merge develop

test/gemm/CMakeLists.txt

-# GEMM XDL
-add_test_executable(test_gemm_xdl_fp32 gemm_xdl_fp32.cpp)
-target_link_libraries(test_gemm_xdl_fp32 PRIVATE host_tensor)
-target_link_libraries(test_gemm_xdl_fp32 PRIVATE device_gemm_instance)
+add_test_executable(test_gemm_fp32 gemm_fp32.cpp)
+target_link_libraries(test_gemm_fp32 PRIVATE host_tensor)
+target_link_libraries(test_gemm_fp32 PRIVATE device_gemm_instance)
 
-add_test_executable(test_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
-target_link_libraries(test_gemm_xdl_fp16 PRIVATE host_tensor)
-target_link_libraries(test_gemm_xdl_fp16 PRIVATE device_gemm_instance)
+add_test_executable(test_gemm_fp16 gemm_fp16.cpp)
+target_link_libraries(test_gemm_fp16 PRIVATE host_tensor)
+target_link_libraries(test_gemm_fp16 PRIVATE device_gemm_instance)
 
-add_test_executable(test_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
-target_link_libraries(test_gemm_xdl_bf16 PRIVATE host_tensor)
-target_link_libraries(test_gemm_xdl_bf16 PRIVATE device_gemm_instance)
+add_test_executable(test_gemm_bf16 gemm_bf16.cpp)
+target_link_libraries(test_gemm_bf16 PRIVATE host_tensor)
+target_link_libraries(test_gemm_bf16 PRIVATE device_gemm_instance)
 
-add_test_executable(test_gemm_xdl_int8 gemm_xdl_int8.cpp)
-target_link_libraries(test_gemm_xdl_int8 PRIVATE host_tensor)
-target_link_libraries(test_gemm_xdl_int8 PRIVATE device_gemm_instance)
-
-# GEMM DL
-add_test_executable(test_gemm_dl_fp32 gemm_dl_fp32.cpp)
-target_link_libraries(test_gemm_dl_fp32 PRIVATE host_tensor)
-target_link_libraries(test_gemm_dl_fp32 PRIVATE device_gemm_instance)
-
-add_test_executable(test_gemm_dl_fp16 gemm_dl_fp16.cpp)
-target_link_libraries(test_gemm_dl_fp16 PRIVATE host_tensor)
-target_link_libraries(test_gemm_dl_fp16 PRIVATE device_gemm_instance)
-
-add_test_executable(test_gemm_dl_int8 gemm_dl_int8.cpp)
-target_link_libraries(test_gemm_dl_int8 PRIVATE host_tensor)
-TArget_link_libraries(test_gemm_dl_int8 PRIVATE device_gemm_instance)
+add_test_executable(test_gemm_int8 gemm_int8.cpp)
+target_link_libraries(test_gemm_int8 PRIVATE host_tensor)
+target_link_libraries(test_gemm_int8 PRIVATE device_gemm_instance)

test/gemm/gemm_bf16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <algorithm>
+#include <cstdlib>
+#include <iostream>
+#include <numeric>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/host_tensor/device_memory.hpp"
+#include "ck/library/host_tensor/host_tensor.hpp"
+#include "ck/library/host_tensor/host_tensor_generator.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+#include "test/gemm/gemm_util.hpp"
+
+int main()
+{
+    using ADataType   = ck::bhalf_t;
+    using BDataType   = ck::bhalf_t;
+    using CDataType   = ck::bhalf_t;
+    using AccDataType = float;
+
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+    auto test = [&](auto a_layout, auto b_layout, auto c_layout) {
+        bool pass = true;
+
+        using DeviceOp = ck::tensor_operation::device::DeviceGemm<decltype(a_layout),
+                                                                  decltype(b_layout),
+                                                                  decltype(c_layout),
+                                                                  ADataType,
+                                                                  BDataType,
+                                                                  CDataType,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>;
+
+        const auto gemmPtrs =
+            ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+                DeviceOp>::GetInstances();
+
+        for(auto& gemmPtr : gemmPtrs)
+        {
+            pass &= ck::gemm_util::TestGemm<std::unique_ptr<DeviceOp>,
+                                            ADataType,
+                                            BDataType,
+                                            CDataType,
+                                            AccDataType,
+                                            decltype(a_layout),
+                                            decltype(b_layout),
+                                            decltype(c_layout),
+                                            PassThrough,
+                                            PassThrough,
+                                            PassThrough>{}(gemmPtr);
+        }
+
+        return pass;
+    };
+
+    bool pass = test(Row{}, Row{}, Row{}) && test(Row{}, Col{}, Row{}) &&
+                test(Col{}, Row{}, Row{}) && test(Col{}, Col{}, Row{});
+
+    std::cout << "TestGemm ..... " << (pass ? "SUCCESS" : "FAILURE") << std::endl;
+    return pass ? 0 : 1;
+}

test/gemm/gemm_dl_fp16.cpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <algorithm>
-#include <cstdlib>
-#include <iostream>
-#include <numeric>
-#include <tuple>
-#include <vector>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-
-#include "test/gemm/gemm_util.hpp"
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using DeviceGemmNoOpPtr =
-    ck::tensor_operation::device::DeviceGemmPtr<ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough>;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace device_gemm_instance {
-
-void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-
-} // namespace device_gemm_instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-int main()
-{
-    using ADataType   = ck::half_t;
-    using BDataType   = ck::half_t;
-    using CDataType   = ck::half_t;
-    using AccDataType = float;
-
-    using RowMajor    = ck::tensor_layout::gemm::RowMajor;
-    using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
-
-    bool res = true;
-
-    std::vector<DeviceGemmNoOpPtr> gemmPtrs;
-
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
-    return res ? 0 : 1;
-}

test/gemm/gemm_dl_fp32.cpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <algorithm>
-#include <cstdlib>
-#include <iostream>
-#include <numeric>
-#include <tuple>
-#include <vector>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-
-#include "test/gemm/gemm_util.hpp"
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using DeviceGemmNoOpPtr =
-    ck::tensor_operation::device::DeviceGemmPtr<ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough>;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace device_gemm_instance {
-
-void add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_dl_f32_f32_f32_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_dl_f32_f32_f32_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-
-} // namespace device_gemm_instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-int main()
-{
-    using ADataType   = float;
-    using BDataType   = float;
-    using CDataType   = float;
-    using AccDataType = float;
-
-    using RowMajor    = ck::tensor_layout::gemm::RowMajor;
-    using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
-
-    bool res = true;
-    std::vector<DeviceGemmNoOpPtr> gemmPtrs;
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_f32_f32_f32_km_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_f32_f32_f32_mk_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
-    return res ? 0 : 1;
-}

test/gemm/gemm_dl_int8.cpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <algorithm>
-#include <cstdlib>
-#include <iostream>
-#include <numeric>
-#include <tuple>
-#include <vector>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-
-#include "test/gemm/gemm_util.hpp"
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using DeviceGemmNoOpPtr =
-    ck::tensor_operation::device::DeviceGemmPtr<ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough>;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace device_gemm_instance {
-
-void add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-
-} // namespace device_gemm_instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-int main()
-{
-    using ADataType   = int8_t;
-    using BDataType   = int8_t;
-    using CDataType   = int8_t;
-    using AccDataType = int;
-
-    using RowMajor    = ck::tensor_layout::gemm::RowMajor;
-    using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
-
-    bool res = true;
-    std::vector<DeviceGemmNoOpPtr> gemmPtrs;
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
-    return res ? 0 : 1;
-}

test/gemm/gemm_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <algorithm>
+#include <cstdlib>
+#include <iostream>
+#include <numeric>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/host_tensor/device_memory.hpp"
+#include "ck/library/host_tensor/host_tensor.hpp"
+#include "ck/library/host_tensor/host_tensor_generator.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+#include "test/gemm/gemm_util.hpp"
+
+int main()
+{
+    using ADataType   = ck::half_t;
+    using BDataType   = ck::half_t;
+    using CDataType   = ck::half_t;
+    using AccDataType = float;
+
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+    auto test = [&](auto a_layout, auto b_layout, auto c_layout) {
+        bool pass = true;
+
+        using DeviceOp = ck::tensor_operation::device::DeviceGemm<decltype(a_layout),
+                                                                  decltype(b_layout),
+                                                                  decltype(c_layout),
+                                                                  ADataType,
+                                                                  BDataType,
+                                                                  CDataType,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>;
+
+        const auto gemmPtrs =
+            ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+                DeviceOp>::GetInstances();
+
+        for(auto& gemmPtr : gemmPtrs)
+        {
+            pass &= ck::gemm_util::TestGemm<std::unique_ptr<DeviceOp>,
+                                            ADataType,
+                                            BDataType,
+                                            CDataType,
+                                            AccDataType,
+                                            decltype(a_layout),
+                                            decltype(b_layout),
+                                            decltype(c_layout),
+                                            PassThrough,
+                                            PassThrough,
+                                            PassThrough>{}(gemmPtr);
+        }
+
+        return pass;
+    };
+
+    bool pass = test(Row{}, Row{}, Row{}) && test(Row{}, Col{}, Row{}) &&
+                test(Col{}, Row{}, Row{}) && test(Col{}, Col{}, Row{});
+
+    std::cout << "TestGemm ..... " << (pass ? "SUCCESS" : "FAILURE") << std::endl;
+    return pass ? 0 : 1;
+}

test/gemm/gemm_fp32.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <algorithm>
+#include <cstdlib>
+#include <iostream>
+#include <numeric>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/host_tensor/device_memory.hpp"
+#include "ck/library/host_tensor/host_tensor.hpp"
+#include "ck/library/host_tensor/host_tensor_generator.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+#include "test/gemm/gemm_util.hpp"
+
+int main()
+{
+    using ADataType   = float;
+    using BDataType   = float;
+    using CDataType   = float;
+    using AccDataType = float;
+
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+    auto test = [&](auto a_layout, auto b_layout, auto c_layout) {
+        bool pass = true;
+
+        using DeviceOp = ck::tensor_operation::device::DeviceGemm<decltype(a_layout),
+                                                                  decltype(b_layout),
+                                                                  decltype(c_layout),
+                                                                  ADataType,
+                                                                  BDataType,
+                                                                  CDataType,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>;
+
+        const auto gemmPtrs =
+            ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+                DeviceOp>::GetInstances();
+
+        for(auto& gemmPtr : gemmPtrs)
+        {
+            pass &= ck::gemm_util::TestGemm<std::unique_ptr<DeviceOp>,
+                                            ADataType,
+                                            BDataType,
+                                            CDataType,
+                                            AccDataType,
+                                            decltype(a_layout),
+                                            decltype(b_layout),
+                                            decltype(c_layout),
+                                            PassThrough,
+                                            PassThrough,
+                                            PassThrough>{}(gemmPtr);
+        }
+
+        return pass;
+    };
+
+    bool pass = test(Row{}, Row{}, Row{}) && test(Row{}, Col{}, Row{}) &&
+                test(Col{}, Row{}, Row{}) && test(Col{}, Col{}, Row{});
+
+    std::cout << "TestGemm ..... " << (pass ? "SUCCESS" : "FAILURE") << std::endl;
+    return pass ? 0 : 1;
+}

test/gemm/gemm_fp64.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <algorithm>
+#include <cstdlib>
+#include <iostream>
+#include <numeric>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/host_tensor/device_memory.hpp"
+#include "ck/library/host_tensor/host_tensor.hpp"
+#include "ck/library/host_tensor/host_tensor_generator.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+#include "test/gemm/gemm_util.hpp"
+
+int main()
+{
+    using ADataType   = double;
+    using BDataType   = double;
+    using CDataType   = double;
+    using AccDataType = double;
+
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+    auto test = [&](auto a_layout, auto b_layout, auto c_layout) {
+        bool pass = true;
+
+        using DeviceOp = ck::tensor_operation::device::DeviceGemm<decltype(a_layout),
+                                                                  decltype(b_layout),
+                                                                  decltype(c_layout),
+                                                                  ADataType,
+                                                                  BDataType,
+                                                                  CDataType,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>;
+
+        const auto gemmPtrs =
+            ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+                DeviceOp>::GetInstances();
+
+        for(auto& gemmPtr : gemmPtrs)
+        {
+            pass &= ck::gemm_util::TestGemm<std::unique_ptr<DeviceOp>,
+                                            ADataType,
+                                            BDataType,
+                                            CDataType,
+                                            AccDataType,
+                                            decltype(a_layout),
+                                            decltype(b_layout),
+                                            decltype(c_layout),
+                                            PassThrough,
+                                            PassThrough,
+                                            PassThrough>{}(gemmPtr);
+        }
+
+        return pass;
+    };
+
+    bool pass = test(Row{}, Row{}, Row{}) && test(Row{}, Col{}, Row{}) &&
+                test(Col{}, Row{}, Row{}) && test(Col{}, Col{}, Row{});
+
+    std::cout << "TestGemm ..... " << (pass ? "SUCCESS" : "FAILURE") << std::endl;
+    return pass ? 0 : 1;
+}

test/gemm/gemm_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <algorithm>
+#include <cstdlib>
+#include <iostream>
+#include <numeric>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/host_tensor/device_memory.hpp"
+#include "ck/library/host_tensor/host_tensor.hpp"
+#include "ck/library/host_tensor/host_tensor_generator.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+#include "test/gemm/gemm_util.hpp"
+
+int main()
+{
+    using ADataType   = int8_t;
+    using BDataType   = int8_t;
+    using CDataType   = int8_t;
+    using AccDataType = int32_t;
+
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+    auto test = [&](auto a_layout, auto b_layout, auto c_layout) {
+        bool pass = true;
+
+        using DeviceOp = ck::tensor_operation::device::DeviceGemm<decltype(a_layout),
+                                                                  decltype(b_layout),
+                                                                  decltype(c_layout),
+                                                                  ADataType,
+                                                                  BDataType,
+                                                                  CDataType,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>;
+
+        const auto gemmPtrs =
+            ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+                DeviceOp>::GetInstances();
+
+        for(auto& gemmPtr : gemmPtrs)
+        {
+            pass &= ck::gemm_util::TestGemm<std::unique_ptr<DeviceOp>,
+                                            ADataType,
+                                            BDataType,
+                                            CDataType,
+                                            AccDataType,
+                                            decltype(a_layout),
+                                            decltype(b_layout),
+                                            decltype(c_layout),
+                                            PassThrough,
+                                            PassThrough,
+                                            PassThrough>{}(gemmPtr);
+        }
+
+        return pass;
+    };
+
+    bool pass = test(Row{}, Row{}, Row{}) && test(Row{}, Col{}, Row{}) &&
+                test(Col{}, Row{}, Row{}) && test(Col{}, Col{}, Row{});
+
+    std::cout << "TestGemm ..... " << (pass ? "SUCCESS" : "FAILURE") << std::endl;
+    return pass ? 0 : 1;
+}

test/gemm/gemm_util.hpp

@@ -159,7 +159,7 @@ struct TestGemm
         return std::make_tuple(a_m_k, b_k_n, c_m_n_host_result, c_m_n_device_result);
     }
 
-    auto operator()(DeviceGemmPtr_& gemmPtr)
+    auto operator()(const DeviceGemmPtr_& gemmPtr)
     {
         std::cout << "ALayout = " << ALayout{}.name << ", BLayout = " << BLayout{}.name
                   << ", CLayout = " << CLayout{}.name << std::endl;

test/gemm/gemm_xdl_bf16.cpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <algorithm>
-#include <cstdlib>
-#include <iostream>
-#include <numeric>
-#include <tuple>
-#include <vector>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-
-#include "test/gemm/gemm_util.hpp"
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using DeviceGemmNoOpPtr =
-    ck::tensor_operation::device::DeviceGemmPtr<ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough>;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace device_gemm_instance {
-void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances(
-    std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(
-    std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(
-    std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instances(
-    std::vector<DeviceGemmNoOpPtr>&);
-} // namespace device_gemm_instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-int main()
-{
-    using ADataType   = ck::bhalf_t;
-    using BDataType   = ck::bhalf_t;
-    using CDataType   = ck::bhalf_t;
-    using AccDataType = float;
-
-    using RowMajor    = ck::tensor_layout::gemm::RowMajor;
-    using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
-
-    bool res = true;
-    std::vector<DeviceGemmNoOpPtr> gemmPtrs;
-
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
-    return res ? 0 : 1;
-}

test/gemm/gemm_xdl_fp16.cpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <algorithm>
-#include <cstdlib>
-#include <iostream>
-#include <numeric>
-#include <tuple>
-#include <vector>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-
-#include "test/gemm/gemm_util.hpp"
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using DeviceGemmNoOpPtr =
-    ck::tensor_operation::device::DeviceGemmPtr<ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough>;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace device_gemm_instance {
-void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-
-#if 0
-void add_device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-#endif
-
-void add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-
-void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(
-    std::vector<DeviceGemmNoOpPtr>&);
-} // namespace device_gemm_instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-int main()
-{
-    using ADataType   = ck::half_t;
-    using BDataType   = ck::half_t;
-    using CDataType   = ck::half_t;
-    using AccDataType = float;
-
-    using RowMajor    = ck::tensor_layout::gemm::RowMajor;
-    using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
-
-    bool res = true;
-    std::vector<DeviceGemmNoOpPtr> gemmPtrs;
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(gemmPtrs);
-#if 0
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instances(gemmPtrs);
-#endif
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(gemmPtrs);
-#if 0
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances(gemmPtrs);
-#endif
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(gemmPtrs);
-#if 0
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(gemmPtrs);
-#endif
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(gemmPtrs);
-#if 0
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(gemmPtrs);
-#endif
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(gemmPtrs);
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
-    return res ? 0 : 1;
-}

test/gemm/gemm_xdl_fp32.cpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <algorithm>
-#include <cstdlib>
-#include <iostream>
-#include <numeric>
-#include <tuple>
-#include <vector>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-
-#include "test/gemm/gemm_util.hpp"
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using DeviceGemmNoOpPtr =
-    ck::tensor_operation::device::DeviceGemmPtr<ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough>;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace device_gemm_instance {
-void add_device_gemm_xdl_f32_f32_f32_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_f32_f32_f32_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-
-#if 0
-void add_device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-#endif
-
-void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-
-} // namespace device_gemm_instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-int main()
-{
-    using ADataType   = float;
-    using BDataType   = float;
-    using CDataType   = float;
-    using AccDataType = float;
-
-    using RowMajor    = ck::tensor_layout::gemm::RowMajor;
-    using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
-
-    bool res = true;
-    std::vector<DeviceGemmNoOpPtr> gemmPtrs;
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f32_f32_f32_km_kn_mn_instances(gemmPtrs);
-#if 0
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instances(gemmPtrs);
-#endif
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f32_f32_f32_km_nk_mn_instances(gemmPtrs);
-#if 0
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instances(gemmPtrs);
-#endif
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances(gemmPtrs);
-#if 0
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instances(gemmPtrs);
-#endif
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(gemmPtrs);
-#if 0
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instances(gemmPtrs);
-#endif
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
-    return res ? 0 : 1;
-}

test/gemm/gemm_xdl_fp64.cpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <algorithm>
-#include <cstdlib>
-#include <iostream>
-#include <numeric>
-#include <tuple>
-#include <vector>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-
-#include "test/gemm/gemm_util.hpp"
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using DeviceGemmNoOpPtr =
-    ck::tensor_operation::device::DeviceGemmPtr<ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough>;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace device_gemm_instance {
-void add_device_gemm_xdl_f64_f64_f64_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_f64_f64_f64_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_f64_f64_f64_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_f64_f64_f64_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-
-} // namespace device_gemm_instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-inline std::string get_device_name()
-{
-    hipDeviceProp_t props{};
-    int device;
-    auto status = hipGetDevice(&device);
-    if(status != hipSuccess)
-    {
-        return std::string();
-    }
-
-    status = hipGetDeviceProperties(&props, device);
-    if(status != hipSuccess)
-    {
-        return std::string();
-    }
-    const std::string name(props.gcnArchName);
-
-    return name;
-}
-
-int main()
-{
-    if(get_device_name().find("gfx90a") == std::string::npos)
-    {
-        std::cout << "TestGemm ..... SUCCESS" << std::endl;
-        return 0;
-    }
-    using ADataType   = double;
-    using BDataType   = double;
-    using CDataType   = double;
-    using AccDataType = double;
-
-    using RowMajor    = ck::tensor_layout::gemm::RowMajor;
-    using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
-
-    bool res = true;
-    std::vector<DeviceGemmNoOpPtr> gemmPtrs;
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f64_f64_f64_km_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f64_f64_f64_km_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f64_f64_f64_mk_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_f64_f64_f64_mk_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-    std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
-    return res ? 0 : 1;
-}

test/gemm/gemm_xdl_int8.cpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <algorithm>
-#include <cstdlib>
-#include <iostream>
-#include <numeric>
-#include <tuple>
-#include <vector>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-
-#include "test/gemm/gemm_util.hpp"
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using DeviceGemmNoOpPtr =
-    ck::tensor_operation::device::DeviceGemmPtr<ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough>;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace device_gemm_instance {
-void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
-} // namespace device_gemm_instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-int main()
-{
-    using ADataType   = int8_t;
-    using BDataType   = int8_t;
-    using CDataType   = int8_t;
-    using AccDataType = int32_t;
-
-    using RowMajor    = ck::tensor_layout::gemm::RowMajor;
-    using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
-
-    std::vector<DeviceGemmNoOpPtr> gemmPtrs;
-    bool res = true;
-
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       ColumnMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       RowMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    gemmPtrs.clear();
-    ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(gemmPtrs);
-
-    for(auto& gemmPtr : gemmPtrs)
-    {
-        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
-                                       ADataType,
-                                       BDataType,
-                                       CDataType,
-                                       AccDataType,
-                                       RowMajor,
-                                       ColumnMajor,
-                                       RowMajor,
-                                       PassThrough,
-                                       PassThrough,
-                                       PassThrough>{}(gemmPtr);
-    }
-
-    std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
-    return res ? 0 : 1;
-}

test/gemm_split_k/gemm_split_k.cpp

@@ -11,6 +11,8 @@
 #include "ck/tensor_operation/gpu/device/device_gemm_xdl_splitk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp"
+
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/host_tensor/device_memory.hpp"
 #include "ck/library/host_tensor/host_tensor.hpp"
@@ -27,30 +29,6 @@ enum struct GemmMatrixLayout
     KM_NK_MN, // 3
 };
 
-using DeviceGemmSplitKNoOpPtr = ck::tensor_operation::device::DeviceGemmSplitKPtr<
-    ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::PassThrough>;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace device_gemm_instance {
-
-void add_device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instances(
-    std::vector<DeviceGemmSplitKNoOpPtr>&);
-void add_device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instances(
-    std::vector<DeviceGemmSplitKNoOpPtr>&);
-void add_device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instances(
-    std::vector<DeviceGemmSplitKNoOpPtr>&);
-void add_device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instances(
-    std::vector<DeviceGemmSplitKNoOpPtr>&);
-
-} // namespace device_gemm_instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
 template <typename T>
 static bool check_out(const Tensor<T>& ref, const Tensor<T>& result)
 {
@@ -82,6 +60,11 @@ struct gemmArgs
 
 int test_gemm(const gemmArgs& args)
 {
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
     bool a_row_major, b_row_major, c_row_major;
 
     switch(args.layout)
@@ -152,64 +135,79 @@ int test_gemm(const gemmArgs& args)
     b_device_buf.ToDevice(b_k_n.mData.data());
     c_device_buf.ToDevice(c_m_n_device_result.mData.data());
 
-    // add device GEMM instances
-    std::vector<DeviceGemmSplitKNoOpPtr> gemm_ptrs;
+    auto test = [&](auto a_layout, auto b_layout, auto c_layout) {
+        bool success = false;
+
+        using DeviceOp = ck::tensor_operation::device::DeviceGemmSplitK<decltype(a_layout),
+                                                                        decltype(b_layout),
+                                                                        decltype(c_layout),
+                                                                        float,
+                                                                        float,
+                                                                        float,
+                                                                        PassThrough,
+                                                                        PassThrough,
+                                                                        PassThrough>;
+
+        const auto gemm_ptrs =
+            ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+                DeviceOp>::GetInstances();
+
+        for(auto& gemm_ptr : gemm_ptrs)
+        {
+            auto argument_ptr =
+                gemm_ptr->MakeArgumentPointer(static_cast<float*>(a_device_buf.GetDeviceBuffer()),
+                                              static_cast<float*>(b_device_buf.GetDeviceBuffer()),
+                                              static_cast<float*>(c_device_buf.GetDeviceBuffer()),
+                                              args.M,
+                                              args.N,
+                                              args.K,
+                                              args.StrideA,
+                                              args.StrideB,
+                                              args.StrideC,
+                                              ck::tensor_operation::element_wise::PassThrough{},
+                                              ck::tensor_operation::element_wise::PassThrough{},
+                                              ck::tensor_operation::element_wise::PassThrough{},
+                                              args.KBatch);
+
+            auto invoker_ptr = gemm_ptr->MakeInvokerPointer();
+
+            if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
+            {
+                invoker_ptr->Run(argument_ptr.get());
+
+                c_device_buf.FromDevice(c_m_n_device_result.mData.data());
+
+                if(!check_out(c_m_n_host_result, c_m_n_device_result))
+                {
+                    success = false;
+                    break;
+                }
+                success = true;
+            }
+        }
+
+        return success;
+    };
+
+    bool success = false;
 
     if(args.layout == GemmMatrixLayout::MK_KN_MN)
     {
-        ck::tensor_operation::device::device_gemm_instance::
-            add_device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instances(gemm_ptrs);
+        success = test(Row{}, Row{}, Row{});
     }
     else if(args.layout == GemmMatrixLayout::MK_NK_MN)
     {
-        ck::tensor_operation::device::device_gemm_instance::
-            add_device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instances(gemm_ptrs);
+        success = test(Row{}, Col{}, Row{});
     }
     else if(args.layout == GemmMatrixLayout::KM_KN_MN)
     {
-        ck::tensor_operation::device::device_gemm_instance::
-            add_device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instances(gemm_ptrs);
+        success = test(Col{}, Row{}, Row{});
     }
     else
     {
-        ck::tensor_operation::device::device_gemm_instance::
-            add_device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instances(gemm_ptrs);
+        success = test(Col{}, Col{}, Row{});
     }
 
-    bool success = false;
-    for(auto& gemm_ptr : gemm_ptrs)
-    {
-        auto argument_ptr =
-            gemm_ptr->MakeArgumentPointer(static_cast<float*>(a_device_buf.GetDeviceBuffer()),
-                                          static_cast<float*>(b_device_buf.GetDeviceBuffer()),
-                                          static_cast<float*>(c_device_buf.GetDeviceBuffer()),
-                                          args.M,
-                                          args.N,
-                                          args.K,
-                                          args.StrideA,
-                                          args.StrideB,
-                                          args.StrideC,
-                                          ck::tensor_operation::element_wise::PassThrough{},
-                                          ck::tensor_operation::element_wise::PassThrough{},
-                                          ck::tensor_operation::element_wise::PassThrough{},
-                                          args.KBatch);
-
-        auto invoker_ptr = gemm_ptr->MakeInvokerPointer();
-
-        if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
-        {
-            invoker_ptr->Run(argument_ptr.get());
-
-            c_device_buf.FromDevice(c_m_n_device_result.mData.data());
-
-            if(!check_out(c_m_n_host_result, c_m_n_device_result))
-            {
-                success = false;
-                break;
-            }
-            success = true;
-        }
-    }
     auto error_code = 0;
     if(success)
     {

test/grouped_gemm/grouped_gemm_fp16.cpp

@@ -28,7 +28,7 @@ using DeviceGroupedGemmPtr_ = ck::tensor_operation::device::DeviceGroupedGemmPtr
 namespace ck {
 namespace tensor_operation {
 namespace device {
-namespace device_grouped_gemm_instance {
+namespace instance {
 void add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(
     std::vector<DeviceGroupedGemmPtr_>&);
 }
@@ -197,7 +197,7 @@ bool TestGroupedGemm(DeviceGroupedGemmPtr_& groupedGemmPtr)
 int main()
 {
     std::vector<DeviceGroupedGemmPtr_> groupedGemmPtrs;
-    ck::tensor_operation::device::device_grouped_gemm_instance::
+    ck::tensor_operation::device::instance::
         add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(groupedGemmPtrs);
 
     bool res = true;

test/softmax/CMakeLists.txt

@@ -2,7 +2,10 @@ add_custom_target(test_softmax)
 
 add_gtest_executable(test_softmax_fp32 test_softmax_fp32.cpp)
 add_gtest_executable(test_softmax_fp16 test_softmax_fp16.cpp)
+add_gtest_executable(test_softmax_int8 test_softmax_int8.cpp)
 target_link_libraries(test_softmax_fp32 PRIVATE host_tensor)
 target_link_libraries(test_softmax_fp16 PRIVATE host_tensor)
+target_link_libraries(test_softmax_int8 PRIVATE host_tensor)
 add_dependencies(test_softmax test_softmax_fp32)
-add_dependencies(test_softmax test_softmax_fp16)
\ No newline at end of file
+add_dependencies(test_softmax test_softmax_fp16)
+add_dependencies(test_softmax test_softmax_int8)
\ No newline at end of file

test/softmax/test_softmax_fp16.cpp

@@ -15,14 +15,19 @@ class TestSoftmaxFP16 : public ck::TestSoftmax<Tuple>
 // clang-format off
 using KernelTypes = ::testing::Types<
 // InDataType, AccDataType, OutDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, InSrcVectorDim, InSrcVectorSize, OutDstVectorSize>
+    std::tuple<ck::half_t, float, float, I<3>, I<1>, I<256>, I<8>, I<32>, I<1>, I<8>, I<1>, I<8>, I<4>>, // mixed precision
     std::tuple<ck::half_t, float, ck::half_t, I<3>, I<1>, I<256>, I<8>, I<32>, I<1>, I<8>, I<1>, I<8>, I<8>>,
     std::tuple<ck::half_t, float, ck::half_t, I<3>, I<1>, I<256>, I<4>, I<64>, I<1>, I<8>, I<1>, I<8>, I<8>>,
     std::tuple<ck::half_t, float, ck::half_t, I<3>, I<1>, I<256>, I<2>, I<128>, I<1>, I<8>, I<1>, I<8>, I<8>>,
     std::tuple<ck::half_t, float, ck::half_t, I<3>, I<1>, I<256>, I<1>, I<256>, I<1>, I<8>, I<1>, I<8>, I<8>>,
+    std::tuple<ck::half_t, float, ck::half_t, I<3>, I<1>, I<256>, I<1>, I<256>, I<1>, I<16>, I<1>, I<8>, I<8>>,
+    std::tuple<ck::half_t, float, ck::half_t, I<3>, I<1>, I<256>, I<1>, I<256>, I<1>, I<32>, I<1>, I<8>, I<8>>,
     std::tuple<ck::half_t, float, ck::half_t, I<3>, I<2>, I<256>, I<8>, I<32>, I<1>, I<8>, I<1>, I<8>, I<8>>,
     std::tuple<ck::half_t, float, ck::half_t, I<3>, I<2>, I<256>, I<4>, I<64>, I<1>, I<8>, I<1>, I<8>, I<8>>,
     std::tuple<ck::half_t, float, ck::half_t, I<3>, I<2>, I<256>, I<2>, I<128>, I<1>, I<8>, I<1>, I<8>, I<8>>,
-    std::tuple<ck::half_t, float, ck::half_t, I<3>, I<2>, I<256>, I<1>, I<256>, I<1>, I<8>, I<1>, I<8>, I<8>>
+    std::tuple<ck::half_t, float, ck::half_t, I<3>, I<2>, I<256>, I<1>, I<256>, I<1>, I<8>, I<1>, I<8>, I<8>>,
+    std::tuple<ck::half_t, float, ck::half_t, I<3>, I<2>, I<256>, I<1>, I<256>, I<1>, I<16>, I<1>, I<8>, I<8>>,
+    std::tuple<ck::half_t, float, ck::half_t, I<3>, I<2>, I<256>, I<1>, I<256>, I<1>, I<32>, I<1>, I<8>, I<8>>
     >;
 // clang-format on
 TYPED_TEST_SUITE(TestSoftmaxFP16, KernelTypes);

test/softmax/test_softmax_fp32.cpp

@@ -15,14 +15,19 @@ class TestSoftmaxFP32 : public ck::TestSoftmax<Tuple>
 // clang-format off
 using KernelTypes = ::testing::Types<
 // InDataType, AccDataType, OutDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, InSrcVectorDim, InSrcVectorSize, OutDstVectorSize>
+    std::tuple<float, float, ck::half_t, I<3>, I<2>, I<256>, I<1>, I<256>, I<1>, I<8>, I<1>, I<4>, I<8>>, // mixed precision
     std::tuple<float, float, float, I<3>, I<1>, I<256>, I<8>, I<32>, I<1>, I<4>, I<1>, I<4>, I<4>>,
     std::tuple<float, float, float, I<3>, I<1>, I<256>, I<4>, I<64>, I<1>, I<4>, I<1>, I<4>, I<4>>,
     std::tuple<float, float, float, I<3>, I<1>, I<256>, I<2>, I<128>, I<1>, I<4>, I<1>, I<4>, I<4>>,
     std::tuple<float, float, float, I<3>, I<1>, I<256>, I<1>, I<256>, I<1>, I<4>, I<1>, I<4>, I<4>>,
+    std::tuple<float, float, float, I<3>, I<1>, I<256>, I<1>, I<256>, I<1>, I<8>, I<1>, I<4>, I<4>>,
+    std::tuple<float, float, float, I<3>, I<1>, I<256>, I<1>, I<256>, I<1>, I<16>, I<1>, I<4>, I<4>>,
     std::tuple<float, float, float, I<3>, I<2>, I<256>, I<8>, I<32>, I<1>, I<4>, I<1>, I<4>, I<4>>,
     std::tuple<float, float, float, I<3>, I<2>, I<256>, I<4>, I<64>, I<1>, I<4>, I<1>, I<4>, I<4>>,
     std::tuple<float, float, float, I<3>, I<2>, I<256>, I<2>, I<128>, I<1>, I<4>, I<1>, I<4>, I<4>>,
-    std::tuple<float, float, float, I<3>, I<2>, I<256>, I<1>, I<256>, I<1>, I<4>, I<1>, I<4>, I<4>>
+    std::tuple<float, float, float, I<3>, I<2>, I<256>, I<1>, I<256>, I<1>, I<4>, I<1>, I<4>, I<4>>,
+    std::tuple<float, float, float, I<3>, I<2>, I<256>, I<1>, I<256>, I<1>, I<8>, I<1>, I<4>, I<4>>,
+    std::tuple<float, float, float, I<3>, I<2>, I<256>, I<1>, I<256>, I<1>, I<16>, I<1>, I<4>, I<4>>
     >;
 // clang-format on
 TYPED_TEST_SUITE(TestSoftmaxFP32, KernelTypes);