merge from public repo

library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_ngchw_gkyxc_ngkhw_f32_mem_inter_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_xdl_ngchw_gkyxc_ngkhw_f32_mem_inter_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NGCHW,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NGKHW,
+                                                                F32,
+                                                                F32,
+                                                                Empty_Tuple,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f32_mem_instances<2,
+                                                                                 NGCHW,
+                                                                                 GKYXC,
+                                                                                 Empty_Tuple,
+                                                                                 NGKHW,
+                                                                                 ConvFwdDefault,
+                                                                                 Interwave>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_ngchw_gkyxc_ngkhw_f32_mem_intra_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_xdl_ngchw_gkyxc_ngkhw_f32_mem_intra_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NGCHW,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NGKHW,
+                                                                F32,
+                                                                F32,
+                                                                Empty_Tuple,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f32_mem_instances<2,
+                                                                                 NGCHW,
+                                                                                 GKYXC,
+                                                                                 Empty_Tuple,
+                                                                                 NGKHW,
+                                                                                 ConvFwdDefault,
+                                                                                 Intrawave>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/merged_groups/device_grouped_conv2d_fwd_xdl_merged_groups_ngchw_gkyxc_ngkhw_f16_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_merged_groups_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_xdl_merged_groups_ngchw_gkyxc_ngkhw_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NGCHW,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NGKHW,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_merged_groups_f16_instances<2,
+                                                                NGCHW,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NGKHW,
+                                                                ConvFwdDefault>{});
+
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_merged_groups_f16_instances<2,
+                                                                NGCHW,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NGKHW,
+                                                                ConvFwd3x3>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/merged_groups/device_grouped_conv2d_fwd_xdl_merged_groups_ngchw_gkyxc_ngkhw_f32_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_merged_groups_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_xdl_merged_groups_ngchw_gkyxc_ngkhw_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NGCHW,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NGKHW,
+                                                                F32,
+                                                                F32,
+                                                                Empty_Tuple,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_merged_groups_f32_instances<2,
+                                                                NGCHW,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NGKHW,
+                                                                ConvFwdDefault>{});
+
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_merged_groups_f32_instances<2,
+                                                                NGCHW,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NGKHW,
+                                                                ConvFwd3x3>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt

 set(DEVICE_MAXPOOL_BWD_INSTANCES)
 list(APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_f16_instance.cpp
                                          device_max_pool_bwd_bf16_instance.cpp
-                                         device_max_pool_bwd_f32_instance.cpp)
+                                         device_max_pool_bwd_f32_instance.cpp
+                                         device_max_pool_bwd_f8_instance.cpp
+                                         device_max_pool_bwd_int8_instance.cpp)
 add_instance_library(device_max_pool_bwd_instance ${DEVICE_MAXPOOL_BWD_INSTANCES})

library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_f8_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "max_pool_bwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_maxpool_bwd_f8_instances(
+    std::vector<std::unique_ptr<DeviceMaxPoolBwd<F8, I32, F8>>>& instances)
+{
+    add_device_operation_instances(instances, device_maxpool_bwd_instances<F8, I32, F8>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_int8_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "max_pool_bwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_maxpool_bwd_int8_instances(
+    std::vector<std::unique_ptr<DeviceMaxPoolBwd<I8, I32, I8>>>& instances)
+{
+    add_device_operation_instances(instances, device_maxpool_bwd_instances<I8, I32, I8>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/max_pool_bwd/max_pool_bwd_instance_common.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -17,6 +17,8 @@ namespace instance {
 using I32  = int32_t;
 using F16  = ck::half_t;
 using BF16 = ck::bhalf_t;
+using I8   = int8_t;
+using F8   = ck::f8_t;
 using F32  = float;
 
