Merge branch 'develop' into aosewski/ggemm_multi_d2

library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp

@@ -3,12 +3,23 @@
 
 #pragma once
 
-#include <iostream>
+#include <cmath>
+#include <cstdlib>
+#include <numeric>
 #include <type_traits>
-#include <sstream>
+#include <vector>
 
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_base.hpp"
+
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/fill.hpp"
 #include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -22,6 +33,7 @@ namespace host {
 //             Supports both GNCHW/NGCHW as well as GNHWC/NHWGC physical layout
 //             as long as dimensions in tensor descriptor is in GNCHW order
 //
+// @tparam     NDimSpatial  Number of spatial dimensions.
 // @tparam     InDataType               Input tensor data type.
 // @tparam     WeiDataType              Weights tensor data type.
 // @tparam     OutDataType              Output tensor data type.
@@ -29,7 +41,9 @@ namespace host {
 //                                      operation.
 // @tparam     WeiElementwiseOperation  Functor for weights tensor elementwise
 //                                      operation.
-// @tparam     NDimSpatial  Number of spatial dimensions.
+// @tparam     NumAElementwiseTensor  Number of A elementwise tensors.
+// @tparam     NumBElementwiseTensor  Number of B elementwise tensors.
+// @tparam     NumDElementwiseTensor  Number of D elementwise tensors.
 //
 // input descriptor in [G, N, C, Do, Ho, Wo] order
 // weight descriptor in [G, K, C, Z, Y, X] order
@@ -42,25 +56,35 @@ template <ck::index_t NDimSpatial,
           typename InElementwiseOperation,
           typename WeiElementwiseOperation,
           typename OutElementwiseOperation,
+          ck::index_t NumAElementwiseTensor                                         = 0,
+          ck::index_t NumBElementwiseTensor                                         = 0,
+          ck::index_t NumDElementwiseTensor                                         = 0,
           typename std::enable_if<NDimSpatial >= 1 && NDimSpatial <= 3, bool>::type = false>
 struct ReferenceConvFwd : public device::BaseOperator
 {
     // Argument
     struct Argument : public device::BaseArgument
     {
-        Argument(const Tensor<InDataType>& input,
-                 const Tensor<WeiDataType>& weight,
-                 Tensor<OutDataType>& output,
-                 std::vector<ck::index_t> conv_filter_strides,
-                 std::vector<ck::index_t> conv_filter_dilations,
-                 std::vector<ck::index_t> input_left_pads,
-                 std::vector<ck::index_t> input_right_pads,
-                 InElementwiseOperation in_element_op,
-                 WeiElementwiseOperation wei_element_op,
-                 OutElementwiseOperation out_element_op)
+        Argument(
+            const Tensor<InDataType>& input,
+            const Tensor<WeiDataType>& weight,
+            Tensor<OutDataType>& output,
+            std::vector<ck::index_t> conv_filter_strides,
+            std::vector<ck::index_t> conv_filter_dilations,
+            std::vector<ck::index_t> input_left_pads,
+            std::vector<ck::index_t> input_right_pads,
+            InElementwiseOperation in_element_op,
+            WeiElementwiseOperation wei_element_op,
+            OutElementwiseOperation out_element_op,
+            const std::array<Tensor<InDataType>, NumAElementwiseTensor>& elementwise_a_tensors,
+            const std::array<Tensor<WeiDataType>, NumBElementwiseTensor>& elementwise_b_tensors,
+            const std::array<Tensor<OutDataType>, NumDElementwiseTensor>& elementwise_d_tensors)
             : input_{input},
               weight_{weight},
               output_{output},
+              elementwise_a_tensors_{elementwise_a_tensors},
+              elementwise_b_tensors_{elementwise_b_tensors},
+              elementwise_d_tensors_{elementwise_d_tensors},
               conv_strides_{conv_filter_strides},
               conv_dilations_{conv_filter_dilations},
               in_left_pads_{input_left_pads},
@@ -75,6 +99,10 @@ struct ReferenceConvFwd : public device::BaseOperator
         const Tensor<WeiDataType>& weight_;
         Tensor<OutDataType>& output_;
 
+        const std::array<Tensor<InDataType>, NumAElementwiseTensor>& elementwise_a_tensors_;
+        const std::array<Tensor<WeiDataType>, NumBElementwiseTensor>& elementwise_b_tensors_;
+        const std::array<Tensor<OutDataType>, NumDElementwiseTensor>& elementwise_d_tensors_;
+
         std::vector<index_t> conv_strides_;
         std::vector<index_t> conv_dilations_;
         std::vector<index_t> in_left_pads_;
@@ -114,25 +142,43 @@ struct ReferenceConvFwd : public device::BaseOperator
                             if(wi >= 0 &&
                                ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[3])
                             {
-                                float v_in;
-                                float v_wei;
-
-                                arg.in_element_op_(
-                                    v_in, ck::type_convert<float>(arg.input_(g, n, c, wi)));
-
-                                arg.wei_element_op_(
-                                    v_wei, ck::type_convert<float>(arg.weight_(g, k, c, x)));
-
-                                v_acc += v_in * v_wei;
+                                InDataType v_in;
+                                WeiDataType v_wei;
+
+                                ExecuteElementwiseOp(arg.in_element_op_,
+                                                     arg.elementwise_a_tensors_,
+                                                     Number<NumAElementwiseTensor>{},
+                                                     v_in,
+                                                     arg.input_(g, n, c, wi),
+                                                     g,
+                                                     n,
+                                                     c,
+                                                     wi);
+                                ExecuteElementwiseOp(arg.wei_element_op_,
+                                                     arg.elementwise_b_tensors_,
+                                                     Number<NumBElementwiseTensor>{},
+                                                     v_wei,
+                                                     arg.weight_(g, k, c, x),
+                                                     g,
+                                                     k,
+                                                     c,
+                                                     x);
+                                v_acc +=
+                                    ck::type_convert<float>(v_in) * ck::type_convert<float>(v_wei);
                             }
                         }
                     }
-
-                    float v_out;
-
-                    arg.out_element_op_(v_out, v_acc);
-
-                    arg.output_(g, n, k, wo) = ck::type_convert<OutDataType>(v_out);
+                    OutDataType v_acc_converted = ck::type_convert<OutDataType>(v_acc);
+                    OutDataType& v_out          = arg.output_(g, n, k, wo);
+                    ExecuteElementwiseOp(arg.out_element_op_,
+                                         arg.elementwise_d_tensors_,
+                                         Number<NumDElementwiseTensor>{},
+                                         v_out,
+                                         v_acc_converted,
+                                         g,
+                                         n,
+                                         k,
+                                         wo);
                 };
 
                 make_ParallelTensorFunctor(func,
@@ -169,26 +215,47 @@ struct ReferenceConvFwd : public device::BaseOperator
                                    wi >= 0 &&
                                    ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[4])
                                 {
-                                    float v_in;
-                                    float v_wei;
-
-                                    arg.in_element_op_(
-                                        v_in, ck::type_convert<float>(arg.input_(g, n, c, hi, wi)));
-
-                                    arg.wei_element_op_(
-                                        v_wei, ck::type_convert<float>(arg.weight_(g, k, c, y, x)));
-
-                                    v_acc += v_in * v_wei;
+                                    InDataType v_in;
+                                    WeiDataType v_wei;
+
+                                    ExecuteElementwiseOp(arg.in_element_op_,
+                                                         arg.elementwise_a_tensors_,
+                                                         Number<NumAElementwiseTensor>{},
+                                                         v_in,
+                                                         arg.input_(g, n, c, hi, wi),
+                                                         g,
+                                                         n,
+                                                         c,
+                                                         hi,
+                                                         wi);
+                                    ExecuteElementwiseOp(arg.wei_element_op_,
+                                                         arg.elementwise_b_tensors_,
+                                                         Number<NumBElementwiseTensor>{},
+                                                         v_wei,
+                                                         arg.weight_(g, k, c, y, x),
+                                                         g,
+                                                         k,
+                                                         c,
+                                                         y,
+                                                         x);
+                                    v_acc += ck::type_convert<float>(v_in) *
+                                             ck::type_convert<float>(v_wei);
                                 }
                             }
                         }
                     }
-
-                    float v_out;
-
-                    arg.out_element_op_(v_out, v_acc);
-
-                    arg.output_(g, n, k, ho, wo) = ck::type_convert<OutDataType>(v_out);
+                    OutDataType v_acc_converted = ck::type_convert<OutDataType>(v_acc);
+                    OutDataType& v_out          = arg.output_(g, n, k, ho, wo);
+                    ExecuteElementwiseOp(arg.out_element_op_,
+                                         arg.elementwise_d_tensors_,
+                                         Number<NumDElementwiseTensor>{},
+                                         v_out,
+                                         v_acc_converted,
+                                         g,
+                                         n,
+                                         k,
+                                         ho,
+                                         wo);
                 };
 
                 make_ParallelTensorFunctor(func,
@@ -235,29 +302,51 @@ struct ReferenceConvFwd : public device::BaseOperator
                                        ck::type_convert<std::size_t>(wi) <
                                            arg.input_.GetLengths()[5])
                                     {
-                                        float v_in;
-                                        float v_wei;
-
-                                        arg.in_element_op_(v_in,
-                                                           ck::type_convert<float>(
-                                                               arg.input_(g, n, c, di, hi, wi)));
-
-                                        arg.wei_element_op_(
-                                            v_wei,
-                                            ck::type_convert<float>(arg.weight_(g, k, c, z, y, x)));
-
-                                        v_acc += v_in * v_wei;
+                                        InDataType v_in;
+                                        WeiDataType v_wei;
+
+                                        ExecuteElementwiseOp(arg.in_element_op_,
+                                                             arg.elementwise_a_tensors_,
+                                                             Number<NumAElementwiseTensor>{},
+                                                             v_in,
+                                                             arg.input_(g, n, c, di, hi, wi),
+                                                             g,
+                                                             n,
+                                                             c,
+                                                             di,
+                                                             hi,
+                                                             wi);
+                                        ExecuteElementwiseOp(arg.wei_element_op_,
+                                                             arg.elementwise_b_tensors_,
+                                                             Number<NumBElementwiseTensor>{},
+                                                             v_wei,
+                                                             arg.weight_(g, k, c, z, y, x),
+                                                             g,
+                                                             k,
+                                                             c,
+                                                             z,
+                                                             y,
+                                                             x);
+                                        v_acc += ck::type_convert<float>(v_in) *
+                                                 ck::type_convert<float>(v_wei);
                                     }
                                 }
                             }
                         }
                     }
-
-                    float v_out;
-
-                    arg.out_element_op_(v_out, v_acc);
-
-                    arg.output_(g, n, k, d_o, ho, wo) = ck::type_convert<OutDataType>(v_out);
+                    OutDataType v_acc_converted = ck::type_convert<OutDataType>(v_acc);
+                    OutDataType& v_out          = arg.output_(g, n, k, d_o, ho, wo);
+                    ExecuteElementwiseOp(arg.out_element_op_,
+                                         arg.elementwise_d_tensors_,
+                                         Number<NumDElementwiseTensor>{},
+                                         v_out,
+                                         v_acc_converted,
+                                         g,
+                                         n,
+                                         k,
+                                         d_o,
+                                         ho,
+                                         wo);
                 };
 
                 make_ParallelTensorFunctor(func,
@@ -280,6 +369,36 @@ struct ReferenceConvFwd : public device::BaseOperator
         }
     };
 
