Adding Instances and Examples for FP8-based Scaled Convolution with ReLU...

Adding Instances and Examples for FP8-based Scaled Convolution with ReLU Activation and AMAX Reduction. (#1469)

* Enable CMakePresets build

* Verify Convolution, Scaling and ReLU algorithms.

* Add tensor element-wise scale and type cast operation.

* Reduction implemented but does not work.

* Exploration of Reduction functionality.

* Completed example for Convolution scaled with ReLu activation and AMAX reduction.

* WIP: Add required instances for convolution.

* WIP: Create client example. Implement convolution stage.

* Add elementwise instances.

* Add elementwise scale + convert example.

* Add reduction instances.

* WIP: Client example for AMAX reduction.

* WIP: Add instances for multistage reduction.

* WIP: Implementation of multistage reduction.

* Refactoring.

* Clean up.

* Guard off FP8 instances when the data type is not available.

* Improve output readability.

* Addressing reviewer's comments.

client_example/24_grouped_conv_activation/CMakeLists.txt

@@ -47,6 +47,14 @@ target_link_libraries(client_conv3d_fwd_convscale_add_fp8 PRIVATE composable_ker
 add_executable(client_conv3d_fwd_convscale_relu_fp8
                grouped_convnd_fwd_convscale_relu/conv3d_fwd_convscale_relu_fp8.cpp)
 target_link_libraries(client_conv3d_fwd_convscale_relu_fp8 PRIVATE composable_kernel::device_conv_operations)
+# Fwd convscale + ReLU + AMAX
+add_executable(client_conv3d_fwd_convscale_relu_amax_fp8
+               grouped_convnd_fwd_convscale_reduce/conv3d_fwd_convscale_relu_amax_fp8.cpp)
+target_link_libraries(client_conv3d_fwd_convscale_relu_amax_fp8 
+                      PRIVATE composable_kernel::device_conv_operations 
+                              composable_kernel::device_other_operations 
+                              composable_kernel::device_reduction_operations 
+                              utility)
 # Fwd convscale
 add_executable(client_conv3d_fwd_convscale_fp8
                grouped_convnd_fwd_convscale/conv3d_fwd_convscale_fp8.cpp)

client_example/24_grouped_conv_activation/grouped_convnd_fwd_convscale_reduce/conv3d_fwd_convscale_relu_amax_fp8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType       = ck::f8_t;
+using WeiDataType      = ck::f8_t;
+using CShuffleDataType = float;
+using ConvOutDataType  = float;    // data type of convolution result
+using OutDataType      = ck::f8_t; // data type of final result
+using AComputeDataType = ck::f8_t;
+using BComputeDataType = ck::f8_t;
+
+using ConvElementOp = ConvScaleRelu;
+
+using InLayout  = ck::tensor_layout::convolution::NDHWGC;
+using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout = ck::tensor_layout::convolution::NDHWGK;
+
+constexpr auto ReduceOpId = ck::ReduceTensorOp::AMAX;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t N             = 64;
+static constexpr ck::index_t K             = 128;
+static constexpr ck::index_t C             = 64;
+static constexpr ck::index_t Z             = 3;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Di            = 28;
+static constexpr ck::index_t Hi            = 28;
+static constexpr ck::index_t Wi            = 3;
+static constexpr ck::index_t Do            = 28;
+static constexpr ck::index_t Ho            = 28;
+static constexpr ck::index_t Wo            = 3;
+
+int main()
+{
+    return run_grouped_conv_fwd_convscale_reduce<NumDimSpatial,
+                                                 InDataType,
+                                                 WeiDataType,
+                                                 ConvOutDataType,
+                                                 OutDataType,
+                                                 ConvElementOp,
+                                                 ReduceOpId,
+                                                 InLayout,
+                                                 WeiLayout,
+                                                 OutLayout,
+                                                 3,
+                                                 AComputeDataType,
+                                                 BComputeDataType>(
+               {N, Di, Hi, Wi, G, C}, {G, K, Z, Y, X, C}, {N, Do, Ho, Wo, G, K})
+               ? EXIT_SUCCESS
+               : EXIT_FAILURE;
+}

example/62_convnd_activ/CMakeLists.txt

@@ -3,6 +3,7 @@ add_subdirectory(convinvscale)
 add_subdirectory(convscale)
 add_subdirectory(convscale_relu)
 add_subdirectory(convscale_add)
+add_subdirectory(convscale_reduce)
 add_subdirectory(multi_AB)
 add_subdirectory(unary)
 

example/62_convnd_activ/convscale_reduce/CMakeLists.txt

+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+     add_custom_target(example_convnd_activ_xdl_convscale_reduce)
+     add_example_executable(example_convnd_fwd_xdl_convscale_relu_amax_fp8 convnd_fwd_xdl_convscale_relu_amax_fp8.cpp)
+     add_example_dependencies(example_convnd_activ_xdl_convscale_reduce example_convnd_fwd_xdl_convscale_relu_amax_fp8 )
+
+     set(target 1)
+ endif()
+endforeach()

example/62_convnd_activ/convscale_reduce/convnd_fwd_xdl_convscale_relu_amax_fp8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_convscale_reduce_common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp"
+
+using InDataType       = ck::f8_t;
+using WeiDataType      = ck::f8_t;
+using AccDataType      = float;
+using CShuffleDataType = float;
+using ConvOutDataType  = float;    // data type of convolution result
+using OutDataType      = ck::f8_t; // data type of final result
+using AComputeDataType = ck::f8_t;
+using BComputeDataType = ck::f8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using OutElementOp = ConvScaleRelu;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<
+        NDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        AccDataType,
+        CShuffleDataType,
+        ck::Tuple<>,
+        ConvOutDataType,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        1,           //
+        256,         // BlockSize
+        128,         // MPerBlock
+        256,         // NPerBlock
+        32,          // KPerBlock
+        8,           // AK1
+        8,           // BK1
+        32,          // MPerXdl
+        32,          // NPerXdl
+        2,           // MXdlPerWave
+        4,           // NXdlPerWave
+        S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
+        S<1, 0, 2>,  // ABlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // ABlockTransferSrcAccessOrder
+        2,           // ABlockTransferSrcVectorDim
+        8,           // ABlockTransferSrcScalarPerVector
+        8,           // ABlockTransferDstScalarPerVector_AK1
+        1,           // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        8,           // BBlockTransferSrcScalarPerVector
+        8,           // BBlockTransferDstScalarPerVector_BK1
+        1,           // BBlockLdsExtraN
+        1,
+        1,
+        S<1, 32, 1, 8>,
+        8,
+        AComputeDataType,
+        BComputeDataType>;
+
+#include "run_convnd_fwd_example.inc"
+
+int main(int argc, char* argv[]) { return run_convnd_fwd_example(argc, argv) ? 0 : 1; }

example/62_convnd_activ/convscale_reduce/run_convnd_fwd_example.inc

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+bool run_convnd_fwd_example(int argc, char* argv[])
+{
+    print_helper_msg();
+
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    ck::utils::conv::ConvParam conv_param{
+        2, 1, 128, 256, 192, {3, 3}, {71, 71}, {2, 2}, {1, 1}, {1, 1}, {1, 1}};
+
+    if(argc == 1)
+    {
+        // use default
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else
+    {
+        do_verification                   = std::stoi(argv[1]);
+        init_method                       = std::stoi(argv[2]);
+        time_kernel                       = std::stoi(argv[3]);
+        const ck::index_t num_dim_spatial = std::stoi(argv[4]);
+
+        conv_param = ck::utils::conv::parse_conv_param(num_dim_spatial, 5, argv);
+    }
+
+    // instantiate in and wei element ops, will
+    // instantiate out_element_op below for every iteration
+    const auto in_element_op  = InElementOp{};
+    const auto wei_element_op = WeiElementOp{};
+
+    const auto run = [&](auto ndim_spatial, auto in_layout, auto wei_layout, auto out_layout) {
+        constexpr ck::index_t ndim_spatial_value = ndim_spatial.value;
+
+        using InLayout  = decltype(in_layout);
+        using WeiLayout = decltype(wei_layout);
+        using OutLayout = decltype(out_layout);
+
+        const auto in_g_n_c_wis_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
+                conv_param);
+
+        const auto wei_g_k_c_xs_desc =
+            ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
+                conv_param);
+
+        const auto out_g_n_k_wos_desc =
+            ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
+                conv_param);
+
+        return run_grouped_conv_fwd<
+            ndim_spatial_value,
+            InDataType,
+            WeiDataType,
+            ConvOutDataType,
+            OutDataType,
+            InElementOp,
+            WeiElementOp,
+            OutElementOp,
+            DeviceGroupedConvNDFwdInstance<ndim_spatial_value, InLayout, WeiLayout, OutLayout>>(
+            do_verification,
+            init_method,
+            time_kernel,
+            conv_param,
+            in_g_n_c_wis_desc,
+            wei_g_k_c_xs_desc,
+            out_g_n_k_wos_desc,
+            in_element_op,
+            wei_element_op);
+    };
+
+    namespace ctc = ck::tensor_layout::convolution;
+
+    if(conv_param.num_dim_spatial_ == 1)
+    {
+        return run(ck::Number<1>{}, ctc::GNWC{}, ctc::GKXC{}, ctc::GNWK{});
+    }
+    else if(conv_param.num_dim_spatial_ == 2)
+    {
+        return run(ck::Number<2>{}, ctc::GNHWC{}, ctc::GKYXC{}, ctc::GNHWK{});
+    }
+    else if(conv_param.num_dim_spatial_ == 3)
+    {
+        return run(ck::Number<3>{}, ctc::GNDHWC{}, ctc::GKZYXC{}, ctc::GNDHWK{});
+    }
+
+    return true;
+}

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

@@ -184,6 +184,43 @@ using device_grouped_conv_fwd_xdl_outelementop_bf8_f8_instances = std::tuple<
     // clang-format on
     >;
 
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec,
+          typename OutElementOp>
+using device_grouped_conv_fwd_xdl_outelementop_f8_f8_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| Compute| Compute|
+        //########################################|    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|   TypeA|   TypeB|
+        //########################################|           |       |       |            |       |      |      |        |         |         |      |   Operation|   Operation|     Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|        |        |
+        //########################################|           |       |       |            |       |      |      |        |         |         |      |            |            |              |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |        |        |
+#ifdef CK_ENABLE_FP8
+        // generic instance
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1,     F8,      F8>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,     F8,      F8>,
+
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,     F8,      F8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,  DsLayout,ELayout,    F8,    F8,     F32,      F32,  Tuple<>,   F32, PassThrough, PassThrough,  OutElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,     F8,      F8>
+#endif
+    // clang-format on
+    >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation

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

@@ -8,6 +8,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_abd.hpp"
+#include "ck/tensor_operation/gpu/element/combined_element_wise_operation.hpp"
 #include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
 
@@ -99,6 +100,89 @@ struct DeviceOperationInstanceFactory<
     }
 };
 
+namespace ew            = ck::tensor_operation::element_wise;
+using CombConvScaleRelu = ew::UnaryCombinedOp<ew::Scale, ew::Scale, ew::Relu>;
+
+#ifdef CK_ENABLE_FP8
+void add_device_grouped_conv3d_fwd_xdl_combconvscale_relu_ndhwgc_gkzyxc_ndhwgk_f8_f8_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<>,
+                                                                NDHWGK,
+                                                                F8,
+                                                                F8,
+                                                                ck::Tuple<>,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                CombConvScaleRelu,
+                                                                F8,
+                                                                F8>>>& instances);
+#endif
+
+template <ck::index_t NumDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename DLayouts,
+          typename OutLayout,
+          typename InDataType,
+          typename WeiDataType,
+          typename DDataTypes,
+          typename OutDataType,
+          typename AComputeType,
+          typename BComputeType>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                                                  InLayout,
+                                                                  WeiLayout,
+                                                                  DLayouts,
+                                                                  OutLayout,
+                                                                  InDataType,
+                                                                  WeiDataType,
+                                                                  DDataTypes,
+                                                                  OutDataType,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  CombConvScaleRelu,
+                                                                  AComputeType,
+                                                                  BComputeType>>
+{
+    using DeviceOp = DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                                     InLayout,
+                                                     WeiLayout,
+                                                     DLayouts,
+                                                     OutLayout,
+                                                     InDataType,
+                                                     WeiDataType,
+                                                     DDataTypes,
+                                                     OutDataType,
+                                                     PassThrough,
+                                                     PassThrough,
+                                                     CombConvScaleRelu,
+                                                     AComputeType,
+                                                     BComputeType>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+        if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
+                     is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
+        {
+#ifdef CK_ENABLE_FP8
+            if constexpr(is_same_v<InDataType, f8_t> && is_same_v<WeiDataType, f8_t> &&
+                         is_same_v<OutDataType, F32> && is_same_v<AComputeType, f8_t> &&
+                         is_same_v<BComputeType, f8_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_combconvscale_relu_ndhwgc_gkzyxc_ndhwgk_f8_f8_f32_instances(
+                    op_ptrs);
+            }
+#endif
+        }
+        return op_ptrs;
+    }
+};
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation

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

@@ -70,6 +70,12 @@ void add_device_permute_scale_6d_f32_instances(
         DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, element_wise::Scale, 6>>>&);
 #endif
 
+#ifdef CK_ENABLE_FP8
+void add_device_permute_scale_6d_f32_f8_instances(
+    std::vector<std::unique_ptr<
+        DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F8>, element_wise::Scale, 6>>>&);
+#endif
+
 template <typename InDataTypeTuple,
           typename OutDataTypeTuple,
           typename ElementwiseOperation,
@@ -184,6 +190,13 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_permute_scale_6d_f16_instances(op_ptrs);
             }
+#endif
+#ifdef CK_ENABLE_FP8
+            if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F32>> &&
+                         is_same_v<OutDataTypeTuple, ck::Tuple<F8>>)
+            {
+                add_device_permute_scale_6d_f32_f8_instances(op_ptrs);
+            }
 #endif
         }
         return op_ptrs;

library/include/ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp

@@ -10,6 +10,7 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F8  = ck::f8_t;
 using F16 = ck::half_t;
 using F32 = float;
 
@@ -183,6 +184,51 @@ using device_permute_scale_f32_instances = std::tuple<
          DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F32>, ElementwiseOp,  NDims, 32,   32,  16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
          DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F32>, ElementwiseOp,  NDims, 32,   16,  32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>
     >;
+
+#ifdef CK_ENABLE_FP8
+template <index_t NDims,
+          typename ElementwiseOp>
+using device_permute_scale_f32_f8_instances = std::tuple<
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 256,  64,  64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 256, 128,  32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 256,  32, 128, 4, 4,  ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128,  64,  32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128,  32,  64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128,  16, 128, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128, 128,  16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 64,   32,  32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 64,   16,  64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 64,   64,  16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 32,   32,  16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 32,   16,  32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
+
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 256, 128, 128, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 256, 256,  64, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 256,  64, 256, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128, 128,  64, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128,  64, 128, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128,  32, 256, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128, 256,  32, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 64,   64,  64, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 64,   32, 128, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 64,  128,  32, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 32,   64,  32, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 32,   32,  64, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
+
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 256,  64,  64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 256, 128,  32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 256,  32, 128, 4, 4,  ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128,  64,  32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128,  32,  64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128,  16, 128, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 128, 128,  16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 64,   32,  32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 64,   16,  64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 64,   64,  16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 32,   32,  16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
+         DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp,  NDims, 32,   16,  32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>
+    >;
+#endif
 // clang-format on
 
 } // namespace instance

library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.hpp

@@ -14,15 +14,24 @@ namespace device {
 namespace instance {
 
 // clang-format off
-// InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+//                                                        InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             3,       ReduceAMax,         UnaryAbs,       PassThrough,        false,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             4,       ReduceAMax,         UnaryAbs,       PassThrough,        false,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             1,       ReduceAMax,         UnaryAbs,       PassThrough,        false,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     2,             1,       ReduceAMax,         UnaryAbs,       PassThrough,        false,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             3,       ReduceAMax,         UnaryAbs,       PassThrough,        false,      true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             4,       ReduceAMax,         UnaryAbs,       PassThrough,        false,      true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             1,       ReduceAMax,         UnaryAbs,       PassThrough,        false,      true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     2,             1,       ReduceAMax,         UnaryAbs,       PassThrough,        false,      true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     6,             6,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 6, 6, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     5,             5,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 5, 5, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             4,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     6,             3,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 6, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     5,             3,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 5, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             3,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     3,             3,       ReduceAMax,      PassThrough,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 3, 3, ReduceAMax, PassThrough, PassThrough, true, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     2,             2,       ReduceAMax,      PassThrough,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 2, ReduceAMax, PassThrough, PassThrough, true, false>>&);
+extern template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     1,             1,       ReduceAMax,      PassThrough,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 1, 1, ReduceAMax, PassThrough, PassThrough, true, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_convscale_relu/xdl/device_grouped_conv3d_fwd_xdl_convscale_relu_ndhwgc_gkzyxc_ndhwgk_f8_instance.cpp

@@ -3,6 +3,7 @@
 
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_outelementop_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/tensor_operation/gpu/element/combined_element_wise_operation.hpp"
 #include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
 
 namespace ck {
@@ -57,6 +58,54 @@ void add_device_grouped_conv3d_fwd_xdl_convscale_relu_ndhwgc_gkzyxc_ndhwgk_f8_in
                                                               ConvScaleRelu>{});
 }
 
+namespace ew            = ck::tensor_operation::element_wise;
+using CombConvScaleRelu = ew::UnaryCombinedOp<ew::Scale, ew::Scale, ew::Relu>;
+
+void add_device_grouped_conv3d_fwd_xdl_combconvscale_relu_ndhwgc_gkzyxc_ndhwgk_f8_f8_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<>,
+                                                                NDHWGK,
+                                                                F8,
+                                                                F8,
+                                                                ck::Tuple<>,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                CombConvScaleRelu,
+                                                                F8,
+                                                                F8>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_outelementop_f8_f8_f32_instances<3,
+                                                                     NDHWGC,
+                                                                     GKZYXC,
+                                                                     ck::Tuple<>,
+                                                                     NDHWGK,
+                                                                     ConvFwdDefault,
+                                                                     CombConvScaleRelu>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_outelementop_f8_f8_f32_instances<3,
+                                                                     NDHWGC,
+                                                                     GKZYXC,
+                                                                     ck::Tuple<>,
+                                                                     NDHWGK,
+                                                                     ConvFwd1x1P0,
+                                                                     CombConvScaleRelu>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_outelementop_f8_f8_f32_instances<3,
+                                                                     NDHWGC,
+                                                                     GKZYXC,
+                                                                     ck::Tuple<>,
+                                                                     NDHWGK,
+                                                                     ConvFwd1x1S1P0,
+                                                                     CombConvScaleRelu>{});
+}
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation

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

-add_instance_library(device_permute_scale_instance 
+add_instance_library(device_permute_scale_instance
 	device_permute_scale_1d_fp16_instances.cpp
 	device_permute_scale_2d_fp16_instances.cpp
 	device_permute_scale_3d_fp16_instances.cpp
@@ -10,4 +10,5 @@ add_instance_library(device_permute_scale_instance
 	device_permute_scale_3d_fp32_instances.cpp
 	device_permute_scale_4d_fp32_instances.cpp
 	device_permute_scale_5d_fp32_instances.cpp
-	device_permute_scale_6d_fp32_instances.cpp)
+	device_permute_scale_6d_fp32_instances.cpp
+	device_permute_scale_6d_fp32_fp8_instances.cpp)

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_6d_fp32_fp8_instances.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Scale = element_wise::Scale;
+
+void add_device_permute_scale_6d_f32_f8_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F8>, Scale, 6>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP8
+    add_device_operation_instances(instances, device_permute_scale_f32_f8_instances<6, Scale>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.cpp

@@ -10,15 +10,24 @@ namespace device {
 namespace instance {
 
 // clang-format off
-// InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+//                                                 InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             3,       ReduceAMax,         UnaryAbs,       PassThrough,        false,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             4,       ReduceAMax,         UnaryAbs,       PassThrough,        false,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             1,       ReduceAMax,         UnaryAbs,       PassThrough,        false,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     2,             1,       ReduceAMax,         UnaryAbs,       PassThrough,        false,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             3,       ReduceAMax,         UnaryAbs,       PassThrough,        false,      true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             4,       ReduceAMax,         UnaryAbs,       PassThrough,        false,      true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             1,       ReduceAMax,         UnaryAbs,       PassThrough,        false,      true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     2,             1,       ReduceAMax,         UnaryAbs,       PassThrough,        false,      true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     6,             6,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 6, 6, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     5,             5,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 5, 5, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             4,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     6,             3,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 6, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     5,             3,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 5, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     4,             3,       ReduceAMax,         UnaryAbs,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     3,             3,       ReduceAMax,      PassThrough,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 3, 3, ReduceAMax, PassThrough, PassThrough, true, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     2,             2,       ReduceAMax,      PassThrough,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 2, ReduceAMax, PassThrough, PassThrough, true, false>>&);
+template void add_device_reduce_instance_blockwise<        F32,          F32,          F32,     1,             1,       ReduceAMax,      PassThrough,       PassThrough,         true,     false>(std::vector<DeviceReducePtr<F32, F32, F32, 1, 1, ReduceAMax, PassThrough, PassThrough, true, false>>&);
 // clang-format on
 
 } // namespace instance