chore: import upstream snapshot with attribution

libnd4j/include/loops/cuda/comb_compilation_units/broadcast_exec_broadcast_with_dimension.cu.in

@@ -0,0 +1,74 @@
+/*
+ *  ******************************************************************************
+ *  *
+ *  *
+ *  * 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 3/10/25.
+//
+
+#include <loops/cuda/broadcasting.chpp>
+#include <system/type_boilerplate.h>
+#include <loops/pairwise_instantiations_single.h>
+
+/*
+ * Function instantiation for Broadcast::execBroadcast with dimension
+ */
+
+ITERATE_COMBINATIONS_3(
+    SD_SINGLE_TYPE_@COMB1@,
+    SD_SINGLE_TYPE_@COMB2@,
+    SD_SINGLE_TYPE_@COMB3@,
+    INSTANT_PROCESS_COMBINATION_3,
+    functions::broadcast::Broadcast,
+    ::execBroadcast(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        void const* x,
+        sd::LongType const* xShapeInfo,
+        void const* y,
+        sd::LongType const* yShapeInfo,
+        void* z,
+        sd::LongType const* zShapeInfo,
+        sd::LongType* dimension,
+        sd::LongType dimensionLength,
+        sd::LongType const* tadOnlyShapeInfo,
+        sd::LongType const* tadOffsets,
+        sd::LongType const* tadOnlyShapeInfoZ,
+        sd::LongType const* tadOffsetsZ);
+);
+
+ITERATE_COMBINATIONS_3(
+    SD_SINGLE_TYPE_@COMB1@,
+    SD_SINGLE_TYPE_@COMB2@,
+    SD_SINGLE_TYPE_@COMB3@,
+    INSTANT_PROCESS_COMBINATION_3,
+    functions::broadcast::Broadcast,
+    ::execBroadcast(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        void const* x,
+        sd::LongType const* xShapeInfo,
+        void const* y,
+        sd::LongType const* yShapeInfo,
+        void* z,
+        sd::LongType const* zShapeInfo);
+);
+

libnd4j/include/loops/cuda/comb_compilation_units/broadcast_exec_inverse_broadcast.cu.in

@@ -0,0 +1,55 @@
+/*
+ *  ******************************************************************************
+ *  *
+ *  *
+ *  * 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 3/10/25.
+//
+
+#include <loops/cuda/broadcasting.chpp>
+#include <system/type_boilerplate.h>
+#include <loops/pairwise_instantiations_single.h>
+
+/*
+ * Function instantiation for Broadcast::execInverseBroadcast
+ */
+
+ITERATE_COMBINATIONS_3(
+    SD_SINGLE_TYPE_@COMB1@,
+    SD_SINGLE_TYPE_@COMB2@,
+    SD_SINGLE_TYPE_@COMB3@,
+    INSTANT_PROCESS_COMBINATION_3,
+    functions::broadcast::Broadcast,
+    ::execInverseBroadcast(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        void const* x,
+        sd::LongType const* xShapeInfo,
+        void const* y,
+        sd::LongType const* yShapeInfo,
+        void* z,
+        sd::LongType const* zShapeInfo,
+        sd::LongType* dimension,
+        sd::LongType dimensionLength,
+        sd::LongType const* tadOnlyShapeInfo,
+        sd::LongType const* tadOffsets,
+        sd::LongType const* tadOnlyShapeInfoZ,
+        sd::LongType const* tadOffsetsZ);
+);

libnd4j/include/loops/cuda/comb_compilation_units/broadcast_instantiation_template_3.cu.in

@@ -0,0 +1,96 @@
+/* ******************************************************************************
+ *
+ *
+ * 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 3/10/25.
+//
+
+#include <loops/cuda/broadcasting.chpp>
+#include <system/type_boilerplate.h>
+#include <loops/pairwise_instantiations_single.h>
+
+/*
+ * Function Instantiation:
+ * Broadcast::execBroadcast 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,
+    ::execBroadcast(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        void const* x,
+        sd::LongType const* xShapeInfo,
+        void const* y,
+        sd::LongType const* yShapeInfo,
+        void* z,
+        sd::LongType const* zShapeInfo,
+        sd::LongType* dimension,
+        sd::LongType dimensionLength,
+        sd::LongType const* tadOnlyShapeInfo,
+        sd::LongType const* tadOffsets,
+        sd::LongType const* tadOnlyShapeInfoZ,
+        sd::LongType const* tadOffsetsZ);
+);
+
+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,
+    ::execBroadcast(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        void const* x,
+        sd::LongType const* xShapeInfo,
+        void const* y,
+        sd::LongType const* yShapeInfo,
+        void* z,
+        sd::LongType const* zShapeInfo);
+);
+
+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,
+    ::execInverseBroadcast(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        void const* x,
+        sd::LongType const* xShapeInfo,
+        void const* y,
+        sd::LongType const* yShapeInfo,
+        void* z,
+        sd::LongType const* zShapeInfo,
+        sd::LongType* dimension,
+        sd::LongType dimensionLength,
+        sd::LongType const* tadOnlyShapeInfo,
+        sd::LongType const* tadOffsets,
+        sd::LongType const* tadOnlyShapeInfoZ,
+        sd::LongType const* tadOffsetsZ);
+);

libnd4j/include/loops/cuda/comb_compilation_units/indexreduce_instantiation_template_2.cu.in

@@ -0,0 +1,81 @@
+/* ******************************************************************************
+ *
+ *
+ * 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 3/10/25.
+//
+
+#include <loops/cuda/indexreduce.cu>
+#include <system/type_boilerplate.h>
+#include <loops/pairwise_instantiations_single.h>
+
+/*
+ * Function Instantiation:
+ * IndexReduce::executeIndexReduce instantiated for types in @COMB1@, @COMB2@
+ * Note: Using @COMB1@ for X type and @COMB2@ for Z type (2-type combinations)
+ */
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_INDEXING_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION,
+    functions::indexreduce::IndexReduce,
+    ::executeIndexReduce(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        const int opNum,
+        void const* dx,
+        sd::LongType const* xShapeInfo,
+        sd::LongType xRank,
+        void* extraParams,
+        void* result,
+        sd::LongType const* zShapeInfo,
+        sd::LongType zRank,
+        sd::LongType* dimension,
+        sd::LongType dimensionLength,
+        int postProcessOrNot,
+        sd::LongType* allocationBuffer,
+        void* reductionBuffer,
+        sd::LongType const* tadOnlyShapeInfo,
+        sd::LongType const* tadOffsets);
+);
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_INDEXING_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION,
+    functions::indexreduce::IndexReduce,
+    ::executeIndexReduceScalar(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        const int opNum,
+        void const* dx,
+        sd::LongType const* xShapeInfo,
+        sd::LongType xRank,
+        void* extraParams,
+        void* result,
+        sd::LongType const* zShapeInfo,
+        sd::LongType zRank,
+        sd::LongType* dimension,
+        sd::LongType dimensionLength,
+        int postProcessOrNot,
+        sd::LongType* allocationBuffer,
+        void* reductionBuffer,
+        sd::LongType const* tadOnlyShapeInfo,
+        sd::LongType const* tadOffsets);
+);

libnd4j/include/loops/cuda/comb_compilation_units/pairwise_exec_cuda_shaped.cu.in

@@ -0,0 +1,50 @@
+/*
+ *  ******************************************************************************
+ *  *
+ *  *
+ *  * 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 3/10/25.
+//
+
+#include <loops/cuda/pairwise.chpp>
+#include <system/type_boilerplate.h>
+#include <loops/pairwise_instantiations_single.h>
+
+/*
+ * Function instantiation for PairWiseTransform::executeCudaShaped
+ */
+
+ITERATE_COMBINATIONS_3(
+    SD_SINGLE_TYPE_@COMB1@,
+    SD_SINGLE_TYPE_@COMB2@,
+    SD_SINGLE_TYPE_@COMB3@,
+    INSTANT_PROCESS_COMBINATION_3,
+    functions::pairwise_transforms::PairWiseTransform,
+    ::executeCudaShaped(
+        dim3& launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        const void *vx,
+        const sd::LongType *xShapeInfo,
+        const void *vy,
+        const sd::LongType *yShapeInfo,
+        void *vz,
+        const sd::LongType *zShapeInfo,
+        void *vextraParams);
+);

libnd4j/include/loops/cuda/comb_compilation_units/pairwise_instantiation_template_3.cu.in

@@ -0,0 +1,49 @@
+/* ******************************************************************************
+ *
+ *
+ * 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 3/10/25.
+//
+
+#include <loops/cuda/pairwise.chpp>
+#include <system/type_boilerplate.h>
+#include <loops/pairwise_instantiations_single.h>
+
+/*
+ * Function Instantiation:
+ * PairWiseTransform::executeCudaShaped 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,
+    ::executeCudaShaped(
+        dim3& launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        const void *vx,
+        const sd::LongType *xShapeInfo,
+        const void *vy,
+        const sd::LongType *yShapeInfo,
+        void *vz,
+        const sd::LongType *zShapeInfo,
+        void *vextraParams);
+);

libnd4j/include/loops/cuda/comb_compilation_units/reduce3_instantiation_template_2.cu.in

@@ -0,0 +1,104 @@
+/* ******************************************************************************
+ *
+ *
+ * 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 3/10/25.
+//
+
+#include <loops/cuda/reduce3.chpp>
+#include <system/type_boilerplate.h>
+#include <loops/pairwise_instantiations_single.h>
+
+/*
+ * Function Instantiation:
+ * Reduce3::exec instantiated for types in @COMB1@, @COMB2@
+ * Note: Using @COMB1@ for X type and @COMB2@ for Z type (2-type combinations)
+ */
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION,
+    functions::reduce3::Reduce3,
+    ::exec(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        void const* vx,
+        sd::LongType const* xShapeInfo,
+        void const* vy,
+        sd::LongType const* yShapeInfo,
+        void* extraParams,
+        void* vz,
+        sd::LongType const* zShapeInfo,
+        sd::LongType* dimension,
+        sd::LongType dimensionLength,
+        int postProcessOrNot,
+        sd::LongType* allocationPointer,
+        sd::LongType const* tadOnlyShapeInfo,
+        sd::LongType const* tadOffsets,
+        sd::LongType const* yTadOnlyShapeInfo,
+        sd::LongType const* yTadOffsets);
+);
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION,
+    functions::reduce3::Reduce3,
+    ::execAll(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        void const* vx,
+        sd::LongType const* xShapeInfo,
+        void const* vy,
+        sd::LongType const* yShapeInfo,
+        void* extraParams,
+        void* vz,
+        sd::LongType const* zShapeInfo,
+        sd::LongType* dimension,
+        long long int dimensionLength,
+        int postProcessOrNot,
+        sd::LongType* allocationPointer,
+        sd::LongType const* tadOnlyShapeInfo,
+        sd::LongType const* tadOffsets,
+        sd::LongType const* yTadOnlyShapeInfo,
+        sd::LongType const* yTadOffsets);
+);
+
+ITERATE_COMBINATIONS(
+    SD_COMMON_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION,
+    functions::reduce3::Reduce3,
+    ::execScalar(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        void const* vx,
+        sd::LongType const* xShapeInfo,
+        void const* vy,
+        sd::LongType const* yShapeInfo,
+        void* extraParams,
+        void* vz,
+        sd::LongType const* zShapeInfo,
+        sd::LongType* allocationPointer,
+        void* reductionBuffer,
+        sd::LongType const* tadOnlyShapeInfo);
+);

libnd4j/include/loops/cuda/comb_compilation_units/reduce_float_instantiation_template_2.cu.in

@@ -0,0 +1,73 @@
+/* ******************************************************************************
+ *
+ *
+ * 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 3/10/25.
+//
+
+#include <loops/cuda/reduce/reduce_float.chpp>
+#include <system/type_boilerplate.h>
+#include <loops/pairwise_instantiations_single.h>
+
+/*
+ * Function Instantiation:
+ * ReduceFloatFunction::execReduce instantiated for types in @COMB1@, @COMB2@
+ * Note: Using @COMB1@ for X type and @COMB2@ for Z type (2-type combinations)
+ */
+
+ITERATE_COMBINATIONS(
+    SD_NUMERIC_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION,
+    functions::reduce::ReduceFloatFunction,
+    ::execReduce(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        const int opNum,
+        const void* x,
+        const sd::LongType* dXShapeInfo,
+        const sd::LongType* hXShapeInfo,
+        void* extraParams,
+        void* vreductionBuffer,
+        void* z,
+        const sd::LongType* dZShapeInfo,
+        const sd::LongType* hZShapeInfo,
+        const sd::LongType* dims);
+);
+
+ITERATE_COMBINATIONS(
+    SD_NUMERIC_TYPES_PART_@COMB1@,
+    SD_FLOAT_TYPES_PART_@COMB2@,
+    INSTANT_PROCESS_COMBINATION,
+    functions::reduce::ReduceFloatFunction,
+    ::execReduceScalar(
+        dim3 launchDims,
+        cudaStream_t* stream,
+        const int opNum,
+        const void* x,
+        const sd::LongType* xShapeInfo,
+        const sd::LongType* hXShapeInfo,
+        void* extraParams,
+        void* z,
+        const sd::LongType* dZShapeInfo,
+        const sd::LongType* hZShapeInfo,
+        sd::LongType* dimension,
+        long long int dimensionLength,
+        void* reductionBuffer,
+        const sd::LongType* tadOnlyShapeInfo);
+);

libnd4j/include/loops/cuda/comb_compilation_units/scalar_exec_cuda_along_dimension.cu.in

@@ -0,0 +1,55 @@
+/*
+ *  ******************************************************************************
+ *  *
+ *  *
+ *  * 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 3/10/25.
+//
+
+#include <loops/cuda/scalar.chpp>
+#include <system/type_boilerplate.h>
+#include <loops/pairwise_instantiations_single.h>
+
+/*
+ * Function instantiation for ScalarTransform::executeCudaAlongDimension
+ */
+
+ITERATE_COMBINATIONS_3(
+    SD_SINGLE_TYPE_@COMB1@,
+    SD_SINGLE_TYPE_@COMB2@,
+    SD_SINGLE_TYPE_@COMB3@,
+    INSTANT_PROCESS_COMBINATION_3,
+    functions::scalar::ScalarTransform,
+    ::executeCudaAlongDimension(
+        dim3& launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        const void *vx,
+        const sd::LongType *xShapeInfo,
+        void *vz,
+        const sd::LongType *zShapeInfo,
+        const void *vscalars,
+        void *vextraParams,
+        sd::LongType *dimension,
+        sd::LongType dimensionLength,
+        const sd::LongType *tadShapeInfo,
+        const sd::LongType *tadOffsets,
+        const sd::LongType *tadShapeInfoZ,
+        const sd::LongType *tadOffsetsZ);
+);

libnd4j/include/loops/cuda/comb_compilation_units/scalar_exec_cuda_shaped.cu.in

@@ -0,0 +1,51 @@
+/*
+ *  ******************************************************************************
+ *  *
+ *  *
+ *  * 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 3/10/25.
+//
+
+#include <loops/cuda/scalar.chpp>
+#include <system/type_boilerplate.h>
+#include <loops/pairwise_instantiations_single.h>
+
+/*
+ * Function instantiation for ScalarTransform::executeCudaShaped
+ */
+
+ITERATE_COMBINATIONS_3(
+    SD_SINGLE_TYPE_@COMB1@,
+    SD_SINGLE_TYPE_@COMB2@,
+    SD_SINGLE_TYPE_@COMB3@,
+    INSTANT_PROCESS_COMBINATION_3,
+    functions::scalar::ScalarTransform,
+    ::executeCudaShaped(
+        dim3& launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        const void *vx,
+        const sd::LongType *xShapeInfo,
+        const sd::LongType *hxShapeInfo,
+        void *vz,
+        const sd::LongType *zShapeInfo,
+        const sd::LongType *hzShapeInfo,
+        const void *vscalar,
+        void *vextraParams);
+);

libnd4j/include/loops/cuda/comb_compilation_units/scalar_instantiation_template_3.cu.in

@@ -0,0 +1,74 @@
+/* ******************************************************************************
+ *
+ *
+ * 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 3/10/25.
+//
+
+#include <loops/cuda/scalar.chpp>
+#include <system/type_boilerplate.h>
+#include <loops/pairwise_instantiations_single.h>
+
+/*
+ * Function Instantiation:
+ * ScalarTransform::executeCudaShaped 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,
+    ::executeCudaShaped(
+        dim3& launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        const void *vx,
+        const sd::LongType *xShapeInfo,
+        const sd::LongType *hxShapeInfo,
+        void *vz,
+        const sd::LongType *zShapeInfo,
+        const sd::LongType *hzShapeInfo,
+        const void *vscalar,
+        void *vextraParams);
+);
+
+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,
+    ::executeCudaAlongDimension(
+        dim3& launchDims,
+        cudaStream_t* stream,
+        int opNum,
+        const void *vx,
+        const sd::LongType *xShapeInfo,
+        void *vz,
+        const sd::LongType *zShapeInfo,
+        const void *vscalars,
+        void *vextraParams,
+        sd::LongType *dimension,
+        sd::LongType dimensionLength,
+        const sd::LongType *tadShapeInfo,
+        const sd::LongType *tadOffsets,
+        const sd::LongType *tadShapeInfoZ,
+        const sd::LongType *tadOffsetsZ);
+);