Skip to content

GitLab

"cacheflow/git@developer.sourcefind.cn:SIYIXNI/vllm.git" did not exist on "cfae35b861c5fc0c9f3689f99c7aba2e4501beb8"
Commit cc8df39e authored by carlushuang's avatar carlushuang
Browse files

Merge remote-tracking branch 'origin/develop' into cpu_avx2

parents 0b9fe840 98e1e2d0
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 1875 additions and 15 deletions
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp 0 → 100644
View file @ cc8df39e
#include <stdlib.h>
#include "config.hpp"
#include "device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_instance {
using DataType = int8_t;
using AccType = int32_t;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances =
std::tuple<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
#if 1
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
#endif
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 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<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
//##############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 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<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<DeviceConvBwdDataPtr<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 device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp 0 → 100644
View file @ cc8df39e
#include <stdlib.h>
#include "config.hpp"
#include "device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_instance {
using BF16 = ushort;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// 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<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp 0 → 100644
View file @ cc8df39e
#include <stdlib.h>
#include "config.hpp"
#include "device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_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 ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// 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<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
#if 1
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
#endif
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp 0 → 100644
View file @ cc8df39e
#include <stdlib.h>
#include "config.hpp"
#include "device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_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 ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// 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<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp 0 → 100644
View file @ cc8df39e
#include <stdlib.h>
#include "config.hpp"
#include "device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_instance {
using DataType = int8_t;
using AccType = int32_t;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// 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<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
#if 1
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
#endif
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 2, 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
//##############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
#if 1
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
#endif
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 2, 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_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp 0 → 100644
View file @ cc8df39e
#include <stdlib.h>
#include "config.hpp"
#include "device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_instance {
using BF16 = ushort;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances =
std::tuple<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_bf16_instances =
std::tuple<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | ./ | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances{});
add_device_operation_instances(
instances, device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_bf16_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp 0 → 100644
View file @ cc8df39e
#include <stdlib.h>
#include "config.hpp"
#include "device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_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 ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances =
std::tuple<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
#if 1
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
#endif
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>
// clang-format on
>;
using device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f16_instances =
std::tuple<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances{});
add_device_operation_instances(
instances, device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f16_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp 0 → 100644
View file @ cc8df39e
#include <stdlib.h>
#include "config.hpp"
#include "device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_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 ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances =
std::tuple<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f32_instances =
std::tuple<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances{});
add_device_operation_instances(
instances, device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f32_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp 0 → 100644
View file @ cc8df39e
#include <stdlib.h>
#include "config.hpp"
#include "device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_instance {
using DataType = int8_t;
using AccType = int32_t;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances =
std::tuple<
// clang-format off
//#############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
#if 1
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
#endif
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 3, 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>
// clang-format on
>;
using device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_int8_instances =
std::tuple<
// clang-format off
//##############################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| 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| Dim| 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|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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>,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 3, 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_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances{});
add_device_operation_instances(
instances, device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_int8_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
profiler/CMakeLists.txt
View file @ cc8df39e
...@@ -32,7 +32,7 @@ set(PROFILER_SOURCE ...@@ -32,7 +32,7 @@ set(PROFILER_SOURCE
src/profile_conv_fwd_bias_relu.cpp src/profile_conv_fwd_bias_relu.cpp
src/profile_conv_fwd_bias_relu_add.cpp src/profile_conv_fwd_bias_relu_add.cpp
src/profile_conv_fwd_bias_relu_atomic_add.cpp src/profile_conv_fwd_bias_relu_atomic_add.cpp
src/profile_conv_bwd_data.cpp src/profile_convnd_bwd_data.cpp
src/profile_reduce.cpp src/profile_reduce.cpp
src/profile_grouped_gemm.cpp src/profile_grouped_gemm.cpp
) )
...@@ -50,7 +50,7 @@ target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_instance) ...@@ -50,7 +50,7 @@ target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_instance)
target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_instance) target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_instance)
target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_add_instance) target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_add_instance)
target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_atomic_add_instance) target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_atomic_add_instance)
target_link_libraries(ckProfiler PRIVATE device_conv2d_bwd_data_instance) target_link_libraries(ckProfiler PRIVATE device_convnd_bwd_data_instance)
target_link_libraries(ckProfiler PRIVATE device_reduce_instance) target_link_libraries(ckProfiler PRIVATE device_reduce_instance)
target_link_libraries(ckProfiler PRIVATE device_reduce_instance) target_link_libraries(ckProfiler PRIVATE device_reduce_instance)
target_link_libraries(ckProfiler PRIVATE device_grouped_gemm_instance) target_link_libraries(ckProfiler PRIVATE device_grouped_gemm_instance)
profiler/include/profile_conv_bwd_data_impl.hpp
View file @ cc8df39e
...@@ -42,6 +42,7 @@ template <int NDimSpatial, ...@@ -42,6 +42,7 @@ template <int NDimSpatial,
typename InDataType, typename InDataType,
typename WeiDataType, typename WeiDataType,
typename OutDataType, typename OutDataType,
typename AccDataType,
typename InLayout, typename InLayout,
typename WeiLayout, typename WeiLayout,
typename OutLayout> typename OutLayout>
...@@ -123,6 +124,7 @@ void profile_conv_bwd_data_impl(int do_verification, ...@@ -123,6 +124,7 @@ void profile_conv_bwd_data_impl(int do_verification,
ck::tensor_operation::host::ReferenceConvBwdData<InDataType, ck::tensor_operation::host::ReferenceConvBwdData<InDataType,
WeiDataType, WeiDataType,
OutDataType, OutDataType,
AccDataType,
InElementOp, InElementOp,
WeiElementOp, WeiElementOp,
OutElementOp>; OutElementOp>;
......
profiler/include/profile_convnd_bwd_data_impl.hpp 0 → 100644
View file @ cc8df39e
This diff is collapsed.
profiler/src/profile_conv_bwd_data.cpp
View file @ cc8df39e
...@@ -89,6 +89,7 @@ int profile_conv_bwd_data(int argc, char* argv[]) ...@@ -89,6 +89,7 @@ int profile_conv_bwd_data(int argc, char* argv[])
float, float,
float, float,
float, float,
float,
ck::tensor_layout::convolution::NHWC, ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC, ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK>( ck::tensor_layout::convolution::NHWK>(
...@@ -114,6 +115,7 @@ int profile_conv_bwd_data(int argc, char* argv[]) ...@@ -114,6 +115,7 @@ int profile_conv_bwd_data(int argc, char* argv[])
ck::half_t, ck::half_t,
ck::half_t, ck::half_t,
ck::half_t, ck::half_t,
float,
ck::tensor_layout::convolution::NHWC, ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC, ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK>( ck::tensor_layout::convolution::NHWK>(
...@@ -139,6 +141,7 @@ int profile_conv_bwd_data(int argc, char* argv[]) ...@@ -139,6 +141,7 @@ int profile_conv_bwd_data(int argc, char* argv[])
uint16_t, uint16_t,
uint16_t, uint16_t,
uint16_t, uint16_t,
float,
ck::tensor_layout::convolution::NHWC, ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC, ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK>( ck::tensor_layout::convolution::NHWK>(
...@@ -164,6 +167,7 @@ int profile_conv_bwd_data(int argc, char* argv[]) ...@@ -164,6 +167,7 @@ int profile_conv_bwd_data(int argc, char* argv[])
int8_t, int8_t,
int8_t, int8_t,
int8_t, int8_t,
int32_t,
ck::tensor_layout::convolution::NHWC, ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC, ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK>( ck::tensor_layout::convolution::NHWK>(
......
profiler/src/profile_convnd_bwd_data.cpp 0 → 100644
View file @ cc8df39e
This diff is collapsed.
profiler/src/profiler.cpp
View file @ cc8df39e
...@@ -15,7 +15,7 @@ int profile_conv_fwd(int, char*[]); ...@@ -15,7 +15,7 @@ int profile_conv_fwd(int, char*[]);
int profile_conv_fwd_bias_relu(int, char*[]); int profile_conv_fwd_bias_relu(int, char*[]);
int profile_conv_fwd_bias_relu_add(int, char*[]); int profile_conv_fwd_bias_relu_add(int, char*[]);
int profile_conv_fwd_bias_relu_atomic_add(int, char*[]); int profile_conv_fwd_bias_relu_atomic_add(int, char*[]);
int profile_conv_bwd_data(int, char*[]); int profile_convnd_bwd_data(int, char*[], int);
int profile_reduce(int, char*[]); int profile_reduce(int, char*[]);
int main(int argc, char* argv[]) int main(int argc, char* argv[])
...@@ -64,9 +64,17 @@ int main(int argc, char* argv[]) ...@@ -64,9 +64,17 @@ int main(int argc, char* argv[])
{ {
return profile_conv_fwd_bias_relu_atomic_add(argc, argv); return profile_conv_fwd_bias_relu_atomic_add(argc, argv);
} }
else if(strcmp(argv[1], "conv_bwd") == 0) else if(strcmp(argv[1], "conv1d_bwd_data") == 0)
{ {
return profile_conv_bwd_data(argc, argv); return profile_convnd_bwd_data(argc, argv, 1);
}
else if(strcmp(argv[1], "conv2d_bwd_data") == 0)
{
return profile_convnd_bwd_data(argc, argv, 2);
}
else if(strcmp(argv[1], "conv3d_bwd_data") == 0)
{
return profile_convnd_bwd_data(argc, argv, 3);
} }
else if(strcmp(argv[1], "reduce") == 0) else if(strcmp(argv[1], "reduce") == 0)
{ {
...@@ -85,8 +93,11 @@ int main(int argc, char* argv[]) ...@@ -85,8 +93,11 @@ int main(int argc, char* argv[])
" conv_fwd_bias_relu: ForwardConvolution+Bias+ReLU\n" " conv_fwd_bias_relu: ForwardConvolution+Bias+ReLU\n"
" conv_fwd_bias_relu_add: ForwardConvolution+Bias+ReLU+Add\n" " conv_fwd_bias_relu_add: ForwardConvolution+Bias+ReLU+Add\n"
" conv_fwd_bias_relu_atomic_add: ForwardConvolution+Bias+ReLU+AtomicAdd\n" " conv_fwd_bias_relu_atomic_add: ForwardConvolution+Bias+ReLU+AtomicAdd\n"
" conv_bwd: BackwardConvolution\n" " conv1d_bwd_data: BackwardConvolution data 1 dim\n"
" reduce: Reduce\n"); " conv2d_bwd_data: BackwardConvolution data 2 dim\n"
" conv3d_bwd_data: BackwardConvolution data 3 dim\n"
" grouped_gemm: Grouped Gemm\n"
" reduce: REDUCE\n");
// clang-format on // clang-format on
return 0; return 0;
......
test/CMakeLists.txt
View file @ cc8df39e
...@@ -42,6 +42,5 @@ add_subdirectory(gemm_reduce) ...@@ -42,6 +42,5 @@ add_subdirectory(gemm_reduce)
add_subdirectory(batched_gemm) add_subdirectory(batched_gemm)
add_subdirectory(grouped_gemm) add_subdirectory(grouped_gemm)
add_subdirectory(convnd_fwd) add_subdirectory(convnd_fwd)
add_subdirectory(conv2d_bwd_data)
add_subdirectory(reduce) add_subdirectory(reduce)
add_subdirectory(cpu_ukernel) add_subdirectory(cpu_ukernel)
test/conv2d_bwd_data/conv2d_bwd_data.cpp
View file @ cc8df39e
...@@ -121,15 +121,17 @@ int main(int argc, char* argv[]) ...@@ -121,15 +121,17 @@ int main(int argc, char* argv[])
exit(1); exit(1);
} }
auto Run = [&](auto input_type, auto wei_type, auto out_type) { auto Run = [&](auto input_type, auto wei_type, auto out_type, auto acc_type) {
using InDataType = decltype(input_type); using InDataType = decltype(input_type);
using WeiDataType = decltype(wei_type); using WeiDataType = decltype(wei_type);
using OutDataType = decltype(out_type); using OutDataType = decltype(out_type);
using AccDataType = decltype(acc_type);
using ReferenceConvBwdInstance = using ReferenceConvBwdInstance =
ck::tensor_operation::host::ReferenceConvBwdData<InDataType, ck::tensor_operation::host::ReferenceConvBwdData<InDataType,
WeiDataType, WeiDataType,
OutDataType, OutDataType,
AccDataType,
InElementOp, InElementOp,
WeiElementOp, WeiElementOp,
OutElementOp>; OutElementOp>;
...@@ -293,33 +295,33 @@ int main(int argc, char* argv[]) ...@@ -293,33 +295,33 @@ int main(int argc, char* argv[])
if(success) if(success)
{ {
std::cout << "test conv2d bwd : Pass" << std::endl; std::cout << "test conv2d bwd : Pass" << std::endl;
return 0;
} }
else else
{ {
std::cout << "test conv2d bwd: Fail " << std::endl; std::cout << "test conv2d bwd: Fail " << std::endl;
return -1;
} }
}; };
if(data_type == 0) if(data_type == 0)
{ {
Run(F32(), F32(), F32()); return Run(F32(), F32(), F32(), F32());
} }
else if(data_type == 1) else if(data_type == 1)
{ {
Run(F16(), F16(), F16()); return Run(F16(), F16(), F16(), F32());
} }
else if(data_type == 2) else if(data_type == 2)
{ {
Run(BF16(), BF16(), BF16()); return Run(BF16(), BF16(), BF16(), F32());
} }
else if(data_type == 3) else if(data_type == 3)
{ {
Run(INT8(), INT8(), INT8()); return Run(INT8(), INT8(), INT8(), int());
} }
else else
{ {
return 1; return 1;
} }
return 0;
} }
test/convnd_bwd_data/CMakeLists.txt 0 → 100644
View file @ cc8df39e
include_directories(BEFORE
${PROJECT_SOURCE_DIR}/profiler/include
${PROJECT_SOURCE_DIR}/external/include/half
)
add_test_executable(test_convnd_bwd_data convnd_bwd_data.cpp)
target_link_libraries(test_convnd_bwd_data PRIVATE host_tensor)
target_link_libraries(test_convnd_bwd_data PRIVATE device_convnd_bwd_data_instance)
test/convnd_bwd_data/convnd_bwd_data.cpp 0 → 100644
View file @ cc8df39e
This diff is collapsed.
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