complex contraction

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/device_normalization_bwd_gamma_beta.hpp"
+#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_bwd_gamma_beta.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+// M is invarient dimension, K is reduced dimension
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename GridwiseReduction,
+          typename DYDataType,
+          typename XDataType,
+          typename MeanInvStdDataType,
+          typename DGammaDataType,
+          typename DBetaDataType,
+          typename GridDesc_M_K,
+          typename GridDesc_M>
+__global__ void
+kernel_normalization_bwd_gamma_beta(const GridDesc_M_K dy_grid_desc_m_k,
+                                    const GridDesc_M_K x_grid_desc_m_k,
+                                    const GridDesc_M_K mean_grid_desc_m_k,
+                                    const GridDesc_M_K inv_std_grid_desc_m_k,
+                                    const GridDesc_M dgamma_grid_desc_m,
+                                    const GridDesc_M dbeta_grid_desc_m,
+                                    index_t num_k_block_tile_iteration,
+                                    const DYDataType* const __restrict__ p_dy_global,
+                                    const XDataType* const __restrict__ p_x_global,
+                                    const MeanInvStdDataType* const __restrict__ p_mean_global,
+                                    const MeanInvStdDataType* const __restrict__ p_inv_std_global,
+                                    DGammaDataType* const __restrict__ p_dgamma_global,
+                                    DBetaDataType* const __restrict__ p_dbeta_global)
+{
+    GridwiseReduction::Run(dy_grid_desc_m_k,
+                           x_grid_desc_m_k,
+                           mean_grid_desc_m_k,
+                           inv_std_grid_desc_m_k,
+                           dgamma_grid_desc_m,
+                           dbeta_grid_desc_m,
+                           num_k_block_tile_iteration,
+                           p_dy_global,
+                           p_x_global,
+                           p_mean_global,
+                           p_inv_std_global,
+                           p_dgamma_global,
+                           p_dbeta_global);
+};
+
+template <typename DYDataType,
+          typename XDataType,
+          typename MeanInvStdDataType,
+          typename ComputeDataType,
+          typename DGammaDataType,
+          typename DBetaDataType,
+          index_t Rank,
+          index_t NumReduceDim,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          bool IsDYFastestDimReduced,
+          index_t DYSrcVectorSize,
+          bool IsXFastestDimReduced,
+          index_t XSrcVectorSize,
+          bool IsMeanInvStdFastestDimReduced,
+          index_t MeanInvStdSrcVectorSize,
+          index_t DGammaDstVectorSize,
+          index_t DBetaDstVectorSize>
+struct DeviceNormalizationBwdGammaBetaImpl
+    : public DeviceNormalizationBwdGammaBeta<DYDataType,
+                                             XDataType,
+                                             MeanInvStdDataType,
+                                             DGammaDataType,
+                                             DBetaDataType,
+                                             Rank,
+                                             NumReduceDim>
+{
+
+    static constexpr index_t DYSrcVectorDim         = IsDYFastestDimReduced ? 1 : 0;
+    static constexpr index_t XSrcVectorDim          = IsXFastestDimReduced ? 1 : 0;
+    static constexpr index_t MeanInvStdSrcVectorDim = IsMeanInvStdFastestDimReduced ? 1 : 0;
+
+    static_assert(BlockSize == MThreadClusterSize * KThreadClusterSize);
+
+    static_assert(((DYSrcVectorDim == 0 && MThreadSliceSize % DYSrcVectorSize == 0) ||
+                   (DYSrcVectorDim == 1 && KThreadSliceSize % DYSrcVectorSize == 0)),
+                  "Invalid thread slice sizes and/or dy vector sizes configuration, please check!");
+
+    static_assert(((XSrcVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
+                   (XSrcVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0)),
+                  "Invalid thread slice sizes and/or x vector sizes configuration, please check!");
+
+    static_assert(
+        ((MThreadSliceSize % DGammaDstVectorSize == 0) ||
+         (MThreadSliceSize % DBetaDstVectorSize == 0)),
+        "Invalid thread slice sizes and/or Gamma and beta vector sizes configuration, please "
+        "check!");
+
+    static_assert(
+        (MeanInvStdSrcVectorDim == 0 && MThreadSliceSize % MeanInvStdSrcVectorSize == 0) ||
+            (MeanInvStdSrcVectorDim == 1 && KThreadSliceSize % MeanInvStdSrcVectorSize == 0),
+        "Invalid thread slice sizes and/or mean and inverse std vector sizes configuration, please "
+        "check!");
+
+    static constexpr index_t NumInvariantDim = Rank - NumReduceDim;
+    static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
+
+    static constexpr bool reduceAllDim = (NumInvariantDim == 0);
+    static_assert(!reduceAllDim);
+
+    static auto MakeSrc2dDescriptor(const std::vector<index_t>& inLengths,
+                                    const std::vector<index_t>& inStrides,
+                                    int numBlockTileIteration)
+    {
+        const auto tupleSrcLengths = make_tuple_from_array(inLengths, Number<Rank>{});
+        const auto tupleSrcStrides = make_tuple_from_array(inStrides, Number<Rank>{});
+
+        const auto inDesc = make_naive_tensor_descriptor(tupleSrcLengths, tupleSrcStrides);
+
+        const auto in_grid_desc_m_k = [&]() {
+            using InvariantDims = typename arithmetic_sequence_gen<0, NumInvariantDim, 1>::type;
+            using ReduceDims    = typename arithmetic_sequence_gen<NumInvariantDim, Rank, 1>::type;
+
+            const auto reduceDimLengths =
+                make_tuple_from_array_and_index_seq(inLengths, ReduceDims{});
+            const auto invariantDimLengths =
+                make_tuple_from_array_and_index_seq(inLengths, InvariantDims{});
+
+            return transform_tensor_descriptor(inDesc,
+                                               make_tuple(make_merge_transform(invariantDimLengths),
+                                                          make_merge_transform(reduceDimLengths)),
+                                               make_tuple(InvariantDims{}, ReduceDims{}),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }();
+
+        const auto invariantLength = in_grid_desc_m_k.GetLength(Number<0>{});
+        const auto reduceLength    = in_grid_desc_m_k.GetLength(Number<1>{});
+
+        const auto inPad_M =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+        const auto inPad_K = K_BlockTileSize * numBlockTileIteration - reduceLength;
+
+        auto in_grid_desc_m_k_padded = transform_tensor_descriptor(
+            in_grid_desc_m_k,
+            make_tuple(make_right_pad_transform(invariantLength, inPad_M),
+                       make_right_pad_transform(reduceLength, inPad_K)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return in_grid_desc_m_k_padded;
+    }
+
+    static auto MakeDst1dDescriptor(const std::vector<index_t>& outLengths,
+                                    const std::vector<index_t>& outStrides)
+    {
+        const auto tupleDstLengths =
+            generate_tuple([&](auto I) { return outLengths[I]; }, Number<NumInvariantDim>{});
+        const auto tupleDstStrides =
+            generate_tuple([&](auto I) { return outStrides[I]; }, Number<NumInvariantDim>{});
+
+        auto outDesc = make_naive_tensor_descriptor(tupleDstLengths, tupleDstStrides);
+
+        auto out_grid_desc_m = transform_tensor_descriptor(
+            outDesc,
+            make_tuple(make_merge_transform(tupleDstLengths)),
+            make_tuple(typename arithmetic_sequence_gen<0, NumInvariantDim, 1>::type{}),
+            make_tuple(Sequence<0>{}));
+
+        const auto invariantLength = out_grid_desc_m.GetLength(Number<0>{});
+
+        const auto outPad =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+
+        auto out_grid_desc_m_padded = transform_tensor_descriptor(
+            out_grid_desc_m,
+            make_tuple(make_right_pad_transform(invariantLength, outPad)),
+            make_tuple(Sequence<0>{}),
+            make_tuple(Sequence<0>{}));
+        return (out_grid_desc_m_padded);
+    };
+
+    using GridDesc_M_K = decltype(MakeSrc2dDescriptor({1}, {1}, 1));
+    using GridDesc_M   = decltype(MakeDst1dDescriptor({1}, {1}));
+
+    using GridwiseNormalizationBwdGammaBeta =
+        GridwiseNormalizationBwdGammaBeta_mk_to_k<DYDataType,
+                                                  XDataType,
+                                                  MeanInvStdDataType,
+                                                  ComputeDataType,
+                                                  DGammaDataType,
+                                                  DBetaDataType,
+                                                  GridDesc_M_K,
+                                                  GridDesc_M,
+                                                  BlockSize,
+                                                  MThreadClusterSize,
+                                                  KThreadClusterSize,
+                                                  MThreadSliceSize,
+                                                  KThreadSliceSize,
+                                                  DYSrcVectorDim,
+                                                  DYSrcVectorSize,
+                                                  XSrcVectorDim,
+                                                  XSrcVectorSize,
+                                                  MeanInvStdSrcVectorDim,
+                                                  MeanInvStdSrcVectorSize,
+                                                  DGammaDstVectorSize,
+                                                  DBetaDstVectorSize>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const std::vector<index_t> inLengths,
+                 const std::vector<index_t> dyStrides,
+                 const std::vector<index_t> xStrides,
+                 const std::vector<index_t> meanStrides,
+                 const std::vector<index_t> invStdStrides,
+                 const std::vector<index_t> outLengths,
+                 const std::vector<index_t> dgammaStrides,
+                 const std::vector<index_t> dbetaStrides,
+                 const std::vector<index_t> reduceDims,
+                 const DYDataType* p_dy,
+                 const XDataType* p_x,
+                 const MeanInvStdDataType* p_mean,
+                 const MeanInvStdDataType* p_invStd,
+                 DGammaDataType* p_dgamma,
+                 DBetaDataType* p_dbeta)
+            : p_dy_(p_dy),
+              p_x_(p_x),
+              p_mean_(p_mean),
+              p_invStd_(p_invStd),
+              p_dgamma_(p_dgamma),
+              p_dbeta_(p_dbeta),
+              outLengths_{outLengths},
+              dgammaStrides_{dgammaStrides},
+              dbetaStrides_{dbetaStrides}
+        {
+            inLengths_   = shuffle_tensor_dimensions<Rank, NumReduceDim>(inLengths, reduceDims);
+            dyStrides_   = shuffle_tensor_dimensions<Rank, NumReduceDim>(dyStrides, reduceDims);
+            xStrides_    = shuffle_tensor_dimensions<Rank, NumReduceDim>(xStrides, reduceDims);
+            meanStrides_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(meanStrides, reduceDims);
+            invStdStrides_ =
+                shuffle_tensor_dimensions<Rank, NumReduceDim>(invStdStrides, reduceDims);
+
+            std::tie(MRaw_, KRaw_) = get_2d_lengths<Rank, NumReduceDim>(inLengths_);
+
+            numBlockTileIteration_ = math::integer_divide_ceil(KRaw_, K_BlockTileSize);
+
+            gridSize_ = math::integer_divide_ceil(MRaw_, M_BlockTileSize);
+
+            dy_grid_desc_m_k_ = MakeSrc2dDescriptor(inLengths_, dyStrides_, numBlockTileIteration_);
+            x_grid_desc_m_k_  = MakeSrc2dDescriptor(inLengths_, xStrides_, numBlockTileIteration_);
+            mean_grid_desc_m_k_ =
+                MakeSrc2dDescriptor(inLengths_, meanStrides_, numBlockTileIteration_);
+            inv_std_grid_desc_m_k_ =
+                MakeSrc2dDescriptor(inLengths_, invStdStrides_, numBlockTileIteration_);
+
+            dgamma_grid_desc_m_ = MakeDst1dDescriptor(outLengths_, dgammaStrides_);
+            dbeta_grid_desc_m_  = MakeDst1dDescriptor(outLengths_, dbetaStrides_);
+        }
+
+        const DYDataType* p_dy_;
+        const XDataType* p_x_;
+        const MeanInvStdDataType* p_mean_;
+        const MeanInvStdDataType* p_invStd_;
+        DGammaDataType* p_dgamma_;
+        DBetaDataType* p_dbeta_;
+
+        std::vector<index_t> inLengths_;
+        std::vector<index_t> dyStrides_;
+        std::vector<index_t> xStrides_;
+        std::vector<index_t> meanStrides_;
+        std::vector<index_t> invStdStrides_;
+        std::vector<index_t> outLengths_;
+        std::vector<index_t> dgammaStrides_;
+        std::vector<index_t> dbetaStrides_;
+
+        int numBlockTileIteration_;
+        size_t gridSize_;
+
+        // Source descriptor
+        GridDesc_M_K dy_grid_desc_m_k_;
+        GridDesc_M_K x_grid_desc_m_k_;
+        GridDesc_M_K mean_grid_desc_m_k_;
+        GridDesc_M_K inv_std_grid_desc_m_k_;
+
+        // Destination descriptor
+        GridDesc_M dgamma_grid_desc_m_;
+        GridDesc_M dbeta_grid_desc_m_;
+
+        index_t MRaw_; // invarient length
+        index_t KRaw_; // reduce length
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            const auto kernel_main =
+                kernel_normalization_bwd_gamma_beta<GridwiseNormalizationBwdGammaBeta,
+                                                    DYDataType,
+                                                    XDataType,
+                                                    MeanInvStdDataType,
+                                                    DGammaDataType,
+                                                    DBetaDataType,
+                                                    GridDesc_M_K,
+                                                    GridDesc_M>;
+
+            return launch_and_time_kernel(stream_config,
+                                          kernel_main,
+                                          dim3(arg.gridSize_),
+                                          dim3(BlockSize),
+                                          0,
+                                          arg.dy_grid_desc_m_k_,
+                                          arg.x_grid_desc_m_k_,
+                                          arg.mean_grid_desc_m_k_,
+                                          arg.inv_std_grid_desc_m_k_,
+                                          arg.dgamma_grid_desc_m_,
+                                          arg.dbeta_grid_desc_m_,
+                                          arg.numBlockTileIteration_,
+                                          arg.p_dy_,
+                                          arg.p_x_,
+                                          arg.p_mean_,
+                                          arg.p_invStd_,
+                                          arg.p_dgamma_,
+                                          arg.p_dbeta_);
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    template <index_t SrcVectorDim, index_t SrcVectorSize>
+    bool IsSrcVectorDimSizeValid(const std::vector<index_t>& lengths,
+                                 const std::vector<index_t>& strides)
+    {
+        if constexpr(SrcVectorSize == 1)
+            return true;
+
+        // Fastest dimension is not reduced
+        if constexpr(SrcVectorDim == 0)
+        {
+            if constexpr(NumInvariantDim == 0)
+                return false;
+
+            if(strides[NumInvariantDim - 1] != 1)
+                return false;
+
+            if(lengths[NumInvariantDim - 1] % SrcVectorSize != 0)
+                return false;
+        }
+        else // Fastest dimension is reduced
+        {
+            if(strides[Rank - 1] != 1)
+                return false;
+
+            if(lengths[Rank - 1] % SrcVectorSize != 0)
+                return false;
+        };
+
+        return true;
+    }
+
+    template <index_t DstVectorSize>
+    bool IsDstVectorSizeValid(const std::vector<index_t>& lengths,
+                              const std::vector<index_t>& strides)
+    {
+        if constexpr(DstVectorSize == 1)
+            return true;
+
+        if(strides[NumInvariantDim - 1] != 1)
+            return false;
+
+        if(lengths[NumInvariantDim - 1] % DstVectorSize != 0)
+            return false;
+
+        return true;
+    }
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        const Argument* p_arg_ = dynamic_cast<const Argument*>(p_arg);
+
+        bool pass = true;
+        pass &= IsSrcVectorDimSizeValid<DYSrcVectorDim, DYSrcVectorSize>(p_arg_->inLengths_,
+                                                                         p_arg_->dyStrides_);
+        pass &= IsSrcVectorDimSizeValid<XSrcVectorDim, XSrcVectorSize>(p_arg_->inLengths_,
+                                                                       p_arg_->xStrides_);
+        pass &= IsSrcVectorDimSizeValid<MeanInvStdSrcVectorDim, MeanInvStdSrcVectorSize>(
+            p_arg_->inLengths_, p_arg_->meanStrides_);
+        pass &= IsSrcVectorDimSizeValid<MeanInvStdSrcVectorDim, MeanInvStdSrcVectorSize>(
+            p_arg_->inLengths_, p_arg_->invStdStrides_);
+
+        pass &=
+            IsDstVectorSizeValid<DGammaDstVectorSize>(p_arg_->outLengths_, p_arg_->dgammaStrides_);
+        pass &=
+            IsDstVectorSizeValid<DBetaDstVectorSize>(p_arg_->outLengths_, p_arg_->dbetaStrides_);
+
+        return pass;
+    }
+
+    std::unique_ptr<BaseArgument> MakeArgumentPointer(const std::vector<index_t> inLengths,
+                                                      const std::vector<index_t> dyStrides,
+                                                      const std::vector<index_t> xStrides,
+                                                      const std::vector<index_t> meanStrides,
+                                                      const std::vector<index_t> invStdStrides,
+                                                      const std::vector<index_t> outLengths,
+                                                      const std::vector<index_t> dgammaStrides,
+                                                      const std::vector<index_t> dbetaStrides,
+                                                      const std::vector<index_t> reduceDims,
+                                                      const void* p_dy,
+                                                      const void* p_x,
+                                                      const void* p_mean,
+                                                      const void* p_invStd,
+                                                      void* p_dgamma,
+                                                      void* p_dbeta) override
+    {
+        if(inLengths.size() != Rank || dyStrides.size() != Rank || xStrides.size() != Rank ||
+           meanStrides.size() != Rank || invStdStrides.size() != Rank)
+            throw std::runtime_error("dimension is incorrect");
+
+        if(outLengths.size() != NumInvariantDim || dgammaStrides.size() != NumInvariantDim ||
+           dbetaStrides.size() != NumInvariantDim)
+            throw std::runtime_error("dimension is incorrect");
+
+        return std::make_unique<Argument>(inLengths,
+                                          dyStrides,
+                                          xStrides,
+                                          meanStrides,
+                                          invStdStrides,
+                                          outLengths,
+                                          dgammaStrides,
+                                          dbetaStrides,
+                                          reduceDims,
+                                          static_cast<const DYDataType*>(p_dy),
+                                          static_cast<const XDataType*>(p_x),
+                                          static_cast<const MeanInvStdDataType*>(p_mean),
+                                          static_cast<const MeanInvStdDataType*>(p_invStd),
+                                          static_cast<DGammaDataType*>(p_dgamma),
+                                          static_cast<DBetaDataType*>(p_dbeta));
+    }
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp

@@ -85,10 +85,13 @@ struct Add
 
 struct ScaleAdd
 {
-    __host__ __device__ ScaleAdd(float scale) : scale_(scale) {}
+    __host__ __device__ ScaleAdd(float scale = 1.f) : scale_(scale) {}
 
     template <typename Y, typename X0, typename X1>
-    __host__ __device__ constexpr void operator()(Y& y, const X0& x0, const X1& x1) const;
+    __host__ __device__ constexpr void operator()(Y& y, const X0& x0, const X1& x1) const
+    {
+        y = ck::type_convert<Y>(scale_ * ck::type_convert<float>(x0) + ck::type_convert<float>(x1));
+    }
 
     template <>
     __host__ __device__ void

include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp

@@ -355,8 +355,8 @@ struct UnarySquare
     template <typename T>
     __host__ __device__ void operator()(T& y, const T& x) const
     {
-        static_assert(is_same_v<T, float> || is_same_v<T, double> || is_same_v<T, int32_t> ||
-                          is_same_v<T, int8_t>
+        static_assert(is_same_v<T, float> || is_same_v<T, half_t> || is_same_v<T, double> ||
+                          is_same_v<T, int32_t> || is_same_v<T, int8_t>
 #ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
                           || is_same_v<T, int4_t>
 #endif

include/ck/tensor_operation/gpu/grid/gridwise_elementwise_1d_scale.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/tensor_description/cluster_descriptor.hpp"
+#include "ck/utility/data_type.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+
+template <typename GridwiseElementwise1dFunctor,
+          typename InGrid1dDescTuple,
+          typename OutGrid1dDescTuple,
+          typename InDataTypePointerTuple,
+          typename OutDataTypePointerTuple,
+          typename ElementwiseOperation,
+          typename UnaryOperation,
+          typename Scale>
+__global__ void kernel_elementwise_1d(const InGrid1dDescTuple in_grid_1d_desc_tuple,
+                                      const OutGrid1dDescTuple out_grid_1d_desc_tuple,
+                                      const InDataTypePointerTuple p_in_global_tuple,
+                                      const OutDataTypePointerTuple p_out_global_tuple,
+                                      const ElementwiseOperation elementwise_op,
+                                      const UnaryOperation unary_op,
+                                      const Scale scale_op)
+{
+    GridwiseElementwise1dFunctor::Run(in_grid_1d_desc_tuple,
+                                      out_grid_1d_desc_tuple,
+                                      p_in_global_tuple,
+                                      p_out_global_tuple,
+                                      elementwise_op,
+                                      unary_op,
+                                      scale_op);
+}
+
+template <typename InGrid1dDescTuple,
+          typename OutGrid1dDescTuple,
+          typename InDataTypePointerTuple,
+          typename OutDataTypePointerTuple,
+          typename ElementwiseOperation,
+          typename UnaryOperation,
+          typename Scale,
+          index_t MPerThread,
+          typename InScalarPerVectorSeq,
+          typename OutScalarPerVectorSeq>
+struct GridwiseElementwise_1D
+{
+    static constexpr index_t NumInput  = InDataTypePointerTuple::Size();
+    static constexpr index_t NumOutput = OutDataTypePointerTuple::Size();
+
+    static_assert(NumInput == InScalarPerVectorSeq::Size() &&
+                      NumOutput == OutScalarPerVectorSeq::Size() &&
+                      NumInput == InGrid1dDescTuple::Size() &&
+                      NumOutput == OutGrid1dDescTuple::Size(),
+                  "Tuple size is inconsistent with the number of in/out!");
+
+    static constexpr auto I0 = Number<0>{};
+
+    static constexpr auto thread_buffer_desc_m =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MPerThread>{}));
+
+    using PassThroughOp = tensor_operation::element_wise::PassThrough;
+
+    __device__ static void Run(const InGrid1dDescTuple in_grid_1d_desc_tuple,
+                               const OutGrid1dDescTuple out_grid_1d_desc_tuple,
+                               const InDataTypePointerTuple p_in_global_tuple,
+                               const OutDataTypePointerTuple p_out_global_tuple,
+                               const ElementwiseOperation elementwise_op,
+                               const UnaryOperation unary_op,
+                               const Scale scale_op)
+    {
+        const index_t thread_global_id = get_thread_global_1d_id();
+
+        auto in_thread_buf_tuple = generate_tuple(
+            [&](auto I) {
+                using DataTypePointer = remove_cvref_t<decltype(InDataTypePointerTuple{}[I])>;
+                using DataType        = remove_cv_t<remove_pointer_t<DataTypePointer>>;
+
+                return StaticBuffer<AddressSpaceEnum::Vgpr, DataType, MPerThread, true>{};
+            },
+            Number<NumInput>{});
+
+        auto out_thread_buf_tuple = generate_tuple(
+            [&](auto I) {
+                using DataTypePointer = remove_cvref_t<decltype(OutDataTypePointerTuple{}[I])>;
+                using DataType        = remove_pointer_t<DataTypePointer>;
+
+                return StaticBuffer<AddressSpaceEnum::Vgpr, DataType, MPerThread, true>{};
+            },
+            Number<NumOutput>{});
+
+        auto in_global_buf_tuple = generate_tuple(
+            [&](auto I) {
+                static_assert(in_grid_1d_desc_tuple[I].GetNumOfDimension() == 1);
+
+                return make_dynamic_buffer<AddressSpaceEnum::Global>(
+                    p_in_global_tuple[I], in_grid_1d_desc_tuple[I].GetElementSpaceSize());
+            },
+            Number<NumInput>{});
+
+        auto out_global_buf_tuple = generate_tuple(
+            [&](auto I) {
+                static_assert(out_grid_1d_desc_tuple[I].GetNumOfDimension() == 1);
+
+                return make_dynamic_buffer<AddressSpaceEnum::Global>(
+                    p_out_global_tuple[I], out_grid_1d_desc_tuple[I].GetElementSpaceSize());
+            },
+            Number<NumOutput>{});
+
+        const auto thread_global_offset = make_multi_index(thread_global_id * MPerThread);
+
+        const index_t blockSize    = get_block_size();
+        const index_t blockPerGrid = get_grid_size();
+        const auto M               = in_grid_1d_desc_tuple[I0].GetLength(I0);
+        const index_t loop_step    = blockPerGrid * blockSize * MPerThread;
+        const auto loop_step_index = make_multi_index(loop_step);
+
+        auto in_global_load_tuple = generate_tuple(
+            [&](auto I) {
+                using DataTypePointer = remove_cvref_t<decltype(InDataTypePointerTuple{}[I])>;
+                using DataType        = remove_cv_t<remove_pointer_t<DataTypePointer>>;
+
+                return ThreadwiseTensorSliceTransfer_v2<DataType,
+                                                        DataType,
+                                                        decltype(in_grid_1d_desc_tuple[I]),
+                                                        decltype(thread_buffer_desc_m),
+                                                        Sequence<MPerThread>, // SliceLengths
+                                                        Sequence<0>,          // DimAccessOrder
+                                                        0,                    // SrcVectorDim
+                                                        InScalarPerVectorSeq::At(
+                                                            I), // ScalarPerVector
+                                                        1,      // SrcScalarStrideInVector
+                                                        false>{in_grid_1d_desc_tuple[I],
+                                                               thread_global_offset};
+            },
+            Number<NumInput>{});
+
+        auto out_global_store_tuple = generate_tuple(
+            [&](auto I) {
+                using DataTypePointer = remove_cvref_t<decltype(OutDataTypePointerTuple{}[I])>;
+                using DataType        = remove_pointer_t<DataTypePointer>;
+
+                return ThreadwiseTensorSliceTransfer_v1r3<DataType,
+                                                          DataType,
+                                                          decltype(thread_buffer_desc_m),
+                                                          decltype(out_grid_1d_desc_tuple[I]),
+                                                          PassThroughOp,
+                                                          Sequence<MPerThread>, // SliceLengths
+                                                          Sequence<0>,          // DimAccessOrder
+                                                          0,                    // SrcVectorDim
+                                                          OutScalarPerVectorSeq::At(I),
+                                                          InMemoryDataOperationEnum::Set,
+                                                          1,
+                                                          false>(
+                    out_grid_1d_desc_tuple[I], thread_global_offset, PassThroughOp{});
+            },
+            Number<NumOutput>{});
+
+        index_t num_iter = M / (loop_step);
+        do
+        {
+            static_for<0, NumInput, 1>{}([&](auto I) {
+                in_global_load_tuple(I).Run(in_grid_1d_desc_tuple[I],
+                                            in_global_buf_tuple[I],
+                                            thread_buffer_desc_m,
+                                            make_tuple(I0),
+                                            in_thread_buf_tuple(I));
+
+                in_global_load_tuple(I).MoveSrcSliceWindow(in_grid_1d_desc_tuple[I],
+                                                           loop_step_index);
+            });
+
+            static_for<0, MPerThread, 1>{}([&](auto iM) {
+                // get reference to in data
+                auto uop_data_refs = generate_tie(
+                    // return type should be lvalue
+                    [&](auto I) -> auto& { return in_thread_buf_tuple(I)(iM); },
+                    Number<NumInput>{});
+
+                // get reference to dst data
+                auto out_data_refs = generate_tie(
+                    // return type should be lvalue
+                    [&](auto I) -> auto& { return out_thread_buf_tuple(I)(iM); },
+                    Number<NumOutput>{});
+
+                unpack2(unary_op, uop_data_refs, uop_data_refs);
+
+                auto sop_in_data_refs = generate_tie(
+                    // return type should be lvalue
+                    [&](auto I) -> auto& { return in_thread_buf_tuple(I)(iM); },
+                    Number<NumInput>{});
+
+                auto sop_out_data_refs = generate_tie(
+                    // return type should be lvalue
+                    [&](auto I) -> auto& { return in_thread_buf_tuple(I)(iM); },
+                    Number<NumInput>{});
+
+                unpack2(scale_op, sop_out_data_refs, sop_in_data_refs);
+
+                const auto in_data_refs = generate_tie(
+                    // return type should be lvalue
+                    [&](auto I) -> const auto& { return in_thread_buf_tuple(I)(iM); },
+                    Number<NumInput>{});
+
+                unpack2(elementwise_op, out_data_refs, in_data_refs);
+            });
+
+            static_for<0, NumOutput, 1>{}([&](auto I) {
+                out_global_store_tuple(I).Run(thread_buffer_desc_m,
+                                              make_tuple(I0),
+                                              out_thread_buf_tuple[I],
+                                              out_grid_1d_desc_tuple[I],
+                                              out_global_buf_tuple(I));
+
+                out_global_store_tuple(I).MoveDstSliceWindow(out_grid_1d_desc_tuple[I],
+                                                             loop_step_index);
+            });
+        } while(--num_iter);
+    }
+};
+
+} // namespace ck

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

@@ -203,7 +203,7 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
     // A desc for source in blockwise copy
     template <typename AGridDesc_M_K>
     __host__ __device__ static constexpr auto
-    MakeAGridDescriptor_AK0_M_AK1(const AGridDesc_M_K& a_grid_desc_m_k)
+    MakeDefaultAGridDescriptor_AK0_M_AK1(const AGridDesc_M_K& a_grid_desc_m_k)
     {
         const auto M = a_grid_desc_m_k.GetLength(I0);
         const auto K = a_grid_desc_m_k.GetLength(I1);
@@ -219,17 +219,17 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
 
     template <typename AsGridDesc_M_K>
     __host__ __device__ static constexpr auto
-    MakeAsGridDescriptor_AK0_M_AK1(const AsGridDesc_M_K& as_grid_desc_m_k)
+    MakeDefaultAsGridDescriptor_AK0_M_AK1(const AsGridDesc_M_K& as_grid_desc_m_k)
     {
         return generate_tuple(
-            [&](auto i) { return MakeAGridDescriptor_AK0_M_AK1(as_grid_desc_m_k[i]); },
+            [&](auto i) { return MakeDefaultAGridDescriptor_AK0_M_AK1(as_grid_desc_m_k[i]); },
             Number<NumATensor>{});
     }
 
     // B desc for source in blockwise copy
     template <typename BGridDesc_N_K>
     __host__ __device__ static constexpr auto
-    MakeBGridDescriptor_BK0_N_BK1(const BGridDesc_N_K& b_grid_desc_n_k)
+    MakeDefaultBGridDescriptor_BK0_N_BK1(const BGridDesc_N_K& b_grid_desc_n_k)
     {
         const auto N = b_grid_desc_n_k.GetLength(I0);
         const auto K = b_grid_desc_n_k.GetLength(I1);
@@ -245,10 +245,10 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
 
     template <typename BsGridDesc_N_K>
     __host__ __device__ static constexpr auto
-    MakeBsGridDescriptor_BK0_N_BK1(const BsGridDesc_N_K& bs_grid_desc_n_k)
+    MakeDefaultBsGridDescriptor_BK0_N_BK1(const BsGridDesc_N_K& bs_grid_desc_n_k)
     {
         return generate_tuple(
-            [&](auto i) { return MakeBGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k[i]); },
+            [&](auto i) { return MakeDefaultBGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k[i]); },
             Number<NumBTensor>{});
     }
 
@@ -288,7 +288,7 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
     // return block_id to E matrix tile idx (m0, n0) mapping
     template <typename EGridDesc_M_N>
     __host__ __device__ static constexpr auto
-    MakeBlock2ETileMap(const EGridDesc_M_N& e_grid_desc_m_n)
+    MakeDefaultBlock2ETileMap(const EGridDesc_M_N& e_grid_desc_m_n)
     {
         return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, EGridDesc_M_N>(
             e_grid_desc_m_n);
@@ -591,6 +591,9 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
             generate_tuple([&](auto) { return make_multi_index(0, m_block_data_idx_on_grid, 0); },
                            Number<NumATensor>{});
 
+        static_assert(ABlockTransferSrcScalarPerVector == ABlockTransferDstScalarPerVector_AK1,
+                      "Src and Dst ScalarPerVector must be the same");
+
         auto a_blockwise_copy = ThreadGroupTensorSliceTransfer_v7r2<
             ThisThreadBlock,
             AsDataType,
@@ -619,6 +622,9 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
             generate_tuple([&](auto) { return make_multi_index(0, n_block_data_idx_on_grid, 0); },
                            Number<NumBTensor>{});
 
+        static_assert(BBlockTransferSrcScalarPerVector == BBlockTransferDstScalarPerVector_BK1,
+                      "Src and Dst ScalarPerVector must be the same");
+
         auto b_blockwise_copy = ThreadGroupTensorSliceTransfer_v7r2<
             ThisThreadBlock,
             BsDataType,
@@ -1005,9 +1011,9 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
         const auto e_grid_desc_m_n = MakeEGridDescriptor_M_N<ELayout, GemmSpec>(M, N, StrideE);
 
         // tensor descriptors for block/thread-wise copy
-        const auto as_grid_desc_ak0_m_ak1 = MakeAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k);
+        const auto as_grid_desc_ak0_m_ak1 = MakeDefaultAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k);
 
-        const auto bs_grid_desc_bk0_n_bk1 = MakeBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k);
+        const auto bs_grid_desc_bk0_n_bk1 = MakeDefaultBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k);
 
         const auto ds_grid_desc_mblock_mperblock_nblock_nperblock =
             MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n);

include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_bwd_gamma_beta.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp"
+
+namespace ck {
+
+// dgamma = reduce_sum(dy * (x - mean) * inv_std)
+// dbeta = reduce_sum(dy)
+template <typename DYDataType,
+          typename XDataType,
+          typename MeanInvStdDataType,
+          typename ComputeDataType,
+          typename DGammaDataType,
+          typename DBetaDataType,
+          typename GridDesc_M_K,
+          typename GridDesc_M,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t DYSrcVectorDim,
+          index_t DYSrcVectorSize,
+          index_t XSrcVectorDim,
+          index_t XSrcVectorSize,
+          index_t MeanInvStdSrcVectorDim,
+          index_t MeanInvStdSrcVectorSize,
+          index_t DGammaDstVectorSize,
+          index_t DBetaDstVectorSize>
+struct GridwiseNormalizationBwdGammaBeta_mk_to_k
+{
+    // if we just check ThreadSliceSize  & VectorSize == 0, the performance may be poor
+    static_assert(((DYSrcVectorDim == 0 && MThreadSliceSize == DYSrcVectorSize) ||
+                   (DYSrcVectorDim == 1 && KThreadSliceSize == DYSrcVectorSize)),
+                  "Invalid thread slice sizes and/or dy vector sizes configuration, please check!");
+
+    static_assert(((XSrcVectorDim == 0 && MThreadSliceSize == XSrcVectorSize) ||
+                   (XSrcVectorDim == 1 && KThreadSliceSize == XSrcVectorSize)),
+                  "Invalid thread slice sizes and/or x vector sizes configuration, please check!");
+
+    using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>;
+
+    using DYThreadBufferDimAccessOrder =
+        typename conditional<DYSrcVectorDim == 0, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    using XThreadBufferDimAccessOrder =
+        typename conditional<XSrcVectorDim == 0, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    using MeanInvStdThreadBufferDimAccessOrder =
+        typename conditional<MeanInvStdSrcVectorDim == 0, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    using ThreadClusterArrangeOrder = DYThreadBufferDimAccessOrder;
+
+    static constexpr auto thread_cluster_desc =
+        make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
+
+    using ThreadBufferLengths_M_K = Sequence<MThreadSliceSize, KThreadSliceSize>;
+    using ThreadBufferLengths_M   = Sequence<MThreadSliceSize>;
+
+    static constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<KThreadSliceSize>{}));
+
+    static constexpr auto thread_buffer_desc_m =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}));
+
+    using PassThroughOp = tensor_operation::element_wise::PassThrough;
+
+    using BlockwiseSumReduce = PartitionedBlockwiseReduction<ComputeDataType,
+                                                             BlockSize,
+                                                             ThreadClusterLengths_M_K,
+                                                             ThreadClusterArrangeOrder,
+                                                             reduce::Add,
+                                                             true>;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+
+    static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
+
+    __device__ static void Run(const GridDesc_M_K& dy_grid_desc_m_k,
+                               const GridDesc_M_K& x_grid_desc_m_k,
+                               const GridDesc_M_K& mean_grid_desc_m_k,
+                               const GridDesc_M_K& inv_std_grid_desc_m_k,
+                               const GridDesc_M& dgamma_grid_desc_m,
+                               const GridDesc_M& dbeta_grid_desc_m,
+                               index_t num_k_block_tile_iteration,
+                               const DYDataType* const __restrict__ p_dy_global,
+                               const XDataType* const __restrict__ p_x_global,
+                               const MeanInvStdDataType* const __restrict__ p_mean_global,
+                               const MeanInvStdDataType* const __restrict__ p_inv_std_global,
+                               DGammaDataType* const __restrict__ p_dgamma_global,
+                               DBetaDataType* const __restrict__ p_dbeta_global)
+    {
+        // LDS
+        __shared__ ComputeDataType p_reduce_work_buffer[BlockSize];
+
+        auto reduce_work_buf =
+            make_dynamic_buffer<AddressSpaceEnum::Lds>(p_reduce_work_buffer, BlockSize);
+
+        // Global
+        const auto dy_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_dy_global, dy_grid_desc_m_k.GetElementSpaceSize());
+
+        const auto x_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_x_global, x_grid_desc_m_k.GetElementSpaceSize());
+
+        const auto mean_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_mean_global, mean_grid_desc_m_k.GetElementSpaceSize());
+
+        const auto inv_std_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_inv_std_global, inv_std_grid_desc_m_k.GetElementSpaceSize());
+
+        auto dgamma_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_dgamma_global, dgamma_grid_desc_m.GetElementSpaceSize());
+
+        auto dbeta_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_dbeta_global, dbeta_grid_desc_m.GetElementSpaceSize());
+
+        // VGPR
+        auto dy_thread_buf = StaticBuffer<AddressSpaceEnum::Vgpr,
+                                          ComputeDataType,
+                                          MThreadSliceSize * KThreadSliceSize,
+                                          true>{};
+
+        auto x_thread_buf = StaticBuffer<AddressSpaceEnum::Vgpr,
+                                         ComputeDataType,
+                                         MThreadSliceSize * KThreadSliceSize,
+                                         true>{};
+
+        auto mean_thread_buf = StaticBuffer<AddressSpaceEnum::Vgpr,
+                                            ComputeDataType,
+                                            MThreadSliceSize * KThreadSliceSize,
+                                            true>{};
+
+        auto inv_std_thread_buf = StaticBuffer<AddressSpaceEnum::Vgpr,
+                                               ComputeDataType,
+                                               MThreadSliceSize * KThreadSliceSize,
+                                               true>{};
+
+        auto dgamma_thread_buf =
+            StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>{};
+
+        auto dbeta_thread_buf =
+            StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>{};
+
+        const index_t thread_local_id = get_thread_local_1d_id();
+        const index_t block_global_id = get_block_1d_id();
+
+        const auto thread_cluster_idx =
+            thread_cluster_desc.CalculateBottomIndex(make_multi_index(thread_local_id));
+
+        const auto thread_m_cluster_id = thread_cluster_idx[I0];
+        const auto thread_k_cluster_id = thread_cluster_idx[I1];
+
+        // IO
+        auto threadwise_dy_load = ThreadwiseTensorSliceTransfer_v2<DYDataType,
+                                                                   ComputeDataType,
+                                                                   GridDesc_M_K,
+                                                                   decltype(thread_buffer_desc_m_k),
+                                                                   ThreadBufferLengths_M_K,
+                                                                   DYThreadBufferDimAccessOrder,
+                                                                   DYSrcVectorDim,
+                                                                   DYSrcVectorSize,
+                                                                   1,
+                                                                   true>(
+            dy_grid_desc_m_k,
+            make_multi_index(block_global_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize,
+                             thread_k_cluster_id * KThreadSliceSize));
+
+        auto threadwise_x_load = ThreadwiseTensorSliceTransfer_v2<XDataType,
+                                                                  ComputeDataType,
+                                                                  GridDesc_M_K,
+                                                                  decltype(thread_buffer_desc_m_k),
+                                                                  ThreadBufferLengths_M_K,
+                                                                  XThreadBufferDimAccessOrder,
+                                                                  XSrcVectorDim,
+                                                                  XSrcVectorSize,
+                                                                  1,
+                                                                  true>(
+            x_grid_desc_m_k,
+            make_multi_index(block_global_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize,
+                             thread_k_cluster_id * KThreadSliceSize));
+
+        auto threadwise_mean_load =
+            ThreadwiseTensorSliceTransfer_v2<MeanInvStdDataType,
+                                             ComputeDataType,
+                                             GridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_k),
+                                             ThreadBufferLengths_M_K,
+                                             MeanInvStdThreadBufferDimAccessOrder,
+                                             MeanInvStdSrcVectorDim,
+                                             MeanInvStdSrcVectorSize,
+                                             1,
+                                             true>(
+                mean_grid_desc_m_k,
+                make_multi_index(block_global_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id * KThreadSliceSize));
+
+        auto threadwise_inv_std_load =
+            ThreadwiseTensorSliceTransfer_v2<MeanInvStdDataType,
+                                             ComputeDataType,
+                                             GridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_k),
+                                             ThreadBufferLengths_M_K,
+                                             MeanInvStdThreadBufferDimAccessOrder,
+                                             MeanInvStdSrcVectorDim,
+                                             MeanInvStdSrcVectorSize,
+                                             1,
+                                             true>(
+                inv_std_grid_desc_m_k,
+                make_multi_index(block_global_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id * KThreadSliceSize));
+
+        auto threadwise_dgamma_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               DGammaDataType,
+                                               decltype(thread_buffer_desc_m),
+                                               GridDesc_M,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M,
+                                               Sequence<0>,
+                                               0,
+                                               DGammaDstVectorSize,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                dgamma_grid_desc_m,
+                make_multi_index(block_global_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize),
+                PassThroughOp{});
+
+        auto threadwise_dbeta_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               DBetaDataType,
+                                               decltype(thread_buffer_desc_m),
+                                               GridDesc_M,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M,
+                                               Sequence<0>,
+                                               0,
+                                               DBetaDstVectorSize,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                dbeta_grid_desc_m,
+                make_multi_index(block_global_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize),
+                PassThroughOp{});
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            dgamma_thread_buf(I) = type_convert<ComputeDataType>(0.0f);
+            dbeta_thread_buf(I)  = type_convert<ComputeDataType>(0.0f);
+        });
+
+        constexpr auto thread_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileSize);
+
+        for(index_t reducedTiles = 0; reducedTiles < num_k_block_tile_iteration; ++reducedTiles)
+        {
+            threadwise_dy_load.Run(dy_grid_desc_m_k,
+                                   dy_global_val_buf,
+                                   thread_buffer_desc_m_k,
+                                   make_tuple(I0, I0),
+                                   dy_thread_buf);
+
+            threadwise_x_load.Run(x_grid_desc_m_k,
+                                  x_global_val_buf,
+                                  thread_buffer_desc_m_k,
+                                  make_tuple(I0, I0),
+                                  x_thread_buf);
+
+            threadwise_mean_load.Run(mean_grid_desc_m_k,
+                                     mean_global_val_buf,
+                                     thread_buffer_desc_m_k,
+                                     make_tuple(I0, I0),
+                                     mean_thread_buf);
+
+            threadwise_inv_std_load.Run(inv_std_grid_desc_m_k,
+                                        inv_std_global_val_buf,
+                                        thread_buffer_desc_m_k,
+                                        make_tuple(I0, I0),
+                                        inv_std_thread_buf);
+
+            threadwise_dy_load.MoveSrcSliceWindow(dy_grid_desc_m_k, thread_copy_fwd_step_m_k);
+            threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
+            threadwise_mean_load.MoveSrcSliceWindow(mean_grid_desc_m_k, thread_copy_fwd_step_m_k);
+            threadwise_inv_std_load.MoveSrcSliceWindow(inv_std_grid_desc_m_k,
+                                                       thread_copy_fwd_step_m_k);
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                constexpr auto offset_m =
+                    Number<thread_buffer_desc_m.CalculateOffset(make_tuple(iM))>{};
+
+                static_for<0, KThreadSliceSize, 1>{}([&](auto iK) {
+                    constexpr auto offset_m_k =
+                        Number<thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK))>{};
+
+                    dgamma_thread_buf(offset_m) +=
+                        dy_thread_buf[offset_m_k] * inv_std_thread_buf[offset_m_k] *
+                        (x_thread_buf[offset_m_k] - mean_thread_buf[offset_m_k]);
+
+                    dbeta_thread_buf(offset_m) += dy_thread_buf[offset_m_k];
+                });
+            });
+        }
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            if constexpr(I > 0)
+                block_sync_lds();
+
+            BlockwiseSumReduce::Reduce(reduce_work_buf, dbeta_thread_buf(I));
+            block_sync_lds();
+            BlockwiseSumReduce::Reduce(reduce_work_buf, dgamma_thread_buf(I));
+        });
+
+        if(thread_k_cluster_id == 0)
+        {
+            threadwise_dgamma_store.Run(thread_buffer_desc_m,
+                                        make_tuple(I0),
+                                        dgamma_thread_buf,
+                                        dgamma_grid_desc_m,
+                                        dgamma_global_val_buf);
+
+            threadwise_dbeta_store.Run(thread_buffer_desc_m,
+                                       make_tuple(I0),
+                                       dbeta_thread_buf,
+                                       dbeta_grid_desc_m,
+                                       dbeta_global_val_buf);
+        }
+    }
+};
+
+} // namespace ck

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
@@ -42,28 +56,35 @@ template <ck::index_t NDimSpatial,
           typename InElementwiseOperation,
           typename WeiElementwiseOperation,
           typename OutElementwiseOperation,
-          ck::index_t NumDTensor                                                    = 0,
+          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,
-                 const std::array<Tensor<OutDataType>, NumDTensor>& d_tensors)
+        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},
-              d_tensors_{d_tensors},
+              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},
@@ -78,7 +99,9 @@ struct ReferenceConvFwd : public device::BaseOperator
         const Tensor<WeiDataType>& weight_;
         Tensor<OutDataType>& output_;
 
-        const std::array<Tensor<OutDataType>, NumDTensor>& d_tensors_;
+        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_;
@@ -119,42 +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);
                             }
                         }
                     }
-
-                    OutDataType v_out;
                     OutDataType v_acc_converted = ck::type_convert<OutDataType>(v_acc);
-                    if constexpr(NumDTensor == 0)
-                    {
-                        arg.out_element_op_(v_out, v_acc_converted);
-                    }
-                    else if constexpr(NumDTensor == 1)
-                    {
-                        arg.out_element_op_(v_out, v_acc_converted, arg.d_tensors_[0](g, n, k, wo));
-                    }
-                    else if constexpr(NumDTensor == 2)
-                    {
-                        arg.out_element_op_(v_out,
-                                            v_acc_converted,
-                                            arg.d_tensors_[0](g, n, k, wo),
-                                            arg.d_tensors_[1](g, n, k, wo));
-                    }
-                    else
-                    {
-                        throw std::runtime_error("Output ElementOp not supported in reference.");
-                    }
-                    arg.output_(g, n, k, wo) = v_out;
+                    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,
@@ -191,44 +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);
                                 }
                             }
                         }
                     }
-
-                    OutDataType v_out;
                     OutDataType v_acc_converted = ck::type_convert<OutDataType>(v_acc);
-                    if constexpr(NumDTensor == 0)
-                    {
-                        arg.out_element_op_(v_out, v_acc_converted);
-                    }
-                    else if constexpr(NumDTensor == 1)
-                    {
-                        arg.out_element_op_(
-                            v_out, v_acc_converted, arg.d_tensors_[0](g, n, k, ho, wo));
-                    }
-                    else if constexpr(NumDTensor == 2)
-                    {
-                        arg.out_element_op_(v_out,
-                                            v_acc_converted,
-                                            arg.d_tensors_[0](g, n, k, ho, wo),
-                                            arg.d_tensors_[1](g, n, k, ho, wo));
-                    }
-                    else
-                    {
-                        throw std::runtime_error("Output ElementOp not supported in reference.");
-                    }
-                    arg.output_(g, n, k, ho, wo) = v_out;
+                    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,
@@ -275,47 +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);
                                     }
                                 }
                             }
                         }
                     }
-
-                    OutDataType v_out;
                     OutDataType v_acc_converted = ck::type_convert<OutDataType>(v_acc);
-                    if constexpr(NumDTensor == 0)
-                    {
-                        arg.out_element_op_(v_out, v_acc_converted);
-                    }
-                    else if constexpr(NumDTensor == 1)
-                    {
-                        arg.out_element_op_(
-                            v_out, v_acc_converted, arg.d_tensors_[0](g, n, k, d_o, ho, wo));
-                    }
-                    else if constexpr(NumDTensor == 2)
-                    {
-                        arg.out_element_op_(v_out,
-                                            v_acc_converted,
-                                            arg.d_tensors_[0](g, n, k, d_o, ho, wo),
-                                            arg.d_tensors_[1](g, n, k, d_o, ho, wo));
-                    }
-                    else
-                    {
-                        throw std::runtime_error("Output ElementOp not supported in reference.");
-                    }
-                    arg.output_(g, n, k, d_o, ho, wo) = v_out;
+                    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,
@@ -338,6 +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
@@ -349,17 +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,
-                             const std::array<Tensor<OutDataType>, NumDTensor>& d_tensors = {})
+    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,
@@ -371,7 +435,9 @@ struct ReferenceConvFwd : public device::BaseOperator
                         in_element_op,
                         wei_element_op,
                         out_element_op,
-                        d_tensors};
+                        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_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/gpu/gemm.hpp

@@ -377,7 +377,7 @@ struct DeviceOperationInstanceFactory<
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<CLayout, Row>)
             {
-                add_device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances(op_ptrs);
+                /// add_device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances(op_ptrs);
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f32_f32_f32_mk_kn_mn_instances(op_ptrs);
 #endif
@@ -386,7 +386,7 @@ struct DeviceOperationInstanceFactory<
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
-                add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(op_ptrs);
+                /// add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(op_ptrs);
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances(op_ptrs);
 #endif
@@ -395,7 +395,7 @@ struct DeviceOperationInstanceFactory<
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
                               is_same_v<CLayout, Row>)
             {
-                add_device_gemm_xdl_f32_f32_f32_km_kn_mn_instances(op_ptrs);
+                /// add_device_gemm_xdl_f32_f32_f32_km_kn_mn_instances(op_ptrs);
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(op_ptrs);
 #endif
@@ -404,7 +404,7 @@ struct DeviceOperationInstanceFactory<
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
-                add_device_gemm_xdl_f32_f32_f32_km_nk_mn_instances(op_ptrs);
+                /// add_device_gemm_xdl_f32_f32_f32_km_nk_mn_instances(op_ptrs);
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f32_f32_f32_km_nk_mn_instances(op_ptrs);
 #endif
@@ -418,7 +418,7 @@ struct DeviceOperationInstanceFactory<
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<CLayout, Row>)
             {
-                add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                /// add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
@@ -430,7 +430,7 @@ struct DeviceOperationInstanceFactory<
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
-                add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                /// add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs);
@@ -443,7 +443,7 @@ struct DeviceOperationInstanceFactory<
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
                               is_same_v<CLayout, Row>)
             {
-                add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
+                /// add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
@@ -455,7 +455,7 @@ struct DeviceOperationInstanceFactory<
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
-                add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
+                /// add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);

