Merge pull request #9 from ROCmSoftwarePlatform/mergedown-from-public

Merge down from public repo

example/15_grouped_gemm/grouped_gemm_multiple_d_dl_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <algorithm>
+#include <cstddef>
+#include <initializer_list>
+#include <iostream>
+#include <numeric>
+#include <stdexcept>
+#include <string>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/utility/sequence.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType   = F16;
+using BDataType   = F16;
+using AccDataType = F32;
+using DsDataType  = ck::Tuple<>;
+using EDataType   = F16;
+
+using ALayout  = Row;
+using BLayout  = Row;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = PassThrough;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::
+        //   ##################| ALayout| BLayout| DsLayout| ELayout|     AData|     BData|     AccData|     DsData|     EData|           A|           B|          CDE|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        //   ##################|        |        |         |        |      Type|      Type|        Type|       Type|      Type| Elementwise| Elementwise|  Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        //   ##################|        |        |         |        |          |          |            |           |          |   Operation|   Operation|    Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        //   ##################|        |        |         |        |          |          |            |           |          |            |            |             |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+  DeviceGroupedGemmMultipleD_Dl< ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, DsDataType, EDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>;
+// clang-format on
+
+#include "run_grouped_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_grouped_gemm_example(argc, argv); }

example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc

@@ -26,7 +26,7 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
     {
         split_k = 1;
     }
-    
+
     const auto in_g_n_c_wis_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<
             InputLayout<NDimSpatial>>(conv_param);

example/29_batched_gemm_bias_e_permute/CMakeLists.txt

 add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp)
 
-if(GPU_TARGETS MATCHES "gfx1100")
+if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
     add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
 endif()

example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt

@@ -16,7 +16,7 @@ if(USE_BITINT_EXTENSION_INT4)
   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
 endif() # USE_BITINT_EXTENSION_INT4
 
-if(GPU_TARGETS MATCHES "gfx1100")
+if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
 endif()
 

example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_xdl_fp16.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 

example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_wmma_example.inc

@@ -74,8 +74,8 @@ using DeviceConvFwdInstance =
         8,           // BBlockTransferSrcScalarPerVector
         8,           // BBlockTransferDstScalarPerVector_BK1
         true,        // BBlockLdsExtraN
-        1,
-        1,
+        4,
+        2,
         S<1, 32, 1, 8>,
         8>;
 

example/40_conv2d_fwd_quantization/CMakeLists.txt

+# Conv perlayer quantization
+add_example_executable(example_conv2d_fwd_dl_perlayer_quantization_int8 conv2d_fwd_dl_perlayer_quantization_int8.cpp)
+add_example_executable(example_conv2d_fwd_xdl_perlayer_quantization_int8 conv2d_fwd_xdl_perlayer_quantization_int8.cpp)
+
+# Conv perchannel quantization
+add_example_executable(example_conv2d_fwd_dl_perchannel_quantization_int8 conv2d_fwd_dl_perchannel_quantization_int8.cpp)
+add_example_executable(example_conv2d_fwd_xdl_perchannel_quantization_int8 conv2d_fwd_xdl_perchannel_quantization_int8.cpp)
+
+# Conv + bias + relu perlayer quantization
+add_example_executable(example_conv2d_fwd_dl_bias_relu_perlayer_quantization_int8 conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp)
+add_example_executable(example_conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8 conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp)
+
+# Conv + bias + relu perchannel quantization
+add_example_executable(example_conv2d_fwd_dl_bias_relu_perchannel_quantization_int8 conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp)
+add_example_executable(example_conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8 conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp)
+
+# Conv + bias + tanh perlayer quantization
+add_example_executable(example_conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8 conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp)
+
+# Conv + bias + tanh perchannel quantization
+add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)

example/40_conv2d_fwd_quantization/common.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType           = int8_t;
+using WeiDataType          = int8_t;
+using BiasDataType         = int32_t;
+using RequantScaleDataType = float;
+using AccDataType          = int32_t;
+using OutDataType          = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<ActivationOp>;
+
+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 BiasLayout,
+          typename RequantScaleLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<BiasDataType, RequantScaleDataType>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout, RequantScaleLayout>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_bias_perchannel_quantization_example.inc"
+
+int main()
+{
+    const auto out_element_op = OutElementOp{ActivationOp{}};
+    run_conv2d_fwd_bias_perchannel_quantization_example(out_element_op);
+};