 template <typename DOutDataType, typename IndexDataType, typename DInDataType>

library/src/tensor_operation_instance/gpu/mha/CMakeLists.txt

 set(FMHA_CPP_FOLDER ${CMAKE_CURRENT_BINARY_DIR})
 set(FMHA_SRC_FOLDER ${CMAKE_SOURCE_DIR}/example/ck_tile/01_fmha/)
 set(CK_TILE_SRC_FOLDER ${CMAKE_SOURCE_DIR}/include/ck_tile/)
-# python stuff
+
+# Usage: for customized Python location cmake -DCK_USE_ALTERNATIVE_PYTHON="/opt/Python-3.8.13/bin/python3.8"
+# CK Codegen requires dataclass which is added in Python 3.7
+# Python version 3.8 is required for general good practice as it is default for Ubuntu 20.04
 if(NOT CK_USE_ALTERNATIVE_PYTHON)
-  find_package(PythonInterp 3 REQUIRED)
+   find_package(PythonInterp 3 REQUIRED)
 else()
-  message("Using alternative python version")
-  set(EXTRA_PYTHON_PATH)
-  string(REPLACE "/bin/python3.8" "" EXTRA_PYTHON_PATH "${CK_USE_ALTERNATIVE_PYTHON}")
-  message("alternative python path is: ${EXTRA_PYTHON_PATH}")
-  find_package(Python3 3.6 COMPONENTS Interpreter REQUIRED)
-  add_definitions(-DPython3_EXECUTABLE="${CK_USE_ALTERNATIVE_PYTHON}")
-  set(Python3_EXECUTABLE "${CK_USE_ALTERNATIVE_PYTHON}")
-  set(PYTHON_EXECUTABLE "${CK_USE_ALTERNATIVE_PYTHON}")
-  set(ENV{LD_LIBRARY_PATH} "${EXTRA_PYTHON_PATH}/lib:$ENV{LD_LIBRARY_PATH}")
+   message("Using alternative python version")
+   set(EXTRA_PYTHON_PATH)
+   # this is overly restrictive, we may need to be more flexible on the following
+   string(REPLACE "/bin/python3.8" "" EXTRA_PYTHON_PATH "${CK_USE_ALTERNATIVE_PYTHON}")
+   message("alternative python path is: ${EXTRA_PYTHON_PATH}")
+   find_package(Python3 3.6 COMPONENTS Interpreter REQUIRED)
+   add_definitions(-DPython3_EXECUTABLE="${CK_USE_ALTERNATIVE_PYTHON}")
+   set(Python3_EXECUTABLE "${CK_USE_ALTERNATIVE_PYTHON}")
+   set(PYTHON_EXECUTABLE "${CK_USE_ALTERNATIVE_PYTHON}")
+   set(ENV{LD_LIBRARY_PATH} "${EXTRA_PYTHON_PATH}/lib:$ENV{LD_LIBRARY_PATH}")
 endif()
 
 rocm_install(DIRECTORY ${CK_TILE_SRC_FOLDER} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/ck_tile)
@@ -23,18 +27,38 @@ rocm_install(FILES ${MHA_HEADERS} DESTINATION include/ck_tile/ops)
 # headers for building lib
 file(COPY ${MHA_HEADERS} DESTINATION ${FMHA_CPP_FOLDER})
 
+# Delete the blob file if it exists to avoid append of old content.
+if(EXISTS ${FMHA_CPP_FOLDER}/blob_list.txt)
+    file(REMOVE ${FMHA_CPP_FOLDER}/blob_list.txt)
+endif()
+
+set(FMHA_KNOWN_APIS "fwd,fwd_splitkv,fwd_appendkv,bwd")
+
 # generate a list of kernels, but not actually emit files at config stage
+# Note: The receipt 3 arg filters the generated backwards instances to reduce compilation time.
+# With receipt 3 set, we are generating instances for datatype == {fp16 || bfp16}, bias == {no || alibi}, deterministic == off, and dpad == dvpad.
 execute_process(
-  COMMAND ${PYTHON_EXECUTABLE} ${CMAKE_SOURCE_DIR}/example/ck_tile/01_fmha/generate.py
+  COMMAND ${PYTHON_EXECUTABLE} ${FMHA_SRC_FOLDER}/generate.py
   --list_blobs ${FMHA_CPP_FOLDER}/blob_list.txt
+  --api ${FMHA_KNOWN_APIS}
+  --receipt 3
+  RESULT_VARIABLE ret
 )
-file(STRINGS ${FMHA_CPP_FOLDER}/blob_list.txt FMHA_FWD_GEN_BLOBS)
+if(ret AND NOT ret EQUAL 0)
+  message( FATAL_ERROR "CK Tile MHA FAILED to genrate a list of kernels via Python.")
+else()
+  file(STRINGS ${FMHA_CPP_FOLDER}/blob_list.txt FMHA_GEN_BLOBS)
+endif()
 
-# actually generate the cpp files
+# actually generate the kernel content now
+# Note: The receipt 3 arg filters the generated backwards instances to reduce compilation time.
+# With receipt 3 set, we are generating instances for datatype == {fp16 || bfp16}, bias == {no || alibi}, deterministic == off, and dpad == dvpad.
 add_custom_command(
-  OUTPUT ${FMHA_FWD_GEN_BLOBS}
-  COMMAND ${PYTHON_EXECUTABLE} ${CMAKE_SOURCE_DIR}/example/ck_tile/01_fmha/generate.py
+  OUTPUT ${FMHA_GEN_BLOBS}
+  COMMAND ${PYTHON_EXECUTABLE} ${FMHA_SRC_FOLDER}/generate.py
   --output_dir ${FMHA_CPP_FOLDER}
+  --api ${FMHA_KNOWN_APIS}
+  --receipt 3
   COMMENT "Generating mha kernel (cpp) files now ..."
   VERBATIM
 )
@@ -43,17 +67,15 @@ add_custom_command(
 # have filename. Since, it was cauing the cmake
 # to throw "File name too long"
 set(device_files)
-foreach(filepath IN LISTS FMHA_FWD_GEN_BLOBS)
+foreach(filepath IN LISTS FMHA_GEN_BLOBS)
     get_filename_component(filename ${filepath} NAME)
     # Append the filename to the device_files list
     list(APPEND device_files ${filename})
 endforeach()
-add_custom_target(generate_cpp_files DEPENDS ${FMHA_FWD_GEN_BLOBS})
+add_custom_target(generate_cpp_files DEPENDS ${FMHA_GEN_BLOBS})
 
