refactored deviceBatchedGemm; removed GridwiseBatchedGemm; added fp32 and int8 to profiler (#120)

changed long_index_t to index_t when computing memory offset

uncomment other ops in profiler

added test for batched_gemm

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

@@ -10,12 +10,68 @@
 #include "tensor_layout.hpp"
 #include "tensor_descriptor.hpp"
 #include "tensor_descriptor_helper.hpp"
-#include "gridwise_batched_gemm_xdlops_v2r3.hpp"
+#include "gridwise_gemm_xdlops_v2r3.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 
+template <typename GridwiseGemm,
+          typename FloatAB,
+          typename FloatC,
+          typename AGridDesc_K0_M_K1,
+          typename BGridDesc_K0_N_K1,
+          typename CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          typename ComputeBasePrtOfBatch,
+          typename Block2CTileMap,
+          bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_batched_gemm_xdlops_v2r3(
+            const FloatAB* __restrict__ p_a_grid,
+            const FloatAB* __restrict__ p_b_grid,
+            FloatC* __restrict__ p_c_grid,
+            const index_t num_batches,
+            const AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1,
+            const BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1,
+            const CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CElementwiseOperation c_element_op,
+            const ComputeBasePrtOfBatch compute_base_ptr_of_batch_,
+            const Block2CTileMap block_2_ctile_map)
+{
+    const index_t num_blocks_per_batch =
+        __builtin_amdgcn_readfirstlane(get_grid_size() / num_batches);
+    const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
+
+    const long_index_t a_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_base_ptr_of_batch_.GetABasePtr(g_idx)));
+    const long_index_t b_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_base_ptr_of_batch_.GetBBasePtr(g_idx)));
+    const long_index_t c_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_base_ptr_of_batch_.GetCBasePtr(g_idx)));
+
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid + a_batch_offset,
+                                                  p_b_grid + b_batch_offset,
+                                                  p_c_grid + c_batch_offset,
+                                                  p_shared,
+                                                  a_grid_desc_k0_m_k1,
+                                                  b_grid_desc_k0_n_k1,
+                                                  c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                                                  a_element_op,
+                                                  b_element_op,
+                                                  c_element_op,
+                                                  block_2_ctile_map);
+}
+
 template <typename ADataType,
           typename BDataType,
           typename CDataType,
@@ -35,14 +91,14 @@ template <typename ADataType,
           ck::index_t NPerXDL,
           ck::index_t MXdlPerWave,
           ck::index_t NXdlPerWave,
-          typename ABlockTransferThreadClusterLengths_G_K0_M_K1,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
           typename ABlockTransferThreadClusterArrangeOrder,
           typename ABlockTransferSrcAccessOrder,
           ck::index_t ABlockTransferSrcVectorDim,
           ck::index_t ABlockTransferSrcScalarPerVector,
           ck::index_t ABlockTransferDstScalarPerVector_K1,
           bool ABlockLdsAddExtraM,
-          typename BBlockTransferThreadClusterLengths_G_K0_N_K1,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
           typename BBlockTransferThreadClusterArrangeOrder,
           typename BBlockTransferSrcAccessOrder,
           ck::index_t BBlockTransferSrcVectorDim,
@@ -57,149 +113,215 @@ struct DeviceBatchedGemmXdl
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
     static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
 
     static constexpr auto K1Number = Number<K1>{};
 
-    static auto
-    MakeAGridDescriptor_G_K0_M_K1(index_t BatchCount, index_t M, index_t K, index_t StrideA)
+    static auto MakeAGridDescriptor_K0_M_K1(index_t M, index_t K, index_t StrideA)
     {
         assert(K % K1 == 0);
 
         const index_t K0 = K / K1;
 
-        const auto a_grid_desc_g_m_k = [&]() {
+        const auto a_grid_desc_m_k = [&]() {
             if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
             {
-                return make_naive_tensor_descriptor(make_tuple(BatchCount, M, K),
-                                                    make_tuple(M * StrideA, StrideA, I1));
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
             }
             else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
             {
-                return make_naive_tensor_descriptor(make_tuple(BatchCount, M, K),
-                                                    make_tuple(K * StrideA, I1, StrideA));
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, M));
             }
         }();
 
         const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
 
-        const auto a_grid_desc_g_k0_mp_k1 =
-            transform_tensor_descriptor(a_grid_desc_g_m_k,
-                                        make_tuple(make_pass_through_transform(BatchCount),
-                                                   make_unmerge_transform(make_tuple(K0, K1Number)),
+        const auto a_grid_desc_k0_mp_k1 =
+            transform_tensor_descriptor(a_grid_desc_m_k,
+                                        make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
                                                    make_right_pad_transform(M, PadM)),
-                                        make_tuple(Sequence<0>{}, Sequence<2>{}, Sequence<1>{}),
-                                        make_tuple(Sequence<0>{}, Sequence<1, 3>{}, Sequence<2>{}));
+                                        make_tuple(Sequence<1>{}, Sequence<0>{}),
+                                        make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
 
-        return a_grid_desc_g_k0_mp_k1;
+        return a_grid_desc_k0_mp_k1;
     }
 
-    static auto
-    MakeBGridDescriptor_G_K0_N_K1(index_t BatchCount, index_t K, index_t N, index_t StrideB)
+    static auto MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t StrideB)
     {
         assert(K % K1 == 0);
 
         const index_t K0 = K / K1;
 
-        const auto b_grid_desc_g_k_n = [&]() {
+        const auto b_grid_desc_k_n = [&]() {
             if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
             {
-                return make_naive_tensor_descriptor(make_tuple(BatchCount, K, N),
-                                                    make_tuple(K * StrideB, StrideB, I1));
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
             }
             else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
             {
-                return make_naive_tensor_descriptor(make_tuple(BatchCount, K, N),
-                                                    make_tuple(N * StrideB, I1, StrideB));
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, K));
             }
         }();
 
         const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
 
-        const auto b_grid_desc_g_k0_np_k1 =
-            transform_tensor_descriptor(b_grid_desc_g_k_n,
-                                        make_tuple(make_pass_through_transform(BatchCount),
-                                                   make_unmerge_transform(make_tuple(K0, K1Number)),
+        const auto b_grid_desc_k0_np_k1 =
+            transform_tensor_descriptor(b_grid_desc_k_n,
+                                        make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
                                                    make_right_pad_transform(N, PadN)),
-                                        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
-                                        make_tuple(Sequence<0>{}, Sequence<1, 3>{}, Sequence<2>{}));
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                        make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
 
-        return b_grid_desc_g_k0_np_k1;
+        return b_grid_desc_k0_np_k1;
     }
 
-    static auto MakeCGridDescriptor_G_M_N(index_t BatchCount, index_t M, index_t N, index_t StrideC)
+    static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
     {
-        const auto c_grid_desc_g_m_n = [&]() {
+        const auto c_grid_desc_m_n = [&]() {
             if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
             {
-                return make_naive_tensor_descriptor(make_tuple(BatchCount, M, N),
-                                                    make_tuple(M * StrideC, StrideC, I1));
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
             }
             else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
             {
-                return make_naive_tensor_descriptor(make_tuple(BatchCount, M, N),
-                                                    make_tuple(N * StrideC, I1, StrideC));
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, M));
             }
         }();
 
         const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
         const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
 
-        const auto c_grid_desc_g_mp_np =
-            transform_tensor_descriptor(c_grid_desc_g_m_n,
-                                        make_tuple(make_pass_through_transform(BatchCount),
-                                                   make_right_pad_transform(M, PadM),
-                                                   make_right_pad_transform(N, PadN)),
-                                        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
-                                        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+        const auto c_grid_desc_mp_np = transform_tensor_descriptor(
+            c_grid_desc_m_n,
+            make_tuple(make_right_pad_transform(M, PadM), make_right_pad_transform(N, PadN)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
 
-        return c_grid_desc_g_mp_np;
+        return c_grid_desc_mp_np;
     }
 
-    using AGridDesc_G_K0_M_K1 = decltype(MakeAGridDescriptor_G_K0_M_K1(1, 1, 1, 1));
-    using BGridDesc_G_K0_N_K1 = decltype(MakeBGridDescriptor_G_K0_N_K1(1, 1, 1, 1));
-    using CGridDesc_G_M_N     = decltype(MakeCGridDescriptor_G_M_N(1, 1, 1, 1));
-
-    // GridwiseBatchedGemm
-    using GridwiseBatchedGemm = GridwiseBatchedGemm_gk0mk1_gk0nk1_gmn_xdlops_v2r3<
-        BlockSize,
-        ADataType, // TODO: distinguish A/B datatype
-        AccDataType,
-        CDataType,
-        InMemoryDataOperationEnum_t::Set,
-        AGridDesc_G_K0_M_K1,
-        BGridDesc_G_K0_N_K1,
-        CGridDesc_G_M_N,
-        AElementwiseOperation,
-        BElementwiseOperation,
-        CElementwiseOperation,
-        MPerBlock,
-        NPerBlock,
-        K0PerBlock,
-        MPerXDL,
-        NPerXDL,
-        K1,
-        MXdlPerWave,
-        NXdlPerWave,
-        ABlockTransferThreadClusterLengths_G_K0_M_K1,
-        ABlockTransferThreadClusterArrangeOrder,
-        ABlockTransferSrcAccessOrder,
-        ABlockTransferSrcVectorDim,
-        ABlockTransferSrcScalarPerVector,
-        ABlockTransferDstScalarPerVector_K1,
-        false, // AThreadTransferSrcResetCoordinateAfterRun,
-        ABlockLdsAddExtraM,
-        BBlockTransferThreadClusterLengths_G_K0_N_K1,
-        BBlockTransferThreadClusterArrangeOrder,
-        BBlockTransferSrcAccessOrder,
-        BBlockTransferSrcVectorDim,
-        BBlockTransferSrcScalarPerVector,
-        BBlockTransferDstScalarPerVector_K1,
-        false, // BThreadTransferSrcResetCoordinateAfterRun,
-        BBlockLdsAddExtraN,
-        Sequence<0, 1, 3, 5, 6, 7, 2, 4, 8>, // CThreadTransferSrcDstAccessOrder,
-        CThreadTransferSrcDstVectorDim,
-        CThreadTransferDstScalarPerVector>;
+    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
+    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
+    using CGridDesc_M_N     = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
+
+    struct Block2CTileMapMaker
+    {
+        Block2CTileMapMaker(index_t num_batches) : num_batches_(num_batches) {}
+
+        __host__ __device__ constexpr auto
+        MakeBlock2CTileMap(const CGridDesc_M_N& c_grid_desc_m_n, index_t M01, index_t N01)
+        {
+            const auto M = c_grid_desc_m_n.GetLength(I0);
+            const auto N = c_grid_desc_m_n.GetLength(I1);
+
+            constexpr auto M1 = Number<MPerBlock>{};
+            constexpr auto N1 = Number<NPerBlock>{};
+
+            const auto M0 = M / M1;
+            const auto N0 = N / N1;
+
+            const auto M00 = M0 / M01;
+            const auto N00 = N0 / N01;
+
+            const auto g_m00_m01_n00_n01_to_m0_n0_block_cluster_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_insert_transform(num_batches_),
+                               make_unmerge_transform(make_tuple(M00, M01)),
+                               make_unmerge_transform(make_tuple(N00, N01))),
+                    make_tuple(Sequence<>{}, Sequence<0>{}, Sequence<1>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1, 3>{}, Sequence<2, 4>{}));
+
+            const auto globalblockid_to_m00_m01_n00_n01_block_cluster_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_merge_transform(make_tuple(num_batches_, M00, N00, M01, N01))),
+                    make_tuple(Sequence<0, 1, 2, 3, 4>{}),
+                    make_tuple(Sequence<0>{}));
+
+            const auto globalblockid_to_m0_n0_block_cluster_adaptor =
+                chain_tensor_adaptors(g_m00_m01_n00_n01_to_m0_n0_block_cluster_adaptor,
+                                      globalblockid_to_m00_m01_n00_n01_block_cluster_adaptor);
+
+            return globalblockid_to_m0_n0_block_cluster_adaptor;
+        }
+
+        private:
+        index_t num_batches_;
+    };
+
+    struct ComputeBasePtrOfStridedBatch
+    {
+        ComputeBasePtrOfStridedBatch(index_t BatchStrideA,
+                                     index_t BatchStrideB,
+                                     index_t BatchStrideC)
+            : BatchStrideA_(BatchStrideA), BatchStrideB_(BatchStrideB), BatchStrideC_(BatchStrideC)
+        {
+        }
+
+        __host__ __device__ constexpr long_index_t GetABasePtr(index_t g_idx) const
+        {
+            return g_idx * static_cast<long_index_t>(BatchStrideA_);
+        }
+
+        __host__ __device__ constexpr long_index_t GetBBasePtr(index_t g_idx) const
+        {
+            return g_idx * static_cast<long_index_t>(BatchStrideB_);
+        }
+
+        __host__ __device__ constexpr long_index_t GetCBasePtr(index_t g_idx) const
+        {
+            return g_idx * static_cast<long_index_t>(BatchStrideC_);
+        }
+
+        private:
+        index_t BatchStrideA_;
+        index_t BatchStrideB_;
+        index_t BatchStrideC_;
+    };
+
+    // GridwiseGemm
+    using GridwiseGemm =
+        GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3<BlockSize,
+                                                ADataType, // TODO: distinguish A/B datatype
+                                                AccDataType,
+                                                CDataType,
+                                                InMemoryDataOperationEnum_t::Set,
+                                                AGridDesc_K0_M_K1,
+                                                BGridDesc_K0_N_K1,
+                                                CGridDesc_M_N,
+                                                AElementwiseOperation,
+                                                BElementwiseOperation,
+                                                CElementwiseOperation,
+                                                MPerBlock,
+                                                NPerBlock,
+                                                K0PerBlock,
+                                                MPerXDL,
+                                                NPerXDL,
+                                                K1,
+                                                MXdlPerWave,
+                                                NXdlPerWave,
+                                                ABlockTransferThreadClusterLengths_K0_M_K1,
+                                                ABlockTransferThreadClusterArrangeOrder,
+                                                ABlockTransferSrcAccessOrder,
+                                                ABlockTransferSrcVectorDim,
+                                                ABlockTransferSrcScalarPerVector,
+                                                ABlockTransferDstScalarPerVector_K1,
+                                                false, // AThreadTransferSrcResetCoordinateAfterRun,
+                                                ABlockLdsAddExtraM,
+                                                BBlockTransferThreadClusterLengths_K0_N_K1,
+                                                BBlockTransferThreadClusterArrangeOrder,
+                                                BBlockTransferSrcAccessOrder,
+                                                BBlockTransferSrcVectorDim,
+                                                BBlockTransferSrcScalarPerVector,
+                                                BBlockTransferDstScalarPerVector_K1,
+                                                false, // BThreadTransferSrcResetCoordinateAfterRun,
+                                                BBlockLdsAddExtraN,
+                                                Sequence<2, 3, 0, 1, 7, 5, 4, 6>,
+                                                CThreadTransferSrcDstVectorDim,
+                                                CThreadTransferDstScalarPerVector>;
+
+    using CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 =
+        decltype(GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(CGridDesc_M_N{}));
+    using Block2CTileMap =
+        decltype(Block2CTileMapMaker{1}.MakeBlock2CTileMap(CGridDesc_M_N{}, 1, 1));
 
     // Argument
     struct Argument : public BaseArgument
@@ -222,10 +344,16 @@ struct DeviceBatchedGemmXdl
             : p_a_grid_{p_a_grid},
               p_b_grid_{p_b_grid},
               p_c_grid_{p_c_grid},
-              a_grid_desc_g_k0_m_k1_{},
-              b_grid_desc_g_k0_n_k1_{},
-              c_grid_desc_g_m_n_{},
-              c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2_{},
+              BatchCount_(BatchCount),
+              a_grid_desc_k0_m_k1_{
+                  DeviceBatchedGemmXdl::MakeAGridDescriptor_K0_M_K1(M, K, StrideA)},
+              b_grid_desc_k0_n_k1_{
+                  DeviceBatchedGemmXdl::MakeBGridDescriptor_K0_N_K1(K, N, StrideB)},
+              c_grid_desc_m_n_{DeviceBatchedGemmXdl::MakeCGridDescriptor_M_N(M, N, StrideC)},
+              c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_{},
+              compute_base_ptr_of_batch_{a_grid_desc_k0_m_k1_.GetElementSpaceSize(),
+                                         b_grid_desc_k0_n_k1_.GetElementSpaceSize(),
+                                         c_grid_desc_m_n_.GetElementSpaceSize()},
               block_2_ctile_map_{},
               M01_{M01},
               N01_{N01},
@@ -233,22 +361,14 @@ struct DeviceBatchedGemmXdl
               b_element_op_{b_element_op},
               c_element_op_{c_element_op}
         {
-            a_grid_desc_g_k0_m_k1_ =
-                DeviceBatchedGemmXdl::MakeAGridDescriptor_G_K0_M_K1(BatchCount, M, K, StrideA);
-            b_grid_desc_g_k0_n_k1_ =
-                DeviceBatchedGemmXdl::MakeBGridDescriptor_G_K0_N_K1(BatchCount, K, N, StrideB);
-            c_grid_desc_g_m_n_ =
-                DeviceBatchedGemmXdl::MakeCGridDescriptor_G_M_N(BatchCount, M, N, StrideC);
-
-            if(GridwiseBatchedGemm::CheckValidity(
-                   a_grid_desc_g_k0_m_k1_, b_grid_desc_g_k0_n_k1_, c_grid_desc_g_m_n_, M01_, N01_))
+            if(GridwiseGemm::CheckValidity(
+                   a_grid_desc_k0_m_k1_, b_grid_desc_k0_n_k1_, c_grid_desc_m_n_, M01_, N01_))
             {
-                c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2_ =
-                    GridwiseBatchedGemm::MakeCGridDescriptor_G_M0_N0_M1_N1_M2_M3_M4_N2(
-                        c_grid_desc_g_m_n_);
+                c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_ =
+                    GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(c_grid_desc_m_n_);
 
                 block_2_ctile_map_ =
-                    GridwiseBatchedGemm::MakeDefaultBlock2CTileMap(c_grid_desc_g_m_n_, M01, N01);
+                    Block2CTileMapMaker{BatchCount}.MakeBlock2CTileMap(c_grid_desc_m_n_, M01, N01);
             }
         }
 
