Navi21 gemm (#197)

* start adding navi21 GEMM

* navi_gemm_km_kn_mn_fp32 compiles and passes one test.

* rename variables and functions in gridwise_gemm_dlops_v1r3

* add other 3 layouts; format instance

* adding more tuning parameters

add tuning parameters for other 3 layouts

* add gemm_dlops_f16

* tmp

* add dependence of DeviceGemm::IsSupportedArg() on arch

* minor changes

* minor changes

* minor changes

* minor changes

* minor changes

* minor changes

* minor changes

* push gemm_dlops into profiler

* minor changes

* if using xdl or dlops is moved into profiler_gemm_impl

* minor changes

* minor changes

* remove is_xdl from profile_gemm_impl

* make IsSupportedArg dependent on arch for other device_gemm

* minor changes

* minor changes

* fix a bug in f_generate_tensor_value

* add 64x64x64 for gemm_dlops_int8

* add 64x64x64 for gemm_dlops_int8

* comment out 3 layouts in gemm_dlops_int8; add 32x32x32 for gemm_dlops_int8; init A values to 1

* fix

* start fixing tuning parameters

* monir

* minor changes

* minor changes

* minor changes

* fixing

* adding example

* adding example

* adding example

* add gemm fp32 example

* clean up

* use 128x128x16 as MNK tile in navi21 gemm example

* bug fix

* fix test

* use new block c tile

* clean

* fix build
Co-authored-by: Chao Liu <chao.liu2@amd.com>
Co-authored-by: shaojiewang <wsjmessi@163.com>

test/CMakeLists.txt

@@ -2,6 +2,7 @@ include_directories(BEFORE
     ${PROJECT_SOURCE_DIR}/
     ${PROJECT_SOURCE_DIR}/include/ck
     ${PROJECT_SOURCE_DIR}/include/ck/utility
+    ${PROJECT_SOURCE_DIR}/include/ck/host_utility
     ${PROJECT_SOURCE_DIR}/include/ck/tensor_description
     ${PROJECT_SOURCE_DIR}/include/ck/tensor
     ${PROJECT_SOURCE_DIR}/include/ck/problem_transform

test/gemm/CMakeLists.txt

-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)
+# 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_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_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_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_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_int8 gemm_int8.cpp)
-target_link_libraries(test_gemm_int8 PRIVATE host_tensor)
-target_link_libraries(test_gemm_int8 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)

test/gemm/gemm_dl_fp16.cpp

+#include <algorithm>
+#include <cstdlib>
+#include <half.hpp>
+#include <iostream>
+#include <numeric>
+#include <tuple>
+#include <vector>
+
+#include "../gemm/gemm_util.hpp"
+#include "config.hpp"
+#include "print.hpp"
+#include "device.hpp"
+#include "host_tensor.hpp"
+#include "host_tensor_generator.hpp"
+#include "host_gemm.hpp"
+#include "device_tensor.hpp"
+#include "device_gemm_dl.hpp"
+#include "element_wise_operation.hpp"
+#include "reference_gemm.hpp"
+#include "gemm_specialization.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 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,
+                                       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,
+                                       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,
+                                       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,
+                                       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

+#include <algorithm>
+#include <cstdlib>
+#include <half.hpp>
+#include <iostream>
+#include <numeric>
+#include <tuple>
+#include <vector>
+
+#include "../gemm/gemm_util.hpp"
+#include "config.hpp"
+#include "print.hpp"
+#include "device.hpp"
+#include "host_tensor.hpp"
+#include "host_tensor_generator.hpp"
+#include "host_gemm.hpp"
+#include "device_tensor.hpp"
+#include "device_gemm_dl.hpp"
+#include "element_wise_operation.hpp"
+#include "reference_gemm.hpp"
+#include "gemm_specialization.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 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,
+                                       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,
+                                       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,
+                                       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,
+                                       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

+#include <algorithm>
+#include <cstdlib>
+#include <half.hpp>
+#include <iostream>
+#include <numeric>
+#include <tuple>
+#include <vector>
+
+#include "../gemm/gemm_util.hpp"
+#include "config.hpp"
+#include "print.hpp"
+#include "device.hpp"
+#include "host_tensor.hpp"
+#include "host_tensor_generator.hpp"
+#include "host_gemm.hpp"
+#include "device_tensor.hpp"
+#include "device_gemm_dl.hpp"
+#include "element_wise_operation.hpp"
+#include "reference_gemm.hpp"
+#include "gemm_specialization.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 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,
+                                       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,
+                                       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,
+                                       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,
+                                       RowMajor,
+                                       ColumnMajor,
+                                       RowMajor,
+                                       PassThrough,
+                                       PassThrough,
+                                       PassThrough>{}(gemmPtr);
+    }
+
+    std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
+    return res ? 0 : 1;
+}

test/gemm/gemm_util.hpp

@@ -60,7 +60,7 @@ template <typename DeviceGemmPtr_,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           typename CElementwiseOperation>
-void RunDeviceGEMM(DeviceGemmPtr_& gemmPtr,
+bool RunDeviceGEMM(DeviceGemmPtr_& gemmPtr,
                    const ck::gemm_util::GemmParams& params,
                    const Tensor<ADataType>& A,
                    const Tensor<BDataType>& B,
@@ -73,9 +73,6 @@ void RunDeviceGEMM(DeviceGemmPtr_& gemmPtr,
     DeviceMem b_k_n_device_buf(sizeof(BDataType) * B.mDesc.GetElementSpace());
     DeviceMem c_m_n_device_buf(sizeof(CDataType) * C.mDesc.GetElementSpace());
 
-    a_m_k_device_buf.ToDevice(A.mData.data());
-    b_k_n_device_buf.ToDevice(B.mData.data());
-
     auto invoker_ptr = gemmPtr->MakeInvokerPointer();
     auto argument_ptr =
         gemmPtr->MakeArgumentPointer(static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
@@ -91,15 +88,23 @@ void RunDeviceGEMM(DeviceGemmPtr_& gemmPtr,
                                      b_element_op,
                                      c_element_op);
 
-    if(!gemmPtr->IsSupportedArgument(argument_ptr.get()))
+    if(gemmPtr->IsSupportedArgument(argument_ptr.get()))
     {
-        throw std::runtime_error(
-            "wrong! device_gemm with the specified compilation parameters does "
-            "not support this GEMM problem");
+        a_m_k_device_buf.ToDevice(A.mData.data());
+        b_k_n_device_buf.ToDevice(B.mData.data());
+        invoker_ptr->Run(argument_ptr.get());
+        c_m_n_device_buf.FromDevice(C.mData.data());
+
+        return true;
     }
+    else
+    {
+        std::cout << "device_gemm with the specified compilation parameters does "
+                     "not support this GEMM problem"
+                  << std::endl;
 
-    invoker_ptr->Run(argument_ptr.get());
-    c_m_n_device_buf.FromDevice(C.mData.data());
+        return false;
+    }
 }
 
 template <typename DeviceGemmPtr_,
@@ -188,28 +193,35 @@ struct TestGemm
             a, b, c_host, a_element_op, b_element_op, c_element_op);
 
         // Act
-        ck::gemm_util::RunDeviceGEMM(
+        bool is_supported = ck::gemm_util::RunDeviceGEMM(
             gemmPtr, params, a, b, c_device, a_element_op, b_element_op, c_element_op);
 
-        // Assert
-        bool res = false;
-        if(std::is_same<CDataType, float>::value)
+        if(is_supported)
         {
-            res = ck::utils::check_err(c_device.mData, c_host.mData);
-            std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
+            // Assert
+            bool res = false;
+            if(std::is_same<CDataType, float>::value)
+            {
+                res = ck::utils::check_err(c_device.mData, c_host.mData);
+                std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
+            }
+            else if(std::is_same<CDataType, ck::half_t>::value)
+            {
+                res = ck::utils::check_err(c_device.mData, c_host.mData);
+                std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
+            }
+            else if(std::is_same<CDataType, int8_t>::value)
+            {
+                res = ck::utils::check_err(c_device.mData, c_host.mData);
+                std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
+            }
+
+            return res;
         }
-        else if(std::is_same<CDataType, ck::half_t>::value)
+        else
         {
-            res = ck::utils::check_err(c_device.mData, c_host.mData);
-            std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
+            return true;
         }
-        else if(std::is_same<CDataType, int8_t>::value)
-        {
-            res = ck::utils::check_err(c_device.mData, c_host.mData);
-            std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
-        }
-
-        return res;
     }
 };
 

test/gemm/gemm_int8.cpp → test/gemm/gemm_xdl_int8.cpp

@@ -31,14 +31,10 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace device_gemm_instance {
-void add_device_gemm_xdl_c_shuffle_int8_int8_int8_km_kn_mn_instances(
-    std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_int8_int8_int8_km_nk_mn_instances(
-    std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_int8_int8_int8_mk_nk_mn_instances(
-    std::vector<DeviceGemmNoOpPtr>&);
-void add_device_gemm_xdl_c_shuffle_int8_int8_int8_mk_kn_mn_instances(
-    std::vector<DeviceGemmNoOpPtr>&);
+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
@@ -57,7 +53,7 @@ int main()
     bool res = true;
 
     ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_int8_int8_int8_km_kn_mn_instances(gemmPtrs);
+        add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(gemmPtrs);
 
     for(auto& gemmPtr : gemmPtrs)
     {
@@ -75,7 +71,7 @@ int main()
 
     gemmPtrs.clear();
     ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_int8_int8_int8_km_nk_mn_instances(gemmPtrs);
+        add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(gemmPtrs);
 
     for(auto& gemmPtr : gemmPtrs)
     {
@@ -93,7 +89,7 @@ int main()
 
     gemmPtrs.clear();
     ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_int8_int8_int8_mk_kn_mn_instances(gemmPtrs);
+        add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(gemmPtrs);
 
     for(auto& gemmPtr : gemmPtrs)
     {
@@ -111,7 +107,7 @@ int main()
 
     gemmPtrs.clear();
     ck::tensor_operation::device::device_gemm_instance::
-        add_device_gemm_xdl_c_shuffle_int8_int8_int8_mk_nk_mn_instances(gemmPtrs);
+        add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(gemmPtrs);
 
     for(auto& gemmPtr : gemmPtrs)
     {