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 || GINKGO_ENABLE_BFLOAT16
107 public ConvertibleTo<Csr<next_precision<ValueType, 2>, IndexType>>,
109#if GINKGO_ENABLE_HALF && GINKGO_ENABLE_BFLOAT16
110 public ConvertibleTo<Csr<next_precision<ValueType, 3>, IndexType>>,
125 remove_complex<Csr<ValueType, IndexType>>>,
128 friend class Coo<ValueType, IndexType>;
129 friend class Dense<ValueType>;
131 friend class Ell<ValueType, IndexType>;
132 friend class Hybrid<ValueType, IndexType>;
133 friend class Sellp<ValueType, IndexType>;
135 friend class Fbcsr<ValueType, IndexType>;
136 friend class CsrBuilder<ValueType, IndexType>;
160 using value_type = ValueType;
161 using index_type = IndexType;
176 friend class automatical;
217 virtual std::shared_ptr<strategy_type>
copy() = 0;
220 void set_name(std::string name) { name_ = name; }
242 auto host_mtx_exec = mtx_row_ptrs.
get_executor()->get_master();
244 const bool is_mtx_on_host{host_mtx_exec ==
246 const index_type* row_ptrs{};
247 if (is_mtx_on_host) {
250 row_ptrs_host = mtx_row_ptrs;
253 auto num_rows = mtx_row_ptrs.
get_size() - 1;
254 max_length_per_row_ = 0;
255 for (
size_type i = 0; i < num_rows; i++) {
256 max_length_per_row_ = std::max(max_length_per_row_,
257 row_ptrs[i + 1] - row_ptrs[i]);
261 int64_t
clac_size(
const int64_t nnz)
override {
return 0; }
263 index_type get_max_length_per_row() const noexcept
265 return max_length_per_row_;
268 std::shared_ptr<strategy_type>
copy()
override
270 return std::make_shared<classical>();
274 index_type max_length_per_row_;
293 int64_t
clac_size(
const int64_t nnz)
override {
return 0; }
295 std::shared_ptr<strategy_type>
copy()
override
297 return std::make_shared<merge_path>();
318 int64_t
clac_size(
const int64_t nnz)
override {
return 0; }
320 std::shared_ptr<strategy_type>
copy()
override
322 return std::make_shared<cusparse>();
342 int64_t
clac_size(
const int64_t nnz)
override {
return 0; }
344 std::shared_ptr<strategy_type>
copy()
override
346 return std::make_shared<sparselib>();
372 :
load_balance(exec->get_num_warps(), exec->get_warp_size())
381 :
load_balance(exec->get_num_warps(), exec->get_warp_size(), false)
392 :
load_balance(exec->get_num_subgroups(), 32, false,
"intel")
407 bool cuda_strategy =
true,
408 std::string strategy_name =
"none")
411 warp_size_(warp_size),
412 cuda_strategy_(cuda_strategy),
413 strategy_name_(strategy_name)
422 auto host_srow_exec = mtx_srow->
get_executor()->get_master();
423 auto host_mtx_exec = mtx_row_ptrs.
get_executor()->get_master();
424 const bool is_srow_on_host{host_srow_exec ==
426 const bool is_mtx_on_host{host_mtx_exec ==
430 const index_type* row_ptrs{};
432 if (is_srow_on_host) {
435 srow_host = *mtx_srow;
438 if (is_mtx_on_host) {
441 row_ptrs_host = mtx_row_ptrs;
447 const auto num_rows = mtx_row_ptrs.
get_size() - 1;
448 const auto num_elems = row_ptrs[num_rows];
449 const auto bucket_divider =
450 num_elems > 0 ?
ceildiv(num_elems, warp_size_) : 1;
451 for (
size_type i = 0; i < num_rows; i++) {
455 if (bucket < nwarps) {
461 srow[i] += srow[i - 1];
463 if (!is_srow_on_host) {
464 *mtx_srow = srow_host;
471 if (warp_size_ > 0) {
473 if (nnz >=
static_cast<int64_t
>(2e8)) {
475 }
else if (nnz >=
static_cast<int64_t
>(2e7)) {
477 }
else if (nnz >=
static_cast<int64_t
>(2e6)) {
479 }
else if (nnz >=
static_cast<int64_t
>(2e5)) {
482 if (strategy_name_ ==
"intel") {
484 if (nnz >=
static_cast<int64_t
>(2e8)) {
486 }
else if (nnz >=
static_cast<int64_t
>(2e7)) {
490#if GINKGO_HIP_PLATFORM_HCC
491 if (!cuda_strategy_) {
493 if (nnz >=
static_cast<int64_t
>(1e7)) {
495 }
else if (nnz >=
static_cast<int64_t
>(1e6)) {
501 auto nwarps = nwarps_ * multiple;
508 std::shared_ptr<strategy_type>
copy()
override
510 return std::make_shared<load_balance>(
511 nwarps_, warp_size_, cuda_strategy_, strategy_name_);
518 std::string strategy_name_;
525 const index_type nvidia_row_len_limit = 1024;
528 const index_type nvidia_nnz_limit{
static_cast<index_type
>(1e6)};
531 const index_type amd_row_len_limit = 768;
534 const index_type amd_nnz_limit{
static_cast<index_type
>(1e8)};
537 const index_type intel_row_len_limit = 25600;
540 const index_type intel_nnz_limit{
static_cast<index_type
>(3e8)};
560 :
automatical(exec->get_num_warps(), exec->get_warp_size())
569 :
automatical(exec->get_num_warps(), exec->get_warp_size(), false)
580 :
automatical(exec->get_num_subgroups(), 32, false,
"intel")
595 bool cuda_strategy =
true,
596 std::string strategy_name =
"none")
599 warp_size_(warp_size),
600 cuda_strategy_(cuda_strategy),
601 strategy_name_(strategy_name),
602 max_length_per_row_(0)
611 index_type nnz_limit = nvidia_nnz_limit;
612 index_type row_len_limit = nvidia_row_len_limit;
613 if (strategy_name_ ==
"intel") {
614 nnz_limit = intel_nnz_limit;
615 row_len_limit = intel_row_len_limit;
617#if GINKGO_HIP_PLATFORM_HCC
618 if (!cuda_strategy_) {
619 nnz_limit = amd_nnz_limit;
620 row_len_limit = amd_row_len_limit;
623 auto host_mtx_exec = mtx_row_ptrs.
get_executor()->get_master();
624 const bool is_mtx_on_host{host_mtx_exec ==
627 const index_type* row_ptrs{};
628 if (is_mtx_on_host) {
631 row_ptrs_host = mtx_row_ptrs;
634 const auto num_rows = mtx_row_ptrs.
get_size() - 1;
635 if (row_ptrs[num_rows] > nnz_limit) {
637 cuda_strategy_, strategy_name_);
638 if (is_mtx_on_host) {
639 actual_strategy.
process(mtx_row_ptrs, mtx_srow);
641 actual_strategy.
process(row_ptrs_host, mtx_srow);
643 this->set_name(actual_strategy.
get_name());
645 index_type maxnum = 0;
646 for (
size_type i = 0; i < num_rows; i++) {
647 maxnum = std::max(maxnum, row_ptrs[i + 1] - row_ptrs[i]);
649 if (maxnum > row_len_limit) {
651 nwarps_, warp_size_, cuda_strategy_, strategy_name_);
652 if (is_mtx_on_host) {
653 actual_strategy.
process(mtx_row_ptrs, mtx_srow);
655 actual_strategy.
process(row_ptrs_host, mtx_srow);
657 this->set_name(actual_strategy.
get_name());
660 if (is_mtx_on_host) {
661 actual_strategy.
process(mtx_row_ptrs, mtx_srow);
662 max_length_per_row_ =
663 actual_strategy.get_max_length_per_row();
665 actual_strategy.
process(row_ptrs_host, mtx_srow);
666 max_length_per_row_ =
667 actual_strategy.get_max_length_per_row();
669 this->set_name(actual_strategy.
get_name());
676 return std::make_shared<load_balance>(
677 nwarps_, warp_size_, cuda_strategy_, strategy_name_)
681 index_type get_max_length_per_row() const noexcept
683 return max_length_per_row_;
686 std::shared_ptr<strategy_type>
copy()
override
688 return std::make_shared<automatical>(
689 nwarps_, warp_size_, cuda_strategy_, strategy_name_);
696 std::string strategy_name_;
697 index_type max_length_per_row_;
707#if GINKGO_ENABLE_HALF || GINKGO_ENABLE_BFLOAT16
719#if GINKGO_ENABLE_HALF && GINKGO_ENABLE_BFLOAT16
759 void read(
const mat_data& data)
override;
761 void read(
const device_mat_data& data)
override;
763 void read(device_mat_data&& data)
override;
765 void write(mat_data& data)
const override;
793 std::unique_ptr<Permutation<IndexType>> value_permutation;
844 bool invert =
false)
const;
891 bool invert =
false)
const;
923 bool invert =
false)
const;
925 std::unique_ptr<LinOp>
permute(
928 std::unique_ptr<LinOp> inverse_permute(
931 std::unique_ptr<LinOp> row_permute(
934 std::unique_ptr<LinOp> column_permute(
937 std::unique_ptr<LinOp> inverse_row_permute(
940 std::unique_ptr<LinOp> inverse_column_permute(
960 bool is_sorted_by_column_index()
const;
967 value_type*
get_values() noexcept {
return values_.get_data(); }
978 return values_.get_const_data();
1009 return col_idxs_.get_const_data();
1028 return row_ptrs_.get_const_data();
1036 index_type*
get_srow() noexcept {
return srow_.get_data(); }
1047 return srow_.get_const_data();
1057 return srow_.get_size();
1067 return values_.get_size();
1086 strategy_ = std::move(strategy->copy());
1099 GKO_ASSERT_EQUAL_DIMENSIONS(alpha,
dim<2>(1, 1));
1112 GKO_ASSERT_EQUAL_DIMENSIONS(alpha,
dim<2>(1, 1));
1124 static std::unique_ptr<Csr>
create(std::shared_ptr<const Executor> exec,
1125 std::shared_ptr<strategy_type> strategy);
1139 std::shared_ptr<const Executor> exec,
const dim<2>& size = {},
1141 std::shared_ptr<strategy_type> strategy =
nullptr);
1163 std::shared_ptr<const Executor> exec,
const dim<2>& size,
1166 std::shared_ptr<strategy_type> strategy =
nullptr);
1172 template <
typename InputValueType,
typename InputColumnIndexType,
1173 typename InputRowPtrType>
1175 "explicitly construct the gko::array argument instead of passing "
1176 "initializer lists")
1179 std::initializer_list<InputValueType> values,
1180 std::initializer_list<InputColumnIndexType> col_idxs,
1181 std::initializer_list<InputRowPtrType> row_ptrs)
1204 std::shared_ptr<const Executor> exec,
const dim<2>& size,
1205 gko::detail::const_array_view<ValueType>&& values,
1206 gko::detail::const_array_view<IndexType>&& col_idxs,
1207 gko::detail::const_array_view<IndexType>&& row_ptrs,
1208 std::shared_ptr<strategy_type> strategy =
nullptr);
1238 const span& row_span,
const span& column_span)
const;
1265 Csr(std::shared_ptr<const Executor> exec,
const dim<2>& size = {},
1267 std::shared_ptr<strategy_type> strategy =
nullptr);
1269 Csr(std::shared_ptr<const Executor> exec,
const dim<2>& size,
1272 std::shared_ptr<strategy_type> strategy =
nullptr);
1274 void apply_impl(
const LinOp* b,
LinOp* x)
const override;
1276 void apply_impl(
const LinOp* alpha,
const LinOp* b,
const LinOp* beta,
1277 LinOp* x)
const override;
1280 static std::shared_ptr<strategy_type> make_default_strategy(
1281 std::shared_ptr<const Executor> exec)
1283 auto cuda_exec = std::dynamic_pointer_cast<const CudaExecutor>(exec);
1284 auto hip_exec = std::dynamic_pointer_cast<const HipExecutor>(exec);
1285 auto dpcpp_exec = std::dynamic_pointer_cast<const DpcppExecutor>(exec);
1286 std::shared_ptr<strategy_type> new_strategy;
1288 new_strategy = std::make_shared<automatical>(cuda_exec);
1289 }
else if (hip_exec) {
1290 new_strategy = std::make_shared<automatical>(hip_exec);
1291 }
else if (dpcpp_exec) {
1292 new_strategy = std::make_shared<automatical>(dpcpp_exec);
1294 new_strategy = std::make_shared<classical>();
1296 return new_strategy;
1300 template <
typename CsrType>
1301 void convert_strategy_helper(CsrType* result)
const
1304 std::shared_ptr<typename CsrType::strategy_type> new_strat;
1306 new_strat = std::make_shared<typename CsrType::classical>();
1307 }
else if (
dynamic_cast<merge_path*
>(strat)) {
1308 new_strat = std::make_shared<typename CsrType::merge_path>();
1309 }
else if (
dynamic_cast<cusparse*
>(strat)) {
1310 new_strat = std::make_shared<typename CsrType::cusparse>();
1311 }
else if (
dynamic_cast<sparselib*
>(strat)) {
1312 new_strat = std::make_shared<typename CsrType::sparselib>();
1314 auto rexec = result->get_executor();
1316 std::dynamic_pointer_cast<const CudaExecutor>(rexec);
1317 auto hip_exec = std::dynamic_pointer_cast<const HipExecutor>(rexec);
1319 std::dynamic_pointer_cast<const DpcppExecutor>(rexec);
1324 std::make_shared<typename CsrType::load_balance>(
1327 new_strat = std::make_shared<typename CsrType::automatical>(
1330 }
else if (hip_exec) {
1333 std::make_shared<typename CsrType::load_balance>(
1336 new_strat = std::make_shared<typename CsrType::automatical>(
1339 }
else if (dpcpp_exec) {
1342 std::make_shared<typename CsrType::load_balance>(
1345 new_strat = std::make_shared<typename CsrType::automatical>(
1350 auto this_cuda_exec =
1351 std::dynamic_pointer_cast<const CudaExecutor>(
1353 auto this_hip_exec =
1354 std::dynamic_pointer_cast<const HipExecutor>(
1356 auto this_dpcpp_exec =
1357 std::dynamic_pointer_cast<const DpcppExecutor>(
1359 if (this_cuda_exec) {
1362 std::make_shared<typename CsrType::load_balance>(
1366 std::make_shared<typename CsrType::automatical>(
1369 }
else if (this_hip_exec) {
1372 std::make_shared<typename CsrType::load_balance>(
1376 std::make_shared<typename CsrType::automatical>(
1379 }
else if (this_dpcpp_exec) {
1382 std::make_shared<typename CsrType::load_balance>(
1386 std::make_shared<typename CsrType::automatical>(
1394 new_strat = std::make_shared<typename CsrType::classical>();
1398 result->set_strategy(new_strat);
1406 srow_.resize_and_reset(strategy_->clac_size(values_.get_size()));
1407 strategy_->process(row_ptrs_, &srow_);
1416 virtual void scale_impl(
const LinOp* alpha);
1424 virtual void inv_scale_impl(
const LinOp* alpha);
1427 std::shared_ptr<strategy_type> strategy_;
1428 array<value_type> values_;
1429 array<index_type> col_idxs_;
1430 array<index_type> row_ptrs_;
1431 array<index_type> srow_;
1433 void add_scaled_identity_impl(
const LinOp* a,
const LinOp* b)
override;
1446template <
typename ValueType,
typename IndexType>
1447void strategy_rebuild_helper(Csr<ValueType, IndexType>* result)
1449 using load_balance =
typename Csr<ValueType, IndexType>::load_balance;
1450 using automatical =
typename Csr<ValueType, IndexType>::automatical;
1451 auto strategy = result->get_strategy();
1452 auto executor = result->get_executor();
1453 if (std::dynamic_pointer_cast<load_balance>(strategy)) {
1455 std::dynamic_pointer_cast<const HipExecutor>(executor)) {
1456 result->set_strategy(std::make_shared<load_balance>(exec));
1457 }
else if (
auto exec = std::dynamic_pointer_cast<const CudaExecutor>(
1459 result->set_strategy(std::make_shared<load_balance>(exec));
1461 }
else if (std::dynamic_pointer_cast<automatical>(strategy)) {
1463 std::dynamic_pointer_cast<const HipExecutor>(executor)) {
1464 result->set_strategy(std::make_shared<automatical>(exec));
1465 }
else if (
auto exec = std::dynamic_pointer_cast<const CudaExecutor>(
1467 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:687
std::shared_ptr< const Executor > get_executor() const noexcept
Returns the Executor associated with the array.
Definition array.hpp:703
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:696
size_type get_size() const noexcept
Returns the number of elements in the array.
Definition array.hpp:670
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:65
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:686
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:594
automatical()
Creates an automatical strategy.
Definition csr.hpp:549
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:674
automatical(std::shared_ptr< const CudaExecutor > exec)
Creates an automatical strategy with CUDA executor.
Definition csr.hpp:559
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:605
automatical(std::shared_ptr< const DpcppExecutor > exec)
Creates an automatical strategy with Dpcpp executor.
Definition csr.hpp:579
automatical(std::shared_ptr< const HipExecutor > exec)
Creates an automatical strategy with HIP executor.
Definition csr.hpp:568
classical is a strategy_type which uses the same number of threads on each row.
Definition csr.hpp:232
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:239
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:268
classical()
Creates a classical strategy.
Definition csr.hpp:237
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:261
cusparse is a strategy_type which uses the sparselib csr.
Definition csr.hpp:307
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:318
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:320
cusparse()
Creates a cusparse strategy.
Definition csr.hpp:312
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:314
load_balance is a strategy_type which uses the load balance algorithm.
Definition csr.hpp:353
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:416
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:508
load_balance(std::shared_ptr< const HipExecutor > exec)
Creates a load_balance strategy with HIP executor.
Definition csr.hpp:380
load_balance()
Creates a load_balance strategy.
Definition csr.hpp:361
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:469
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:406
load_balance(std::shared_ptr< const CudaExecutor > exec)
Creates a load_balance strategy with CUDA executor.
Definition csr.hpp:371
load_balance(std::shared_ptr< const DpcppExecutor > exec)
Creates a load_balance strategy with DPCPP executor.
Definition csr.hpp:391
merge_path is a strategy_type which uses the merge_path algorithm.
Definition csr.hpp:282
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:293
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:295
merge_path()
Creates a merge_path strategy.
Definition csr.hpp:287
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:289
sparselib is a strategy_type which uses the sparselib csr.
Definition csr.hpp:331
int64_t clac_size(const int64_t nnz) override
Computes the srow size according to the number of nonzeros.
Definition csr.hpp:342
void process(const array< index_type > &mtx_row_ptrs, array< index_type > *mtx_srow) override
Computes srow according to row pointers.
Definition csr.hpp:338
sparselib()
Creates a sparselib strategy.
Definition csr.hpp:336
std::shared_ptr< strategy_type > copy() override
Copy a strategy.
Definition csr.hpp:344
strategy_type is to decide how to set the csr algorithm.
Definition csr.hpp:175
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:193
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:184
CSR is a matrix format which stores only the nonzero coefficients by compressing each row of the matr...
Definition csr.hpp:126
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:1026
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:1045
void set_strategy(std::shared_ptr< strategy_type > strategy)
Set the strategy.
Definition csr.hpp:1084
void inv_scale(ptr_param< const LinOp > alpha)
Scales the matrix with the inverse of a scalar.
Definition csr.hpp:1109
index_type * get_srow() noexcept
Returns the starting rows.
Definition csr.hpp:1036
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:1055
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:1017
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:976
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:1065
std::shared_ptr< strategy_type > get_strategy() const noexcept
Returns the strategy.
Definition csr.hpp:1074
const index_type * get_const_col_idxs() const noexcept
Returns the column indexes of the matrix.
Definition csr.hpp:1007
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:1096
value_type * get_values() noexcept
Returns the values of the matrix.
Definition csr.hpp:967
index_type * get_col_idxs() noexcept
Returns the column indexes of the matrix.
Definition csr.hpp:998
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:120
This class is a utility which efficiently implements the diagonal matrix (a linear operator which sca...
Definition diagonal.hpp:56
ELL is a matrix format where stride with explicit zeros is used such that all rows have the same numb...
Definition ell.hpp:66
Fixed-block compressed sparse row storage matrix format.
Definition fbcsr.hpp:116
HYBRID is a matrix format which splits the matrix into ELLPACK and COO format.
Definition hybrid.hpp:57
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:58
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:24
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: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
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:614
std::size_t size_type
Integral type used for allocation quantities.
Definition types.hpp:90
constexpr T min(const T &x, const T &y)
Returns the smaller of the arguments.
Definition math.hpp:750
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
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
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
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
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