@@ -256,12 +376,13 @@ struct DeviceBatchedGemmXdl
         const ADataType* p_a_grid_;
         const BDataType* p_b_grid_;
         CDataType* p_c_grid_;
-        AGridDesc_G_K0_M_K1 a_grid_desc_g_k0_m_k1_;
-        BGridDesc_G_K0_N_K1 b_grid_desc_g_k0_n_k1_;
-        CGridDesc_G_M_N c_grid_desc_g_m_n_;
-        typename GridwiseBatchedGemm::CGridDesc_G_M0_N0_M1_N1_M2_M3_M4_N2
-            c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2_;
-        typename GridwiseBatchedGemm::DefaultBlock2CTileMap block_2_ctile_map_;
+        index_t BatchCount_;
+        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
+        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
+        CGridDesc_M_N c_grid_desc_m_n_;
+        CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_;
+        ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch_;
+        Block2CTileMap block_2_ctile_map_;
         index_t M01_;
         index_t N01_;
         AElementwiseOperation a_element_op_;
@@ -277,57 +398,51 @@ struct DeviceBatchedGemmXdl
         float Run(const Argument& arg, int nrepeat = 1)
         {
             {
-                std::cout << "arg.a_grid_desc_g_k0_m_k1_{"
-                          << arg.a_grid_desc_g_k0_m_k1_.GetLength(I0) << ", "
-                          << arg.a_grid_desc_g_k0_m_k1_.GetLength(I1) << ", "
-                          << arg.a_grid_desc_g_k0_m_k1_.GetLength(I2) << ", "
-                          << arg.a_grid_desc_g_k0_m_k1_.GetLength(I3) << "}" << std::endl;
-
-                std::cout << "arg.b_grid_desc_g_k0_n_k1_{"
-                          << arg.b_grid_desc_g_k0_n_k1_.GetLength(I0) << ", "
-                          << arg.b_grid_desc_g_k0_n_k1_.GetLength(I1) << ", "
-                          << arg.b_grid_desc_g_k0_n_k1_.GetLength(I2) << ", "
-                          << arg.b_grid_desc_g_k0_n_k1_.GetLength(I3) << "}" << std::endl;
-
-                std::cout << "arg.c_grid_desc_g_m_n_{" << arg.c_grid_desc_g_m_n_.GetLength(I0)
-                          << ", " << arg.c_grid_desc_g_m_n_.GetLength(I1) << ", "
-                          << arg.c_grid_desc_g_m_n_.GetLength(I2) << "}" << std::endl;
+                std::cout << "arg.a_grid_desc_k0_m_k1_{" << arg.a_grid_desc_k0_m_k1_.GetLength(I0)
+                          << ", " << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.b_grid_desc_k0_n_k1_{" << arg.b_grid_desc_k0_n_k1_.GetLength(I0)
+                          << ", " << arg.b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.c_grid_desc_m_n_{" << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
+                          << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
             }
 
-            if(!GridwiseBatchedGemm::CheckValidity(arg.a_grid_desc_g_k0_m_k1_,
-                                                   arg.b_grid_desc_g_k0_n_k1_,
-                                                   arg.c_grid_desc_g_m_n_,
-                                                   arg.M01_,
-                                                   arg.N01_))
+            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
+                                            arg.b_grid_desc_k0_n_k1_,
+                                            arg.c_grid_desc_m_n_,
+                                            arg.M01_,
+                                            arg.N01_))
             {
                 throw std::runtime_error(
                     "wrong! GridwiseBatchedGemm_km_kn_m0m1n0n1_xdlops_v2r3 has invalid setting");
             }
 
             const index_t grid_size =
-                GridwiseBatchedGemm::CalculateGridSize(arg.c_grid_desc_g_m_n_);
+                GridwiseGemm::CalculateGridSize(arg.c_grid_desc_m_n_) * arg.BatchCount_;
 
-            const auto K0 = arg.a_grid_desc_g_k0_m_k1_.GetLength(I1);
+            const auto K0 = arg.a_grid_desc_k0_m_k1_.GetLength(I0);
 
-            const bool has_main_k0_block_loop =
-                GridwiseBatchedGemm::CalculateHasMainK0BlockLoop(K0);
+            const bool has_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
 
             float ave_time = 0;
 
             if(has_main_k0_block_loop)
             {
                 const auto kernel = kernel_batched_gemm_xdlops_v2r3<
-                    GridwiseBatchedGemm,
+                    GridwiseGemm,
                     ADataType, // TODO: distiguish A/B datatype
                     CDataType,
-                    remove_reference_t<DeviceBatchedGemmXdl::AGridDesc_G_K0_M_K1>,
-                    remove_reference_t<DeviceBatchedGemmXdl::BGridDesc_G_K0_N_K1>,
-                    remove_reference_t<
-                        typename GridwiseBatchedGemm::CGridDesc_G_M0_N0_M1_N1_M2_M3_M4_N2>,
+                    remove_reference_t<DeviceBatchedGemmXdl::AGridDesc_K0_M_K1>,
+                    remove_reference_t<DeviceBatchedGemmXdl::BGridDesc_K0_N_K1>,
+                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
                     AElementwiseOperation,
                     BElementwiseOperation,
                     CElementwiseOperation,
-                    remove_reference_t<typename GridwiseBatchedGemm::DefaultBlock2CTileMap>,
+                    ComputeBasePtrOfStridedBatch,
+                    remove_reference_t<Block2CTileMap>,
                     true>;
 
                 ave_time = launch_and_time_kernel(kernel,
@@ -338,28 +453,30 @@ struct DeviceBatchedGemmXdl
                                                   arg.p_a_grid_,
                                                   arg.p_b_grid_,
                                                   arg.p_c_grid_,
-                                                  arg.a_grid_desc_g_k0_m_k1_,
-                                                  arg.b_grid_desc_g_k0_n_k1_,
-                                                  arg.c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2_,
+                                                  arg.BatchCount_,
+                                                  arg.a_grid_desc_k0_m_k1_,
+                                                  arg.b_grid_desc_k0_n_k1_,
+                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
                                                   arg.a_element_op_,
                                                   arg.b_element_op_,
                                                   arg.c_element_op_,
+                                                  arg.compute_base_ptr_of_batch_,
                                                   arg.block_2_ctile_map_);
             }
             else
             {
                 const auto kernel = kernel_batched_gemm_xdlops_v2r3<
-                    GridwiseBatchedGemm,
+                    GridwiseGemm,
                     ADataType, // TODO: distiguish A/B datatype
                     CDataType,
-                    remove_reference_t<DeviceBatchedGemmXdl::AGridDesc_G_K0_M_K1>,
-                    remove_reference_t<DeviceBatchedGemmXdl::BGridDesc_G_K0_N_K1>,
-                    remove_reference_t<
-                        typename GridwiseBatchedGemm::CGridDesc_G_M0_N0_M1_N1_M2_M3_M4_N2>,
+                    remove_reference_t<DeviceBatchedGemmXdl::AGridDesc_K0_M_K1>,
+                    remove_reference_t<DeviceBatchedGemmXdl::BGridDesc_K0_N_K1>,
+                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
                     AElementwiseOperation,
                     BElementwiseOperation,
                     CElementwiseOperation,
-                    remove_reference_t<typename GridwiseBatchedGemm::DefaultBlock2CTileMap>,
+                    ComputeBasePtrOfStridedBatch,
+                    remove_reference_t<Block2CTileMap>,
                     false>;
 
                 ave_time = launch_and_time_kernel(kernel,
@@ -370,12 +487,14 @@ struct DeviceBatchedGemmXdl
                                                   arg.p_a_grid_,
                                                   arg.p_b_grid_,
                                                   arg.p_c_grid_,
-                                                  arg.a_grid_desc_g_k0_m_k1_,
-                                                  arg.b_grid_desc_g_k0_n_k1_,
-                                                  arg.c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2_,
+                                                  arg.BatchCount_,
+                                                  arg.a_grid_desc_k0_m_k1_,
+                                                  arg.b_grid_desc_k0_n_k1_,
+                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
                                                   arg.a_element_op_,
                                                   arg.b_element_op_,
                                                   arg.c_element_op_,
+                                                  arg.compute_base_ptr_of_batch_,
                                                   arg.block_2_ctile_map_);
             }
 
@@ -397,11 +516,11 @@ struct DeviceBatchedGemmXdl
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        return GridwiseBatchedGemm::CheckValidity(arg.a_grid_desc_g_k0_m_k1_,
-                                                  arg.b_grid_desc_g_k0_n_k1_,
-                                                  arg.c_grid_desc_g_m_n_,
-                                                  arg.M01_,
-                                                  arg.N01_);
+        return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
+                                           arg.b_grid_desc_k0_n_k1_,
+                                           arg.c_grid_desc_m_n_,
+                                           arg.M01_,
+                                           arg.N01_);
     }
 
     // polymorphic

include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_xdlops_v2r3.hpp

-#ifndef CK_GRIDWISE_BATCHED_GEMM_XDLOPS_V2R3_HPP
-#define CK_GRIDWISE_BATCHED_GEMM_XDLOPS_V2R3_HPP
-
-#include "common_header.hpp"
-#include "multi_index_transform_helper.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-#include "blockwise_gemm_xdlops.hpp"
-#include "blockwise_tensor_slice_transfer_v4r1.hpp"
-#include "threadwise_tensor_slice_transfer.hpp"
-
-namespace ck {
-
-template <typename GridwiseBatchedGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_G_K0_M_K1,
-          typename BGridDesc_G_K0_N_K1,
-          typename CGridDesc_G_M0_N0_M1_N1_M2_M3_M4_N2,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation,
-          typename Block2CTileMap,
-          bool HasMainKBlockLoop>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_batched_gemm_xdlops_v2r3(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            FloatC* __restrict__ p_c_grid,
-            const AGridDesc_G_K0_M_K1 a_grid_desc_g_k0_m_k1,
-            const BGridDesc_G_K0_N_K1 b_grid_desc_g_k0_n_k1,
-            const CGridDesc_G_M0_N0_M1_N1_M2_M3_M4_N2 c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2,
-            const AElementwiseOperation a_element_op,
-            const BElementwiseOperation b_element_op,
-            const CElementwiseOperation c_element_op,
-            const Block2CTileMap block_2_ctile_map)
-{
-    __shared__ char p_shared[GridwiseBatchedGemm::GetSharedMemoryNumberOfByte()];
-
-    GridwiseBatchedGemm::template Run<HasMainKBlockLoop>(p_a_grid,
-                                                         p_b_grid,
-                                                         p_c_grid,
-                                                         p_shared,
-                                                         a_grid_desc_g_k0_m_k1,
-                                                         b_grid_desc_g_k0_n_k1,
-                                                         c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2,
-                                                         a_element_op,
-                                                         b_element_op,
-                                                         c_element_op,
-                                                         block_2_ctile_map);
-}
-
-template <index_t BlockSize,
-          typename FloatAB,
-          typename FloatAcc,
-          typename FloatC,
-          InMemoryDataOperationEnum_t CGlobalMemoryDataOperation,
-          typename AGridDesc_G_K0_M_K1,
-          typename BGridDesc_G_K0_N_K1,
-          typename CGridDesc_G_M_N,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation,
-          index_t MPerBlock,
-          index_t NPerBlock,
-          index_t K0PerBlock,
-          index_t MPerXDL,
-          index_t NPerXDL,
-          index_t K1Value,
-          index_t MXdlPerWave,
-          index_t NXdlPerWave,
-          typename ABlockTransferThreadClusterLengths_G_K0_M_K1,
-          typename ABlockTransferThreadClusterArrangeOrder,
-          typename ABlockTransferSrcAccessOrder,
-          index_t ABlockTransferSrcVectorDim,
-          index_t ABlockTransferSrcScalarPerVector,
-          index_t ABlockTransferDstScalarPerVector_K1,
-          bool AThreadTransferSrcResetCoordinateAfterRun,
-          bool ABlockLdsExtraM,
-          typename BBlockTransferThreadClusterLengths_G_K0_N_K1,
-          typename BBlockTransferThreadClusterArrangeOrder,
-          typename BBlockTransferSrcAccessOrder,
-          index_t BBlockTransferSrcVectorDim,
-          index_t BBlockTransferSrcScalarPerVector,
-          index_t BBlockTransferDstScalarPerVector_K1,
-          bool BThreadTransferSrcResetCoordinateAfterRun,
-          bool BBlockLdsExtraN,
-          typename CThreadTransferSrcDstAccessOrder,
-          index_t CThreadTransferSrcDstVectorDim,
-          index_t CThreadTransferDstScalarPerVector>
-struct GridwiseBatchedGemm_gk0mk1_gk0nk1_gmn_xdlops_v2r3
-{
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
-    static constexpr auto I4 = Number<4>{};
-    static constexpr auto I5 = Number<5>{};
-    static constexpr auto I6 = Number<6>{};
-    static constexpr auto I7 = Number<7>{};
-    static constexpr auto I8 = Number<8>{};
-
-    // K1 should be Number<...>
-    static constexpr auto K1 = Number<K1Value>{};
-
-    __host__ __device__ static constexpr auto
-    GetABlockDescriptor_BatchCount_K0PerBlock_MPerBlock_K1()
-    {
-        constexpr auto max_lds_align = K1;
-
-        // A matrix in LDS memory, dst of blockwise copy
-        constexpr auto a_block_desc_g_k0_m_k1 = [&]() {
-            if constexpr(ABlockLdsExtraM)
-            {
-                return make_naive_tensor_descriptor(
-                    make_tuple(I1, Number<K0PerBlock>{}, Number<MPerBlock>{}, K1),
-                    make_tuple(Number<K0PerBlock>{} * Number<MPerBlock + 1>{} * K1,
-                               Number<MPerBlock + 1>{} * K1,
-                               K1,
-                               I1));
-            }
-            else
-            {
-                return make_naive_tensor_descriptor_aligned(
-                    make_tuple(I1, Number<K0PerBlock>{}, Number<MPerBlock>{}, K1), max_lds_align);
-            }
-        }();
-
-        return a_block_desc_g_k0_m_k1;
-    }
-
-    __host__ __device__ static constexpr auto
-    GetBBlockDescriptor_BatchCount_K0PerBlock_NPerBlock_K1()
-    {
-        constexpr auto max_lds_align = K1;
-
-        // B matrix in LDS memory, dst of blockwise copy
-        constexpr auto b_block_desc_g_k0_n_k1 = [&]() {
-            if constexpr(BBlockLdsExtraN)
-            {
-                return make_naive_tensor_descriptor(
-                    make_tuple(I1, Number<K0PerBlock>{}, Number<NPerBlock>{}, K1),
-                    make_tuple(Number<K0PerBlock>{} * Number<NPerBlock + 1>{} * K1,
-                               Number<NPerBlock + 1>{} * K1,
-                               K1,
-                               I1));
-            }
-            else
-            {
-                return make_naive_tensor_descriptor_aligned(
-                    make_tuple(I1, Number<K0PerBlock>{}, Number<NPerBlock>{}, K1), max_lds_align);
-            }
-        }();
-
-        return b_block_desc_g_k0_n_k1;
-    }
-
-    __host__ __device__ static constexpr auto GetABlockDescriptor_K0PerBlock_MPerBlock_K1()
-    {
-        constexpr auto a_block_desc_g_k0_m_k1 =
-            GetABlockDescriptor_BatchCount_K0PerBlock_MPerBlock_K1();
-
-        constexpr auto K0 = a_block_desc_g_k0_m_k1.GetLength(I1);
-        constexpr auto M  = a_block_desc_g_k0_m_k1.GetLength(I2);
-
-        constexpr auto a_block_desc_k0_m_k1 = transform_tensor_descriptor(
-            a_block_desc_g_k0_m_k1,
-            make_tuple(make_freeze_transform(I0),
-                       make_pass_through_transform(K0),
-                       make_pass_through_transform(M),
-                       make_pass_through_transform(K1)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<>{}, Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-
-        return a_block_desc_k0_m_k1;
-    }
-
-    __host__ __device__ static constexpr auto GetBBlockDescriptor_K0PerBlock_NPerBlock_K1()
-    {
-        constexpr auto b_block_desc_g_k0_n_k1 =
-            GetBBlockDescriptor_BatchCount_K0PerBlock_NPerBlock_K1();
-
-        constexpr auto K0 = b_block_desc_g_k0_n_k1.GetLength(I1);
-        constexpr auto N  = b_block_desc_g_k0_n_k1.GetLength(I2);
-
-        constexpr auto b_block_desc_k0_n_k1 = transform_tensor_descriptor(
-            b_block_desc_g_k0_n_k1,
-            make_tuple(make_freeze_transform(I0),
-                       make_pass_through_transform(K0),
-                       make_pass_through_transform(N),
-                       make_pass_through_transform(K1)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<>{}, Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-
-        return b_block_desc_k0_n_k1;
-    }
-
-    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
-    {
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_desc_g_k0_m_k1 =
-            GetABlockDescriptor_BatchCount_K0PerBlock_MPerBlock_K1();
-
-        constexpr auto b_block_desc_g_k0_n_k1 =
-            GetBBlockDescriptor_BatchCount_K0PerBlock_NPerBlock_K1();
-
-        constexpr auto max_lds_align = K1;
-
-        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
-            a_block_desc_g_k0_m_k1.GetElementSpaceSize(), max_lds_align);
-
-        constexpr auto b_block_space_size_aligned = math::integer_least_multiple(
-            b_block_desc_g_k0_n_k1.GetElementSpaceSize(), max_lds_align);
-
-        return (a_block_space_size_aligned + b_block_space_size_aligned) * sizeof(FloatAB);
-    }
-
-    // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
-    __host__ __device__ static constexpr bool
-    CheckValidity(const AGridDesc_G_K0_M_K1& a_grid_desc_g_k0_m_k1,
-                  const BGridDesc_G_K0_N_K1& b_grid_desc_g_k0_n_k1,
-                  const CGridDesc_G_M_N& c_grid_desc_g_m_n,
-                  index_t M01,
-                  index_t N01)
-    {
-        static_assert(is_known_at_compile_time<remove_cv_t<decltype(K1)>>::value,
-                      "wrong! K1 need to be known at compile-time");
-
-        static_assert((MPerBlock % (MPerXDL * MXdlPerWave) == 0) &&
-                          (NPerBlock % (NXdlPerWave * NPerXDL)) == 0,
-                      "Invalid tuning param!");
-
-        // const auto G  = a_grid_desc_g_k0_m_k1.GetLength(I0);
-        const auto K0 = a_grid_desc_g_k0_m_k1.GetLength(I1);
-        const auto M  = a_grid_desc_g_k0_m_k1.GetLength(I2);
-        const auto N  = b_grid_desc_g_k0_n_k1.GetLength(I2);
-
-        if(!(M == c_grid_desc_g_m_n.GetLength(I1) && N == c_grid_desc_g_m_n.GetLength(I2) &&
-             K0 == b_grid_desc_g_k0_n_k1.GetLength(I1) &&
-             K1 == a_grid_desc_g_k0_m_k1.GetLength(I3) &&
-             K1 == b_grid_desc_g_k0_n_k1.GetLength(I3)))
-            return false;
-
-        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K0 % K0PerBlock == 0))
-            return false;
-
-        // check M01, N01
-        constexpr auto M1 = Number<MPerBlock>{};
-        constexpr auto N1 = Number<NPerBlock>{};
-
-        const auto M0 = M / M1;
-        const auto N0 = N / N1;
-
-        if(!(M0 % M01 == 0 && N0 % N01 == 0))
-            return false;
-
-        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
-        return true;
-    }
-
-    __host__ __device__ static constexpr index_t
-    CalculateGridSize(const CGridDesc_G_M_N& c_grid_desc_g_m_n)
-    {
-        const auto G = c_grid_desc_g_m_n.GetLength(I0);
-        const auto M = c_grid_desc_g_m_n.GetLength(I1);
-        const auto N = c_grid_desc_g_m_n.GetLength(I2);
-
-        const index_t grid_size = G * (M / MPerBlock) * (N / NPerBlock);
-
-        return grid_size;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasMainK0BlockLoop(index_t K0)
-    {
-        const bool has_main_k0_block_loop = (K0 / K0PerBlock) > 1;
-
-        return has_main_k0_block_loop;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeCGridDescriptor_G_M0_N0_M1_N1_M2_M3_M4_N2(const CGridDesc_G_M_N& c_grid_desc_g_m_n)
-    {
-        constexpr auto max_lds_align = K1;
-
-        // A matrix in LDS memory, dst of blockwise copy
-        constexpr auto a_block_desc_k0_m_k1 = [&]() {
-            if constexpr(ABlockLdsExtraM)
-            {
-                return make_naive_tensor_descriptor(
-                    make_tuple(Number<K0PerBlock>{}, Number<MPerBlock>{}, K1),
-                    make_tuple(Number<MPerBlock + 1>{} * K1, K1, I1));
-            }
-            else
-            {
-                return make_naive_tensor_descriptor_aligned(
-                    make_tuple(Number<K0PerBlock>{}, Number<MPerBlock>{}, K1), max_lds_align);
-            }
-        }();
-
-        // B matrix in LDS memory, dst of blockwise copy
-        constexpr auto b_block_desc_k0_n_k1 = [&]() {
-            if constexpr(BBlockLdsExtraN)
-            {
-                return make_naive_tensor_descriptor(
-                    make_tuple(Number<K0PerBlock>{}, Number<NPerBlock>{}, K1),
-                    make_tuple(Number<NPerBlock + 1>{} * K1, K1, I1));
-            }
-            else
-            {
-                return make_naive_tensor_descriptor_aligned(
-                    make_tuple(Number<K0PerBlock>{}, Number<NPerBlock>{}, K1), max_lds_align);
-            }
-        }();
-
-        using BlockwiseGemm =
-            BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                FloatAB,
-                                                                FloatAcc,
-                                                                decltype(a_block_desc_k0_m_k1),
-                                                                decltype(b_block_desc_k0_n_k1),
-                                                                MPerXDL,
-                                                                NPerXDL,
-                                                                MXdlPerWave,
-                                                                NXdlPerWave,
-                                                                K1>;
-
-        return BlockwiseGemm::MakeCGridDescriptor_G_M0_N0_M1_N1_M2_M3_M4_N2(c_grid_desc_g_m_n);
-    }
-
-    // return block_id to C matrix tile idx (m0, n0) mapping
-    __host__ __device__ static constexpr auto
-    MakeDefaultBlock2CTileMap(const CGridDesc_G_M_N& c_grid_desc_g_m_n, index_t M01, index_t N01)
-    {
-        const auto G = c_grid_desc_g_m_n.GetLength(I0);
-        const auto M = c_grid_desc_g_m_n.GetLength(I1);
-        const auto N = c_grid_desc_g_m_n.GetLength(I2);
-
-        constexpr auto M1 = Number<MPerBlock>{};
-        constexpr auto N1 = Number<NPerBlock>{};
-
-        const auto M0 = M / M1;
-        const auto N0 = N / N1;
-
-        const auto M00 = M0 / M01;
-        const auto N00 = N0 / N01;
-
-        const auto g_m00_m01_n00_n01_to_m0_n0_block_cluster_adaptor =
-            make_single_stage_tensor_adaptor(
-                make_tuple(make_pass_through_transform(G),
-                           make_unmerge_transform(make_tuple(M00, M01)),
-                           make_unmerge_transform(make_tuple(N00, N01))),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
-                make_tuple(Sequence<0>{}, Sequence<1, 3>{}, Sequence<2, 4>{}));
-
-        const auto cblockid_to_g_m00_m01_n00_n01_block_cluster_adaptor =
-            make_single_stage_tensor_adaptor(
-                make_tuple(make_merge_transform(make_tuple(G, M00, N00, M01, N01))),
-                make_tuple(Sequence<0, 1, 2, 3, 4>{}),
-                make_tuple(Sequence<0>{}));
-
-        const auto cblockid_to_g_m0_n0_block_cluster_adaptor =
-            chain_tensor_adaptors(g_m00_m01_n00_n01_to_m0_n0_block_cluster_adaptor,
-                                  cblockid_to_g_m00_m01_n00_n01_block_cluster_adaptor);
-
-        return cblockid_to_g_m0_n0_block_cluster_adaptor;
-    }
-
-    using CGridDesc_G_M0_N0_M1_N1_M2_M3_M4_N2 =
-        decltype(MakeCGridDescriptor_G_M0_N0_M1_N1_M2_M3_M4_N2(CGridDesc_G_M_N{}));
-    using DefaultBlock2CTileMap = decltype(MakeDefaultBlock2CTileMap(CGridDesc_G_M_N{}, 1, 1));
-
-    template <bool HasMainKBlockLoop, typename Block2CTileMap = DefaultBlock2CTileMap>
-    __device__ static void
-    Run(const FloatAB* __restrict__ p_a_grid,
-        const FloatAB* __restrict__ p_b_grid,
-        FloatC* __restrict__ p_c_grid,
-        void* __restrict__ p_shared,
-        const AGridDesc_G_K0_M_K1& a_grid_desc_g_k0_m_k1,
-        const BGridDesc_G_K0_N_K1& b_grid_desc_g_k0_n_k1,
-        const CGridDesc_G_M0_N0_M1_N1_M2_M3_M4_N2& c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2,
-        const AElementwiseOperation& a_element_op,
-        const BElementwiseOperation& b_element_op,
-        const CElementwiseOperation& c_element_op,
-        const Block2CTileMap& block_2_ctile_map)
-    {
-        const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum_t::Global>(
-            p_a_grid, a_grid_desc_g_k0_m_k1.GetElementSpaceSize());
-        const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum_t::Global>(
-            p_b_grid, b_grid_desc_g_k0_n_k1.GetElementSpaceSize());
-        auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum_t::Global>(
-            p_c_grid, c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2.GetElementSpaceSize());
-
-        const auto K0 = a_grid_desc_g_k0_m_k1.GetLength(I1);
-
-        // divide block work by [M, N]
-        const auto block_work_idx =
-            block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
-
-        // HACK: this force m/n_block_data_idx_on_grid into SGPR
-        const index_t g_idx_on_grid = __builtin_amdgcn_readfirstlane(block_work_idx[I0]);
-
-        const index_t m_block_data_idx_on_grid =
-            __builtin_amdgcn_readfirstlane(block_work_idx[I1] * MPerBlock);
-
-        const index_t n_block_data_idx_on_grid =
-            __builtin_amdgcn_readfirstlane(block_work_idx[I2] * NPerBlock);
-
-        // lds max alignment
-        constexpr auto max_lds_align = K1;
-
-        // A matrix in LDS memory, dst of blockwise copy
-        constexpr auto a_block_desc_g_k0_m_k1 =
-            GetABlockDescriptor_BatchCount_K0PerBlock_MPerBlock_K1();
-
-        // B matrix in LDS memory, dst of blockwise copy
-        constexpr auto b_block_desc_g_k0_n_k1 =
-            GetBBlockDescriptor_BatchCount_K0PerBlock_NPerBlock_K1();
-
-        // A matrix blockwise copy
-        auto a_blockwise_copy =
-            BlockwiseTensorSliceTransfer_v4r1<BlockSize,
-                                              AElementwiseOperation,
-                                              ck::tensor_operation::element_wise::PassThrough,
-                                              InMemoryDataOperationEnum_t::Set,
-                                              Sequence<1, K0PerBlock, MPerBlock, K1>,
-                                              ABlockTransferThreadClusterLengths_G_K0_M_K1,
-                                              ABlockTransferThreadClusterArrangeOrder,
-                                              FloatAB,
-                                              FloatAB,
-                                              decltype(a_grid_desc_g_k0_m_k1),
-                                              decltype(a_block_desc_g_k0_m_k1),
-                                              ABlockTransferSrcAccessOrder,
-                                              Sequence<0, 2, 1, 3>,
-                                              ABlockTransferSrcVectorDim,
-                                              3,
-                                              ABlockTransferSrcScalarPerVector,
-                                              ABlockTransferDstScalarPerVector_K1,
-                                              1,
-                                              1,
-                                              AThreadTransferSrcResetCoordinateAfterRun,
-                                              true>(
-                a_grid_desc_g_k0_m_k1,
-                make_multi_index(g_idx_on_grid, 0, m_block_data_idx_on_grid, 0),
-                a_element_op,
-                a_block_desc_g_k0_m_k1,
-                make_multi_index(0, 0, 0, 0),
-                ck::tensor_operation::element_wise::PassThrough{});
-
-        // B matrix blockwise copy
-        auto b_blockwise_copy =
-            BlockwiseTensorSliceTransfer_v4r1<BlockSize,
-                                              BElementwiseOperation,
-                                              ck::tensor_operation::element_wise::PassThrough,
-                                              InMemoryDataOperationEnum_t::Set,
-                                              Sequence<1, K0PerBlock, NPerBlock, K1>,
-                                              BBlockTransferThreadClusterLengths_G_K0_N_K1,
-                                              BBlockTransferThreadClusterArrangeOrder,
-                                              FloatAB,
-                                              FloatAB,
-                                              decltype(b_grid_desc_g_k0_n_k1),
-                                              decltype(b_block_desc_g_k0_n_k1),
-                                              BBlockTransferSrcAccessOrder,
-                                              Sequence<0, 2, 1, 3>,
-                                              BBlockTransferSrcVectorDim,
-                                              3,
-                                              BBlockTransferSrcScalarPerVector,
-                                              BBlockTransferDstScalarPerVector_K1,
-                                              1,
-                                              1,
-                                              BThreadTransferSrcResetCoordinateAfterRun,
-                                              true>(
-                b_grid_desc_g_k0_n_k1,
-                make_multi_index(g_idx_on_grid, 0, n_block_data_idx_on_grid, 0),
-                b_element_op,
-                b_block_desc_g_k0_n_k1,
-                make_multi_index(0, 0, 0, 0),
-                ck::tensor_operation::element_wise::PassThrough{});
-
-        // GEMM definition
-        //   c_mtx += transpose(a_mtx) * b_mtx
-        //     a_mtx[K0PerBlock, MPerBlock] is in LDS
-        //     b_mtx[K0PerBlock, NPerBlock] is in LDS
-        //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
-        //       register
-        // sanity check
-
-        constexpr auto a_block_desc_k0_m_k1 = GetABlockDescriptor_K0PerBlock_MPerBlock_K1();
-        constexpr auto b_block_desc_k0_n_k1 = GetBBlockDescriptor_K0PerBlock_NPerBlock_K1();
-
-        auto blockwise_gemm =
-            BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                FloatAB,
-                                                                FloatAcc,
-                                                                decltype(a_block_desc_k0_m_k1),
-                                                                decltype(b_block_desc_k0_n_k1),
-                                                                MPerXDL,
-                                                                NPerXDL,
-                                                                MXdlPerWave,
-                                                                NXdlPerWave,
-                                                                K1>{};
-
-        auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
-            a_block_desc_g_k0_m_k1.GetElementSpaceSize(), max_lds_align);
-
-        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum_t::Lds>(
-            static_cast<FloatAB*>(p_shared), a_block_desc_g_k0_m_k1.GetElementSpaceSize());
-
-        auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum_t::Lds>(
-            static_cast<FloatAB*>(p_shared) + a_block_space_size_aligned,
-            b_block_desc_g_k0_n_k1.GetElementSpaceSize());
-
-        constexpr auto a_block_slice_copy_step = make_multi_index(0, K0PerBlock, 0, 0);
-        constexpr auto b_block_slice_copy_step = make_multi_index(0, K0PerBlock, 0, 0);
-
-        // preload data into LDS
-        {
-            a_blockwise_copy.RunRead(a_grid_desc_g_k0_m_k1, a_grid_buf);
-            b_blockwise_copy.RunRead(b_grid_desc_g_k0_n_k1, b_grid_buf);
-
-            a_blockwise_copy.RunWrite(a_block_desc_g_k0_m_k1, a_block_buf);
-            b_blockwise_copy.RunWrite(b_block_desc_g_k0_n_k1, b_block_buf);
-        }
-
-        // Initialize C
-        c_thread_buf.Clear();
-
-        // main body
-        if constexpr(HasMainKBlockLoop)
-        {
-            index_t k0_block_data_begin = 0;
-
-            do
-            {
-                a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc_g_k0_m_k1, a_block_slice_copy_step);
-                b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_g_k0_n_k1, b_block_slice_copy_step);
-
-                a_blockwise_copy.RunRead(a_grid_desc_g_k0_m_k1, a_grid_buf);
-
-                block_sync_lds();
-
-                b_blockwise_copy.RunRead(b_grid_desc_g_k0_n_k1, b_grid_buf);
-
-                blockwise_gemm.Run(a_block_buf, b_block_buf, c_thread_buf);
-
-                block_sync_lds();
-
-                a_blockwise_copy.RunWrite(a_block_desc_g_k0_m_k1, a_block_buf);
-                b_blockwise_copy.RunWrite(b_block_desc_g_k0_n_k1, b_block_buf);
-
-                k0_block_data_begin += K0PerBlock;
-            } while(k0_block_data_begin < (K0 - K0PerBlock));
-        }
-
-        // tail
-        {
-            block_sync_lds();
-
-            blockwise_gemm.Run(a_block_buf, b_block_buf, c_thread_buf);
-        }
-
-        // output: register to global memory
-        {
-            constexpr auto c_thread_desc_g_m0_n0_m1_n1_m2_m3_m4_n2 =
-                blockwise_gemm.GetCThreadDescriptor_G_M0_N0_M1_N1_M2_M3_M4_N2();
-
-            constexpr auto c_block_desc_g_m0_n0_m1_n1_m2_m3_m4_n2 =
-                blockwise_gemm.GetCBlockDescriptor_G_M0_N0_M1_N1_M2_M3_M4_N2();
-
-            // constexpr auto G  = c_block_desc_g_m0_n0_m1_n1_m2_m3_m4_n2.GetLength(I0);
-            constexpr auto M0 = c_block_desc_g_m0_n0_m1_n1_m2_m3_m4_n2.GetLength(I1);
-            constexpr auto N0 = c_block_desc_g_m0_n0_m1_n1_m2_m3_m4_n2.GetLength(I2);
-            constexpr auto M1 = c_block_desc_g_m0_n0_m1_n1_m2_m3_m4_n2.GetLength(I3);
-            constexpr auto N1 = c_block_desc_g_m0_n0_m1_n1_m2_m3_m4_n2.GetLength(I4);
-            constexpr auto M2 = c_block_desc_g_m0_n0_m1_n1_m2_m3_m4_n2.GetLength(I5);
-            constexpr auto M3 = c_block_desc_g_m0_n0_m1_n1_m2_m3_m4_n2.GetLength(I6);
-            constexpr auto M4 = c_block_desc_g_m0_n0_m1_n1_m2_m3_m4_n2.GetLength(I7);
-            constexpr auto N2 = c_block_desc_g_m0_n0_m1_n1_m2_m3_m4_n2.GetLength(I8);
-
-            // calculate origin of thread output tensor on global memory
-            //     blockwise GEMM c matrix starting index
-            const auto c_thread_mtx_on_block =
-                blockwise_gemm.CalculateCThreadOriginDataIndex(I0, I0, I0, I0);
-
-            const index_t m_thread_data_on_grid =
-                m_block_data_idx_on_grid + c_thread_mtx_on_block[I0];
-
-            const index_t n_thread_data_on_grid =
-                n_block_data_idx_on_grid + c_thread_mtx_on_block[I1];
-
-            const auto m_thread_data_on_grid_to_m0_m1_m2_m3_m4_adaptor =
-                make_single_stage_tensor_adaptor(
-                    make_tuple(make_merge_transform(make_tuple(M0, M1, M2, M3, M4))),
-                    make_tuple(Sequence<0, 1, 2, 3, 4>{}),
-                    make_tuple(Sequence<0>{}));
-
-            const auto m_thread_data_on_grid_idx =
-                m_thread_data_on_grid_to_m0_m1_m2_m3_m4_adaptor.CalculateBottomIndex(
-                    make_multi_index(m_thread_data_on_grid));
-
-            const auto n_thread_data_on_grid_to_n0_n1_n2_adaptor = make_single_stage_tensor_adaptor(
-                make_tuple(make_merge_transform(make_tuple(N0, N1, N2))),
-                make_tuple(Sequence<0, 1, 2>{}),
-                make_tuple(Sequence<0>{}));
-
-            const auto n_thread_data_on_grid_idx =
-                n_thread_data_on_grid_to_n0_n1_n2_adaptor.CalculateBottomIndex(
-                    make_multi_index(n_thread_data_on_grid));
-
-            auto c_thread_copy = ThreadwiseTensorSliceTransfer_v1r3<
-                FloatAcc,
-                FloatC,
-                decltype(c_thread_desc_g_m0_n0_m1_n1_m2_m3_m4_n2),
-                decltype(c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2),
-                CElementwiseOperation,
-                Sequence<I1, M0, N0, I1, I1, M2, I1, M4, I1>,
-                CThreadTransferSrcDstAccessOrder,
-                CThreadTransferSrcDstVectorDim,
-                CThreadTransferDstScalarPerVector,
-                CGlobalMemoryDataOperation,
-                1,
-                true>{c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2,
-                      make_multi_index(g_idx_on_grid,
-                                       m_thread_data_on_grid_idx[I0],
-                                       n_thread_data_on_grid_idx[I0],
-                                       m_thread_data_on_grid_idx[I1],
-                                       n_thread_data_on_grid_idx[I1],
-                                       m_thread_data_on_grid_idx[I2],
-                                       m_thread_data_on_grid_idx[I3],
-                                       m_thread_data_on_grid_idx[I4],
-                                       n_thread_data_on_grid_idx[I2]),
-                      c_element_op};
-
-            c_thread_copy.Run(c_thread_desc_g_m0_n0_m1_n1_m2_m3_m4_n2,
-                              make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                              c_thread_buf,
-                              c_grid_desc_g_m0_n0_m1_n1_m2_m3_m4_n2,
-                              c_grid_buf);
-        }
-    }
-};
-
-} // namespace ck
-#endif