+    template <typename... Args,
+              typename ElementwiseOp,
+              typename ElementwiseTensor,
+              typename NumTensor,
+              typename T>
+    static void ExecuteElementwiseOp(ElementwiseOp& elementwise_op,
+                                     ElementwiseTensor& elementwise_tensors,
+                                     NumTensor,
+                                     T& y,
+                                     const T& x,
+                                     Args... dims)
+    {
+        if constexpr(NumTensor::value == 0)
+        {
+            elementwise_op(y, x);
+        }
+        else if constexpr(NumTensor::value == 1)
+        {
+            elementwise_op(y, x, elementwise_tensors[0](dims...));
+        }
+        else if constexpr(NumTensor::value == 2)
+        {
+            elementwise_op(y, x, elementwise_tensors[0](dims...), elementwise_tensors[1](dims...));
+        }
+        else
+        {
+            throw std::runtime_error("ElementOp not supported in reference.");
+        }
+    }
+
     static constexpr bool IsValidCompilationParameter()
     {
         // TODO: properly implement this check
@@ -291,16 +410,20 @@ struct ReferenceConvFwd : public device::BaseOperator
         return NDimSpatial >= 1 && NDimSpatial <= 3;
     }
 
-    static auto MakeArgument(const Tensor<InDataType>& input,
-                             const Tensor<WeiDataType>& weight,
-                             Tensor<OutDataType>& output,
-                             std::vector<ck::index_t> conv_filter_strides,
-                             std::vector<ck::index_t> conv_filter_dilations,
-                             std::vector<ck::index_t> input_left_pads,
-                             std::vector<ck::index_t> input_right_pads,
-                             InElementwiseOperation in_element_op,
-                             WeiElementwiseOperation wei_element_op,
-                             OutElementwiseOperation out_element_op)
+    static auto MakeArgument(
+        const Tensor<InDataType>& input,
+        const Tensor<WeiDataType>& weight,
+        Tensor<OutDataType>& output,
+        std::vector<ck::index_t> conv_filter_strides,
+        std::vector<ck::index_t> conv_filter_dilations,
+        std::vector<ck::index_t> input_left_pads,
+        std::vector<ck::index_t> input_right_pads,
+        InElementwiseOperation in_element_op,
+        WeiElementwiseOperation wei_element_op,
+        OutElementwiseOperation out_element_op,
+        const std::array<Tensor<InDataType>, NumAElementwiseTensor>& elementwise_a_tensors  = {},
+        const std::array<Tensor<WeiDataType>, NumBElementwiseTensor>& elementwise_b_tensors = {},
+        const std::array<Tensor<OutDataType>, NumDElementwiseTensor>& elementwise_d_tensors = {})
     {
         return Argument{input,
                         weight,
@@ -311,7 +434,10 @@ struct ReferenceConvFwd : public device::BaseOperator
                         input_right_pads,
                         in_element_op,
                         wei_element_op,
-                        out_element_op};
+                        out_element_op,
+                        elementwise_a_tensors,
+                        elementwise_b_tensors,
+                        elementwise_d_tensors};
     }
 
     static auto MakeInvoker() { return Invoker{}; }

library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm_bwd.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+#include <vector>
+#include <algorithm>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+
+template <typename DYDataType,
+          typename XDataType,
+          typename GammaDataType,
+          typename MeanInvStdDataType,
+          typename DGammaDataType,
+          typename DBetaDataType,
+          typename DXDataType,
+          typename ComputeDataType>
+struct ReferenceGroupnormBwd : public device::BaseOperator
+{
+    // Argument
+    struct Argument : public device::BaseArgument
+    {
+        Argument(const Tensor<DYDataType>& dy_nhwgc,
+                 const Tensor<XDataType>& x_nhwgc,
+                 const Tensor<GammaDataType>& gamma_gc,
+                 const Tensor<MeanInvStdDataType>& mean_ng,
+                 const Tensor<MeanInvStdDataType>& inv_std_ng,
+                 Tensor<DGammaDataType>& dgamma_gc,
+                 Tensor<DBetaDataType>& dbeta_gc,
+                 Tensor<DXDataType>& dx_nhwgc,
+                 const std::vector<index_t> lengths)
+            : dy_nhwgc_(dy_nhwgc),
+              x_nhwgc_(x_nhwgc),
+              gamma_gc_(gamma_gc),
+              mean_ng_(mean_ng),
+              inv_std_ng_(inv_std_ng),
+              dgamma_gc_(dgamma_gc),
+              dbeta_gc_(dbeta_gc),
+              dx_nhwgc_(dx_nhwgc),
+              lengths_(lengths)
+        {
+        }
+
+        const Tensor<DYDataType>& dy_nhwgc_;
+        const Tensor<XDataType>& x_nhwgc_;
+        const Tensor<GammaDataType>& gamma_gc_;
+        const Tensor<MeanInvStdDataType>& mean_ng_;
+        const Tensor<MeanInvStdDataType>& inv_std_ng_;
+        Tensor<DGammaDataType>& dgamma_gc_;
+        Tensor<DBetaDataType>& dbeta_gc_;
+        Tensor<DXDataType>& dx_nhwgc_;
+        std::vector<index_t> lengths_;
+    };
+
+    // Invoker
+    struct Invoker : public device::BaseInvoker
+    {
+        float Run(const Argument& arg)
+        {
+            int N = arg.lengths_[0];
+            int H = arg.lengths_[1];
+            int W = arg.lengths_[2];
+            int G = arg.lengths_[3];
+            int C = arg.lengths_[4];
+
+            // Calculate dgamma and dbeta
+            for(int g = 0; g < G; ++g)
+                for(int c = 0; c < C; ++c)
+                {
+                    ComputeDataType dgamma = 0;
+                    ComputeDataType dbeta  = 0;
+
+                    for(int n = 0; n < N; ++n)
+                        for(int h = 0; h < H; ++h)
+                            for(int w = 0; w < W; ++w)
+                            {
+                                ComputeDataType dy =
+                                    ck::type_convert<ComputeDataType>(arg.dy_nhwgc_(n, h, w, g, c));
+                                ComputeDataType x =
+                                    ck::type_convert<ComputeDataType>(arg.x_nhwgc_(n, h, w, g, c));
+                                ComputeDataType mean =
+                                    ck::type_convert<ComputeDataType>(arg.mean_ng_(n, g));
+                                ComputeDataType rstd =
+                                    ck::type_convert<ComputeDataType>(arg.inv_std_ng_(n, g));
+                                dgamma += dy * rstd * (x - mean);
+                                dbeta += dy;
+                            }
+                    arg.dgamma_gc_(g, c) = ck::type_convert<DGammaDataType>(dgamma);
+                    arg.dbeta_gc_(g, c)  = ck::type_convert<DBetaDataType>(dbeta);
+                }
+
+            // Calculate dx
+            int reduce_size = H * W * C;
+            for(int n = 0; n < N; ++n)
+                for(int g = 0; g < G; ++g)
+                {
+                    ComputeDataType ds = 0;
+                    ComputeDataType db = 0;
+
+                    ComputeDataType mean = ck::type_convert<ComputeDataType>(arg.mean_ng_(n, g));
+                    ComputeDataType rstd = ck::type_convert<ComputeDataType>(arg.inv_std_ng_(n, g));
+
+                    for(int h = 0; h < H; ++h)
+                        for(int w = 0; w < W; ++w)
+                            for(int c = 0; c < C; ++c)
+                            {
+                                ComputeDataType dy =
+                                    ck::type_convert<ComputeDataType>(arg.dy_nhwgc_(n, h, w, g, c));
+                                ComputeDataType x =
+                                    ck::type_convert<ComputeDataType>(arg.x_nhwgc_(n, h, w, g, c));
+                                ComputeDataType gamma =
+                                    ck::type_convert<ComputeDataType>(arg.gamma_gc_(g, c));
+
+                                ds += dy * gamma * x;
+                                db += dy * gamma;
+                            }
+
+                    for(int h = 0; h < H; ++h)
+                        for(int w = 0; w < W; ++w)
+                            for(int c = 0; c < C; ++c)
+                            {
+                                ComputeDataType dy =
+                                    ck::type_convert<ComputeDataType>(arg.dy_nhwgc_(n, h, w, g, c));
+                                ComputeDataType x =
+                                    ck::type_convert<ComputeDataType>(arg.x_nhwgc_(n, h, w, g, c));
+                                ComputeDataType gamma =
+                                    ck::type_convert<ComputeDataType>(arg.gamma_gc_(g, c));
+
+                                ComputeDataType b =
+                                    (db * mean - ds) * rstd * rstd * rstd / reduce_size;
+                                ComputeDataType c1 = -b * mean - db * rstd / reduce_size;
+                                arg.dx_nhwgc_(n, h, w, g, c) =
+                                    ck::type_convert<DXDataType>(dy * gamma * rstd + b * x + c1);
+                            }
+                }
+
+            return 0;
+        }
+
+        float Run(const device::BaseArgument* p_arg,
+                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
+
+    static auto MakeArgument(const Tensor<DYDataType>& dy_nhwgc,
+                             const Tensor<XDataType>& x_nhwgc,
+                             const Tensor<GammaDataType>& gamma_gc,
+                             const Tensor<MeanInvStdDataType>& mean_ng,
+                             const Tensor<MeanInvStdDataType>& inv_std_ng,
+                             Tensor<DGammaDataType>& dgamma_gc,
+                             Tensor<DBetaDataType>& dbeta_gc,
+                             Tensor<DXDataType>& dx_nhwgc,
+                             const std::vector<index_t> lengths)
+    {
+        return Argument{dy_nhwgc,
+                        x_nhwgc,
+                        gamma_gc,
+                        mean_ng,
+                        inv_std_ng,
+                        dgamma_gc,
+                        dbeta_gc,
+                        dx_nhwgc,
+                        lengths};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferenceGroupnormBwd"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp

@@ -19,9 +19,7 @@ namespace host {
  * \brief Reference implementation for image to column.
  *
  * Input tensor descriptor has [G, N, C, Di, Hi, Wi] data layout.
- * G must be equal to 1. Memory layout is [G, N, Di, Hi, Wi, C].
- * Output tensor descriptor has [N * Do * Ho * Wo, Z * Y * X * C] data layout.
- * Memory layout is the same.
+ * Output tensor descriptor has [G * N * Do * Ho * Wo, Z * Y * X * C] data layout.
  *
  * \tparam NDimSpatial Number of spatial dimensions.
  * \tparam ImageLayout Image Layout.
@@ -95,18 +93,19 @@ struct ReferenceImageToColumn : public device::BaseOperator
         float Run(const Argument& arg)
         {
             if(!(arg.input_.GetNumOfDimension() == NDimSpatial + 3 &&
-                 arg.output_.GetNumOfDimension() == 2))
+                 arg.output_.GetNumOfDimension() == 3))
             {
                 throw std::runtime_error("wrong! inconsistent dimension");
             }
 
+            const index_t G = arg.input_.GetLengths()[0];
             const index_t N = arg.input_.GetLengths()[1];
             const index_t C = arg.input_.GetLengths()[2];
 
             if constexpr(NDimSpatial == 1)
             {
                 const index_t Wo = arg.output_spatial_lengths_[0];
-                auto func        = [&](auto n, auto wo) {
+                auto func        = [&](auto g, auto n, auto wo) {
                     index_t row    = n * Wo + wo;
                     index_t column = 0;
 
@@ -121,15 +120,15 @@ struct ReferenceImageToColumn : public device::BaseOperator
                             if(wi >= 0 &&
                                ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[3])
                             {
-                                InDataType v_in          = arg.input_(0, n, c, wi);
-                                arg.output_(row, column) = ck::type_convert<OutDataType>(v_in);
+                                InDataType v_in             = arg.input_(g, n, c, wi);
+                                arg.output_(g, row, column) = ck::type_convert<OutDataType>(v_in);
                             }
                             column++;
                         }
                     }
                 };
 
-                make_ParallelTensorFunctor(func, N, Wo)(std::thread::hardware_concurrency());
+                make_ParallelTensorFunctor(func, G, N, Wo)(std::thread::hardware_concurrency());
 
                 return 0;
             }
@@ -138,7 +137,7 @@ struct ReferenceImageToColumn : public device::BaseOperator
                 const index_t Ho = arg.output_spatial_lengths_[0];
                 const index_t Wo = arg.output_spatial_lengths_[1];
 
-                auto func = [&](auto n, auto ho, auto wo) {
+                auto func = [&](auto g, auto n, auto ho, auto wo) {
                     index_t row    = n * Ho * Wo + ho * Wo + wo;
                     index_t column = 0;
 
@@ -162,8 +161,9 @@ struct ReferenceImageToColumn : public device::BaseOperator
                                    wi >= 0 &&
                                    ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[4])
                                 {
-                                    InDataType v_in          = arg.input_(0, n, c, hi, wi);
-                                    arg.output_(row, column) = ck::type_convert<OutDataType>(v_in);
+                                    InDataType v_in = arg.input_(g, n, c, hi, wi);
+                                    arg.output_(g, row, column) =
+                                        ck::type_convert<OutDataType>(v_in);
                                 }
                                 column++;
                             }
@@ -171,7 +171,7 @@ struct ReferenceImageToColumn : public device::BaseOperator
                     }
                 };
 
-                make_ParallelTensorFunctor(func, N, Ho, Wo)(std::thread::hardware_concurrency());
+                make_ParallelTensorFunctor(func, G, N, Ho, Wo)(std::thread::hardware_concurrency());
 
                 return 0;
             }
@@ -181,7 +181,7 @@ struct ReferenceImageToColumn : public device::BaseOperator
                 const index_t Ho = arg.output_spatial_lengths_[1];
                 const index_t Wo = arg.output_spatial_lengths_[2];
 
-                auto func = [&](auto n, auto d_o, auto ho, auto wo) {
+                auto func = [&](auto g, auto n, auto d_o, auto ho, auto wo) {
                     index_t row    = n * Do * Ho * Wo + d_o * Ho * Wo + ho * Wo + wo;
                     index_t column = 0;
 
@@ -213,8 +213,8 @@ struct ReferenceImageToColumn : public device::BaseOperator
                                        ck::type_convert<std::size_t>(wi) <
                                            arg.input_.GetLengths()[5])
                                     {
-                                        InDataType v_in = arg.input_(0, n, c, di, hi, wi);
-                                        arg.output_(row, column) =
+                                        InDataType v_in = arg.input_(g, n, c, di, hi, wi);
+                                        arg.output_(g, row, column) =
                                             ck::type_convert<OutDataType>(v_in);
                                     }
                                     column++;
@@ -224,7 +224,7 @@ struct ReferenceImageToColumn : public device::BaseOperator
                     }
                 };
 
-                make_ParallelTensorFunctor(func, N, Do, Ho, Wo)(
+                make_ParallelTensorFunctor(func, G, N, Do, Ho, Wo)(
                     std::thread::hardware_concurrency());
 
                 return 0;
@@ -267,8 +267,9 @@ struct ReferenceImageToColumn : public device::BaseOperator
             C * ck::accumulate_n<index_t>(
                     arg.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
 
-        if(!(arg.output_.GetLengths()[0] == static_cast<std::size_t>(NDoHoWo) &&
-             arg.output_.GetLengths()[1] == static_cast<std::size_t>(CZYX)))
+        if(!(arg.output_.GetLengths()[0] == static_cast<std::size_t>(G) &&
+             arg.output_.GetLengths()[1] == static_cast<std::size_t>(NDoHoWo) &&
+             arg.output_.GetLengths()[2] == static_cast<std::size_t>(CZYX)))
         {
             return false;
         }

library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp

@@ -28,7 +28,8 @@ template <typename XDataType,
 struct ReferenceLayernorm : public device::BaseOperator
 {
     // TODO - support generic layernorm
-    static_assert((Rank == 2 && NumReduceDim == 1), "Only support 2D version so far");
+    static_assert((Rank == 2 && NumReduceDim == 1) || (Rank == 4 && NumReduceDim == 3),
+                  "Only support 2D & 4D version so far");
 
     // Argument
     struct Argument : public device::BaseArgument
@@ -71,7 +72,7 @@ struct ReferenceLayernorm : public device::BaseOperator
     // Invoker
     struct Invoker : public device::BaseInvoker
     {
-        float Run(const Argument& arg)
+        float Run2D(const Argument& arg)
         {
             int M = arg.lengths_[0];
             int N = arg.lengths_[1];
@@ -117,6 +118,71 @@ struct ReferenceLayernorm : public device::BaseOperator
             return 0;
         }
 
+        float Run4D(const Argument& arg)
+        {
+            int N = arg.lengths_[0];
+            int H = arg.lengths_[1];
+            int W = arg.lengths_[2];
+            int C = arg.lengths_[3];
+
+            Tensor<ComputeDataType> mean({N});
+            Tensor<ComputeDataType> var({N});
+
+            int reduce_length = H * W * C;
+
+            for(int n = 0; n < N; ++n)
+            {
+                mean(n) = 0;
+                var(n)  = 0;
+
+                for(int h = 0; h < H; ++h)
+                    for(int w = 0; w < W; ++w)
+                        for(int c = 0; c < C; ++c)
+                        {
+                            auto x_val = ck::type_convert<ComputeDataType>(arg.x_m_n_(n, h, w, c));
+                            mean(n) += x_val;
+                            var(n) += x_val * x_val;
+                        }
+
+                mean(n) = mean(n) / reduce_length;
+                var(n)  = (var(n) / reduce_length) - (mean(n) * mean(n));
+            }
+
+            for(int n = 0; n < N; ++n)
+            {
+                ComputeDataType divisor =
+                    static_cast<ComputeDataType>(1) / ck::math::sqrt(var(n) + arg.epsilon_);
+
+                for(int h = 0; h < H; ++h)
+                    for(int w = 0; w < W; ++w)
+                        for(int c = 0; c < C; ++c)
+                        {
+                            auto x_val = ck::type_convert<ComputeDataType>(arg.x_m_n_(n, h, w, c));
+                            auto gamma_val =
+                                ck::type_convert<ComputeDataType>(arg.gamma_n_(h, w, c));
+                            auto beta_val = ck::type_convert<ComputeDataType>(arg.beta_n_(h, w, c));
+                            auto y_val    = (x_val - mean(n)) * divisor;
+                            y_val         = (y_val * gamma_val) + beta_val;
+                            arg.y_elementwise_op_(y_val, y_val);
+                            arg.y_m_n_(n, h, w, c) = ck::type_convert<YDataType>(y_val);
+                        }
+                arg.save_mean_m_(n)    = ck::type_convert<SaveMeanInvStdDataType>(mean(n));
+                arg.save_inv_std_m_(n) = ck::type_convert<SaveMeanInvStdDataType>(divisor);
+            }
+
+            return 0;
+        }
+
+        float Run(const Argument& arg)
+        {
+            if(arg.lengths_.size() == 2)
+                return Run2D(arg);
+            else if(arg.lengths_.size() == 4)
+                return Run4D(arg);
+
+            return 0;
+        }
+
         float Run(const device::BaseArgument* p_arg,
                   const StreamConfig& /* stream_config */ = StreamConfig{}) override
         {
@@ -134,17 +200,16 @@ struct ReferenceLayernorm : public device::BaseOperator
     {
         const Argument* p_arg_ = dynamic_cast<const Argument*>(p_arg);
 
-        // TODO - support generic layernorm
-        if(p_arg_->lengths_.size() != 2)
-            return false;
-
-        if(p_arg_->reduceDims_.size() != 1)
-            return false;
+        if(p_arg_->lengths_.size() == 2 && p_arg_->reduceDims_.size() == 1 &&
+           p_arg_->reduceDims_[0] == 1)
+            return true;
 
-        if(p_arg_->reduceDims_[0] != 1)
-            return false;
+        else if(p_arg_->lengths_.size() == 4 && p_arg_->reduceDims_.size() == 3 &&
+                p_arg_->reduceDims_[0] == 1 && p_arg_->reduceDims_[1] == 2 &&
+                p_arg_->reduceDims_[2] == 3)
+            return true;
 
-        return true;
+        return false;
     }
 
     static auto MakeArgument(const Tensor<XDataType>& x_m_n,

library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm_bwd.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+#include <vector>
+#include <algorithm>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+
+template <typename DYDataType,
+          typename XDataType,
+          typename GammaDataType,
+          typename MeanInvStdDataType,
+          typename DGammaDataType,
+          typename DBetaDataType,
+          typename DXDataType,
+          typename ComputeDataType>
+struct ReferenceLayernormBwd : public device::BaseOperator
+{
+    // Argument
+    struct Argument : public device::BaseArgument
+    {
+        Argument(const Tensor<DYDataType>& dy_m_n,
+                 const Tensor<XDataType>& x_m_n,
+                 const Tensor<GammaDataType>& gamma_n,
+                 const Tensor<MeanInvStdDataType>& mean_m,
+                 const Tensor<MeanInvStdDataType>& inv_std_m,
+                 Tensor<DGammaDataType>& dgamma_n,
+                 Tensor<DBetaDataType>& dbeta_n,
+                 Tensor<DXDataType>& dx_m_n,
+                 const std::vector<index_t> lengths)
+            : dy_m_n_(dy_m_n),
+              x_m_n_(x_m_n),
+              gamma_n_(gamma_n),
+              mean_m_(mean_m),
+              inv_std_m_(inv_std_m),
+              dgamma_n_(dgamma_n),
+              dbeta_n_(dbeta_n),
+              dx_m_n_(dx_m_n),
+              lengths_(lengths)
+        {
+        }
+
+        const Tensor<DYDataType>& dy_m_n_;
+        const Tensor<XDataType>& x_m_n_;
+        const Tensor<GammaDataType>& gamma_n_;
+        const Tensor<MeanInvStdDataType>& mean_m_;
+        const Tensor<MeanInvStdDataType>& inv_std_m_;
+        Tensor<DGammaDataType>& dgamma_n_;
+        Tensor<DBetaDataType>& dbeta_n_;
+        Tensor<DXDataType>& dx_m_n_;
+        std::vector<index_t> lengths_;
+    };
+
+    // Invoker
+    struct Invoker : public device::BaseInvoker
+    {
+        float Run(const Argument& arg)
+        {
+            int M = arg.lengths_[0];
+            int N = arg.lengths_[1];
+
+            // Calculate dgamma and dbeta
+            for(int n = 0; n < N; ++n)
+            {
+                ComputeDataType dgamma = 0;
+                ComputeDataType dbeta  = 0;
+
+                for(int m = 0; m < M; ++m)
+                {
+                    ComputeDataType dy   = ck::type_convert<ComputeDataType>(arg.dy_m_n_(m, n));
+                    ComputeDataType x    = ck::type_convert<ComputeDataType>(arg.x_m_n_(m, n));
+                    ComputeDataType mean = ck::type_convert<ComputeDataType>(arg.mean_m_(m));
+                    ComputeDataType rstd = ck::type_convert<ComputeDataType>(arg.inv_std_m_(m));
+                    dgamma += dy * rstd * (x - mean);
+                    dbeta += dy;
+                }
+                arg.dgamma_n_(n) = ck::type_convert<DGammaDataType>(dgamma);
+                arg.dbeta_n_(n)  = ck::type_convert<DBetaDataType>(dbeta);
+            }
+
+            // Calculate dx
+            for(int m = 0; m < M; ++m)
+            {
+                ComputeDataType ds = 0;
+                ComputeDataType db = 0;
+
+                ComputeDataType mean = ck::type_convert<ComputeDataType>(arg.mean_m_(m));
+                ComputeDataType rstd = ck::type_convert<ComputeDataType>(arg.inv_std_m_(m));
+
+                for(int n = 0; n < N; ++n)
+                {
+                    ComputeDataType dy    = ck::type_convert<ComputeDataType>(arg.dy_m_n_(m, n));
+                    ComputeDataType x     = ck::type_convert<ComputeDataType>(arg.x_m_n_(m, n));
+                    ComputeDataType gamma = ck::type_convert<ComputeDataType>(arg.gamma_n_(n));
+
+                    ds += dy * gamma * x;
+                    db += dy * gamma;
+                }
+
+                for(int n = 0; n < N; ++n)
+                {
+                    ComputeDataType dy    = ck::type_convert<ComputeDataType>(arg.dy_m_n_(m, n));
+                    ComputeDataType x     = ck::type_convert<ComputeDataType>(arg.x_m_n_(m, n));
+                    ComputeDataType gamma = ck::type_convert<ComputeDataType>(arg.gamma_n_(n));
+
+                    ComputeDataType b = (db * mean - ds) * rstd * rstd * rstd / N;
+                    ComputeDataType c = -b * mean - db * rstd / N;
+
+                    arg.dx_m_n_(m, n) = ck::type_convert<DXDataType>(dy * gamma * rstd + b * x + c);
+                }
+            }
+
+            return 0;
+        }
+
+        float Run(const device::BaseArgument* p_arg,
+                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
+
+    static auto MakeArgument(const Tensor<DYDataType>& dy_m_n,
+                             const Tensor<XDataType>& x_m_n,
+                             const Tensor<GammaDataType>& gamma_n,
+                             const Tensor<MeanInvStdDataType>& mean_m,
+                             const Tensor<MeanInvStdDataType>& inv_std_m,
+                             Tensor<DGammaDataType>& dgamma_n,
+                             Tensor<DBetaDataType>& dbeta_n,
+                             Tensor<DXDataType>& dx_m_n,
+                             const std::vector<index_t> lengths)
+    {
+        return Argument{
+            dy_m_n, x_m_n, gamma_n, mean_m, inv_std_m, dgamma_n, dbeta_n, dx_m_n, lengths};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferenceLayernormBwd"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp

@@ -25,6 +25,8 @@ using BF8  = ck::bf8_t;
 
 using Empty_Tuple = ck::Tuple<>;
 
+using BF16_Tuple = ck::Tuple<BF16>;
+
 using F16_Tuple     = ck::Tuple<F16>;
 using F16_F16_Tuple = ck::Tuple<F16, F16>;
 

library/include/ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+#pragma once
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16  = ck::half_t;
+using BF16 = ck::bhalf_t;
+using F32  = float;
+using F64  = double;
+
+using F16_Tuple   = ck::Tuple<F16>;
+using BF16_Tuple  = ck::Tuple<BF16>;
+using F32_Tuple   = ck::Tuple<F32>;
+using F64_Tuple   = ck::Tuple<F64>;
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_kk_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         Compute|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            Data|
+        //#####################################|        |        |        |           |          |            |                 |           |          |      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            Type|
+        //#####################################|        |        |        |           |          |            |                 |           |          |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4, ComputeDataType>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_kn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         Compute|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            Data|
+        //#####################################|        |        |        |           |          |            |                 |           |          |      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            Type|
+        //#####################################|        |        |        |           |          |            |                 |           |          |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   1,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   1,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   1,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_mk_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         Compute|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            Data|
+        //#####################################|        |        |        |           |          |            |                 |           |          |      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            Type|
+        //#####################################|        |        |        |           |          |            |                 |           |          |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   4,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   4,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   4,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_mn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         Compute|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            Data|
+        //#####################################|        |        |        |           |          |            |                 |           |          |      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            Type|
+        //#####################################|        |        |        |           |          |            |                 |           |          |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   1,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4, ComputeDataType>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_f64_kk_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         Compute|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            Data|
+        //#####################################|        |        |        |           |          |            |                 |           |          |      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            Type|
+        //#####################################|        |        |        |           |          |            |                 |           |          |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    32,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    32,    64,    16,   2,   2,   16,   16,    2,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1, ComputeDataType>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_f64_kn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         Compute|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            Data|
+        //#####################################|        |        |        |           |          |            |                 |           |          |      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            Type|
+        //#####################################|        |        |        |           |          |            |                 |           |          |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   1,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   1,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_f64_mk_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         Compute|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            Data|
+        //#####################################|        |        |        |           |          |            |                 |           |          |      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            Type|
+        //#####################################|        |        |        |           |          |            |                 |           |          |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,              1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,              1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,              1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,              1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,              1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,              1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   2,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,              1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,              1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   2,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,              1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,              1, ComputeDataType>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_f64_mn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         Compute|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            Data|
+        //#####################################|        |        |        |           |          |            |                 |           |          |      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            Type|
+        //#####################################|        |        |        |           |          |            |                 |           |          |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1, ComputeDataType>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

@@ -17,7 +17,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 #ifdef CK_ENABLE_FP32
-// float
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -28,7 +27,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F32>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -40,7 +40,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F32>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -52,7 +53,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F32>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -64,10 +66,115 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
-#endif
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances);
+#endif // CK_ENABLE_FP32
+
 #ifdef CK_ENABLE_FP64
-// double
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -78,7 +185,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F64>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -90,7 +198,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F64>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -102,7 +211,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F64>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -114,8 +224,170 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
-#endif
+                                                           Bilinear,
+                                                           F64>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP64
+
+#ifdef CK_ENABLE_FP16
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP16
+
+#ifdef CK_ENABLE_BF16
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP16
+
 // Contraction + Bilinear
 template <index_t NumDimM,
           index_t NumDimN,
@@ -123,7 +395,8 @@ template <index_t NumDimM,
           typename ADataType,
           typename BDataType,
           typename DDataType,
-          typename EDataType>
+          typename EDataType,
+          typename ComputeDataType>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContractionMultipleD<
     NumDimM,
     NumDimN,
@@ -134,7 +407,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     EDataType,
     ck::tensor_operation::element_wise::PassThrough,
     ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::Bilinear>>
+    ck::tensor_operation::element_wise::Bilinear,
+    ComputeDataType>>
 {
     using DeviceOp = DeviceContractionMultipleD<NumDimM,
                                                 NumDimN,
@@ -145,45 +419,125 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
                                                 EDataType,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::Bilinear>;
+                                                ck::tensor_operation::element_wise::Bilinear,
+                                                ComputeDataType>;
 
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 #ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
-                     is_same_v<DDataType, float> && is_same_v<EDataType, float>)
+                     is_same_v<EDataType, float>)
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance(
-                    op_ptrs);
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, ck::half_t>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, ck::bhalf_t>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance(
+                        op_ptrs);
+                }
             }
         }
-#endif
+#endif // CK_ENABLE_FP32
 #ifdef CK_ENABLE_FP64
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
-                     is_same_v<DDataType, double> && is_same_v<EDataType, double>)
+                     is_same_v<EDataType, double>)
+        {
+            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
+            {
+                if constexpr(is_same_v<ComputeDataType, double>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance(
+                        op_ptrs);
+                }
+            }
+        }
+#endif // CK_ENABLE_FP64
+#ifdef CK_ENABLE_FP16
+        if constexpr(is_same_v<ADataType, ck::half_t> && is_same_v<BDataType, ck::half_t> &&
+                     is_same_v<EDataType, ck::half_t>)
+        {
+            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
+            {
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance(
+                        op_ptrs);
+                }
+            }
+        }
+#endif // CK_ENABLE_FP16
+#ifdef CK_ENABLE_BF16
+        if constexpr(is_same_v<ADataType, ck::bhalf_t> && is_same_v<BDataType, ck::bhalf_t> &&
+                     is_same_v<EDataType, ck::bhalf_t>)
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
-                    op_ptrs);
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance(
+                        op_ptrs);
+                }
             }
         }
-#endif
+#endif // CK_ENABLE_BF16
         return op_ptrs;
     }
 };

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

@@ -17,7 +17,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 #ifdef CK_ENABLE_FP32
-// float
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -28,7 +27,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F32>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -40,7 +40,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F32>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -52,7 +53,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F32>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -64,10 +66,115 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
-#endif
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances);
+#endif // CK_ENABLE_FP32
+
 #ifdef CK_ENABLE_FP64
-// double
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -78,7 +185,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F64>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -90,7 +198,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F64>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -102,7 +211,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F64>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -114,15 +224,178 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
-#endif
+                                                           Scale,
+                                                           F64>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP64
+
+#ifdef CK_ENABLE_FP16
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP16
+
+#ifdef CK_ENABLE_BF16
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP16
+
 // Contraction + Scale
 template <index_t NumDimM,
           index_t NumDimN,
           index_t NumDimK,
           typename ADataType,
           typename BDataType,
-          typename EDataType>
+          typename EDataType,
+          typename ComputeDataType>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContractionMultipleD<
     NumDimM,
     NumDimN,
@@ -133,7 +406,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     EDataType,
     ck::tensor_operation::element_wise::PassThrough,
     ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::Scale>>
+    ck::tensor_operation::element_wise::Scale,
+    ComputeDataType>>
 {
     using DeviceOp = DeviceContractionMultipleD<NumDimM,
                                                 NumDimN,
@@ -144,7 +418,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
                                                 EDataType,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::Scale>;
+                                                ck::tensor_operation::element_wise::Scale,
+                                                ComputeDataType>;
 
     static auto GetInstances()
     {
@@ -155,34 +430,113 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance(
-                    op_ptrs);
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, ck::half_t>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, ck::bhalf_t>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance(
+                        op_ptrs);
+                }
             }
         }
-#endif
+#endif // CK_ENABLE_FP32
 #ifdef CK_ENABLE_FP64
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
                      is_same_v<EDataType, double>)
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
-                    op_ptrs);
+                if constexpr(is_same_v<ComputeDataType, double>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance(
+                        op_ptrs);
+                }
+            }
+        }
+#endif // CK_ENABLE_FP64
+#ifdef CK_ENABLE_FP16
+        if constexpr(is_same_v<ADataType, ck::half_t> && is_same_v<BDataType, ck::half_t> &&
+                     is_same_v<EDataType, ck::half_t>)
+        {
+            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
+            {
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance(
+                        op_ptrs);
+                }
+            }
+        }
+#endif // CK_ENABLE_FP16
+#ifdef CK_ENABLE_BF16
+        if constexpr(is_same_v<ADataType, ck::bhalf_t> && is_same_v<BDataType, ck::bhalf_t> &&
+                     is_same_v<EDataType, ck::bhalf_t>)
+        {
+            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
+            {
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance(
+                        op_ptrs);
+                }
             }
         }
-#endif
+#endif // CK_ENABLE_BF16
         return op_ptrs;
     }
 };

