Merge branch 'develop' into transpose_5d

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

@@ -18,16 +18,18 @@ namespace host {
 /**
  * \brief Reference implementation for image to column.
  *
- * Tensor descriptor has [G, N, C, Di, Hi, Wi] data layout.
+ * Input tensor descriptor has [G, N, C, Di, Hi, Wi] data layout.
  * G must be equal to 1. Memory layout is [G, N, Di, Hi, Wi, C].
+ * Output tensor descriptor has [N * Do * Ho * Wo, Z * Y * X * C] data layout.
+ * Memory layout is the same.
  *
  * \tparam NDimSpatial Number of spatial dimensions.
- * \tparam InputLayout Input Layout.
+ * \tparam ImageLayout Image Layout.
  * \tparam InDataType Input Data Type.
  * \tparam OutDataType Output Data Type.
  */
 template <ck::index_t NDimSpatial,
-          typename InputLayout,
+          typename ImageLayout,
           typename InDataType,
           typename OutDataType,
           typename std::enable_if<NDimSpatial >= 1 && NDimSpatial <= 3, bool>::type = false>
@@ -240,8 +242,8 @@ struct ReferenceImageToColumn : public device::BaseOperator
     {
         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>))
+        if constexpr(!(std::is_same_v<ImageLayout, GNWC> || std::is_same_v<ImageLayout, GNHWC> ||
+                       std::is_same_v<ImageLayout, GNDHWC>))
         {
             return false;
         }

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+// Image to Column
+// nhwc, 1d
+void add_device_image_to_column_nwc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+// nhwc, 2d
+void add_device_image_to_column_nhwc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+// nhwc, 3d
+void add_device_image_to_column_ndhwc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+
+// Column to Image
+// nhwc, 1d
+void add_device_column_to_image_nwc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
+// nhwc, 2d
+void add_device_column_to_image_nhwc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
+// nhwc, 3d
+void add_device_column_to_image_ndhwc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
+
+template <ck::index_t NumDimSpatial,
+          typename ImageLayout,
+          typename InDataType,
+          typename OutDataType,
+          typename ConvTensorRearrangeOp>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceConvTensorRearrange<NumDimSpatial,
+                                                            ImageLayout,
+                                                            InDataType,
+                                                            OutDataType,
+                                                            ConvTensorRearrangeOp>>
+{
+    using DeviceOp = DeviceConvTensorRearrange<NumDimSpatial,
+                                               ImageLayout,
+                                               InDataType,
+                                               OutDataType,
+                                               ConvTensorRearrangeOp>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ConvTensorRearrangeOp, ImageToColumn>)
+        {
+            if constexpr(NumDimSpatial == 1 && is_same_v<ImageLayout, GNWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_nwc_1d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_image_to_column_nwc_1d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_image_to_column_nwc_1d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_image_to_column_nwc_1d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, GNHWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_nhwc_2d_f32_instances(op_ptrs);
+                }
+                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<ImageLayout, GNDHWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_ndhwc_3d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_image_to_column_ndhwc_3d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_image_to_column_ndhwc_3d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_image_to_column_ndhwc_3d_i8_instances(op_ptrs);
+                }
+            }
+        }
+        else if constexpr(is_same_v<ConvTensorRearrangeOp, ColumnToImage>)
+        {
+            if constexpr(NumDimSpatial == 1 && is_same_v<ImageLayout, GNWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_nwc_1d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_nwc_1d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_nwc_1d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_nwc_1d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, GNHWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_nhwc_2d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_nhwc_2d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_nhwc_2d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_nhwc_2d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 3 && is_same_v<ImageLayout, GNDHWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_ndhwc_3d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_ndhwc_3d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_ndhwc_3d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_ndhwc_3d_i8_instances(op_ptrs);
+                }
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_column_to_image_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 namespace ck::conv_tensor_rearrange_op;
+
+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_column_to_image_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|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        // generic instance
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    16,    16,   S<8, 8>,     1>,
+
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,    64,    64, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     8>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_column_to_image_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|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        // generic instance
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,    64,    16,    16,   S<8, 8>,     1>,
+        
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,   256,    64,    64, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     8>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_column_to_image_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|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        // generic instance
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F32,         F32,    64,    16,    16,   S<8, 8>,     1>,
+        
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F32,         F32,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F32,         F32,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F32,         F32,   256,    64,    64, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F32,         F32,   256,   128,   128, S<16, 16>,     4>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_column_to_image_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|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        // generic instance
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    16,    16,   S<8, 8>,     1>,
+
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,    64,    64, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     8>,
+        DeviceColumnToImageImpl<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

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

@@ -13,6 +13,7 @@ namespace device {
 namespace instance {
 
 using namespace ck::tensor_layout::convolution;
+using namespace ck::conv_tensor_rearrange_op;
 
 using BF16 = ck::bhalf_t;
 using F16  = ck::half_t;
@@ -28,17 +29,12 @@ using device_image_to_column_bf16_instances = std::tuple<
         //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
         //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
         //#####################|           |         |           |            |      |      |      |          |       |
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,     8,     8,   S<8, 8>,     1>,
+        // generic instance
         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>
@@ -52,17 +48,13 @@ using device_image_to_column_f16_instances = std::tuple<
         //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
         //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
         //#####################|           |         |           |            |      |      |      |          |       |
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,     8,     8,   S<8, 8>,     1>,
+        // generic instance        
         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>
@@ -76,15 +68,11 @@ using device_image_to_column_f32_instances = std::tuple<
         //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
         //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
         //#####################|           |         |           |            |      |      |      |          |       |
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,    64,     8,     8,   S<8, 8>,     1>,
+        // generic instance      
         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
@@ -97,17 +85,13 @@ using device_image_to_column_i8_instances = std::tuple<
         //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
         //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
         //#####################|           |         |           |            |      |      |      |          |       |
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,     8,     8,   S<8, 8>,     1>,
+        // generic instance
         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>,

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

@@ -312,6 +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,
@@ -505,6 +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 +68,8 @@ 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,
@@ -120,7 +121,7 @@ void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(
                                                     PassThrough,
                                                     PassThrough,
                                                     Bilinear>>>& instances);
-
+#endif
 // GEMM + Bilinear
 template <typename ALayout,
           typename BLayout,
@@ -158,7 +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>)
         {
@@ -187,7 +188,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                     op_ptrs);
             }
         }
-        else if constexpr(is_same_v<ADataType, std::int8_t> && is_same_v<BDataType, std::int8_t> &&
+#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> &&
@@ -211,7 +214,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                 add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
@@ -220,4 +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/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 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_backward_data_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_bwd_data_multiple_d_wmma_cshuffle.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+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 ConvBwdDataDefault =
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
+
+static constexpr auto ConvBwdData1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
+
+template <index_t NDSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_wmma_f16_instances = std::tuple<
+    // clang-format off
+        //########################################|    NumDim|       A|       B|       Ds|       E| AData| BData|    AccData|  CShuffle|      DsData| EData|           A|           B|          CDE|    ConvForward| Block|  MPer|  NPer| K0Per| 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|       Type|  DataType|        Type|  Type| Elementwise| Elementwise|  Elementwise| 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
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 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<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        // blocksize=256
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 256,   128,   256,     8,  8,    16,   16,       4,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 256,    64,   256,     8,  8,    16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 256,   128,   256,     8,  8,    16,   16,       4,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 256,   128,    64,     8,  8,    16,   16,       4,       1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        // blocksize=128
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 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<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 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<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 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<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,   128,   128,     8,  8,    16,   16,       4,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,    32,   256,     8,  8,    16,   16,       1,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        // blocksize=64
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,    64,     8,  8,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    64,    64,     8,  8,    16,   16,       2,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,    64,     8,  8,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,   128,     8,  8,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        // blocksize=32
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    16,    64,     8,  8,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    64,    32,     8,  8,    16,   16,       4,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    32,    32,     8,  8,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    16,    32,     8,  8,    16,   16,       1,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              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,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_wmma_i8_instances = std::tuple<
+    // clang-format off
+        //########################################|    NumDim|       A|       B|       Ds|       E| AData| BData|    AccData|  CShuffle|      DsData| EData|           A|           B|          CDE|    ConvForward| Block|  MPer|  NPer| K0Per| 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|       Type|  DataType|        Type|  Type| Elementwise| Elementwise|  Elementwise| 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
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 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<0, 2, 1>,     S<0, 2, 1>,             1,            1,              16,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        // blocksize=256
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 256,    64,   256,     8,  16,    16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        // blocksize=128
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,    64,   256,     8,  16,    16,   16,       2,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 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<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,   128,   256,     8,  16,    16,   16,       4,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,    32,   256,     8,  16,    16,   16,       1,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,   256,   128,     8,  16,    16,   16,       8,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,      
+        // blocksize=64
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,   128,     8,  16,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    64,   128,     8,  16,    16,   16,       2,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,   128,     8,  16,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,   64,      8,  16,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        // blocksize=32
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    16,    64,     8,  16,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    64,    64,     8,  16,    16,   16,       4,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    32,    32,     8,  16,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    16,    64,     8,  16,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           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_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp

@@ -18,6 +18,8 @@ namespace instance {
 using BF16 = ck::bhalf_t;
 using F16  = ck::half_t;
 using F32  = float;
+using BF8  = ck::bf8_t;
+using F8   = ck::f8_t;
 
 using Empty_Tuple = ck::Tuple<>;
 
@@ -143,6 +145,43 @@ using device_grouped_conv_bwd_data_xdl_f32_instances =
         // clang-format on
         >;
 
+// f16_f16_f16_comp_f8
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_xdl_input_fp16_comp_bf8f8_instances =
+    std::tuple<
+        // clang-format off
+         // ##############################################|       NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+         // ##############################################|    Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+        // generic instance
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,            1,            1,     S<1, 16, 1, 4>,                1,  LoopScheduler::Default, BF8, F8>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,            1,            1,     S<1, 32, 1, 4>,                1,  LoopScheduler::Default, BF8, F8>,
+
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8,  LoopScheduler::Default, BF8, F8>
+        // clang-format on
+        >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation

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

@@ -19,6 +19,14 @@ using BF16 = ck::bhalf_t;
 using F16  = ck::half_t;
 using F32  = float;
 
+#ifdef CK_ENABLE_FP8
+using F8 = ck::f8_t;
+#endif
+
+#ifdef CK_ENABLE_BF8
+using BF8 = ck::bf8_t;
+#endif
+
 using Empty_Tuple = ck::Tuple<>;
 
 template <ck::index_t... Is>
@@ -133,6 +141,43 @@ using device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances = std::tuple<
     // clang-format on
     >;
 
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_comp_bf8_f8_instances = std::tuple<
+// clang-format off
+        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|              ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer| Compute| Compute|
+        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                    Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|   TypeA|   TypeB|
+        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|            Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|        |        |
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |        |        |
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+        // generic instance
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              2,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              2,              4,      true,           1,           1,   S<1, 16, 1, 4>,               2,       BF8,     F8>,   
+        // instance for small conv.K
+        // for fp16 conv.K and conv.C must be divisible by 2
+        // since half_t atomic_add require scalar_per_x_vector % 2 == 0
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              2,              1,      true,           1,           1,   S<1, 32, 1, 4>,               2,       BF8,     F8>,  
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              2,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8,       BF8,     F8>,
+
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8,       BF8,     F8>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8,       BF8,     F8>
+#endif
+    // clang-format on
+    >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation

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

@@ -13,6 +13,10 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+#ifdef CK_ENABLE_FP8
+using F8 = ck::f8_t;
+#endif
+
 using BF16 = ck::bhalf_t;
 using F16  = ck::half_t;
 using F32  = float;
@@ -174,6 +178,42 @@ using device_grouped_conv_fwd_xdl_int8_instances = std::tuple<
     // clang-format on
     >;
 
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_f16_comp_f8_instances = std::tuple<
+// clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer| ComputeType|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            |
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            |
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |            |
+#ifdef CK_ENABLE_FP8
+        // generic instance
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1,         F8>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>
+#endif
+    // clang-format on
+    >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation

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

@@ -30,6 +30,34 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(
@@ -61,6 +89,35 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 #endif
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  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,
@@ -75,6 +132,35 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(
@@ -106,6 +192,35 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 #endif
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  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(
@@ -121,6 +236,34 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
@@ -152,6 +295,35 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 #endif
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  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,
@@ -166,6 +338,34 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
@@ -197,13 +397,61 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 #endif
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+#endif
+#if defined CK_ENABLE_FP16 && defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_input_f16_comp_bf8f8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  BF8,
+                                                                  F8>>>& instances);
+#endif
 template <ck::index_t NumDimSpatial,
           typename OutLayout,
           typename WeiLayout,
           typename InLayout,
           typename OutDataType,
           typename WeiDataType,
-          typename InDataType>
+          typename InDataType,
+          typename ComputeTypeA,
+          typename ComputeTypeB>
 struct DeviceOperationInstanceFactory<
     ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD<
         NumDimSpatial,
@@ -217,7 +465,9 @@ struct DeviceOperationInstanceFactory<
         InDataType,
         ck::tensor_operation::element_wise::PassThrough,
         ck::tensor_operation::element_wise::PassThrough,
-        ck::tensor_operation::element_wise::PassThrough>>
+        ck::tensor_operation::element_wise::PassThrough,
+        ComputeTypeA,
+        ComputeTypeB>>
 {
     using DeviceOp =
         DeviceGroupedConvBwdDataMultipleD<NumDimSpatial,
@@ -231,7 +481,9 @@ struct DeviceOperationInstanceFactory<
                                           InDataType,
                                           ck::tensor_operation::element_wise::PassThrough,
                                           ck::tensor_operation::element_wise::PassThrough,
-                                          ck::tensor_operation::element_wise::PassThrough>;
+                                          ck::tensor_operation::element_wise::PassThrough,
+                                          ComputeTypeA,
+                                          ComputeTypeB>;
 
     static auto GetInstances()
     {
@@ -247,6 +499,10 @@ struct DeviceOperationInstanceFactory<
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_1x1s1p0_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP32
@@ -263,6 +519,15 @@ struct DeviceOperationInstanceFactory<
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_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_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_instances(op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_1x1s1p0_instances(
+                        op_ptrs);
+                }
 #endif
             }
             else if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
@@ -273,6 +538,10 @@ struct DeviceOperationInstanceFactory<
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_1x1s1p0_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP32
@@ -289,6 +558,15 @@ struct DeviceOperationInstanceFactory<
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_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_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_instances(op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_1x1s1p0_instances(
+                        op_ptrs);
+                }
 #endif
             }
         }
@@ -304,6 +582,10 @@ struct DeviceOperationInstanceFactory<
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
                         op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_f16_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_f16_1x1s1p0_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP32
@@ -321,6 +603,16 @@ struct DeviceOperationInstanceFactory<
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_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_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_i8_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_i8_1x1s1p0_instances(
+                        op_ptrs);
+                }
 #endif
             }
             else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
@@ -328,10 +620,24 @@ struct DeviceOperationInstanceFactory<
             {
 #ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
-                             is_same_v<OutDataType, F16>)
+                             is_same_v<OutDataType, F16> && is_same_v<ComputeTypeA, F16> &&
+                             is_same_v<ComputeTypeB, F16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
                         op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_f16_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_f16_1x1s1p0_instances(
+                        op_ptrs);
+                }
+#endif
+#if defined CK_ENABLE_FP16 && defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+                else if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                                  is_same_v<OutDataType, F16> && is_same_v<ComputeTypeA, bf8_t> &&
+                                  is_same_v<ComputeTypeB, f8_t>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_input_f16_comp_bf8f8_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP32
@@ -349,6 +655,16 @@ struct DeviceOperationInstanceFactory<
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_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_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_1x1s1p0_instances(
+                        op_ptrs);
+                }
 #endif
             }
         }

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

@@ -216,6 +216,21 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 #endif
+#if defined CK_ENABLE_FP16 && defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_f8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           BF8,
+                                                           F8>>>& instances);
+#endif
 
 #ifdef DL_KERNELS
 // dl
@@ -464,7 +479,9 @@ template <ck::index_t NumDimSpatial,
           typename OutLayout,
           typename InDataType,
           typename WeiDataType,
-          typename OutDataType>
+          typename OutDataType,
+          typename ComputeTypeA,
+          typename ComputeTypeB>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvBwdWeight<
     NumDimSpatial,
     InLayout,
@@ -475,7 +492,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     OutDataType,
     ck::tensor_operation::element_wise::PassThrough,
     ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::PassThrough>>
+    ck::tensor_operation::element_wise::PassThrough,
+    ComputeTypeA,
+    ComputeTypeB>>
 {
     using DeviceOp = DeviceGroupedConvBwdWeight<NumDimSpatial,
                                                 InLayout,
@@ -486,7 +505,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                                                 OutDataType,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough>;
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                ComputeTypeA,
+                                                ComputeTypeB>;
 
     static auto GetInstances()
     {
@@ -706,7 +727,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #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>)
+                                  is_same_v<OutDataType, half_t> &&
+                                  is_same_v<ComputeTypeA, half_t> &&
+                                  is_same_v<ComputeTypeB, half_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
@@ -728,6 +751,15 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
+#if defined CK_ENABLE_FP16 && defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t> &&
+                                  is_same_v<ComputeTypeA, bf8_t> && is_same_v<ComputeTypeB, f8_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_f8_instances(
+                        op_ptrs);
+                }
 #endif
             }
         }

