Ginkgo Generated from develop branch based on develop. Ginkgo version 1.10.0
A numerical linear algebra library targeting many-core architectures
 
Loading...
Searching...
No Matches
vector.hpp
1// SPDX-FileCopyrightText: 2017 - 2025 The Ginkgo authors
2//
3// SPDX-License-Identifier: BSD-3-Clause
4
5#ifndef GKO_PUBLIC_CORE_DISTRIBUTED_VECTOR_HPP_
6#define GKO_PUBLIC_CORE_DISTRIBUTED_VECTOR_HPP_
7
8
9#include <ginkgo/config.hpp>
10
11
12#if GINKGO_BUILD_MPI
13
14
15#include <ginkgo/core/base/dense_cache.hpp>
16#include <ginkgo/core/base/lin_op.hpp>
17#include <ginkgo/core/base/mpi.hpp>
18#include <ginkgo/core/distributed/base.hpp>
19#include <ginkgo/core/matrix/dense.hpp>
20
21
22namespace gko {
23namespace experimental {
24namespace distributed {
25namespace detail {
26
27
28template <typename ValueType>
29class VectorCache;
30
31
32} // namespace detail
33
34
35template <typename LocalIndexType, typename GlobalIndexType>
36class Partition;
37
38
66template <typename ValueType = double>
67class Vector
68 : public EnableLinOp<Vector<ValueType>>,
69 public ConvertibleTo<Vector<next_precision<ValueType>>>,
70#if GINKGO_ENABLE_HALF
71 public ConvertibleTo<Vector<next_precision<next_precision<ValueType>>>>,
72#endif
73 public EnableAbsoluteComputation<remove_complex<Vector<ValueType>>>,
74 public DistributedBase {
75 friend class EnablePolymorphicObject<Vector, LinOp>;
76 friend class Vector<to_complex<ValueType>>;
77 friend class Vector<remove_complex<ValueType>>;
78 friend class Vector<previous_precision<ValueType>>;
79 friend class detail::VectorCache<ValueType>;
80
81public:
82 using EnableLinOp<Vector>::convert_to;
83 using EnableLinOp<Vector>::move_to;
84 using ConvertibleTo<Vector<next_precision<ValueType>>>::convert_to;
85 using ConvertibleTo<Vector<next_precision<ValueType>>>::move_to;
86
87 using value_type = ValueType;
88 using absolute_type = remove_complex<Vector>;
89 using real_type = absolute_type;
90 using complex_type = Vector<to_complex<value_type>>;
91 using local_vector_type = gko::matrix::Dense<value_type>;
92
99 static std::unique_ptr<Vector> create_with_config_of(
101
102
114 static std::unique_ptr<Vector> create_with_type_of(
115 ptr_param<const Vector> other, std::shared_ptr<const Executor> exec);
116
129 static std::unique_ptr<Vector> create_with_type_of(
130 ptr_param<const Vector> other, std::shared_ptr<const Executor> exec,
131 const dim<2>& global_size, const dim<2>& local_size, size_type stride);
132
148 ptr_param<const Partition<int64, int64>> partition);
149
151 ptr_param<const Partition<int32, int64>> partition);
152
154 ptr_param<const Partition<int32, int32>> partition);
155
166 ptr_param<const Partition<int64, int64>> partition);
167
169 ptr_param<const Partition<int32, int64>> partition);
170
172 ptr_param<const Partition<int32, int32>> partition);
173
174 void convert_to(Vector<next_precision<ValueType>>* result) const override;
175
176 void move_to(Vector<next_precision<ValueType>>* result) override;
177
178#if GINKGO_ENABLE_HALF
179 friend class Vector<previous_precision<previous_precision<ValueType>>>;
180 using ConvertibleTo<
181 Vector<next_precision<next_precision<ValueType>>>>::convert_to;
182 using ConvertibleTo<
183 Vector<next_precision<next_precision<ValueType>>>>::move_to;
184
185 void convert_to(Vector<next_precision<next_precision<ValueType>>>* result)
186 const override;
187
188 void move_to(
189 Vector<next_precision<next_precision<ValueType>>>* result) override;
190#endif
191
192 std::unique_ptr<absolute_type> compute_absolute() const override;
193
195
200 std::unique_ptr<complex_type> make_complex() const;
201
208
213 std::unique_ptr<real_type> get_real() const;
214
218 void get_real(ptr_param<real_type> result) const;
219
224 std::unique_ptr<real_type> get_imag() const;
225
230 void get_imag(ptr_param<real_type> result) const;
231
237 void fill(ValueType value);
238
249
260
271
281
292
306 array<char>& tmp) const;
307
318 ptr_param<LinOp> result) const;
319
333 array<char>& tmp) const;
334
344
357
366 void compute_norm2(ptr_param<LinOp> result) const;