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

@@ -19,109 +19,214 @@ namespace instance {
 
 using namespace ck::conv_tensor_rearrange_op;
 
+// GNWC/GNHWC/GNDHWC
 // Image to Column
-// nhwc, 1d
-void add_device_image_to_column_nwc_1d_bf16_instances(
+// GNWC, 1d
+void add_device_image_to_column_gnwc_1d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nwc_1d_f16_instances(
+void add_device_image_to_column_gnwc_1d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nwc_1d_f32_instances(
+void add_device_image_to_column_gnwc_1d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nwc_1d_i8_instances(
+void add_device_image_to_column_gnwc_1d_i8_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ImageToColumn>>>&
         instances);
-// nhwc, 2d
-void add_device_image_to_column_nhwc_2d_bf16_instances(
+// GNHWC, 2d
+void add_device_image_to_column_gnhwc_2d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nhwc_2d_f16_instances(
+void add_device_image_to_column_gnhwc_2d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nhwc_2d_f32_instances(
+void add_device_image_to_column_gnhwc_2d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nhwc_2d_i8_instances(
+void add_device_image_to_column_gnhwc_2d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ImageToColumn>>>&
         instances);
-// nhwc, 3d
-void add_device_image_to_column_ndhwc_3d_bf16_instances(
+// GNDHWC, 3d
+void add_device_image_to_column_gndhwc_3d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_ndhwc_3d_f16_instances(
+void add_device_image_to_column_gndhwc_3d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_ndhwc_3d_f32_instances(
+void add_device_image_to_column_gndhwc_3d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_ndhwc_3d_i8_instances(
+void add_device_image_to_column_gndhwc_3d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ImageToColumn>>>&
         instances);
 
 // Column to Image
-// nhwc, 1d
-void add_device_column_to_image_nwc_1d_bf16_instances(
+// GNWC, 1d
+void add_device_column_to_image_gnwc_1d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nwc_1d_f16_instances(
+void add_device_column_to_image_gnwc_1d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nwc_1d_f32_instances(
+void add_device_column_to_image_gnwc_1d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nwc_1d_i8_instances(
+void add_device_column_to_image_gnwc_1d_i8_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ColumnToImage>>>&
         instances);
-// nhwc, 2d
-void add_device_column_to_image_nhwc_2d_bf16_instances(
+// GNHWC, 2d
+void add_device_column_to_image_gnhwc_2d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nhwc_2d_f16_instances(
+void add_device_column_to_image_gnhwc_2d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nhwc_2d_f32_instances(
+void add_device_column_to_image_gnhwc_2d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nhwc_2d_i8_instances(
+void add_device_column_to_image_gnhwc_2d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ColumnToImage>>>&
         instances);
-// nhwc, 3d
-void add_device_column_to_image_ndhwc_3d_bf16_instances(
+// GNDHWC, 3d
+void add_device_column_to_image_gndhwc_3d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_ndhwc_3d_f16_instances(
+void add_device_column_to_image_gndhwc_3d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_ndhwc_3d_f32_instances(
+void add_device_column_to_image_gndhwc_3d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_ndhwc_3d_i8_instances(
+void add_device_column_to_image_gndhwc_3d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ColumnToImage>>>&
         instances);
+// NWGC/NHWGC/NDHWGC
+// Image to Column
+// NWGC, 1d
+void add_device_image_to_column_nwgc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwgc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwgc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwgc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+// NHWGC, 2d
+void add_device_image_to_column_nhwgc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwgc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwgc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwgc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+// NDHWGC, 3d
+void add_device_image_to_column_ndhwgc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwgc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwgc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwgc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+
+// Column to Image
+// NWGC, 1d
+void add_device_column_to_image_nwgc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwgc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwgc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwgc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
+// NHWGC, 2d
+void add_device_column_to_image_nhwgc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwgc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwgc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwgc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
+// NDHWGC, 3d
+void add_device_column_to_image_ndhwgc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwgc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwgc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwgc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
 
 template <ck::index_t NumDimSpatial,
           typename ImageLayout,
@@ -151,60 +256,120 @@ struct DeviceOperationInstanceFactory<
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_image_to_column_nwc_1d_f32_instances(op_ptrs);
+                    add_device_image_to_column_gnwc_1d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_image_to_column_nwc_1d_f16_instances(op_ptrs);
+                    add_device_image_to_column_gnwc_1d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_image_to_column_nwc_1d_bf16_instances(op_ptrs);
+                    add_device_image_to_column_gnwc_1d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_image_to_column_nwc_1d_i8_instances(op_ptrs);
+                    add_device_image_to_column_gnwc_1d_i8_instances(op_ptrs);
                 }
             }
             else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, GNHWC>)
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_image_to_column_nhwc_2d_f32_instances(op_ptrs);
+                    add_device_image_to_column_gnhwc_2d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_image_to_column_nhwc_2d_f16_instances(op_ptrs);
+                    add_device_image_to_column_gnhwc_2d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_image_to_column_nhwc_2d_bf16_instances(op_ptrs);
+                    add_device_image_to_column_gnhwc_2d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_image_to_column_nhwc_2d_i8_instances(op_ptrs);
+                    add_device_image_to_column_gnhwc_2d_i8_instances(op_ptrs);
                 }
             }
             else if constexpr(NumDimSpatial == 3 && is_same_v<ImageLayout, GNDHWC>)
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_image_to_column_ndhwc_3d_f32_instances(op_ptrs);
+                    add_device_image_to_column_gndhwc_3d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_image_to_column_ndhwc_3d_f16_instances(op_ptrs);
+                    add_device_image_to_column_gndhwc_3d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_image_to_column_ndhwc_3d_bf16_instances(op_ptrs);
+                    add_device_image_to_column_gndhwc_3d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_image_to_column_ndhwc_3d_i8_instances(op_ptrs);
+                    add_device_image_to_column_gndhwc_3d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 1 && is_same_v<ImageLayout, NWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_nwgc_1d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_image_to_column_nwgc_1d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_image_to_column_nwgc_1d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_image_to_column_nwgc_1d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, NHWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_nhwgc_2d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_image_to_column_nhwgc_2d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_image_to_column_nhwgc_2d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_image_to_column_nhwgc_2d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 3 && is_same_v<ImageLayout, NDHWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_ndhwgc_3d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_image_to_column_ndhwgc_3d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_image_to_column_ndhwgc_3d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_image_to_column_ndhwgc_3d_i8_instances(op_ptrs);
                 }
             }
         }
@@ -214,60 +379,120 @@ struct DeviceOperationInstanceFactory<
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_column_to_image_nwc_1d_f32_instances(op_ptrs);
+                    add_device_column_to_image_gnwc_1d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_column_to_image_nwc_1d_f16_instances(op_ptrs);
+                    add_device_column_to_image_gnwc_1d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_column_to_image_nwc_1d_bf16_instances(op_ptrs);
+                    add_device_column_to_image_gnwc_1d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_column_to_image_nwc_1d_i8_instances(op_ptrs);
+                    add_device_column_to_image_gnwc_1d_i8_instances(op_ptrs);
                 }
             }
             else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, GNHWC>)
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_column_to_image_nhwc_2d_f32_instances(op_ptrs);
+                    add_device_column_to_image_gnhwc_2d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_column_to_image_nhwc_2d_f16_instances(op_ptrs);
+                    add_device_column_to_image_gnhwc_2d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_column_to_image_nhwc_2d_bf16_instances(op_ptrs);
+                    add_device_column_to_image_gnhwc_2d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_column_to_image_nhwc_2d_i8_instances(op_ptrs);
+                    add_device_column_to_image_gnhwc_2d_i8_instances(op_ptrs);
                 }
             }
             else if constexpr(NumDimSpatial == 3 && is_same_v<ImageLayout, GNDHWC>)
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_column_to_image_ndhwc_3d_f32_instances(op_ptrs);
+                    add_device_column_to_image_gndhwc_3d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_gndhwc_3d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_gndhwc_3d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_gndhwc_3d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 1 && is_same_v<ImageLayout, NWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_nwgc_1d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_nwgc_1d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_nwgc_1d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_nwgc_1d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, NHWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_nhwgc_2d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_nhwgc_2d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_nhwgc_2d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_nhwgc_2d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 3 && is_same_v<ImageLayout, NDHWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_ndhwgc_3d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_column_to_image_ndhwc_3d_f16_instances(op_ptrs);
+                    add_device_column_to_image_ndhwgc_3d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_column_to_image_ndhwc_3d_bf16_instances(op_ptrs);
+                    add_device_column_to_image_ndhwgc_3d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_column_to_image_ndhwc_3d_i8_instances(op_ptrs);
+                    add_device_column_to_image_ndhwgc_3d_i8_instances(op_ptrs);
                 }
             }
         }

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

@@ -328,7 +328,18 @@ void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances(
 void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
+
+void add_device_gemm_xdl_c_shuffle_f16_f8_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_c_shuffle_f16_f8_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
 #endif
+
 template <typename ALayout,
           typename BLayout,
           typename CLayout,
@@ -548,6 +559,20 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances(op_ptrs);
             }
         }
+        else if constexpr(is_same_v<ADataType, ck::half_t> && is_same_v<BDataType, ck::f8_t> &&
+                          is_same_v<CDataType, ck::half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_c_shuffle_f16_f8_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_c_shuffle_f16_f8_f16_mk_nk_mn_instances(op_ptrs);
+            }
+        }
 #endif
         return op_ptrs;
     }

library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp

@@ -3,7 +3,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
@@ -55,24 +55,24 @@ using device_grouped_conv_fwd_xdl_bf16_instances = std::tuple<
         //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
         // generic instance
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
         // instances for small conv.K and conv.C
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
     // clang-format on
     >;
 
