Skip to content

GitLab

Commit 6d9a07d7 authored by Jun Liu's avatar Jun Liu
Browse files

Merge branch 'develop' into amd-develop

parents b30d416c a776978c
Hide whitespace changes
Inline Side-by-side
Showing with 755 additions and 11 deletions
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data_bilinear/CMakeLists.txt 0 → 100644
View file @ 6d9a07d7
set(GROUPED_CONV3D_BWD_DATA_BILINEAR
xdl/device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp)
add_instance_library(device_grouped_conv3d_bwd_data_bilinear_instance ${GROUPED_CONV3D_BWD_DATA_BILINEAR})
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data_bilinear/xdl/device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_bilinear_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
// g, k]
void add_device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
NDHWGK,
GKZYXC,
Tuple<NDHWGC>,
NDHWGC,
BF16,
BF16,
Tuple<BF16>,
BF16,
PassThrough,
PassThrough,
Bilinear>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bilinear_bf16_instances<3,
NDHWGK,
GKZYXC,
Tuple<NDHWGC>,
NDHWGC,
ConvBwdDataDefault>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(instances,
device_grouped_conv_bwd_data_xdl_bilinear_bf16_instances<
3,
NDHWGK,
GKZYXC,
Tuple<NDHWGC>,
NDHWGC,
ConvBwdDataFilter1x1Stride1Pad0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data_bilinear/xdl/device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_bilinear_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
// g, k]
void add_device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgk_gkzyxc_ndhwgc_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
NDHWGK,
GKZYXC,
Tuple<NDHWGC>,
NDHWGC,
F16,
F16,
Tuple<F16>,
F16,
PassThrough,
PassThrough,
Bilinear>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bilinear_f16_instances<3,
NDHWGK,
GKZYXC,
Tuple<NDHWGC>,
NDHWGC,
ConvBwdDataDefault>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bilinear_f16_instances<3,
NDHWGK,
GKZYXC,
Tuple<NDHWGC>,
NDHWGC,
ConvBwdDataFilter1x1Stride1Pad0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data_bilinear/xdl/device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_bilinear_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
// g, k]
void add_device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgk_gkzyxc_ndhwgc_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
NDHWGK,
GKZYXC,
Tuple<NDHWGC>,
NDHWGC,
F32,
F32,
Tuple<F32>,
F32,
PassThrough,
PassThrough,
Bilinear>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bilinear_f32_instances<3,
NDHWGK,
GKZYXC,
Tuple<NDHWGC>,
NDHWGC,
ConvBwdDataDefault>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bilinear_f32_instances<3,
NDHWGK,
GKZYXC,
Tuple<NDHWGC>,
NDHWGC,
ConvBwdDataFilter1x1Stride1Pad0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_bilinear/CMakeLists.txt 0 → 100644
View file @ 6d9a07d7
set(GROUPED_CONV3D_FWD_BILINEAR
xdl/device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp)
add_instance_library(device_grouped_conv3d_fwd_bilinear_instance ${GROUPED_CONV3D_FWD_BILINEAR})
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_bilinear/xdl/device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_bilinear_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<NDHWGK>,
NDHWGK,
BF16,
BF16,
ck::Tuple<BF16>,
BF16,
PassThrough,
PassThrough,
Bilinear>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_bf16_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_bf16_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_bf16_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_bilinear/xdl/device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_bilinear_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<NDHWGK>,
NDHWGK,
F16,
F16,
ck::Tuple<F16>,
F16,
PassThrough,
PassThrough,
Bilinear>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_f16_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_f16_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_f16_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_bilinear/xdl/device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_bilinear_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<NDHWGK>,
NDHWGK,
F32,
F32,
ck::Tuple<F32>,
F32,
PassThrough,
PassThrough,
Bilinear>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_f32_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_f32_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_f32_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_bilinear/xdl/device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_bilinear_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_bilinear_ndhwgc_gkzyxc_ndhwgk_int8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<NDHWGK>,
NDHWGK,
int8_t,
int8_t,
ck::Tuple<int8_t>,
int8_t,
PassThrough,
PassThrough,
Bilinear>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_int8_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_int8_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bilinear_int8_instances<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
View file @ 6d9a07d7
......@@ -5,6 +5,8 @@ list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16
device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_kn_mn_instance.cpp
device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_nk_mn_instance.cpp
device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp
device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp)
device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp
device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_kn_mn_instance.cpp
device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_nk_mn_instance.cpp)
add_instance_library(device_grouped_gemm_fixed_nk_instance ${GROUPED_GEMM_FIXED_NK_INSTANCES})
library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_kn_mn_instance.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using I8 = int8_t;
using BF16 = ck::bhalf_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using DsDataType = ck::Tuple<>;
using DsLayout = ck::Tuple<>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
using device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_kn_mn_irregular_tile_instances = std::tuple<
// clang-format off
//############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 64, 32, 8, 2, 32, 32, 2, 1, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 16,16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 2, 32, 32, 2, 2, S< 1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S< 1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 2, 32, 32, 2, 2, S< 1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S< 1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8>
// clang-format on
>;
void add_device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_kn_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
Row,
DsLayout,
Row,
BF16,
I8,
DsDataType,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_kn_mn_irregular_tile_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_nk_mn_instance.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using I8 = int8_t;
using BF16 = ck::bhalf_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using DsDataType = ck::Tuple<>;
using DsLayout = ck::Tuple<>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
using device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_nk_mn_irregular_tile_instances = std::tuple<
// clang-format off
//############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 256, 64, 8, 8, 32, 32, 2, 4, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 64, 8, 8, 32, 32, 2, 2, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 64, 64, 8, 8, 32, 32, 2, 1, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 128, 64, 8, 8, 32, 32, 1, 2, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 64, 8, 8, 32, 32, 4, 2, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 64, 8, 8, 32, 32, 2, 2, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 64, 8, 8, 32, 32, 2, 2, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 32, 64, 8, 8, 32, 32, 2, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 32, 128, 64, 8, 8, 32, 32, 1, 2, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 32, 256, 64, 8, 8, 32, 32, 1, 4, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 64, 8, 8, 32, 32, 2, 2, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 64, 8, 8, 32, 32, 2, 1, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 32, 64, 64, 8, 8, 32, 32, 1, 2, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
void add_device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_nk_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
Col,
DsLayout,
Row,
BF16,
I8,
DsDataType,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_nk_mn_irregular_tile_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/permute_scale/CMakeLists.txt
View file @ 6d9a07d7
add_instance_library(device_permute_scale_instance
device_permute_scale_instances.cpp)
device_permute_scale_1d_instances.cpp
device_permute_scale_2d_instances.cpp
device_permute_scale_3d_instances.cpp
device_permute_scale_4d_instances.cpp
device_permute_scale_5d_instances.cpp
device_permute_scale_6d_instances.cpp)
library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_1d_instances.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_permute_scale_1d_f16_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 1>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f16_instances<1>{});
}
void add_device_permute_scale_1d_f32_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 1>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f32_instances<1>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_2d_instances.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_permute_scale_2d_f16_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 2>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f16_instances<2>{});
}
void add_device_permute_scale_2d_f32_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 2>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f32_instances<2>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_3d_instances.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_permute_scale_3d_f16_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 3>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f16_instances<3>{});
}
void add_device_permute_scale_3d_f32_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 3>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f32_instances<3>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_4d_instances.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_permute_scale_4d_f16_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 4>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f16_instances<4>{});
}
void add_device_permute_scale_4d_f32_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 4>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f32_instances<4>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_5d_instances.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_permute_scale_5d_f16_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 5>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f16_instances<5>{});
}
void add_device_permute_scale_5d_f32_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 5>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f32_instances<5>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_6d_instances.cpp 0 → 100644
View file @ 6d9a07d7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_permute_scale_6d_f16_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 6>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f16_instances<6>{});
}
void add_device_permute_scale_6d_f32_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 6>>>& instances)
{
add_device_operation_instances(instances, device_permute_scale_f32_instances<6>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
profiler/README.md
View file @ 6d9a07d7
......@@ -37,9 +37,9 @@ Best Perf: 1.1933 ms, 107.977 TFlops, 79.0848 GB/s
################ op datatype in_layout wei_layout out_layout verify init log repeat N__ K___ C___ Y X Hi__ Wi__ Strides Dilations LeftPads RightPads
./bin/ckProfiler conv2d_fwd 1 1 1 1 1 1 0 5 128 256 192 3 3 71 71 2 2 1 1 1 1 1 1
```
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
```
```bash
in_n_c_hi_wi: dim 4, lengths {128, 192, 71, 71}, strides {967872, 1, 13632, 192}
wei_k_c_y_x: dim 4, lengths {256, 192, 3, 3}, strides {1728, 1, 576, 192}
out_n_k_ho_wo: dim 4, lengths {128, 256, 36, 36}, strides {331776, 1, 9216, 256}
......@@ -104,6 +104,7 @@ arg.b_grid_desc_k0_n0_n1_k1_{2048, 4096, 2}
arg.e_grid_desc_m_n_{ 4096, 4096}
....
Best Perf: 58.0306 ms, 37.8942 TFlops, 27.7545 GB/s
```
## Profile grouped convolution backward data kernels
```bash
# arg1: tensor operation (grouped_conv_bwd_data: Grouped Convolution Backward Data)
......@@ -129,10 +130,11 @@ Best Perf: 58.0306 ms, 37.8942 TFlops, 27.7545 GB/s
################ op datatype layout verify init log time Ndims G N K C Y X Hi Wi Sy Sx Dy Dx LeftPy LeftPx RightPy RightPx
./bin/ckProfiler grouped_conv_bwd_data 1 0 1 1 0 1 2 32 4 192 192 3 3 28 28 1 1 1 1 1 1 1 1
```
```
Result (MI100, FP16, GNHWC_GKYXC_GNHWK)
```
```bash
out: dim 5, lengths {32, 4, 192, 28, 28}, strides {602112, 150528, 1, 5376, 192}
wei: dim 5, lengths {32, 192, 192, 3, 3}, strides {331776, 1728, 1, 576, 192}
in: dim 5, lengths {32, 4, 192, 28, 28}, strides {602112, 150528, 1, 5376, 192}
......@@ -173,10 +175,11 @@ GB/s: 127.947
################ op datatype layout verify init log time Ndims G N K C Y X Hi Wi Sy Sx Dy Dx LeftPy LeftPx RightPy RightPx SplitK
./bin/ckProfiler grouped_conv_bwd_weight 1 1 0 1 0 1 2 32 256 256 512 3 3 28 28 1 1 1 1 1 0 0 0 1
```
```
Result (MI100, FP16, GNHWC_GKYXC_GNHWK)
```
```bash
input: dim 5, lengths {32, 512, 1024, 28, 28}, strides {411041792, 802816, 1, 28672, 1024}
weight: dim 5, lengths {32, 512, 1024, 3, 3}, strides {4718592, 9216, 1, 3072, 1024}
output: dim 5, lengths {32, 512, 512, 26, 26}, strides {177209344, 346112, 1, 13312, 512}
......@@ -190,8 +193,9 @@ GB/s: 69.2301
Note: This kernel use atomic add, this will cause output buffer to be accumulated multiple times, causing verification failure. To work around it, do not use CK's own timer and do verification at the same time.
## Profile image to column/column to image kernels
```bash
# arg1: tensor operation (" OP_NAME ": " OP_DESC ")
# arg1: tensor operation ( conv_tensor_rearrange : Conv Tensor Rearrange )
# arg2: data type (0: Input fp32, Weight fp32, Output fp32
# 1: Input fp16, Weight fp16, Output fp16
# 2: Input bf16, Weight bf16, Output bf16
......@@ -216,10 +220,11 @@ Note: This kernel use atomic add, this will cause output buffer to be accumulate
################ op datatype layout verify init log time opType Ndims G N K C Y X Hi Wi Sy Sx Dy Dx LeftPy LeftPx RightPy RightPx
./bin/ckProfiler conv_tensor_rearrange 0 0 0 1 0 1 0 2 1 256 1 512 3 3 28 28 1 1 1 1 0 0 0 0
```
```
Result (MI210, FP32, NHWC)
```
```bash
input: dim 5, lengths {1, 256, 512, 28, 28}, strides {102760448, 401408, 1, 14336, 512}
output: dim 2, lengths {173056, 4608}, strides {4608, 1}
....
......@@ -229,3 +234,30 @@ avg_time: 3.12326
GB/s: 2042.59
```
Note: Column to image kernel adds to the output memory, this will cause output buffer to be accumulated multiple times, causing verification failure. To work around it, do not use CK's own timer and do verification at the same time.
## Profile Permute scale kernels
```bash
# arg1: tensor operation ( permute_scale : Permute Scale )
# arg2: data type (0: Input fp32, Output fp32
# 1: Input fp16, Output fp16
# arg4: verification (0: no, 1: yes)
# arg5: initialization (0: no init, 1: integer value, 2: decimal value)
# arg6: print tensor value (0: no; 1: yes)
# arg7: time kernel (0: no, 1: yes)
# from arg8: tensor lengths
# input strides
# output strides
################ op datatype verify init log time dim0 dim1 dim2 in_stride0 in_stride1 in_stride2 out_stride0 out_stride1 out_stride2
./bin/ckProfiler permute_scale 0 1 1 0 1 64 64 64 4096 64 1 1 64 4096
```
Result (MI100, FP32)
```bash
A: dim 3, lengths {64, 64, 64}, strides {4096, 64, 1}
B: dim 3, lengths {64, 64, 64}, strides {1, 64, 4096}
....
Best perf = 0.0146878 ms, 142.782 GB/s, DeviceElementwiseNormalizationImpl<3, 2>
```
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