Skip to content

GitLab

Commit ef326c73 authored by Alan Turner's avatar Alan Turner
Browse files

Merge remote-tracking branch 'origin/develop' into migraphx-update

parents b7775add e4dfe4d8
Hide whitespace changes
Inline Side-by-side
Showing with 474 additions and 234 deletions
client_example/14_instance_id/CMakeLists.txt
View file @ ef326c73
add_executable(client_batchnorm_fwd_instance_id batchnorm_fwd_instance_id.cpp)
target_link_libraries(client_batchnorm_fwd_instance_id PRIVATE composable_kernel::device_operations)
target_link_libraries(client_batchnorm_fwd_instance_id PRIVATE composable_kernel::device_other_operations)
client_example/14_instance_id/batchnorm_fwd_instance_id.cpp
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include <functional>
#include <numeric>
......
client_example/15_convnd_bwd_data/CMakeLists.txt
View file @ ef326c73
add_executable(client_conv3d_bwd_data_fp16 conv3d_bwd_data_fp16.cpp)
add_executable(client_conv3d_bwd_data_fp32 conv3d_bwd_data_fp32.cpp)
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_conv3d_bwd_data_fp16 conv3d_bwd_data_fp16.cpp)
add_executable(client_conv3d_bwd_data_fp32 conv3d_bwd_data_fp32.cpp)
target_link_libraries(client_conv3d_bwd_data_fp16 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_conv3d_bwd_data_fp32 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_conv3d_bwd_data_fp16 PRIVATE composable_kernel::device_conv_operations)
target_link_libraries(client_conv3d_bwd_data_fp32 PRIVATE composable_kernel::device_conv_operations)
endif()
client_example/15_convnd_bwd_data/conv3d_bwd_data_fp16.cpp
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
......
client_example/15_convnd_bwd_data/conv3d_bwd_data_fp32.cpp
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
......
client_example/15_gemm_add_multiply/CMakeLists.txt
View file @ ef326c73
add_executable(client_gemm_add_multiply gemm_add_multiply.cpp)
target_link_libraries(client_gemm_add_multiply PRIVATE composable_kernel::device_operations)
\ No newline at end of file
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_gemm_add_multiply gemm_add_multiply.cpp)
target_link_libraries(client_gemm_add_multiply PRIVATE composable_kernel::device_gemm_operations)
endif()
client_example/15_gemm_add_multiply/gemm_add_multiply.cpp
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include <iomanip>
#include <vector>
......@@ -204,6 +204,7 @@ int main(int argc, char* argv[])
<< best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
// run the best intance
if(found)
{
auto& op_ptr = op_ptrs[best_op_id];
......
client_example/15_reduce/CMakeLists.txt
View file @ ef326c73
add_executable(client_reduce_nhwc_c reduce_nhwc_c.cpp)
target_link_libraries(client_reduce_nhwc_c PRIVATE composable_kernel::device_operations)
target_link_libraries(client_reduce_nhwc_c PRIVATE composable_kernel::device_reduction_operations)
client_example/15_reduce/reduce_nhwc_c.cpp
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include <functional>
#include <numeric>
......