 add_instance_library(device_mha_instance ${device_files})
 
-
-
 if (TARGET device_mha_instance)
   add_dependencies(device_mha_instance generate_cpp_files)
 endif()

library/src/tensor_operation_instance/gpu/pool2d_fwd/CMakeLists.txt

+set(DEVICE_POOL2D_FWD_INSTANCES)
+list(APPEND DEVICE_POOL2D_FWD_INSTANCES device_avg_pool2d_fwd_nhwc_f16_instance.cpp
+                                        device_max_pool2d_fwd_nhwc_f16_instance.cpp
+                                        device_avg_pool2d_fwd_nhwc_f32_instance.cpp
+                                        device_max_pool2d_fwd_nhwc_f32_instance.cpp
+                                        device_avg_pool2d_fwd_nhwc_bf16_instance.cpp
+                                        device_max_pool2d_fwd_nhwc_bf16_instance.cpp
+                                        device_avg_pool2d_fwd_nhwc_i8_instance.cpp
+                                        device_max_pool2d_fwd_nhwc_i8_instance.cpp
+                                        device_avg_pool2d_fwd_nhwc_f8_instance.cpp
+                                        device_max_pool2d_fwd_nhwc_f8_instance.cpp)
+add_instance_library(device_pool2d_fwd_instance ${DEVICE_POOL2D_FWD_INSTANCES})

library/src/tensor_operation_instance/gpu/pool2d_fwd/device_avg_pool2d_fwd_nhwc_bf16_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool2d_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool2d_fwd_nhwc_bf16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<4, 2, BF16, BF16, I32, NHWC, NHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<BF16, BF16, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/pool2d_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool2d_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool2d_fwd_nhwc_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, NHWC, NHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/pool2d_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool2d_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool2d_fwd_nhwc_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, NHWC, NHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/pool2d_fwd/device_avg_pool2d_fwd_nhwc_f8_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool2d_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool2d_fwd_nhwc_f8_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F8, F8, I32, NHWC, NHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F8, F8, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/pool2d_fwd/device_avg_pool2d_fwd_nhwc_i8_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool2d_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool2d_fwd_nhwc_i8_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, I8, I8, I32, NHWC, NHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<I8, I8, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/pool2d_fwd/device_max_pool2d_fwd_nhwc_bf16_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool2d_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool2d_fwd_nhwc_bf16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<4, 2, BF16, BF16, I32, NHWC, NHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<BF16, BF16, I32, F32, ReduceOpId, false>{});
+}
+
+void add_device_pool2d_fwd_nhwc_index_bf16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<4, 2, BF16, BF16, I32, NHWC, NHWC, ReduceOpId, true>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<BF16, BF16, I32, F32, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/pool2d_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool2d_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool2d_fwd_nhwc_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, NHWC, NHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F32, ReduceOpId, false>{});
+}
+
+void add_device_pool2d_fwd_nhwc_index_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, NHWC, NHWC, ReduceOpId, true>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F32, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/pool2d_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool2d_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool2d_fwd_nhwc_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, NHWC, NHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
+}
+
+void add_device_pool2d_fwd_nhwc_index_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, NHWC, NHWC, ReduceOpId, true>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/pool2d_fwd/device_max_pool2d_fwd_nhwc_f8_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool2d_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool2d_fwd_nhwc_f8_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F8, F8, I32, NHWC, NHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F8, F8, I32, F32, ReduceOpId, false>{});
+}
+
+void add_device_pool2d_fwd_nhwc_index_f8_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F8, F8, I32, NHWC, NHWC, ReduceOpId, true>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F8, F8, I32, F32, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/pool2d_fwd/device_max_pool2d_fwd_nhwc_i8_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool2d_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool2d_fwd_nhwc_i8_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, I8, I8, I32, NHWC, NHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<I8, I8, I32, F32, ReduceOpId, false>{});
+}
+
+void add_device_pool2d_fwd_nhwc_index_i8_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, I8, I8, I32, NHWC, NHWC, ReduceOpId, true>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<I8, I8, I32, F32, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck