Merge branch 'develop' of...

Merge branch 'develop' of https://github.com/ROCmSoftwarePlatform/composable_kernel into add_fp16_wmma_conv_instance

library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvBwdDataDefault = ConvolutionBackwardDataSpecialization::Default;
+
+static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
+    ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
+
+// f16_f16_f32_f16
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_xdl_f16_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,      F16, 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,              8,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,            1,            1,     S<1, 16, 1, 4>,                1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,        S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,            1,            1,     S<1, 32, 1, 4>,                1>,
+
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,        S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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,              8,              8,         1,        S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, 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>
+        // clang-format on
+        >;
+
+// bf16_bf16_f32_bf16
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_xdl_bf16_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,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,         1,    S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,            1,            1,     S<1, 16, 1, 4>,                1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,    S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,            1,            1,     S<1, 32, 1, 4>,                1>,
+
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,   S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,   S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,    S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,    S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,    S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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,              8,              8,         1,    S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  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>
+    // clang-format on
+    >;
+
+// f32_f32_f32_f32
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_xdl_f32_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,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  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>
+        // clang-format on
+        >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

@@ -39,10 +39,16 @@ template <ck::index_t NDimSpatial,
           ConvolutionBackwardWeightSpecialization ConvSpec>
 using device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_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|
-        //#########################################|     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|
-        //#########################################| 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|                |
+        //#########################################|     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|
+        //#########################################|     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|
+        //#########################################| 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|                |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              1,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              1,              4,      true,           1,           1,   S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              1,              1,      true,           1,           1,   S<1, 32, 1, 4>,               1>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              1,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
+
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
@@ -66,10 +72,18 @@ template <ck::index_t NDimSpatial,
           ConvolutionBackwardWeightSpecialization ConvSpec>
 using device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_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|
-        //#########################################|     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|
-        //#########################################| 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|                |
+        //#########################################|     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|
+        //#########################################|     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|
+        //#########################################| 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|                |
+        // 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>,   
+        // 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>,  
+        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>,
+
         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>,
         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>,   
         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>,   
@@ -93,10 +107,16 @@ template <ck::index_t NDimSpatial,
           ConvolutionBackwardWeightSpecialization ConvSpec>
 using device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_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|
-        //#########################################|     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|
-        //#########################################| 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|                |
+        //#########################################|     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|
+        //#########################################|     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|
+        //#########################################| 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|                |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     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,              1,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              1,              4,      true,           1,           1,   S<1, 16, 1, 4>,               1>,   
+        // instance for small conv.K
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     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,              1,              1,      true,           1,           1,   S<1, 32, 1, 4>,               1>, 
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     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,              1,              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>,               4>,
+
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     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>,               4>,
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     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>,               4>,   
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     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>,               4>,   

library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp → library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
+#include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
-#include "device_grouped_conv2d_fwd_common.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto ConvFwdOddC =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 template <typename InLayout,
           typename WeiLayout,
           typename DsLayout,
@@ -22,6 +53,11 @@ using device_grouped_conv2d_fwd_dl_f16_instances = std::tuple<
            // ########################################|     Spatial|   Type|    Type|        Type|    Type|    Type|         |          |    Layout|          | Elementwise|  Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
            // ########################################|            |       |        |            |        |        |         |          |          |          |   Operation|    Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
            // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        // generic instances
+        // TODO: Change to ScalarPerVector = 1 when inner_product<half_t, half_t, float> will be supported
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,     8,    16,     4,     2,  2,          1,          2,      1,       S<4, 2>,       S<1, 1>,      S<2, 1, 2, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<1, 1, 1, 2>,        S<2, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 2>,
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 2>,
+
         DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
     // clang-format on
     >;
@@ -40,6 +76,10 @@ using device_grouped_conv2d_fwd_dl_f32_instances = std::tuple<
            // ########################################|     Spatial|   Type|    Type|        Type|    Type|    Type|         |          |    Layout|          | Elementwise|  Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
            // ########################################|            |       |        |            |        |        |         |          |          |          |   Operation|    Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
            // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        // generic instances
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F32,     F32,  DsDatatype,     F32,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough, PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,     8,    16,      4,    2,  1,          1,          2,      1,       S<4, 2>,       S<1, 1>,      S<2, 1, 2, 1>,      S<1, 1, 8, 1>,    S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 1, 1, 1>,        S<2, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F32,     F32,  DsDatatype,     F32,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough, PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
+
         DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F32,     F32,  DsDatatype,     F32,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough, PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
     // clang-format on
     >;

