Merge remote-tracking branch 'origin/develop' into stream-k-initial-impl

example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt

@@ -20,4 +20,5 @@ if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS M
 endif()
 if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
+  add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
 endif()

example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common_wmma.hpp"
+
+// kernel data types
+using InKernelDataType       = I8;
+using WeiKernelDataType      = I8;
+using AccDataType            = I32;
+using CShuffleDataType       = I8;
+using BiasKernelDataType     = I8;
+using ResidualKernelDataType = I8;
+using OutKernelDataType      = I8;
+
+// tensor data types
+using InUserDataType  = InKernelDataType;
+using WeiUserDataType = WeiKernelDataType;
+using OutUserDataType = OutKernelDataType;
+
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using OutElementOp = ck::tensor_operation::element_wise::AddReluAdd;
+
+#include "run_grouped_conv_fwd_bias_relu_add_wmma_example.inc"
+
+int main(int argc, char* argv[]) { return !run_grouped_conv_fwd_bias_relu_add_example(argc, argv); }

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
 #pragma once
+
 #include <iostream>
 #include <vector>
 

include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp

@@ -73,6 +73,11 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
     static constexpr auto I2 = Number<2>{};
     static constexpr auto I3 = Number<3>{};
 
+    // TODO: should be exposed as Tparams.
+    static constexpr index_t NumGemmKPrefetchStage = 1;
+    static constexpr LoopScheduler LoopSched       = make_default_loop_scheduler();
+    static constexpr PipelineVersion PipelineVer   = PipelineVersion::v2;
+
     using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
         BlockSize,
         ADataType, // TODO: distinguish A/B datatype
@@ -85,6 +90,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
         BElementwiseOperation,
         CElementwiseOperation,
         GemmSpec,
+        NumGemmKPrefetchStage,
         MPerBlock,
         NPerBlock,
         K0PerBlock,
@@ -112,7 +118,9 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
         CShuffleMRepeatPerShuffle,
         CShuffleNRepeatPerShuffle,
         CBlockTransferScalarPerVector_NWaveNPerXDL,
-        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>;
+        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        LoopSched,
+        PipelineVer>;
 
     using Argument              = typename GridwiseGemm::Argument;
     using DefaultBlock2CTileMap = typename GridwiseGemm::DefaultBlock2CTileMap;
@@ -257,7 +265,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                         StrideC,
                         GridwiseGemm::CalculateMPadded(M),
                         GridwiseGemm::CalculateNPadded(N),
-                        GridwiseGemm::CalculateKPadded(K),
+                        GridwiseGemm::CalculateKPadded(K, KBatch),
                         GridwiseGemm::CalculateK0(K, KBatch),
                         KBatch};
     }
@@ -290,7 +298,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                           StrideC,
                                           GridwiseGemm::CalculateMPadded(M),
                                           GridwiseGemm::CalculateNPadded(N),
-                                          GridwiseGemm::CalculateKPadded(K),
+                                          GridwiseGemm::CalculateKPadded(K, KBatch),
                                           GridwiseGemm::CalculateK0(K, KBatch),
                                           KBatch);
     }

include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp

@@ -85,7 +85,7 @@ template <typename ALayout,
           typename BElementwiseOperation,
           typename CDEElementwiseOperation,
           GemmSpecialization GemmSpec,
-          ck::index_t NumPrefetch,
+          ck::index_t NumGemmKPrefetchStage,
           ck::index_t BlockSize,
           ck::index_t MPerBlock,
           ck::index_t NPerBlock,
@@ -152,6 +152,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         BElementwiseOperation,
         CDEElementwiseOperation,
         GemmSpec,
+        NumGemmKPrefetchStage,
         MPerBlock,
         NPerBlock,
         K0PerBlock,
@@ -179,7 +180,9 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         CShuffleMXdlPerWavePerShuffle,
         CShuffleNXdlPerWavePerShuffle,
         CDEBlockTransferScalarPerVector_NPerBlock,
-        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>;
+        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        LoopSched,
+        PipelineVersion::v2>;
 
     using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
     using Block2ETileMapKSplit =
@@ -265,8 +268,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                 const index_t k_padded = GridwiseGemm::CalculateKPadded(K, K_BATCH);
                 const index_t k0       = GridwiseGemm::CalculateK0(K, K_BATCH);
 
-                const auto c_grid_desc_m_n =
-                    GridwiseGemm::MakeCGridDescriptor_M_N(M, N, m_padded, n_padded, stride_c);
+                const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, stride_c);
 
                 const auto local_b2c_tile_map =
                     Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
@@ -319,8 +321,8 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                 const index_t k_padded = GridwiseGemm::CalculateKPadded(karg.K, K_BATCH);
                 const index_t k0       = GridwiseGemm::CalculateK0(karg.K, K_BATCH);
 
-                const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(
-                    karg.M, karg.N, karg.MPadded, karg.NPadded, karg.StrideC);
+                const auto c_grid_desc_m_n =
+                    GridwiseGemm::MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC);
 
                 const auto local_b2c_tile_map =
                     Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
@@ -501,6 +503,11 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         if((ck::type_convert<ck::index_t>(arg.gemm_kernel_args_.size()) +
             arg.skipped_group_count_) != arg.group_count_)
         {
+#if DEBUG_LOG
+            std::cout << "The group count is not equal to sum of skipped groups "
+                         "and kernel args size!"
+                      << std::endl;
+#endif // DEBUG_LOG
             return false;
         }
 
@@ -509,14 +516,15 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         {
             const auto& a        = arg.gemm_kernel_args_[i].karg_;
             bool group_arg_valid = GridwiseGemm::CheckValidity(a);
-#if DEBUG_LOG
             if(not group_arg_valid)
             {
-                std::cout << "[" << __func__ << "] group id: " << i << " is not supported!\n";
+#if DEBUG_LOG
+                std::cout << "[" << __func__ << "] group id: " << i
+                          << " has invalid GridwiseGemm settings!" << std::endl;
                 a.Print();
-            }
 #endif // DEBUG_LOG
-            supported &= group_arg_valid;
+            }
+            supported = supported && group_arg_valid;
         }
         return supported;
     }

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

@@ -111,30 +111,37 @@ struct BlockToCTileMap_M00_N0_M01
 
 // Rows of column-vectors
 // This C-tile map dynamically adjusts M01 when C-tile index is out of range
-template <index_t MPerBlock, index_t NPerBlock, typename CGridDesc_M_N>
-struct BlockToCTileMap_M00_N0_M01Adapt
+template <index_t MPerBlock, index_t NPerBlock, typename CGridDesc_M_N = void>
+struct BlockToCTileMap_M00_N0_M01Adapt;
+
+template <index_t MPerBlock, index_t NPerBlock>
+struct BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>
 {
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
 
     __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt() = default;
 
-    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(const CGridDesc_M_N& c_grid_desc_m_n,
-                                                        index_t M01 = 8)
-        : M01_(M01), c_grid_desc_m_n_(c_grid_desc_m_n)
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(const BlockToCTileMap_M00_N0_M01Adapt&) =
+        default;
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(BlockToCTileMap_M00_N0_M01Adapt&&) =
+        default;
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt&
+    operator=(const BlockToCTileMap_M00_N0_M01Adapt&) = default;
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt&
+    operator=(BlockToCTileMap_M00_N0_M01Adapt&&) = default;
+
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(index_t M, index_t N, index_t M01 = 8)
+        : M_(M), N_(N), M01_(M01)
     {
     }
 
-    __host__ constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n) const
+    __host__ static constexpr index_t CalculateGridSize(index_t M, index_t N)
     {
-        const auto M0 = math::integer_divide_ceil(c_grid_desc_m_n.GetLength(I0), MPerBlock);
-        const auto N0 = math::integer_divide_ceil(c_grid_desc_m_n.GetLength(I1), NPerBlock);
-
-        const index_t grid_size = M0 * N0;
+        const auto M0 = math::integer_divide_ceil(M, MPerBlock);
+        const auto N0 = math::integer_divide_ceil(N, NPerBlock);
 
-        return grid_size;
+        return M0 * N0;
     }
 
     template <typename TopIdx>
@@ -142,8 +149,8 @@ struct BlockToCTileMap_M00_N0_M01Adapt
     {
         auto block_1d_id = idx_top[I0];
 
-        const auto M0 = math::integer_divide_ceil(c_grid_desc_m_n_.GetLength(I0), MPerBlock);
-        const auto N0 = math::integer_divide_ceil(c_grid_desc_m_n_.GetLength(I1), NPerBlock);
+        const auto M0 = math::integer_divide_ceil(M_, MPerBlock);
+        const auto N0 = math::integer_divide_ceil(N_, NPerBlock);
 
         block_1d_id = block_1d_id % (M0 * N0); // swallow batch index
 
@@ -211,11 +218,36 @@ struct BlockToCTileMap_M00_N0_M01Adapt
         return true; // always valid provided that user gets grid size from CalculateGridSize()
     }
 
