device_grouped_conv_bwd_weight.hpp Source File

device_grouped_conv_bwd_weight.hpp Source File#

Composable Kernel: device_grouped_conv_bwd_weight.hpp Source File
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
#pragma once
 
#include <array>
 
#include "ck/tensor_operation/gpu/device/device_base.hpp"
 
namespace ck {
namespace tensor_operation {
namespace device {
 
#define DISABLE_SPLIT_K_AUTODEDUCE_FOR_ONE_STAGE_KERNELS 1
 
template <ck::index_t NDimSpatial,
          typename InLayout,
          typename WeiLayout,
          typename OutLayout,
          typename InDataType,
          typename WeiDataType,
          typename OutDataType,
          typename InElementwiseOperation,
          typename WeiElementwiseOperation,
          typename OutElementwiseOperation,
          typename ComputeTypeA = InDataType,
          typename ComputeTypeB = ComputeTypeA>
struct DeviceGroupedConvBwdWeight : public BaseOperator
{
    virtual std::unique_ptr<BaseArgument>
    MakeArgumentPointer(const void* p_in,
                        void* p_wei,
                        const void* p_out,
                        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
                        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
                        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
                        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
                        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
                        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
                        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
                        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
                        InElementwiseOperation in_element_op,
                        WeiElementwiseOperation wei_element_op,
                        OutElementwiseOperation out_element_op,
                        ck::index_t split_k) = 0;
 
    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
};
 
} // namespace device
} // namespace tensor_operation
} // namespace ck