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

@@ -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"
@@ -50,12 +50,12 @@ using device_grouped_conv_fwd_xdl_scaleadd_scaleadd_relu_bf16_instances = std::t
         //########################################|           |       |       |            |       |      |      |        |         |            |      |   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, 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>,
+        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
-        DeviceGroupedConvFwdMultipleD_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>,
-        DeviceGroupedConvFwdMultipleD_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,    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>,
 
-        DeviceGroupedConvFwdMultipleD_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>
+        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
     >;
 
@@ -72,12 +72,12 @@ using device_grouped_conv_fwd_xdl_scaleadd_scaleadd_relu_f16_instances = std::tu
         //########################################|           |       |       |            |       |      |      |        |         |            |      |   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, 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>,
+        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
-        DeviceGroupedConvFwdMultipleD_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>,
-        DeviceGroupedConvFwdMultipleD_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,    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>,
 
-        DeviceGroupedConvFwdMultipleD_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>
+        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
     >;
 
@@ -94,12 +94,12 @@ using device_grouped_conv_fwd_xdl_scaleadd_scaleadd_relu_f32_instances = std::tu
         //########################################|           |       |       |            |       |      |      |        |         |            |      |   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, 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>,
+        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
-        DeviceGroupedConvFwdMultipleD_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>,
-        DeviceGroupedConvFwdMultipleD_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,    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>,
 
-        DeviceGroupedConvFwdMultipleD_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>
+        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
     >;
 
@@ -116,12 +116,12 @@ using device_grouped_conv_fwd_xdl_scaleadd_scaleadd_relu_int8_instances = std::t
         //########################################|           |       |       |            |       |      |      |        |         |                   |      |   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, 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>,
+        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
-        DeviceGroupedConvFwdMultipleD_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>,
-        DeviceGroupedConvFwdMultipleD_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,    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>,
 
-        DeviceGroupedConvFwdMultipleD_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>
+        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
     >;
 

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()
     {

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

@@ -7,7 +7,7 @@
 #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"
 
@@ -24,66 +24,66 @@ using ScaleAddScaleAddRelu = ck::tensor_operation::element_wise::ScaleAddScaleAd
 #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<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              ck::Tuple<NDHWGK, NDHWGK>,
-                                                              NDHWGK,
-                                                              BF16,
-                                                              BF16,
-                                                              ck::Tuple<BF16, BF16>,
-                                                              BF16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              ScaleAddScaleAddRelu>>>& 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<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              ck::Tuple<NDHWGK, NDHWGK>,
-                                                              NDHWGK,
-                                                              F16,
-                                                              F16,
-                                                              ck::Tuple<F16, F16>,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              ScaleAddScaleAddRelu>>>& 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<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              ck::Tuple<NDHWGK, NDHWGK>,
-                                                              NDHWGK,
-                                                              F32,
-                                                              F32,
-                                                              ck::Tuple<F32, F32>,
-                                                              F32,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              ScaleAddScaleAddRelu>>>& 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<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              ck::Tuple<NDHWGK, NDHWGK>,
-                                                              NDHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              ck::Tuple<F32, F32>,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              ScaleAddScaleAddRelu>>>& 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,
@@ -96,7 +96,7 @@ template <ck::index_t NumDimSpatial,
           typename DDataTypes,
           typename OutDataType,
           typename ComputeType>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
     NumDimSpatial,
     InLayout,
     WeiLayout,
@@ -112,19 +112,19 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     ComputeType>>
 {
     using DeviceOp =
-        DeviceGroupedConvFwdMultipleD<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>;
+        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()
     {

library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp

@@ -7,7 +7,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.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/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -20,96 +20,96 @@ namespace instance {
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(
     std::vector<
-        std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      NHWGC,
-                                                      GKYXC,
-                                                      GK_GK_Tuple,
-                                                      NHWGK,
-                                                      int8_t,
-                                                      int8_t,
-                                                      I32_F32_Tuple,
-                                                      int8_t,
-                                                      PassThrough,
-                                                      PassThrough,
-                                                      Add_Activation_Mul2_Clamp<PassThrough>>>>&
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        GK_GK_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        I32_F32_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Add_Activation_Mul2_Clamp<PassThrough>>>>&
         instances);
 
 void add_device_conv2d_dl_bias_relu_perchannel_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              GK_GK_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              I32_F32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Add_Activation_Mul2_Clamp<Relu>>>>&
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                GK_GK_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                I32_F32_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                Add_Activation_Mul2_Clamp<Relu>>>>&
         instances);
 
 void add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(
     std::vector<
-        std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      NHWGC,
-                                                      GKYXC,
-                                                      GK_GK_Tuple,
-                                                      NHWGK,
-                                                      int8_t,
-                                                      int8_t,
-                                                      I32_F32_Tuple,
-                                                      int8_t,
-                                                      PassThrough,
-                                                      PassThrough,
-                                                      Add_Mul2_Activation_Mul_Clamp<TanH>>>>&
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        GK_GK_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        I32_F32_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Add_Mul2_Activation_Mul_Clamp<TanH>>>>&
         instances);
 #endif
 void add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(
     std::vector<
-        std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      NHWGC,
-                                                      GKYXC,
-                                                      GK_GK_Tuple,
-                                                      NHWGK,
-                                                      int8_t,
-                                                      int8_t,
-                                                      I32_F32_Tuple,
-                                                      int8_t,
-                                                      PassThrough,
-                                                      PassThrough,
-                                                      Add_Activation_Mul2_Clamp<PassThrough>>>>&
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        GK_GK_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        I32_F32_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Add_Activation_Mul2_Clamp<PassThrough>>>>&
         instances);
 
 void add_device_conv2d_xdl_bias_relu_perchannel_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              GK_GK_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              I32_F32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Add_Activation_Mul2_Clamp<Relu>>>>&
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                GK_GK_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                I32_F32_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                Add_Activation_Mul2_Clamp<Relu>>>>&
         instances);
 
 void add_device_conv2d_xdl_bias_tanh_perchannel_quantization_int8_instances(
     std::vector<
-        std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      NHWGC,
-                                                      GKYXC,
-                                                      GK_GK_Tuple,
-                                                      NHWGK,
-                                                      int8_t,
-                                                      int8_t,
-                                                      I32_F32_Tuple,
-                                                      int8_t,
-                                                      PassThrough,
-                                                      PassThrough,
-                                                      Add_Mul2_Activation_Mul_Clamp<TanH>>>>&
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        GK_GK_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        I32_F32_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Add_Mul2_Activation_Mul_Clamp<TanH>>>>&
         instances);
 
 // piecewise activation function
@@ -123,7 +123,7 @@ template <ck::index_t NumDimSpatial,
           typename DsDataType,
           typename OutDataType,
           typename Activation>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
     NumDimSpatial,
     InLayout,
     WeiLayout,
@@ -137,18 +137,19 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     ck::tensor_operation::element_wise::PassThrough,
     Add_Activation_Mul2_Clamp<Activation>>>
 {
-    using DeviceOp = DeviceGroupedConvFwdMultipleD<NumDimSpatial,
-                                                   InLayout,
-                                                   WeiLayout,
-                                                   DsLayout,
-                                                   OutLayout,
-                                                   InDataType,
-                                                   WeiDataType,
-                                                   DsDataType,
-                                                   OutDataType,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   Add_Activation_Mul2_Clamp<Activation>>;
+    using DeviceOp =
+        DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        DsLayout,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        DsDataType,
+                                        OutDataType,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        Add_Activation_Mul2_Clamp<Activation>>;
 
     static auto GetInstances()
     {
@@ -193,7 +194,7 @@ template <ck::index_t NumDimSpatial,
           typename DsDataType,
           typename OutDataType,
           typename Activation>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
     NumDimSpatial,
     InLayout,
     WeiLayout,
@@ -207,18 +208,19 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     ck::tensor_operation::element_wise::PassThrough,
     Add_Mul2_Activation_Mul_Clamp<Activation>>>
 {
-    using DeviceOp = DeviceGroupedConvFwdMultipleD<NumDimSpatial,
-                                                   InLayout,
-                                                   WeiLayout,
-                                                   DsLayout,
-                                                   OutLayout,
-                                                   InDataType,
-                                                   WeiDataType,
-                                                   DsDataType,
-                                                   OutDataType,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   Add_Mul2_Activation_Mul_Clamp<Activation>>;
+    using DeviceOp =
+        DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        DsLayout,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        DsDataType,
+                                        OutDataType,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        Add_Mul2_Activation_Mul_Clamp<Activation>>;
 
     static auto GetInstances()
     {

library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp

@@ -7,7 +7,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.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/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -20,94 +20,96 @@ namespace instance {
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(
     std::vector<
-        std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      NHWGC,
-                                                      GKYXC,
-                                                      GK_Tuple,
-                                                      NHWGK,
-                                                      int8_t,
-                                                      int8_t,
-                                                      I32_Tuple,
-                                                      int8_t,
-                                                      PassThrough,
-                                                      PassThrough,
-                                                      Add_Activation_Mul_Clamp<PassThrough>>>>&
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        GK_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        I32_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Add_Activation_Mul_Clamp<PassThrough>>>>&
         instances);
 
 void add_device_conv2d_dl_bias_relu_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              I32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Add_Activation_Mul_Clamp<Relu>>>>&
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                GK_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                I32_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                Add_Activation_Mul_Clamp<Relu>>>>&
         instances);
 
 void add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              I32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Add_Mul_Activation_Mul_Clamp<TanH>>>>&
+    std::vector<
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        GK_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        I32_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Add_Mul_Activation_Mul_Clamp<TanH>>>>&
         instances);
 #endif
 void add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(
     std::vector<
-        std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      NHWGC,
-                                                      GKYXC,
-                                                      GK_Tuple,
-                                                      NHWGK,
-                                                      int8_t,
-                                                      int8_t,
-                                                      I32_Tuple,
-                                                      int8_t,
-                                                      PassThrough,
-                                                      PassThrough,
-                                                      Add_Activation_Mul_Clamp<PassThrough>>>>&
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        GK_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        I32_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Add_Activation_Mul_Clamp<PassThrough>>>>&
         instances);
 
 void add_device_conv2d_xdl_bias_relu_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              I32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Add_Activation_Mul_Clamp<Relu>>>>&
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                GK_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                I32_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                Add_Activation_Mul_Clamp<Relu>>>>&
         instances);
 
 void add_device_conv2d_xdl_bias_tanh_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              I32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Add_Mul_Activation_Mul_Clamp<TanH>>>>&
+    std::vector<
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        GK_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        I32_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Add_Mul_Activation_Mul_Clamp<TanH>>>>&
         instances);
 
 // piecewise activation function
@@ -121,7 +123,7 @@ template <ck::index_t NumDimSpatial,
           typename DsDataType,
           typename OutDataType,
           typename Activation>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
     NumDimSpatial,
     InLayout,
     WeiLayout,
@@ -135,18 +137,19 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     ck::tensor_operation::element_wise::PassThrough,
     Add_Activation_Mul_Clamp<Activation>>>
 {
-    using DeviceOp = DeviceGroupedConvFwdMultipleD<NumDimSpatial,
-                                                   InLayout,
-                                                   WeiLayout,
-                                                   DsLayout,
-                                                   OutLayout,
-                                                   InDataType,
-                                                   WeiDataType,
-                                                   DsDataType,
-                                                   OutDataType,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   Add_Activation_Mul_Clamp<Activation>>;
+    using DeviceOp =
+        DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        DsLayout,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        DsDataType,
+                                        OutDataType,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        Add_Activation_Mul_Clamp<Activation>>;
 
     static auto GetInstances()
     {
@@ -191,7 +194,7 @@ template <ck::index_t NumDimSpatial,
           typename DsDataType,
           typename OutDataType,
           typename Activation>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
     NumDimSpatial,
     InLayout,
     WeiLayout,
@@ -205,18 +208,19 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     ck::tensor_operation::element_wise::PassThrough,
     Add_Mul_Activation_Mul_Clamp<Activation>>>
 {
-    using DeviceOp = DeviceGroupedConvFwdMultipleD<NumDimSpatial,
-                                                   InLayout,
-                                                   WeiLayout,
-                                                   DsLayout,
-                                                   OutLayout,
-                                                   InDataType,
-                                                   WeiDataType,
-                                                   DsDataType,
-                                                   OutDataType,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   Add_Mul_Activation_Mul_Clamp<Activation>>;
+    using DeviceOp =
+        DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        DsLayout,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        DsDataType,
+                                        OutDataType,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        Add_Mul_Activation_Mul_Clamp<Activation>>;
 
     static auto GetInstances()
     {

library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp

@@ -7,7 +7,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.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/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -19,63 +19,65 @@ namespace instance {
 #ifdef DL_KERNELS
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_perchannel_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              F32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Activation_Mul2_Clamp<PassThrough>>>>&
+    std::vector<
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        GK_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        F32_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Activation_Mul2_Clamp<PassThrough>>>>&
         instances);
 
 void add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              F32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Activation_Mul2_Clamp<Relu>>>>&
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                GK_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                F32_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                Activation_Mul2_Clamp<Relu>>>>&
         instances);
 #endif
 void add_device_conv2d_xdl_perchannel_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              F32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Activation_Mul2_Clamp<PassThrough>>>>&
+    std::vector<
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        GK_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        F32_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Activation_Mul2_Clamp<PassThrough>>>>&
         instances);
 
 void add_device_conv2d_xdl_relu_perchannel_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              F32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Activation_Mul2_Clamp<Relu>>>>&
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                GK_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                F32_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                Activation_Mul2_Clamp<Relu>>>>&
         instances);
 
 template <ck::index_t NumDimSpatial,
@@ -88,7 +90,7 @@ template <ck::index_t NumDimSpatial,
           typename DsDataType,
           typename OutDataType,
           typename Activation>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
     NumDimSpatial,
     InLayout,
     WeiLayout,
@@ -102,18 +104,19 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     ck::tensor_operation::element_wise::PassThrough,
     Activation_Mul2_Clamp<Activation>>>
 {
-    using DeviceOp = DeviceGroupedConvFwdMultipleD<NumDimSpatial,
-                                                   InLayout,
-                                                   WeiLayout,
-                                                   GK_Tuple,
-                                                   OutLayout,
-                                                   InDataType,
-                                                   WeiDataType,
-                                                   F32_Tuple,
-                                                   OutDataType,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   ck::tensor_operation::element_wise::PassThrough,
-                                                   Activation_Mul2_Clamp<Activation>>;
+    using DeviceOp =
+        DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        GK_Tuple,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        F32_Tuple,
+                                        OutDataType,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        ck::tensor_operation::element_wise::PassThrough,
+                                        Activation_Mul2_Clamp<Activation>>;
 
     static auto GetInstances()
     {

library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp

@@ -7,7 +7,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.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/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -19,63 +19,65 @@ namespace instance {
 #ifdef DL_KERNELS
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Activation_Mul_Clamp<PassThrough>>>>&
+    std::vector<
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        Empty_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Activation_Mul_Clamp<PassThrough>>>>&
         instances);
 
 void add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Activation_Mul_Clamp<Relu>>>>&
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                Activation_Mul_Clamp<Relu>>>>&
         instances);
 #endif
 void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Activation_Mul_Clamp<PassThrough>>>>&
+    std::vector<
+        std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        Empty_Tuple,
+                                                        NHWGK,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        Activation_Mul_Clamp<PassThrough>>>>&
         instances);
 
 void add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Activation_Mul_Clamp<Relu>>>>&
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
+                                                                NHWGC,
+                                                                GKYXC,
+                                                                Empty_Tuple,
+                                                                NHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                Activation_Mul_Clamp<Relu>>>>&
         instances);
 
 template <ck::index_t NumDimSpatial,
@@ -86,7 +88,7 @@ template <ck::index_t NumDimSpatial,
           typename WeiDataType,
           typename OutDataType,
           typename Activation>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
     NumDimSpatial,
     InLayout,
     WeiLayout,
@@ -100,18 +102,19 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     ck::tensor_operation::element_wise::PassThrough,
     Activation_Mul_Clamp<Activation>>>
 {
-    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,
-                                                   Activation_Mul_Clamp<Activation>>;
+    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,
+                                        Activation_Mul_Clamp<Activation>>;
 
     static auto GetInstances()
     {