Add a clear kernel to clear the analysis data.

ginkgo-project · Sep 2, 2019 · d89c1a0 · d89c1a0
1 parent c371d9b
commit d89c1a0
Show file tree

Hide file tree

Showing 9 changed files with 62 additions and 5 deletions.
diff --git a/core/device_hooks/common_kernels.inc.cpp b/core/device_hooks/common_kernels.inc.cpp
@@ -185,6 +185,9 @@ GKO_INSTANTIATE_FOR_EACH_VALUE_TYPE(GKO_DECLARE_CG_STEP_2_KERNEL);
 namespace lower_trs {
 
 
+GKO_DECLARE_LOWER_TRS_CLEAR_KERNEL()
+GKO_NOT_COMPILED(GKO_HOOK_MODULE);
+
 template <typename ValueType, typename IndexType>
 GKO_DECLARE_LOWER_TRS_GENERATE_KERNEL(ValueType, IndexType)
 GKO_NOT_COMPILED(GKO_HOOK_MODULE);

diff --git a/core/solver/lower_trs.cpp b/core/solver/lower_trs.cpp
@@ -53,13 +53,21 @@ namespace solver {
 namespace lower_trs {
 
 
+GKO_REGISTER_OPERATION(clear, lower_trs::clear);
 GKO_REGISTER_OPERATION(generate, lower_trs::generate);
 GKO_REGISTER_OPERATION(solve, lower_trs::solve);
 
 
 } // namespace lower_trs
 
 
+template <typename ValueType, typename IndexType>
+void LowerTrs<ValueType, IndexType>::clear_data() const
+{
+ this->get_executor()->run(lower_trs::make_clear());
+}
+
+
 template <typename ValueType, typename IndexType>
 void LowerTrs<ValueType, IndexType>::generate()
 {

diff --git a/core/solver/lower_trs_kernels.hpp b/core/solver/lower_trs_kernels.hpp
@@ -47,6 +47,10 @@ namespace kernels {
 namespace lower_trs {
 
 
+#define GKO_DECLARE_LOWER_TRS_CLEAR_KERNEL() \
+ void clear(std::shared_ptr<const DefaultExecutor> exec)
+
+
 #define GKO_DECLARE_LOWER_TRS_GENERATE_KERNEL(_vtype, _itype) \
  void generate(std::shared_ptr<const DefaultExecutor> exec, \
  const matrix::Csr<_vtype, _itype> *matrix, \
@@ -60,6 +64,7 @@ namespace lower_trs {
 
 
 #define GKO_DECLARE_ALL_AS_TEMPLATES \
+ GKO_DECLARE_LOWER_TRS_CLEAR_KERNEL(); \
  template <typename ValueType, typename IndexType> \
  GKO_DECLARE_LOWER_TRS_SOLVE_KERNEL(ValueType, IndexType); \
  template <typename ValueType, typename IndexType> \

diff --git a/cuda/solver/lower_trs_kernels.cu b/cuda/solver/lower_trs_kernels.cu
@@ -78,6 +78,25 @@ static cusp_csrsm_data cusp_csrsm_data{};
 #endif
 
 
+void clear(std::shared_ptr<const CudaExecutor> exec)
+{
+#if (defined(CUDA_VERSION) && (CUDA_VERSION > 9100))
+ cusparse::destroy(cusp_csrsm2_data.factor_descr);
+ if (cusp_csrsm2_data.solve_info) {
+ GKO_ASSERT_NO_CUSPARSE_ERRORS(
+ cusparseDestroyCsrsm2Info(cusp_csrsm2_data.solve_info));
+ }
+ if (cusp_csrsm2_data.factor_work_vec != nullptr) {
+ exec->free(cusp_csrsm2_data.factor_work_vec);
+ }
+#elif (defined(CUDA_VERSION) && (CUDA_VERSION < 9200))
+ cusparse::destroy(cusp_csrsm_data.factor_descr);
+ GKO_ASSERT_NO_CUSPARSE_ERRORS(
+ cusparseDestroySolveAnalysisInfo(cusp_csrsm_data.solve_info));
+#endif
+}
+
+
 template <typename ValueType, typename IndexType>
 void generate(std::shared_ptr<const CudaExecutor> exec,
  const matrix::Csr<ValueType, IndexType> *matrix,

diff --git a/cuda/test/solver/lower_trs_kernels.cpp b/cuda/test/solver/lower_trs_kernels.cpp
@@ -75,16 +75,16 @@ class LowerTrs : public ::testing::Test {
 
  std::unique_ptr<Mtx> gen_mtx(int num_rows, int num_cols)
  {
- return gko::test::generate_random_lower_triangular_matrix<Mtx>(
- num_rows, num_cols, false,
+ return gko::test::generate_random_matrix<Mtx>(
+ num_rows, num_cols,
  std::uniform_int_distribution<>(num_cols, num_cols),
  std::normal_distribution<>(-1.0, 1.0), rand_engine, ref);
  }
 
  std::unique_ptr<Mtx> gen_l_mtx(int num_rows, int num_cols)
  {
- return gko::test::generate_random_matrix<Mtx>(
- num_rows, num_cols,
+ return gko::test::generate_random_lower_triangular_matrix<Mtx>(
+ num_rows, num_cols, false,
  std::uniform_int_distribution<>(num_cols, num_cols),
  std::normal_distribution<>(-1.0, 1.0), rand_engine, ref);
  }

diff --git a/include/ginkgo/core/solver/lower_trs.hpp b/include/ginkgo/core/solver/lower_trs.hpp
@@ -117,18 +117,26 @@ class LowerTrs : public EnableLinOp<LowerTrs<ValueType, IndexType>>,
  GKO_ENABLE_LIN_OP_FACTORY(LowerTrs, parameters, Factory);
  GKO_ENABLE_BUILD_METHOD(Factory);
 
+ ~LowerTrs() { this->clear_data(); }
+
 protected:
  void apply_impl(const LinOp *b, LinOp *x) const override;
 
  void apply_impl(const LinOp *alpha, const LinOp *b, const LinOp *beta,
  LinOp *x) const override;
 
+ /**
+ * Clears the held data.
+ */
+ void clear_data() const;
+
  /**
  * Generates the analysis structure from the system matrix and the right
  * hand side needed for the level solver.
  */
  void generate();
 
+
  explicit LowerTrs(std::shared_ptr<const Executor> exec)
  : EnableLinOp<LowerTrs>(std::move(exec))
  {}

diff --git a/omp/solver/lower_trs_kernels.cpp b/omp/solver/lower_trs_kernels.cpp
@@ -58,6 +58,13 @@ namespace omp {
 namespace lower_trs {
 
 
+void clear(std::shared_ptr<const OmpExecutor> exec)
+{
+ // This clear kernel is here to allow for a more sophisticated
+ // implementation as for other executors.
+}
+
+
 template <typename ValueType, typename IndexType>
 void generate(std::shared_ptr<const OmpExecutor> exec,
  const matrix::Csr<ValueType, IndexType> *matrix,

diff --git a/reference/solver/lower_trs_kernels.cpp b/reference/solver/lower_trs_kernels.cpp
@@ -54,6 +54,13 @@ namespace reference {
 namespace lower_trs {
 
 
+void clear(std::shared_ptr<const ReferenceExecutor> exec)
+{
+ // This clear kernel is here to allow for a more sophisticated
+ // implementation as for other executors.
+}
+
+
 template <typename ValueType, typename IndexType>
 void generate(std::shared_ptr<const ReferenceExecutor> exec,
  const matrix::Csr<ValueType, IndexType> *matrix,

diff --git a/reference/test/solver/lower_trs.cpp b/reference/test/solver/lower_trs.cpp
@@ -135,8 +135,8 @@ TEST_F(LowerTrs, CanBeCleared)
 
  auto solver_mtx = lower_trs_solver->get_system_matrix();
 
- ASSERT_EQ(lower_trs_solver->get_size(), gko::dim<2>(0, 0));
  ASSERT_EQ(solver_mtx, nullptr);
+ ASSERT_EQ(lower_trs_solver->get_size(), gko::dim<2>(0, 0));
 }