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

@@ -16,6 +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(
@@ -32,6 +33,7 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(
                                                               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,
@@ -47,6 +49,7 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(
                                                               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,
@@ -62,6 +65,7 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(
                                                               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,
@@ -77,100 +81,90 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(
                                                               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(
+
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
-                                                              BF16,
-                                                              BF16,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
                                                               Empty_Tuple,
-                                                              BF16,
+                                                              int8_t,
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
-#ifdef CK_ENABLE_FP16
-void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
+
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
                                                               Empty_Tuple,
-                                                              F16,
+                                                              int8_t,
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
-#ifdef CK_ENABLE_FP32
-void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
+
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
-                                                              F32,
-                                                              F32,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
                                                               Empty_Tuple,
-                                                              F32,
+                                                              int8_t,
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
-#ifdef DL_KERNELS
-#ifdef CK_ENABLE_FP16
-void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
+
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
                                                               Empty_Tuple,
-                                                              F16,
+                                                              int8_t,
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
-#ifdef CK_ENABLE_FP32
-void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
+
+#ifdef CK_ENABLE_BF16
+// grouped conv2d forward, GNHWC/GKYXC/GNHWK
+void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
                                                               GKYXC,
                                                               Empty_Tuple,
                                                               GNHWK,
-                                                              F32,
-                                                              F32,
+                                                              BF16,
+                                                              BF16,
                                                               Empty_Tuple,
-                                                              F32,
+                                                              BF16,
                                                               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(
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
                                                               GKYXC,
@@ -183,22 +177,26 @@ void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
 
-void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instances(
+#ifdef CK_ENABLE_FP32
+void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
                                                               GKYXC,
                                                               Empty_Tuple,
                                                               GNHWK,
-                                                              F16,
-                                                              F16,
+                                                              F32,
+                                                              F32,
                                                               Empty_Tuple,
-                                                              F16,
+                                                              F32,
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
 
-void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instances(
+#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,
@@ -211,23 +209,8 @@ void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instances(
                                                               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,
@@ -285,22 +268,7 @@ void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instances(
                                                               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(
@@ -317,6 +285,7 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
                                                               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,
@@ -388,63 +357,7 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               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,
@@ -460,6 +373,7 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                               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(
@@ -476,6 +390,7 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(
                                                               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,
@@ -547,6 +462,7 @@ void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instances(
                                                               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,
@@ -562,6 +478,7 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
                                                               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,
@@ -633,6 +550,7 @@ void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instances(
                                                               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(
@@ -649,6 +567,7 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
                                                               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,
@@ -663,7 +582,9 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
 
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -677,7 +598,9 @@ void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
 
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -691,7 +614,9 @@ void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instances
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
 
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -705,7 +630,9 @@ void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instanc
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
 
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -720,6 +647,88 @@ void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP8
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_f8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              F8>>>& instances);
+#endif
+
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
@@ -735,6 +744,7 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
                                                               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,
@@ -807,13 +817,79 @@ void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instances(
                                                               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
+
+#if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
+void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
+void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
+void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F32,
+                                                              F32,
+                                                              Empty_Tuple,
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
 template <ck::index_t NumDimSpatial,
           typename InLayout,
           typename WeiLayout,
           typename OutLayout,
           typename InDataType,
           typename WeiDataType,
-          typename OutDataType>
+          typename OutDataType,
+          typename ComputeType>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
     NumDimSpatial,
     InLayout,
@@ -826,7 +902,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     OutDataType,
     ck::tensor_operation::element_wise::PassThrough,
     ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::PassThrough>>
+    ck::tensor_operation::element_wise::PassThrough,
+    ComputeType>>
 {
     using DeviceOp = DeviceGroupedConvFwdMultipleD<NumDimSpatial,
                                                    InLayout,
@@ -839,7 +916,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                                                    OutDataType,
                                                    ck::tensor_operation::element_wise::PassThrough,
                                                    ck::tensor_operation::element_wise::PassThrough,
-                                                   ck::tensor_operation::element_wise::PassThrough>;
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   ComputeType>;
 
     static auto GetInstances()
     {
@@ -877,7 +955,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             }
 #endif
         }
-        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
+
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
                      is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, GNHWK>)
         {
 #ifdef CK_ENABLE_FP32
@@ -885,25 +964,37 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                          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
+
+#if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
+            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                         is_same_v<OutDataType, float>)
+            {
+                add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
             }
 #endif
+
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 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);
             }
 #endif
+
+#if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
+            {
+                add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
+            }
+#endif
+
 #ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                          is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
@@ -911,8 +1002,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 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> &&
+            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);
@@ -922,33 +1014,43 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             }
 #endif
         }
-        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
+
+        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
+
+#if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
+            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                         is_same_v<OutDataType, float>)
+            {
+                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
             }
 #endif
+
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 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);
+
+#if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
+            {
+                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
             }
 #endif
+
 #ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                          is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
@@ -967,7 +1069,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             }
 #endif
         }
-        else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC> &&
+
+        if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC> &&
                      is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, GNDHWK>)
         {
 #ifdef CK_ENABLE_FP32
@@ -1010,7 +1113,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             }
 #endif
         }
-        else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
+
+        if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
                      is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
         {
 #ifdef CK_ENABLE_FP32
@@ -1020,9 +1124,18 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(op_ptrs);
             }
 #endif
+
+#ifdef CK_ENABLE_FP8
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, ck::f8_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_f8_instances(
+                    op_ptrs);
+            }
+#endif
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                         is_same_v<OutDataType, half_t>)
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, half_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(op_ptrs);
                 add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(op_ptrs);

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/utility/host_tensor_generator.hpp

@@ -95,7 +95,7 @@ struct GeneratorTensor_2<int8_t>
     }
 };
 
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+#if defined CK_ENABLE_FP8
 template <>
 struct GeneratorTensor_2<ck::f8_t>
 {
@@ -111,6 +111,22 @@ struct GeneratorTensor_2<ck::f8_t>
 };
 #endif
 
+#if defined CK_ENABLE_BF8
+template <>
+struct GeneratorTensor_2<ck::bf8_t>
+{
+    int min_value = 0;
+    int max_value = 1;
+
+    template <typename... Is>
+    ck::bf8_t operator()(Is...)
+    {
+        float tmp = (std::rand() % (max_value - min_value)) + min_value;
+        return ck::type_convert<ck::bf8_t>(tmp);
+    }
+};
+#endif
+
 template <typename T>
 struct GeneratorTensor_3
 {
@@ -143,7 +159,7 @@ struct GeneratorTensor_3<ck::bhalf_t>
     }
 };
 
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+#if defined CK_ENABLE_FP8
 template <>
 struct GeneratorTensor_3<ck::f8_t>
 {
@@ -162,6 +178,25 @@ struct GeneratorTensor_3<ck::f8_t>
 };
 #endif
 
+#if defined CK_ENABLE_BF8
+template <>
+struct GeneratorTensor_3<ck::bf8_t>
+{
+    float min_value = 0;
+    float max_value = 1;
+
+    template <typename... Is>
+    ck::bf8_t operator()(Is...)
+    {
+        float tmp = float(std::rand()) / float(RAND_MAX);
+
+        float fp32_tmp = min_value + tmp * (max_value - min_value);
+
+        return ck::type_convert<ck::bf8_t>(fp32_tmp);
+    }
+};
+#endif
+
 template <typename T>
 struct GeneratorTensor_4
 {

library/src/tensor_operation_instance/gpu/CMakeLists.txt

@@ -51,7 +51,7 @@ function(add_instance_library INSTANCE_NAME)
         set(result 0)
     endif()
     #message("add_instance_library returns ${result}")
-    return(PROPAGATE result)
+    set(result ${result} PARENT_SCOPE)
 endfunction(add_instance_library INSTANCE_NAME)
 
 

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

+add_instance_library(device_column_to_image_instance
+   device_column_to_image_nhwc_1d_instance.cpp
+   device_column_to_image_nhwc_2d_instance.cpp
+   device_column_to_image_nhwc_3d_instance.cpp
+)

library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_1d_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_column_to_image_nwc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_column_to_image_bf16_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nwc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_column_to_image_f16_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nwc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_column_to_image_f32_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nwc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_column_to_image_i8_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_2d_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_column_to_image_nhwc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_column_to_image_bf16_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nhwc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_column_to_image_f16_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nhwc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_column_to_image_f32_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nhwc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_column_to_image_i8_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_3d_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_column_to_image_ndhwc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_column_to_image_bf16_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_ndhwc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_column_to_image_f16_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_ndhwc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_column_to_image_f32_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_ndhwc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_column_to_image_i8_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck