Merge branch 'develop' into gemm_v2r3_kpad_fix

include/ck/utility/type_convert.hpp

@@ -80,46 +80,158 @@ inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int8_t>(int8_
     return type_convert<bhalf_t>(x_fp32);
 }
 
+#if defined CK_ENABLE_FP8
 // convert fp32 to fp8
 template <>
 inline __host__ __device__ f8_t type_convert<f8_t, float>(float x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    union
+    {
+        float fval;
+        uint32_t i32val;
+        uint8_t i8val[4]; // not endian independent
+    } val;
+    val.fval      = x;
+    uint32_t ival = 0;
+    ival       = __builtin_amdgcn_cvt_pk_fp8_f32(val.fval, val.fval, ival, false); // false -> WORD0
+    val.i32val = ival;
+    return val.i8val[0];
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
     constexpr uint32_t rng           = 0;
-    return utils::cast_to_f8<float, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
-        x, rng);
+    return utils::
+        cast_to_f8<float, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(x,
+                                                                                               rng);
+#endif
 }
 
 // convert fp8 to fp32
 template <>
 inline __host__ __device__ float type_convert<float, f8_t>(f8_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    float fval;
+    uint32_t i32val = static_cast<uint32_t>(x);
+    fval            = __builtin_amdgcn_cvt_f32_fp8(i32val, 0);
+    // asm volatile("v_cvt_f32_fp8 %0, %1 src0_sel:BYTE_0" : "=v"(fval) : "v"(i32val));
+    return fval;
+#else
     constexpr bool negative_zero_nan = true;
-    return utils::cast_from_f8<float, negative_zero_nan>(x);
+    return utils::cast_from_f8<f8_t, float, negative_zero_nan>(x);
+#endif
 }
 
 // convert fp16 to fp8
 template <>
 inline __host__ __device__ f8_t type_convert<f8_t, half_t>(half_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // convert to float and use native converion
+    return type_convert<f8_t>(type_convert<float>(x));
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
     constexpr uint32_t rng           = 0;
-    return utils::cast_to_f8<half_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+    return utils::
+        cast_to_f8<half_t, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#endif
 }
 
 // convert fp8 to fp16
 template <>
 inline __host__ __device__ half_t type_convert<half_t, f8_t>(f8_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // use native conversion to float and convert to fp16
+    return type_convert<half_t>(type_convert<float>(x));
+#else
+    constexpr bool negative_zero_nan = true;
+    return utils::cast_from_f8<f8_t, half_t, negative_zero_nan>(x);
+#endif
+}
+#endif
+
+#if defined CK_ENABLE_BF8
+// convert fp32 to bf8
+template <>
+inline __host__ __device__ bf8_t type_convert<bf8_t, float>(float x)
+{
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    union
+    {
+        float fval;
+        uint32_t i32val;
+        uint8_t i8val[4]; // not endian independent
+    } val;
+    val.fval      = x;
+    uint32_t ival = 0;
+    ival       = __builtin_amdgcn_cvt_pk_bf8_f32(val.fval, val.fval, ival, false); // false -> WORD0
+    val.i32val = ival;
+    return val.i8val[0];
+#else
     constexpr bool negative_zero_nan = true;
-    return utils::cast_from_f8<half_t, negative_zero_nan>(x);
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
+    constexpr uint32_t rng           = 0;
+    return utils::
+        cast_to_f8<float, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+            x, rng);
+#endif
 }
 
+// convert bf8 to fp32
+template <>
+inline __host__ __device__ float type_convert<float, bf8_t>(bf8_t x)
+{
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    float fval;
+    uint32_t i32val = static_cast<uint32_t>(x);
+    fval            = __builtin_amdgcn_cvt_f32_bf8(i32val, 0);
+    // asm volatile("v_cvt_f32_bf8 %0, %1 src0_sel:BYTE_0" : "=v"(fval) : "v"(i32val));
+    return fval;
+#else
+    constexpr bool negative_zero_nan = true;
+    return utils::cast_from_f8<bf8_t, float, negative_zero_nan>(x);
+#endif
+}
+
+// convert fp16 to bf8
+template <>
+inline __host__ __device__ bf8_t type_convert<bf8_t, half_t>(half_t x)
+{
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // convert to float and use native converion
+    return type_convert<f8_t>(type_convert<float>(x));
+#else
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
+    constexpr uint32_t rng           = 0;
+    return utils::
+        cast_to_f8<half_t, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+            x, rng);
+#endif
+}
+
+// convert bf8 to fp16
+template <>
+inline __host__ __device__ half_t type_convert<half_t, bf8_t>(bf8_t x)
+{
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // use native conversion to float and convert to fp16
+    return type_convert<half_t>(type_convert<float>(x));
+#else
+    constexpr bool negative_zero_nan = true;
+    return utils::cast_from_f8<bf8_t, half_t, negative_zero_nan>(x);
+#endif
+}
+#endif
+
 // Declare a template function for bf16 conversion using RTN
 template <typename Y, typename X>
 __host__ __device__ constexpr Y bf16_convert_rtn(X x);
@@ -181,32 +293,103 @@ inline __host__ __device__ constexpr bhalf_t bf16_convert_rtn<bhalf_t, half_t>(h
 template <typename Y, typename X>
 __host__ __device__ constexpr Y f8_convert_sr(X x);
 
+#if defined CK_ENABLE_FP8
 // convert fp32 to fp8 with stochastic rounding
 template <>
 inline __host__ __device__ f8_t f8_convert_sr<f8_t, float>(float x)
 {
+    constexpr int seed = 42;
+    uint32_t rng       = prand_generator<float, seed>(reinterpret_cast<uintptr_t>(&x), x);
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    union
+    {
+        float fval;
+        uint32_t i32val;
+        uint8_t i8val[4]; // not endian independent
+    } val;
+    val.fval      = x;
+    uint32_t ival = 0;
+    ival          = __builtin_amdgcn_cvt_sr_fp8_f32(val.fval, rng, ival, 0); // 0 pos
+    val.i32val    = ival;
+    return val.i8val[0]; // little endian
+#else
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
+    return utils::
+        cast_to_f8<float, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(x,
+                                                                                               rng);
+#endif
+}
+
+// convert fp16 to fp8 with stochastic rounding
+template <>
+inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
+{
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // convert to float and use native converion
+    return f8_convert_sr<f8_t>(type_convert<float>(x));
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
     constexpr int seed               = 42;
-    // as thread id is not available on host, use 0 for prn generation
+    uint32_t rng = prand_generator<half_t, seed>(reinterpret_cast<uintptr_t>(&x), x);
+    return utils::
+        cast_to_f8<half_t, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+            x, rng);
+#endif
+}
+#endif
+
+#if defined CK_ENABLE_BF8
+// convert fp32 to bf8 with stochastic rounding
+template <>
+inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, float>(float x)
+{
+    constexpr int seed = 42;
     uint32_t rng       = prand_generator<float, seed>(reinterpret_cast<uintptr_t>(&x), x);
-    return utils::cast_to_f8<float, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    union
+    {
+        float fval;
+        uint32_t i32val;
+        uint8_t i8val[4]; // not endian independent
+    } val;
+    val.fval      = x;
+    uint32_t ival = 0;
+    ival          = __builtin_amdgcn_cvt_sr_bf8_f32(val.fval, rng, ival, 0); // 0 pos
+    val.i32val    = ival;
+    return val.i8val[0]; // little endian
+#else
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
+    return utils::
+        cast_to_f8<float, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#endif
 }
 
