From 0f67b3bde810472153b309b747d4ef60042e047c Mon Sep 17 00:00:00 2001
From: Bartlomiej Kocot <barkocot@amd.com>
Date: Tue, 7 Oct 2025 22:20:35 +0000
Subject: [PATCH] fix

---
 ...ice_grouped_conv_bwd_data_xdl_instance.hpp | 118 ++++++++++++++++++
 .../grouped_conv2d_bwd_data/CMakeLists.txt    |  38 +++++-
 .../grouped_conv3d_bwd_data/CMakeLists.txt    |  38 +++++-
 3 files changed, 182 insertions(+), 12 deletions(-)

diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
index 11a8ff8e91..f16a345e14 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
@@ -76,6 +76,47 @@ using device_grouped_conv_bwd_data_xdl_f16_16_16_instances =
         // clang-format on
         >;
 
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_xdl_f16_optimized_loads_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|
+        // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+        // A K1 one access for each thread per load 
+        // 32x32
+        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,    32,    64,  16,   16,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 4, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 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,    32,    32,   8,    8,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 4, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 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,    32,    16,   4,    4,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 4, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 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,    32,    64,  16,   16,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 4, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 16, 1, 16>,               2>,
+        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,    32,    32,   8,    8,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 16, 1, 16>,               2>,
+        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,    32,    16,   4,    4,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 16, 1, 16>,               2>,
+        
+        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,    32,    64,  16,   16,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 4, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 8, 1, 32>,                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,    32,    32,   8,    8,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 8, 1, 32>,                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,    32,    16,   4,    4,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 8, 1, 32>,                1>,
+        // 16x16      
+        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,    16,    64,  16,   16,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 2, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 1, 4>,                4>,
+        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,    16,    32,   8,    8,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 2, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 1, 4>,                4>,
+        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,    16,    16,   4,    4,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 2, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 1, 4>,                4>,
+
+        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,    16,    64,  16,   16,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 4, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 32, 1, 8>,                2>,
+        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,    16,    32,   8,    8,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 32, 1, 8>,                2>,
+        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,    16,    16,   4,    4,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 32, 1, 8>,                2>,
+
+        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,    16,    64,  16,   16,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 4, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 16, 1, 16>,               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,   64,    16,    32,   8,    8,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 16, 1, 16>,               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,   64,    16,    16,   4,    4,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 16, 1, 16>,               1>
+        // clang-format on
+        >;
+
 template <index_t NDimSpatial,
           typename ALayout,
           typename BLayout,
@@ -178,6 +219,48 @@ using device_grouped_conv_bwd_data_xdl_bf16_16_16_instances = std::tuple<
         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,    16,    32,   8,   8,   16,   16,       4,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,        S<4, 2, 1>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
         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,    16,    32,   8,   8,   16,   16,       4,       1,       S<4, 16, 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,              1,              8,         1,            1,            1,     S<1, 16, 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,    64,    64,    16,    32,   8,   8,   16,   16,       4,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,        S<4, 2, 1>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                4>
+
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_xdl_bf16_optimized_loads_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|
+        // ##############################################|              |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+        // A K1 one access for each thread per load 
+        // 32x32
+        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,    32,    64,  16,   16,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 4, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 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,    32,    32,   8,    8,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 4, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 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,    32,    16,   4,    4,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 4, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 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,    32,    64,  16,   16,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 4, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 16, 1, 16>,               2>,
+        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,    32,    32,   8,    8,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 16, 1, 16>,               2>,
+        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,    32,    16,   4,    4,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 16, 1, 16>,               2>,
+        
+        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,    32,    64,  16,   16,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 4, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 8, 1, 32>,                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,    32,    32,   8,    8,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 8, 1, 32>,                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,    32,    16,   4,    4,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 8, 1, 32>,                1>,
+        // 16x16        
+        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,    16,    64,  16,   16,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 2, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 1, 4>,                4>,
+        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,    16,    32,   8,    8,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 2, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 1, 4>,                4>,
+        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,    16,    16,   4,    4,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 2, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 1, 4>,                4>,
+
+        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,    16,    64,  16,   16,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 4, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 32, 1, 8>,                2>,
+        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,    16,    32,   8,    8,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 32, 1, 8>,                2>,
+        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,    16,    16,   4,    4,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 32, 1, 8>,                2>,
+
+        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,    16,    64,  16,   16,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              4,         1,        S<4, 4, 16>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 16, 1, 16>,               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,   64,    16,    32,   8,    8,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 16, 1, 16>,               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,   64,    16,    16,   4,    4,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 16, 1, 16>,               1>
+
     // clang-format on
     >;
 
@@ -257,6 +340,41 @@ using device_grouped_conv_bwd_data_xdl_f32_16_16_instances =
         // clang-format on
         >;
 
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_xdl_f32_optimized_loads_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|
+       // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+       // A K1 one access for each thread per load 
+       // 32x32
+        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,    32,    32,   8,    8,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 4, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         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,    32,    16,   4,    4,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 4, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         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,    32,    32,   8,    8,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 16, 1, 16>,               2>,
+        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,    32,    16,   4,    4,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 16, 1, 16>,               2>,
+        
+        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,    32,    32,   8,    8,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 8, 1, 32>,                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,    32,    16,   4,    4,  32,   32,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 8, 1, 32>,                1>,
+        // 16x16      
+        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,    16,    32,   8,    8,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 2, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 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,   64,    16,    16,   4,    4,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 2, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              1,         1,            1,            1,     S<1, 64, 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,   64,    16,    32,   8,    8,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 32, 1, 8>,                2>,
+        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,    16,    16,   4,    4,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 4>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         1,            1,            1,     S<1, 32, 1, 8>,                2>,
+
+        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,    16,    32,   8,    8,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              4,         1,        S<4, 8, 8>,      S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 16, 1, 16>,               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,   64,    16,    16,   4,    4,  16,   16,       1,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,            1,            1,     S<1, 16, 1, 16>,               1>
+        // clang-format on
+        >;
+
 template <index_t NDimSpatial,
           typename ALayout,
           typename BLayout,
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
index d02210ab57..ac8eb64afb 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
@@ -61,7 +61,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_16_16_instance.in
-  NUM_SHARDS 15
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV2D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -70,7 +70,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_16_16_instance.in
-  NUM_SHARDS 15
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV2D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -79,7 +79,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_16_16_instance.in
-  NUM_SHARDS 12
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV2D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -115,7 +115,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv2d_bwd_data_xdl_ngchw_gkcyx_ngkhw_f16_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv2d_bwd_data_xdl_ngchw_gkcyx_ngkhw_f16_16_16_instance.in
-  NUM_SHARDS 15
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV2D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -124,7 +124,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv2d_bwd_data_xdl_ngchw_gkcyx_ngkhw_bf16_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv2d_bwd_data_xdl_ngchw_gkcyx_ngkhw_bf16_16_16_instance.in
-  NUM_SHARDS 15
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV2D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -133,7 +133,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv2d_bwd_data_xdl_ngchw_gkcyx_ngkhw_f32_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv2d_bwd_data_xdl_ngchw_gkcyx_ngkhw_f32_16_16_instance.in
-  NUM_SHARDS 12
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV2D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -192,6 +192,32 @@ generate_sharded_instantiations(
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
 
+set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
+generate_sharded_instantiations(
+  INSTANCES_NAME device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_optimized_loads_instances
+  TEMPLATE_FILE xdl/device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_optimized_loads_instance.in
+  NUM_SHARDS 18
+  SRC_LIST GROUPED_CONV2D_BWD_DATA
+  OUTPUT_DIR ${GENERATED_DIR}/xdl
+)
+
+set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
+generate_sharded_instantiations(
+  INSTANCES_NAME device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_optimized_loads_instances
+  TEMPLATE_FILE xdl/device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_optimized_loads_instance.in
+  NUM_SHARDS 18
+  SRC_LIST GROUPED_CONV2D_BWD_DATA
+  OUTPUT_DIR ${GENERATED_DIR}/xdl
+)
+
+set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
+generate_sharded_instantiations(
+  INSTANCES_NAME device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_optimized_loads_instances
+  TEMPLATE_FILE xdl/device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_optimized_loads_instance.in
+  NUM_SHARDS 12
+  SRC_LIST GROUPED_CONV2D_BWD_DATA
+  OUTPUT_DIR ${GENERATED_DIR}/xdl
+)
        
 set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
index ad1b85401d..05ffedaa5f 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
@@ -61,7 +61,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_16_16_instance.in
-  NUM_SHARDS 15
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV3D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -70,7 +70,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_16_16_instance.in
-  NUM_SHARDS 15
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV3D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -79,7 +79,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_16_16_instance.in
-  NUM_SHARDS 12
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV3D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -115,7 +115,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkczyx_ngkdhw_f16_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkczyx_ngkdhw_f16_16_16_instance.in
-  NUM_SHARDS 15
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV3D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -124,7 +124,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkczyx_ngkdhw_bf16_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkczyx_ngkdhw_bf16_16_16_instance.in
-  NUM_SHARDS 15
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV3D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -133,7 +133,7 @@ set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(
   INSTANCES_NAME device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkczyx_ngkdhw_f32_16_16_instances
   TEMPLATE_FILE xdl/device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkczyx_ngkdhw_f32_16_16_instance.in
-  NUM_SHARDS 12
+  NUM_SHARDS 6
   SRC_LIST GROUPED_CONV3D_BWD_DATA
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
@@ -192,6 +192,32 @@ generate_sharded_instantiations(
   OUTPUT_DIR ${GENERATED_DIR}/xdl
 )
 
+set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
+generate_sharded_instantiations(
+  INSTANCES_NAME device_grouped_conv3d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_optimized_loads_instances
+  TEMPLATE_FILE xdl/device_grouped_conv3d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_optimized_loads_instance.in
+  NUM_SHARDS 18
+  SRC_LIST GROUPED_CONV3D_BWD_DATA
+  OUTPUT_DIR ${GENERATED_DIR}/xdl
+)
+
+set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
+generate_sharded_instantiations(
+  INSTANCES_NAME device_grouped_conv3d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_optimized_loads_instances
+  TEMPLATE_FILE xdl/device_grouped_conv3d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_optimized_loads_instance.in
+  NUM_SHARDS 18
+  SRC_LIST GROUPED_CONV3D_BWD_DATA
+  OUTPUT_DIR ${GENERATED_DIR}/xdl
+)
+
+set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
+generate_sharded_instantiations(
+  INSTANCES_NAME device_grouped_conv3d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_optimized_loads_instances
+  TEMPLATE_FILE xdl/device_grouped_conv3d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_optimized_loads_instance.in
+  NUM_SHARDS 12
+  SRC_LIST GROUPED_CONV3D_BWD_DATA
+  OUTPUT_DIR ${GENERATED_DIR}/xdl
+)
    
 set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
 generate_sharded_instantiations(