@@ -89,24 +89,24 @@ using device_grouped_conv_fwd_xdl_f16_instances = std::tuple<
         //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
         // generic instance
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
         // instances for small conv.K and conv.C
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
     // clang-format on
     >;
 
@@ -123,24 +123,24 @@ using device_grouped_conv_fwd_xdl_f32_instances = std::tuple<
         //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
         // generic instance
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
         // instances for small conv.K and conv.C
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
     // clang-format on
     >;
 
@@ -157,24 +157,24 @@ using device_grouped_conv_fwd_xdl_int8_instances = std::tuple<
         //########################################|           |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|           |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
         // generic instance
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
         // instances for small conv.K and conv.C
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
     // clang-format on
     >;
 
@@ -192,24 +192,24 @@ using device_grouped_conv_fwd_xdl_f16_comp_f8_instances = std::tuple<
         //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |            |
 #ifdef CK_ENABLE_FP8
         // generic instance
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1,         F8>,
         // instances for small conv.K and conv.C
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
-
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>
 #endif
     // clang-format on
     >;

library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_scaleadd_ab_instance.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using ScaleAdd    = ck::tensor_operation::element_wise::ScaleAdd;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+static constexpr auto ConvFwd1x1P0   = ConvolutionForwardSpecialization::Filter1x1Pad0;
+static constexpr auto ConvFwd1x1S1P0 = ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+static constexpr auto ConvFwdOddC =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
+
+static constexpr auto GemmMNKPadding = GemmSpecialization::MNKPadding;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_scaleadd_ab_bf16_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E|                  AData|              BData| AccData| CShuffle|       Ds| EData|           A|           B|             CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|                   Type|               Type|    Type| DataType| DataType|  Type| Elementwise| Elementwise|     Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |                       |                   |        |         |         |      |   Operation|   Operation|       Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |                       |                   |        |         |         |      |            |            |                |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,  ck::Tuple<BF16,  BF16>, ck::Tuple<BF16,  BF16>,     F32, BF16, ck::Tuple<>, BF16,  ScaleAdd, ScaleAdd, PassThrough,         ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,  ck::Tuple<BF16,  BF16>, ck::Tuple<BF16,  BF16>,     F32, BF16, ck::Tuple<>, BF16,  ScaleAdd, ScaleAdd, PassThrough,         ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,  ck::Tuple<BF16,  BF16>, ck::Tuple<BF16,  BF16>,     F32, BF16, ck::Tuple<>, BF16,  ScaleAdd, ScaleAdd, PassThrough,         ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,  ck::Tuple<BF16,  BF16>, ck::Tuple<BF16,  BF16>,     F32, BF16, ck::Tuple<>, BF16,  ScaleAdd, ScaleAdd, PassThrough,         ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_scaleadd_ab_f16_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E|                  AData|              BData| AccData| CShuffle|       Ds| EData|           A|           B|             CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|                   Type|               Type|    Type| DataType| DataType|  Type| Elementwise| Elementwise|     Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |                       |                   |        |         |         |      |   Operation|   Operation|       Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |                       |                   |        |         |         |      |            |            |                |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,   ck::Tuple<F16,  F16>, ck::Tuple<F16,  F16>,     F32, F16, ck::Tuple<>, F16,      ScaleAdd, ScaleAdd, PassThrough,         ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,   ck::Tuple<F16,  F16>, ck::Tuple<F16,  F16>,     F32, F16, ck::Tuple<>, F16,      ScaleAdd, ScaleAdd, PassThrough,         ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,   ck::Tuple<F16,  F16>, ck::Tuple<F16,  F16>,     F32, F16, ck::Tuple<>, F16,      ScaleAdd, ScaleAdd, PassThrough,         ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,   ck::Tuple<F16,  F16>, ck::Tuple<F16,  F16>,     F32, F16, ck::Tuple<>, F16,      ScaleAdd, ScaleAdd, PassThrough,         ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_scaleadd_ab_f32_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E|                  AData|              BData| AccData| CShuffle|       Ds| EData|           A|           B|             CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|                   Type|               Type|    Type| DataType| DataType|  Type| Elementwise| Elementwise|     Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |                       |                   |        |         |         |      |   Operation|   Operation|       Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |                       |                   |        |         |         |      |            |            |                |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,   ck::Tuple<F32,  F32>, ck::Tuple<F32,  F32>,     F32, F32, ck::Tuple<>, F32,      ScaleAdd,   ScaleAdd, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
+        // instances for small conv.K and conv.C        
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,   ck::Tuple<F32,  F32>, ck::Tuple<F32,  F32>,     F32, F32, ck::Tuple<>, F32,      ScaleAdd,   ScaleAdd, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,   ck::Tuple<F32,  F32>, ck::Tuple<F32,  F32>,     F32, F32, ck::Tuple<>, F32,      ScaleAdd,   ScaleAdd, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout,   ck::Tuple<F32,  F32>, ck::Tuple<F32,  F32>,     F32, F32, ck::Tuple<>, F32,      ScaleAdd,   ScaleAdd, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_scaleadd_ab_int8_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E|                  AData|              BData| AccData| CShuffle|       Ds| EData|           A|           B|             CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|                   Type|               Type|    Type| DataType| DataType|  Type| Elementwise| Elementwise|     Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |                       |                   |        |         |         |      |   Operation|   Operation|       Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |                       |                   |        |         |         |      |            |            |                |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout, ck::Tuple<int8_t,  int8_t>, ck::Tuple<int8_t,  int8_t>, int32_t, int8_t, ck::Tuple<>, int8_t,   ScaleAdd, ScaleAdd,   PassThrough,      ConvSpec,      GemmMNKPadding, 1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,     2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout, ck::Tuple<int8_t,  int8_t>, ck::Tuple<int8_t,  int8_t>, int32_t, int8_t, ck::Tuple<>, int8_t,   ScaleAdd, ScaleAdd,   PassThrough,      ConvSpec,      GemmMNKPadding, 1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,     2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout, ck::Tuple<int8_t,  int8_t>, ck::Tuple<int8_t,  int8_t>, int32_t, int8_t, ck::Tuple<>, int8_t,   ScaleAdd, ScaleAdd,   PassThrough,      ConvSpec,      GemmMNKPadding, 1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,     2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, ck::Tuple<>,ELayout, ck::Tuple<int8_t,  int8_t>, ck::Tuple<int8_t,  int8_t>, int32_t, int8_t, ck::Tuple<>, int8_t,   ScaleAdd, ScaleAdd,   PassThrough,      ConvSpec,       GemmMNKPadding, 1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,    2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_scaleadd_scaleadd_relu_instance.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough          = ck::tensor_operation::element_wise::PassThrough;
+using ScaleAddScaleAddRelu = ck::tensor_operation::element_wise::ScaleAddScaleAddRelu;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 = ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 = ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto ConvFwdOddC =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
+
+static constexpr auto GemmMNKPadding = GemmSpecialization::MNKPadding;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_scaleadd_scaleadd_relu_bf16_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|             CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise|     Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|       Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |                |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32, BF16, ck::Tuple<BF16, BF16>, BF16, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32, BF16, ck::Tuple<BF16, BF16>, BF16, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32, BF16, ck::Tuple<BF16, BF16>, BF16, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32, BF16, ck::Tuple<BF16, BF16>, BF16, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_scaleadd_scaleadd_relu_f16_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|             CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise|     Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|       Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |                |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32, F16, ck::Tuple<F16, F16>, F16, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32, F16, ck::Tuple<F16, F16>, F16, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32, F16, ck::Tuple<F16, F16>, F16, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32, F16, ck::Tuple<F16, F16>, F16, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_scaleadd_scaleadd_relu_f32_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|             CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise|     Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|       Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |                |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32, F32, ck::Tuple<F32, F32>, F32, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32, F32, ck::Tuple<F32, F32>, F32, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32, F32, ck::Tuple<F32, F32>, F32, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32, F32, ck::Tuple<F32, F32>, F32, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_scaleadd_scaleadd_relu_int8_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|                 Ds| EData|           A|           B|             CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|           DataType|  Type| Elementwise| Elementwise|     Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |      |      |        |         |                   |      |   Operation|   Operation|       Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |      |      |        |         |                   |      |            |            |                |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, ck::Tuple<F32, F32>, int8_t, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, ck::Tuple<F32, F32>, int8_t, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, ck::Tuple<F32, F32>, int8_t, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, ck::Tuple<F32, F32>, int8_t, PassThrough, PassThrough, ScaleAddScaleAddRelu,  ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

@@ -496,7 +496,8 @@ struct DeviceOperationInstanceFactory<
             {
 #ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
-                             is_same_v<OutDataType, F16>)
+                             is_same_v<OutDataType, F16> && is_same_v<ComputeTypeA, F16> &&
+                             is_same_v<ComputeTypeB, F16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(op_ptrs);
                     add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_instances(
@@ -507,14 +508,16 @@ struct DeviceOperationInstanceFactory<
 #endif
 #ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
-                                  is_same_v<OutDataType, F32>)
+                                  is_same_v<OutDataType, F32> && is_same_v<ComputeTypeA, F32> &&
+                                  is_same_v<ComputeTypeB, F32>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
-                                  is_same_v<OutDataType, BF16>)
+                                  is_same_v<OutDataType, BF16> && is_same_v<ComputeTypeA, BF16> &&
+                                  is_same_v<ComputeTypeB, BF16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
                         op_ptrs);
@@ -522,7 +525,9 @@ struct DeviceOperationInstanceFactory<
 #endif
 #ifdef CK_ENABLE_INT8
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                                  is_same_v<OutDataType, int8_t>)
+                                  is_same_v<OutDataType, int8_t> &&
+                                  is_same_v<ComputeTypeA, int8_t> &&
+                                  is_same_v<ComputeTypeB, int8_t>)
                 {
                     add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_instances(op_ptrs);
                     add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_1x1s1p0_instances(
@@ -535,7 +540,8 @@ struct DeviceOperationInstanceFactory<
             {
 #ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
-                             is_same_v<OutDataType, F16>)
+                             is_same_v<OutDataType, F16> && is_same_v<ComputeTypeA, F16> &&
+                             is_same_v<ComputeTypeB, F16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(op_ptrs);
                     add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_instances(
@@ -546,14 +552,16 @@ struct DeviceOperationInstanceFactory<
 #endif
 #ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
-                                  is_same_v<OutDataType, F32>)
+                                  is_same_v<OutDataType, F32> && is_same_v<ComputeTypeA, F32> &&
+                                  is_same_v<ComputeTypeB, F32>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
-                                  is_same_v<OutDataType, BF16>)
+                                  is_same_v<OutDataType, BF16> && is_same_v<ComputeTypeA, BF16> &&
+                                  is_same_v<ComputeTypeB, BF16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
                         op_ptrs);
@@ -561,7 +569,9 @@ struct DeviceOperationInstanceFactory<
 #endif
 #ifdef CK_ENABLE_INT8
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                                  is_same_v<OutDataType, int8_t>)
+                                  is_same_v<OutDataType, int8_t> &&
+                                  is_same_v<ComputeTypeA, int8_t> &&
+                                  is_same_v<ComputeTypeB, int8_t>)
                 {
                     add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_instances(op_ptrs);
                     add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_1x1s1p0_instances(
@@ -578,7 +588,8 @@ struct DeviceOperationInstanceFactory<
             {
 #ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
-                             is_same_v<OutDataType, F16>)
+                             is_same_v<OutDataType, F16> && is_same_v<ComputeTypeA, F16> &&
+                             is_same_v<ComputeTypeB, F16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
                         op_ptrs);
@@ -590,7 +601,8 @@ struct DeviceOperationInstanceFactory<
 #endif
 #ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
-                                  is_same_v<OutDataType, F32>)
+                                  is_same_v<OutDataType, F32> && is_same_v<ComputeTypeA, F32> &&
+                                  is_same_v<ComputeTypeB, F32>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
                         op_ptrs);
@@ -598,7 +610,8 @@ struct DeviceOperationInstanceFactory<
 #endif
 #ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
-                                  is_same_v<OutDataType, BF16>)
+                                  is_same_v<OutDataType, BF16> && is_same_v<ComputeTypeA, BF16> &&
+                                  is_same_v<ComputeTypeB, BF16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
                         op_ptrs);
@@ -606,7 +619,9 @@ struct DeviceOperationInstanceFactory<
 #endif
 #ifdef CK_ENABLE_INT8
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                                  is_same_v<OutDataType, int8_t>)
+                                  is_same_v<OutDataType, int8_t> &&
+                                  is_same_v<ComputeTypeA, int8_t> &&
+                                  is_same_v<ComputeTypeB, int8_t>)
                 {
                     add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_i8_instances(
                         op_ptrs);
@@ -642,7 +657,8 @@ struct DeviceOperationInstanceFactory<
 #endif
 #ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
-                                  is_same_v<OutDataType, F32>)
+                                  is_same_v<OutDataType, F32> && is_same_v<ComputeTypeA, F32> &&
+                                  is_same_v<ComputeTypeB, F32>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
                         op_ptrs);
@@ -650,7 +666,8 @@ struct DeviceOperationInstanceFactory<
 #endif
 #ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
-                                  is_same_v<OutDataType, BF16>)
+                                  is_same_v<OutDataType, BF16> && is_same_v<ComputeTypeA, BF16> &&
+                                  is_same_v<ComputeTypeB, BF16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
                         op_ptrs);
@@ -658,7 +675,9 @@ struct DeviceOperationInstanceFactory<
 #endif
 #ifdef CK_ENABLE_INT8
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                                  is_same_v<OutDataType, int8_t>)
+                                  is_same_v<OutDataType, int8_t> &&
+                                  is_same_v<ComputeTypeA, int8_t> &&
+                                  is_same_v<ComputeTypeB, int8_t>)
                 {
                     add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_instances(
                         op_ptrs);

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

@@ -618,7 +618,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
 #ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, float>)
+                             is_same_v<OutDataType, float> && is_same_v<ComputeTypeA, float> &&
+                             is_same_v<ComputeTypeB, float>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
@@ -628,7 +629,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                             is_same_v<OutDataType, half_t>)
+                             is_same_v<OutDataType, half_t> && is_same_v<ComputeTypeA, half_t> &&
+                             is_same_v<ComputeTypeB, half_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
@@ -638,7 +640,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_BF16
                 if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, ck::bhalf_t>)
+                             is_same_v<OutDataType, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeA, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeB, ck::bhalf_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
@@ -655,21 +659,25 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #ifdef DL_KERNELS
 #ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, float>)
+                             is_same_v<OutDataType, float> && is_same_v<ComputeTypeA, float> &&
+                             is_same_v<ComputeTypeB, float>)
                 {
                     add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instances(op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                             is_same_v<OutDataType, half_t>)
+                             is_same_v<OutDataType, half_t> && is_same_v<ComputeTypeA, half_t> &&
+                             is_same_v<ComputeTypeB, half_t>)
                 {
                     add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instances(op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_BF16
                 if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, ck::bhalf_t>)
+                             is_same_v<OutDataType, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeA, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeB, ck::bhalf_t>)
                 {
                     add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_f32_bf16_instances(
                         op_ptrs);
@@ -685,7 +693,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
 #ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, float>)
+                             is_same_v<OutDataType, float> && is_same_v<ComputeTypeA, float> &&
+                             is_same_v<ComputeTypeB, float>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instances(
@@ -697,7 +706,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                             is_same_v<OutDataType, half_t>)
+                             is_same_v<OutDataType, half_t> && is_same_v<ComputeTypeA, half_t> &&
+                             is_same_v<ComputeTypeB, half_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instances(
@@ -709,7 +719,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_BF16
                 if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, ck::bhalf_t>)
+                             is_same_v<OutDataType, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeA, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeB, ck::bhalf_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
@@ -725,7 +737,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
 #ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, float>)
+                             is_same_v<OutDataType, float> && is_same_v<ComputeTypeA, float> &&
+                             is_same_v<ComputeTypeB, float>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instances(
@@ -737,7 +750,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                             is_same_v<OutDataType, half_t>)
+                             is_same_v<OutDataType, half_t> && is_same_v<ComputeTypeA, half_t> &&
+                             is_same_v<ComputeTypeB, half_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instances(
@@ -749,7 +763,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_BF16
                 if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, ck::bhalf_t>)
+                             is_same_v<OutDataType, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeA, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeB, ck::bhalf_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
@@ -768,7 +784,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
 #ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, float>)
+                             is_same_v<OutDataType, float> && is_same_v<ComputeTypeA, float> &&
+                             is_same_v<ComputeTypeB, float>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instances(
@@ -780,7 +797,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                             is_same_v<OutDataType, half_t>)
+                             is_same_v<OutDataType, half_t> && is_same_v<ComputeTypeA, half_t> &&
+                             is_same_v<ComputeTypeB, half_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instances(
@@ -796,7 +814,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_BF16
                 if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, ck::bhalf_t>)
+                             is_same_v<OutDataType, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeA, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeB, ck::bhalf_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
@@ -808,7 +828,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_INT8
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                                  is_same_v<OutDataType, int8_t>)
+                                  is_same_v<OutDataType, int8_t> &&
+                                  is_same_v<ComputeTypeA, int8_t> &&
+                                  is_same_v<ComputeTypeB, int8_t>)
                 {
                     add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_instances(
                         op_ptrs);
@@ -822,7 +844,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
 #ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, float>)
+                             is_same_v<OutDataType, float> && is_same_v<ComputeTypeA, float> &&
+                             is_same_v<ComputeTypeB, float>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
@@ -851,7 +874,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_BF16
                 if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                             is_same_v<OutDataType, ck::bhalf_t>)
+                             is_same_v<OutDataType, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeA, ck::bhalf_t> &&
+                             is_same_v<ComputeTypeB, ck::bhalf_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
@@ -863,7 +888,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_INT8
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                                  is_same_v<OutDataType, int8_t>)
+                                  is_same_v<OutDataType, int8_t> &&
+                                  is_same_v<ComputeTypeA, int8_t> &&
+                                  is_same_v<ComputeTypeB, int8_t>)
                 {
                     add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instances(
                         op_ptrs);

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

@@ -6,7 +6,7 @@
 #include <vector>
 #include <memory>
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_abd.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
@@ -20,866 +20,866 @@ namespace instance {
 #ifdef CK_ENABLE_BF16
 // grouped conv1d forward, GNWC/GKXC/GNWK
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
-                                                              GNWC,
-                                                              GKXC,
-                                                              Empty_Tuple,
-                                                              GNWK,
-                                                              BF16,
-                                                              BF16,
-                                                              Empty_Tuple,
-                                                              BF16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<1,
+                                                                GNWC,
+                                                                GKXC,
+                                                                Empty_Tuple,
+                                                                GNWK,
+                                                                BF16,
+                                                                BF16,
+                                                                Empty_Tuple,
+                                                                BF16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
-                                                              GNWC,
-                                                              GKXC,
-                                                              Empty_Tuple,
-                                                              GNWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<1,
+                                                                GNWC,
+                                                                GKXC,
+                                                                Empty_Tuple,
+                                                                GNWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
-                                                              GNWC,
-                                                              GKXC,
-                                                              Empty_Tuple,
-                                                              GNWK,
-                                                              F32,
-                                                              F32,
-                                                              Empty_Tuple,
-                                                              F32,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<1,
+                                                                GNWC,
+                                                                GKXC,
+                                                                Empty_Tuple,
+                                                                GNWK,
+                                                                F32,
+                                                                F32,
+                                                                Empty_Tuple,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
-                                                              GNWC,
-                                                              GKXC,
-                                                              Empty_Tuple,
-                                                              GNWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<1,
+                                                                GNWC,
+                                                                GKXC,
+                                                                Empty_Tuple,
+                                                                GNWK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_BF16
 // grouped conv2d forward, GNHWC/GKYXC/GNHWK
 void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              BF16,
-                                                              BF16,
-                                                              Empty_Tuple,
-                                                              BF16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                BF16,
+                                                                BF16,
+                                                                Empty_Tuple,
+                                                                BF16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F32,
-                                                              F32,
-                                                              Empty_Tuple,
-                                                              F32,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                F32,
+                                                                F32,
+                                                                Empty_Tuple,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 #ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              BF16,
-                                                              BF16,
-                                                              Empty_Tuple,
-                                                              BF16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                BF16,
+                                                                BF16,
+                                                                Empty_Tuple,
+                                                                BF16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              F32,
-                                                              F32,
-                                                              Empty_Tuple,
-                                                              F32,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                F32,
+                                                                F32,
+                                                                Empty_Tuple,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_BF16
 // grouped conv3d forward, GNDHWC/GKZYXC/GNDHWK
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              BF16,
-                                                              BF16,
-                                                              Empty_Tuple,
-                                                              BF16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                BF16,
+                                                                BF16,
+                                                                Empty_Tuple,
+                                                                BF16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              F32,
-                                                              F32,
-                                                              Empty_Tuple,
-                                                              F32,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                F32,
+                                                                F32,
+                                                                Empty_Tuple,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              GNDHWC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              GNDHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                GNDHWC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                GNDHWK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_BF16
 // grouped conv3d forward, NDHWGC/GKZYXC/NDHWGK
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              BF16,
-                                                              BF16,
-                                                              Empty_Tuple,
-                                                              BF16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                BF16,
+                                                                BF16,
+                                                                Empty_Tuple,
+                                                                BF16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP8
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_f8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              F8>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                F8>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              F32,
-                                                              F32,
-                                                              Empty_Tuple,
-                                                              F32,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                F32,
+                                                                F32,
+                                                                Empty_Tuple,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
 void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              F32,
-                                                              F32,
-                                                              Empty_Tuple,
-                                                              F32,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                F32,
+                                                                F32,
+                                                                Empty_Tuple,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 
 #endif
 
 #if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
 void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                F16,
+                                                                F16,
+                                                                Empty_Tuple,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 #if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F32,
-                                                              F32,
-                                                              Empty_Tuple,
-                                                              F32,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                GNHWC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                GNHWK,
+                                                                F32,
+                                                                F32,
+                                                                Empty_Tuple,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances);
 #endif
 
 template <ck::index_t NumDimSpatial,
@@ -890,7 +890,7 @@ template <ck::index_t NumDimSpatial,
           typename WeiDataType,
           typename OutDataType,
           typename ComputeType>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
     NumDimSpatial,
     InLayout,
     WeiLayout,
@@ -905,19 +905,20 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     ck::tensor_operation::element_wise::PassThrough,
     ComputeType>>
 {
-    using DeviceOp = DeviceGroupedConvFwdMultipleD<NumDimSpatial,
-                                                   InLayout,
-                                                   WeiLayout,
-                                                   Empty_Tuple,
-                                                   OutLayout,
-                                                   InDataType,
-                                                   WeiDataType,
-                                                   Empty_Tuple,
-                                                   OutDataType,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   ComputeType>;
+    using DeviceOp =
+        DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        Empty_Tuple,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        Empty_Tuple,
+                                        OutDataType,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        ComputeType>;
 
     static auto GetInstances()
     {
@@ -928,28 +929,29 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
         {
 #ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
+                         is_same_v<OutDataType, float> && is_same_v<ComputeType, float>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                         is_same_v<OutDataType, half_t>)
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, half_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
+                         is_same_v<WeiDataType, ck::bhalf_t> &&
+                         is_same_v<OutDataType, ck::bhalf_t> && is_same_v<ComputeType, bhalf_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_INT8
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                         is_same_v<OutDataType, int8_t>)
+                         is_same_v<OutDataType, int8_t> && is_same_v<ComputeType, int8_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(op_ptrs);
             }
@@ -961,7 +963,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
         {
 #ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
+                         is_same_v<OutDataType, float> && is_same_v<ComputeType, float>)
             {
                 add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
             }
@@ -969,7 +971,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 
 #if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
+                         is_same_v<OutDataType, float> && is_same_v<ComputeType, float>)
             {
                 add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
             }
@@ -977,7 +979,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                         is_same_v<OutDataType, half_t>)
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, half_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
@@ -989,7 +991,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 
 #if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                         is_same_v<OutDataType, half_t>)
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, half_t>)
             {
                 add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
             }
@@ -997,7 +999,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 
 #ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
+                         is_same_v<WeiDataType, ck::bhalf_t> &&
+                         is_same_v<OutDataType, ck::bhalf_t> && is_same_v<ComputeType, bhalf_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
             }
@@ -1005,7 +1008,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 
 #ifdef CK_ENABLE_INT8
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                         is_same_v<OutDataType, int8_t>)
+                         is_same_v<OutDataType, int8_t> && is_same_v<ComputeType, int8_t>)
             {
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(op_ptrs);
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instances(op_ptrs);
@@ -1021,7 +1024,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 
 #ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
+                         is_same_v<OutDataType, float> && is_same_v<ComputeType, float>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
             }
@@ -1029,7 +1032,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 
 #if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
+                         is_same_v<OutDataType, float> && is_same_v<ComputeType, float>)
             {
                 add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
             }
@@ -1037,7 +1040,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                         is_same_v<OutDataType, half_t>)
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, half_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
             }
@@ -1045,7 +1048,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 
 #if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                         is_same_v<OutDataType, half_t>)
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, half_t>)
             {
                 add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
             }
@@ -1053,14 +1056,15 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 
 #ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
+                         is_same_v<WeiDataType, ck::bhalf_t> &&
+                         is_same_v<OutDataType, ck::bhalf_t> && is_same_v<ComputeType, bhalf_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_INT8
             else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
+                              is_same_v<OutDataType, int8_t> && is_same_v<ComputeType, int8_t>)
             {
                 add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instances(op_ptrs);
                 add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instances(op_ptrs);
@@ -1075,14 +1079,14 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
         {
 #ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
+                         is_same_v<OutDataType, float> && is_same_v<ComputeType, float>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                         is_same_v<OutDataType, half_t>)
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, half_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(op_ptrs);
                 add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instances(op_ptrs);
@@ -1095,14 +1099,15 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
+                         is_same_v<WeiDataType, ck::bhalf_t> &&
+                         is_same_v<OutDataType, ck::bhalf_t> && is_same_v<ComputeType, bhalf_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_INT8
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                         is_same_v<OutDataType, int8_t>)
+                         is_same_v<OutDataType, int8_t> && is_same_v<ComputeType, int8_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(op_ptrs);
                 add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instances(op_ptrs);
@@ -1119,7 +1124,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
         {
 #ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
+                         is_same_v<OutDataType, float> && is_same_v<ComputeType, float>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(op_ptrs);
             }
@@ -1148,14 +1153,15 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
+                         is_same_v<WeiDataType, ck::bhalf_t> &&
+                         is_same_v<OutDataType, ck::bhalf_t> && is_same_v<ComputeType, bhalf_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_INT8
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                         is_same_v<OutDataType, int8_t>)
+                         is_same_v<OutDataType, int8_t> && is_same_v<ComputeType, int8_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(op_ptrs);
                 add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instances(op_ptrs);

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_abd.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.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 {
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using ScaleAdd    = ck::tensor_operation::element_wise::ScaleAdd;
+
+#ifdef CK_ENABLE_BF16
+// grouped conv3d forward multi AB scaleadd, NDHWGC/GKZYXC/NDHWGK
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<>,
+                                                                NDHWGK,
+                                                                ck::Tuple<BF16, BF16>,
+                                                                ck::Tuple<BF16, BF16>,
+                                                                ck::Tuple<>,
+                                                                BF16,
+                                                                ScaleAdd,
+                                                                ScaleAdd,
+                                                                PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<>,
+                                                                NDHWGK,
+                                                                ck::Tuple<F16, F16>,
+                                                                ck::Tuple<F16, F16>,
+                                                                ck::Tuple<>,
+                                                                F16,
+                                                                ScaleAdd,
+                                                                ScaleAdd,
+                                                                PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP32
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<>,
+                                                                NDHWGK,
+                                                                ck::Tuple<F32, F32>,
+                                                                ck::Tuple<F32, F32>,
+                                                                ck::Tuple<>,
+                                                                F32,
+                                                                ScaleAdd,
+                                                                ScaleAdd,
+                                                                PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<>,
+                                                                NDHWGK,
+                                                                ck::Tuple<int8_t, int8_t>,
+                                                                ck::Tuple<int8_t, int8_t>,
+                                                                ck::Tuple<>,
+                                                                int8_t,
+                                                                ScaleAdd,
+                                                                ScaleAdd,
+                                                                PassThrough>>>& instances);
+#endif
+
+template <ck::index_t NumDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename DLayouts,
+          typename OutLayout,
+          typename InDataType,
+          typename WeiDataType,
+          typename DDataTypes,
+          typename OutDataType,
+          typename ComputeType>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
+    NumDimSpatial,
+    InLayout,
+    WeiLayout,
+    DLayouts,
+    OutLayout,
+    InDataType,
+    WeiDataType,
+    DDataTypes,
+    OutDataType,
+    ck::tensor_operation::element_wise::ScaleAdd,
+    ck::tensor_operation::element_wise::ScaleAdd,
+    ck::tensor_operation::element_wise::PassThrough,
+    ComputeType>>
+{
+    using DeviceOp =
+        DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        DLayouts,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        DDataTypes,
+                                        OutDataType,
+                                        ck::tensor_operation::element_wise::ScaleAdd,
+                                        ck::tensor_operation::element_wise::ScaleAdd,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        ComputeType>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+        if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
+                     is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
+        {
+#ifdef CK_ENABLE_FP32
+            if constexpr(is_same_v<InDataType, ck::Tuple<float, float>> &&
+                         is_same_v<WeiDataType, ck::Tuple<float, float>> &&
+                         is_same_v<OutDataType, float> && is_same_v<ComputeType, float>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+                    op_ptrs);
+            }
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, ck::Tuple<half_t, half_t>> &&
+                         is_same_v<WeiDataType, ck::Tuple<half_t, half_t>> &&
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, half_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+                    op_ptrs);
+            }
+#endif
+#ifdef CK_ENABLE_BF16
+            if constexpr(is_same_v<InDataType, ck::Tuple<ck::bhalf_t, ck::bhalf_t>> &&
+                         is_same_v<WeiDataType, ck::Tuple<ck::bhalf_t, ck::bhalf_t>> &&
+                         is_same_v<OutDataType, ck::bhalf_t> && is_same_v<ComputeType, ck::bhalf_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
+                    op_ptrs);
+            }
+#endif
+#ifdef CK_ENABLE_INT8
+            if constexpr(is_same_v<InDataType, ck::Tuple<int8_t, int8_t>> &&
+                         is_same_v<WeiDataType, ck::Tuple<int8_t, int8_t>> &&
+                         is_same_v<OutDataType, int8_t> && is_same_v<ComputeType, int8_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_int8_instances(
+                    op_ptrs);
+            }
+#endif
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_abd.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.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 {
+
+using PassThrough          = ck::tensor_operation::element_wise::PassThrough;
+using ScaleAddScaleAddRelu = ck::tensor_operation::element_wise::ScaleAddScaleAddRelu;
+
+#ifdef CK_ENABLE_BF16
+// grouped conv3d forward, NDHWGC/GKZYXC/NDHWGK
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                NDHWGK,
+                                                                BF16,
+                                                                BF16,
+                                                                ck::Tuple<BF16, BF16>,
+                                                                BF16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                ScaleAddScaleAddRelu>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                NDHWGK,
+                                                                F16,
+                                                                F16,
+                                                                ck::Tuple<F16, F16>,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                ScaleAddScaleAddRelu>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP32
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                NDHWGK,
+                                                                F32,
+                                                                F32,
+                                                                ck::Tuple<F32, F32>,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                ScaleAddScaleAddRelu>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                NDHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                ck::Tuple<F32, F32>,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                ScaleAddScaleAddRelu>>>& instances);
+#endif
+
+template <ck::index_t NumDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename DLayouts,
+          typename OutLayout,
+          typename InDataType,
+          typename WeiDataType,
+          typename DDataTypes,
+          typename OutDataType,
+          typename ComputeType>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
+    NumDimSpatial,
+    InLayout,
+    WeiLayout,
+    DLayouts,
+    OutLayout,
+    InDataType,
+    WeiDataType,
+    DDataTypes,
+    OutDataType,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::ScaleAddScaleAddRelu,
+    ComputeType>>
+{
+    using DeviceOp =
+        DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        DLayouts,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        DDataTypes,
+                                        OutDataType,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        ck::tensor_operation::element_wise::ScaleAddScaleAddRelu,
+                                        ComputeType>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+        if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
+                     is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
+        {
+#ifdef CK_ENABLE_FP32
+            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                         is_same_v<OutDataType, float>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+                    op_ptrs);
+            }
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, half_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+                    op_ptrs);
+            }
+#endif
+#ifdef CK_ENABLE_BF16
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
+                    op_ptrs);
+            }
+#endif
+#ifdef CK_ENABLE_INT8
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                         is_same_v<OutDataType, int8_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_int8_instances(
+                    op_ptrs);
+            }
+#endif
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

@@ -120,6 +120,32 @@ void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 
+void add_device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F8,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F8,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances);
+
 template <typename ALayout,
           typename BLayout,
           typename ELayout,
@@ -184,6 +210,24 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 add_device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
         }
+        else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, f8_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_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<ADataType, f8_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_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
+            }
+        }
         return op_ptrs;
     }
 };