 +398,6 @@ struct TensorEvaluator<TensorShufflingOp<Shuffle, ArgType>, Device>
template <int StoreMode> EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketReturnType& x)
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
EIGEN_ALIGN_MAX typename internal::remove_const<CoeffReturnType>::type values[PacketSize];
internal::pstore<CoeffReturnType, PacketReturnType>(values, x);
EIGEN_UNROLL_LOOP

2
unsupported/Eigen/CXX11/src/Tensor/TensorTrace.h
View File

@ -245,8 +245,6 @@ struct TensorEvaluator<const TensorTraceOp<Dims, ArgType>, Device>
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const {
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE);
eigen_assert(index + PacketSize - 1 < dimensions().TotalSize());
EIGEN_ALIGN_MAX typename internal::remove_const<CoeffReturnType>::type values[PacketSize];

1
unsupported/Eigen/CXX11/src/Tensor/TensorVolumePatch.h
View File

@ -421,7 +421,6 @@ struct TensorEvaluator<const TensorVolumePatchOp<Planes, Rows, Cols, ArgType>, D
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
{
EIGEN_STATIC_ASSERT((PacketSize > 1), YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
if (m_in_row_strides != 1 || m_in_col_strides != 1 || m_row_inflate_strides != 1 || m_col_inflate_strides != 1 ||

55
unsupported/test/cxx11_tensor_executor.cpp
View File

@ -612,43 +612,42 @@ static void test_async_execute_binary_expr(Device d)
}
}
#ifdef EIGEN_DONT_VECTORIZE
#define VECTORIZABLE(VAL) !EIGEN_DONT_VECTORIZE && VAL
#else
#define VECTORIZABLE(VAL) VAL
#ifndef EIGEN_DONT_VECTORIZE
#define EIGEN_DONT_VECTORIZE 0
#endif
#define VECTORIZABLE(T, VAL) !EIGEN_DONT_VECTORIZE && Eigen::internal::packet_traits<T>::Vectorizable && VAL
#define CALL_SUBTEST_PART(PART) \
CALL_SUBTEST_##PART
#define CALL_SUBTEST_COMBINATIONS(PART, NAME, T, NUM_DIMS) \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, false, TiledEvaluation::Off, ColMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, false, TiledEvaluation::On, ColMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, VECTORIZABLE(true), TiledEvaluation::Off, ColMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, VECTORIZABLE(true), TiledEvaluation::On, ColMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, false, TiledEvaluation::Off, RowMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, false, TiledEvaluation::On, RowMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, VECTORIZABLE(true), TiledEvaluation::Off, RowMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, VECTORIZABLE(true), TiledEvaluation::On, RowMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::Off, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::On, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(true), TiledEvaluation::Off, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(true), TiledEvaluation::On, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::Off, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::On, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(true), TiledEvaluation::Off, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(true), TiledEvaluation::On, RowMajor>(tp_device)))
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, false, TiledEvaluation::Off, ColMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, false, TiledEvaluation::On, ColMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, VECTORIZABLE(T, true), TiledEvaluation::Off, ColMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, VECTORIZABLE(T, true), TiledEvaluation::On, ColMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, false, TiledEvaluation::Off, RowMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, false, TiledEvaluation::On, RowMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, VECTORIZABLE(T, true), TiledEvaluation::Off, RowMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, DefaultDevice, VECTORIZABLE(T, true), TiledEvaluation::On, RowMajor>(default_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::Off, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::On, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(T, true), TiledEvaluation::Off, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(T, true), TiledEvaluation::On, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::Off, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::On, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(T, true), TiledEvaluation::Off, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(T, true), TiledEvaluation::On, RowMajor>(tp_device)))
// NOTE: Currently only ThreadPoolDevice supports async expression evaluation.
#define CALL_ASYNC_SUBTEST_COMBINATIONS(PART, NAME, T, NUM_DIMS) \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::Off, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::On, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(true), TiledEvaluation::Off, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(true), TiledEvaluation::On, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::Off, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::On, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(true), TiledEvaluation::Off, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(true), TiledEvaluation::On, RowMajor>(tp_device)))
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::Off, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::On, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(T, true), TiledEvaluation::Off, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(T, true), TiledEvaluation::On, ColMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::Off, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, false, TiledEvaluation::On, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(T, true), TiledEvaluation::Off, RowMajor>(tp_device))); \
CALL_SUBTEST_PART(PART)((NAME<T, NUM_DIMS, ThreadPoolDevice, VECTORIZABLE(T, true), TiledEvaluation::On, RowMajor>(tp_device)))
EIGEN_DECLARE_TEST(cxx11_tensor_executor) {
Eigen::DefaultDevice default_device;