Commit aea62819 authored by Chaitanya Inumella's avatar Chaitanya Inumella
Browse files

Rebase branch 'develop' of...

Rebase branch 'develop' of https://github.com/ROCmSoftwarePlatform/composable_kernel into contraction_hipTENSOR
parents 75af5450 75ab874e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using BF16 = bhalf_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NWC = ck::tensor_layout::convolution::NWC;
using KXC = ck::tensor_layout::convolution::KXC;
using NWK = ck::tensor_layout::convolution::NWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
using device_conv1d_bwd_data_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(
std::vector<std::unique_ptr<DeviceConvBwdData<1,
NWC,
KXC,
NWK,
BF16,
BF16,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances{});
add_device_operation_instances(
instances, device_conv1d_bwd_data_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NWC = ck::tensor_layout::convolution::NWC;
using KXC = ck::tensor_layout::convolution::KXC;
using NWK = ck::tensor_layout::convolution::NWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
using device_conv1d_bwd_data_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances(
std::vector<std::unique_ptr<
DeviceConvBwdData<1, NWC, KXC, NWK, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances,
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances{});
add_device_operation_instances(
instances, device_conv1d_bwd_data_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NWC = ck::tensor_layout::convolution::NWC;
using KXC = ck::tensor_layout::convolution::KXC;
using NWK = ck::tensor_layout::convolution::NWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>
// clang-format on
>;
using device_conv1d_bwd_data_xdl_nwc_kxc_nwk_1x1_s1_p0_f32_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>
// clang-format on
>;
void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(
std::vector<std::unique_ptr<
DeviceConvBwdData<1, NWC, KXC, NWK, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances,
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances{});
add_device_operation_instances(
instances, device_conv1d_bwd_data_xdl_nwc_kxc_nwk_1x1_s1_p0_f32_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NWC = ck::tensor_layout::convolution::NWC;
using KXC = ck::tensor_layout::convolution::KXC;
using NWK = ck::tensor_layout::convolution::NWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>
// clang-format on
>;
using device_conv1d_bwd_data_xdl_nwc_kxc_nwk_1x1_s1_p0_int8_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 1, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>
// clang-format on
>;
void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(
std::vector<std::unique_ptr<DeviceConvBwdData<1,
NWC,
KXC,
NWK,
int8_t,
int8_t,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances{});
add_device_operation_instances(
instances, device_conv1d_bwd_data_xdl_nwc_kxc_nwk_1x1_s1_p0_int8_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
#device_conv1d_bwd_weight_instance
set(DEVICE_CONV1D_BWD_WEIGHT_INSTANCE_SOURCE
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instance.cpp;
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instance.cpp;
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instance.cpp;
)
add_library(device_conv1d_bwd_weight_instance OBJECT ${DEVICE_CONV1D_BWD_WEIGHT_INSTANCE_SOURCE})
target_compile_features(device_conv1d_bwd_weight_instance PUBLIC)
set_target_properties(device_conv1d_bwd_weight_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
rocm_install(TARGETS device_conv1d_bwd_weight_instance)
clang_tidy_check(device_conv1d_bwd_weight_instance)
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using BF16 = bhalf_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NWC = ck::tensor_layout::convolution::NWC;
using KXC = ck::tensor_layout::convolution::KXC;
using NWK = ck::tensor_layout::convolution::NWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_bf16_f32_bf16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
using device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_f32_bf16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_f32_bf16_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<1,
NWC,
KXC,
NWK,
BF16,
F32,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_bf16_f32_bf16_instances{});
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_f32_bf16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NWC = ck::tensor_layout::convolution::NWC;
using KXC = ck::tensor_layout::convolution::KXC;
using NWK = ck::tensor_layout::convolution::NWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f16_default_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
using device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<1,
NWC,
KXC,
NWK,
F16,
F16,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f16_default_instances{});
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NWC = ck::tensor_layout::convolution::NWC;
using KXC = ck::tensor_layout::convolution::KXC;
using NWK = ck::tensor_layout::convolution::NWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f32_default_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
using device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f32_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<1,
NWC,
KXC,
NWK,
F32,
F32,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f32_default_instances{});
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f32_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
# device_conv1d_fwd_instance
set(DEVICE_CONV1D_FWD_INSTANCE_SOURCE
device_conv1d_fwd_xdl_nwc_kxc_nwk_bf16_instance.cpp;
device_conv1d_fwd_xdl_nwc_kxc_nwk_f16_instance.cpp;
device_conv1d_fwd_xdl_nwc_kxc_nwk_f32_instance.cpp;
device_conv1d_fwd_xdl_nwc_kxc_nwk_int8_instance.cpp;
)
add_library(device_conv1d_fwd_instance OBJECT ${DEVICE_CONV1D_FWD_INSTANCE_SOURCE})
# target_compile_features(device_conv1d_fwd_instance PUBLIC)
set_target_properties(device_conv1d_fwd_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
# install(TARGETS device_conv1d_fwd_instance LIBRARY DESTINATION lib)
clang_tidy_check(device_conv1d_fwd_instance)
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#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/device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F32 = float;
using BF16 = bhalf_t;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
static constexpr auto ConvFwd1x1P0 =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
static constexpr auto ConvFwd1x1S1P0 =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv1d_fwd_xdl_nwc_kxc_nwk_bf16_instances = std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
// clang-format on
>;
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_bf16_instances = std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
// clang-format on
>;
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_instances = std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
// clang-format on
>;
void add_device_conv1d_fwd_xdl_nwc_kxc_nwk_bf16_instances(
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances, device_conv1d_fwd_xdl_nwc_kxc_nwk_bf16_instances{});
add_device_operation_instances(instances,
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_bf16_instances{});
add_device_operation_instances(instances,
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#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/device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
static constexpr auto ConvFwd1x1P0 =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
static constexpr auto ConvFwd1x1S1P0 =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv1d_fwd_xdl_nwc_kxc_nwk_f16_instances = std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
// clang-format on
>;
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_f16_instances = std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
// clang-format on
>;
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances = std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
// clang-format on
>;
void add_device_conv1d_fwd_xdl_nwc_kxc_nwk_f16_instances(
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances, device_conv1d_fwd_xdl_nwc_kxc_nwk_f16_instances{});
add_device_operation_instances(instances,
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_f16_instances{});
add_device_operation_instances(instances,
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#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/device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
static constexpr auto ConvFwd1x1P0 =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
static constexpr auto ConvFwd1x1S1P0 =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
//------------------------------------------------------------------------------
// Conv1D
//------------------------------------------------------------------------------
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using device_conv1d_fwd_xdl_nwc_kxc_nwk_f32_instances = std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
// clang-format on
>;
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_f32_instances = std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
// clang-format on
>;
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_f32_instances = std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
// clang-format on
>;
void add_device_conv1d_fwd_xdl_nwc_kxc_nwk_f32_instances(
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances, device_conv1d_fwd_xdl_nwc_kxc_nwk_f32_instances{});
add_device_operation_instances(instances,
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_f32_instances{});
add_device_operation_instances(instances,
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_f32_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#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/device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
static constexpr auto ConvFwd1x1P0 =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
static constexpr auto ConvFwd1x1S1P0 =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv1d_fwd_xdl_nwc_kxc_nwk_int8_instances =
std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
// clang-format on
>;
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_int8_instances =
std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
// clang-format on
>;
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_int8_instances =
std::tuple<
// clang-format off
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
// clang-format on
>;
void add_device_conv1d_fwd_xdl_nwc_kxc_nwk_int8_instances(
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances, device_conv1d_fwd_xdl_nwc_kxc_nwk_int8_instances{});
add_device_operation_instances(instances,
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_int8_instances{});
add_device_operation_instances(instances,
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_int8_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
...@@ -5,8 +5,11 @@ ...@@ -5,8 +5,11 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp"
// FIXME: retire dedicated 2D version
#include "ck/tensor_operation/gpu/device/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp" #include "ck/tensor_operation/gpu/device/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck { namespace ck {
...@@ -14,13 +17,18 @@ namespace tensor_operation { ...@@ -14,13 +17,18 @@ namespace tensor_operation {
namespace device { namespace device {
namespace instance { namespace instance {
using BF16 = ck::bhalf_t; using BF16 = bhalf_t;
using F32 = float; using F32 = float;
template <ck::index_t... Is> template <ck::index_t... Is>
using S = ck::Sequence<Is...>; using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough; using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault = static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default; ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
...@@ -29,6 +37,52 @@ static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 = ...@@ -29,6 +37,52 @@ static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k] // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances = std::tuple< using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
// FIXME: retire dedicated 2D version
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances = std::tuple<
// clang-format off // clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| //####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar| //####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
...@@ -50,7 +104,8 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances = std::tuple< ...@@ -50,7 +104,8 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances = std::tuple<
// clang-format on // clang-format on
>; >;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances = // FIXME: retire dedicated 2D version
using device_conv_dedidecate_2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances =
std::tuple< std::tuple<
// clang-format off // clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| //####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
...@@ -74,12 +129,26 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances = ...@@ -74,12 +129,26 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances =
>; >;
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances( void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances) std::vector<std::unique_ptr<DeviceConvBwdData<2,
NHWC,
KYXC,
NHWK,
BF16,
BF16,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{ {
add_device_operation_instances(instances, add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances{}); device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances{});
add_device_operation_instances( add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances{}); instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances{});
add_device_operation_instances(
instances, device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances{});
add_device_operation_instances(
instances,
device_conv_dedidecate_2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances{});
} }
} // namespace instance } // namespace instance
......