-    __host__ bool CheckValidity(const CGridDesc_M_N& /* c_grid_desc_m_n */) const { return true; }
-
     private:
+    index_t M_;
+    index_t N_;
     index_t M01_;
-    CGridDesc_M_N c_grid_desc_m_n_;
+};
+
+template <index_t MPerBlock, index_t NPerBlock, typename CGridDesc_M_N>
+struct BlockToCTileMap_M00_N0_M01Adapt : BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>
+{
+    using Parent = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>;
+
+    using Parent::I0;
+    using Parent::I1;
+
+    using Parent::Parent;
+    using Parent::operator=;
+
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(const CGridDesc_M_N& c_grid_desc_m_n,
+                                                        index_t M01 = 8)
+        : Parent(c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1), M01)
+    {
+    }
+
+    __host__ static constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        return Parent::CalculateGridSize(c_grid_desc_m_n.GetLength(I0),
+                                         c_grid_desc_m_n.GetLength(I1));
+    }
+
+    __host__ bool CheckValidity(const CGridDesc_M_N& /* c_grid_desc_m_n */) const { return true; }
 };
 
 // 2D slices of column-vectors in 3D space

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp

@@ -8,14 +8,14 @@
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp"
 #include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
 #include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
 #include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp"
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
 
 namespace ck {
 
@@ -55,6 +55,7 @@ template <index_t BlockSize,
           typename BElementwiseOperation,
           typename CElementwiseOperation,
           tensor_operation::device::GemmSpecialization GemmSpec,
+          index_t NumGemmKPrefetchStage,
           index_t MPerBlock,
           index_t NPerBlock,
           index_t K0PerBlock,
@@ -82,7 +83,9 @@ template <index_t BlockSize,
           index_t CShuffleMRepeatPerShuffle,
           index_t CShuffleNRepeatPerShuffle,
           index_t CBlockTransferScalarPerVector_NWaveNPerXDL,
-          typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>
+          typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          LoopScheduler LoopSched     = make_default_loop_scheduler(),
+          PipelineVersion PipelineVer = PipelineVersion::v1>
 struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 {
     static constexpr auto I0 = Number<0>{};
@@ -99,8 +102,15 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
     static constexpr auto M01 = 1;
     static constexpr auto N01 = 1;
 
+    static constexpr auto gemm_padder =
+        tensor_operation::device::GemmPadder<GemmSpec, index_t, index_t, index_t>{
+            MPerBlock, NPerBlock, K1* K0PerBlock};
+
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
+    using GridwiseGemmPipe = remove_cvref_t<decltype(
+        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+
     struct Argument : public ck::tensor_operation::device::BaseArgument
     {
         const FloatAB* p_a_grid;
@@ -176,12 +186,12 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
     // prefer this to be called on host
     __host__ __device__ static auto CalculateMPadded(index_t M)
     {
-        return (M + MPerBlock - 1) / MPerBlock * MPerBlock;
+        return math::integer_least_multiple(M, MPerBlock);
     }
 
     __host__ __device__ static auto CalculateNPadded(index_t N)
     {
-        return (N + NPerBlock - 1) / NPerBlock * NPerBlock;
+        return math::integer_least_multiple(N, NPerBlock);
     }
 
     __host__ __device__ static auto CalculateK0(index_t K, index_t K_Batch = 1)
@@ -295,8 +305,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         }
     }
 
-    __host__ __device__ static auto
-    MakeCGridDescriptor_M_N(index_t M, index_t N, index_t MPad, index_t NPad, index_t StrideC)
+    __host__ __device__ static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
     {
         const auto c_grid_desc_m_n = [&]() {
             if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
@@ -309,22 +318,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             }
         }();
 
-        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
-        {
-            return transform_tensor_descriptor(c_grid_desc_m_n,
-                                               make_tuple(make_right_pad_transform(M, MPad - M),
-                                                          make_right_pad_transform(N, NPad - N)),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else
-        {
-            return transform_tensor_descriptor(
-                c_grid_desc_m_n,
-                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
+        return gemm_padder.PadCDescriptor_M_N(c_grid_desc_m_n);
     }
 
