Merge #1724 Fix dynamic cast leads the undefined symbol in clang/msys…

…2/windows This PR fixes dynamic cast leads the undefined symbol in clang/msys2/windows and update the stack macro usage Related PR: #1724
ginkgo-project · Nov 22, 2024 · 681caa0 · 681caa0
2 parents 75434f7 + 1bff73b
commit 681caa0
Show file tree

Hide file tree

Showing 32 changed files with 388 additions and 277 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -94,7 +94,13 @@ if(MSVC)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /bigobj")
 endif()
 if(MINGW OR CYGWIN)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wa,-mbig-obj")
+    if(CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+        # Otherwise, dynamic_cast to the class marked by final will be failed.
+        # https://reviews.llvm.org/D154658 should be relevant
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-assume-unique-vtables")
+    else()
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wa,-mbig-obj")
+    endif()
 endif()
 
 # For now, PGI/NVHPC nvc++ compiler doesn't seem to support

diff --git a/common/cuda_hip/solver/batch_bicgstab_launch.hpp b/common/cuda_hip/solver/batch_bicgstab_launch.hpp
@@ -11,6 +11,7 @@
 #include "core/base/batch_struct.hpp"
 #include "core/matrix/batch_struct.hpp"
 #include "core/solver/batch_bicgstab_kernels.hpp"
+#include "core/solver/batch_dispatch.hpp"
 
 
 namespace gko {
@@ -50,32 +51,28 @@ void launch_apply_kernel(
         device_type<_vtype>* const __restrict__ workspace_data,            \
         const int& block_size, const size_t& shared_size)
 
-#define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH(...) \
-    GKO_INSTANTIATE_FOR_EACH_VALUE_TYPE_VARGS(     \
-        GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, __VA_ARGS__)
-
 #define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH_0_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH, 0, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, 0, false)
 #define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH_1_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH, 1, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, 1, false)
 #define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH_2_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH, 2, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, 2, false)
 #define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH_3_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH, 3, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, 3, false)
 #define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH_4_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH, 4, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, 4, false)
 #define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH_5_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH, 5, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, 5, false)
 #define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH_6_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH, 6, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, 6, false)
 #define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH_7_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH, 7, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, 7, false)
 #define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH_8_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH, 8, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, 8, false)
 #define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH_9_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH, 9, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, 9, false)
 #define GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH_9_TRUE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_BICGSTAB_LAUNCH, 9, true)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_BICGSTAB_LAUNCH, 9, true)
 
 
 }  // namespace batch_bicgstab

diff --git a/common/cuda_hip/solver/batch_cg_launch.hpp b/common/cuda_hip/solver/batch_cg_launch.hpp
@@ -11,6 +11,7 @@
 #include "core/base/batch_struct.hpp"
 #include "core/matrix/batch_struct.hpp"
 #include "core/solver/batch_cg_kernels.hpp"
+#include "core/solver/batch_dispatch.hpp"
 
 
 namespace gko {
@@ -50,24 +51,20 @@ void launch_apply_kernel(
         device_type<_vtype>* const __restrict__ workspace_data,               \
         const int& block_size, const size_t& shared_size)
 
-#define GKO_INSTANTIATE_BATCH_CG_LAUNCH(...)                               \
-    GKO_INSTANTIATE_FOR_EACH_VALUE_TYPE_VARGS(GKO_DECLARE_BATCH_CG_LAUNCH, \
-                                              __VA_ARGS__)
-
 #define GKO_INSTANTIATE_BATCH_CG_LAUNCH_0_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_CG_LAUNCH, 0, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_CG_LAUNCH, 0, false)
 #define GKO_INSTANTIATE_BATCH_CG_LAUNCH_1_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_CG_LAUNCH, 1, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_CG_LAUNCH, 1, false)
 #define GKO_INSTANTIATE_BATCH_CG_LAUNCH_2_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_CG_LAUNCH, 2, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_CG_LAUNCH, 2, false)
 #define GKO_INSTANTIATE_BATCH_CG_LAUNCH_3_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_CG_LAUNCH, 3, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_CG_LAUNCH, 3, false)
 #define GKO_INSTANTIATE_BATCH_CG_LAUNCH_4_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_CG_LAUNCH, 4, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_CG_LAUNCH, 4, false)
 #define GKO_INSTANTIATE_BATCH_CG_LAUNCH_5_FALSE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_CG_LAUNCH, 5, false)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_CG_LAUNCH, 5, false)
 #define GKO_INSTANTIATE_BATCH_CG_LAUNCH_5_TRUE \
-    GKO_BATCH_INSTANTIATE_VARGS(GKO_INSTANTIATE_BATCH_CG_LAUNCH, 5, true)
+    GKO_BATCH_INSTANTIATE(GKO_DECLARE_BATCH_CG_LAUNCH, 5, true)
 
 
 }  // namespace batch_cg

diff --git a/core/CMakeLists.txt b/core/CMakeLists.txt
@@ -10,11 +10,13 @@ set(config_source
 if(GINKGO_BUILD_MPI)
     list(APPEND config_source config/schwarz_config.cpp)
 endif()
-# MSVC: To solve LNK1189, we separate the library as a workaround
+# MSVC: LNK1189 issue
+# CLANG in MSYS2 (MINGW): too many exported symbols
+# We separate the library as a workaround to solve this issue
 # To make ginkgo still be the major library, we make the original to ginkgo_core in MSVC/shared
 # TODO: should think another way to solve it like dllexport or def file
 set(ginkgo_core "ginkgo")
-if(MSVC AND BUILD_SHARED_LIBS)
+if((MSVC OR MINGW) AND BUILD_SHARED_LIBS)
     set(ginkgo_core "ginkgo_core")
 endif()
 
@@ -142,8 +144,8 @@ if(GINKGO_BUILD_MPI)
         distributed/preconditioner/schwarz.cpp)
 endif()
 
-# MSVC/shared: make ginkgo be the major library
-if(MSVC AND BUILD_SHARED_LIBS)
+# MSVC or CLANG/msys2 with shared: make ginkgo be the major library
+if((MSVC OR MINGW) AND BUILD_SHARED_LIBS)
     add_library(ginkgo "")
     target_sources(ginkgo PRIVATE ${config_source})
     ginkgo_compile_features(ginkgo)
@@ -161,7 +163,7 @@ ginkgo_compile_features(${ginkgo_core})
 # add a namespace alias so Ginkgo can always be included as Ginkgo::ginkgo
 # regardless of whether it is installed or added as a subdirectory
 add_library(Ginkgo::ginkgo ALIAS ginkgo)
-if(MSVC AND BUILD_SHARED_LIBS)
+if((MSVC OR MINGW) AND BUILD_SHARED_LIBS)
     target_link_libraries(ginkgo PUBLIC ${ginkgo_core})
 endif()
 target_link_libraries(${ginkgo_core}

diff --git a/core/base/batch_instantiation.hpp b/core/base/batch_instantiation.hpp
@@ -0,0 +1,54 @@
+// SPDX-FileCopyrightText: 2024 The Ginkgo authors
+//
+// SPDX-License-Identifier: BSD-3-Clause
+
+#ifndef GKO_PUBLIC_CORE_BASE_BATCH_INSTANTIATION_HPP_
+#define GKO_PUBLIC_CORE_BASE_BATCH_INSTANTIATION_HPP_
+
+#include <ginkgo/core/base/types.hpp>
+#include <ginkgo/core/matrix/batch_csr.hpp>
+#include <ginkgo/core/matrix/batch_dense.hpp>
+#include <ginkgo/core/matrix/batch_ell.hpp>
+#include <ginkgo/core/matrix/batch_identity.hpp>
+#include <ginkgo/core/preconditioner/batch_jacobi.hpp>
+
+
+namespace gko {
+namespace batch {
+
+
+// just make the call list more consistent
+#define GKO_CALL(_macro, ...) GKO_INDIRECT(_macro(__VA_ARGS__))
+
+#define GKO_BATCH_INSTANTIATE_PRECONDITIONER(_next, ...)            \
+    GKO_INDIRECT(_next(__VA_ARGS__, gko::batch::matrix::Identity)); \
+    GKO_INDIRECT(_next(__VA_ARGS__, gko::batch::preconditioner::Jacobi))
+
+#define GKO_BATCH_INSTANTIATE_MATRIX(_next, ...)                 \
+    GKO_INDIRECT(_next(__VA_ARGS__, gko::batch::matrix::Ell));   \
+    GKO_INDIRECT(_next(__VA_ARGS__, gko::batch::matrix::Dense)); \
+    GKO_INDIRECT(_next(__VA_ARGS__, gko::batch::matrix::Csr))
+
+/**
+ * Instantiates a template for each valid combination of value type, batch
+ * matrix type, and batch preconditioner type. This only allows batch matrix
+ * type and preconditioner type also uses the same value type.
+ *
+ * @param args   the first should be a macro which expands the template
+ *               instantiation (not including the leading `template` specifier).
+ *               Should take three arguments, where the first is replaced by the
+ *               value type, the second by the matrix, and the third by the
+ *               preconditioner.
+ *
+ * @note the second and third arguments only accept the base type.s
+ */
+#define GKO_INSTANTIATE_FOR_BATCH_VALUE_MATRIX_PRECONDITIONER(...) \
+    GKO_CALL(GKO_BATCH_INSTANTIATE_MATRIX,                         \
+             GKO_BATCH_INSTANTIATE_PRECONDITIONER,                 \
+             GKO_INSTANTIATE_FOR_EACH_VALUE_TYPE_VARGS, __VA_ARGS__)
+
+
+}  // namespace batch
+}  // namespace gko
+
+#endif  // GKO_PUBLIC_CORE_BASE_BATCH_INSTANTIATION_HPP_
diff --git a/core/device_hooks/common_kernels.inc.cpp b/core/device_hooks/common_kernels.inc.cpp
@@ -5,6 +5,7 @@
 #include <ginkgo/core/base/exception_helpers.hpp>
 #include <ginkgo/core/base/types.hpp>
 
+#include "core/base/batch_instantiation.hpp"
 #include "core/base/batch_multi_vector_kernels.hpp"
 #include "core/base/device_matrix_data_kernels.hpp"
 #include "core/base/index_set_kernels.hpp"
@@ -168,6 +169,13 @@
     _macro(ValueType, ValueTypeKrylovBases) GKO_NOT_COMPILED(GKO_HOOK_MODULE); \
     GKO_INSTANTIATE_FOR_EACH_CB_GMRES_CONST_TYPE(_macro)
 
+#define GKO_STUB_BATCH_VALUE_MATRIX_PRECONDITIONER(_declare, _wrapper)         \
+    template <typename ValueType, typename BatchMatrixType, typename PrecType> \
+    _declare(ValueType, BatchMatrixType, PrecType)                             \
+        GKO_NOT_COMPILED(GKO_HOOK_MODULE);                                     \
+    GKO_INSTANTIATE_FOR_BATCH_VALUE_MATRIX_PRECONDITIONER(_wrapper)
+
+
 namespace gko {
 namespace kernels {
 namespace GKO_HOOK_MODULE {
@@ -421,7 +429,9 @@ GKO_STUB_VALUE_AND_INDEX_TYPE(GKO_DECLARE_DIAGONAL_FILL_IN_MATRIX_DATA_KERNEL);
 namespace batch_bicgstab {
 
 
-GKO_STUB_VALUE_TYPE(GKO_DECLARE_BATCH_BICGSTAB_APPLY_KERNEL);
+GKO_STUB_BATCH_VALUE_MATRIX_PRECONDITIONER(
+    GKO_DECLARE_BATCH_BICGSTAB_APPLY_KERNEL,
+    GKO_DECLARE_BATCH_BICGSTAB_APPLY_KERNEL_WRAPPER);
 
 
 }  // namespace batch_bicgstab
@@ -430,7 +440,9 @@ GKO_STUB_VALUE_TYPE(GKO_DECLARE_BATCH_BICGSTAB_APPLY_KERNEL);
 namespace batch_cg {
 
 
-GKO_STUB_VALUE_TYPE(GKO_DECLARE_BATCH_CG_APPLY_KERNEL);
+GKO_STUB_BATCH_VALUE_MATRIX_PRECONDITIONER(
+    GKO_DECLARE_BATCH_CG_APPLY_KERNEL,
+    GKO_DECLARE_BATCH_CG_APPLY_KERNEL_WRAPPER);
 
 
 }  // namespace batch_cg

diff --git a/core/solver/batch_bicgstab.cpp b/core/solver/batch_bicgstab.cpp
@@ -7,8 +7,14 @@
 #include <ginkgo/core/base/batch_lin_op.hpp>
 #include <ginkgo/core/base/batch_multi_vector.hpp>
 #include <ginkgo/core/base/math.hpp>
+#include <ginkgo/core/matrix/batch_csr.hpp>
+#include <ginkgo/core/matrix/batch_dense.hpp>
+#include <ginkgo/core/matrix/batch_ell.hpp>
+#include <ginkgo/core/matrix/batch_identity.hpp>
+#include <ginkgo/core/preconditioner/batch_jacobi.hpp>
 
 #include "core/base/batch_multi_vector_kernels.hpp"
+#include "core/base/dispatch_helper.hpp"
 #include "core/solver/batch_bicgstab_kernels.hpp"
 
 
@@ -45,14 +51,19 @@ void Bicgstab<ValueType>::solver_apply(
     const MultiVector<ValueType>* b, MultiVector<ValueType>* x,
     log::detail::log_data<remove_complex<ValueType>>* log_data) const
 {
-    using MVec = MultiVector<ValueType>;
     const kernels::batch_bicgstab::settings<remove_complex<ValueType>> settings{
         this->max_iterations_, static_cast<real_type>(this->residual_tol_),
         parameters_.tolerance_type};
     auto exec = this->get_executor();
-    exec->run(bicgstab::make_apply(settings, this->system_matrix_.get(),
-                                   this->preconditioner_.get(), b, x,
-                                   *log_data));
+
+    run<matrix::Dense<ValueType>, matrix::Csr<ValueType>,
+        matrix::Ell<ValueType>>(this->system_matrix_.get(), [&](auto matrix) {
+        run<matrix::Identity<ValueType>, preconditioner::Jacobi<ValueType>>(
+            this->preconditioner_.get(), [&](auto preconditioner) {
+                exec->run(bicgstab::make_apply(settings, matrix, preconditioner,
+                                               b, x, *log_data));
+            });
+    });
 }
 
 

diff --git a/core/solver/batch_bicgstab_kernels.hpp b/core/solver/batch_bicgstab_kernels.hpp
@@ -174,19 +174,25 @@ storage_config compute_shared_storage(const int available_shared_mem,
 }  // namespace batch_bicgstab
 
 
-#define GKO_DECLARE_BATCH_BICGSTAB_APPLY_KERNEL(_type)                       \
+#define GKO_DECLARE_BATCH_BICGSTAB_APPLY_KERNEL(_type, _matrix, _prec)       \
     void apply(                                                              \
         std::shared_ptr<const DefaultExecutor> exec,                         \
         const gko::kernels::batch_bicgstab::settings<remove_complex<_type>>& \
             options,                                                         \
-        const batch::BatchLinOp* a, const batch::BatchLinOp* preconditioner, \
+        const _matrix* a, const _prec* preconditioner,                       \
         const batch::MultiVector<_type>* b, batch::MultiVector<_type>* x,    \
         gko::batch::log::detail::log_data<remove_complex<_type>>& logdata)
 
+#define GKO_DECLARE_BATCH_BICGSTAB_APPLY_KERNEL_WRAPPER(_vtype, _matrix, \
+                                                        _precond)        \
+    GKO_DECLARE_BATCH_BICGSTAB_APPLY_KERNEL(_vtype, _matrix<_vtype>,     \
+                                            _precond<_vtype>)
 
-#define GKO_DECLARE_ALL_AS_TEMPLATES \
-    template <typename ValueType>    \
-    GKO_DECLARE_BATCH_BICGSTAB_APPLY_KERNEL(ValueType)
+
+#define GKO_DECLARE_ALL_AS_TEMPLATES                                           \
+    template <typename ValueType, typename BatchMatrixType, typename PrecType> \
+    GKO_DECLARE_BATCH_BICGSTAB_APPLY_KERNEL(ValueType, BatchMatrixType,        \
+                                            PrecType)
 
 
 GKO_DECLARE_FOR_ALL_EXECUTOR_NAMESPACES(batch_bicgstab,

diff --git a/core/solver/batch_cg.cpp b/core/solver/batch_cg.cpp
@@ -7,11 +7,16 @@
 #include <ginkgo/core/base/batch_lin_op.hpp>
 #include <ginkgo/core/base/batch_multi_vector.hpp>
 #include <ginkgo/core/base/math.hpp>
+#include <ginkgo/core/matrix/batch_csr.hpp>
+#include <ginkgo/core/matrix/batch_dense.hpp>
+#include <ginkgo/core/matrix/batch_ell.hpp>
+#include <ginkgo/core/matrix/batch_identity.hpp>
+#include <ginkgo/core/preconditioner/batch_jacobi.hpp>
 
 #include "core/base/batch_multi_vector_kernels.hpp"
+#include "core/base/dispatch_helper.hpp"
 #include "core/solver/batch_cg_kernels.hpp"
 
-
 namespace gko {
 namespace batch {
 namespace solver {
@@ -49,8 +54,17 @@ void Cg<ValueType>::solver_apply(
         this->max_iterations_, static_cast<real_type>(this->residual_tol_),
         parameters_.tolerance_type};
     auto exec = this->get_executor();
-    exec->run(cg::make_apply(settings, this->system_matrix_.get(),
-                             this->preconditioner_.get(), b, x, *log_data));
+
+    run<batch::matrix::Dense<ValueType>, batch::matrix::Csr<ValueType>,
+        batch::matrix::Ell<ValueType>>(
+        this->system_matrix_.get(), [&](auto matrix) {
+            run<batch::matrix::Identity<ValueType>,
+                batch::preconditioner::Jacobi<ValueType>>(
+                this->preconditioner_.get(), [&](auto preconditioner) {
+                    exec->run(cg::make_apply(settings, matrix, preconditioner,
+                                             b, x, *log_data));
+                });
+        });
 }
 
 

diff --git a/core/solver/batch_cg_kernels.hpp b/core/solver/batch_cg_kernels.hpp
@@ -162,19 +162,22 @@ storage_config compute_shared_storage(const int available_shared_mem,
 }  // namespace batch_cg
 
 
-#define GKO_DECLARE_BATCH_CG_APPLY_KERNEL(_type)                               \
-    void apply(                                                                \
-        std::shared_ptr<const DefaultExecutor> exec,                           \
-        const gko::kernels::batch_cg::settings<remove_complex<_type>>&         \
-            options,                                                           \
-        const batch::BatchLinOp* mat, const batch::BatchLinOp* preconditioner, \
-        const batch::MultiVector<_type>* b, batch::MultiVector<_type>* x,      \
+#define GKO_DECLARE_BATCH_CG_APPLY_KERNEL(_type, _matrix, _prec)          \
+    void apply(                                                           \
+        std::shared_ptr<const DefaultExecutor> exec,                      \
+        const gko::kernels::batch_cg::settings<remove_complex<_type>>&    \
+            options,                                                      \
+        const _matrix* mat, const _prec* preconditioner,                  \
+        const batch::MultiVector<_type>* b, batch::MultiVector<_type>* x, \
         gko::batch::log::detail::log_data<remove_complex<_type>>& logdata)
 
+#define GKO_DECLARE_BATCH_CG_APPLY_KERNEL_WRAPPER(_vtype, _matrix, _precond) \
+    GKO_DECLARE_BATCH_CG_APPLY_KERNEL(_vtype, _matrix<_vtype>, _precond<_vtype>)
 
-#define GKO_DECLARE_ALL_AS_TEMPLATES \
-    template <typename ValueType>    \
-    GKO_DECLARE_BATCH_CG_APPLY_KERNEL(ValueType)
+
+#define GKO_DECLARE_ALL_AS_TEMPLATES                                           \
+    template <typename ValueType, typename BatchMatrixType, typename PrecType> \
+    GKO_DECLARE_BATCH_CG_APPLY_KERNEL(ValueType, BatchMatrixType, PrecType)
 
 
 GKO_DECLARE_FOR_ALL_EXECUTOR_NAMESPACES(batch_cg, GKO_DECLARE_ALL_AS_TEMPLATES);