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 = ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 = ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto ConvFwdOddC =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
+
+static constexpr auto GemmMNKPadding = GemmSpecialization::MNKPadding;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_bf16_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_f16_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        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>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_f32_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_int8_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

@@ -16,7 +16,7 @@ namespace device {
 namespace instance {
 
 // conv2d backward data
-void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
                                                                   GKYXC,
@@ -30,7 +30,7 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
                                                                   GKYXC,
@@ -44,7 +44,7 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
                                                                   GKYXC,
@@ -58,7 +58,7 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
                                                                   GKYXC,
@@ -72,7 +72,7 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
                                                                   GKYXC,
@@ -86,7 +86,7 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
                                                                   GKYXC,
@@ -100,6 +100,91 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
+// conv3d backward data
+void add_device_grouped_conv3d_bwd_data_xdl_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_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_xdl_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_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
 template <ck::index_t NumDimSpatial,
           typename OutLayout,
           typename WeiLayout,
@@ -139,43 +224,96 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
-                     is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, GNHWK>)
+        if constexpr(NumDimSpatial == 2)
         {
-            if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
-                         is_same_v<OutDataType, F16>)
-            {
-                add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
-                              is_same_v<OutDataType, F32>)
+
+            if constexpr(is_same_v<InLayout, GNHWC> && is_same_v<WeiLayout, GKYXC> &&
+                         is_same_v<OutLayout, GNHWK>)
             {
-                add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
+                if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                             is_same_v<OutDataType, F16>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                                  is_same_v<OutDataType, F32>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                                  is_same_v<OutDataType, BF16>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
+                        op_ptrs);
+                }
             }
-            else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
-                              is_same_v<OutDataType, BF16>)
+            else if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
+                              is_same_v<OutLayout, NHWGK>)
             {
-                add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
+                if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                             is_same_v<OutDataType, F16>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                                  is_same_v<OutDataType, F32>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                                  is_same_v<OutDataType, BF16>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
+                        op_ptrs);
+                }
             }
         }
-        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
-                          is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
+        else if constexpr(NumDimSpatial == 3)
         {
-            if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
-                         is_same_v<OutDataType, F16>)
-            {
-                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
-                              is_same_v<OutDataType, F32>)
+
+            if constexpr(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
+                         is_same_v<OutLayout, GNDHWK>)
             {
-                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
+                if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                             is_same_v<OutDataType, F16>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                                  is_same_v<OutDataType, F32>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                                  is_same_v<OutDataType, BF16>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
+                        op_ptrs);
+                }
             }
-            else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
-                              is_same_v<OutDataType, BF16>)
+            else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
+                              is_same_v<OutLayout, NDHWGK>)
             {
-                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(op_ptrs);
+                if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                             is_same_v<OutDataType, F16>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                                  is_same_v<OutDataType, F32>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                                  is_same_v<OutDataType, BF16>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
+                        op_ptrs);
+                }
             }
         }
 

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

@@ -164,6 +164,42 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
 template <ck::index_t NumDimSpatial,
           typename InLayout,
           typename WeiLayout,
@@ -273,8 +309,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
         }
         else if constexpr(NumDimSpatial == 3)
         {
-            if(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
-               is_same_v<OutLayout, GNDHWK>)
+            if constexpr(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
+                         is_same_v<OutLayout, GNDHWK>)
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
@@ -296,6 +332,29 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                         op_ptrs);
                 }
             }
+            else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
+                              is_same_v<OutLayout, NDHWGK>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
+            }
         }
 
         return op_ptrs;

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

@@ -151,6 +151,12 @@ void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               GKYXC,
                                                               Empty_Tuple,
                                                               GNHWK,
+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,
@@ -169,6 +175,16 @@ void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
                                                               int8_t,
                                                               Empty_Tuple,
                                                               int8_t,
+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);
@@ -428,11 +444,13 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
             }
             else 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);
+                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
             }
             else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                               is_same_v<WeiDataType, ck::bhalf_t> &&

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

@@ -10,7 +10,7 @@
 #include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -192,3 +192,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// fp16_output
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         Row_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         F32_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances);
+
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         Row_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         F32_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances);
+
+// fp32_output
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         Row_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         F32_Tuple,
+                                                         F32,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances);
+
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         Row_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         F32_Tuple,
+                                                         F32,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances);
+
+template <typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename EDataType>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceGroupedGemmFixedNK<ALayout,
+                                                           BLayout,
+                                                           Row_Tuple,
+                                                           ELayout,
+                                                           ADataType,
+                                                           BDataType,
+                                                           F32_Tuple,
+                                                           EDataType,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Add>>
+{
+    using DeviceOp = DeviceGroupedGemmFixedNK<ALayout,
+                                              BLayout,
+                                              Row_Tuple,
+                                              ELayout,
+                                              ADataType,
+                                              BDataType,
+                                              F32_Tuple,
+                                              EDataType,
+                                              PassThrough,
+                                              PassThrough,
+                                              Add>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        // fp16_output
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+            }
+        }
+
+        // fp32_output
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<EDataType, float>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_instances(op_ptrs);
+            }
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_instances(op_ptrs);
+            }
+        }
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

