Ginkgo Generated from branch based on main. Ginkgo version 1.11.0
A numerical linear algebra library targeting many-core architectures
Loading...
Searching...
No Matches
sellp.hpp
1// SPDX-FileCopyrightText: 2017 - 2025 The Ginkgo authors
2//
3// SPDX-License-Identifier: BSD-3-Clause
4
5#ifndef GKO_PUBLIC_CORE_MATRIX_SELLP_HPP_
6#define GKO_PUBLIC_CORE_MATRIX_SELLP_HPP_
7
8
9#include <ginkgo/core/base/array.hpp>
10#include <ginkgo/core/base/lin_op.hpp>
11
12
13namespace gko {
14namespace matrix {
15
16
17constexpr int default_slice_size = 64;
18constexpr int default_stride_factor = 1;
19
20
21template <typename ValueType>
22class Dense;
23
24template <typename ValueType, typename IndexType>
25class Csr;
26
42template <typename ValueType = default_precision, typename IndexType = int32>
43class Sellp
44 : public EnableLinOp<Sellp<ValueType, IndexType>>,
45 public ConvertibleTo<Sellp<next_precision<ValueType>, IndexType>>,
46#if GINKGO_ENABLE_HALF || GINKGO_ENABLE_BFLOAT16
47 public ConvertibleTo<Sellp<next_precision<ValueType, 2>, IndexType>>,
48#endif
49#if GINKGO_ENABLE_HALF && GINKGO_ENABLE_BFLOAT16
50 public ConvertibleTo<Sellp<next_precision<ValueType, 3>, IndexType>>,
51#endif
52 public ConvertibleTo<Dense<ValueType>>,
53 public ConvertibleTo<Csr<ValueType, IndexType>>,
54 public DiagonalExtractable<ValueType>,
55 public ReadableFromMatrixData<ValueType, IndexType>,
56 public WritableToMatrixData<ValueType, IndexType>,
58 remove_complex<Sellp<ValueType, IndexType>>> {
59 friend class EnablePolymorphicObject<Sellp, LinOp>;
60 friend class Dense<ValueType>;
61 friend class Csr<ValueType, IndexType>;
62 friend class Sellp<to_complex<ValueType>, IndexType>;
63 GKO_ASSERT_SUPPORTED_VALUE_AND_INDEX_TYPE;
64
65public:
66 using EnableLinOp<Sellp>::convert_to;
67 using EnableLinOp<Sellp>::move_to;
68 using ConvertibleTo<
69 Sellp<next_precision<ValueType>, IndexType>>::convert_to;
70 using ConvertibleTo<Sellp<next_precision<ValueType>, IndexType>>::move_to;
71 using ConvertibleTo<Dense<ValueType>>::convert_to;
72 using ConvertibleTo<Dense<ValueType>>::move_to;
73 using ConvertibleTo<Csr<ValueType, IndexType>>::convert_to;
74 using ConvertibleTo<Csr<ValueType, IndexType>>::move_to;
75 using ReadableFromMatrixData<ValueType, IndexType>::read;
76
77 using value_type = ValueType;
78 using index_type = IndexType;
79 using mat_data = matrix_data<ValueType, IndexType>;
80 using device_mat_data = device_matrix_data<ValueType, IndexType>;
81 using absolute_type = remove_complex<Sellp>;
82
83 friend class Sellp<previous_precision<ValueType>, IndexType>;
84
85 void convert_to(
86 Sellp<next_precision<ValueType>, IndexType>* result) const override;
87
88 void move_to(Sellp<next_precision<ValueType>, IndexType>* result) override;
89
90#if GINKGO_ENABLE_HALF || GINKGO_ENABLE_BFLOAT16
91 friend class Sellp<previous_precision<ValueType, 2>, IndexType>;
92 using ConvertibleTo<
93 Sellp<next_precision<ValueType, 2>, IndexType>>::convert_to;
94 using ConvertibleTo<
95 Sellp<next_precision<ValueType, 2>, IndexType>>::move_to;
96
97 void convert_to(
98 Sellp<next_precision<ValueType, 2>, IndexType>* result) const override;
99
100 void move_to(
101 Sellp<next_precision<ValueType, 2>, IndexType>* result) override;
102#endif
103
104#if GINKGO_ENABLE_HALF && GINKGO_ENABLE_BFLOAT16
105 friend class Sellp<previous_precision<ValueType, 3>, IndexType>;
106 using ConvertibleTo<
107 Sellp<next_precision<ValueType, 3>, IndexType>>::convert_to;
108 using ConvertibleTo<
109 Sellp<next_precision<ValueType, 3>, IndexType>>::move_to;
110
111 void convert_to(
112 Sellp<next_precision<ValueType, 3>, IndexType>* result) const override;
113
114 void move_to(
115 Sellp<next_precision<ValueType, 3>, IndexType>* result) override;
116#endif
117