...@@ -5,8 +5,11 @@ ...@@ -5,8 +5,11 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp"
// FIXME: retire dedicated 2D version
#include "ck/tensor_operation/gpu/device/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp" #include "ck/tensor_operation/gpu/device/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck { namespace ck {
...@@ -20,7 +23,12 @@ using F32 = float; ...@@ -20,7 +23,12 @@ using F32 = float;
template <ck::index_t... Is> template <ck::index_t... Is>
using S = ck::Sequence<Is...>; using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough; using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault = static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default; ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
...@@ -29,6 +37,52 @@ static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 = ...@@ -29,6 +37,52 @@ static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k] // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances = std::tuple< using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
// FIXME: retire dedicated 2D version
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances = std::tuple<
// clang-format off // clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| //####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar| //####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
...@@ -52,7 +106,8 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances = std::tuple< ...@@ -52,7 +106,8 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances = std::tuple<
// clang-format on // clang-format on
>; >;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances = // FIXME: retire dedicated 2D version
using device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances =
std::tuple< std::tuple<
// clang-format off // clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| //####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
...@@ -76,12 +131,25 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances = ...@@ -76,12 +131,25 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances =
>; >;
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances( void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances) std::vector<std::unique_ptr<DeviceConvBwdData<2,
NHWC,
KYXC,
NHWK,
F16,
F16,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{ {
add_device_operation_instances(instances, add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances{}); device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances{});
add_device_operation_instances( add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances{}); instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances{});
add_device_operation_instances(
instances, device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances{});
add_device_operation_instances(
instances, device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances{});
} }
} // namespace instance } // namespace instance
......
...@@ -5,8 +5,11 @@ ...@@ -5,8 +5,11 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp"
// FIXME: retire dedicated 2D version
#include "ck/tensor_operation/gpu/device/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp" #include "ck/tensor_operation/gpu/device/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck { namespace ck {
...@@ -19,7 +22,12 @@ using F32 = float; ...@@ -19,7 +22,12 @@ using F32 = float;
template <ck::index_t... Is> template <ck::index_t... Is>
using S = ck::Sequence<Is...>; using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough; using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault = static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default; ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
...@@ -28,6 +36,52 @@ static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 = ...@@ -28,6 +36,52 @@ static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k] // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances = std::tuple< using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>
// clang-format on
>;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>
// clang-format on
>;
// FIXME: retire dedicated 2D version
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances = std::tuple<
// clang-format off // clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| //####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar| //####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
...@@ -49,7 +103,8 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances = std::tuple< ...@@ -49,7 +103,8 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances = std::tuple<
// clang-format on // clang-format on
>; >;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances = // FIXME: retire dedicated 2D version
using device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances =
std::tuple< std::tuple<
// clang-format off // clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| //####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
...@@ -73,12 +128,25 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances = ...@@ -73,12 +128,25 @@ using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances =
>; >;
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances( void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances) std::vector<std::unique_ptr<DeviceConvBwdData<2,
NHWC,
KYXC,
NHWK,
F32,
F32,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{ {
add_device_operation_instances(instances, add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances{}); device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances{});
add_device_operation_instances( add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances{}); instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances{});
add_device_operation_instances(
instances, device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances{});
add_device_operation_instances(
instances, device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances{});
} }
} // namespace instance } // namespace instance
......