library/include/ck/library/host_tensor/host_tensor.hpp

@@ -317,7 +317,7 @@ float bf16_to_f32_(ck::bhalf_t src_val);
 void bf16_to_f32_(const Tensor<ck::bhalf_t>& src, Tensor<float>& dst);
 
 template <typename T>
-void check_error(const Tensor<T>& ref, const Tensor<T>& result)
+float check_error(const Tensor<T>& ref, const Tensor<T>& result)
 {
     float error     = 0;
     float max_diff  = -1;
@@ -354,6 +354,7 @@ void check_error(const Tensor<T>& ref, const Tensor<T>& result)
 
     std::cout << "error: " << error << std::endl;
     std::cout << "max_diff: " << max_diff << ", " << ref_value << ", " << result_value << std::endl;
+    return max_diff;
 }
 
 template <typename T>

library/include/ck/library/host_tensor/host_tensor_generator.hpp

@@ -93,8 +93,8 @@ struct GeneratorTensor_2<int8_t>
 template <typename T>
 struct GeneratorTensor_3
 {
-    T min_value = 0;
-    T max_value = 1;
+    float min_value = 0;
+    float max_value = 1;
 
     template <typename... Is>
     T operator()(Is...)
@@ -122,22 +122,6 @@ struct GeneratorTensor_3<ck::bhalf_t>
     }
 };
 
-template <>
-struct GeneratorTensor_3<int8_t>
-{
-    float min_value = 0;
-    float max_value = 1;
-
-    template <typename... Is>
-    int8_t operator()(Is...)
-    {
-        int8_t min_tmp = static_cast<int8_t>(min_value);
-        int8_t max_tmp = static_cast<int8_t>(max_value);
-
-        return (std::rand() % (max_tmp - min_tmp)) + min_tmp;
-    }
-};
-
 struct GeneratorTensor_Checkboard
 {
     template <typename... Ts>

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

@@ -4,6 +4,14 @@ set(DEVICE_BATCHED_GEMM_INSTANCE_SOURCE
    device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp;
    device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp;
    device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp;
+   device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp;
+   device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp;
+   device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp;
+   device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp;
+   device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp;
+   device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp;
+   device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp;
+   device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp;
 )
 
 add_library(device_batched_gemm_instance SHARED ${DEVICE_BATCHED_GEMM_INSTANCE_SOURCE})

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp

@@ -21,23 +21,22 @@ using S = ck::Sequence<Is...>;
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 // Compilation parameters for a[k, m] * b[k, n] = c[m, n]
-using device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instances =
-    std::tuple<
-        // clang-format off
-        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
-        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
-        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
-        //##########|      |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              4,              8,      true,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              2,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              2,              8,      true,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              4,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              4,              8,      true,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              4,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              2,              8,      true,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              2,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              4,              8,      true,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              2,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              2,              8,      true,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              4,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              2,              8,      true,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              1,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              1,              8,      true,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              2,              8,      true,               8,               1>
-        // clang-format on
-        >;
+using device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        //##########|        AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+        //##########|         Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+        //##########|             |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+        //##########|             |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1>
+    // clang-format on
+    >;
 
 void add_device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instances(
     std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp

@@ -21,23 +21,22 @@ using S = ck::Sequence<Is...>;
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 // Compilation parameters for a[k, m] * b[n, k] = c[m, n]
-using device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instances =
-    std::tuple<
-        // clang-format off
-        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
-        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
-        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
-        //##########|      |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              4,              8,      true,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              2,              8,      true,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              4,              8,      true,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              2,              8,      true,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              4,              8,      true,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              2,              8,      true,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              2,              8,      true,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,              2,              1,              8,      true,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>
-        // clang-format on
-        >;
+using device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        //##########|        AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+        //##########|         Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+        //##########|             |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+        //##########|             |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>
+    // clang-format on
+    >;
 
 void add_device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instances(
     std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp

@@ -21,27 +21,31 @@ using S = ck::Sequence<Is...>;
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
-using device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances =
-    std::tuple<
-        // clang-format off
-        //####################| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
-        //####################|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
-        //####################|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
-        //####################|      |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              2,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              4,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              4,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              2,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              2,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              4,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              1,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              1,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 64, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              2,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    32,    32,     4,  8,   16,   16,    2,    1,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 32, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              1,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<1, 4, 16, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 16, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              2,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<1, 4, 16, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 16, 1>,     S<0, 1, 3, 2>,     S<0, 1, 3, 2>,             2,              1,              8,      true,               8,               1>
-        // clang-format on
-        >;
+using device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        //#################| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+        //#################|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+        //#################|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+        //#################|      |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1>
+    // clang-format on
+    >;
 
 void add_device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances(
     std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp

@@ -27,19 +27,19 @@ using device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances = std::tuple<
         //#################|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
         //#################|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
         //#################|      |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<1, 4, 16, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 16, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 64, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 32, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<1, 4, 16, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 16, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>,
-        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<1, 4, 16, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,              3,              8,              8,      true,     S<1, 4, 16, 1>,     S<0, 2, 1, 3>,     S<0, 2, 1, 3>,             3,              8,              8,      true,               8,               1>
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>
     // clang-format on
     >;
 

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp

+#include <stdlib.h>
+#include "config.hpp"
+#include "device_batched_gemm_xdl.hpp"
+#include "element_wise_operation.hpp"
+#include "device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace device_batched_gemm_instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        //##########|        AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+        //##########|         Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+        //##########|             |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+        //##########|             |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               7,               1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instances(
+    std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instances{});
+}
+
+} // namespace device_batched_gemm_instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp

+#include <stdlib.h>
+#include "config.hpp"
+#include "device_batched_gemm_xdl.hpp"
+#include "element_wise_operation.hpp"
+#include "device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace device_batched_gemm_instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        //##########|        AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+        //##########|         Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+        //##########|             |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+        //##########|             |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instances(
+    std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instances{});
+}
+
+} // namespace device_batched_gemm_instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp

+#include <stdlib.h>
+#include "config.hpp"
+#include "device_batched_gemm_xdl.hpp"
+#include "element_wise_operation.hpp"
+#include "device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace device_batched_gemm_instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        //##########|        AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+        //##########|         Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+        //##########|             |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+        //##########|             |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               7,               1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instances(
+    std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instances{});
+}
+
+} // namespace device_batched_gemm_instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp

+#include <stdlib.h>
+#include "config.hpp"
+#include "device_batched_gemm_xdl.hpp"
+#include "element_wise_operation.hpp"
+#include "device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace device_batched_gemm_instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        //##########|        AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+        //##########|         Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+        //##########|             |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+        //##########|             |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    64,     4,  4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    32,     4,  4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    32,   128,     4,  4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    32,     4,  4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  F32,   F32,   F32,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    32,    64,     4,  4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,      true,               7,               1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instances(
+    std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instances{});
+}
+
+} // namespace device_batched_gemm_instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp

+#include <stdlib.h>
+#include "config.hpp"
+#include "device_batched_gemm_xdl.hpp"
+#include "element_wise_operation.hpp"
+#include "device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace device_batched_gemm_instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using AData   = int8_t;
+using BData   = int8_t;
+using CData   = int8_t;
+using AccData = int32_t;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        //##########|          AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer|   ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer|  BBlockTransfer|  BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+        //##########|           Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|    ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|        DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|       SrcScalar|       DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+        //##########|               |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave|  Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|     PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |       PerVector|    PerVector_K1|          |                |       PerVector|
+        //##########|               |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                 |               |               |               |               |                 |          |                |               |               |              |                |                |          |                |                |
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    64,     4,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    32,     4,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,    32,   128,     4,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    32,     4,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    64,    32,    64,     4,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,               16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,               16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,               16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,               16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,               16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,               16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,               16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,               16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             1,              16,              16,      true,               7,               1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instances(
+    std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instances{});
+}
+
+} // namespace device_batched_gemm_instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp

+#include <stdlib.h>
+#include "config.hpp"
+#include "device_batched_gemm_xdl.hpp"
+#include "element_wise_operation.hpp"
+#include "device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace device_batched_gemm_instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using AData   = int8_t;
+using BData   = int8_t;
+using CData   = int8_t;
+using AccData = int32_t;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        //##########|          AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+        //##########|           Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+        //##########|               |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+        //##########|               |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    64,     4,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    32,     4,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    32,   128,     4,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    32,     4,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    32,    64,     4,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              16,              16,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,               16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,               16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,               16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,               16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,               16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,               16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,               16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,               16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instances(
+    std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instances{});
+}
+
+} // namespace device_batched_gemm_instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp

+#include <stdlib.h>
+#include "config.hpp"
+#include "device_batched_gemm_xdl.hpp"
+#include "element_wise_operation.hpp"
+#include "device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace device_batched_gemm_instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using AData   = int8_t;
+using BData   = int8_t;
+using CData   = int8_t;
+using AccData = int32_t;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        //##########|          AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer|  ABlockTransfer|  ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer|  BBlockTransfer|  BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+        //##########|           Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|       SrcScalar|       DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|       SrcScalar|       DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+        //##########|               |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |       PerVector|    PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |       PerVector|    PerVector_K1|          |                |       PerVector|
+        //##########|               |      |      |        |        |        |        |            |            |            |      |      |      |      |    |     |     |     |     |                |               |               |               |                |                |          |                |               |               |              |                |                |          |                |                |
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    64,     4,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    32,     4,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,    32,   128,     4,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    32,     4,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    64,    32,    64,     4,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,               16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,               16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,               16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,               16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,               16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,               16,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,               16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,               16,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              16,      true,               7,               1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instances(
+    std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instances{});
+}
+
+} // namespace device_batched_gemm_instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp

+#include <stdlib.h>
+#include "config.hpp"
+#include "device_batched_gemm_xdl.hpp"
+#include "element_wise_operation.hpp"
+#include "device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace device_batched_gemm_instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using AData   = int8_t;
+using BData   = int8_t;
+using CData   = int8_t;
+using AccData = int32_t;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        //##########|          AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer|  ABlockTransfer|  ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer|  BBlockTransfer|  BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+        //##########|           Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|       SrcScalar|       DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|       SrcScalar|       DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+        //##########|               |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |       PerVector|    PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |       PerVector|    PerVector_K1|          |                |       PerVector|
+        //##########|               |      |      |        |        |        |        |            |            |            |      |      |      |      |    |     |     |     |     |                |               |               |               |                |                |          |                |               |               |              |                |                |          |                |                |
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    64,     4,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,   128,    32,     4,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   128,    32,   128,     4,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    64,    32,     4,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>,
+        DeviceBatchedGemmXdl<  AData, BData, CData, AccData,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    32,    64,     4,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,      true,               7,               1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instances(
+    std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instances{});
+}
+
+} // namespace device_batched_gemm_instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

profiler/include/profile_batched_gemm_impl.hpp

@@ -15,6 +15,18 @@ void add_device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances(std::vector<D
 void add_device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances(std::vector<DeviceGemmNoOpPtr>&);
 void add_device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instances(std::vector<DeviceGemmNoOpPtr>&);
 void add_device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instances(std::vector<DeviceGemmNoOpPtr>&);
+void add_device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instances(std::vector<DeviceGemmNoOpPtr>&);
+void add_device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instances(std::vector<DeviceGemmNoOpPtr>&);
+void add_device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instances(std::vector<DeviceGemmNoOpPtr>&);
+void add_device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instances(std::vector<DeviceGemmNoOpPtr>&);
+void add_device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instances(
+    std::vector<DeviceGemmNoOpPtr>&);
+void add_device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instances(
+    std::vector<DeviceGemmNoOpPtr>&);
+void add_device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instances(
+    std::vector<DeviceGemmNoOpPtr>&);
+void add_device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instances(
+    std::vector<DeviceGemmNoOpPtr>&);
 
 } // namespace device_batched_gemm_instance
 } // namespace device
@@ -156,6 +168,70 @@ void profile_batched_gemm_impl(int do_verification,
                 add_device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instances(gemm_ptrs);
         }
     }
+    else if constexpr(is_same<ADataType, float>::value && is_same<BDataType, float>::value &&
+                      is_same<CDataType, float>::value)
+    {
+        if constexpr(is_same<ALayout, tensor_layout::gemm::RowMajor>::value &&
+                     is_same<BLayout, tensor_layout::gemm::RowMajor>::value &&
+                     is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
+        {
+            ck::tensor_operation::device::device_batched_gemm_instance::
+                add_device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instances(gemm_ptrs);
+        }
+        else if constexpr(is_same<ALayout, tensor_layout::gemm::RowMajor>::value &&
+                          is_same<BLayout, tensor_layout::gemm::ColumnMajor>::value &&
+                          is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
+        {
+            ck::tensor_operation::device::device_batched_gemm_instance::
+                add_device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instances(gemm_ptrs);
+        }
+        else if constexpr(is_same<ALayout, tensor_layout::gemm::ColumnMajor>::value &&
+                          is_same<BLayout, tensor_layout::gemm::RowMajor>::value &&
+                          is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
+        {
+            ck::tensor_operation::device::device_batched_gemm_instance::
+                add_device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instances(gemm_ptrs);
+        }
+        else if constexpr(is_same<ALayout, tensor_layout::gemm::ColumnMajor>::value &&
+                          is_same<BLayout, tensor_layout::gemm::ColumnMajor>::value &&
+                          is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
+        {
+            ck::tensor_operation::device::device_batched_gemm_instance::
+                add_device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instances(gemm_ptrs);
+        }
+    }
+    else if constexpr(is_same<ADataType, int8_t>::value && is_same<BDataType, int8_t>::value &&
+                      is_same<CDataType, int8_t>::value)
+    {
+        if constexpr(is_same<ALayout, tensor_layout::gemm::RowMajor>::value &&
+                     is_same<BLayout, tensor_layout::gemm::RowMajor>::value &&
+                     is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
+        {
+            ck::tensor_operation::device::device_batched_gemm_instance::
+                add_device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instances(gemm_ptrs);
+        }
+        else if constexpr(is_same<ALayout, tensor_layout::gemm::RowMajor>::value &&
+                          is_same<BLayout, tensor_layout::gemm::ColumnMajor>::value &&
+                          is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
+        {
+            ck::tensor_operation::device::device_batched_gemm_instance::
+                add_device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instances(gemm_ptrs);
+        }
+        else if constexpr(is_same<ALayout, tensor_layout::gemm::ColumnMajor>::value &&
+                          is_same<BLayout, tensor_layout::gemm::RowMajor>::value &&
+                          is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
+        {
+            ck::tensor_operation::device::device_batched_gemm_instance::
+                add_device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instances(gemm_ptrs);
+        }
+        else if constexpr(is_same<ALayout, tensor_layout::gemm::ColumnMajor>::value &&
+                          is_same<BLayout, tensor_layout::gemm::ColumnMajor>::value &&
+                          is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
+        {
+            ck::tensor_operation::device::device_batched_gemm_instance::
+                add_device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instances(gemm_ptrs);
+        }
+    }
 
     if(gemm_ptrs.size() <= 0)
     {

profiler/src/profile_batched_gemm.cpp

+#include <cstdint>
 #include <iostream>
 #include <numeric>
 #include <initializer_list>
@@ -29,8 +30,9 @@ enum GemmMatrixLayout
 
 enum GemmDataType
 {
-    F32_F32_F32, // 0
-    F16_F16_F16, // 1
+    F32_F32_F32,    // 0
+    F16_F16_F16,    // 1
+    Int8_Int8_Int8, // 2
 };
 
 int profile_batched_gemm(int argc, char* argv[])
@@ -38,7 +40,7 @@ int profile_batched_gemm(int argc, char* argv[])
     if(!(argc == 15))
     {
         printf("arg1: tensor operation (batched_gemm: Batched GEMM)\n");
-        printf("arg2: data type (0: fp32; 1: fp16)\n");
+        printf("arg2: data type (0: fp32; 1: fp16, 2: int8)\n");
         printf("arg3: matrix layout (0: A[g, m, k] * B[g, k, n] = C[g, m, n];\n");
         printf("                     1: A[g, m, k] * B[g, n, k] = C[g, m, n];\n");
         printf("                     2: A[g, k, m] * B[g, k, n] = C[g, m, n];\n");
@@ -146,6 +148,163 @@ int profile_batched_gemm(int argc, char* argv[])
             (StrideB < 0) ? K : StrideB,
             (StrideC < 0) ? N : StrideC);
     }
+    else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        ck::profiler::profile_batched_gemm_impl<float,
+                                                float,
+                                                float,
+                                                ck::tensor_layout::gemm::RowMajor,
+                                                ck::tensor_layout::gemm::RowMajor,
+                                                ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            nrepeat,
+            M,
+            N,
+            K,
+            (StrideA < 0) ? K : StrideA,
+            (StrideB < 0) ? N : StrideB,
+            (StrideC < 0) ? N : StrideC,
+            BatchCount);
+    }
+    else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        ck::profiler::profile_batched_gemm_impl<float,
+                                                float,
+                                                float,
+                                                ck::tensor_layout::gemm::RowMajor,
+                                                ck::tensor_layout::gemm::ColumnMajor,
+                                                ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            nrepeat,
+            M,
+            N,
+            K,
+            (StrideA < 0) ? K : StrideA,
+            (StrideB < 0) ? K : StrideB,
+            (StrideC < 0) ? N : StrideC,
+            BatchCount);
+    }
+    else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        ck::profiler::profile_batched_gemm_impl<float,
+                                                float,
+                                                float,
+                                                ck::tensor_layout::gemm::ColumnMajor,
+                                                ck::tensor_layout::gemm::RowMajor,
+                                                ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            nrepeat,
+            M,
+            N,
+            K,
+            (StrideA < 0) ? M : StrideA,
+            (StrideB < 0) ? N : StrideB,
+            (StrideC < 0) ? N : StrideC);
+    }
+    else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        ck::profiler::profile_batched_gemm_impl<float,
+                                                float,
+                                                float,
+                                                ck::tensor_layout::gemm::ColumnMajor,
+                                                ck::tensor_layout::gemm::ColumnMajor,
+                                                ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            nrepeat,
+            M,
+            N,
+            K,
+            (StrideA < 0) ? M : StrideA,
+            (StrideB < 0) ? K : StrideB,
+            (StrideC < 0) ? N : StrideC);
+    }
+    else if(data_type == GemmDataType::Int8_Int8_Int8 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        ck::profiler::profile_batched_gemm_impl<int8_t,
+                                                int8_t,
+                                                int8_t,
+                                                ck::tensor_layout::gemm::RowMajor,
+                                                ck::tensor_layout::gemm::RowMajor,
+                                                ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            nrepeat,
+            M,
+            N,
+            K,
+            (StrideA < 0) ? K : StrideA,
+            (StrideB < 0) ? N : StrideB,
+            (StrideC < 0) ? N : StrideC,
+            BatchCount);
+    }
+    else if(data_type == GemmDataType::Int8_Int8_Int8 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        ck::profiler::profile_batched_gemm_impl<int8_t,
+                                                int8_t,
+                                                int8_t,
+                                                ck::tensor_layout::gemm::RowMajor,
+                                                ck::tensor_layout::gemm::ColumnMajor,
+                                                ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            nrepeat,
+            M,
+            N,
+            K,
+            (StrideA < 0) ? K : StrideA,
+            (StrideB < 0) ? K : StrideB,
+            (StrideC < 0) ? N : StrideC,
+            BatchCount);
+    }
+    else if(data_type == GemmDataType::Int8_Int8_Int8 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        ck::profiler::profile_batched_gemm_impl<int8_t,
+                                                int8_t,
+                                                int8_t,
+                                                ck::tensor_layout::gemm::ColumnMajor,
+                                                ck::tensor_layout::gemm::RowMajor,
+                                                ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            nrepeat,
+            M,
+            N,
+            K,
+            (StrideA < 0) ? M : StrideA,
+            (StrideB < 0) ? N : StrideB,
+            (StrideC < 0) ? N : StrideC);
+    }
+    else if(data_type == GemmDataType::Int8_Int8_Int8 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        ck::profiler::profile_batched_gemm_impl<int8_t,
+                                                int8_t,
+                                                int8_t,
+                                                ck::tensor_layout::gemm::ColumnMajor,
+                                                ck::tensor_layout::gemm::ColumnMajor,
+                                                ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            nrepeat,
+            M,
+            N,
+            K,
+            (StrideA < 0) ? M : StrideA,
+            (StrideB < 0) ? K : StrideB,
+            (StrideC < 0) ? N : StrideC,
+            BatchCount);
+    }
     else
     {
         throw std::runtime_error("wrong! this GEMM data_type & layout is not implemented");

test/CMakeLists.txt

@@ -39,3 +39,4 @@ add_subdirectory(gemm_split_k)
 add_subdirectory(conv2d_fwd)
 add_subdirectory(convnd_fwd)
 add_subdirectory(conv2d_bwd_data)
+add_subdirectory(batched_gemm)