Skip to content

GitLab

Commit bc641634 authored by Jun Liu's avatar Jun Liu
Browse files

Merge branch 'develop-tmp' into amd-develop

parents f30e5975 a3d9a2cd
Hide whitespace changes
Inline Side-by-side
Showing with 607 additions and 69 deletions
example/27_layernorm2d_fwd/layernorm2d_fwd_fp16.cpp 0 → 100644
View file @ bc641634
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using SaveMeanInvStdDataType = float;
using ComputeDataType = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
#define SAVE_MEAN_INV_STD
constexpr int Rank = 2;
constexpr int NumReduceDim = 1;
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationFwdImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
SaveMeanInvStdDataType,
PassThrough,
Rank,
NumReduceDim,
256, // BlockSize
8, // ClusterM
32, // ClusterK
1, // SliceM
8, // SliceK
1, // XYVectorDim (0=M, 1=K)
8, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
8, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
8, // BetaScalarPerVector
8, // YScalarPerVector
1>; // SaveMeanInvStdScalarPerVector
#include "run_layernorm_example.inc"
int main() { return run_layernorm2d_fwd_example<DeviceInstance>(); }
example/27_layernorm2d_fwd/layernorm2d_fwd_splitk_fp16.cpp 0 → 100644
View file @ bc641634
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using SaveMeanInvStdDataType = float;
using ComputeDataType = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
#define SAVE_MEAN_INV_STD
constexpr int Rank = 2;
constexpr int NumReduceDim = 1;
using DeviceInstance = ck::tensor_operation::device::DeviceNormalizationFwdSplitKImpl<
XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
SaveMeanInvStdDataType,
PassThrough,
Rank,
NumReduceDim,
256, // BlockSize
8, // ClusterM
32, // ClusterK
1, // SliceM
8, // SliceK
1, // XYVectorDim (0=M, 1=K)
8, // XScalarPerVector
1, // GammaVecDim (0=M, 1=K)
8, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
8, // BetaScalarPerVector
8, // YScalarPerVector
1>; // SaveMeanInvStdScalarPerVector
#include "run_layernorm_example.inc"
int main() { return run_layernorm2d_fwd_example<DeviceInstance>(); }
example/27_layernorm/run_layernorm_example.inc → example/27_layernorm2d_fwd/run_layernorm_example.inc
View file @ bc641634
...@@ -4,12 +4,12 @@ ...@@ -4,12 +4,12 @@
#pragma once #pragma once
template <typename DeviceInstance> template <typename DeviceInstance>
int run_groupnorm_example() int run_layernorm2d_fwd_example()
{ {
bool time_kernel = false; bool time_kernel = false;
ck::index_t M = 1024; ck::index_t M = 1024;
ck::index_t N = 1024; ck::index_t N = 1024;
Tensor<XDataType> x({M, N}); Tensor<XDataType> x({M, N});
Tensor<GammaDataType> gamma({N}); Tensor<GammaDataType> gamma({N});
......
example/42_groupnorm/CMakeLists.txt deleted 100644 → 0
View file @ f30e5975
add_example_executable(example_groupnorm_sigmoid_mul_fp16 groupnorm_sigmoid_mul_fp16.cpp)
add_example_executable(example_groupnorm_splitk_fp16 groupnorm_splitk_fp16.cpp)
add_example_executable(example_groupnorm_swish_fp16 groupnorm_swish_fp16.cpp)
example/42_groupnorm/groupnorm_swish_fp16.cpp deleted 100644 → 0
View file @ f30e5975
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using SaveMeanInvStdDataType = float;
using ComputeDataType = float;
using YElementOp = ck::tensor_operation::element_wise::Swish;
#define SAVE_MEAN_INV_STD
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
SaveMeanInvStdDataType,
YElementOp,
Rank,
NumReduceDim,
1024, // BlockSize
1, // ClusterM
1024, // ClusterK
1, // SliceM
32, // SliceK
1, // SrcVecDim (0=M, 1=K)
2, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
2, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
2, // BetaScalarPerVector
2, // YScalarPerVector
1>; // SaveMeanInvStdScalarPerVector
#include "run_groupnorm_example.inc"
int main(int argc, char* argv[]) { run_groupnorm_example(argc, argv); }
example/42_groupnorm_fwd/CMakeLists.txt 0 → 100644
View file @ bc641634
add_example_executable(example_groupnorm_fwd_sigmoid_mul_fp16 groupnorm_fwd_sigmoid_mul_fp16.cpp)
add_example_executable(example_groupnorm_fwd_splitk_fp16 groupnorm_fwd_splitk_fp16.cpp)
add_example_executable(example_groupnorm_fwd_swish_fp16 groupnorm_fwd_swish_fp16.cpp)
example/42_groupnorm/common.hpp → example/42_groupnorm_fwd/common.hpp
View file @ bc641634
...@@ -11,8 +11,8 @@ ...@@ -11,8 +11,8 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/utility/reduction_enums.hpp" #include "ck/utility/reduction_enums.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_normalization_fwd_impl.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_normalization_fwd_splitk_impl.hpp"
#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp" #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
#include "ck/library/utility/fill.hpp" #include "ck/library/utility/fill.hpp"
......