     __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
@@ -383,7 +377,15 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                        GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
         {
             if(!(karg.M % MPerBlock == 0))
+            {
+#if DEBUG_LOG
+                std::cout << "Arg M value is not a multiple of MPerBlock! M: " << karg.M << " "
+                          << __FILE__ << ":" << __LINE__ << ", in function: " << __func__
+                          << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
         if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
                        GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
@@ -391,40 +393,116 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                        GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
         {
             if(!(karg.N % NPerBlock == 0))
+            {
+#if DEBUG_LOG
+                std::cout << "Arg N value is not a multiple of NPerBlock! N: " << karg.N << " "
+                          << __FILE__ << ":" << __LINE__ << ", in function: " << __func__
+                          << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
 
         if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
         {
             if(karg.K % ABlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg K (" << karg.K
+                          << ") value is not a multiple of ABlockTransferSrcScalarPerVector ("
+                          << ABlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
         else
         {
             if(karg.M % ABlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg M (" << karg.M
+                          << ") value is not a multiple of ABlockTransferSrcScalarPerVector ("
+                          << ABlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
 
         if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
         {
             if(karg.N % BBlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg N (" << karg.N
+                          << ") value is not a multiple of BBlockTransferSrcScalarPerVector ("
+                          << BBlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
         else
         {
             if(karg.K % BBlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg K (" << karg.K
+                          << ") value is not a multiple of BBlockTransferSrcScalarPerVector ("
+                          << BBlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
 
         if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
         {
             if(karg.N % CBlockTransferScalarPerVector_NWaveNPerXDL != 0)
+            {
+#if DEBUG_LOG
+                std::cout
+                    << "Arg N (" << karg.N
+                    << ") value is not a multiple of CBlockTransferScalarPerVector_NWaveNPerXDL ("
+                    << CBlockTransferScalarPerVector_NWaveNPerXDL << " )! " << __FILE__ << ":"
+                    << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
         else
         {
             if(karg.M % CBlockTransferScalarPerVector_NWaveNPerXDL != 0)
+            {
+#if DEBUG_LOG
+                std::cout
+                    << "Arg M (" << karg.M
+                    << ") value is not a multiple of CBlockTransferScalarPerVector_NWaveNPerXDL ("
+                    << CBlockTransferScalarPerVector_NWaveNPerXDL << " )! " << __FILE__ << ":"
+                    << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
+        }
+
+        const auto num_k_loop = karg.K0 / K0PerBlock;
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+#if DEBUG_LOG
+            std::cout << "The number of k loops (" << num_k_loop
+                      << ") value is not supported by GridwiseGemm Pipeline."
+                      << " K0: " << karg.K0 << ", K0PerBlock: " << K0PerBlock << " " << __FILE__
+                      << ":" << __LINE__ << ", in function: " << __func__ << std::endl;
+#endif // DEBUG_LOG
+            return false;
         }
 
         return true;
@@ -439,9 +517,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 
     __host__ __device__ static constexpr bool CalculateHasMainK0BlockLoop(index_t K0)
     {
-        const bool has_main_k0_block_loop = K0 > K0PerBlock;
-
-        return has_main_k0_block_loop;
+        const index_t num_loop = K0 / K0PerBlock;
+        return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
     }
 
     template <typename CGridDesc>
@@ -490,7 +567,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         return BlockToCTileMap_3DGrid_KSplit<MPerBlock, NPerBlock>();
     }
 
-    using CGridDesc_M_N         = remove_cvref_t<decltype(MakeCGridDescriptor_M_N(1, 1, 1, 1, 1))>;
+    using CGridDesc_M_N         = remove_cvref_t<decltype(MakeCGridDescriptor_M_N(1, 1, 1))>;
     using DefaultBlock2CTileMap = remove_cvref_t<decltype(MakeDefaultBlock2CTileMap())>;
 
     template <bool HasMainKBlockLoop,
@@ -507,8 +584,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             karg.M, karg.MPadded, karg.K, karg.StrideA, karg.k_batch, karg.K0, karg.KPadded);
         const auto b_b_k0_n_k1_grid_desc = MakeBGridDescriptor_KBatch_K0_N_K1(
             karg.K, karg.NPadded, karg.N, karg.StrideB, karg.k_batch, karg.K0, karg.KPadded);
-        const auto c_grid_desc_m_n =
-            MakeCGridDescriptor_M_N(karg.M, karg.N, karg.MPadded, karg.NPadded, karg.StrideC);
+        const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC);
 
         const auto c_grid_desc_mblock_mperblock_nblock_nperblock =
             MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n);
@@ -680,20 +756,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
         //       register
         // sanity check
-#if 1
-        auto blockwise_gemm =
-            BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                FloatAB,
-                                                                FloatAcc,
-                                                                decltype(a_k0_m_k1_block_desc),
-                                                                decltype(b_k0_n_k1_block_desc),
-                                                                MPerXDL,
-                                                                NPerXDL,
-                                                                MRepeat,
-                                                                NRepeat,
-                                                                K1>{};
-#else
-        auto blockwise_gemm = BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<
+
+        auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             FloatAB,
             FloatAcc,
@@ -703,9 +767,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             NPerXDL,
             MRepeat,
             NRepeat,
-            K1>{};
-
-#endif
+            K1,
+            LoopSched>();
 
         auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
 
@@ -761,7 +824,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                 b_blockwise_copy.RunWrite(b_b_k0_n_k1_block_desc, b_block_buf);
 
                 k0_block_data_begin += K0PerBlock;
-            } while(k0_block_data_begin < (K0 - K0PerBlock));
+            } while(k0_block_data_begin < (karg.K0 - K0PerBlock));
         }
 
         // tail
@@ -772,13 +835,12 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         }
 #else
         // gridwise GEMM pipeline
-        const auto gridwise_gemm_pipeline =
-            GridwiseGemmPipeline_Selector<PipelineVersion::v2, 1, LoopScheduler::Default>();
-
         const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(
             (a_b_k0_m_k1_grid_desc.GetLength(I1) * a_b_k0_m_k1_grid_desc.GetLength(I3)) /
             (K0PerBlock * K1));
 
+        const auto gridwise_gemm_pipeline = GridwiseGemmPipe{};
+
         gridwise_gemm_pipeline.template Run<HasMainKBlockLoop>(a_b_k0_m_k1_grid_desc,
                                                                a_b_k0_m_k1_block_desc,
                                                                a_blockwise_copy,
@@ -993,24 +1055,6 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         }
     }
 
-    template <typename Layout>
-    struct LStr
-    {
-        static std::string Get() { return ""; }
-    };
-
-    template <>
-    struct LStr<ck::tensor_layout::gemm::RowMajor>
-    {
-        static std::string Get() { return "R"; }
-    };
-
-    template <>
-    struct LStr<ck::tensor_layout::gemm::ColumnMajor>
-    {
-        static std::string Get() { return "C"; }
-    };
-
     static std::string GetTypeString()
     {
         auto str = std::stringstream();

include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp

@@ -262,12 +262,12 @@ struct wmma_type<WmmaInstr::wmma_i32_16x16x16_iu8,
 
     template <index_t MPerWmma,
               index_t NPerWmma,
-              bool neg_a,
-              bool neg_b,
-              bool clamp,
               class FloatA,
               class FloatB,
-              class FloatC>
+              class FloatC,
+              bool neg_a = false,
+              bool neg_b = false,
+              bool clamp = false>
     __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
     {
         if constexpr(wave_size == 32)

library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp

@@ -22,21 +22,41 @@ void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -63,21 +83,41 @@ void add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(
         DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -295,6 +335,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
@@ -302,6 +343,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
             }
@@ -310,6 +352,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
@@ -317,6 +360,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
         }
@@ -352,24 +396,28 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(op_ptrs);
             }
         }
 

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

@@ -33,11 +33,19 @@ add_instance_library(device_gemm_instance
    device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
    device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
    device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
    device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
    device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
    device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
    device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
    device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
    device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
    device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
 )

library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp

@@ -34,7 +34,39 @@ using device_gemm_dl_f16_f16_f16_km_kn_mn_instances = std::tuple<
         // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 

library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace 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;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp

@@ -34,7 +34,39 @@ using device_gemm_dl_f16_f16_f16_km_nk_mn_instances = std::tuple<
         // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128       
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 

library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace 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;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128       
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64       
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp

@@ -34,7 +34,39 @@ using device_gemm_dl_f16_f16_f16_mk_kn_mn_instances = std::tuple<
         // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 

library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace 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;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp

@@ -35,7 +35,39 @@ using device_gemm_dl_f16_f16_f16_mk_nk_mn_instances =
         //  ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         //  ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         //  ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
         // clang-format on
         >;