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

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

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

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

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/tensor_operation/gpu/device/device_max_pool_bwd.hpp"
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+#ifdef CK_ENABLE_FP16
+void add_device_maxpool_bwd_f16_instances(
+    std::vector<std::unique_ptr<DeviceMaxPoolBwd<F16, I32, F16>>>&);
+#endif
+#ifdef CK_ENABLE_BF16
+void add_device_maxpool_bwd_bf16_instances(
+    std::vector<std::unique_ptr<DeviceMaxPoolBwd<BF16, I32, BF16>>>&);
+#endif
+#ifdef CK_ENABLE_FP32
+void add_device_maxpool_bwd_f32_instances(
+    std::vector<std::unique_ptr<DeviceMaxPoolBwd<F32, I32, F32>>>&);
+#endif
+template <typename DOutDataType, typename IndexDataType, typename DInDataType>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceMaxPoolBwd<DOutDataType, IndexDataType, DInDataType>>
+{
+    using DeviceOp = DeviceMaxPoolBwd<DOutDataType, IndexDataType, DInDataType>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+#ifdef CK_ENABLE_FP16
+        if constexpr(is_same_v<DOutDataType, F16> && is_same_v<DInDataType, F16> &&
+                     is_same_v<IndexDataType, I32>)
+            add_device_maxpool_bwd_f16_instances(op_ptrs);
+#endif
+#ifdef CK_ENABLE_BF16
+        else if constexpr(is_same_v<DOutDataType, BF16> && is_same_v<DInDataType, BF16> &&
+                          is_same_v<IndexDataType, I32>)
+            add_device_maxpool_bwd_bf16_instances(op_ptrs);
+#endif
+#ifdef CK_ENABLE_FP32
+        else if constexpr(is_same_v<DOutDataType, F32> && is_same_v<DInDataType, F32> &&
+                          is_same_v<IndexDataType, I32>)
+            add_device_maxpool_bwd_f32_instances(op_ptrs);
+#endif
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef CK_ENABLE_FP16
 // FP16
 void add_device_normalization_rank_2_1_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, PassThrough, 2, 1>>>&);
@@ -26,7 +26,8 @@ void add_device_normalization_rank_4_3_f16_instances(
 
 void add_device_normalization_rank_5_3_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, PassThrough, 5, 3>>>&);
-
+#endif
+#ifdef CK_ENABLE_FP32
 // FP32
 void add_device_normalization_rank_2_1_f32_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, PassThrough, 2, 1>>>&);
@@ -36,7 +37,7 @@ void add_device_normalization_rank_4_3_f32_instances(
 
 void add_device_normalization_rank_5_3_f32_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, PassThrough, 5, 3>>>&);
-
+#endif
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
@@ -65,7 +66,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<XDataType, F16> && is_same_v<GammaDataType, F16> &&
                      is_same_v<BetaDataType, F16> && is_same_v<YDataType, F16>)
         {
@@ -82,8 +83,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
                 add_device_normalization_rank_5_3_f16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F32> && is_same_v<GammaDataType, F32> &&
-                          is_same_v<BetaDataType, F32> && is_same_v<YDataType, F32>)
+#endif
+#ifdef CK_ENABLE_FP32
+        if constexpr(is_same_v<XDataType, F32> && is_same_v<GammaDataType, F32> &&
+                     is_same_v<BetaDataType, F32> && is_same_v<YDataType, F32>)
         {
             if constexpr(Rank == 2 && NumReduceDim == 1)
             {
@@ -98,7 +101,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
                 add_device_normalization_rank_5_3_f32_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };

library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr auto InOutRank  = 4;
-static constexpr auto WindowRank = 2;
-
-static constexpr auto MaxOp = ck::ReduceTensorOp::MAX;
-static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
-
-// FP16
-void add_device_pool2d_fwd_nhwc_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, false>>>&);
-
-void add_device_pool2d_fwd_nhwc_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, AvgOp, false>>>&);
-
-// FP16 - return index
-void add_device_pool2d_fwd_nhwc_index_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, true>>>&);
-
-// FP32
-void add_device_pool2d_fwd_nhwc_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, false>>>&);
-
-void add_device_pool2d_fwd_nhwc_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, AvgOp, false>>>&);
-
-// FP32 - return index
-void add_device_pool2d_fwd_nhwc_index_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, true>>>&);
-
-template <typename InDataType,
-          typename OutDataType,
-          typename IndexDataType,
-          ck::ReduceTensorOp ReduceOpId,
-          bool OutputIndex>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFwd<InOutRank,
-                                                                                  WindowRank,
-                                                                                  InDataType,
-                                                                                  OutDataType,
-                                                                                  IndexDataType,
-                                                                                  ReduceOpId,
-                                                                                  OutputIndex>>
-{
-    using DeviceOp = DevicePoolFwd<InOutRank,
-                                   WindowRank,
-                                   InDataType,
-                                   OutDataType,
-                                   IndexDataType,
-                                   ReduceOpId,
-                                   OutputIndex>;
-
-    static auto GetInstances()
-    {
-        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
-        if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
-                     is_same_v<IndexDataType, I32>)
-        {
-            if constexpr(OutputIndex && ReduceOpId == MaxOp)
-            {
-                add_device_pool2d_fwd_nhwc_index_f16_instances(op_ptrs);
-            }
-            else
-            {
-                add_device_pool2d_fwd_nhwc_f16_instances(op_ptrs);
-            }
-        }
-        else if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
-                          is_same_v<IndexDataType, I32>)
-        {
-            if constexpr(OutputIndex && ReduceOpId == MaxOp)
-            {
-                add_device_pool2d_fwd_nhwc_index_f32_instances(op_ptrs);
-            }
-            else
-            {
-                add_device_pool2d_fwd_nhwc_f32_instances(op_ptrs);
-            }
-        }
-
-        return op_ptrs;
-    }
-};
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp

@@ -22,38 +22,56 @@ static constexpr auto WindowRank = 3;
 
 static constexpr auto MaxOp = ck::ReduceTensorOp::MAX;
 static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
-
+#ifdef CK_ENABLE_FP16
 // FP16
 void add_device_pool3d_fwd_ndhwc_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, false>>>&);
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, NDHWC, NDHWC, MaxOp, false>>>&);
 
 void add_device_pool3d_fwd_ndhwc_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, AvgOp, false>>>&);
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, NDHWC, NDHWC, AvgOp, false>>>&);
 
 // FP16 - return index
 void add_device_pool3d_fwd_ndhwc_index_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, true>>>&);
-
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, NDHWC, NDHWC, MaxOp, true>>>&);
+#endif
+#ifdef CK_ENABLE_BF16
+// BF16
+void add_device_pool3d_fwd_ndhwc_bf16_instances(
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, BF16, BF16, I32, NDHWC, NDHWC, MaxOp, false>>>&);
+
+void add_device_pool3d_fwd_ndhwc_bf16_instances(
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, BF16, BF16, I32, NDHWC, NDHWC, AvgOp, false>>>&);
+
+// BF16 - return index
+void add_device_pool3d_fwd_ndhwc_index_bf16_instances(
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, BF16, BF16, I32, NDHWC, NDHWC, MaxOp, true>>>&);
+#endif
+#ifdef CK_ENABLE_FP32
 // FP32
 void add_device_pool3d_fwd_ndhwc_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, false>>>&);
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, NDHWC, NDHWC, MaxOp, false>>>&);
 
 void add_device_pool3d_fwd_ndhwc_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, AvgOp, false>>>&);
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, NDHWC, NDHWC, AvgOp, false>>>&);
 
 // FP32 - return index
 void add_device_pool3d_fwd_ndhwc_index_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, true>>>&);
-
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, NDHWC, NDHWC, MaxOp, true>>>&);
+#endif
 template <typename InDataType,
           typename OutDataType,
           typename IndexDataType,
+          typename InLayout,
+          typename OutLayout,
           ck::ReduceTensorOp ReduceOpId,
           bool OutputIndex>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFwd<InOutRank,
@@ -61,6 +79,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
                                                                                   InDataType,
                                                                                   OutDataType,
                                                                                   IndexDataType,
+                                                                                  InLayout,
+                                                                                  OutLayout,
                                                                                   ReduceOpId,
                                                                                   OutputIndex>>
 {
@@ -69,36 +89,58 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
                                    InDataType,
                                    OutDataType,
                                    IndexDataType,
+                                   InLayout,
+                                   OutLayout,
                                    ReduceOpId,
                                    OutputIndex>;
 
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
-        if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
-                     is_same_v<IndexDataType, I32>)
+        if constexpr(is_same_v<InLayout, NDHWC> && is_same_v<OutLayout, NDHWC>)
         {
-            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
+                         is_same_v<IndexDataType, I32>)
             {
-                add_device_pool3d_fwd_ndhwc_index_f16_instances(op_ptrs);
+                if constexpr(OutputIndex && ReduceOpId == MaxOp)
+                {
+                    add_device_pool3d_fwd_ndhwc_index_f16_instances(op_ptrs);
+                }
+                else
+                {
+                    add_device_pool3d_fwd_ndhwc_f16_instances(op_ptrs);
+                }
             }
-            else
-            {
-                add_device_pool3d_fwd_ndhwc_f16_instances(op_ptrs);
-            }
-        }
-        else if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
-                          is_same_v<IndexDataType, I32>)
-        {
-            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+#endif
+#ifdef CK_ENABLE_BF16
+            else if constexpr(is_same_v<InDataType, BF16> && is_same_v<OutDataType, BF16> &&
+                              is_same_v<IndexDataType, I32>)
             {
-                add_device_pool3d_fwd_ndhwc_index_f32_instances(op_ptrs);
+                if constexpr(OutputIndex && ReduceOpId == MaxOp)
+                {
+                    add_device_pool3d_fwd_ndhwc_index_bf16_instances(op_ptrs);
+                }
+                else
+                {
+                    add_device_pool3d_fwd_ndhwc_bf16_instances(op_ptrs);
+                }
             }
-            else
+#endif
+#ifdef CK_ENABLE_FP32
+            else if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
+                              is_same_v<IndexDataType, I32>)
             {
-                add_device_pool3d_fwd_ndhwc_f32_instances(op_ptrs);
+                if constexpr(OutputIndex && ReduceOpId == MaxOp)
+                {
+                    add_device_pool3d_fwd_ndhwc_index_f32_instances(op_ptrs);
+                }
+                else
+                {
+                    add_device_pool3d_fwd_ndhwc_f32_instances(op_ptrs);
+                }
             }
+#endif
         }
 
         return op_ptrs;

library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.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_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef DL_KERNELS
 // Layout(A, B, C) = [Col, Row, Row]
 void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
@@ -76,7 +76,7 @@ void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
                                                     PassThrough,
                                                     Activation_Mul_Clamp<PassThrough>>>>&
         instances);
-
+#endif
 // Layout(A, B, C) = [Col, Row, Row]
 void add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
@@ -181,7 +181,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
+#endif
                     add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
                 }
             }
@@ -190,7 +192,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
+#endif
                     add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
                 }
             }
@@ -199,7 +203,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
+#endif
                     add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
                 }
             }
@@ -208,7 +214,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
+#endif
                     add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
                 }
             }
@@ -222,3 +230,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
\ No newline at end of file

library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.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_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef DL_KERNELS
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(
     std::vector<
@@ -64,7 +64,7 @@ void add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(
                                                       PassThrough,
                                                       Add_Mul2_Activation_Mul_Clamp<TanH>>>>&
         instances);
-
+#endif
 void add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -163,12 +163,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<Activation, Relu>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_relu_perchannel_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_relu_perchannel_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -229,7 +233,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, TanH>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_tanh_perchannel_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -243,3 +249,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif

library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.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_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef DL_KERNELS
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(
     std::vector<
@@ -63,7 +63,7 @@ void add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(
                                                               PassThrough,
                                                               Add_Mul_Activation_Mul_Clamp<TanH>>>>&
         instances);
-
+#endif
 void add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -161,12 +161,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<Activation, Relu>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_relu_perlayer_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_relu_perlayer_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -227,7 +231,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, TanH>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_tanh_perlayer_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -241,3 +247,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif

library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.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_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef DL_KERNELS
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -47,7 +47,7 @@ void add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(
                                                               PassThrough,
                                                               Activation_Mul2_Clamp<Relu>>>>&
         instances);
-
+#endif
 void add_device_conv2d_xdl_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -128,12 +128,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_perchannel_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_perchannel_quantization_int8_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<Activation, Relu>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_relu_perchannel_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -147,3 +151,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif

library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.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_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef DL_KERNELS
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -47,7 +47,7 @@ void add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(
                                                               PassThrough,
                                                               Activation_Mul_Clamp<Relu>>>>&
         instances);
-
+#endif
 void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -125,12 +125,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_perlayer_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_perlayer_quantization_int8_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<Activation, Relu>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -144,3 +148,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif