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

Replace Eigen type metaprogramming with corresponding std types and make use of alias templates

Browse Source
This commit is contained in:
Erik Schultheis
2022-03-16 16:43:40 +00:00
committed by Antonio Sánchez
parent 514f90c9ff
commit 421cbf0866
191 changed files with 1147 additions and 1221 deletions
Download Patch File Download Diff File Expand all files Collapse all files

47
unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h
View File

@@ -158,7 +158,7 @@ enum class data_source { global_mem, local_mem, private_mem };
*/
template <bool PacketLoad, bool is_coalesced_layout, bool, typename PacketType, typename TensorMapper,
typename StorageIndex>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename ::Eigen::internal::enable_if<PacketLoad, PacketType>::type read(
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::enable_if_t<PacketLoad, PacketType> read(
const TensorMapper &tensorMapper, const StorageIndex &NCIndex, const StorageIndex &CIndex, const StorageIndex &ld) {
const StorageIndex row = (is_coalesced_layout) ? NCIndex : CIndex;
const StorageIndex col = (is_coalesced_layout) ? CIndex : NCIndex;
@@ -188,7 +188,7 @@ static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename ::Eigen::internal::enable_
* \param CIndex: is the contracting dim index
*/
template <bool PacketLoad, bool, bool IsRhs, typename PacketType, typename TensorMapper, typename StorageIndex>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename ::Eigen::internal::enable_if<!PacketLoad, PacketType>::type read(
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::enable_if_t<!PacketLoad, PacketType> read(
const TensorMapper &tensorMapper, const StorageIndex &NCIndex, const StorageIndex &CIndex, const StorageIndex &) {
const StorageIndex row = (IsRhs) ? CIndex : NCIndex;
const StorageIndex col = (IsRhs) ? NCIndex : CIndex;
@@ -218,7 +218,7 @@ static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename ::Eigen::internal::enable_
template <typename StorageIndex, StorageIndex ld, data_source dt, typename PacketType, typename DataScalar>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
typename ::Eigen::internal::enable_if<dt != data_source::global_mem, void>::type
std::enable_if_t<dt != data_source::global_mem, void>
write(PacketType &packet_data, DataScalar ptr) {
EIGEN_CONSTEXPR int PacketSize = Eigen::internal::unpacket_traits<PacketType>::size;
EIGEN_UNROLL_LOOP
@@ -244,8 +244,8 @@ static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
*/
template <data_source dt, typename PacketType, typename DataScalar>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename ::Eigen::internal::enable_if<
Eigen::internal::unpacket_traits<PacketType>::size != 1 && dt == data_source::global_mem, void>::type
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename std::enable_if_t<
Eigen::internal::unpacket_traits<PacketType>::size != 1 && dt == data_source::global_mem, void>
write(PacketType &packet_data, DataScalar *ptr) {
::Eigen::internal::pstoreu<DataScalar, PacketType>(ptr, packet_data);
}
@@ -264,8 +264,8 @@ write(PacketType &packet_data, DataScalar *ptr) {
* \param ptr: a pointer to the local memory
*/
template <data_source dt, typename PacketType, typename DataScalar>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename ::Eigen::internal::enable_if<
Eigen::internal::unpacket_traits<PacketType>::size == 1 && dt == data_source::global_mem, void>::type
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename std::enable_if_t<
Eigen::internal::unpacket_traits<PacketType>::size == 1 && dt == data_source::global_mem, void>
write(PacketType &packet_data, DataScalar *ptr) {
*ptr = packet_data;
}
@@ -321,7 +321,7 @@ struct BlockProperties {
static EIGEN_CONSTEXPR bool packet_load = packet_load_;
typedef typename Eigen::internal::unpacket_traits<PacketType>::type OutScalar;
static EIGEN_CONSTEXPR bool is_rhs = is_rhs_;
typedef typename Eigen::internal::conditional<packet_load, PacketType, OutScalar>::type OutType;
typedef Eigen::std::conditional_t<packet_load, PacketType, OutScalar> OutType;
static EIGEN_CONSTEXPR int elements_per_access = Eigen::internal::unpacket_traits<OutType>::size;
static EIGEN_CONSTEXPR bool is_coalesced_layout = !(is_transposed ^ is_rhs);
static EIGEN_CONSTEXPR int nc_stride = (is_coalesced_layout ? elements_per_access : 1);
@@ -477,8 +477,7 @@ class TensorContractionKernel {
typedef cl::sycl::accessor<OutScalar, 1, cl::sycl::access::mode::read_write, cl::sycl::access::target::local> Scratch;
typedef cl::sycl::multi_ptr<OutScalar, cl::sycl::access::address_space::local_space> local_ptr;
typedef OutScalar * /*cl::sycl::multi_ptr<OutScalar, cl::sycl::access::address_space::private_space>*/ private_ptr;
typedef
typename ::Eigen::internal::conditional<contraction_tp == contraction_type::local, local_ptr, private_ptr>::type
typedef std::conditional_t<contraction_tp == contraction_type::local, local_ptr, private_ptr>
tile_ptr;
static EIGEN_CONSTEXPR StorageIndex LSDL = contraction_tp == contraction_type::local
? Properties::TileSizeDimM + Properties::BC
@@ -544,7 +543,7 @@ class TensorContractionKernel {
template <contraction_type tp = contraction_tp>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
TiledMemory(const ThreadProperties<StorageIndex> &, local_ptr,
typename ::Eigen::internal::enable_if<tp == contraction_type::no_local>::type * = 0)
std::enable_if_t<tp == contraction_type::no_local> * = 0)
: lhs_scratch_extract{},
rhs_scratch_extract{},
lhs_scratch_ptr_compute(lhs_scratch_extract.ptr),
@@ -555,7 +554,7 @@ class TensorContractionKernel {
template <contraction_type tp = contraction_tp>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
TiledMemory(const ThreadProperties<StorageIndex> &thread_properties, local_ptr block_start_ptr,
typename ::Eigen::internal::enable_if<tp == contraction_type::local>::type * = 0)
std::enable_if_t<tp == contraction_type::local> * = 0)
: lhs_scratch_extract{block_start_ptr},
rhs_scratch_extract{lhs_scratch_extract.ptr +
((Properties::DoubleBuffer + 1) * LSDL * Properties::TileSizeDimK)},
@@ -712,7 +711,7 @@ class TensorContractionKernel {
template <typename InputBlockProperties, bool is_internal_block, typename Input, typename PrivateReg,
contraction_type contract_tp = contraction_tp>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
typename ::Eigen::internal::enable_if<contract_tp == contraction_type::no_local>::type
std::enable_if_t<contract_tp == contraction_type::no_local>
extract_block(const Input &inpt, PrivateReg private_ptr, const std::pair<StorageIndex, StorageIndex> &,
const StorageIndex &ncOffset, const StorageIndex cOffset) {
EIGEN_CONSTEXPR StorageIndex LocalThreadSizeNC =
@@ -785,28 +784,28 @@ class TensorContractionKernel {
template <bool db = Properties::DoubleBuffer, contraction_type ctp = contraction_tp>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
typename ::Eigen::internal::enable_if<db && ctp == contraction_type::local>::type
std::enable_if_t<db && ctp == contraction_type::local>
sync_mem(const cl::sycl::nd_item<1> &, bool &db_offset) noexcept {
db_offset = !db_offset;
}
template <bool db = Properties::DoubleBuffer, contraction_type ctp = contraction_tp>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
typename ::Eigen::internal::enable_if<!db && ctp == contraction_type::local>::type
std::enable_if_t<!db && ctp == contraction_type::local>
sync_mem(const cl::sycl::nd_item<1> &itemID, bool &) noexcept {
itemID.barrier(cl::sycl::access::fence_space::local_space);
}
template <contraction_type ctp = contraction_tp>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
typename ::Eigen::internal::enable_if<ctp == contraction_type::no_local>::type
std::enable_if_t<ctp == contraction_type::no_local>
sync_mem(const cl::sycl::nd_item<1> &, bool &) noexcept {
return;
}
template <bool need_sync, contraction_type ctp = contraction_tp>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
typename ::Eigen::internal::enable_if<need_sync && ctp == contraction_type::no_local>::type
std::enable_if_t<need_sync && ctp == contraction_type::no_local>
sync_thread(const cl::sycl::nd_item<1> &
#ifdef EIGEN_SYCL_ARM_GPU_CACHE_OPTIMISATION
itemID
@@ -820,12 +819,12 @@ class TensorContractionKernel {
}
template <bool need_sync, contraction_type ctp = contraction_tp>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
typename ::Eigen::internal::enable_if<need_sync && ctp == contraction_type::local>::type
std::enable_if_t<need_sync && ctp == contraction_type::local>
sync_thread(const cl::sycl::nd_item<1> &itemID) {
itemID.barrier(cl::sycl::access::fence_space::local_space);
}
template <bool need_sync>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename ::Eigen::internal::enable_if<!need_sync>::type sync_thread(
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::enable_if_t<!need_sync> sync_thread(
const cl::sycl::nd_item<1> &) {
return;
}
@@ -896,7 +895,7 @@ class TensorContractionKernel {
template <typename InputBlockProperties, bool is_internal_block, typename Input, typename Local,
contraction_type contract_tp = contraction_tp>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
typename ::Eigen::internal::enable_if<contract_tp == contraction_type::local>::type
std::enable_if_t<contract_tp == contraction_type::local>
extract_block(const Input &inpt, Local local_ptr, const std::pair<StorageIndex, StorageIndex>& local_index,
const StorageIndex &ncOffset, const StorageIndex cOffset) {
EIGEN_CONSTEXPR StorageIndex TileSizeDimNC =
@@ -1294,7 +1293,7 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
typedef TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgType, OutputKernelType>, Device> Self;
typedef TensorContractionEvaluatorBase<Self> Base;
typedef TensorContractionOp<Indices, LeftArgType, RightArgType, OutputKernelType> XprType;
typedef typename internal::remove_const<typename XprType::Scalar>::type Scalar;
typedef std::remove_const_t<typename XprType::Scalar> Scalar;
typedef typename XprType::Index StorageIndex;
typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
@@ -1307,11 +1306,11 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
TripleDim(const StorageIndex M_, const StorageIndex N_, const StorageIndex K_) : M(M_), N(N_), K(K_) {}
};
enum {
Layout = TensorEvaluator<LeftArgType, Device>::Layout,
PacketAccess = (PacketType<CoeffReturnType, Device>::size > 1),
BlockAccess = false,
};
static constexpr int Layout = TensorEvaluator<LeftArgType, Device>::Layout;
static EIGEN_CONSTEXPR int LDims = Base::LDims;
static EIGEN_CONSTEXPR int RDims = Base::RDims;
static EIGEN_CONSTEXPR int ContractDims = Base::ContractDims;
@@ -1329,8 +1328,8 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
typedef TensorEvaluator<typename Base::EvalLeftArgType, Device> LeftEvaluator;
typedef TensorEvaluator<typename Base::EvalRightArgType, Device> RightEvaluator;
typedef typename Eigen::internal::remove_const<typename LeftEvaluator::CoeffReturnType>::type LhsScalar;
typedef typename Eigen::internal::remove_const<typename RightEvaluator::CoeffReturnType>::type RhsScalar;
typedef std::remove_const_t<typename LeftEvaluator::CoeffReturnType> LhsScalar;
typedef std::remove_const_t<typename RightEvaluator::CoeffReturnType> RhsScalar;
typedef typename LeftEvaluator::Dimensions LeftDimensions;
typedef typename RightEvaluator::Dimensions RightDimensions;