Merge branch 'develop' into grouped_conv_3d_layout_fix

client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp

@@ -76,7 +76,7 @@ int main(int argc, char* argv[])
         StrideA  = std::stoi(argv[4]);
         StrideB  = std::stoi(argv[5]);
         StrideD0 = std::stoi(argv[6]);
-        StrideE  = std::stoi(argv[8]);
+        StrideE  = std::stoi(argv[7]);
     }
     else
     {

client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp

@@ -72,7 +72,7 @@ int main(int argc, char* argv[])
 
         StrideA = std::stoi(argv[4]);
         StrideB = std::stoi(argv[5]);
-        StrideE = std::stoi(argv[8]);
+        StrideE = std::stoi(argv[6]);
     }
     else
     {

example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp

@@ -121,7 +121,8 @@ using DeviceOpInstance =
         2,              // CShuffleNXdlPerWavePerShuffle
         S<1, 32, 1, 8>, // CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
         8,              // CShuffleBlockTransferScalarPerVector_NPerBlock
-        MaskingSpec>;   // MaskingSpecialization
+        MaskingSpec,    // MaskingSpecialization
+        1>;
 
 // Ref Gemm0: fp16 in, fp32 out
 using ReferenceGemm0Instance = ck::tensor_operation::host::ReferenceBatchedGemm<ADataType,

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

@@ -197,7 +197,8 @@ template <index_t NumDimG,
           typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
           MaskingSpecialization MaskingSpec,
-          LoopScheduler LoopSched = LoopScheduler::Default>
+          int D0sTransferSrcScalarPerVector = 4,
+          LoopScheduler LoopSched           = LoopScheduler::Default>
 struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
     : public DeviceBatchedGemmSoftmaxGemmPermute<NumDimG,
                                                  NumDimM,
@@ -438,7 +439,8 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         CShuffleBlockTransferScalarPerVector_NPerBlock,
         LoopSched,
         Transform::matrix_padder.PadN,
-        MaskingSpec == MaskingSpecialization::MaskOutUpperTriangle>;
+        MaskingSpec == MaskingSpecialization::MaskOutUpperTriangle,
+        D0sTransferSrcScalarPerVector>;
 
     // Argument
     // FIXME: constness
@@ -530,6 +532,11 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
                 using D0DataType = remove_cvref_t<tuple_element_t<i.value, D0sDataType>>;
                 // D0 pointer
                 p_d0s_grid_(i) = static_cast<const D0DataType*>(p_acc0_biases[i]);
+                // for  check
+                d0s_nl_ns_lengths_strides_[i].push_back(
+                    acc0_biases_gs_ms_ns_lengths[i][NumDimG + NumDimM]);
+                d0s_nl_ns_lengths_strides_[i].push_back(
+                    acc0_biases_gs_ms_ns_strides[i][NumDimG + NumDimM]);
             });
 
             if(GridwiseGemm::CheckValidity(a_grid_desc_ak0_m_ak1_,
@@ -608,6 +615,7 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         std::vector<index_t> b_nz_kz_strides_;
         std::vector<index_t> b1_nz_kz_strides_;
         std::vector<index_t> c_mz_gemm1nz_strides_;
+        std::array<std::vector<ck::index_t>, NumD0Tensor> d0s_nl_ns_lengths_strides_;
 
         index_t batch_count_;
         ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch_;
@@ -772,6 +780,20 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         {
             return false;
         }
+        for(int i = 0; i < NumD0Tensor; i++)
+        {
+            if(arg.d0s_nl_ns_lengths_strides_[i][1] == 1 &&
+               arg.d0s_nl_ns_lengths_strides_[i][0] % D0sTransferSrcScalarPerVector != 0)
+            {
+                std::cout << "first" << std::endl;
+                return false;
+            }
+            if(arg.d0s_nl_ns_lengths_strides_[i][1] != 1 && D0sTransferSrcScalarPerVector != 1)
+            {
+                std::cout << "second" << std::endl;
+                return false;
+            }
+        }
 
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
                                            arg.b_grid_desc_bk0_n_bk1_,

include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp

@@ -80,7 +80,8 @@ template <typename FloatAB,
           LoopScheduler LoopSched,
           bool PadN,
           bool MaskOutUpperTriangle,
-          PipelineVersion PipelineVer = PipelineVersion::v1>
+          int D0sTransferSrcScalarPerVector = 4,
+          PipelineVersion PipelineVer       = PipelineVersion::v1>
 struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
 {
     static_assert(LoopSched == LoopScheduler::Default,
@@ -621,13 +622,13 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
         constexpr auto d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5 =
             make_naive_tensor_descriptor_packed(make_tuple(I1,   // MBlockId
                                                            I1,   // NBlockID
-                                                           I1,   // MRepeat
-                                                           I1,   // NRepeat
-                                                           I1,   // MWaveId
-                                                           I1,   // NWaveId
-                                                           I1,   // MPerXdl
-                                                           I1,   // NGroupNum
-                                                           I1,   // NInputNum
+                                                           m0,   // MRepeat
+                                                           n0,   // NRepeat
+                                                           m1,   // MWaveId
+                                                           n1,   // NWaveId
+                                                           m2,   // MPerXdl
+                                                           n2,   // NGroupNum
+                                                           n3,   // NInputNum
                                                            n4)); // registerNum
 
         auto d0s_thread_buf = generate_tuple(
@@ -644,9 +645,6 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
         const auto wave_id     = GetGemm0WaveIdx();
         const auto wave_m_n_id = GetGemm0WaveMNIdx(wave_id[I2]); // I2: 0~63
 
-        constexpr auto acc0_thread_desc = make_naive_tensor_descriptor_packed(
-            make_tuple(Number<MXdlPerWave>{}, Number<NXdlPerWave>{}, n2, n4));
-
         auto d0s_threadwise_copy = generate_tuple(
             [&](auto i) {
                 using D0DataType = remove_cvref_t<tuple_element_t<i.value, D0sDataType>>;
@@ -655,10 +653,19 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
                     D0DataType,
                     decltype(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i]),
                     decltype(d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5),
-                    Sequence<I1, I1, I1, I1, I1, I1, I1, I1, I1, n4>,
+                    Sequence<I1, // MBlockId
+                             I1, // NBlockID
+                             m0, // MRepeat
+                             n0, // NRepeat
+                             m1, // MWaveId
+                             n1, // NWaveId
+                             m2, // MPerXdl
+                             n2, // NGroupNum
+                             n3, // NInputNum
+                             n4>,
                     Sequence<0, 1, 2, 3, 4, 5, 6, 7, 8, 9>,
                     9,
-                    n4,
+                    D0sTransferSrcScalarPerVector,
                     1,
                     false>(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
                            make_multi_index(block_work_idx[I0], // MBlockId
@@ -884,62 +891,35 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
             // multiple d
             if constexpr(NumD0Tensor)
             {
-                static_for<0, MXdlPerWave, 1>{}([&](auto mr) {
-                    static_for<0, NXdlPerWave, 1>{}([&](auto nr) {
-                        static_for<0, n2, 1>{}([&](auto groupid) {
-                            static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                                d0s_threadwise_copy(i).Run(
-                                    d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                                    d0s_grid_buf[i],
-                                    d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5,
-                                    make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                    d0s_thread_buf(i));
-                            });
-
-                            static_for<0, n4, 1>{}([&](auto i) {
-                                constexpr index_t c_offset = acc0_thread_desc.CalculateOffset(
-                                    make_tuple(mr, nr, groupid, i));
-
-                                // get reference to src data
-                                const auto src_data_refs = generate_tie(
-                                    // return type should be lvalue
-                                    [&](auto iSrc) -> const auto& {
-                                        return d0s_thread_buf[iSrc][i];
-                                    },
-                                    Number<NumD0Tensor>{});
-
-                                // get reference to dst data
-                                auto dst_data_refs = generate_tie(
-                                    // return type should be lvalue
-                                    [&](auto) -> auto& {
-                                        return acc_thread_buf(Number<c_offset>{});
-                                    },
-                                    Number<2>{});
-
-                                unpack2(c0de_element_op, dst_data_refs, src_data_refs);
-                            });
-                            static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                                d0s_threadwise_copy(i).MoveSrcSliceWindow(
-                                    d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                                    make_multi_index(0, 0, 0, 0, 0, 0, 0, 1, 0, 0));
-                            });
-                        });
-                        static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                            d0s_threadwise_copy(i).MoveSrcSliceWindow(
-                                d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                                make_multi_index(0, 0, 0, 1, 0, 0, 0, -n2.value, 0, 0));
-                        });
-                    });
-                    static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                        d0s_threadwise_copy(i).MoveSrcSliceWindow(
-                            d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                            make_multi_index(0, 0, 1, -NXdlPerWave, 0, 0, 0, 0, 0, 0));
-                    });
+                static_assert(NXdlPerWave == n0);
+                static_assert(MXdlPerWave == m0);
+
+                static_for<0, NumD0Tensor, 1>{}([&](auto i) {
+                    d0s_threadwise_copy(i).Run(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
+                                               d0s_grid_buf[i],
+                                               d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5,
+                                               make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
+                                               d0s_thread_buf(i));
+                });
+                static_for<0, m0 * n0 * n2 * n4, 1>{}([&](auto i) {
+                    // get reference to src data
+                    const auto src_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto iSrc) -> const auto& { return d0s_thread_buf[iSrc][i]; },
+                        Number<NumD0Tensor>{});
+
+                    // get reference to dst data
+                    auto dst_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto) -> auto& { return acc_thread_buf(i); },
+                        Number<2>{});
+
+                    unpack2(c0de_element_op, dst_data_refs, src_data_refs);
                 });
                 static_for<0, NumD0Tensor, 1>{}([&](auto i) {
                     d0s_threadwise_copy(i).MoveSrcSliceWindow(
                         d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                        make_multi_index(0, 1, -MXdlPerWave, 0, 0, 0, 0, 0, 0, 0));
+                        make_multi_index(0, 1, 0, 0, 0, 0, 0, 0, 0, 0));
                 });
             }
             else

include/ck/utility/amd_wave_read_first_lane.hpp

@@ -7,6 +7,7 @@
 #include "ck/utility/functional2.hpp"
 #include "ck/utility/math.hpp"
 
+#include <array>
 #include <cstddef>
 #include <cstdint>
 #include <type_traits>
@@ -14,29 +15,83 @@
 namespace ck {
 namespace detail {
 
-template <unsigned Size>
-struct get_unsigned_int;
+template <unsigned SizeInBytes>
+struct get_carrier;
 
 template <>
-struct get_unsigned_int<1>
+struct get_carrier<1>
 {
     using type = uint8_t;
 };
 
 template <>
-struct get_unsigned_int<2>
+struct get_carrier<2>
 {
     using type = uint16_t;
 };
 
 template <>
-struct get_unsigned_int<4>
+struct get_carrier<3>
+{
+    using type = class carrier
+    {
+        using value_type = uint32_t;
+
+        std::array<std::byte, 3> bytes;
+        static_assert(sizeof(bytes) <= sizeof(value_type));
+
+        // replacement of host std::copy_n()
+        template <typename InputIterator, typename Size, typename OutputIterator>
+        __device__ static OutputIterator copy_n(InputIterator from, Size size, OutputIterator to)
+        {
+            if(0 < size)
+            {
+                *to = *from;
+                ++to;
+                for(Size count = 1; count < size; ++count)
+                {
+                    *to = *++from;
+                    ++to;
+                }
+            }
+
+            return to;
+        }
+
+        // method to trigger template substitution failure
+        __device__ carrier(const carrier& other) noexcept
+        {
+            copy_n(other.bytes.begin(), bytes.size(), bytes.begin());
+        }
+
+        public:
+        __device__ carrier& operator=(value_type value) noexcept
+        {
+            copy_n(reinterpret_cast<const std::byte*>(&value), bytes.size(), bytes.begin());
+
+            return *this;
+        }
+
+        __device__ operator value_type() const noexcept
+        {
+            std::byte result[sizeof(value_type)];
+
+            copy_n(bytes.begin(), bytes.size(), result);
+
+            return *reinterpret_cast<const value_type*>(result);
+        }
+    };
+};
+static_assert(sizeof(get_carrier<3>::type) == 3);
+
+template <>
+struct get_carrier<4>
 {
     using type = uint32_t;
 };
 
-template <unsigned Size>
-using get_unsigned_int_t = typename get_unsigned_int<Size>::type;
+template <unsigned SizeInBytes>
+using get_carrier_t = typename get_carrier<SizeInBytes>::type;
 
 } // namespace detail
 
@@ -61,7 +116,7 @@ __device__ auto amd_wave_read_first_lane(const Object& obj)
     constexpr Size CompleteSgprCopyBoundary = ObjectSize - RemainedSize;
     for(Size offset = 0; offset < CompleteSgprCopyBoundary; offset += SgprSize)
     {
-        using Sgpr = detail::get_unsigned_int_t<SgprSize>;
+        using Sgpr = detail::get_carrier_t<SgprSize>;
 
         *reinterpret_cast<Sgpr*>(to_obj + offset) =
             amd_wave_read_first_lane(*reinterpret_cast<const Sgpr*>(from_obj + offset));
@@ -69,9 +124,9 @@ __device__ auto amd_wave_read_first_lane(const Object& obj)
 
     if constexpr(0 < RemainedSize)
     {
-        using Carrier = detail::get_unsigned_int_t<RemainedSize>;
+        using Carrier = detail::get_carrier_t<RemainedSize>;
 
-        *reinterpret_cast<Carrier>(to_obj + CompleteSgprCopyBoundary) = amd_wave_read_first_lane(
+        *reinterpret_cast<Carrier*>(to_obj + CompleteSgprCopyBoundary) = amd_wave_read_first_lane(
             *reinterpret_cast<const Carrier*>(from_obj + CompleteSgprCopyBoundary));
     }
 

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <vector>
+#include <memory>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+template <typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename EDataType>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatchedGemmMultiD<
+    ALayout,
+    BLayout,
+    Empty_Tuple,
+    ELayout,
+    ADataType,
+    BDataType,
+    Empty_Tuple,
+    EDataType,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough>>
+{
+    using DeviceOp = DeviceBatchedGemmMultiD<ALayout,
+                                             BLayout,
+                                             Empty_Tuple,
+                                             ELayout,
+                                             ADataType,
+                                             BDataType,
+                                             Empty_Tuple,
+                                             EDataType,
+                                             ck::tensor_operation::element_wise::PassThrough,
+                                             ck::tensor_operation::element_wise::PassThrough,
+                                             ck::tensor_operation::element_wise::PassThrough>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
+                          is_same_v<EDataType, int8_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

+add_instance_library(device_batched_gemm_multi_d_instance
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
+)

library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_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;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           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|  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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_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;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           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|  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
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_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;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           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|  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       
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_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;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           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|  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       
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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       
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_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;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           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|  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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_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;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           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|  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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_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;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           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|  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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_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;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           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|  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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, 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<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
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           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|  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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       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, 4>,     S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       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, 4>,     S<2, 1, 8, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       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, 4>,     S<2, 1, 8, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       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, 4>,     S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       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, 4>,     S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       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, 4>,     S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       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, 4>,     S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       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, 4>,     S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       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, 4>,     S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 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, 4>,     S<2, 1, 4, 4>,        S<4, 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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  4,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 4>,       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, 4>,     S<4, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  4,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 4>,       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, 4>,     S<4, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  4,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 4>,       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, 4>,     S<1, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  4,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 4>,       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, 4>,     S<1, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        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, 4>,     S<4, 1, 4, 4>,        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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  4,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 4>,       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, 4>,     S<8, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       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, 4>,     S<8, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 4>,       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, 4>,     S<1, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       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, 4>,     S<1, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        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, 4>,     S<1, 1, 4, 4>,        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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        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, 4>,     S<1, 1, 4, 4>,        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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        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, 4>,     S<1, 1, 4, 4>,        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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        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, 4>,     S<1, 1, 4, 4>,        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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           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|  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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       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, 4>,      S<2, 1, 8, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       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, 4>,      S<2, 1, 8, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       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, 4>,      S<2, 1, 8, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       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, 4>,      S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       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, 4>,      S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       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, 4>,      S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       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, 4>,      S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       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, 4>,      S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       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, 4>,      S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       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, 4>,      S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       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, 4>,      S<2, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 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, 4>,      S<2, 1, 4, 4>,       S<4, 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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 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, 4>,      S<2, 1, 4, 4>,       S<4, 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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 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, 4>,      S<2, 1, 4, 4>,       S<4, 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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 4>,        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, 4>,      S<4, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        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, 4>,      S<4, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       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, 4>,      S<8, 1, 4, 4>,      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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       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, 4>,      S<1, 1, 4, 4>,      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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        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, 4>,      S<1, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        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, 4>,      S<1, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        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, 4>,      S<1, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        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, 4>,      S<1, 1, 4, 4>,       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, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           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|  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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       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, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       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, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       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, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       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, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       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, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       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, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       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, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       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, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       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, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 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, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    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,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, 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
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, 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>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, 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
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck