fix errors

include/ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp

@@ -96,51 +96,6 @@ struct DeviceGemmMultipleDSplitK : public BaseOperator
     virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
 };
 
-// GEMM:
-//   input : A[M, K], B[K, N],
-//   input : D0[M, N], D1[M, N], ...
-//   output : E[M, N]
-//   C = a_op(A) * b_op(B)
-//   E = cde_op(C, D0, D1, ...)
-// Assume:
-//   D0, D1, ... and E have the same layout
-template <typename ALayout,
-          typename BLayout,
-          typename DsLayout,
-          typename ELayout,
-          typename ADataType,
-          typename BDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation>
-struct DeviceGemmMultipleDSplitKBPreShuffle : public BaseOperator
-{
-    static constexpr index_t NumDTensor = DsDataType::Size();
-
-    virtual std::unique_ptr<BaseArgument>
-    MakeArgumentPointer(const void* p_a,
-                        const void* p_b,
-                        std::array<const void*, NumDTensor> p_ds,
-                        void* p_e,
-                        ck::index_t M,
-                        ck::index_t N,
-                        ck::index_t K,
-                        ck::index_t StrideA,
-                        ck::index_t StrideB,
-                        std::array<ck::index_t, NumDTensor> StrideDs,
-                        ck::index_t StrideE,
-                        ck::index_t KBatch,
-                        AElementwiseOperation a_element_op,
-                        BElementwiseOperation b_element_op,
-                        CDEElementwiseOperation cde_element_op) = 0;
-
-    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
-
-    virtual int GetPreShuffleParameters() = 0;
-};
-
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck

profiler/src/profile_gemm_multiply_multiply.cpp

@@ -175,7 +175,7 @@ int profile_gemm_multiply_multiply(int argc, char* argv[])
     else if(data_type == GemmDataType::INT8_INT8_BF16 && layout == GemmMatrixLayout::MK_NK_MN)
     {
         return profile(
-            F8{}, F8{}, F8{}, F32{}, F32{}, F32{}, F16{}, Row{}, Col{}, Row{}, Col{}, Row{});
+            I8{}, I8{}, I8{}, I32{}, F32{}, F32{}, BF16{}, Row{}, Col{}, Row{}, Col{}, Row{});
     }
     else if(data_type == GemmDataType::INT8_INT8_F16 && layout == GemmMatrixLayout::MK_NK_MN)
     {