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType   = int8_t;
+using WeiDataType  = int8_t;
+using BiasDataType = int32_t;
+using AccDataType  = int32_t;
+using OutDataType  = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<ActivationOp>;
+
+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 BiasLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<BiasDataType>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_bias_perlayer_quantization_example.inc"
+
+int main()
+{
+    float requant_scale       = 0.5f;
+    const auto out_element_op = OutElementOp{requant_scale, ActivationOp{}};
+    run_conv2d_fwd_bias_perlayer_quantization_example(out_element_op);
+}

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType           = int8_t;
+using WeiDataType          = int8_t;
+using BiasDataType         = int32_t;
+using RequantScaleDataType = float;
+using AccDataType          = int32_t;
+using OutDataType          = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::TanH;
+using OutElementOp =
+    ck::tensor_operation::element_wise::Add_Mul2_Activation_Mul_Clamp<ActivationOp>;
+
+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 BiasLayout,
+          typename RequantScaleLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<BiasDataType, RequantScaleDataType>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout, RequantScaleLayout>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_bias_perchannel_quantization_example.inc"
+
+int main()
+{
+    float scale_z_inv         = 0.5f;
+    const auto out_element_op = OutElementOp{scale_z_inv, ActivationOp{}};
+    run_conv2d_fwd_bias_perchannel_quantization_example(out_element_op);
+};

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType   = int8_t;
+using WeiDataType  = int8_t;
+using BiasDataType = int32_t;
+using AccDataType  = int32_t;
+using OutDataType  = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::TanH;
+using OutElementOp = ck::tensor_operation::element_wise::Add_Mul_Activation_Mul_Clamp<ActivationOp>;
+
+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 BiasLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<BiasDataType>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_bias_perlayer_quantization_example.inc"
+
+int main()
+{
+    float scale_acc           = 0.5f;
+    float scale_z_inv         = 0.5f;
+    const auto out_element_op = OutElementOp{scale_z_inv, scale_acc, ActivationOp{}};
+    run_conv2d_fwd_bias_perlayer_quantization_example(out_element_op);
+}

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType           = int8_t;
+using WeiDataType          = int8_t;
+using RequantScaleDataType = float;
+using AccDataType          = int32_t;
+using OutDataType          = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Activation_Mul2_Clamp<ActivationOp>;
+
+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 RequantScaleLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<RequantScaleDataType>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<RequantScaleLayout>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_perchannel_quantization_example.inc"
+
+int main()
+{
+    const auto out_element_op = OutElementOp{ActivationOp{}};
+    run_conv2d_fwd_perchannel_quantization_example(out_element_op);
+}

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType  = int8_t;
+using WeiDataType = int8_t;
+using AccDataType = int32_t;
+using OutDataType = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = PassThrough;
+using OutElementOp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
+
+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::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_perlayer_quantization_example.inc"
+
+int main()
+{
+    float requant_scale       = 0.5f;
+    const auto out_element_op = OutElementOp{requant_scale, ActivationOp{}};
+    run_conv2d_fwd_perlayer_quantization_example(out_element_op);
+}

example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+
+using InDataType           = int8_t;
+using WeiDataType          = int8_t;
+using BiasDataType         = int32_t;
+using RequantScaleDataType = float;
+using AccDataType          = int32_t;
+using CShuffleDataType     = AccDataType;
+using OutDataType          = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<ActivationOp>;
+
+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 BiasLayout,
+          typename RequantScaleLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
+        NDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout, RequantScaleLayout>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        AccDataType,
+        CShuffleDataType,
+        ck::Tuple<BiasDataType, RequantScaleDataType>,
+        OutDataType,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        1,           //
+        256,         // BlockSize
+        128,         // MPerBlock
+        256,         // NPerBlock
+        64,          // KPerBlock
+        16,          // AK1
+        16,          // 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
+        16,          // ABlockTransferSrcScalarPerVector
+        16,          // ABlockTransferDstScalarPerVector_AK1
+        1,           // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        16,          // BBlockTransferSrcScalarPerVector
+        16,          // BBlockTransferDstScalarPerVector_BK1
+        1,           // BBlockLdsExtraN
+        1,
+        1,
+        S<1, 64, 1, 4>,
+        8>;
+
+#include "run_conv2d_fwd_bias_perchannel_quantization_example.inc"
+
+int main()
+{
+    const auto out_element_op = OutElementOp{ActivationOp{}};
+    run_conv2d_fwd_bias_perchannel_quantization_example(out_element_op);
+};

example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+
+using InDataType       = int8_t;
+using WeiDataType      = int8_t;
+using BiasDataType     = int32_t;
+using AccDataType      = int32_t;
+using CShuffleDataType = AccDataType;
+using OutDataType      = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<ActivationOp>;
+
+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 BiasLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
+        NDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        AccDataType,
+        CShuffleDataType,
+        ck::Tuple<BiasDataType>,
+        OutDataType,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        1,           //
+        256,         // BlockSize
+        128,         // MPerBlock
+        256,         // NPerBlock
+        64,          // KPerBlock
+        16,          // AK1
+        16,          // 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
+        16,          // ABlockTransferSrcScalarPerVector
+        16,          // ABlockTransferDstScalarPerVector_AK1
+        1,           // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        16,          // BBlockTransferSrcScalarPerVector
+        16,          // BBlockTransferDstScalarPerVector_BK1
+        1,           // BBlockLdsExtraN
+        1,
+        1,
+        S<1, 64, 1, 4>,
+        8>;
+
+#include "run_conv2d_fwd_bias_perlayer_quantization_example.inc"
+
+int main()
+{
+    float requant_scale       = 0.5f;
+    const auto out_element_op = OutElementOp{requant_scale, ActivationOp{}};
+    run_conv2d_fwd_bias_perlayer_quantization_example(out_element_op);
+}

example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perchannel_quantization_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+
+using InDataType           = int8_t;
+using WeiDataType          = int8_t;
+using RequantScaleDataType = float;
+using AccDataType          = int32_t;
+using CShuffleDataType     = AccDataType;
+using OutDataType          = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Activation_Mul2_Clamp<ActivationOp>;
+
+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 RequantScaleLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
+        NDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<RequantScaleLayout>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        AccDataType,
+        CShuffleDataType,
+        ck::Tuple<RequantScaleDataType>,
+        OutDataType,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        1,           //
+        256,         // BlockSize
+        128,         // MPerBlock
+        256,         // NPerBlock
+        64,          // KPerBlock
+        16,          // AK1
+        16,          // 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
+        16,          // ABlockTransferSrcScalarPerVector
+        16,          // ABlockTransferDstScalarPerVector_AK1
+        1,           // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        16,          // BBlockTransferSrcScalarPerVector
+        16,          // BBlockTransferDstScalarPerVector_BK1
+        1,           // BBlockLdsExtraN
+        1,
+        1,
+        S<1, 64, 1, 4>,
+        8>;
+
+#include "run_conv2d_fwd_perchannel_quantization_example.inc"
+
+int main()
+{
+    const auto out_element_op = OutElementOp{ActivationOp{}};
+    run_conv2d_fwd_perchannel_quantization_example(out_element_op);
+}

example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+
+using InDataType       = int8_t;
+using WeiDataType      = int8_t;
+using AccDataType      = int32_t;
+using CShuffleDataType = AccDataType;
+using OutDataType      = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = PassThrough;
+using OutElementOp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
+
+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::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
+        NDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        AccDataType,
+        CShuffleDataType,
+        ck::Tuple<>,
+        OutDataType,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        1,           //
+        256,         // BlockSize
+        128,         // MPerBlock
+        256,         // NPerBlock
+        64,          // KPerBlock
+        16,          // AK1
+        16,          // 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
+        16,          // ABlockTransferSrcScalarPerVector
+        16,          // ABlockTransferDstScalarPerVector_AK1
+        1,           // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        16,          // BBlockTransferSrcScalarPerVector
+        16,          // BBlockTransferDstScalarPerVector_BK1
+        1,           // BBlockLdsExtraN
+        1,
+        1,
+        S<1, 64, 1, 4>,
+        16>;
+
+#include "run_conv2d_fwd_perlayer_quantization_example.inc"
+
+int main()
+{
+    float requant_scale       = 0.5f;
+    const auto out_element_op = OutElementOp{requant_scale, ActivationOp{}};
+    run_conv2d_fwd_perlayer_quantization_example(out_element_op);
+}

example/44_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp → example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/algorithm.hpp"
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/utility/convolution_parameter.hpp"
-#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
-
-using InDataType           = int8_t;
-using WeiDataType          = int8_t;
-using BiasDataType         = int32_t;
-using RequantScaleDataType = float;
-using AccDataType          = int32_t;
-using CShuffleDataType     = int32_t;
-using OutDataType          = int8_t;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
-using InElementOp  = PassThrough;
-using WeiElementOp = PassThrough;
-using ActivationOp = ck::tensor_operation::element_wise::Relu;
-using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<ActivationOp>;
-
-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 BiasLayout,
-          typename RequantScaleLayout,
-          typename OutLayout>
-using DeviceGroupedConvNDFwdInstance =
-    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
-        NDimSpatial,
-        InLayout,
-        WeiLayout,
-        ck::Tuple<BiasLayout, RequantScaleLayout>,
-        OutLayout,
-        InDataType,
-        WeiDataType,
-        AccDataType,
-        CShuffleDataType,
-        ck::Tuple<BiasDataType, RequantScaleDataType>,
-        OutDataType,
-        InElementOp,
-        WeiElementOp,
-        OutElementOp,
-        ConvSpec,    // ConvForwardSpecialization
-        GemmSpec,    // GemmSpecialization
-        1,           //
-        256,         // BlockSize
-        128,         // MPerBlock
-        256,         // NPerBlock
-        64,          // KPerBlock
-        16,          // AK1
-        16,          // 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
-        16,          // ABlockTransferSrcScalarPerVector
-        16,          // ABlockTransferDstScalarPerVector_AK1
-        1,           // ABlockLdsExtraM
-        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
-        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
-        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
-        2,           // BBlockTransferSrcVectorDim
-        16,          // BBlockTransferSrcScalarPerVector
-        16,          // BBlockTransferDstScalarPerVector_BK1
-        1,           // BBlockLdsExtraN
-        1,
-        1,
-        S<1, 64, 1, 4>,
-        8>;
-
+#pragma once
 template <ck::index_t NDimSpatial,
           typename InDataType,
           typename WeiDataType,
@@ -219,12 +129,12 @@ bool run_grouped_conv_fwd(bool do_verification,
 
     if(do_verification)
     {
-        Tensor<CShuffleDataType> c_host(out_g_n_k_wos_desc);
+        Tensor<AccDataType> c_host(out_g_n_k_wos_desc);
 
         auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
                                                                      InDataType,
                                                                      WeiDataType,
-                                                                     CShuffleDataType,
+                                                                     AccDataType,
                                                                      InElementOp,
                                                                      WeiElementOp,
                                                                      PassThrough>();
@@ -257,7 +167,7 @@ bool run_grouped_conv_fwd(bool do_verification,
     return (pass ? 0 : 1);
 }
 
-int main()
+int run_conv2d_fwd_bias_perchannel_quantization_example(const OutElementOp& out_element_op)
 {
     bool do_verification           = true;
     bool time_kernel               = true;
@@ -268,7 +178,7 @@ int main()
         1,            // group
         4,            // batch
         64,           // output channels
-        32,           // input chanels
+        192,          // input chanels
         {3, 3},       // weight HW
         {71, 71},     // x HW
         {2, 2},       // strides
@@ -279,7 +189,6 @@ int main()
 
     const auto in_element_op  = InElementOp{};
     const auto wei_element_op = WeiElementOp{};
-    const auto out_element_op = OutElementOp{ActivationOp{}};
 
     using InLayout           = ck::tensor_layout::convolution::GNHWC;
     using WeiLayout          = ck::tensor_layout::convolution::GKYXC;
@@ -312,8 +221,6 @@ int main()
     const auto out_g_n_k_wos_desc =
         ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(conv_param);
 
-    std::cout << out_g_n_k_wos_desc << std::endl;
-
     using deviceOp = DeviceGroupedConvNDFwdInstance<ndim_spatial,
                                                     InLayout,
                                                     WeiLayout,

example/44_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp → example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/algorithm.hpp"
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/utility/convolution_parameter.hpp"
-#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
-
-using InDataType       = int8_t;
-using WeiDataType      = int8_t;
-using BiasDataType     = int32_t;
-using AccDataType      = int32_t;
-using CShuffleDataType = int32_t;
-using OutDataType      = int8_t;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
-using InElementOp  = PassThrough;
-using WeiElementOp = PassThrough;
-using ActivationOp = ck::tensor_operation::element_wise::Relu;
-using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<ActivationOp>;
-
-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 BiasLayout,
-          typename OutLayout>
-using DeviceGroupedConvNDFwdInstance =
-    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
-        NDimSpatial,
-        InLayout,
-        WeiLayout,
-        ck::Tuple<BiasLayout>,
-        OutLayout,
-        InDataType,
-        WeiDataType,
-        AccDataType,
-        CShuffleDataType,
-        ck::Tuple<BiasDataType>,
-        OutDataType,
-        InElementOp,
-        WeiElementOp,
-        OutElementOp,
-        ConvSpec,    // ConvForwardSpecialization
-        GemmSpec,    // GemmSpecialization
-        1,           //
-        256,         // BlockSize
-        128,         // MPerBlock
-        256,         // NPerBlock
-        64,          // KPerBlock
-        16,          // AK1
-        16,          // 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
-        16,          // ABlockTransferSrcScalarPerVector
-        16,          // ABlockTransferDstScalarPerVector_AK1
-        1,           // ABlockLdsExtraM
-        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
-        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
-        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
-        2,           // BBlockTransferSrcVectorDim
-        16,          // BBlockTransferSrcScalarPerVector
-        16,          // BBlockTransferDstScalarPerVector_BK1
-        1,           // BBlockLdsExtraN
-        1,
-        1,
-        S<1, 64, 1, 4>,
-        8>;
+#pragma once
 
 template <ck::index_t NDimSpatial,
           typename InDataType,
@@ -205,12 +118,12 @@ bool run_grouped_conv_fwd(bool do_verification,
 
     if(do_verification)
     {
-        Tensor<CShuffleDataType> c_host(out_g_n_k_wos_desc);
+        Tensor<AccDataType> c_host(out_g_n_k_wos_desc);
 
         auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
                                                                      InDataType,
                                                                      WeiDataType,
-                                                                     CShuffleDataType,
+                                                                     AccDataType,
                                                                      InElementOp,
                                                                      WeiElementOp,
                                                                      PassThrough>();
@@ -242,7 +155,7 @@ bool run_grouped_conv_fwd(bool do_verification,
     return (pass ? 0 : 1);
 }
 
-int main()
+int run_conv2d_fwd_bias_perlayer_quantization_example(const OutElementOp& out_element_op)
 {
     bool do_verification           = true;
     bool time_kernel               = true;
@@ -253,7 +166,7 @@ int main()
         1,            // group
         4,            // batch
         64,           // output channels
-        32,           // input chanels
+        192,          // input chanels
         {3, 3},       // weight HW
         {71, 71},     // x HW
         {2, 2},       // strides
@@ -264,7 +177,6 @@ int main()
 
     const auto in_element_op  = InElementOp{};
     const auto wei_element_op = WeiElementOp{};
-    const auto out_element_op = OutElementOp{0.5f, ActivationOp{}};
 
     using InLayout   = ck::tensor_layout::convolution::GNHWC;
     using WeiLayout  = ck::tensor_layout::convolution::GKYXC;
@@ -294,8 +206,6 @@ int main()
     const auto out_g_n_k_wos_desc =
         ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(conv_param);
 
-    std::cout << out_g_n_k_wos_desc << std::endl;
-
     return run_grouped_conv_fwd<
         ndim_spatial,
         InDataType,