5#ifndef GKO_PUBLIC_CORE_MATRIX_CSR_HPP_
6#define GKO_PUBLIC_CORE_MATRIX_CSR_HPP_
9#include <ginkgo/core/base/array.hpp>
10#include <ginkgo/core/base/index_set.hpp>
11#include <ginkgo/core/base/lin_op.hpp>
12#include <ginkgo/core/base/math.hpp>
13#include <ginkgo/core/matrix/permutation.hpp>
14#include <ginkgo/core/matrix/scaled_permutation.hpp>
21template <
typename ValueType>
24template <
typename ValueType>
27template <
typename ValueType,
typename IndexType>
30template <
typename ValueType,
typename IndexType>
33template <
typename ValueType,
typename IndexType>
36template <
typename ValueType,
typename IndexType>
39template <
typename ValueType,
typename IndexType>
42template <
typename ValueType,
typename IndexType>
45template <
typename ValueType,
typename IndexType>
48template <
typename ValueType,
typename IndexType>
51template <
typename IndexType>
58template <
typename ValueType = default_precision,
typename IndexType =
int32>
103template <
typename ValueType = default_precision,
typename IndexType =
int32>
105 public ConvertibleTo<Csr<next_precision<ValueType>, IndexType>>,
106#if GINKGO_ENABLE_HALF
108 Csr<next_precision<next_precision<ValueType>>, IndexType>>,
123 remove_complex<Csr<ValueType, IndexType>>>,
126 friend class Coo<ValueType, IndexType>;
127 friend class Dense<ValueType>;
129 friend class Ell<ValueType, IndexType>;
130 friend class Hybrid<ValueType, IndexType>;
131 friend class Sellp<ValueType, IndexType>;
133 friend class Fbcsr<ValueType, IndexType>;
134 friend class CsrBuilder<ValueType, IndexType>;
158 using value_type = ValueType;
159 using index_type = IndexType;
174 friend class automatical;
215 virtual std::shared_ptr<strategy_type>
copy() = 0;
218 void set_name(std::string name) { name_ = name; }
240 auto host_mtx_exec = mtx_row_ptrs.
get_executor()->get_master();
242 const bool is_mtx_on_host{host_mtx_exec ==
244 const index_type* row_ptrs{};
245 if (is_mtx_on_host) {
248 row_ptrs_host = mtx_row_ptrs;
251 auto num_rows = mtx_row_ptrs.
get_size() - 1;
252 max_length_per_row_ = 0;
253 for (
size_type i = 0; i < num_rows; i++) {
254 max_length_per_row_ = std::max(max_length_per_row_,
255 row_ptrs[i + 1] - row_ptrs[i]);
259 int64_t
clac_size(
const int64_t nnz)
override {
return 0; }
261 index_type get_max_length_per_row() const noexcept
263 return max_length_per_row_;
266 std::shared_ptr<strategy_type>
copy()
override
268 return std::make_shared<classical>();
272 index_type max_length_per_row_;
291 int64_t
clac_size(
const int64_t nnz)
override {
return 0; }
293 std::shared_ptr<strategy_type>
copy()
override
295 return std::make_shared<merge_path>();
316 int64_t
clac_size(
const int64_t nnz)
override {
return 0; }
318 std::shared_ptr<strategy_type>
copy()
override
320 return std::make_shared<cusparse>();
340 int64_t
clac_size(
const int64_t nnz)
override {
return 0; }
342 std::shared_ptr<strategy_type>
copy()
override
344 return std::make_shared<sparselib>();
370 :
load_balance(exec->get_num_warps(), exec->get_warp_size())
379 :
load_balance(exec->get_num_warps(), exec->get_warp_size(), false)
390 :
load_balance(exec->get_num_subgroups(), 32, false,
"intel")
405 bool cuda_strategy =
true,
406 std::string strategy_name =
"none")
409 warp_size_(warp_size),
410 cuda_strategy_(cuda_strategy),
411 strategy_name_(strategy_name)
420 auto host_srow_exec = mtx_srow->
get_executor()->get_master();
421 auto host_mtx_exec = mtx_row_ptrs.
get_executor()->get_master();
422 const bool is_srow_on_host{host_srow_exec ==
424 const bool is_mtx_on_host{host_mtx_exec ==
428 const index_type* row_ptrs{};
430 if (is_srow_on_host) {
433 srow_host = *mtx_srow;
436 if (is_mtx_on_host) {
439 row_ptrs_host = mtx_row_ptrs;
445 const auto num_rows = mtx_row_ptrs.
get_size() - 1;
446 const auto num_elems = row_ptrs[num_rows];
447 const auto bucket_divider =
448 num_elems > 0 ?
ceildiv(num_elems, warp_size_) : 1;
449 for (
size_type i = 0; i < num_rows; i++) {
453 if (bucket < nwarps) {
459 srow[i] += srow[i - 1];
461 if (!is_srow_on_host) {
462 *mtx_srow = srow_host;
469 if (warp_size_ > 0) {
471 if (nnz >=
static_cast<int64_t
>(2e8)) {
473 }
else if (nnz >=
static_cast<int64_t
>(2e7)) {
475 }
else if (nnz >=
static_cast<int64_t
>(2e6)) {
477 }
else if (nnz >=
static_cast<int64_t
>(2e5)) {
480 if (strategy_name_ ==
"intel") {
482 if (nnz >=
static_cast<int64_t
>(2e8)) {
484 }
else if (nnz >=
static_cast<int64_t
>(2e7)) {
488#if GINKGO_HIP_PLATFORM_HCC
489 if (!cuda_strategy_) {
491 if (nnz >=
static_cast<int64_t
>(1e7)) {
493 }
else if (nnz >=
static_cast<int64_t
>(1e6)) {
499 auto nwarps = nwarps_ * multiple;
506 std::shared_ptr<strategy_type>
copy()
override
508 return std::make_shared<load_balance>(
509 nwarps_, warp_size_, cuda_strategy_, strategy_name_);
516 std::string strategy_name_;
523 const index_type nvidia_row_len_limit = 1024;
526 const index_type nvidia_nnz_limit{
static_cast<index_type
>(1e6)};
529 const index_type amd_row_len_limit = 768;
532 const index_type amd_nnz_limit{
static_cast<index_type
>(1e8)};
535 const index_type intel_row_len_limit = 25600;
538 const index_type intel_nnz_limit{
static_cast<index_type
>(3e8)};
558 :
automatical(exec->get_num_warps(), exec->get_warp_size())
567 :
automatical(exec->get_num_warps(), exec->get_warp_size(), false)
578 :
automatical(exec->get_num_subgroups(), 32, false,
"intel")
593 bool cuda_strategy =
true,
594 std::string strategy_name =
"none")
597 warp_size_(warp_size),
598 cuda_strategy_(cuda_strategy),
599 strategy_name_(strategy_name),
600 max_length_per_row_(0)
609 index_type nnz_limit = nvidia_nnz_limit;
610 index_type row_len_limit = nvidia_row_len_limit;
611 if (strategy_name_ ==
"intel") {
612 nnz_limit = intel_nnz_limit;
613 row_len_limit = intel_row_len_limit;
615#if GINKGO_HIP_PLATFORM_HCC
616 if (!cuda_strategy_) {
617 nnz_limit = amd_nnz_limit;
618 row_len_limit = amd_row_len_limit;
621 auto host_mtx_exec = mtx_row_ptrs.
get_executor()->get_master();
622 const bool is_mtx_on_host{host_mtx_exec ==
625 const index_type* row_ptrs{};
626 if (is_mtx_on_host) {
629 row_ptrs_host = mtx_row_ptrs;
632 const auto num_rows = mtx_row_ptrs.
get_size() - 1;
633 if (row_ptrs[num_rows] > nnz_limit) {
635 cuda_strategy_, strategy_name_);
636 if (is_mtx_on_host) {
637 actual_strategy.
process(mtx_row_ptrs, mtx_srow);
639 actual_strategy.
process(row_ptrs_host, mtx_srow);
641 this->set_name(actual_strategy.
get_name());
643 index_type maxnum = 0;
644 for (
size_type i = 0; i < num_rows; i++) {
645 maxnum = std::max(maxnum, row_ptrs[i + 1] - row_ptrs[i]);
647 if (maxnum > row_len_limit) {
649 nwarps_, warp_size_, cuda_strategy_, strategy_name_);
650 if (is_mtx_on_host) {
651 actual_strategy.
process(mtx_row_ptrs, mtx_srow);
653 actual_strategy.
process(row_ptrs_host, mtx_srow);
655 this->set_name(actual_strategy.
get_name());
658 if (is_mtx_on_host) {
659 actual_strategy.
process(mtx_row_ptrs, mtx_srow);
660 max_length_per_row_ =
661 actual_strategy.get_max_length_per_row();
663 actual_strategy.
process(row_ptrs_host, mtx_srow);
664 max_length_per_row_ =
665 actual_strategy.get_max_length_per_row();
667 this->set_name(actual_strategy.
get_name());
674 return std::make_shared<load_balance>(
675 nwarps_, warp_size_, cuda_strategy_, strategy_name_)
679 index_type get_max_length_per_row() const noexcept
681 return max_length_per_row_;
684 std::shared_ptr<strategy_type>
copy()
override
686 return std::make_shared<automatical>(
687 nwarps_, warp_size_, cuda_strategy_, strategy_name_);
694 std::string strategy_name_;
695 index_type max_length_per_row_;
698 friend class Csr<previous_precision<ValueType>, IndexType>;
705#if GINKGO_ENABLE_HALF
706 friend class Csr<previous_precision<previous_precision<ValueType>>,
714 result)
const override;
748 void read(
const mat_data& data)
override;
750 void read(
const device_mat_data& data)
override;
752 void read(device_mat_data&& data)
override;
754 void write(mat_data& data)
const override;
782 std::unique_ptr<Permutation<IndexType>> value_permutation;
833 bool invert =
false)
const;
880 bool invert =
false)
const;
912 bool invert =
false)
const;
914 std::unique_ptr<LinOp>
permute(
917 std::unique_ptr<LinOp> inverse_permute(
920 std::unique_ptr<LinOp> row_permute(
923 std::unique_ptr<LinOp> column_permute(
926 std::unique_ptr<LinOp> inverse_row_permute(
929 std::unique_ptr<LinOp> inverse_column_permute(
949 bool is_sorted_by_column_index()
const;
956 value_type*
get_values() noexcept {
return values_.get_data(); }
967 return values_.get_const_data();
998 return col_idxs_.get_const_data();
1017 return row_ptrs_.get_const_data();
1025 index_type*
get_srow() noexcept {
return srow_.get_data(); }
1036 return srow_.get_const_data();
1046 return srow_.get_size();
1056 return values_.get_size();
1075 strategy_ = std::move(strategy->copy());
1088 GKO_ASSERT_EQUAL_DIMENSIONS(alpha,
dim<2>(1, 1));
1101 GKO_ASSERT_EQUAL_DIMENSIONS(alpha,
dim<2>(1, 1));
1113 static std::unique_ptr<Csr>
create(std::shared_ptr<const Executor> exec,
1114 std::shared_ptr<strategy_type> strategy);
1128 std::shared_ptr<const Executor> exec,
const dim<2>& size = {},
1130 std::shared_ptr<strategy_type> strategy =
nullptr);
1152 std::shared_ptr<const Executor> exec,
const dim<2>& size,
1155 std::shared_ptr<strategy_type> strategy =
nullptr);
1161 template <
typename InputValueType,
typename InputColumnIndexType,
1162 typename InputRowPtrType>
1164 "explicitly construct the gko::array argument instead of passing "
1165 "initializer lists")
1168 std::initializer_list<InputValueType> values,
1169 std::initializer_list<InputColumnIndexType> col_idxs,
1170 std::initializer_list<InputRowPtrType> row_ptrs)
1193 std::shared_ptr<const Executor> exec,
const dim<2>& size,
1194 gko::detail::const_array_view<ValueType>&& values,
1195 gko::detail::const_array_view<IndexType>&& col_idxs,
1196 gko::detail::const_array_view<IndexType>&& row_ptrs,
1197 std::shared_ptr<strategy_type> strategy =