example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp → example/42_groupnorm_fwd/groupnorm_fwd_sigmoid_mul_fp16.cpp
View file @ bc641634
...@@ -37,29 +37,29 @@ struct YElementOp ...@@ -37,29 +37,29 @@ struct YElementOp
}; };
using DeviceInstance = using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationImpl<XDataType, ck::tensor_operation::device::DeviceNormalizationFwdImpl<XDataType,
GammaDataType, GammaDataType,
BetaDataType, BetaDataType,
ComputeDataType, ComputeDataType,
YDataType, YDataType,
SaveMeanInvStdDataType, SaveMeanInvStdDataType,
YElementOp, YElementOp,
Rank, Rank,
NumReduceDim, NumReduceDim,
1024, // BlockSize 1024, // BlockSize
1, // ClusterM 1, // ClusterM
1024, // ClusterK 1024, // ClusterK
1, // SliceM 1, // SliceM
32, // SliceK 32, // SliceK
1, // SrcVecDim (0=M, 1=K) 1, // SrcVecDim (0=M, 1=K)
2, // SrcScalarPerVector 2, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K) 1, // GammaVecDim (0=M, 1=K)
2, // GammaScalarPerVector 2, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K) 1, // BetaVecDim (0=M, 1=K)
2, // BetaScalarPerVector 2, // BetaScalarPerVector
2, // YScalarPerVector 2, // YScalarPerVector
1>; // SaveMeanInvStdScalarPerVector 1>; // SaveMeanInvStdScalarPerVector
#include "run_groupnorm_example.inc" #include "run_groupnorm_fwd_example.inc"
int main(int argc, char* argv[]) { run_groupnorm_example(argc, argv); } int main(int argc, char* argv[]) { run_groupnorm_fwd_example(argc, argv); }
example/42_groupnorm_fwd/groupnorm_fwd_splitk_fp16.cpp 0 → 100644
View file @ bc641634
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using SaveMeanInvStdDataType = float;
using ComputeDataType = float;
using YElementOp = ck::tensor_operation::element_wise::Swish;
#define SAVE_MEAN_INV_STD
using DeviceInstance = ck::tensor_operation::device::DeviceNormalizationFwdSplitKImpl<
XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
SaveMeanInvStdDataType,
YElementOp,
Rank,
NumReduceDim,
256, // BlockSize
1, // ClusterM
256, // ClusterK
1, // SliceM
16, // SliceK
1, // SrcVecDim (0=M, 1=K)
2, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
2, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
2, // BetaScalarPerVector
2, // YScalarPerVector
1>; // SaveMeanInvStdScalarPerVector
#include "run_groupnorm_fwd_example.inc"
int main(int argc, char* argv[]) { run_groupnorm_fwd_example(argc, argv); }
example/42_groupnorm_fwd/groupnorm_fwd_swish_fp16.cpp 0 → 100644
View file @ bc641634
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using SaveMeanInvStdDataType = float;
using ComputeDataType = float;
using YElementOp = ck::tensor_operation::element_wise::Swish;
#define SAVE_MEAN_INV_STD
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationFwdImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
SaveMeanInvStdDataType,
YElementOp,
Rank,
NumReduceDim,
1024, // BlockSize
1, // ClusterM
1024, // ClusterK
1, // SliceM
32, // SliceK
1, // SrcVecDim (0=M, 1=K)
2, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
2, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
2, // BetaScalarPerVector
2, // YScalarPerVector
1>; // SaveMeanInvStdScalarPerVector
#include "run_groupnorm_fwd_example.inc"
int main(int argc, char* argv[]) { run_groupnorm_fwd_example(argc, argv); }
example/42_groupnorm/run_groupnorm_example.inc → example/42_groupnorm_fwd/run_groupnorm_fwd_example.inc
View file @ bc641634
...@@ -3,7 +3,7 @@ ...@@ -3,7 +3,7 @@
#pragma once #pragma once
int run_groupnorm_example(int argc, char* argv[]) int run_groupnorm_fwd_example(int argc, char* argv[])
{ {
ck::index_t N = 32; ck::index_t N = 32;
ck::index_t H = 16; ck::index_t H = 16;
......