client_example/16_convnd_fwd/CMakeLists.txt
View file @ ef326c73
add_executable(client_conv3d_fwd_fp16 conv3d_fwd_fp16.cpp)
add_executable(client_conv3d_fwd_fp32 conv3d_fwd_fp32.cpp)
if((DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
add_executable(client_conv3d_fwd_fp16 conv3d_fwd_fp16.cpp)
target_link_libraries(client_conv3d_fwd_fp16 PRIVATE composable_kernel::device_conv_operations)
target_link_libraries(client_conv3d_fwd_fp16 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_conv3d_fwd_fp32 PRIVATE composable_kernel::device_operations)
endif()
if((DTYPES MATCHES "fp8") OR (NOT DEFINED DTYPES AND GPU_TARGETS MATCHES "gfx94"))
add_executable(client_conv3d_fwd_fp16_comp_fp8 conv3d_fwd_fp16_comp_fp8.cpp)
target_link_libraries(client_conv3d_fwd_fp16_comp_fp8 PRIVATE composable_kernel::device_conv_operations)
endif()
if((DTYPES MATCHES "fp32") OR NOT DEFINED DTYPES)
add_executable(client_conv3d_fwd_fp32 conv3d_fwd_fp32.cpp)
target_link_libraries(client_conv3d_fwd_fp32 PRIVATE composable_kernel::device_conv_operations)
endif()
client_example/16_convnd_fwd/common.hpp
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iomanip>
......@@ -11,7 +11,7 @@
#include "ck/ck.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp"
#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp"
#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_abd.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
......@@ -94,7 +94,9 @@ template <ck::index_t NumDimSpatial,
typename InLayout,
typename WeiLayout,
typename OutLayout,
ck::index_t NumNonSpatialDim = 3>
ck::index_t NumNonSpatialDim = 3,
typename AComputeType = InDataType,
typename BComputeType = AComputeType>
bool run_grouped_conv_fwd(std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> in_lengths,
std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> wei_lengths,
std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> out_lengths)
......@@ -173,18 +175,20 @@ bool run_grouped_conv_fwd(std::array<ck::index_t, NumDimSpatial + NumNonSpatialD
std::size_t flop = GetFlops<NumDimSpatial>(out_lengths, wei_lengths);
std::size_t num_bytes = in_mem_size + wei_mem_size + out_mem_size;
using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
InLayout,
WeiLayout,
ck::Tuple<>,
OutLayout,
InDataType,
WeiDataType,
ck::Tuple<>,
OutDataType,
PassThrough,
PassThrough,
PassThrough>;
using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
InLayout,
WeiLayout,
ck::Tuple<>,
OutLayout,
InDataType,
WeiDataType,
ck::Tuple<>,
OutDataType,
PassThrough,
PassThrough,
PassThrough,
AComputeType,
BComputeType>;
// get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances();
......
client_example/16_convnd_fwd/conv3d_fwd_fp16.cpp
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
......
client_example/16_convnd_fwd/conv3d_fwd_fp16_comp_fp8.cpp 0 → 100644
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
using InDataType = ck::half_t;
using WeiDataType = ck::half_t;
using OutDataType = ck::half_t;
using InLayout = ck::tensor_layout::convolution::NDHWGC;
using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
using OutLayout = ck::tensor_layout::convolution::NDHWGK;
static constexpr ck::index_t NumDimSpatial = 3;
static constexpr ck::index_t G = 1;
static constexpr ck::index_t N = 64;
static constexpr ck::index_t K = 128;
static constexpr ck::index_t C = 64;
static constexpr ck::index_t Z = 3;
static constexpr ck::index_t Y = 3;
static constexpr ck::index_t X = 3;
static constexpr ck::index_t Di = 28;
static constexpr ck::index_t Hi = 28;
static constexpr ck::index_t Wi = 3;
static constexpr ck::index_t Do = 28;
static constexpr ck::index_t Ho = 28;
static constexpr ck::index_t Wo = 3;
int main()
{
return run_grouped_conv_fwd<NumDimSpatial,
InDataType,
WeiDataType,
OutDataType,
InLayout,
WeiLayout,
OutLayout,
3,
ck::f8_t>(
{N, Di, Hi, Wi, G, C}, {G, K, Z, Y, X, C}, {N, Do, Ho, Wo, G, K})
? EXIT_SUCCESS
: EXIT_FAILURE;
}
client_example/16_convnd_fwd/conv3d_fwd_fp32.cpp
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
......
client_example/17_grouped_gemm_fastgelu/CMakeLists.txt
View file @ ef326c73
add_executable(client_grouped_gemm_fastgelu grouped_gemm_fastgelu.cpp)
target_link_libraries(client_grouped_gemm_fastgelu PRIVATE composable_kernel::device_operations)
\ No newline at end of file
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_grouped_gemm_fastgelu grouped_gemm_fastgelu.cpp)
target_link_libraries(client_grouped_gemm_fastgelu PRIVATE composable_kernel::device_gemm_operations)
endif()
client_example/17_grouped_gemm_fastgelu/grouped_gemm_fastgelu.cpp
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include <iomanip>
#include <iostream>
......
