ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-15 18:42:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

Support multi AB for grouped conv fwd xdl (#1027)

Browse Source 
* Support multi AB for grouped conv fwd xdl

* Add instances

* Add client example

* Add example

* Add interface test

* Minor fixes

Minor fixes

Minor fixes

* Comment fixes

* Fixes

* Reference fix

* Test xdl fixes

* Improve multi_ab interface test

[ROCm/composable_kernel commit: 49e52bb357]
This commit is contained in:
Bartłomiej Kocot
2023-11-10 15:54:44 +01:00
committed by GitHub
parent 283920969f
commit 05d18a052b
40 changed files with 2235 additions and 365 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
example/62_conv_fwd_activ/CMakeLists.txt
View File

@@ -30,6 +30,15 @@ foreach(gpu IN LISTS GPU_TARGETS)
# Elu
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)
# ScaleAdd on A and B
add_example_executable(example_conv_fwd_xdl_scaleadd_ab_fp16 multi_AB/conv_fwd_xdl_scaleadd_ab_fp16.cpp)
add_example_dependencies(example_convnd_fwd_activ_xdl example_conv_fwd_xdl_scaleadd_ab_fp16)
add_example_executable(example_conv_fwd_xdl_scaleadd_ab_fp32 multi_AB/conv_fwd_xdl_scaleadd_ab_fp32.cpp)
add_example_dependencies(example_convnd_fwd_activ_xdl example_conv_fwd_xdl_scaleadd_ab_fp32)
add_example_executable(example_conv_fwd_xdl_scaleadd_ab_bf16 multi_AB/conv_fwd_xdl_scaleadd_ab_bf16.cpp)
add_example_dependencies(example_convnd_fwd_activ_xdl example_conv_fwd_xdl_scaleadd_ab_bf16)
add_example_executable(example_conv_fwd_xdl_scaleadd_ab_int8 multi_AB/conv_fwd_xdl_scaleadd_ab_int8.cpp)
add_example_dependencies(example_convnd_fwd_activ_xdl example_conv_fwd_xdl_scaleadd_ab_int8)
# 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)

21
example/62_conv_fwd_activ/convnd_fwd_xdl_scaleadd_scaleadd_relu_fp16.cpp
View File

@@ -226,14 +226,17 @@ bool run_grouped_conv_fwd(bool do_verification,
if(do_verification)
{
auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
NumDs>();
auto ref_conv =
ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
0, /*Num A Elementwise Tensors*/
0, /*Num B Elementwise Tensors*/
NumDs>();
auto ref_invoker = ref_conv.MakeInvoker();
auto ref_argument = ref_conv.MakeArgument(in,
@@ -246,6 +249,8 @@ bool run_grouped_conv_fwd(bool do_verification,
in_element_op,
wei_element_op,
out_element_op,
{},
{},
d_tensors);
ref_invoker.Run(ref_argument);

26
example/62_conv_fwd_activ/multi_AB/conv_fwd_xdl_scaleadd_ab_bf16.cpp Normal file
View File

@@ -0,0 +1,26 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
#include "convnd_fwd_activ_multi_ab_common.hpp"
using DataType = ck::bhalf_t;
using AccDataType = float;
using InDataType = DataType;
using WeiDataType = DataType;
using OutDataType = DataType;
using ADataTypes = ck::Tuple<DataType, DataType>;
using BDataTypes = ck::Tuple<DataType, DataType>;
using InElementOp = ck::tensor_operation::element_wise::ScaleAdd;
using WeiElementOp = ck::tensor_operation::element_wise::ScaleAdd;
using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDMultiABFwdInstance<DataType,
AccDataType,
ADataTypes,
BDataTypes,
InElementOp,
WeiElementOp>;
#include "../run_convnd_fwd_activ_example.inc"
int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }

26
example/62_conv_fwd_activ/multi_AB/conv_fwd_xdl_scaleadd_ab_fp16.cpp Normal file
View File

@@ -0,0 +1,26 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
#include "convnd_fwd_activ_multi_ab_common.hpp"
using DataType = ck::half_t;
using AccDataType = float;
using InDataType = DataType;
using WeiDataType = DataType;
using OutDataType = DataType;
using ADataTypes = ck::Tuple<DataType, DataType>;
using BDataTypes = ck::Tuple<DataType, DataType>;
using InElementOp = ck::tensor_operation::element_wise::ScaleAdd;
using WeiElementOp = ck::tensor_operation::element_wise::ScaleAdd;
using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDMultiABFwdInstance<DataType,
AccDataType,
ADataTypes,
BDataTypes,
InElementOp,
WeiElementOp>;
#include "../run_convnd_fwd_activ_example.inc"
int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }

26
example/62_conv_fwd_activ/multi_AB/conv_fwd_xdl_scaleadd_ab_fp32.cpp Normal file
View File

@@ -0,0 +1,26 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
#include "convnd_fwd_activ_multi_ab_common.hpp"
using DataType = float;
using AccDataType = float;
using InDataType = DataType;
using WeiDataType = DataType;
using OutDataType = DataType;
using ADataTypes = ck::Tuple<DataType, DataType>;
using BDataTypes = ck::Tuple<DataType, DataType>;
using InElementOp = ck::tensor_operation::element_wise::ScaleAdd;
using WeiElementOp = ck::tensor_operation::element_wise::ScaleAdd;
using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDMultiABFwdInstance<DataType,
AccDataType,
ADataTypes,
BDataTypes,
InElementOp,
WeiElementOp>;
#include "../run_convnd_fwd_activ_example.inc"
int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }

26
example/62_conv_fwd_activ/multi_AB/conv_fwd_xdl_scaleadd_ab_int8.cpp Normal file
View File

@@ -0,0 +1,26 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
#include "convnd_fwd_activ_multi_ab_common.hpp"
using DataType = int8_t;
using AccDataType = int32_t;
using InDataType = DataType;
using WeiDataType = DataType;
using OutDataType = DataType;
using ADataTypes = ck::Tuple<DataType, DataType>;
using BDataTypes = ck::Tuple<DataType, DataType>;
using InElementOp = ck::tensor_operation::element_wise::ScaleAdd;
using WeiElementOp = ck::tensor_operation::element_wise::ScaleAdd;
using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDMultiABFwdInstance<DataType,
AccDataType,
ADataTypes,
BDataTypes,
InElementOp,
WeiElementOp>;
#include "../run_convnd_fwd_activ_example.inc"
int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }

266
example/62_conv_fwd_activ/multi_AB/convnd_fwd_activ_multi_ab_common.hpp Normal file
View File

@@ -0,0 +1,266 @@
// 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;
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 OutElementOp = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvSpec =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
template <typename DataType,
typename AccDataType,
typename InDataTypes,
typename WeiDataTypes,
typename InElementOp,
typename WeiElementOp>
using DeviceGroupedConvNDMultiABFwdInstance =
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
NDimSpatial,
InLayout,
WeiLayout,
ck::Tuple<>,
OutLayout,
InDataTypes,
WeiDataTypes,
AccDataType,
DataType,
ck::Tuple<>,
DataType,
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>;
namespace {
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 NumAs = 2;
constexpr ck::index_t NumBs = 2;
Tensor<InDataType> in(in_g_n_c_wis_desc);
Tensor<InDataType> in_bias(in_g_n_c_wis_desc);
Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
Tensor<WeiDataType> wei_bias(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::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});
in_bias.GenerateTensorValue(GeneratorTensor_2<InDataType>{-2, 2});
wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-2, 2});
wei_bias.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-2, 2});
break;
default:
in.GenerateTensorValue(GeneratorTensor_3<InDataType>{-1.0, 1.0});
in_bias.GenerateTensorValue(GeneratorTensor_3<InDataType>{-1.0, 1.0});
wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.05, 0.05});
wei_bias.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-1.0, 1.0});
}
DeviceMem in_device_buf(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
DeviceMem in_bias_device_buf(sizeof(InDataType) * in_bias.mDesc.GetElementSpaceSize());
DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
DeviceMem wei_bias_device_buf(sizeof(WeiDataType) * wei_bias.mDesc.GetElementSpaceSize());
DeviceMem out_device_buf(sizeof(OutDataType) * out_device.mDesc.GetElementSpaceSize());
in_device_buf.ToDevice(in.mData.data());
in_bias_device_buf.ToDevice(in_bias.mData.data());
wei_device_buf.ToDevice(wei.mData.data());
wei_bias_device_buf.ToDevice(wei_bias.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);
std::array<const void*, NumAs> as{in_device_buf.GetDeviceBuffer(),
in_bias_device_buf.GetDeviceBuffer()};
std::array<const void*, NumBs> bs{wei_device_buf.GetDeviceBuffer(),
wei_bias_device_buf.GetDeviceBuffer()};
std::array<const void*, 0> ds{};
// do Conv
auto conv = DeviceConvNDFwdInstance{};
auto invoker = conv.MakeInvoker();
auto argument = conv.MakeArgument(as,
bs,
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,
{},
{},
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(
"wrong! device_conv with the specified compilation parameters does "
"not support this Conv problem");
}
float avg_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = conv_param.GetFlops() +
2 * conv_param.GetOutputByte<InDataType>() / sizeof(InDataType) +
2 * conv_param.GetOutputByte<WeiDataType>() / sizeof(WeiDataType);
std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>() +
conv_param.GetInputByte<InDataType>() +
conv_param.GetWeightByte<WeiDataType>();
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)
{
const std::array<Tensor<InDataType>, NumAs - 1> elementwise_a_tensors = {in_bias};
const std::array<Tensor<WeiDataType>, NumBs - 1> elementwise_b_tensors = {wei_bias};
auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
NumAs - 1,
NumBs - 1>();
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,
elementwise_a_tensors,
elementwise_b_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