example/52_im2col_col2im/column_to_image_f32.cpp
View file @ bc641634
...@@ -20,7 +20,7 @@ using DeviceColToImgInstance = ck::tensor_operation::device::DeviceColumnToImage ...@@ -20,7 +20,7 @@ using DeviceColToImgInstance = ck::tensor_operation::device::DeviceColumnToImage
bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::ConvParam& conv_params) bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::ConvParam& conv_params)
{ {
const auto G = conv_params.G_;
const auto N = conv_params.N_; const auto N = conv_params.N_;
const auto C = conv_params.C_; const auto C = conv_params.C_;
...@@ -31,7 +31,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv ...@@ -31,7 +31,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv
C * ck::accumulate_n<ck::index_t>( C * ck::accumulate_n<ck::index_t>(
conv_params.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>()); conv_params.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
const auto in_desc = HostTensorDescriptor({NDoHoWo, CZYX}); const auto in_desc = HostTensorDescriptor({G, NDoHoWo, CZYX});
const auto out_desc = const auto out_desc =
ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(conv_params); ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(conv_params);
...@@ -39,7 +39,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv ...@@ -39,7 +39,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv
std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{}; std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
std::array<ck::index_t, NDimSpatial> output_spatial_lengths{}; std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
std::array<ck::index_t, NDimSpatial + 3> image_g_n_c_wis_strides{}; std::array<ck::index_t, NDimSpatial + 3> image_g_n_c_wis_strides{};
std::array<ck::index_t, 2> gemm_m_k_strides{}; std::array<ck::index_t, 3> gemm_g_m_k_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_strides{}; std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_dilations{}; std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
std::array<ck::index_t, NDimSpatial> input_left_pads{}; std::array<ck::index_t, NDimSpatial> input_left_pads{};
...@@ -50,7 +50,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv ...@@ -50,7 +50,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv
copy(conv_params.input_spatial_lengths_, input_spatial_lengths); copy(conv_params.input_spatial_lengths_, input_spatial_lengths);
copy(conv_params.filter_spatial_lengths_, filter_spatial_lengths); copy(conv_params.filter_spatial_lengths_, filter_spatial_lengths);
copy(conv_params.output_spatial_lengths_, output_spatial_lengths); copy(conv_params.output_spatial_lengths_, output_spatial_lengths);
copy(in_desc.GetStrides(), gemm_m_k_strides); copy(in_desc.GetStrides(), gemm_g_m_k_strides);
copy(out_desc.GetStrides(), image_g_n_c_wis_strides); copy(out_desc.GetStrides(), image_g_n_c_wis_strides);
copy(conv_params.conv_filter_strides_, conv_filter_strides); copy(conv_params.conv_filter_strides_, conv_filter_strides);
copy(conv_params.conv_filter_dilations_, conv_filter_dilations); copy(conv_params.conv_filter_dilations_, conv_filter_dilations);
...@@ -86,13 +86,14 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv ...@@ -86,13 +86,14 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv
auto invoker = col2img.MakeInvoker(); auto invoker = col2img.MakeInvoker();
auto argument = col2img.MakeArgument(in_device_buf.GetDeviceBuffer(), auto argument = col2img.MakeArgument(in_device_buf.GetDeviceBuffer(),
out_device_buf.GetDeviceBuffer(), out_device_buf.GetDeviceBuffer(),
G,
N, N,
C, C,
input_spatial_lengths, input_spatial_lengths,
filter_spatial_lengths, filter_spatial_lengths,
output_spatial_lengths, output_spatial_lengths,
image_g_n_c_wis_strides, image_g_n_c_wis_strides,
gemm_m_k_strides, gemm_g_m_k_strides,
conv_filter_strides, conv_filter_strides,
conv_filter_dilations, conv_filter_dilations,
input_left_pads, input_left_pads,
...@@ -108,7 +109,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv ...@@ -108,7 +109,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv
} }
float ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel}); float ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
std::size_t num_btype = NDoHoWo * CZYX * (sizeof(OutDataType) + sizeof(InDataType)); std::size_t num_btype = G * NDoHoWo * CZYX * (sizeof(OutDataType) + sizeof(InDataType));
float gb_per_sec = num_btype / 1.E6 / ave_time; float gb_per_sec = num_btype / 1.E6 / ave_time;
std::cout << "Perf: " << ave_time << " ms, " << gb_per_sec << " GB/s" << std::endl; std::cout << "Perf: " << ave_time << " ms, " << gb_per_sec << " GB/s" << std::endl;
......