...@@ -5,8 +5,11 @@ ...@@ -5,8 +5,11 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp"
// FIXME: retire dedicated 2D version
#include "ck/tensor_operation/gpu/device/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp" #include "ck/tensor_operation/gpu/device/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck { namespace ck {
...@@ -14,13 +17,15 @@ namespace tensor_operation { ...@@ -14,13 +17,15 @@ namespace tensor_operation {
namespace device { namespace device {
namespace instance { namespace instance {
using DataType = int8_t;
using AccType = int32_t;
template <ck::index_t... Is> template <ck::index_t... Is>
using S = ck::Sequence<Is...>; using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough; using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault = static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default; ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
...@@ -30,56 +35,116 @@ static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 = ...@@ -30,56 +35,116 @@ static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k] // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances = std::tuple< using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances = std::tuple<
// clang-format off // clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| //##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar| //##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| //##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1> DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>
// clang-format on
>;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances = std::tuple<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvNdBwdDataNwcKxcNwk_Xdl< 2, int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>
// clang-format on // clang-format on
>; >;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances = // FIXME: retire dedicated 2D version
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances = std::tuple<
// clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>
// clang-format on
>;
// FIXME: retire dedicated 2D version
using device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances =
std::tuple< std::tuple<
// clang-format off // clang-format off
//#####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| //#####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//#####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar| //#####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| //#####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //#####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>, DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1> DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>
// clang-format on // clang-format on
>; >;
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances( void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances) std::vector<std::unique_ptr<DeviceConvBwdData<2,
NHWC,
KYXC,
NHWK,
int8_t,
int8_t,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{ {
add_device_operation_instances(instances, add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances{}); device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances{});
add_device_operation_instances( add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances{}); instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances{});
add_device_operation_instances(
instances, device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances{});
add_device_operation_instances(
instances, device_conv_dedicated_2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances{});
} }
} // namespace instance } // namespace instance
......
# device_conv2d_bwd_weight_instance #device_conv2d_bwd_weight_instance
set(DEVICE_CONV2D_BWD_WEIGHT_INSTANCE_SOURCE set(DEVICE_CONV2D_BWD_WEIGHT_INSTANCE_SOURCE
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instance.cpp; device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instance.cpp;
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instance.cpp; device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instance.cpp;
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp;
) )
add_library(device_conv2d_bwd_weight_instance OBJECT ${DEVICE_CONV2D_BWD_WEIGHT_INSTANCE_SOURCE}) add_library(device_conv2d_bwd_weight_instance OBJECT ${DEVICE_CONV2D_BWD_WEIGHT_INSTANCE_SOURCE})
target_compile_features(device_conv2d_bwd_weight_instance PUBLIC) target_compile_features(device_conv2d_bwd_weight_instance PUBLIC)
set_target_properties(device_conv2d_bwd_weight_instance PROPERTIES POSITION_INDEPENDENT_CODE ON) set_target_properties(device_conv2d_bwd_weight_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
rocm_install(TARGETS device_conv2d_bwd_weight_instance) rocm_install(TARGETS device_conv2d_bwd_weight_instance)
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using BF16 = bhalf_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_bf16_f32_bf16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
using device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_f32_bf16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
void add_device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_f32_bf16_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<2,
NHWC,
KYXC,
NHWK,
BF16,
F32,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances, device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_bf16_f32_bf16_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_f32_bf16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
...@@ -5,8 +5,11 @@ ...@@ -5,8 +5,11 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp"
// TODO: retire dedicated 2d version
#include "ck/tensor_operation/gpu/device/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp" #include "ck/tensor_operation/gpu/device/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck { namespace ck {
...@@ -20,36 +23,105 @@ using F32 = float; ...@@ -20,36 +23,105 @@ using F32 = float;
template <ck::index_t... Is> template <ck::index_t... Is>
using S = ck::Sequence<Is...>; using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough; using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_default_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
using device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
// TODO: retire dedicated 2d version
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k] // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances = std::tuple< using device_conv_dedicated_2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances = std::tuple<
// clang-format off // clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer| //#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector| //#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl| //#################################################################################| | | | | Operation| Operation| Operation| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| | //#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>, DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8> DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on // clang-format on
>; >;
void add_device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances( void add_device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances(
std::vector<DeviceConvBwdWeightPtr<PassThrough, PassThrough, PassThrough>>& instances) std::vector<std::unique_ptr<DeviceConvBwdWeight<2,
NHWC,
KYXC,
NHWK,
F16,
F16,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{ {
add_device_operation_instances(instances, add_device_operation_instances(
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances{}); instances, device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_default_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances{});
add_device_operation_instances(
instances, device_conv_dedicated_2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances{});
} }
} // namespace instance } // namespace instance
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment