chore: import upstream snapshot with attribution

libnd4j/include/loops/cpu/comb_compilation_units/broadcast_bool_instantiation_template.cpp.in

@@ -0,0 +1,86 @@
+/* ******************************************************************************
+*
+*
+* This program and the accompanying materials are made available under the
+* terms of the Apache License, Version 2.0 which is available at
+* https://www.apache.org/licenses/LICENSE-2.0.
+*
+*  See the NOTICE file distributed with this work for additional
+*  information regarding copyright ownership.
+* Unless required by applicable law or agreed to in writing, software
+* distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+* License for the specific language governing permissions and limitations
+* under the License.
+*
+* SPDX-License-Identifier: Apache-2.0
+******************************************************************************/
+
+//
+// BroadcastBool template instantiations
+//
+
+#include <loops/pairwise_instantiations.h>
+#include <types/type_promote.h>
+#include <loops/broadcasting_bool.h>
+#include <loops/cpu/broadcasting_bool.hpp>
+/*
+*
+* Function Instantiation:
+* BroadcastBool::exec instantiated for types in @COMB1@, @COMB2@
+*/
+
+ITERATE_COMBINATIONS(
+   SD_COMMON_TYPES_PART_@COMB1@,
+   SD_COMMON_TYPES_PART_@COMB2@,
+   INSTANT_PROCESS_COMBINATION,
+   functions::broadcast::BroadcastBool,
+   ::exec(int opNum, const void *x, const sd::LongType *xShapeInfo, const void *y,
+          const sd::LongType *yShapeInfo, void *z, const sd::LongType *zShapeInfo,
+          void *extraParams, sd::LongType *dimension, sd::LongType dimensionLength,
+          const sd::LongType *tadOnlyShapeInfo, const sd::LongType *tadOffsets,
+          const sd::LongType *tadOnlyShapeInfoZ, const sd::LongType *tadOffsetsZ,
+          sd::LongType start, sd::LongType stop)
+)
+
+/*
+*
+* Function Instantiation:
+* BroadcastBool::exec (simple version) instantiated for types in @COMB1@, @COMB2@
+*/
+
+ITERATE_COMBINATIONS(
+   SD_COMMON_TYPES_PART_@COMB1@,
+   SD_COMMON_TYPES_PART_@COMB2@,
+   INSTANT_PROCESS_COMBINATION,
+   functions::broadcast::BroadcastBool,
+   ::exec(int opNum, const void *x, const sd::LongType *xShapeInfo, const void *y,
+          const sd::LongType *yShapeInfo, void *z, const sd::LongType *zShapeInfo,
+          void *extraParams)
+)
+
+/*
+*
+* Function Instantiation:
+* BroadcastBool::execInverse instantiated for types in @COMB1@, @COMB2@
+*/
+
+ITERATE_COMBINATIONS(
+   SD_COMMON_TYPES_PART_@COMB1@,
+   SD_COMMON_TYPES_PART_@COMB2@,
+   INSTANT_PROCESS_COMBINATION,
+   functions::broadcast::BroadcastBool,
+   ::execInverse(int opNum, const void *x, const sd::LongType *xShapeInfo, const void *y,
+                 const sd::LongType *yShapeInfo, void *z, const sd::LongType *zShapeInfo,
+                 void *extraParams, sd::LongType *dimension, sd::LongType dimensionLength,
+                 const sd::LongType *tadOnlyShapeInfo, const sd::LongType *tadOffsets,
+                 const sd::LongType *tadOnlyShapeInfoZ, const sd::LongType *tadOffsetsZ,
+                 sd::LongType start, sd::LongType stop)
+)
+
+/*
+*
+* Explicit instantiation for bool types (since BroadcastBool typically uses bool output)
+*/
+
+template class functions::broadcast::BroadcastBool<bool, bool>;

libnd4j/include/loops/cpu/comb_compilation_units/broadcast_bool_instantiation_template_2.cpp.in

@@ -0,0 +1,31 @@
+/* ******************************************************************************
+*
+*
+* This program and the accompanying materials are made available under the
+* terms of the Apache License, Version 2.0 which is available at
+* https://www.apache.org/licenses/LICENSE-2.0.
+*
+*  See the NOTICE file distributed with this work for additional
+*  information regarding copyright ownership.
+* Unless required by applicable law or agreed to in writing, software
+* distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+* License for the specific language governing permissions and limitations
+* under the License.
+*
+* SPDX-License-Identifier: Apache-2.0
+******************************************************************************/
+
+//
+// BroadcastBool template instantiations - Direct instantiation version
+// This file is designed for 2-type combinations
+//
+
+#include <loops/cpu/broadcasting_bool.hpp>
+#include <types/types.h>
+
+// Direct instantiations will be inserted here by the CMake template system
+// The handler will generate instantiations like:
+// template void functions::broadcast::BroadcastBool<float, bool>::exec(...);
+// template void functions::broadcast::BroadcastBool<double, bool>::exec(...);
+// etc.

libnd4j/include/loops/cpu/comb_compilation_units/broadcast_instantiation_template_3.cpp.in

@@ -0,0 +1,95 @@
+/* ******************************************************************************
+*
+*
+* This program and the accompanying materials are made available under the
+* terms of the Apache License, Version 2.0 which is available at
+* https://www.apache.org/licenses/LICENSE-2.0.
+*
+*  See the NOTICE file distributed with this work for additional
+*  information regarding copyright ownership.
+* Unless required by applicable law or agreed to in writing, software
+* distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+* License for the specific language governing permissions and limitations
+* under the License.
+*
+* SPDX-License-Identifier: Apache-2.0
+******************************************************************************/
+
+//
+// Broadcast template instantiations
+//
+
+#include <loops/pairwise_instantiations.h>
+#include <types/type_promote.h>
+#include <loops/cpu/broadcasting.hpp>
+
+/*
+*
+* Function Instantiation:
+* Broadcast::exec instantiated for types in @COMB1@, @COMB2@, @COMB3@
+*/
+
+ITERATE_COMBINATIONS_3(
+   SD_COMMON_TYPES_PART_@COMB1@,
+   SD_COMMON_TYPES_PART_@COMB2@,
+   SD_COMMON_TYPES_PART_@COMB3@,
+   INSTANT_PROCESS_COMBINATION_3,
+   functions::broadcast::Broadcast,
+   ::exec(int opNum, const void *x, const sd::LongType *xShapeInfo, const void *y,
+          const sd::LongType *yShapeInfo, void *z, const sd::LongType *zShapeInfo,
+          sd::LongType *dimension, sd::LongType dimensionLength,
+          const sd::LongType *tadOnlyShapeInfo, const sd::LongType *tadOffsets,
+          const sd::LongType *tadOnlyShapeInfoZ, const sd::LongType *tadOffsetsZ,
+          sd::LoopKind::Kind loopKind, sd::LongType start, sd::LongType stop)
+)
+
+/*
+*
+* Function Instantiation:
+* Broadcast::exec (simple version) instantiated for types in @COMB1@, @COMB2@, @COMB3@
+*/
+
+ITERATE_COMBINATIONS_3(
+   SD_COMMON_TYPES_PART_@COMB1@,
+   SD_COMMON_TYPES_PART_@COMB2@,
+   SD_COMMON_TYPES_PART_@COMB3@,
+   INSTANT_PROCESS_COMBINATION_3,
+   functions::broadcast::Broadcast,
+   ::exec(int opNum, const void *x, const sd::LongType *xShapeInfo, const void *y,
+          const sd::LongType *yShapeInfo, void *z, const sd::LongType *zShapeInfo)
+)
+
+/*
+*
+* Function Instantiation:
+* Broadcast::execInverse instantiated for types in @COMB1@, @COMB2@, @COMB3@
+*/
+
+ITERATE_COMBINATIONS_3(
+   SD_COMMON_TYPES_PART_@COMB1@,
+   SD_COMMON_TYPES_PART_@COMB2@,
+   SD_COMMON_TYPES_PART_@COMB3@,
+   INSTANT_PROCESS_COMBINATION_3,
+   functions::broadcast::Broadcast,
+   ::execInverse(int opNum, const void *x, const sd::LongType *xShapeInfo, const void *y,
+                 const sd::LongType *yShapeInfo, void *z, const sd::LongType *zShapeInfo,
+                 sd::LongType *dimension, sd::LongType dimensionLength,
+                 const sd::LongType *tadOnlyShapeInfo, const sd::LongType *tadOffsets,
+                 const sd::LongType *tadOnlyShapeInfoZ, const sd::LongType *tadOffsetsZ,
+                 sd::LongType start, sd::LongType stop)
+)
+
+/*
+*
+* Function Instantiation:
+* BroadcastBool::exec instantiated for bool types
+*/
+
+template void functions::broadcast::BroadcastBool<bool, bool>::exec(
+   int opNum, const void *x, const sd::LongType *xShapeInfo, const void *y,
+   const sd::LongType *yShapeInfo, void *z, const sd::LongType *zShapeInfo,
+   void *extraParams, sd::LongType *dimension, sd::LongType dimensionLength,
+   const sd::LongType *tadOnlyShapeInfo, const sd::LongType *tadOffsets,
+   const sd::LongType *tadOnlyShapeInfoZ, const sd::LongType *tadOffsetsZ,
+   sd::LongType start, sd::LongType stop);

libnd4j/include/loops/cpu/comb_compilation_units/broadcast_int_instantiation_template_1.cpp.in

@@ -0,0 +1,57 @@
+/* ******************************************************************************
+*
+*
+* This program and the accompanying materials are made available under the
+* terms of the Apache License, Version 2.0 which is available at
+* https://www.apache.org/licenses/LICENSE-2.0.
+*
+*  See the NOTICE file distributed with this work for additional
+*  information regarding copyright ownership.
+* Unless required by applicable law or agreed to in writing, software
+* distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+* License for the specific language governing permissions and limitations
+* under the License.
+*
+* SPDX-License-Identifier: Apache-2.0
+******************************************************************************/
+
+//
+// BroadcastInt template instantiations
+//
+
+#include <loops/cpu/broadcasting_int.hpp>
+#include <loops/pairwise_instantiations.h>
+#include <system/type_boilerplate.h>
+
+/*
+* Macro for instantiating BroadcastInt for a single type
+* This follows the pattern where BroadcastInt is specialized for integer types only
+*/
+
+#define INSTANTIATE_BROADCAST_INT(TYPE_TUPLE) \
+    template void functions::broadcast::BroadcastInt<GET_SECOND(TYPE_TUPLE)>::exec( \
+        int opNum, const void *x, const sd::LongType *xShapeInfo, const void *y, \
+        const sd::LongType *yShapeInfo, void *z, const sd::LongType *zShapeInfo, \
+        sd::LongType *dimension, sd::LongType dimensionLength, \
+        const sd::LongType *tadOnlyShapeInfo, const sd::LongType *tadOffsets, \
+        const sd::LongType *tadOnlyShapeInfoZ, const sd::LongType *tadOffsetsZ, \
+        sd::LongType start, sd::LongType stop); \
+    \
+    template void functions::broadcast::BroadcastInt<GET_SECOND(TYPE_TUPLE)>::exec( \
+        int opNum, const void *x, const sd::LongType *xShapeInfo, const void *y, \
+        const sd::LongType *yShapeInfo, void *z, const sd::LongType *zShapeInfo); \
+    \
+    template void functions::broadcast::BroadcastInt<GET_SECOND(TYPE_TUPLE)>::execInverse( \
+        int opNum, const void *x, const sd::LongType *xShapeInfo, const void *y, \
+        const sd::LongType *yShapeInfo, void *z, const sd::LongType *zShapeInfo, \
+        sd::LongType *dimension, sd::LongType dimensionLength, \
+        const sd::LongType *tadOnlyShapeInfo, const sd::LongType *tadOffsets, \
+        const sd::LongType *tadOnlyShapeInfoZ, const sd::LongType *tadOffsetsZ, \
+        sd::LongType start, sd::LongType stop);
+
+/*
+* Function Instantiation:
+* BroadcastInt instantiated for all integer types in SD_INTEGER_TYPES_PART_@COMB1@
+*/
+ITERATE_LIST((SD_INTEGER_TYPES_PART_@COMB1@), INSTANTIATE_BROADCAST_INT)

libnd4j/include/loops/cpu/comb_compilation_units/indexreduction_instantiation_template_2.cpp.in

@@ -0,0 +1,35 @@
+/*
+ *  ******************************************************************************
+ *  *
+ *  *
+ *  * This program and the accompanying materials are made available under the
+ *  * terms of the Apache License, Version 2.0 which is available at
+ *  * https://www.apache.org/licenses/LICENSE-2.0.
+ *  *
+ *  * See the NOTICE file distributed with this work for additional
+ *  * information regarding copyright ownership.
+ *  * Unless required by applicable law or agreed to in writing, software
+ *  * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ *  * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ *  * License for the specific language governing permissions and limitations
+ *  * under the License.
+ *  *
+ *  * SPDX-License-Identifier: Apache-2.0
+ *  *****************************************************************************
+ */
+
+//
+// Created by agibsonccc on 12/26/24.
+//
+
+#include <helpers/cpu/loops/IndexReductionLoops.hpp>
+#include <system/type_boilerplate.h>
+// Note: Instantiations are generated to prevent compiler memory issues
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_COMMON_TYPES_PART_@COMB2@,
+    template class sd::IndexReductionLoops,
+    IndexReductionLoops,
+    ;
+)

libnd4j/include/loops/cpu/comb_compilation_units/pairwise_instantiation_template_2.cpp.in

@@ -0,0 +1,40 @@
+// pairwise_instantiation_template_3.cpp.in
+
+/*
+ *  ******************************************************************************
+ *  *
+ *  *
+ *  * This program and the accompanying materials are made available under the
+ *  * terms of the Apache License, Version 2.0 which is available at
+ *  * https://www.apache.org/licenses/LICENSE-2.0.
+ *  *
+ *  * See the NOTICE file distributed with this work for additional
+ *  * information regarding copyright ownership.
+ *  * Unless required by applicable law or agreed to in writing, software
+ *  * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ *  * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ *  * License for the specific language governing permissions and limitations
+ *  * under the License.
+ *  *
+ *  * SPDX-License-Identifier: Apache-2.0
+ *  *****************************************************************************
+ */
+
+//
+// Created by agibsonccc on 11/22/24.
+//
+
+#include <loops/pairwise_instantiations.h>
+#include <types/type_promote.h>
+#include <loops/cpu/pairwise.hpp>
+// Note: Instantiations are generated to prevent compiler memory issues
+
+
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_COMMON_TYPES_PART_@COMB2@,
+    CALLBACK_INSTANTIATE_PROMOTE,
+    promote,
+    ;
+)

libnd4j/include/loops/cpu/comb_compilation_units/pairwise_instantiation_template_3.cpp.in

@@ -0,0 +1,44 @@
+/*
+*  ******************************************************************************
+*  *
+*  *
+*  * This program and the accompanying materials are made available under the
+*  * terms of the Apache License, Version 2.0 which is available at
+*  * https://www.apache.org/licenses/LICENSE-2.0.
+*  *
+*  * See the NOTICE file distributed with this work for additional
+*  * information regarding copyright ownership.
+*  * Unless required by applicable law or agreed to in writing, software
+*  * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+*  * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+*  * License for the specific language governing permissions and limitations
+*  * under the License.
+*  *
+*  * SPDX-License-Identifier: Apache-2.0
+*  *****************************************************************************
+ */
+
+//
+// Created by agibsonccc on 11/22/24.
+//
+
+#include <loops/pairwise_instantiations.h>
+#include <types/type_promote.h>
+#include <loops/cpu/pairwise.hpp>
+
+/*
+*
+* Function Instantiation:
+* PairWiseTransform::exec instantiated for types in @COMB1@, @COMB2@, @COMB3@
+ */
+
+ITERATE_COMBINATIONS_3(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_COMMON_TYPES_PART_@COMB2@,
+    SD_COMMON_TYPES_PART_@COMB3@,
+    INSTANT_PROCESS_COMBINATION_3,
+    functions::pairwise_transforms::PairWiseTransform,
+    ::exec(int opNum, const void *x, const sd::LongType *xShapeInfo, const void *y,
+     const sd::LongType *yShapeInfo, void *z, const sd::LongType *zShapeInfo,
+     void *extraParams, sd::LongType start, sd::LongType stop)
+)

libnd4j/include/loops/cpu/comb_compilation_units/random_instantiation_template_1.cpp.in