example/52_im2col_col2im/image_to_column_f32.cpp
View file @ bc641634
...@@ -20,7 +20,7 @@ using DeviceImgToColInstance = ck::tensor_operation::device::DeviceImageToColumn ...@@ -20,7 +20,7 @@ using DeviceImgToColInstance = ck::tensor_operation::device::DeviceImageToColumn
bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::ConvParam& conv_params) bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::ConvParam& conv_params)
{ {
const auto G = conv_params.G_;
const auto N = conv_params.N_; const auto N = conv_params.N_;
const auto C = conv_params.C_; const auto C = conv_params.C_;
...@@ -33,13 +33,13 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv ...@@ -33,13 +33,13 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
const auto in_desc = const auto in_desc =
ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(conv_params); ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(conv_params);
const auto out_desc = HostTensorDescriptor({NDoHoWo, CZYX}); const auto out_desc = HostTensorDescriptor({G, NDoHoWo, CZYX});
std::array<ck::index_t, NDimSpatial> input_spatial_lengths{}; std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{}; std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
std::array<ck::index_t, NDimSpatial> output_spatial_lengths{}; std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
std::array<ck::index_t, NDimSpatial + 3> image_g_n_c_wis_strides{}; std::array<ck::index_t, NDimSpatial + 3> image_g_n_c_wis_strides{};
std::array<ck::index_t, 2> gemm_m_k_strides{}; std::array<ck::index_t, 3> gemm_g_m_k_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_strides{}; std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_dilations{}; std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
std::array<ck::index_t, NDimSpatial> input_left_pads{}; std::array<ck::index_t, NDimSpatial> input_left_pads{};
...@@ -51,7 +51,7 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv ...@@ -51,7 +51,7 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
copy(conv_params.filter_spatial_lengths_, filter_spatial_lengths); copy(conv_params.filter_spatial_lengths_, filter_spatial_lengths);
copy(conv_params.output_spatial_lengths_, output_spatial_lengths); copy(conv_params.output_spatial_lengths_, output_spatial_lengths);
copy(in_desc.GetStrides(), image_g_n_c_wis_strides); copy(in_desc.GetStrides(), image_g_n_c_wis_strides);
copy(out_desc.GetStrides(), gemm_m_k_strides); copy(out_desc.GetStrides(), gemm_g_m_k_strides);
copy(conv_params.conv_filter_strides_, conv_filter_strides); copy(conv_params.conv_filter_strides_, conv_filter_strides);
copy(conv_params.conv_filter_dilations_, conv_filter_dilations); copy(conv_params.conv_filter_dilations_, conv_filter_dilations);
copy(conv_params.input_left_pads_, input_left_pads); copy(conv_params.input_left_pads_, input_left_pads);
...@@ -86,13 +86,14 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv ...@@ -86,13 +86,14 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
auto invoker = img2col.MakeInvoker(); auto invoker = img2col.MakeInvoker();
auto argument = img2col.MakeArgument(in_device_buf.GetDeviceBuffer(), auto argument = img2col.MakeArgument(in_device_buf.GetDeviceBuffer(),
out_device_buf.GetDeviceBuffer(), out_device_buf.GetDeviceBuffer(),
G,
N, N,
C, C,
input_spatial_lengths, input_spatial_lengths,
filter_spatial_lengths, filter_spatial_lengths,
output_spatial_lengths, output_spatial_lengths,
image_g_n_c_wis_strides, image_g_n_c_wis_strides,
gemm_m_k_strides, gemm_g_m_k_strides,
conv_filter_strides, conv_filter_strides,
conv_filter_dilations, conv_filter_dilations,
input_left_pads, input_left_pads,
...@@ -108,7 +109,7 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv ...@@ -108,7 +109,7 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
} }
float ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel}); float ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
std::size_t num_btype = NDoHoWo * CZYX * (sizeof(OutDataType) + sizeof(InDataType)); std::size_t num_btype = G * NDoHoWo * CZYX * (sizeof(OutDataType) + sizeof(InDataType));
float gb_per_sec = num_btype / 1.E6 / ave_time; float gb_per_sec = num_btype / 1.E6 / ave_time;
std::cout << "Perf: " << ave_time << " ms, " << gb_per_sec << " GB/s" << std::endl; std::cout << "Perf: " << ave_time << " ms, " << gb_per_sec << " GB/s" << std::endl;
......
example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
View file @ bc641634
...@@ -34,6 +34,7 @@ using AccDataType = F32; ...@@ -34,6 +34,7 @@ using AccDataType = F32;
using CShuffleDataType = F32; using CShuffleDataType = F32;
using DDataType = F16; using DDataType = F16;
using EDataType = F16; using EDataType = F16;
using ComputeDataType = F16;
static constexpr ck::index_t NumDimM = 2; static constexpr ck::index_t NumDimM = 2;
static constexpr ck::index_t NumDimN = 2; static constexpr ck::index_t NumDimN = 2;
...@@ -291,6 +292,7 @@ int main(int argc, char* argv[]) ...@@ -291,6 +292,7 @@ int main(int argc, char* argv[])
BDataType, BDataType,
CShuffleDataType, CShuffleDataType,
AccDataType, AccDataType,
ComputeDataType,
PassThrough, PassThrough,
BElementOp>; BElementOp>;
......
example/62_conv_fwd_activ/CMakeLists.txt
View file @ bc641634
...@@ -30,6 +30,9 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -30,6 +30,9 @@ foreach(gpu IN LISTS GPU_TARGETS)
# Elu # Elu
add_example_executable(example_convnd_fwd_xdl_elu_fp16 convnd_fwd_xdl_elu_fp16.cpp) add_example_executable(example_convnd_fwd_xdl_elu_fp16 convnd_fwd_xdl_elu_fp16.cpp)
add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_elu_fp16) add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_elu_fp16)
# ScaleAdd ScaleAdd Relu
add_example_executable(example_convnd_fwd_xdl_scaleadd_scaleadd_relu_fp16 convnd_fwd_xdl_scaleadd_scaleadd_relu_fp16.cpp)
add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_scaleadd_scaleadd_relu_fp16)
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
example/62_conv_fwd_activ/convnd_fwd_activ_common.hpp
View file @ bc641634
...@@ -190,9 +190,8 @@ bool run_grouped_conv_fwd(bool do_verification, ...@@ -190,9 +190,8 @@ bool run_grouped_conv_fwd(bool do_verification,
if(!conv.IsSupportedArgument(argument)) if(!conv.IsSupportedArgument(argument))
{ {
throw std::runtime_error( throw std::runtime_error("The device op with the specified compilation parameters does "
"wrong! device_conv with the specified compilation parameters does " "not support this convolution problem.");
"not support this Conv problem");
} }
float avg_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel}); float avg_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
......
example/62_conv_fwd_activ/convnd_fwd_xdl_scaleadd_scaleadd_relu_fp16.cpp 0 → 100644
View file @ bc641634
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iostream>
#include <numeric>
#include <type_traits>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
#include "ck/library/utility/algorithm.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/convolution_parameter.hpp"
#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
constexpr ck::index_t NDimSpatial = 3;
using InDataType = ck::half_t;
using WeiDataType = ck::half_t;
using AccDataType = float;
using CShuffleDataType = ck::half_t;
using OutDataType = ck::half_t;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using InLayout = ck::tensor_layout::convolution::GNDHWC;
using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
using OutLayout = ck::tensor_layout::convolution::GNDHWK;
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::ScaleAddScaleAddRelu;
static constexpr auto ConvSpec =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
template <typename OutElementOp>
using DeviceGroupedConvNDFwdInstance =
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
NDimSpatial,
InLayout,
WeiLayout,
ck::Tuple<OutLayout, OutLayout>,
OutLayout,
InDataType,
WeiDataType,
AccDataType,
CShuffleDataType,
ck::Tuple<OutDataType, OutDataType>,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
ConvSpec, // ConvForwardSpecialization
GemmSpec, // GemmSpecialization
1, //
256, // BlockSize
128, // MPerBlock
256, // NPerBlock
32, // KPerBlock
8, // AK1
8, // BK1
32, // MPerXdl
32, // NPerXdl
2, // MXdlPerWave
4, // NXdlPerWave
S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // ABlockTransferSrcAccessOrder
2, // ABlockTransferSrcVectorDim
8, // ABlockTransferSrcScalarPerVector
8, // ABlockTransferDstScalarPerVector_AK1
1, // ABlockLdsExtraM
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // BBlockTransferSrcAccessOrder
2, // BBlockTransferSrcVectorDim
8, // BBlockTransferSrcScalarPerVector
8, // BBlockTransferDstScalarPerVector_BK1
1, // BBlockLdsExtraN
1,
1,
S<1, 32, 1, 8>,
8>;
using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDFwdInstance<OutElementOp>;
namespace {
// Use custom implementation to pass two more tensors for post op
template <ck::index_t NDimSpatial,
typename InDataType,
typename WeiDataType,
typename OutDataType,
typename InElementOp,
typename WeiElementOp,
typename OutElementOp,
typename DeviceConvNDFwdInstance>
bool run_grouped_conv_fwd(bool do_verification,
int init_method,
bool time_kernel,
const ck::utils::conv::ConvParam& conv_param,
const HostTensorDescriptor& in_g_n_c_wis_desc,
const HostTensorDescriptor& wei_g_k_c_xs_desc,
const HostTensorDescriptor& out_g_n_k_wos_desc,
const InElementOp& in_element_op,
const WeiElementOp& wei_element_op,
const OutElementOp& out_element_op)
{
constexpr ck::index_t NumDs = 2;
Tensor<InDataType> in(in_g_n_c_wis_desc);
Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
Tensor<OutDataType> out_host(out_g_n_k_wos_desc);
Tensor<OutDataType> out_device(out_g_n_k_wos_desc);
std::array<Tensor<OutDataType>, NumDs> d_tensors = {Tensor<OutDataType>(out_g_n_k_wos_desc),
Tensor<OutDataType>(out_g_n_k_wos_desc)};
std::cout << "in: " << in.mDesc << std::endl;
std::cout << "wei: " << wei.mDesc << std::endl;
std::cout << "out: " << out_host.mDesc << std::endl;
switch(init_method)
{
case 0: break;
case 1:
in.GenerateTensorValue(GeneratorTensor_2<InDataType>{-2, 2});
wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-2, 2});
d_tensors[0].GenerateTensorValue(GeneratorTensor_2<OutDataType>{-2, 2});
d_tensors[1].GenerateTensorValue(GeneratorTensor_2<OutDataType>{-2, 2});
break;
default:
in.GenerateTensorValue(GeneratorTensor_3<InDataType>{-1.0, 1.0});
wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.05, 0.05});
d_tensors[0].GenerateTensorValue(GeneratorTensor_3<OutDataType>{-0.05, 0.05});
d_tensors[1].GenerateTensorValue(GeneratorTensor_3<OutDataType>{-0.05, 0.05});
}
DeviceMem in_device_buf(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
DeviceMem d0_buf(sizeof(OutDataType) * d_tensors[0].mDesc.GetElementSpaceSize());
DeviceMem d1_buf(sizeof(OutDataType) * d_tensors[1].mDesc.GetElementSpaceSize());
DeviceMem out_device_buf(sizeof(OutDataType) * out_device.mDesc.GetElementSpaceSize());
in_device_buf.ToDevice(in.mData.data());
wei_device_buf.ToDevice(wei.mData.data());
d0_buf.ToDevice(d_tensors[0].mData.data());
d1_buf.ToDevice(d_tensors[1].mData.data());
std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_lengths{};
std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_strides{};
std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_lengths{};
std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_strides{};
std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_lengths{};
std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
std::array<ck::index_t, NDimSpatial> input_left_pads{};
std::array<ck::index_t, NDimSpatial> input_right_pads{};
auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
copy(in_g_n_c_wis_desc.GetLengths(), a_g_n_c_wis_lengths);
copy(in_g_n_c_wis_desc.GetStrides(), a_g_n_c_wis_strides);
copy(wei_g_k_c_xs_desc.GetLengths(), b_g_k_c_xs_lengths);
copy(wei_g_k_c_xs_desc.GetStrides(), b_g_k_c_xs_strides);
copy(out_g_n_k_wos_desc.GetLengths(), e_g_n_k_wos_lengths);
copy(out_g_n_k_wos_desc.GetStrides(), e_g_n_k_wos_strides);
copy(conv_param.conv_filter_strides_, conv_filter_strides);
copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
copy(conv_param.input_left_pads_, input_left_pads);
copy(conv_param.input_right_pads_, input_right_pads);
const std::array<const void*, NumDs> ds = {d0_buf.GetDeviceBuffer(), d1_buf.GetDeviceBuffer()};
auto conv = DeviceConvNDFwdInstance{};
auto invoker = conv.MakeInvoker();
auto argument = conv.MakeArgument(in_device_buf.GetDeviceBuffer(),
wei_device_buf.GetDeviceBuffer(),
ds,
out_device_buf.GetDeviceBuffer(),
a_g_n_c_wis_lengths,
a_g_n_c_wis_strides,
b_g_k_c_xs_lengths,
b_g_k_c_xs_strides,
std::array<std::array<ck::index_t, NDimSpatial + 3>, NumDs>{
e_g_n_k_wos_lengths, e_g_n_k_wos_lengths},
std::array<std::array<ck::index_t, NDimSpatial + 3>, NumDs>{
e_g_n_k_wos_strides, e_g_n_k_wos_strides},
e_g_n_k_wos_lengths,
e_g_n_k_wos_strides,
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads,
in_element_op,
wei_element_op,
out_element_op);
if(!conv.IsSupportedArgument(argument))
{
throw std::runtime_error("The device op with the specified compilation parameters does "
"not support this convolution problem.");
}
float avg_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop =
conv_param.GetFlops() + 2 * conv_param.GetOutputByte<OutDataType>() / sizeof(OutDataType);
std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>() +
2 * conv_param.GetOutputByte<OutDataType>();
float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
float gb_per_sec = num_btype / 1.E6 / avg_time;
std::cout << "Perf: " << avg_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
<< conv.GetTypeString() << std::endl;
if(do_verification)
{
auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
NumDs>();
auto ref_invoker = ref_conv.MakeInvoker();
auto ref_argument = ref_conv.MakeArgument(in,
wei,
out_host,
conv_param.conv_filter_strides_,
conv_param.conv_filter_dilations_,
conv_param.input_left_pads_,
conv_param.input_right_pads_,
in_element_op,
wei_element_op,
out_element_op,
d_tensors);
ref_invoker.Run(ref_argument);
out_device_buf.FromDevice(out_device.mData.data());
return ck::utils::check_err(out_device, out_host, "Error: incorrect results!");
}
return true;
}
} // namespace
#include "run_convnd_fwd_activ_example.inc"
int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }
example/63_layernorm4d_fwd/CMakeLists.txt 0 → 100644
View file @ bc641634
add_example_executable(example_layernorm4d_fwd_fp16 layernorm4d_fwd_fp16.cpp)
add_example_executable(example_layernorm4d_fwd_splitk_fp16 layernorm4d_fwd_splitk_fp16.cpp)
example/63_layernorm4d_fwd/common.hpp 0 → 100644
View file @ bc641634
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include <getopt.h>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_fwd_impl.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_fwd_splitk_impl.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_common_util.hpp"
#include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp"
example/63_layernorm4d_fwd/layernorm4d_fwd_fp16.cpp 0 → 100644
View file @ bc641634
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using SaveMeanInvStdDataType = float;
using ComputeDataType = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
#define SAVE_MEAN_INV_STD
constexpr int Rank = 4;
constexpr int NumReduceDim = 3;
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationFwdImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
SaveMeanInvStdDataType,
PassThrough,
Rank,
NumReduceDim,
256, // BlockSize
8, // ClusterM
32, // ClusterK
1, // SliceM
8, // SliceK
1, // XYVectorDim (0=M, 1=K)
8, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
8, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
8, // BetaScalarPerVector
8, // YScalarPerVector
1>; // SaveMeanInvStdScalarPerVector
#include "run_layernorm4d_fwd_example.inc"
int main() { return run_layernorm4d_fwd_example<DeviceInstance>(); }
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