client_example/18_groupnorm/CMakeLists.txt
View file @ ef326c73
add_executable(client_groupnorm_swish groupnorm_swish.cpp)
target_link_libraries(client_groupnorm_swish PRIVATE composable_kernel::device_operations)
add_executable(client_groupnorm_bwd_data groupnorm_bwd_data.cpp)
target_link_libraries(client_groupnorm_bwd_data PRIVATE composable_kernel::device_other_operations)
add_executable(client_groupnorm_bwd_gamma_beta groupnorm_bwd_gamma_beta.cpp)
target_link_libraries(client_groupnorm_bwd_gamma_beta PRIVATE composable_kernel::device_other_operations)
add_executable(client_groupnorm_swish_fwd groupnorm_swish_fwd.cpp)
target_link_libraries(client_groupnorm_swish_fwd PRIVATE composable_kernel::device_other_operations)
client_example/18_groupnorm/groupnorm_bwd_data.cpp 0 → 100644
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include <iomanip>
#include <vector>
#include <iostream>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_normalization_bwd_data.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/gpu/groupnorm_bwd_data.hpp"
using DYDataType = float;
using XDataType = float;
using GammaDataType = float;
using MeanInvStdDataType = float;
using DXDataType = float;
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
struct SimpleDeviceMem
{
SimpleDeviceMem() = delete;
SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
{
(void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
}
void* GetDeviceBuffer() { return p_mem_; }
~SimpleDeviceMem() { (void)hipFree(p_mem_); }
void* p_mem_;
};
int main(int argc, char* argv[])
{
ck::index_t N = 32;
ck::index_t H = 16;
ck::index_t W = 16;
ck::index_t G = 64;
ck::index_t C = 128;
std::size_t length = N * H * W * G * C;
std::vector<ck::index_t> strideDy = {H * W * G * C, W * G * C, G * C, C, 1};
std::vector<ck::index_t> strideX = strideDy;
std::vector<ck::index_t> strideDx = strideDy;
std::vector<ck::index_t> strideGamma = {0, 0, 0, C, 1};
std::vector<ck::index_t> strideMeanInvStd = {G, 0, 0, 1, 0};
SimpleDeviceMem dy_dev(sizeof(DYDataType) * length);
SimpleDeviceMem x_dev(sizeof(XDataType) * length);
SimpleDeviceMem gamma_dev(sizeof(GammaDataType) * G * C);
SimpleDeviceMem mean_dev(sizeof(MeanInvStdDataType) * N * G);
SimpleDeviceMem inv_std_dev(sizeof(MeanInvStdDataType) * N * G);
SimpleDeviceMem dx_dev(sizeof(DXDataType) * length);
using DeviceOp = ck::tensor_operation::device::DeviceNormalizationBwdData<DYDataType,
XDataType,
GammaDataType,
MeanInvStdDataType,
DXDataType,
Rank,
NumReduceDim>;
// get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances();
std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
std::string best_op_name;
bool found = false;
int best_op_id = -1;
float best_ave_time = std::numeric_limits<float>::max();
float best_gb_per_sec = 0;
// profile device operation instances
std::cout << "Run all instances and do timing" << std::endl;
for(int i = 0; i < op_ptrs.size(); ++i)
{
auto& op_ptr = op_ptrs[i];
auto argument_ptr = op_ptr->MakeArgumentPointer({N, H, W, G, C},
strideDy,
strideX,
strideGamma,
strideMeanInvStd,
strideMeanInvStd,
strideDx,
{1, 2, 4}, // reduceDims
dy_dev.GetDeviceBuffer(),
x_dev.GetDeviceBuffer(),
gamma_dev.GetDeviceBuffer(),
mean_dev.GetDeviceBuffer(),
inv_std_dev.GetDeviceBuffer(),
dx_dev.GetDeviceBuffer());
auto invoker_ptr = op_ptr->MakeInvokerPointer();
std::string op_name = op_ptr->GetTypeString();
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
{
size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
SimpleDeviceMem workspace(workspace_sz);
op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
std::size_t num_byte = sizeof(DYDataType) * length + sizeof(XDataType) * length +
sizeof(GammaDataType) * G * C +
sizeof(MeanInvStdDataType) * N * G * 2 +
sizeof(DXDataType) * length;
float gb_per_sec = num_byte / 1.E6 / ave_time;
std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
<< op_name << std::endl;
if(ave_time < best_ave_time)
{
found = true;
best_op_id = i;
best_op_name = op_name;
best_ave_time = ave_time;
best_gb_per_sec = gb_per_sec;
}
}
else
{
std::cout << op_name << " does not support this problem" << std::endl;
}
}
// run the best intance
if(found)
{
std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
<< best_op_name << std::endl;
auto& op_ptr = op_ptrs[best_op_id];
std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
<< std::endl;
auto argument_ptr = op_ptr->MakeArgumentPointer({N, H, W, G, C},
strideDy,
strideX,
strideGamma,
strideMeanInvStd,
strideMeanInvStd,
strideDx,
{1, 2, 4}, // reduceDims
dy_dev.GetDeviceBuffer(),
x_dev.GetDeviceBuffer(),
gamma_dev.GetDeviceBuffer(),
mean_dev.GetDeviceBuffer(),
inv_std_dev.GetDeviceBuffer(),
dx_dev.GetDeviceBuffer());
auto invoker_ptr = op_ptr->MakeInvokerPointer();
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
{
size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
SimpleDeviceMem workspace(workspace_sz);
op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
}
std::cout << "Done" << std::endl;
}
return 0;
}
client_example/18_groupnorm/groupnorm_bwd_gamma_beta.cpp 0 → 100644
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include <iomanip>
#include <vector>
#include <iostream>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/device_normalization_bwd_gamma_beta.hpp"
#include "ck/library/tensor_operation_instance/gpu/groupnorm_bwd_gamma_beta.hpp"
using DYDataType = float;
using XDataType = float;
using GammaDataType = float;
using MeanInvStdDataType = float;
using DGammaDataType = float;
using DBetaDataType = float;
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
struct SimpleDeviceMem
{
SimpleDeviceMem() = delete;
SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
{
(void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
}
void* GetDeviceBuffer() { return p_mem_; }
~SimpleDeviceMem() { (void)hipFree(p_mem_); }
void* p_mem_;
};
int main(int argc, char* argv[])
{
ck::index_t N = 32;
ck::index_t H = 16;
ck::index_t W = 16;
ck::index_t G = 64;
ck::index_t C = 128;
std::size_t length = N * H * W * G * C;
std::vector<ck::index_t> strideDy = {H * W * G * C, W * G * C, G * C, C, 1};
std::vector<ck::index_t> strideX = strideDy;
std::vector<ck::index_t> strideMeanInvStd = {G, 0, 0, 1, 0};
std::vector<ck::index_t> strideDGammaBeta = {C, 1};
SimpleDeviceMem dy_dev(sizeof(DYDataType) * length);
SimpleDeviceMem x_dev(sizeof(XDataType) * length);
SimpleDeviceMem mean_dev(sizeof(MeanInvStdDataType) * N * G);
SimpleDeviceMem inv_std_dev(sizeof(MeanInvStdDataType) * N * G);
SimpleDeviceMem dgamma_dev(sizeof(DGammaDataType) * G * C);
SimpleDeviceMem dbeta_dev(sizeof(DBetaDataType) * G * C);
using DeviceOp =
ck::tensor_operation::device::DeviceNormalizationBwdGammaBeta<DYDataType,
XDataType,
MeanInvStdDataType,
DGammaDataType,
DBetaDataType,
Rank,
NumReduceDim>;
// get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances();
std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
std::string best_op_name;
bool found = false;
int best_op_id = -1;
float best_ave_time = std::numeric_limits<float>::max();
float best_gb_per_sec = 0;
// profile device operation instances
std::cout << "Run all instances and do timing" << std::endl;
std::size_t num_bytes = sizeof(DYDataType) * length + sizeof(XDataType) * length +
sizeof(GammaDataType) * G * C + sizeof(MeanInvStdDataType) * N * G * 2 +
sizeof(DGammaDataType) * G * C + sizeof(DBetaDataType) * G * C;
for(int i = 0; i < op_ptrs.size(); ++i)
{
auto& op_ptr = op_ptrs[i];
auto argument_ptr = op_ptr->MakeArgumentPointer({N, H, W, G, C},
strideDy,
strideX,
strideMeanInvStd,
strideMeanInvStd,
{G, C},
strideDGammaBeta,
strideDGammaBeta,
{0, 1, 2}, // reduceDims
dy_dev.GetDeviceBuffer(),
x_dev.GetDeviceBuffer(),
mean_dev.GetDeviceBuffer(),
inv_std_dev.GetDeviceBuffer(),
dgamma_dev.GetDeviceBuffer(),
dbeta_dev.GetDeviceBuffer());
auto invoker_ptr = op_ptr->MakeInvokerPointer();
std::string op_name = op_ptr->GetTypeString();
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
{
size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
SimpleDeviceMem workspace(workspace_sz);
op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
float gb_per_sec = num_bytes / 1.E6 / ave_time;
std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
<< op_name << std::endl;
if(ave_time < best_ave_time)
{
found = true;
best_op_id = i;
best_op_name = op_name;
best_ave_time = ave_time;
best_gb_per_sec = gb_per_sec;
}
}
else
{
std::cout << op_name << " does not support this problem" << std::endl;
}
}
// run the best intance
if(found)
{
std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
<< best_op_name << std::endl;
auto& op_ptr = op_ptrs[best_op_id];
std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
<< std::endl;
auto argument_ptr = op_ptr->MakeArgumentPointer({N, H, W, G, C},
strideDy,
strideX,
strideMeanInvStd,
strideMeanInvStd,
{G, C},
strideDGammaBeta,
strideDGammaBeta,
{0, 1, 2}, // reduceDims
dy_dev.GetDeviceBuffer(),
x_dev.GetDeviceBuffer(),
mean_dev.GetDeviceBuffer(),
inv_std_dev.GetDeviceBuffer(),
dgamma_dev.GetDeviceBuffer(),
dbeta_dev.GetDeviceBuffer());
auto invoker_ptr = op_ptr->MakeInvokerPointer();
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
{
size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
SimpleDeviceMem workspace(workspace_sz);
op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
}
std::cout << "Done" << std::endl;
}
return 0;
}
client_example/18_groupnorm/groupnorm_swish.cpp deleted 100644 → 0
View file @ b7775add
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iomanip>
#include <vector>
#include <iostream>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_normalization.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/gpu/normalization_swish.hpp"
using XDataType = ck::half_t;
using GammaDataType = float;
using BetaDataType = float;
using YDataType = ck::half_t;
using ComputeDataType = float;
using Swish = ck::tensor_operation::element_wise::Swish;
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
struct SimpleDeviceMem
{
SimpleDeviceMem() = delete;
SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
{
(void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
}
void* GetDeviceBuffer() { return p_mem_; }
~SimpleDeviceMem() { (void)hipFree(p_mem_); }
void* p_mem_;
};
int main(int argc, char* argv[])
{
ck::index_t N = 32;
ck::index_t H = 16;
ck::index_t W = 16;
ck::index_t G = 64;
ck::index_t C = 128;
std::size_t xy_size = N * H * W * G * C;
std::size_t gamma_beta_size = G * C;
std::vector<ck::index_t> xy_strides = {H * W * G * C, W * G * C, G * C, C, 1};
std::vector<ck::index_t> gamma_beta_strides = {0, 0, 0, C, 1};
SimpleDeviceMem x_device_buf(sizeof(XDataType) * xy_size);
SimpleDeviceMem gamma_device_buf(sizeof(GammaDataType) * gamma_beta_size);
SimpleDeviceMem beta_device_buf(sizeof(BetaDataType) * gamma_beta_size);
SimpleDeviceMem y_device_buf(sizeof(YDataType) * xy_size);
using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
Swish,
Rank,
NumReduceDim>;
// get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances();
std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
const auto& generic_op_ptr = op_ptrs[0];
auto generic_argument_ptr =
generic_op_ptr->MakeArgumentPointer({N, H, W, G, C}, // lengths
xy_strides, // xStrides
gamma_beta_strides, // gammaStrides
gamma_beta_strides, // betaStrides
xy_strides, // yStrides
{1, 2, 4}, // reduceDims
1e-6,
x_device_buf.GetDeviceBuffer(),
gamma_device_buf.GetDeviceBuffer(),
beta_device_buf.GetDeviceBuffer(),
y_device_buf.GetDeviceBuffer(),
nullptr,
nullptr,
Swish{});
if(!generic_op_ptr->IsSupportedArgument(generic_argument_ptr.get()))
{
throw std::runtime_error(
"The generic kernel instance should be able to support any input shapes");
};
std::string best_op_name;
bool found = false;
int best_op_id = -1;
float best_ave_time = std::numeric_limits<float>::max();
float best_gb_per_sec = 0;
// profile device operation instances
std::cout << "Run all instances and do timing" << std::endl;
for(int i = 0; i < op_ptrs.size(); ++i)
{
auto& op_ptr = op_ptrs[i];
auto argument_ptr = op_ptr->MakeArgumentPointer({N, H, W, G, C}, // lengths
xy_strides, // xStrides
gamma_beta_strides, // gammaStrides
gamma_beta_strides, // betaStrides
xy_strides, // yStrides
{1, 2, 4}, // reduceDims
1e-6,
x_device_buf.GetDeviceBuffer(),
gamma_device_buf.GetDeviceBuffer(),
beta_device_buf.GetDeviceBuffer(),
y_device_buf.GetDeviceBuffer(),
nullptr,
nullptr,
Swish{});
auto invoker_ptr = op_ptr->MakeInvokerPointer();
std::string op_name = op_ptr->GetTypeString();
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
{
float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
std::size_t num_byte =
sizeof(XDataType) * xy_size + sizeof(GammaDataType) * gamma_beta_size +
sizeof(BetaDataType) * gamma_beta_size + sizeof(YDataType) * xy_size;
float gb_per_sec = num_byte / 1.E6 / ave_time;
std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
<< op_name << std::endl;
if(ave_time < best_ave_time)
{
found = true;
best_op_id = i;
best_op_name = op_name;
best_ave_time = ave_time;
best_gb_per_sec = gb_per_sec;
}
}
else
{
std::cout << op_name << " does not support this problem" << std::endl;
}
}
// run the best intance
if(found)
{
std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
<< best_op_name << std::endl;
auto& op_ptr = op_ptrs[best_op_id];
std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
<< std::endl;
auto argument_ptr = op_ptr->MakeArgumentPointer({N, H, W, G, C}, // lengths
xy_strides, // xStrides
gamma_beta_strides, // gammaStrides
gamma_beta_strides, // betaStrides
xy_strides, // yStrides
{1, 2, 4}, // reduceDims
1e-6,
x_device_buf.GetDeviceBuffer(),
gamma_device_buf.GetDeviceBuffer(),
beta_device_buf.GetDeviceBuffer(),
y_device_buf.GetDeviceBuffer(),
nullptr,
nullptr,
Swish{});
auto invoker_ptr = op_ptr->MakeInvokerPointer();
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
{
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
}
std::cout << "Done" << std::endl;
}
return 0;
}
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