118 void convert_to(Dense<ValueType>* other) const override;
119
120 void move_to(Dense<ValueType>* other) override;
121
122 void convert_to(Csr<ValueType, IndexType>* other) const override;
123
124 void move_to(Csr<ValueType, IndexType>* other) override;
125
126 void read(const mat_data& data) override;
127
128 void read(const device_mat_data& data) override;
129
130 void read(device_mat_data&& data) override;
131
132 void write(mat_data& data) const override;
133
134 std::unique_ptr<Diagonal<ValueType>> extract_diagonal() const override;
135
136 std::unique_ptr<absolute_type> compute_absolute() const override;
137
139
145 value_type* get_values() noexcept { return values_.get_data(); }
146
154 const value_type* get_const_values() const noexcept
155 {
156 return values_.get_const_data();
157 }
158
164 index_type* get_col_idxs() noexcept { return col_idxs_.get_data(); }
165
173 const index_type* get_const_col_idxs() const noexcept
174 {
175 return col_idxs_.get_const_data();
176 }
177
184 {
185 return slice_lengths_.get_data();
186 }
187
195 const size_type* get_const_slice_lengths() const noexcept
196 {
197 return slice_lengths_.get_const_data();
198 }
199
205 size_type* get_slice_sets() noexcept { return slice_sets_.get_data(); }
206
214 const size_type* get_const_slice_sets() const noexcept
215 {
216 return slice_sets_.get_const_data();
217 }
218
224 size_type get_slice_size() const noexcept { return slice_size_; }
225
231 size_type get_stride_factor() const noexcept { return stride_factor_; }
232
238 size_type get_total_cols() const noexcept
239 {
240 return values_.get_size() / slice_size_;
241 }
242
249 {
250 return values_.get_size();
251 }
252
265 value_type& val_at(size_type row, size_type slice_set,
266 size_type idx) noexcept
267 {
268 return values_.get_data()[this->linearize_index(row, slice_set, idx)];
269 }
270
274 value_type val_at(size_type row, size_type slice_set,
275 size_type idx) const noexcept
276 {
277 return values_
278 .get_const_data()[this->linearize_index(row, slice_set, idx)];
279 }
280
293 index_type& col_at(size_type row, size_type slice_set,
294 size_type idx) noexcept
295 {
296 return this->get_col_idxs()[this->linearize_index(row, slice_set, idx)];
297 }
298
302 index_type col_at(size_type row, size_type slice_set,
303 size_type idx) const noexcept
304 {
305 return this
306 ->get_const_col_idxs()[this->linearize_index(row, slice_set, idx)];
307 }
308
319 static std::unique_ptr<Sellp> create(std::shared_ptr<const Executor> exec,
320 const dim<2>& size = {},
321 size_type total_cols = 0);
322
335 static std::unique_ptr<Sellp> create(std::shared_ptr<const Executor> exec,
336 const dim<2>& size,
337 size_type slice_size,
338 size_type stride_factor,
339 size_type total_cols);
340
346
353
358 Sellp(const Sellp&);
359
366
367protected:
368 Sellp(std::shared_ptr<const Executor> exec, const dim<2>& size = {},
369 size_type total_cols = {});
370
371 Sellp(std::shared_ptr<const Executor> exec, const dim<2>& size,
372 size_type slice_size, size_type stride_factor, size_type total_cols);
373
374 void apply_impl(const LinOp* b, LinOp* x) const override;
375
376 void apply_impl(const LinOp* alpha, const LinOp* b, const LinOp* beta,
377 LinOp* x) const override;
378
379 size_type linearize_index(size_type row, size_type slice_set,
380 size_type col) const noexcept
381 {
382 return (slice_set + col) * slice_size_ + row;
383 }
384
385private:
386 array<value_type> values_;
387 array<index_type> col_idxs_;
388 array<size_type> slice_lengths_;
389 array<size_type> slice_sets_;
390 size_type slice_size_;
391 size_type stride_factor_;
392};
393
394
395} // namespace matrix
396} // namespace gko
397
398
399#endif // GKO_PUBLIC_CORE_MATRIX_SELLP_HPP_
The diagonal of a LinOp implementing this interface can be extracted.
Definition lin_op.hpp:743
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
A LinOp implementing this interface can read its data from a matrix_data structure.
Definition lin_op.hpp:605
A LinOp implementing this interface can write its data to a matrix_data structure.
Definition lin_op.hpp:660
This type is a device-side equivalent to matrix_data.
Definition device_matrix_data.hpp:36
CSR is a matrix format which stores only the nonzero coefficients by compressing each row of the matr...
Definition csr.hpp:126
Dense is a matrix format which explicitly stores all values of the matrix.
Definition dense.hpp:120
size_type get_slice_size() const noexcept
Returns the size of a slice.
Definition sellp.hpp:224
value_type & val_at(size_type row, size_type slice_set, size_type idx) noexcept
Returns the idx-th non-zero element of the row-th row with slice_set slice set.
Definition sellp.hpp:265
static std::unique_ptr< Sellp > create(std::shared_ptr< const Executor > exec, const dim< 2 > &size, size_type slice_size, size_type stride_factor, size_type total_cols)
Creates an uninitialized Sellp matrix of the specified size.
index_type & col_at(size_type row, size_type slice_set, size_type idx) noexcept
Returns the idx-th column index of the row-th row with slice_set slice set.
Definition sellp.hpp:293
Sellp(const Sellp &)
Copy-assigns a Sellp matrix.
std::unique_ptr< absolute_type > compute_absolute() const override
Gets the AbsoluteLinOp.
value_type * get_values() noexcept
Returns the values of the matrix.
Definition sellp.hpp:145
Sellp & operator=(Sellp &&)
Move-assigns a Sellp matrix.
size_type get_stride_factor() const noexcept
Returns the stride factor(t) of SELL-P.
Definition sellp.hpp:231
const size_type * get_const_slice_sets() const noexcept
Returns the offsets of slices.
Definition sellp.hpp:214
index_type * get_col_idxs() noexcept
Returns the column indexes of the matrix.
Definition sellp.hpp:164
size_type get_total_cols() const noexcept
Returns the total column number.
Definition sellp.hpp:238
index_type col_at(size_type row, size_type slice_set, size_type idx) const noexcept
Returns the idx-th column index of the row-th row with slice_set slice set.
Definition sellp.hpp:302
size_type * get_slice_lengths() noexcept
Returns the lengths(columns) of slices.
Definition sellp.hpp:183
Sellp & operator=(const Sellp &)
Copy-assigns a Sellp matrix.
void compute_absolute_inplace() override
Compute absolute inplace on each element.
static std::unique_ptr< Sellp > create(std::shared_ptr< const Executor > exec, const dim< 2 > &size={}, size_type total_cols=0)
Creates an uninitialized Sellp matrix of the specified size.
size_type * get_slice_sets() noexcept
Returns the offsets of slices.
Definition sellp.hpp:205
size_type get_num_stored_elements() const noexcept
Returns the number of elements explicitly stored in the matrix.
Definition sellp.hpp:248
value_type val_at(size_type row, size_type slice_set, size_type idx) const noexcept
Returns the idx-th non-zero element of the row-th row with slice_set slice set.
Definition sellp.hpp:274
const index_type * get_const_col_idxs() const noexcept
Returns the column indexes of the matrix.
Definition sellp.hpp:173
Sellp(Sellp &&)
Move-assigns a Sellp matrix.
std::unique_ptr< Diagonal< ValueType > > extract_diagonal() const override
Extracts the diagonal entries of the matrix into a vector.
const size_type * get_const_slice_lengths() const noexcept
Returns the lengths(columns) of slices.
Definition sellp.hpp:195
const value_type * get_const_values() const noexcept
Returns the values of the matrix.
Definition sellp.hpp:154
The matrix namespace.
Definition dense_cache.hpp:24
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:264
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:283
std::size_t size_type
Integral type used for allocation quantities.
Definition types.hpp:90
typename detail::find_precision_impl< T, -step >::type previous_precision
Obtains the previous move type of T in the singly-linked precision corresponding bfloat16/half.
Definition math.hpp:473
typename detail::find_precision_impl< T, step >::type next_precision
Obtains the next move type of T in the singly-linked precision corresponding bfloat16/half.
Definition math.hpp:466
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