@@ -0,0 +1,66 @@
+/* ******************************************************************************
+ *
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Apache License, Version 2.0 which is available at
+ * https://www.apache.org/licenses/LICENSE-2.0.
+ *
+ *  See the NOTICE file distributed with this work for additional
+ *  information regarding copyright ownership.
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ * License for the specific language governing permissions and limitations
+ * under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ ******************************************************************************/
+
+//
+// Created by agibsonccc on 1/3/25.
+//
+
+#include <loops/cpu/random.hpp>
+#include <loops/pairwise_instantiations.h>
+#include <system/type_boilerplate.h>
+
+/*
+ * Function Instantiation:
+ * RandomFunction instantiated for float types in @COMB1@
+ */
+
+#define LIST_CALLBACK_EXEC_TRANSFORM_1(INPUT) \
+    template void functions::random::RandomFunction<GET_SECOND(INPUT)>::execTransform( \
+        int opNum, \
+        sd::Pointer state, \
+        const void *x, \
+        const sd::LongType *xShapeInfo, \
+        void *z, \
+        const sd::LongType *zShapeInfo, \
+        void *extraArguments);
+
+ITERATE_LIST((SD_COMMON_TYPES_PART_@COMB1@), LIST_CALLBACK_EXEC_TRANSFORM_1)
+
+#define LIST_CALLBACK_EXEC_TRANSFORM_2(INPUT) \
+    template void functions::random::RandomFunction<GET_SECOND(INPUT)>::execTransform( \
+        int opNum, \
+        sd::Pointer state, \
+        const void *x, \
+        const sd::LongType *xShapeInfo, \
+        const void *y, \
+        const sd::LongType *yShapeInfo, \
+        void *z, \
+        const sd::LongType *zShapeInfo, \
+        void *extraArguments);
+
+ITERATE_LIST((SD_COMMON_TYPES_PART_@COMB1@), LIST_CALLBACK_EXEC_TRANSFORM_2)
+
+#define LIST_CALLBACK_EXEC_TRANSFORM_3(INPUT) \
+    template void functions::random::RandomFunction<GET_SECOND(INPUT)>::execTransform( \
+        int opNum, \
+        sd::Pointer state, \
+        void *z, \
+        const sd::LongType *zShapeInfo, \
+        void *extraArguments);
+
+ITERATE_LIST((SD_COMMON_TYPES_PART_@COMB1@), LIST_CALLBACK_EXEC_TRANSFORM_3)

libnd4j/include/loops/cpu/comb_compilation_units/reduce3_instantiation_template_2.cpp.in

@@ -0,0 +1,62 @@
+/* ******************************************************************************
+ *
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Apache License, Version 2.0 which is available at
+ * https://www.apache.org/licenses/LICENSE-2.0.
+ *
+ *  See the NOTICE file distributed with this work for additional
+ *  information regarding copyright ownership.
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ * License for the specific language governing permissions and limitations
+ * under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ ******************************************************************************/
+
+//
+// Created by agibsonccc on 1/3/25.
+//
+
+#include <loops/cpu/reduce3.hpp>
+#include <loops/pairwise_instantiations.h>
+#include <system/type_boilerplate.h>
+
+/*
+ * Function Instantiation:
+ * Reduce3 instantiated for types @COMB1@, @COMB2@
+ */
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce3::Reduce3,
+    ::execScalar(const int opNum, const void *vx, const sd::LongType *xShapeInfo, void *extraParamsVals, const void *vy, const sd::LongType *yShapeInfo, void *vz, const sd::LongType *zShapeInfo)
+)
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce3::Reduce3,
+    ::exec(int opNum, const void *vx, const sd::LongType *xShapeInfo, void *extraParamsVals, const void *vy, const sd::LongType *yShapeInfo, void *vz, const sd::LongType *zShapeInfo, sd::LongType *dimension, sd::LongType dimensionLength, sd::LongType start, sd::LongType stop)
+)
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce3::Reduce3,
+    ::exec(int opNum, const void *vx, const sd::LongType *xShapeInfo, void *extraParamsVals, const void *vy, const sd::LongType *yShapeInfo, void *vz, const sd::LongType *zShapeInfo, sd::LongType *dimension, sd::LongType dimensionLength, const sd::LongType *tadShapeInfo, const sd::LongType *tadOffsets, sd::LongType start, sd::LongType stop)
+)
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce3::Reduce3,
+    ::execAll(int opNum, const void *vx, const sd::LongType *xShapeInfo, void *extraParamsVals, const void *vy, const sd::LongType *yShapeInfo, void *vz, const sd::LongType *zShapeInfo, sd::LongType *dimension, sd::LongType dimensionLength, const sd::LongType *xTadShapeInfo, const sd::LongType *xOffsets, const sd::LongType *yTadShapeInfo, const sd::LongType *yOffsets, sd::LongType start, sd::LongType stop)
+)

libnd4j/include/loops/cpu/comb_compilation_units/reduce_bool_instantiation_template_2.cpp.in

@@ -0,0 +1,54 @@
+/* ******************************************************************************
+ *
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Apache License, Version 2.0 which is available at
+ * https://www.apache.org/licenses/LICENSE-2.0.
+ *
+ *  See the NOTICE file distributed with this work for additional
+ *  information regarding copyright ownership.
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ * License for the specific language governing permissions and limitations
+ * under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ ******************************************************************************/
+
+//
+// Created by agibsonccc on 1/3/25.
+//
+
+#include <loops/cpu/reduce/reduce_bool.hpp>
+#include <loops/pairwise_instantiations.h>
+#include <system/type_boilerplate.h>
+
+/*
+ * Function Instantiation:
+ * ReduceBoolFunction instantiated for types @COMB1@, @COMB2@
+ */
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_BOOL_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce::ReduceBoolFunction,
+    ::execScalar(const int opNum, const void *x, const sd::LongType *xShapeInfo, void *extraParams)
+)
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_BOOL_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce::ReduceBoolFunction,
+    ::execScalar(const int opNum, const void *x, const sd::LongType *xShapeInfo, void *extraParams, void *z, const sd::LongType *zShapeInfo)
+)
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_BOOL_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce::ReduceBoolFunction,
+    ::exec(int opNum, sd::memory::Workspace *workspace, const void *vx, const sd::LongType *xShapeInfo, void *vextraParams, void *vz, const sd::LongType *zShapeInfo, const sd::LongType *dims)
+)

libnd4j/include/loops/cpu/comb_compilation_units/reduce_float_instantiation_template_2.cpp.in

@@ -0,0 +1,54 @@
+/* ******************************************************************************
+ *
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Apache License, Version 2.0 which is available at
+ * https://www.apache.org/licenses/LICENSE-2.0.
+ *
+ *  See the NOTICE file distributed with this work for additional
+ *  information regarding copyright ownership.
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ * License for the specific language governing permissions and limitations
+ * under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ ******************************************************************************/
+
+//
+// Created by agibsonccc on 1/3/25.
+//
+
+#include <loops/cpu/reduce/reduce_float.hpp>
+#include <loops/pairwise_instantiations.h>
+#include <system/type_boilerplate.h>
+
+/*
+ * Function Instantiation:
+ * ReduceFloatFunction instantiated for types @COMB1@, @COMB2@
+ */
+
+ITERATE_COMBINATIONS(
+    SD_NUMERIC_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce::ReduceFloatFunction,
+    ::execScalar(const int opNum, const void *x, const sd::LongType *xShapeInfo, void *extraParams)
+)
+
+ITERATE_COMBINATIONS(
+    SD_NUMERIC_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce::ReduceFloatFunction,
+    ::execScalar(const int opNum, const void *x, const sd::LongType *xShapeInfo, void *extraParams, void *z, const sd::LongType *zShapeInfo)
+)
+
+ITERATE_COMBINATIONS(
+    SD_NUMERIC_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce::ReduceFloatFunction,
+    ::exec(int opNum, sd::memory::Workspace *workspace, const void *vx, const sd::LongType *xShapeInfo, void *vextraParams, void *vz, const sd::LongType *zShapeInfo, const sd::LongType *dims)
+)

libnd4j/include/loops/cpu/comb_compilation_units/reduce_long_instantiation_template_2.cpp.in

@@ -0,0 +1,54 @@
+/* ******************************************************************************
+ *
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Apache License, Version 2.0 which is available at
+ * https://www.apache.org/licenses/LICENSE-2.0.
+ *
+ *  See the NOTICE file distributed with this work for additional
+ *  information regarding copyright ownership.
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ * License for the specific language governing permissions and limitations
+ * under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ ******************************************************************************/
+
+//
+// Created by agibsonccc on 1/3/25.
+//
+
+#include <loops/cpu/reduce/reduce_long.hpp>
+#include <loops/pairwise_instantiations.h>
+#include <system/type_boilerplate.h>
+
+/*
+ * Function Instantiation:
+ * ReduceLongFunction instantiated for types @COMB1@, @COMB2@
+ */
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_LONG_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce::ReduceLongFunction,
+    ::execScalar(const int opNum, const void *x, const sd::LongType *xShapeInfo, void *extraParams)
+)
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_LONG_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce::ReduceLongFunction,
+    ::execScalar(const int opNum, const void *x, const sd::LongType *xShapeInfo, void *extraParams, void *z, const sd::LongType *zShapeInfo)
+)
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_LONG_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION_2,
+    functions::reduce::ReduceLongFunction,
+    ::exec(const int opNum, sd::memory::Workspace *workspace, const void *vx, const sd::LongType *xShapeInfo, void *vextraParams, void *vz, const sd::LongType *zShapeInfo, sd::LongType *dims)
+)

libnd4j/include/loops/cpu/comb_compilation_units/reduce_same_instantiation_template_1.cpp.in

@@ -0,0 +1,63 @@
+/* ******************************************************************************
+ *
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Apache License, Version 2.0 which is available at
+ * https://www.apache.org/licenses/LICENSE-2.0.
+ *
+ *  See the NOTICE file distributed with this work for additional
+ *  information regarding copyright ownership.
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ * License for the specific language governing permissions and limitations
+ * under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ ******************************************************************************/
+
+//
+// Created by agibsonccc on 1/3/25.
+//
+
+#include <loops/cpu/reduce/reduce_same.hpp>
+#include <loops/pairwise_instantiations.h>
+#include <system/type_boilerplate.h>
+
+/*
+ * Function Instantiation:
+ * ReduceSameFunction instantiated for common types in @COMB1@
+ */
+
+#define LIST_CALLBACK_EXEC_SCALAR_1(INPUT) \
+    template void* functions::reduce::ReduceSameFunction<GET_SECOND(INPUT)>::execScalar( \
+        const int opNum, \
+        const void *x, \
+        const sd::LongType *xShapeInfo, \
+        void *extraParams);
+
+ITERATE_LIST((SD_COMMON_TYPES_PART_@COMB1@), LIST_CALLBACK_EXEC_SCALAR_1)
+
+#define LIST_CALLBACK_EXEC_SCALAR_2(INPUT) \
+    template void functions::reduce::ReduceSameFunction<GET_SECOND(INPUT)>::execScalar( \
+        const int opNum, \
+        const void *x, \
+        const sd::LongType *xShapeInfo, \
+        void *extraParams, \
+        void *z, \
+        const sd::LongType *zShapeInfo);
+
+ITERATE_LIST((SD_COMMON_TYPES_PART_@COMB1@), LIST_CALLBACK_EXEC_SCALAR_2)
+
+#define LIST_CALLBACK_EXEC(INPUT) \
+    template void functions::reduce::ReduceSameFunction<GET_SECOND(INPUT)>::exec( \
+        int opNum, \
+        sd::memory::Workspace *workspace, \
+        const void *vx, \
+        const sd::LongType *xShapeInfo, \
+        void *vextraParams, \
+        void *vz, \
+        const sd::LongType *zShapeInfo, \
+        const sd::LongType *dims);
+
+ITERATE_LIST((SD_COMMON_TYPES_PART_@COMB1@), LIST_CALLBACK_EXEC)

libnd4j/include/loops/cpu/comb_compilation_units/scalar_instantiation_template_3.cpp.in

@@ -0,0 +1,42 @@
+/* ******************************************************************************
+ *
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Apache License, Version 2.0 which is available at
+ * https://www.apache.org/licenses/LICENSE-2.0.
+ *
+ *  See the NOTICE file distributed with this work for additional
+ *  information regarding copyright ownership.
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ * License for the specific language governing permissions and limitations
+ * under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ ******************************************************************************/
+
+//
+// Created by agibsonccc on 12/30/24.
+//
+
+#include <loops/pairwise_instantiations.h>
+#include <types/type_promote.h>
+#include <loops/cpu/scalar.hpp>
+
+/*
+ *
+ * Function Instantiation:
+ * ScalarTransform::transform instantiated for types in @COMB1@, @COMB2@, @COMB3@
+ */
+
+ITERATE_COMBINATIONS_3(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_COMMON_TYPES_PART_@COMB2@,
+    SD_COMMON_TYPES_PART_@COMB3@,
+    INSTANT_PROCESS_COMBINATION_3,
+    functions::scalar::ScalarTransform,
+    ::transform(int opNum, const void *x, const sd::LongType *xShapeInfo, void *result,
+                const sd::LongType *resultShapeInfo, const void *scalar, void *extraParams, sd::LongType start,
+                sd::LongType stop)
+)