367
380
388 void compute_norm1(ptr_param<LinOp> result) const;
389
402
411 void compute_mean(ptr_param<LinOp> result) const;
412
424 void compute_mean(ptr_param<LinOp> result, array<char>& tmp) const;
425
436 value_type& at_local(size_type row, size_type col) noexcept;
437
441 value_type at_local(size_type row, size_type col) const noexcept;
442
457 ValueType& at_local(size_type idx) noexcept;
458
462 ValueType at_local(size_type idx) const noexcept;
463
469 value_type* get_local_values();
470
478 const value_type* get_const_local_values() const;
479
485 const local_vector_type* get_local_vector() const;
486
494 std::unique_ptr<const real_type> create_real_view() const;
495
499 std::unique_ptr<real_type> create_real_view();
500
501 size_type get_stride() const noexcept { return local_.get_stride(); }
502
514 static std::unique_ptr<Vector> create(std::shared_ptr<const Executor> exec,
516 dim<2> global_size, dim<2> local_size,
517 size_type stride);
518
530 static std::unique_ptr<Vector> create(std::shared_ptr<const Executor> exec,
532 dim<2> global_size = {},
533 dim<2> local_size = {});
534
552 static std::unique_ptr<Vector> create(
553 std::shared_ptr<const Executor> exec, mpi::communicator comm,
554 dim<2> global_size, std::unique_ptr<local_vector_type> local_vector);
555
574 static std::unique_ptr<Vector> create(
575 std::shared_ptr<const Executor> exec, mpi::communicator comm,
576 std::unique_ptr<local_vector_type> local_vector);
577
590 static std::unique_ptr<const Vector> create_const(
591 std::shared_ptr<const Executor> exec, mpi::communicator comm,
592 dim<2> global_size,
593 std::unique_ptr<const local_vector_type> local_vector);
594
607 static std::unique_ptr<const Vector> create_const(
608 std::shared_ptr<const Executor> exec, mpi::communicator comm,
609 std::unique_ptr<const local_vector_type> local_vector);
610
611protected:
612 Vector(std::shared_ptr<const Executor> exec, mpi::communicator comm,
613 dim<2> global_size, dim<2> local_size, size_type stride);
614
615 explicit Vector(std::shared_ptr<const Executor> exec,
616 mpi::communicator comm, dim<2> global_size = {},
617 dim<2> local_size = {});
618
619 Vector(std::shared_ptr<const Executor> exec, mpi::communicator comm,
620 dim<2> global_size, std::unique_ptr<local_vector_type> local_vector);
621
622 Vector(std::shared_ptr<const Executor> exec, mpi::communicator comm,
623 std::unique_ptr<local_vector_type> local_vector);
624
625 void resize(dim<2> global_size, dim<2> local_size);
626
627 template <typename LocalIndexType, typename GlobalIndexType>
628 void read_distributed_impl(
630 const Partition<LocalIndexType, GlobalIndexType>* partition);
631
632 void apply_impl(const LinOp*, LinOp*) const override;
633
634 void apply_impl(const LinOp*, const LinOp*, const LinOp*,
635 LinOp*) const override;
636
643 virtual std::unique_ptr<Vector> create_with_same_config() const;
644
657 virtual std::unique_ptr<Vector> create_with_type_of_impl(
658 std::shared_ptr<const Executor> exec, const dim<2>& global_size,
659 const dim<2>& local_size, size_type stride) const;
660
661private:
662 local_vector_type local_;
663 ::gko::detail::DenseCache<ValueType> host_reduction_buffer_;
664 ::gko::detail::DenseCache<remove_complex<ValueType>> host_norm_buffer_;
665};
666
667
668} // namespace distributed
669} // namespace experimental
670
671
672namespace detail {
673
674
675template <typename TargetType>
676struct conversion_target_helper;
677
678
688template <typename ValueType>
689struct conversion_target_helper<experimental::distributed::Vector<ValueType>> {
690 using target_type = experimental::distributed::Vector<ValueType>;
691 using source_type =
692 experimental::distributed::Vector<previous_precision<ValueType>>;
693
694 static std::unique_ptr<target_type> create_empty(const source_type* source)
695 {
696 return target_type::create(source->get_executor(),
697 source->get_communicator());
698 }
699
700 // Allow to create_empty of the same type
701 // For distributed case, next<next<V>> will be V in the candidate list.
702 // TODO: decide to whether to add this or add condition to the list
703 static std::unique_ptr<target_type> create_empty(const target_type* source)
704 {
705 return target_type::create(source->get_executor(),
706 source->get_communicator());
707 }
708
709#if GINKGO_ENABLE_HALF
710 using snd_source_type = experimental::distributed::Vector<
711 previous_precision<previous_precision<ValueType>>>;
712
713 static std::unique_ptr<target_type> create_empty(
714 const snd_source_type* source)
715 {
716 return target_type::create(source->get_executor(),
717 source->get_communicator());
718 }
719#endif
720};
721
722
723} // namespace detail
724} // namespace gko
725
726
727#endif // GINKGO_BUILD_MPI
728
729
730#endif // GKO_PUBLIC_CORE_DISTRIBUTED_VECTOR_HPP_
ConvertibleTo interface is used to mark that the implementer can be converted to the object of Result...
Definition polymorphic_object.hpp:479
The EnableAbsoluteComputation mixin provides the default implementations of compute_absolute_linop an...
Definition lin_op.hpp:794
The EnableLinOp mixin can be used to provide sensible default implementations of the majority of the ...
Definition lin_op.hpp:879
This mixin inherits from (a subclass of) PolymorphicObject and provides a base implementation of a ne...
Definition polymorphic_object.hpp:668
Definition lin_op.hpp:117
LinOp(const LinOp &)=default
Copy-constructs a LinOp.
An array is a container which encapsulates fixed-sized arrays, stored on the Executor tied to the arr...
Definition array.hpp:166
This type is a device-side equivalent to matrix_data.
Definition device_matrix_data.hpp:36
Represents a partition of a range of indices [0, size) into a disjoint set of parts.
Definition partition.hpp:83
value_type at_local(size_type row, size_type col) const noexcept
void compute_mean(ptr_param< LinOp > result) const
Computes the column-wise mean of this (multi-)vector using a global reduction.
static std::unique_ptr< Vector > create(std::shared_ptr< const Executor > exec, mpi::communicator comm, dim< 2 > global_size, dim< 2 > local_size, size_type stride)
Creates an empty distributed vector with a specified size.
void compute_norm2(ptr_param< LinOp > result) const
Computes the Euclidean (L^2) norm of this (multi-)vector using a global reduction.
void read_distributed(const matrix_data< ValueType, int64 > &data, ptr_param< const Partition< int64, int64 > > partition)
Reads a vector from the matrix_data structure and a global row partition.
void make_complex(ptr_param< complex_type > result) const
Writes a complex copy of the original vectors to given complex vectors.
std::unique_ptr< real_type > create_real_view()
Create a real view of the (potentially) complex original multi-vector.
void compute_squared_norm2(ptr_param< LinOp > result, array< char > &tmp) const
Computes the square of the column-wise Euclidean ( ) norm of this (multi-)vector using a global reduc...
static std::unique_ptr< Vector > create(std::shared_ptr< const Executor > exec, mpi::communicator comm, dim< 2 > global_size={}, dim< 2 > local_size={})
Creates an empty distributed vector with a specified size.
std::unique_ptr< real_type > get_real() const
Creates new real vectors and extracts the real part of the original vectors into that.
static std::unique_ptr< const Vector > create_const(std::shared_ptr< const Executor > exec, mpi::communicator comm, dim< 2 > global_size, std::unique_ptr< const local_vector_type > local_vector)
Creates a constant (immutable) distributed Vector from a constant local vector.
std::unique_ptr< const real_type > create_real_view() const
Create a real view of the (potentially) complex original multi-vector.
void fill(ValueType value)
Fill the distributed vectors with a given value.
static std::unique_ptr< Vector > create_with_type_of(ptr_param< const Vector > other, std::shared_ptr< const Executor > exec, const dim< 2 > &global_size, const dim< 2 > &local_size, size_type stride)
Creates an Vector with the same type as another Vector, but on a different executor and with a differ...
static std::unique_ptr< Vector > create_with_config_of(ptr_param< const Vector > other)
Creates a distributed Vector with the same size and stride as another Vector.
value_type & at_local(size_type row, size_type col) noexcept
Returns a single element of the multi-vector.
value_type * get_local_values()
Returns a pointer to the array of local values of the multi-vector.
void compute_norm2(ptr_param< LinOp > result, array< char > &tmp) const
Computes the Euclidean (L^2) norm of this (multi-)vector using a global reduction.
const value_type * get_const_local_values() const
Returns a pointer to the array of local values of the multi-vector.
void compute_conj_dot(ptr_param< const LinOp > b, ptr_param< LinOp > result) const
Computes the column-wise dot product of this (multi-)vector and conj(b) using a global reduction.
void compute_absolute_inplace() override
Compute absolute inplace on each element.
void get_real(ptr_param< real_type > result) const
Extracts the real part of the original vectors into given real vectors.
void compute_squared_norm2(ptr_param< LinOp > result) const
Computes the square of the column-wise Euclidean ( ) norm of this (multi-)vector using a global reduc...
void sub_scaled(ptr_param< const LinOp > alpha, ptr_param< const LinOp > b)
Subtracts b scaled by alpha from the vectors (aka: BLAS axpy).
void compute_dot(ptr_param< const LinOp > b, ptr_param< LinOp > result) const
Computes the column-wise dot product of this (multi-)vector and b using a global reduction.
static std::unique_ptr< const Vector > create_const(std::shared_ptr< const Executor > exec, mpi::communicator comm, std::unique_ptr< const local_vector_type > local_vector)
Creates a constant (immutable) distributed Vector from a constant local vector.
void compute_mean(ptr_param< LinOp > result, array< char > &tmp) const
Computes the column-wise arithmetic mean of this (multi-)vector using a global reduction.
std::unique_ptr< complex_type > make_complex() const
Creates a complex copy of the original vectors.
void compute_norm1(ptr_param< LinOp > result) const
Computes the column-wise (L^1) norm of this (multi-)vector.
static std::unique_ptr< Vector > create(std::shared_ptr< const Executor > exec, mpi::communicator comm, dim< 2 > global_size, std::unique_ptr< local_vector_type > local_vector)
Creates a distributed vector from local vectors with a specified size.
void get_imag(ptr_param< real_type > result) const
Extracts the imaginary part of the original vectors into given real vectors.
std::unique_ptr< absolute_type > compute_absolute() const override
Gets the AbsoluteLinOp.
void compute_dot(ptr_param< const LinOp > b, ptr_param< LinOp > result, array< char > &tmp) const
Computes the column-wise dot product of this (multi-)vector and b using a global reduction.
std::unique_ptr< real_type > get_imag() const
Creates new real vectors and extracts the imaginary part of the original vectors into that.
void inv_scale(ptr_param< const LinOp > alpha)
Scales the vectors with the inverse of a scalar.
void add_scaled(ptr_param< const LinOp > alpha, ptr_param< const LinOp > b)
Adds b scaled by alpha to the vectors (aka: BLAS axpy).
void scale(ptr_param< const LinOp > alpha)
Scales the vectors with a scalar (aka: BLAS scal).
void read_distributed(const device_matrix_data< ValueType, int64 > &data, ptr_param< const Partition< int64, int64 > > partition)
Reads a vector from the device_matrix_data structure and a global row partition.
const local_vector_type * get_local_vector() const
Direct (read) access to the underlying local local_vector_type vectors.
static std::unique_ptr< Vector > create_with_type_of(ptr_param< const Vector > other, std::shared_ptr< const Executor > exec)
Creates an empty Vector with the same type as another Vector, but on a different executor.
ValueType & at_local(size_type idx) noexcept
Returns a single element of the multi-vector.
void compute_norm1(ptr_param< LinOp > result, array< char > &tmp) const
Computes the column-wise (L^1) norm of this (multi-)vector using a global reduction.
ValueType at_local(size_type idx) const noexcept
static std::unique_ptr< Vector > create(std::shared_ptr< const Executor > exec, mpi::communicator comm, std::unique_ptr< local_vector_type > local_vector)
Creates a distributed vector from local vectors.
void compute_conj_dot(ptr_param< const LinOp > b, ptr_param< LinOp > result, array< char > &tmp) const
Computes the column-wise dot product of this (multi-)vector and conj(b) using a global reduction.
A thin wrapper of MPI_Comm that supports most MPI calls.
Definition mpi.hpp:416
Dense is a matrix format which explicitly stores all values of the matrix.
Definition dense.hpp:117
This class is used for function parameters in the place of raw pointers.
Definition utils_helper.hpp:41
The distributed namespace.
Definition polymorphic_object.hpp:19
The Ginkgo namespace.
Definition abstract_factory.hpp:20
typename detail::remove_complex_s< T >::type remove_complex
Obtain the type which removed the complex of complex/scalar type or the template parameter of class b...
Definition math.hpp:260
typename detail::to_complex_s< T >::type to_complex
Obtain the type which adds the complex of complex/scalar type or the template parameter of class by a...
Definition math.hpp:279
std::size_t size_type
Integral type used for allocation quantities.
Definition types.hpp:89
next_precision_base< T > next_precision
Obtains the next type in the singly-linked precision list with half.
Definition math.hpp:445
A type representing the dimensions of a multidimensional object.
Definition dim.hpp:26
This structure is used as an intermediate data type to store a sparse matrix.
Definition matrix_data.hpp:126