nullptr);
1227 const span& row_span,
const span& column_span)
const;
1254 Csr(std::shared_ptr<const Executor> exec,
const dim<2>& size = {},
1256 std::shared_ptr<strategy_type> strategy =
nullptr);
1258 Csr(std::shared_ptr<const Executor> exec,
const dim<2>& size,
1261 std::shared_ptr<strategy_type> strategy =
nullptr);
1263 void apply_impl(
const LinOp* b,
LinOp* x)
const override;
1265 void apply_impl(
const LinOp* alpha,
const LinOp* b,
const LinOp* beta,
1266 LinOp* x)
const override;
1269 static std::shared_ptr<strategy_type> make_default_strategy(
1270 std::shared_ptr<const Executor> exec)
1272 auto cuda_exec = std::dynamic_pointer_cast<const CudaExecutor>(exec);
1273 auto hip_exec = std::dynamic_pointer_cast<const HipExecutor>(exec);
1274 auto dpcpp_exec = std::dynamic_pointer_cast<const DpcppExecutor>(exec);
1275 std::shared_ptr<strategy_type> new_strategy;
1277 new_strategy = std::make_shared<automatical>(cuda_exec);
1278 }
else if (hip_exec) {
1279 new_strategy = std::make_shared<automatical>(hip_exec);
1280 }
else if (dpcpp_exec) {
1281 new_strategy = std::make_shared<automatical>(dpcpp_exec);
1283 new_strategy = std::make_shared<classical>();
1285 return new_strategy;
1289 template <
typename CsrType>
1290 void convert_strategy_helper(CsrType* result)
const
1293 std::shared_ptr<typename CsrType::strategy_type> new_strat;
1295 new_strat = std::make_shared<typename CsrType::classical>();
1296 }
else if (
dynamic_cast<merge_path*
>(strat)) {
1297 new_strat = std::make_shared<typename CsrType::merge_path>();
1298 }
else if (
dynamic_cast<cusparse*
>(strat)) {
1299 new_strat = std::make_shared<typename CsrType::cusparse>();
1300 }
else if (
dynamic_cast<sparselib*
>(strat)) {
1301 new_strat = std::make_shared<typename CsrType::sparselib>();
1303 auto rexec = result->get_executor();
1305 std::dynamic_pointer_cast<const CudaExecutor>(rexec);
1306 auto hip_exec = std::dynamic_pointer_cast<const HipExecutor>(rexec);
1308 std::dynamic_pointer_cast<const DpcppExecutor>(rexec);
1313 std::make_shared<typename CsrType::load_balance>(
1316 new_strat = std::make_shared<typename CsrType::automatical>(
1319 }
else if (hip_exec) {
1322 std::make_shared<typename CsrType::load_balance>(
1325 new_strat = std::make_shared<typename CsrType::automatical>(
1328 }
else if (dpcpp_exec) {
1331 std::make_shared<typename CsrType::load_balance>(
1334 new_strat = std::make_shared<typename CsrType::automatical>(
1339 auto this_cuda_exec =
1340 std::dynamic_pointer_cast<const CudaExecutor>(
1342 auto this_hip_exec =
1343 std::dynamic_pointer_cast<const HipExecutor>(
1345 auto this_dpcpp_exec =
1346 std::dynamic_pointer_cast<const DpcppExecutor>(
1348 if (this_cuda_exec) {
1351 std::make_shared<typename CsrType::load_balance>(
1355 std::make_shared<typename CsrType::automatical>(
1358 }
else if (this_hip_exec) {
1361 std::make_shared<typename CsrType::load_balance>(
1365 std::make_shared<typename CsrType::automatical>(
1368 }
else if (this_dpcpp_exec) {
1371 std::make_shared<typename CsrType::load_balance>(
1375 std::make_shared<typename CsrType::automatical>(
1383 new_strat = std::make_shared<typename CsrType::classical>();
1387 result->set_strategy(new_strat);
1395 srow_.resize_and_reset(strategy_->clac_size(values_.get_size()));
1396 strategy_->process(row_ptrs_, &srow_);
1405 virtual void scale_impl(
const LinOp* alpha);
1413 virtual void inv_scale_impl(
const LinOp* alpha);
1416 std::shared_ptr<strategy_type> strategy_;
1417 array<value_type> values_;
1418 array<index_type> col_idxs_;
1419 array<index_type> row_ptrs_;
1420 array<index_type> srow_;
1422 void add_scaled_identity_impl(
const LinOp* a,
const LinOp* b)
override;
1435template <
typename ValueType,
typename IndexType>
1436void strategy_rebuild_helper(Csr<ValueType, IndexType>* result)
1438 using load_balance =
typename Csr<ValueType, IndexType>::load_balance;
1439 using automatical =
typename Csr<ValueType, IndexType>::automatical;
1440 auto strategy = result->get_strategy();
1441 auto executor = result->get_executor();
1442 if (std::dynamic_pointer_cast<load_balance>(strategy)) {
1444 std::dynamic_pointer_cast<const HipExecutor>(executor)) {
1445 result->set_strategy(std::make_shared<load_balance>(exec));
1446 }
else if (
auto exec = std::dynamic_pointer_cast<const CudaExecutor>(
1448 result->set_strategy(std::make_shared<load_balance>(exec));
1450 }
else if (std::dynamic_pointer_cast<automatical>(strategy)) {
1452 std::dynamic_pointer_cast<const HipExecutor>(executor)) {
1453 result->set_strategy(std::make_shared<automatical>(exec));
1454 }
else if (
auto exec = std::dynamic_pointer_cast<const CudaExecutor>(
1456 result->set_strategy(std::make_shared<automatical>(exec));
ConvertibleTo interface is used to mark that the implementer can be converted to the object of Result...
Definition polymorphic_object.hpp:479
This is the Executor subclass which represents the CUDA device.
Definition executor.hpp:1542
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
The first step in using the Ginkgo library consists of creating an executor.
Definition executor.hpp:615
Definition lin_op.hpp:117
LinOp(const LinOp &)=default
Copy-constructs a LinOp.
This is the Executor subclass which represents the OpenMP device (typically CPU).
Definition executor.hpp:1387
Linear operators which support permutation should implement the Permutable interface.
Definition lin_op.hpp:484
std::shared_ptr< const Executor > get_executor() const noexcept
Returns the Executor of the object.
Definition polymorphic_object.hpp:243
A LinOp implementing this interface can read its data from a matrix_data structure.
Definition lin_op.hpp:605
Adds the operation M <- a I + b M for matrix M, identity operator I and scalars a and b,...
Definition lin_op.hpp:818
Linear operators which support transposition should implement the Transposable interface.
Definition lin_op.hpp:433
A LinOp implementing this interface can write its data to a matrix_data structure.
Definition lin_op.hpp:660
An array is a container which encapsulates fixed-sized arrays, stored on the Executor tied to the arr...
Definition array.hpp:166
value_type * get_data() noexcept
Returns a pointer to the block of memory used to store the elements of the array.
Definition array.hpp:673
std::shared_ptr< const Executor > get_executor() const noexcept
Returns the Executor associated with the array.
Definition array.hpp:689
const value_type * get_const_data() const noexcept
Returns a constant pointer to the block of memory used to store the elements of the array.
Definition array.hpp:682
size_type get_size() const noexcept
Returns the number of elements in the array.
Definition array.hpp:656
This type is a device-side equivalent to matrix_data.
Definition device_matrix_data.hpp:36
An index set class represents an ordered set of intervals.
Definition index_set.hpp:56
COO stores a matrix in the coordinate matrix format.
Definition coo.hpp:63
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:684
automatical(int64_t nwarps, int warp_size=32, bool cuda_strategy=true, std::string strategy_name="none")
Creates an automatical strategy with specified parameters.
Definition csr.hpp:592
automatical()
Creates an automatical strategy.
Definition csr.hpp:547
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:672
automatical(std::shared_ptr< const CudaExecutor > exec)
Creates an automatical strategy with CUDA executor.
Definition csr.hpp:557
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:603
automatical(std::shared_ptr< const DpcppExecutor > exec)
Creates an automatical strategy with Dpcpp executor.
Definition csr.hpp:577
automatical(std::shared_ptr< const HipExecutor > exec)
Creates an automatical strategy with HIP executor.
Definition csr.hpp:566
classical is a strategy_type which uses the same number of threads on each row.
Definition csr.hpp:230
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:237
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:266
classical()
Creates a classical strategy.
Definition csr.hpp:235
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:259
cusparse is a strategy_type which uses the sparselib csr.
Definition csr.hpp:305
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:316
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:318
cusparse()
Creates a cusparse strategy.
Definition csr.hpp:310
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:312
load_balance is a strategy_type which uses the load balance algorithm.
Definition csr.hpp:351
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:414
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:506
load_balance(std::shared_ptr< const HipExecutor > exec)
Creates a load_balance strategy with HIP executor.
Definition csr.hpp:378
load_balance()
Creates a load_balance strategy.
Definition csr.hpp:359
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:467
load_balance(int64_t nwarps, int warp_size=32, bool cuda_strategy=true, std::string strategy_name="none")
Creates a load_balance strategy with specified parameters.
Definition csr.hpp:404
load_balance(std::shared_ptr< const CudaExecutor > exec)
Creates a load_balance strategy with CUDA executor.
Definition csr.hpp:369
load_balance(std::shared_ptr< const DpcppExecutor > exec)
Creates a load_balance strategy with DPCPP executor.
Definition csr.hpp:389
merge_path is a strategy_type which uses the merge_path algorithm.
Definition csr.hpp:280
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:291
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:293
merge_path()
Creates a merge_path strategy.
Definition csr.hpp:285
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:287
sparselib is a strategy_type which uses the sparselib csr.
Definition csr.hpp:329
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:340
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:336
sparselib()
Creates a sparselib strategy.
Definition csr.hpp:334
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:342
strategy_type is to decide how to set the csr algorithm.
Definition csr.hpp:173
virtual int64_t clac_size(const int64_t nnz)=0
Computes the srow size according to the number of nonzeros.
std::string get_name()
Returns the name of strategy.
Definition csr.hpp:191
virtual std::shared_ptr< strategy_type > copy()=0
Copy a strategy.
virtual void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow)=0
Computes srow according to row pointers.
strategy_type(std::string name)
Creates a strategy_type.
Definition csr.hpp:182
CSR is a matrix format which stores only the nonzero coefficients by compressing each row of the matr...
Definition csr.hpp:124
std::pair< std::unique_ptr< Csr >, permuting_reuse_info > permute_reuse(ptr_param< const Permutation< index_type > > permutation, permute_mode mode=permute_mode::symmetric) const
Computes the operations necessary to propagate changed values from a matrix A to a permuted matrix.
Csr & operator=(const Csr &)
Copy-assigns a Csr matrix.
std::unique_ptr< Csr > scale_permute(ptr_param< const ScaledPermutation< value_type, index_type > > permutation, permute_mode=permute_mode::symmetric) const
Creates a scaled and permuted copy of this matrix.
std::unique_ptr< absolute_type > compute_absolute() const override
Gets the AbsoluteLinOp.
const index_type * get_const_row_ptrs() const noexcept
Returns the row pointers of the matrix.
Definition csr.hpp:1015
std::unique_ptr< Csr< ValueType, IndexType > > create_submatrix(const span &row_span, const span &column_span) const
Creates a submatrix from this Csr matrix given row and column spans.
static std::unique_ptr< Csr > create(std::shared_ptr< const Executor > exec, const dim< 2 > &size={}, size_type num_nonzeros={}, std::shared_ptr< strategy_type > strategy=nullptr)
Creates an uninitialized CSR matrix of the specified size.
const index_type * get_const_srow() const noexcept
Returns the starting rows.
Definition csr.hpp:1034
void set_strategy(std::shared_ptr< strategy_type > strategy)
Set the strategy.
Definition csr.hpp:1073
void inv_scale(ptr_param< const LinOp > alpha)
Scales the matrix with the inverse of a scalar.
Definition csr.hpp:1098
index_type * get_srow() noexcept
Returns the starting rows.
Definition csr.hpp:1025
static std::unique_ptr< Csr > create(std::shared_ptr< const Executor > exec, std::shared_ptr< strategy_type > strategy)
Creates an uninitialized CSR matrix of the specified size.
size_type get_num_srow_elements() const noexcept
Returns the number of the srow stored elements (involved warps)
Definition csr.hpp:1044
std::pair< std::unique_ptr< Csr >, permuting_reuse_info > permute_reuse(ptr_param< const Permutation< index_type > > row_permutation, ptr_param< const Permutation< index_type > > column_permutation, bool invert=false) const
Computes the operations necessary to propagate changed values from a matrix A to a permuted matrix.
std::unique_ptr< Csr< ValueType, IndexType > > create_submatrix(const index_set< IndexType > &row_index_set, const index_set< IndexType > &column_index_set) const
Creates a submatrix from this Csr matrix given row and column index_set objects.
std::unique_ptr< Diagonal< ValueType > > extract_diagonal() const override
Extracts the diagonal entries of the matrix into a vector.
static std::unique_ptr< Csr > create(std::shared_ptr< const Executor > exec, const dim< 2 > &size, array< value_type > values, array< index_type > col_idxs, array< index_type > row_ptrs, std::shared_ptr< strategy_type > strategy=nullptr)
Creates a CSR matrix from already allocated (and initialized) row pointer, column index and value arr...
index_type * get_row_ptrs() noexcept
Returns the row pointers of the matrix.
Definition csr.hpp:1006
std::unique_ptr< Csr > permute(ptr_param< const Permutation< index_type > > permutation, permute_mode mode=permute_mode::symmetric) const
Creates a permuted copy of this matrix with the given permutation .
std::unique_ptr< const Dense< ValueType > > create_const_value_view() const
Creates a const Dense view of the value array of this matrix as a column vector of dimensions nnz x 1...
static std::unique_ptr< const Csr > create_const(std::shared_ptr< const Executor > exec, const dim< 2 > &size, gko::detail::const_array_view< ValueType > &&values, gko::detail::const_array_view< IndexType > &&col_idxs, gko::detail::const_array_view< IndexType > &&row_ptrs, std::shared_ptr< strategy_type > strategy=nullptr)
Creates a constant (immutable) Csr matrix from a set of constant arrays.
Csr(const Csr &)
Copy-constructs a Csr matrix.
Csr & operator=(Csr &&)
Move-assigns a Csr matrix.
std::unique_ptr< LinOp > transpose() const override
Returns a LinOp representing the transpose of the Transposable object.
const value_type * get_const_values() const noexcept
Returns the values of the matrix.
Definition csr.hpp:965
void compute_absolute_inplace() override
Compute absolute inplace on each element.
size_type get_num_stored_elements() const noexcept
Returns the number of elements explicitly stored in the matrix.
Definition csr.hpp:1054
std::shared_ptr< strategy_type > get_strategy() const noexcept
Returns the strategy.
Definition csr.hpp:1063
const index_type * get_const_col_idxs() const noexcept
Returns the column indexes of the matrix.
Definition csr.hpp:996
void sort_by_column_index()
Sorts all (value, col_idx) pairs in each row by column index.
std::pair< std::unique_ptr< Csr >, permuting_reuse_info > transpose_reuse() const
Computes the necessary data to update a transposed matrix from its original matrix.
std::unique_ptr< Csr > scale_permute(ptr_param< const ScaledPermutation< value_type, index_type > > row_permutation, ptr_param< const ScaledPermutation< value_type, index_type > > column_permutation, bool invert=false) const
Creates a scaled and permuted copy of this matrix.
std::unique_ptr< Dense< ValueType > > create_value_view()
Creates a Dense view of the value array of this matrix as a column vector of dimensions nnz x 1.
void scale(ptr_param< const LinOp > alpha)
Scales the matrix with a scalar.
Definition csr.hpp:1085
value_type * get_values() noexcept
Returns the values of the matrix.
Definition csr.hpp:956
index_type * get_col_idxs() noexcept
Returns the column indexes of the matrix.
Definition csr.hpp:987
Csr(Csr &&)
Move-constructs a Csr matrix.
std::unique_ptr< Csr > permute(ptr_param< const Permutation< index_type > > row_permutation, ptr_param< const Permutation< index_type > > column_permutation, bool invert=false) const
Creates a non-symmetrically permuted copy of this matrix with the given row and column permutations...
std::unique_ptr< LinOp > conj_transpose() const override
Returns a LinOp representing the conjugate transpose of the Transposable object.
Dense is a matrix format which explicitly stores all values of the matrix.
Definition dense.hpp:117
This class is a utility which efficiently implements the diagonal matrix (a linear operator which sca...
Definition diagonal.hpp:53
ELL is a matrix format where stride with explicit zeros is used such that all rows have the same numb...
Definition ell.hpp:64
Fixed-block compressed sparse row storage matrix format.
Definition fbcsr.hpp:113
HYBRID is a matrix format which splits the matrix into ELLPACK and COO format.
Definition hybrid.hpp:55
Permutation is a matrix format that represents a permutation matrix, i.e.
Definition permutation.hpp:112
ScaledPermutation is a matrix combining a permutation with scaling factors.
Definition scaled_permutation.hpp:38
SELL-P is a matrix format similar to ELL format.
Definition sellp.hpp:55
SparsityCsr is a matrix format which stores only the sparsity pattern of a sparse matrix by compressi...
Definition sparsity_csr.hpp:56
This class is used for function parameters in the place of raw pointers.
Definition utils_helper.hpp:41
The matrix namespace.
Definition dense_cache.hpp:15
permute_mode
Specifies how a permutation will be applied to a matrix.
Definition permutation.hpp:42
@ symmetric
The rows and columns will be permuted.
Definition permutation.hpp:53
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
void write(StreamType &&os, MatrixPtrType &&matrix, layout_type layout=detail::mtx_io_traits< std::remove_cv_t< detail::pointee< MatrixPtrType > > >::default_layout)
Writes a matrix into an output stream in matrix market format.
Definition mtx_io.hpp:295
constexpr int64 ceildiv(int64 num, int64 den)
Performs integer division with rounding up.
Definition math.hpp:590
std::size_t size_type
Integral type used for allocation quantities.
Definition types.hpp:89
constexpr T min(const T &x, const T &y)
Returns the smaller of the arguments.
Definition math.hpp:719
std::unique_ptr< MatrixType > read(StreamType &&is, MatrixArgs &&... args)
Reads a matrix stored in matrix market format from an input stream.
Definition mtx_io.hpp:159
next_precision_base< T > next_precision
Obtains the next type in the singly-linked precision list with half.
Definition math.hpp:445
detail::temporary_clone< detail::pointee< Ptr > > make_temporary_clone(std::shared_ptr< const Executor > exec, Ptr &&ptr)
Creates a temporary_clone.
Definition temporary_clone.hpp:208
A type representing the dimensions of a multidimensional object.
Definition dim.hpp:26
permuting_reuse_info()
Creates an empty reuse info.
void update_values(ptr_param< const Csr > input, ptr_param< Csr > output) const
Propagates the values from an input matrix to the transformed matrix.
permuting_reuse_info(std::unique_ptr< Permutation< index_type > > value_permutation)
Creates a reuse info structure from its value permutation.
This structure is used as an intermediate data type to store a sparse matrix.
Definition matrix_data.hpp:126
A span is a lightweight structure used to create sub-ranges from other ranges.
Definition range.hpp:46