-// convert fp16 to fp8 with stochastic rounding
+// convert fp16 to bf8 with stochastic rounding
 template <>
-inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
+inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, half_t>(half_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // convert to float and use native converion
+    return f8_convert_sr<f8_t>(type_convert<float>(x));
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
     constexpr int seed               = 42;
     // as thread id is not available on host, use 0 for prn generation
     uint32_t rng = prand_generator<half_t, seed>(reinterpret_cast<uintptr_t>(&x), x);
-    return utils::cast_to_f8<half_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+    return utils::
+        cast_to_f8<half_t, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#endif
 }
+#endif
 
 } // namespace ck

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

@@ -20,7 +20,8 @@ template <typename ADataType,
           typename AccDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
-          typename CElementwiseOperation>
+          typename CElementwiseOperation,
+          typename ComputType = ADataType>
 struct ReferenceGemm : public device::BaseOperator
 {
     // Argument
@@ -64,8 +65,8 @@ struct ReferenceGemm : public device::BaseOperator
 
                 for(int k = 0; k < K; ++k)
                 {
-                    ADataType v_a;
-                    BDataType v_b;
+                    ComputType v_a;
+                    ComputType v_b;
 
                     // use PassThrough instead of ConvertBF16RTN for reference calculation
                     if constexpr(is_same_v<AElementwiseOperation,

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <type_traits>
+#include <sstream>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+
+/**
+ * \brief Reference implementation for image to column.
+ *
+ * Tensor descriptor has [G, N, C, Di, Hi, Wi] data layout.
+ * G must be equal to 1. Memory layout is [G, N, Di, Hi, Wi, C].
+ *
+ * \tparam NDimSpatial Number of spatial dimensions.
+ * \tparam InputLayout Input Layout.
+ * \tparam InDataType Input Data Type.
+ * \tparam OutDataType Output Data Type.
+ */
+template <ck::index_t NDimSpatial,
+          typename InputLayout,
+          typename InDataType,
+          typename OutDataType,
+          typename std::enable_if<NDimSpatial >= 1 && NDimSpatial <= 3, bool>::type = false>
+struct ReferenceImageToColumn : public device::BaseOperator
+{
+    // Argument
+    struct Argument : public device::BaseArgument
+    {
+        public:
+        Argument(const Tensor<InDataType>& input,
+                 Tensor<OutDataType>& output,
+                 std::vector<ck::index_t> filter_spatial_lengths,
+                 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)
+            : input_{input},
+              output_{output},
+              conv_strides_{conv_filter_strides},
+              conv_dilations_{conv_filter_dilations},
+              in_left_pads_{input_left_pads},
+              in_right_pads_{input_right_pads},
+              filter_spatial_lengths_{filter_spatial_lengths}
+        {
+            initOutputSpatialLengths();
+        }
+
+        const Tensor<InDataType>& input_;
+        Tensor<OutDataType>& output_;
+
+        std::vector<index_t> conv_strides_;
+        std::vector<index_t> conv_dilations_;
+        std::vector<index_t> in_left_pads_;
+        std::vector<index_t> in_right_pads_;
+
+        std::vector<index_t> filter_spatial_lengths_;
+        std::vector<index_t> output_spatial_lengths_;
+
+        private:
+        void initOutputSpatialLengths()
+        {
+            constexpr auto input_offset_to_spatial = 3;
+
+            for(ck::index_t i = 0; i < NDimSpatial; ++i)
+            {
+                // XEff = (X - 1) * conv_dilation_w + 1;
+                // Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
+                const ck::index_t x_eff = (filter_spatial_lengths_[i] - 1) * conv_dilations_[i] + 1;
+
+                output_spatial_lengths_.push_back(
+                    (input_.GetLengths()[i + input_offset_to_spatial] + in_left_pads_[i] +
+                     in_right_pads_[i] - x_eff) /
+                        conv_strides_[i] +
+                    1);
+            }
+        }
+    };
+
+    struct Invoker : public device::BaseInvoker
+    {
+        using Argument = ReferenceImageToColumn::Argument;
+
+        float Run(const Argument& arg)
+        {
+            if(!(arg.input_.GetNumOfDimension() == NDimSpatial + 3 &&
+                 arg.output_.GetNumOfDimension() == 2))
+            {
+                throw std::runtime_error("wrong! inconsistent dimension");
+            }
+
+            const index_t N = arg.input_.GetLengths()[1];
+            const index_t C = arg.input_.GetLengths()[2];
+
+            if constexpr(NDimSpatial == 1)
+            {
+                const index_t Wo = arg.output_spatial_lengths_[0];
+                auto func        = [&](auto n, auto wo) {
+                    index_t row    = n * Wo + wo;
+                    index_t column = 0;
+
+                    for(index_t x = 0; x < arg.filter_spatial_lengths_[0]; ++x)
+                    {
+                        auto wi = static_cast<ck::long_index_t>(wo * arg.conv_strides_[0]) +
+                                  static_cast<ck::long_index_t>(x * arg.conv_dilations_[0]) -
+                                  static_cast<ck::long_index_t>(arg.in_left_pads_[0]);
+
+                        for(index_t c = 0; c < C; ++c)
+                        {
+                            if(wi >= 0 &&
+                               ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[3])
+                            {
+                                InDataType v_in          = arg.input_(0, n, c, wi);
+                                arg.output_(row, column) = ck::type_convert<OutDataType>(v_in);
+                            }
+                            column++;
+                        }
+                    }
+                };
+
+                make_ParallelTensorFunctor(func, N, Wo)(std::thread::hardware_concurrency());
+
+                return 0;
+            }
+            else if constexpr(NDimSpatial == 2)
+            {
+                const index_t Ho = arg.output_spatial_lengths_[0];
+                const index_t Wo = arg.output_spatial_lengths_[1];
+
+                auto func = [&](auto n, auto ho, auto wo) {
+                    index_t row    = n * Ho * Wo + ho * Wo + wo;
+                    index_t column = 0;
+
+                    for(index_t y = 0; y < arg.filter_spatial_lengths_[0]; ++y)
+                    {
+                        auto hi = static_cast<ck::long_index_t>(ho * arg.conv_strides_[0]) +
+                                  static_cast<ck::long_index_t>(y * arg.conv_dilations_[0]) -
+                                  static_cast<ck::long_index_t>(arg.in_left_pads_[0]);
+
+                        for(index_t x = 0; x < arg.filter_spatial_lengths_[1]; ++x)
+                        {
+                            auto wi = static_cast<ck::long_index_t>(wo * arg.conv_strides_[1]) +
+                                      static_cast<ck::long_index_t>(x * arg.conv_dilations_[1]) -
+                                      static_cast<ck::long_index_t>(arg.in_left_pads_[1]);
+
+                            for(index_t c = 0; c < C; ++c)
+                            {
+
+                                if(hi >= 0 &&
+                                   ck::type_convert<std::size_t>(hi) < arg.input_.GetLengths()[3] &&
+                                   wi >= 0 &&
+                                   ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[4])
+                                {
+                                    InDataType v_in          = arg.input_(0, n, c, hi, wi);
+                                    arg.output_(row, column) = ck::type_convert<OutDataType>(v_in);
+                                }
+                                column++;
+                            }
+                        }
+                    }
+                };
+
+                make_ParallelTensorFunctor(func, N, Ho, Wo)(std::thread::hardware_concurrency());
+
+                return 0;
+            }
+            else if constexpr(NDimSpatial == 3)
+            {
+                const index_t Do = arg.output_spatial_lengths_[0];
+                const index_t Ho = arg.output_spatial_lengths_[1];
+                const index_t Wo = arg.output_spatial_lengths_[2];
+
+                auto func = [&](auto n, auto d_o, auto ho, auto wo) {
+                    index_t row    = n * Do * Ho * Wo + d_o * Ho * Wo + ho * Wo + wo;
+                    index_t column = 0;
+
+                    for(index_t z = 0; z < arg.filter_spatial_lengths_[0]; ++z)
+                    {
+                        auto di = static_cast<ck::long_index_t>(d_o * arg.conv_strides_[0]) +
+                                  static_cast<ck::long_index_t>(z * arg.conv_dilations_[0]) -
+                                  static_cast<ck::long_index_t>(arg.in_left_pads_[0]);
+                        for(index_t y = 0; y < arg.filter_spatial_lengths_[1]; ++y)
+                        {
+                            auto hi = static_cast<ck::long_index_t>(ho * arg.conv_strides_[1]) +
+                                      static_cast<ck::long_index_t>(y * arg.conv_dilations_[1]) -
+                                      static_cast<ck::long_index_t>(arg.in_left_pads_[1]);
+                            for(index_t x = 0; x < arg.filter_spatial_lengths_[2]; ++x)
+                            {
+                                auto wi =
+                                    static_cast<ck::long_index_t>(wo * arg.conv_strides_[2]) +
+                                    static_cast<ck::long_index_t>(x * arg.conv_dilations_[2]) -
+                                    static_cast<ck::long_index_t>(arg.in_left_pads_[2]);
+                                for(index_t c = 0; c < C; ++c)
+                                {
+                                    if(di >= 0 &&
+                                       ck::type_convert<std::size_t>(di) <
+                                           arg.input_.GetLengths()[3] &&
+                                       hi >= 0 &&
+                                       ck::type_convert<std::size_t>(hi) <
+                                           arg.input_.GetLengths()[4] &&
+                                       wi >= 0 &&
+                                       ck::type_convert<std::size_t>(wi) <
+                                           arg.input_.GetLengths()[5])
+                                    {
+                                        InDataType v_in = arg.input_(0, n, c, di, hi, wi);
+                                        arg.output_(row, column) =
+                                            ck::type_convert<OutDataType>(v_in);
+                                    }
+                                    column++;
+                                }
+                            }
+                        }
+                    }
+                };
+
+                make_ParallelTensorFunctor(func, N, Do, Ho, Wo)(
+                    std::thread::hardware_concurrency());
+
+                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()
+    {
+        using namespace tensor_layout::convolution;
+
+        if constexpr(!(std::is_same_v<InputLayout, GNWC> || std::is_same_v<InputLayout, GNHWC> ||
+                       std::is_same_v<InputLayout, GNDHWC>))
+        {
+            return false;
+        }
+        if constexpr(!(NDimSpatial >= 1 && NDimSpatial <= 3))
+        {
+            return false;
+        }
+        return true;
+    }
+
+    bool IsSupportedArgument(const Argument& arg)
+    {
+        const ck::index_t G = arg.input_.GetLengths()[0];
+        const ck::index_t N = arg.input_.GetLengths()[1];
+        const ck::index_t C = arg.input_.GetLengths()[2];
+
+        const index_t NDoHoWo =
+            N * ck::accumulate_n<index_t>(
+                    arg.output_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+        const index_t CZYX =
+            C * ck::accumulate_n<index_t>(
+                    arg.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+
+        if(!(arg.output_.GetLengths()[0] == static_cast<std::size_t>(NDoHoWo) &&
+             arg.output_.GetLengths()[1] == static_cast<std::size_t>(CZYX)))
+        {
+            return false;
+        }
+
+        if(G != 1)
+        {
+            return false;
+        }
+        return true;
+    }
+
+    bool IsSupportedArgument(const device::BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const Tensor<InDataType>& input,
+                             Tensor<OutDataType>& output,
+                             std::vector<ck::index_t> filter_spatial_lengths,
+                             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)
+    {
+        return Argument{input,
+                        output,
+                        filter_spatial_lengths,
+                        conv_filter_strides,
+                        conv_filter_dilations,
+                        input_left_pads,
+                        input_right_pads};
+    }
+
+    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 << "ReferenceImageToColumn"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck

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

@@ -17,10 +17,15 @@ namespace instance {
 using F64  = double;
 using F32  = float;
 using F16  = ck::half_t;
-using F8   = ck::f8_t;
 using BF16 = ck::bhalf_t;
 using I8   = int8_t;
 using I32  = int32_t;
+#if defined CK_ENABLE_FP8
+using F8 = ck::f8_t;
+#endif
+#if defined CK_ENABLE_BF8
+using BF8 = ck::bf8_t;
+#endif
 
 using Empty_Tuple = ck::Tuple<>;
 
@@ -31,6 +36,7 @@ using F64_Tuple     = ck::Tuple<F64>;
 using F32_Tuple     = ck::Tuple<F32>;
 using I32_Tuple     = ck::Tuple<I32>;
 using I32_F32_Tuple = ck::Tuple<I32, F32>;
+using I8_Tuple      = ck::Tuple<I8>;
 
 using F32_F32_Tuple = ck::Tuple<F32, F32>;
 

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

@@ -16,26 +16,26 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// FP16
+#ifdef CK_ENABLE_FP16
 void add_device_batchnorm_backward_rank_4_3_f16_instances(
     std::vector<std::unique_ptr<
         DeviceBatchNormBwd<F16, F32, F32, F32, F16, F32, F32, PassThrough, 4, 3>>>&);
-
-// FP32
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_batchnorm_backward_rank_4_3_f32_instances(
     std::vector<std::unique_ptr<
         DeviceBatchNormBwd<F32, F32, F32, F32, F32, F32, F32, PassThrough, 4, 3>>>&);
-
-// BF16
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_batchnorm_backward_rank_4_3_bf16_instances(
     std::vector<std::unique_ptr<
         DeviceBatchNormBwd<BF16, F32, F32, F32, BF16, F32, F32, PassThrough, 4, 3>>>&);
-
-// FP64
+#endif
+#ifdef CK_ENABLE_FP64
 void add_device_batchnorm_backward_rank_4_3_f64_instances(
     std::vector<std::unique_ptr<
         DeviceBatchNormBwd<F64, F64, F64, F64, F64, F64, F64, PassThrough, 4, 3>>>&);
-
+#endif
 template <typename XDataType,
           typename DxDataType,
           typename DyDataType,
@@ -72,7 +72,7 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<XDataType, F16> && is_same_v<DxDataType, F32> &&
                      is_same_v<DyDataType, F32> && is_same_v<AccDataType, F32> &&
                      is_same_v<ScaleDataType, F16> && is_same_v<DscaleDbiasDataType, F32> &&
@@ -83,7 +83,9 @@ struct DeviceOperationInstanceFactory<
                 add_device_batchnorm_backward_rank_4_3_f16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F32> && is_same_v<DxDataType, F32> &&
+#endif
+#ifdef CK_ENABLE_FP32
+        if constexpr(is_same_v<XDataType, F32> && is_same_v<DxDataType, F32> &&
                      is_same_v<DyDataType, F32> && is_same_v<AccDataType, F32> &&
                      is_same_v<ScaleDataType, F32> && is_same_v<DscaleDbiasDataType, F32> &&
                      is_same_v<MeanVarDataType, F32>)
@@ -93,7 +95,9 @@ struct DeviceOperationInstanceFactory<
                 add_device_batchnorm_backward_rank_4_3_f32_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, BF16> && is_same_v<DxDataType, F32> &&
+#endif
+#ifdef CK_ENABLE_BF16
+        if constexpr(is_same_v<XDataType, BF16> && is_same_v<DxDataType, F32> &&
                      is_same_v<DyDataType, F32> && is_same_v<AccDataType, F32> &&
                      is_same_v<ScaleDataType, BF16> && is_same_v<DscaleDbiasDataType, F32> &&
                      is_same_v<MeanVarDataType, F32>)
@@ -103,7 +107,9 @@ struct DeviceOperationInstanceFactory<
                 add_device_batchnorm_backward_rank_4_3_bf16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F64> && is_same_v<DxDataType, F64> &&
+#endif
+#ifdef CK_ENABLE_FP64
+        if constexpr(is_same_v<XDataType, F64> && is_same_v<DxDataType, F64> &&
                      is_same_v<DyDataType, F64> && is_same_v<AccDataType, F64> &&
                      is_same_v<ScaleDataType, F64> && is_same_v<DscaleDbiasDataType, F64> &&
                      is_same_v<MeanVarDataType, F64>)
@@ -113,7 +119,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_batchnorm_backward_rank_4_3_f64_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };

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

@@ -16,26 +16,26 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// FP16
+#ifdef CK_ENABLE_FP16
 void add_device_batchnorm_forward_rank_4_3_f16_instances(
     std::vector<
         std::unique_ptr<DeviceBatchNormFwd<F16, F16, F32, F16, F16, F32, PassThrough, 4, 3>>>&);
-
-// FP32
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_batchnorm_forward_rank_4_3_f32_instances(
     std::vector<
         std::unique_ptr<DeviceBatchNormFwd<F32, F32, F32, F32, F32, F32, PassThrough, 4, 3>>>&);
-
-// BF16
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_batchnorm_forward_rank_4_3_bf16_instances(
     std::vector<
         std::unique_ptr<DeviceBatchNormFwd<BF16, BF16, F32, BF16, BF16, F32, PassThrough, 4, 3>>>&);
-
-// FP64
+#endif
+#ifdef CK_ENABLE_FP64
 void add_device_batchnorm_forward_rank_4_3_f64_instances(
     std::vector<
         std::unique_ptr<DeviceBatchNormFwd<F64, F64, F64, F64, F64, F64, PassThrough, 4, 3>>>&);
-
+#endif
 template <typename XDataType,
           typename YDataType,
           typename AccDataType,
@@ -69,7 +69,7 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<XDataType, F16> && is_same_v<YDataType, F16> &&
                      is_same_v<AccDataType, F32> && is_same_v<ScaleDataType, F16> &&
                      is_same_v<BiasDataType, F16> && is_same_v<MeanVarDataType, F32>)
@@ -79,7 +79,9 @@ struct DeviceOperationInstanceFactory<
                 add_device_batchnorm_forward_rank_4_3_f16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F32> && is_same_v<YDataType, F32> &&
+#endif
+#ifdef CK_ENABLE_FP32
+        if constexpr(is_same_v<XDataType, F32> && is_same_v<YDataType, F32> &&
                      is_same_v<AccDataType, F32> && is_same_v<ScaleDataType, F32> &&
                      is_same_v<BiasDataType, F32> && is_same_v<MeanVarDataType, F32>)
         {
@@ -88,7 +90,9 @@ struct DeviceOperationInstanceFactory<
                 add_device_batchnorm_forward_rank_4_3_f32_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, BF16> && is_same_v<YDataType, BF16> &&
+#endif
+#ifdef CK_ENABLE_BF16
+        if constexpr(is_same_v<XDataType, BF16> && is_same_v<YDataType, BF16> &&
                      is_same_v<AccDataType, F32> && is_same_v<ScaleDataType, BF16> &&
                      is_same_v<BiasDataType, BF16> && is_same_v<MeanVarDataType, F32>)
         {
@@ -97,7 +101,9 @@ struct DeviceOperationInstanceFactory<
                 add_device_batchnorm_forward_rank_4_3_bf16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F64> && is_same_v<YDataType, F64> &&
+#endif
+#ifdef CK_ENABLE_FP64
+        if constexpr(is_same_v<XDataType, F64> && is_same_v<YDataType, F64> &&
                      is_same_v<AccDataType, F64> && is_same_v<ScaleDataType, F64> &&
                      is_same_v<BiasDataType, F64> && is_same_v<MeanVarDataType, F64>)
         {
@@ -106,7 +112,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_batchnorm_forward_rank_4_3_f64_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };

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

@@ -16,38 +16,38 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// FP16
+#ifdef CK_ENABLE_FP16
 void add_device_batchnorm_infer_rank_4_f16_instances(
     std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
         ck::Tuple<F16, F32, F32, F16, F16>,
         ck::Tuple<F16>,
         ck::tensor_operation::element_wise::NormalizeInInfer,
         4>>>&);
-
-// FP32
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_batchnorm_infer_rank_4_f32_instances(
     std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
         ck::Tuple<F32, F32, F32, F32, F32>,
         ck::Tuple<F32>,
         ck::tensor_operation::element_wise::NormalizeInInfer,
         4>>>&);
-
-// BF16
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_batchnorm_infer_rank_4_bf16_instances(
     std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
         ck::Tuple<BF16, F32, F32, BF16, BF16>,
         ck::Tuple<BF16>,
         ck::tensor_operation::element_wise::NormalizeInInfer,
         4>>>&);
-
-// FP64
+#endif
+#ifdef CK_ENABLE_FP64
 void add_device_batchnorm_infer_rank_4_f64_instances(
     std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
         ck::Tuple<F64, F64, F64, F64, F64>,
         ck::Tuple<F64>,
         ck::tensor_operation::element_wise::NormalizeInInfer,
         4>>>&);
-
+#endif
 template <typename XDataType,
           typename YDataType,
           typename ScaleDataType,
@@ -69,7 +69,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceElemen
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<XDataType, F16> && is_same_v<YDataType, F16> &&
                      is_same_v<ScaleDataType, F16> && is_same_v<BiasDataType, F16> &&
                      is_same_v<MeanVarDataType, F32>)
@@ -79,7 +79,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceElemen
                 add_device_batchnorm_infer_rank_4_f16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F32> && is_same_v<YDataType, F32> &&
+#endif
+#ifdef CK_ENABLE_FP32
+        if constexpr(is_same_v<XDataType, F32> && is_same_v<YDataType, F32> &&
                      is_same_v<ScaleDataType, F32> && is_same_v<BiasDataType, F32> &&
                      is_same_v<MeanVarDataType, F32>)
         {
@@ -88,7 +90,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceElemen
                 add_device_batchnorm_infer_rank_4_f32_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, BF16> && is_same_v<YDataType, BF16> &&
+#endif
+#ifdef CK_ENABLE_BF16
+        if constexpr(is_same_v<XDataType, BF16> && is_same_v<YDataType, BF16> &&
                      is_same_v<ScaleDataType, BF16> && is_same_v<BiasDataType, BF16> &&
                      is_same_v<MeanVarDataType, F32>)
         {
@@ -97,7 +101,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceElemen
                 add_device_batchnorm_infer_rank_4_bf16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F64> && is_same_v<YDataType, F64> &&
+#endif
+#ifdef CK_ENABLE_FP64
+        if constexpr(is_same_v<XDataType, F64> && is_same_v<YDataType, F64> &&
                      is_same_v<ScaleDataType, F64> && is_same_v<BiasDataType, F64> &&
                      is_same_v<MeanVarDataType, F64>)
         {
@@ -106,7 +112,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceElemen
                 add_device_batchnorm_infer_rank_4_f64_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };

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

@@ -23,12 +23,17 @@ void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances(
+void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
+void add_device_gemm_dpp_f16_f16_f16_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -38,12 +43,17 @@ void add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances(
+void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
+void add_device_gemm_dpp_f16_f16_f16_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -53,12 +63,17 @@ void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances(
+void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
+void add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -68,12 +83,17 @@ void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances(
+void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
+void add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -292,6 +312,23 @@ void add_device_gemm_xdl_f64_f64_f64_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F64, F64, F64, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
+#ifdef CK_ENABLE_FP8
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
+
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
+
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
+
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
+#endif
 template <typename ALayout,
           typename BLayout,
           typename CLayout,
@@ -374,7 +411,8 @@ struct DeviceOperationInstanceFactory<
 #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);
-                add_device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
             }
@@ -385,7 +423,8 @@ struct DeviceOperationInstanceFactory<
 #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);
-                add_device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
@@ -397,7 +436,8 @@ struct DeviceOperationInstanceFactory<
 #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);
-                add_device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_km_kn_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs);
             }
@@ -408,7 +448,8 @@ struct DeviceOperationInstanceFactory<
 #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);
-                add_device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_km_nk_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
@@ -481,6 +522,32 @@ struct DeviceOperationInstanceFactory<
 #endif
             }
         }
+#endif
+#ifdef CK_ENABLE_FP8
+        else if constexpr(is_same_v<ADataType, ck::f8_t> && is_same_v<BDataType, ck::f8_t> &&
+                          is_same_v<CDataType, ck::f8_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances(op_ptrs);
+            }
+        }
 #endif
         return op_ptrs;
     }

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

@@ -11,12 +11,12 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef CK_ENABLE_FP16
+
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef CK_ENABLE_FP16
 void add_device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
                                                     Row,
@@ -68,7 +68,60 @@ void add_device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance
                                                     PassThrough,
                                                     PassThrough,
                                                     Bilinear>>>& instances);
+#endif
+#ifdef CK_ENABLE_INT8
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances);
+
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances);
 
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Row,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances);
+
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Col,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances);
+#endif
 // GEMM + Bilinear
 template <typename ALayout,
           typename BLayout,
@@ -106,7 +159,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                      is_same_v<DDataType, half_t> && is_same_v<EDataType, half_t>)
         {
@@ -135,7 +188,33 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                     op_ptrs);
             }
         }
-
+#endif
+#ifdef CK_ENABLE_INT8
+        if constexpr(is_same_v<ADataType, std::int8_t> && is_same_v<BDataType, std::int8_t> &&
+                     is_same_v<DDataType, std::int8_t> && is_same_v<EDataType, std::int8_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(op_ptrs);
+            }
+        }
+#endif
         return op_ptrs;
     }
 };
@@ -144,4 +223,3 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
-#endif

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

@@ -45,6 +45,7 @@ void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_
                                                     PassThrough,
                                                     MultiplyAdd>>>&);
 
+#if defined CK_ENABLE_FP8
 void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
                                                     Row,
@@ -70,6 +71,7 @@ void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_m
                                                     PassThrough,
                                                     PassThrough,
                                                     MultiplyAdd>>>&);
+#endif
 
 // GEMM + Multiply + Add
 template <typename ALayout,
@@ -131,6 +133,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
             }
         }
 
+#if defined CK_ENABLE_FP8
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, f8_t> &&
                      is_same_v<D0DataType, float> && is_same_v<D1DataType, float> &&
                      is_same_v<EDataType, half_t>)
@@ -150,6 +153,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                     op_ptrs);
             }
         }
+#endif
 
         return op_ptrs;
     }

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

@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef CK_ENABLE_FP16
 void add_device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -36,7 +36,8 @@ void add_device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -56,7 +57,8 @@ void add_device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
+#if(defined(CK_ENABLE_FP16) || defined(CK_ENABLE_FP8))
 void add_device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Col, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -96,6 +98,7 @@ void add_device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Row, Col, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
+#endif
 
 template <typename ADataType,
           typename BDataType,
@@ -127,7 +130,7 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
                      is_same_v<CDataType, float>)
         {
@@ -152,6 +155,8 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instances(op_ptrs);
             }
         }
+#endif
+#ifdef CK_ENABLE_FP16
         else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                           is_same_v<CDataType, half_t>)
         {
@@ -176,6 +181,8 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
         }
+#endif
+#if(defined(CK_ENABLE_FP16) || defined(CK_ENABLE_FP8))
         else if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, half_t> &&
                           is_same_v<CDataType, half_t>)
         {
@@ -224,7 +231,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };

library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, 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_bwd_weight_dl.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::tensor_layout::convolution;
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvBwdWeightDefault =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
+
+static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_dl_f32_instances = std::tuple<
+    // clang-format off
+        //############################|        Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|              ConvBackward| Block|  MPer|  NPer| K0Per| K1|  M1Per|  N1Per|   KPer|  M1N1Thread|  M1N1Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|         ABlockTransfer|     ABlockTransfer|         ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|         BBlockTransfer|     BBlockTransfer|         BBlockTransfer|   CThreadTransfer| CThreadTransfer|    CThreadTransfer| 
+        //############################|        Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                    Weight|  Size| Block| Block| Block|   | Thread| Thread| Thread| ClusterM1Xs| ClusterN1Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| SrcDstAccessOrder| SrcDstVectorDim| DstScalarPerVector|
+        //############################|    Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|            Specialization|      |      |      |      |   |       |       |       |            |            |       _K0_M0_M1_K1|         _K0_M0_M1_K1|   ArrangeOrder|               |           _K0_M0_M1_K1| ContiguousDimOrder|           _K0_M0_M1_K1|       _K0_N0_N1_K1|         _K0_N0_N1_K1|   ArrangeOrder|               |           _K0_N0_N1_K1| ContiguousDimOrder|           _K0_N0_N1_K1|                  |                |                   |
+        //############################|           |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |       |       |       |            |            |                   |                     |               |               |                       |                   |                       |                   |                     |               |               |                       |                   |                       |                  |                |                   |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Dl< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,    16,  1,      4,      4,      1,     S<8, 2>,     S<8, 2>,   S<1, 8, 1, 1, 1>,   S<1, 2, 1, 128, 1>, S<0, 2, 3, 1, 4>, S<0, 2, 3, 1, 4>,   S<1, 1, 1, 1, 1>,   S<0, 2, 3, 1, 4>,       S<1, 1, 1, 1, 1>,   S<1, 1, 1, 8, 1>,   S<1, 16, 1, 16, 1>, S<0, 1, 4, 2, 3>, S<0, 1, 4, 2, 3>,   S<1, 1, 1, 1, 1>,   S<0, 1, 4, 2, 3>,       S<1, 1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,             5,                   1>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_dl_f16_instances = std::tuple<
+    // clang-format off
+        //############################|        Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|              ConvBackward| Block|  MPer|  NPer| K0Per| K1|  M1Per|  N1Per|   KPer|  M1N1Thread|  M1N1Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|         ABlockTransfer|     ABlockTransfer|         ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|         BBlockTransfer|     BBlockTransfer|         BBlockTransfer|   CThreadTransfer| CThreadTransfer|    CThreadTransfer| 
+        //############################|        Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                    Weight|  Size| Block| Block| Block|   | Thread| Thread| Thread| ClusterM1Xs| ClusterN1Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| SrcDstAccessOrder| SrcDstVectorDim| DstScalarPerVector|
+        //############################|    Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|            Specialization|      |      |      |      |   |       |       |       |            |            |       _K0_M0_M1_K1|         _K0_M0_M1_K1|   ArrangeOrder|               |           _K0_M0_M1_K1| ContiguousDimOrder|           _K0_M0_M1_K1|       _K0_N0_N1_K1|         _K0_N0_N1_K1|   ArrangeOrder|               |           _K0_N0_N1_K1| ContiguousDimOrder|           _K0_N0_N1_K1|                  |                |                   |
+        //############################|           |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |       |       |       |            |            |                   |                     |               |               |                       |                   |                       |                   |                     |               |               |                       |                   |                       |                  |                |                   |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Dl< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,    16,  1,      4,      4,      1,     S<8, 2>,     S<8, 2>,   S<1, 8, 1, 1, 1>,   S<1, 2, 1, 128, 1>, S<0, 2, 3, 1, 4>, S<0, 2, 3, 1, 4>,   S<1, 1, 1, 1, 1>,   S<0, 2, 3, 1, 4>,       S<1, 1, 1, 1, 1>,   S<1, 1, 1, 8, 1>,   S<1, 16, 1, 16, 1>, S<0, 1, 4, 2, 3>, S<0, 1, 4, 2, 3>,   S<1, 1, 1, 1, 1>,   S<0, 1, 4, 2, 3>,       S<1, 1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,             5,                   1>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_dl_bf16_instances = std::tuple<
+    // clang-format off
+        //############################|        Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|              ConvBackward| Block|  MPer|  NPer| K0Per| K1|  M1Per|  N1Per|   KPer|  M1N1Thread|  M1N1Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|         ABlockTransfer|     ABlockTransfer|         ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|         BBlockTransfer|     BBlockTransfer|         BBlockTransfer|   CThreadTransfer| CThreadTransfer|    CThreadTransfer| 
+        //############################|        Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                    Weight|  Size| Block| Block| Block|   | Thread| Thread| Thread| ClusterM1Xs| ClusterN1Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| SrcDstAccessOrder| SrcDstVectorDim| DstScalarPerVector|
+        //############################|    Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|            Specialization|      |      |      |      |   |       |       |       |            |            |       _K0_M0_M1_K1|         _K0_M0_M1_K1|   ArrangeOrder|               |           _K0_M0_M1_K1| ContiguousDimOrder|           _K0_M0_M1_K1|       _K0_N0_N1_K1|         _K0_N0_N1_K1|   ArrangeOrder|               |           _K0_N0_N1_K1| ContiguousDimOrder|           _K0_N0_N1_K1|                  |                |                   |
+        //############################|           |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |       |       |       |            |            |                   |                     |               |               |                       |                   |                       |                   |                     |               |               |                       |                   |                       |                  |                |                   |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Dl< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,    16,  1,      4,      4,      1,     S<8, 2>,     S<8, 2>,   S<1, 8, 1, 1, 1>,   S<1, 2, 1, 128, 1>, S<0, 2, 3, 1, 4>, S<0, 2, 3, 1, 4>,   S<1, 1, 1, 1, 1>,   S<0, 2, 3, 1, 4>,       S<1, 1, 1, 1, 1>,   S<1, 1, 1, 8, 1>,   S<1, 16, 1, 16, 1>, S<0, 1, 4, 2, 3>, S<0, 1, 4, 2, 3>,   S<1, 1, 1, 1, 1>,   S<0, 1, 4, 2, 3>,       S<1, 1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,             5,                   1>
+    // 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_dl_instance.hpp

@@ -53,6 +53,10 @@ using device_grouped_conv2d_fwd_dl_f16_instances = std::tuple<
            // ########################################|     Spatial|   Type|    Type|        Type|    Type|    Type|         |          |    Layout|          | Elementwise|  Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
            // ########################################|            |       |        |            |        |        |         |          |          |          |   Operation|    Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
            // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        // generic instances
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,     8,    16,     4,     2,  1,          1,          2,      1,       S<4, 2>,       S<1, 1>,      S<2, 1, 2, 1>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 1, 1, 1>,        S<2, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
+
         DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
     // clang-format on
     >;
@@ -71,6 +75,10 @@ using device_grouped_conv2d_fwd_dl_f32_instances = std::tuple<
            // ########################################|     Spatial|   Type|    Type|        Type|    Type|    Type|         |          |    Layout|          | Elementwise|  Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
            // ########################################|            |       |        |            |        |        |         |          |          |          |   Operation|    Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
            // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        // generic instances
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F32,     F32,  DsDatatype,     F32,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough, PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,     8,    16,      4,    2,  1,          1,          2,      1,       S<4, 2>,       S<1, 1>,      S<2, 1, 2, 1>,      S<1, 1, 8, 1>,    S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 1, 1, 1>,        S<2, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F32,     F32,  DsDatatype,     F32,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough, PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
+
         DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F32,     F32,  DsDatatype,     F32,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough, PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
     // clang-format on
     >;

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.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"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+using I8   = int8_t;
+using I32  = int32_t;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto ConvFwdOddC =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <index_t NDSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_wmma_f16_instances = std::tuple<
+    // clang-format off
+        //########################################|    NumDim|       A|       B|       Ds|       E| AData| BData|         Ds|  EData| AccData| CShuffle|            A|           B|          CDE|    ConvForward|           GEMM| Block|  MPer|  NPer|  KPer| K1|  MPer| NPer| MRepeat| NRepeat|  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|   DataType|   Type|    Type| DataType|  Elementwise| Elementwise|  Elementwise| Specialization| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|          |        |        |         |        |      |      |           |       |        |         |    Operation|   Operation|    Operation|               |               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|          |        |        |         |        |      |      |           |       |        |         |             |            |             |               |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  8,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        // blocksize=256
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     4,  8,    16,   16,       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_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,    64,   256,     4,  8,    16,   16,       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_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   256,    64,     4,  8,    16,   16,       8,       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_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     8,  8,    16,   16,       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>,
+        // blocksize=128
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  8,    16,   16,       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_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     8,  8,    16,   16,       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_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     4,  8,    16,   16,       2,       4,     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_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     8,  8,    16,   16,       2,       4,     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_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     4,  8,    16,   16,       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, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     8,  8,    16,   16,       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, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    32,   256,     4,  8,    16,   16,       1,       8,     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_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   256,    32,     4,  8,    16,   16,       8,       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>,      
+        // blocksize=64
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    64,     4,  8,    16,   16,       1,       4,     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, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    64,    32,     4,  8,    16,   16,       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, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    32,     8,  8,    16,   16,       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, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,   128,     4,  8,    16,   16,       1,       8,     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, 32, 1, 2>,               8>,
+        // blocksize=32
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    64,     4,  8,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    64,    16,     4,  8,    16,   16,       4,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    32,    32,     4,  8,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    16,     4,  8,    16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>
+    // clang-format on
+    >;
+
+template <index_t NDSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_wmma_i8_instances = std::tuple<
+    // clang-format off
+        //########################################|    NumDim|       A|       B|       Ds|       E| AData| BData|         Ds|  EData| AccData| CShuffle|            A|           B|          CDE|    ConvForward|           GEMM| Block|  MPer|  NPer|  KPer| K1|  MPer| NPer| MRepeat| NRepeat|  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|   DataType|   Type|    Type| DataType|  Elementwise| Elementwise|  Elementwise| Specialization| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|          |        |        |         |        |      |      |           |       |        |         |    Operation|   Operation|    Operation|               |               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|          |        |        |         |        |      |      |           |       |        |         |             |            |             |               |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        //generic instance
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  16,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,               1,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,               1,              16,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        // blocksize=256
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     4,  16,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,    64,   256,     4,  16,    16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   256,    64,     4,  16,    16,   16,       8,       1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     8,  16,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        // blocksize=128
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  16,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     8,  16,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     4,  16,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     8,  16,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     4,  16,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     8,  16,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    32,   256,     4,  16,    16,   16,       1,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   256,    32,     4,  16,    16,   16,       8,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,      
+        // blocksize=64
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    64,     4,  16,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    64,    32,     4,  16,    16,   16,       2,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    32,     8,  16,    16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,   128,     4,  16,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        // blocksize=32
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    64,     4,  16,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    64,    16,     4,  16,    16,   16,       4,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    32,    32,     4,  16,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    16,     4,  16,    16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

@@ -16,6 +16,7 @@ namespace device {
 namespace instance {
 
 // conv2d backward data
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
@@ -29,7 +30,8 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
@@ -43,7 +45,8 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
@@ -57,7 +60,8 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
@@ -71,7 +75,8 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
@@ -85,7 +90,8 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
@@ -99,8 +105,9 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
 // conv3d backward data
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   GNDHWK,
@@ -114,7 +121,8 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   GNDHWK,
@@ -128,7 +136,8 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   GNDHWK,
@@ -142,7 +151,8 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   NDHWGK,
@@ -156,7 +166,8 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   NDHWGK,
@@ -170,7 +181,8 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   NDHWGK,
@@ -184,7 +196,7 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
 template <ck::index_t NumDimSpatial,
           typename OutLayout,
           typename WeiLayout,
@@ -230,42 +242,54 @@ struct DeviceOperationInstanceFactory<
             if constexpr(is_same_v<InLayout, GNHWC> && is_same_v<WeiLayout, GKYXC> &&
                          is_same_v<OutLayout, GNHWK>)
             {
+#ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
                                   is_same_v<OutDataType, F32>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
                                   is_same_v<OutDataType, BF16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
             else if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
                               is_same_v<OutLayout, NHWGK>)
             {
+#ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
                                   is_same_v<OutDataType, F32>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
                                   is_same_v<OutDataType, BF16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
         }
         else if constexpr(NumDimSpatial == 3)
@@ -274,46 +298,58 @@ struct DeviceOperationInstanceFactory<
             if constexpr(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
                          is_same_v<OutLayout, GNDHWK>)
             {
+#ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
                                   is_same_v<OutDataType, F32>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
                                   is_same_v<OutDataType, BF16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
             else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
                               is_same_v<OutLayout, NDHWGK>)
             {
+#ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
                                   is_same_v<OutDataType, F32>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
                                   is_same_v<OutDataType, BF16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
         }
 

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

@@ -17,7 +17,9 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+// xdl
 // conv1d backward weight
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -29,7 +31,8 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_insta
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -41,7 +44,8 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -53,8 +57,9 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
 // conv2d backward weight
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -66,7 +71,8 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_in
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -78,7 +84,8 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -90,7 +97,8 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            NHWGC,
@@ -102,7 +110,8 @@ void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_in
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            NHWGC,
@@ -114,7 +123,8 @@ void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            NHWGC,
@@ -126,8 +136,9 @@ void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
 // conv3d backward weight
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -139,7 +150,8 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -151,7 +163,8 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -163,7 +176,8 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            NDHWGC,
@@ -175,7 +189,8 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            NDHWGC,
@@ -187,7 +202,8 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            NDHWGC,
@@ -199,6 +215,248 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
+#endif
+
+#ifdef DL_KERNELS
+// dl
+// conv1d backward weight
+#ifdef CK_ENABLE_BF16
+void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           GNWC,
+                                                           GKXC,
+                                                           GNWK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           GNWC,
+                                                           GKXC,
+                                                           GNWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP32
+void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           GNWC,
+                                                           GKXC,
+                                                           GNWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_BF16
+void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           NWGC,
+                                                           GKXC,
+                                                           NWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           NWGC,
+                                                           GKXC,
+                                                           NWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP32
+void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           NWGC,
+                                                           GKXC,
+                                                           NWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+// conv2d backward weight
+#ifdef CK_ENABLE_BF16
+void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP32
+void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_BF16
+void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP32
+void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+// conv3d backward weight
+#ifdef CK_ENABLE_BF16
+void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP32
+void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_BF16
+void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#ifdef CK_ENABLE_FP32
+void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+#endif
 
 template <ck::index_t NumDimSpatial,
           typename InLayout,
@@ -239,23 +497,68 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             if constexpr(is_same_v<InLayout, GNWC> && is_same_v<WeiLayout, GKXC> &&
                          is_same_v<OutLayout, GNWK>)
             {
+#ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
+#endif
                     add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
+#endif
                     add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
+            }
+            else if constexpr(is_same_v<InLayout, NWGC> && is_same_v<WeiLayout, GKXC> &&
+                              is_same_v<OutLayout, NWGK>)
+            {
+#ifdef DL_KERNELS
+#ifdef CK_ENABLE_FP32
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instances(op_ptrs);
+                }
+#endif
+#ifdef CK_ENABLE_FP16
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instances(op_ptrs);
+                }
+#endif
+#ifdef CK_ENABLE_BF16
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
+#endif
+#endif
             }
         }
         else if constexpr(NumDimSpatial == 2)
@@ -263,48 +566,84 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             if constexpr(is_same_v<InLayout, GNHWC> && is_same_v<WeiLayout, GKYXC> &&
                          is_same_v<OutLayout, GNHWK>)
             {
+#ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
             else if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
                               is_same_v<OutLayout, NHWGK>)
             {
+#ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
         }
         else if constexpr(NumDimSpatial == 3)
@@ -312,48 +651,84 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             if constexpr(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
                          is_same_v<OutLayout, GNDHWK>)
             {
+#ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
             else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
                               is_same_v<OutLayout, NDHWGK>)
             {
+#ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
         }
 

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

@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 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,
@@ -31,7 +31,8 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(
                                                               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,
@@ -45,7 +46,8 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(
                                                               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,
@@ -59,7 +61,8 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(
                                                               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,
@@ -73,7 +76,8 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(
                                                               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,
@@ -88,7 +92,8 @@ void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
                                                               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,
@@ -102,7 +107,8 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               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,
@@ -116,7 +122,9 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef DL_KERNELS
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -130,7 +138,8 @@ void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -144,8 +153,156 @@ void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
+#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);
+
+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);
 
+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);
+
+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);
+
+#ifdef 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);
+#endif
+#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);
+
+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);
+
+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);
+
+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);
+#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);
+
+#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,
@@ -159,6 +316,63 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
                                                               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);
+
+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);
+
+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);
+
+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);
 
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -173,7 +387,65 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               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);
 
+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);
+
+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);
+
+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);
+#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,
@@ -187,7 +459,8 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                               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,
@@ -202,7 +475,8 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(
                                                               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,
@@ -217,6 +491,63 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
                                                               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);
+
+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);
+
+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);
+
+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);
+#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,
@@ -230,7 +561,8 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
                                                               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,
@@ -245,6 +577,63 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(
                                                               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);
+
+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);
+
+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);
+
+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);
+#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,
@@ -259,7 +648,8 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
                                                               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,
@@ -274,6 +664,63 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
+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);
+
+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);
+
+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);
+
+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);
+#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,
@@ -287,7 +734,8 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
                                                               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,
@@ -302,6 +750,63 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
                                                               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);
+
+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);
+
+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);
+
+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);
+#endif
+
 template <ck::index_t NumDimSpatial,
           typename InLayout,
           typename WeiLayout,
@@ -343,119 +848,210 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
         if constexpr(NumDimSpatial == 1 && is_same_v<InLayout, GNWC> &&
                      is_same_v<WeiLayout, GKXC> && is_same_v<OutLayout, GNWK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef CK_ENABLE_BF16
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+#endif
+#ifdef CK_ENABLE_INT8
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(op_ptrs);
             }
+#endif
         }
         else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
                           is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, GNHWK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
+#endif
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
+#endif
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef CK_ENABLE_BF16
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
             }
+#endif
+#ifdef CK_ENABLE_INT8
+            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                              is_same_v<OutDataType, int8_t>)
+            {
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instances(op_ptrs);
+            }
+#endif
         }
         else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                           is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
+#ifdef DL_KERNELS
+                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
+#endif
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
+#ifdef DL_KERNELS
+                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
+#endif
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef CK_ENABLE_BF16
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(op_ptrs);
             }
+#endif
+#ifdef CK_ENABLE_INT8
+            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                              is_same_v<OutDataType, int8_t>)
+            {
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instances(op_ptrs);
+            }
+#endif
         }
         else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC> &&
                           is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, GNDHWK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef CK_ENABLE_BF16
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+#endif
+#ifdef CK_ENABLE_INT8
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instances(op_ptrs);
             }
+#endif
         }
         else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
                           is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef CK_ENABLE_BF16
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+#endif
+#ifdef CK_ENABLE_INT8
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instances(op_ptrs);
             }
+#endif
         }
 
         return op_ptrs;

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// fp16_output
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+// fp8_inputB
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F8,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F8,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+// i8_inputB
+void add_device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         I8,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         I8,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+template <typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename EDataType>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceGroupedGemmFixedNK<ALayout,
+                                                           BLayout,
+                                                           Empty_Tuple,
+                                                           ELayout,
+                                                           ADataType,
+                                                           BDataType,
+                                                           Empty_Tuple,
+                                                           EDataType,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>
+{
+    using DeviceOp = DeviceGroupedGemmFixedNK<ALayout,
+                                              BLayout,
+                                              Empty_Tuple,
+                                              ELayout,
+                                              ADataType,
+                                              BDataType,
+                                              Empty_Tuple,
+                                              EDataType,
+                                              PassThrough,
+                                              PassThrough,
+                                              PassThrough>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        // fp16_output
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+            }
+        }
+
+        // fp8_input
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, f8_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instances(op_ptrs);
+            }
+        }
+
+        // i8_input
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, int8_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instances(op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_image_to_column.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// nhwc, 1d
+void add_device_image_to_column_nhwc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, BF16, BF16>>>& instances);
+
+void add_device_image_to_column_nhwc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F16, F16>>>& instances);
+
+void add_device_image_to_column_nhwc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F32, F32>>>& instances);
+
+void add_device_image_to_column_nhwc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, int8_t, int8_t>>>& instances);
+// nhwc, 2d
+void add_device_image_to_column_nhwc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, BF16, BF16>>>& instances);
+
+void add_device_image_to_column_nhwc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F16, F16>>>& instances);
+
+void add_device_image_to_column_nhwc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F32, F32>>>& instances);
+
+void add_device_image_to_column_nhwc_2d_i8_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, int8_t, int8_t>>>& instances);
+// nhwc, 3d
+void add_device_image_to_column_nhwc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, BF16, BF16>>>& instances);
+
+void add_device_image_to_column_nhwc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F16, F16>>>& instances);
+
+void add_device_image_to_column_nhwc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F32, F32>>>& instances);
+
+void add_device_image_to_column_nhwc_3d_i8_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, int8_t, int8_t>>>& instances);
+
+template <ck::index_t NumDimSpatial, typename InLayout, typename InDataType, typename OutDataType>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::
+        DeviceImageToColumn<NumDimSpatial, InLayout, InDataType, OutDataType>>
+{
+    using DeviceOp = DeviceImageToColumn<NumDimSpatial, InLayout, InDataType, OutDataType>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(NumDimSpatial == 1 && is_same_v<InLayout, GNWC>)
+        {
+            if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+            {
+                add_device_image_to_column_nhwc_1d_f32_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+            {
+                add_device_image_to_column_nhwc_1d_f16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                              is_same_v<OutDataType, ck::bhalf_t>)
+            {
+                add_device_image_to_column_nhwc_1d_bf16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+            {
+                add_device_image_to_column_nhwc_1d_i8_instances(op_ptrs);
+            }
+        }
+        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC>)
+        {
+            if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+            {
+                add_device_image_to_column_nhwc_2d_f32_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+            {
+                add_device_image_to_column_nhwc_2d_f16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                              is_same_v<OutDataType, ck::bhalf_t>)
+            {
+                add_device_image_to_column_nhwc_2d_bf16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+            {
+                add_device_image_to_column_nhwc_2d_i8_instances(op_ptrs);
+            }
+        }
+        else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC>)
+        {
+            if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+            {
+                add_device_image_to_column_nhwc_3d_f32_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+            {
+                add_device_image_to_column_nhwc_3d_f16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                              is_same_v<OutDataType, ck::bhalf_t>)
+            {
+                add_device_image_to_column_nhwc_3d_bf16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+            {
+                add_device_image_to_column_nhwc_3d_i8_instances(op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, 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_image_to_column_impl.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::tensor_layout::convolution;
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_image_to_column_bf16_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,     8,     8,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    16,    16,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    16,    16,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    64,    64,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,    16,    16, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,    64,    64, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,    64,    64, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     8>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_image_to_column_f16_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,     8,     8,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,    16,    16,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    16,    16,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    64,    64,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,    16,    16, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,    64,    64, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,    64,    64, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     8>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_image_to_column_f32_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,    64,     8,     8,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,    64,    16,    16,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   128,    16,    16,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   128,    64,    64,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,    16,    16, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,    64,    64, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,   128,   128, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,    64,    64, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,   128,   128, S<16, 16>,     4>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_image_to_column_i8_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,     8,     8,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    16,    16,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    16,    16,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    64,    64,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,    16,    16, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,    64,    64, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,    64,    64, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   256,   256, S<16, 16>,     16>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck