Skip to content

GitLab

Unverified Commit 500fa995 authored by Chao Liu's avatar Chao Liu Committed by GitHub
Browse files

Clean up conv example, Instances, profiler and test (#324) 

* convnd_fwd fp16 example

* update example

* update example

* update instance

* updating refernce conv

* update reference conv

* update conv fwd profiler

* update conv 1d and 3d instance

* update include path

* clean

* update profiler for conv bwd data and weight

* update conv bwd weight

* clean

* update conv example

* update profiler for conv bwd weight

* update ckprofiler for conv bwd data

* fix reference conv bwd data bug; update conv bwd data test

* update examples

* fix initialization issue

* update test for conv fwd

* clean

* clean

* remove test case too sensitive to error threshhold

* fix test

* clean

* fix build

* adding conv multiple d

* adding conv multiple D

* add matrix padder

* add gemm padding to convnd

* adding group conv

* update gemm multi-d

* refactor

* refactor

* refactor

* clean

* clean

* refactor

* refactor

* reorg

* add ds

* add bias

* clean

* add G

* adding group

* adding group

* adding group

* update Tensor

* clean

* update example

* update DeviceGemmMultipleD_Xdl_CShuffle

* update conv bwd-data and bwd-weight

* upate contraction example

* update gemm and batch gemm with e permute

* fix example build

* instance for grouped conv1d

* update example

* adding group conv instance

* update gemm bilinear instance

* update gemm+add+add+fastgelu instance

* update profiler

* update profiler

* update test

* update test and client example

* clean

* add grouped conv into profiler

* update profiler

* clean

* add test grouped conv, update all conv test to gtest

* update test
parent 85978e02
Expand all Show whitespace changes
Inline Side-by-side
Showing with 1044 additions and 1919 deletions
example/09_convnd_fwd/convnd_fwd_xdl_bf16.cpp 0 → 100644
View file @ 500fa995
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "convnd_fwd_common.hpp"
#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
using InDataType = ck::bhalf_t;
using WeiDataType = ck::bhalf_t;
using AccDataType = float;
using CShuffleDataType = float;
using OutDataType = ck::bhalf_t;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
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 <ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout>
using DeviceGroupedConvNDFwdInstance =
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
NDimSpatial,
InLayout,
WeiLayout,
ck::Tuple<>,
OutLayout,
InDataType,
WeiDataType,
AccDataType,
CShuffleDataType,
ck::Tuple<>,
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>;
int main(int argc, char* argv[])
{
namespace ctc = ck::tensor_layout::convolution;
print_helper_msg();
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
ck::utils::conv::ConvParam conv_param{
2, 1, 128, 256, 192, {3, 3}, {71, 71}, {2, 2}, {1, 1}, {1, 1}, {1, 1}};
if(argc == 1)
{
// use default
}
else if(argc == 4)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
}
else
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
const ck::index_t num_dim_spatial = std::stoi(argv[4]);
conv_param = ck::utils::conv::parse_conv_param(num_dim_spatial, 5, argv);
}
const auto in_element_op = InElementOp{};
const auto wei_element_op = WeiElementOp{};
const auto out_element_op = OutElementOp{};
if(conv_param.num_dim_spatial_ == 1)
{
using InLayout = ctc::GNWC;
using WeiLayout = ctc::GKXC;
using OutLayout = ctc::GNWK;
const auto in_g_n_c_wis_desc =
ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
conv_param);
const auto wei_g_k_c_xs_desc =
ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
conv_param);
const auto out_g_n_k_wos_desc =
ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
conv_param);
return run_grouped_conv_fwd<
1,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
DeviceGroupedConvNDFwdInstance<1, InLayout, WeiLayout, OutLayout>>(do_verification,
init_method,
time_kernel,
conv_param,
in_g_n_c_wis_desc,
wei_g_k_c_xs_desc,
out_g_n_k_wos_desc,
in_element_op,
wei_element_op,
out_element_op);
}
else if(conv_param.num_dim_spatial_ == 2)
{
using InLayout = ctc::GNHWC;
using WeiLayout = ctc::GKYXC;
using OutLayout = ctc::GNHWK;
const auto in_g_n_c_wis_desc =
ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
conv_param);
const auto wei_g_k_c_xs_desc =
ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
conv_param);
const auto out_g_n_k_wos_desc =
ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
conv_param);
return run_grouped_conv_fwd<
2,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
DeviceGroupedConvNDFwdInstance<2, InLayout, WeiLayout, OutLayout>>(do_verification,
init_method,
time_kernel,
conv_param,
in_g_n_c_wis_desc,
wei_g_k_c_xs_desc,
out_g_n_k_wos_desc,
in_element_op,
wei_element_op,
out_element_op);
}
else if(conv_param.num_dim_spatial_ == 3)
{
using InLayout = ctc::GNDHWC;
using WeiLayout = ctc::GKZYXC;
using OutLayout = ctc::GNDHWK;
const auto in_g_n_c_wis_desc =
ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
conv_param);
const auto wei_g_k_c_xs_desc =
ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
conv_param);
const auto out_g_n_k_wos_desc =
ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
conv_param);
return run_grouped_conv_fwd<
3,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
DeviceGroupedConvNDFwdInstance<3, InLayout, WeiLayout, OutLayout>>(do_verification,
init_method,
time_kernel,
conv_param,
in_g_n_c_wis_desc,
wei_g_k_c_xs_desc,
out_g_n_k_wos_desc,
in_element_op,
wei_element_op,
out_element_op);
}
return 0;
}
example/10_conv2d_bwd_data/CMakeLists.txt deleted 100644 → 0
View file @ 85978e02
add_example_executable(example_conv2d_bwd_data_xdl conv2d_bwd_data_xdl.cpp)
target_link_libraries(example_conv2d_bwd_data_xdl PRIVATE conv_util)
example/10_conv2d_bwd_data/README.md deleted 100644 → 0
View file @ 85978e02
# Instructions for ```example_conv2d_bwd_data_xdl``` Example
## Run ```example_conv2d_bwd_data_xdl```
```bash
#arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1)
#arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx
./bin/example_conv2d_bwd_data_xdl 0 1 5
```
Result
```
in_n_c_hi_wi: dim 4, lengths {128, 256, 71, 71}, strides {1290496, 1, 18176, 256}
wei_k_c_y_x: dim 4, lengths {256, 256, 3, 3}, strides {2304, 1, 768, 256}
out_n_k_ho_wo: dim 4, lengths {128, 256, 36, 36}, strides {331776, 1, 9216, 256}
arg.a_grid_desc_k0_m_k1_container_{128, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{128, 256, 8}
arg.c_grid_desc_m_n_container_{ 175232, 256}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 4, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {2738, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
arg.a_grid_desc_k0_m_k1_container_{64, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{64, 256, 8}
arg.c_grid_desc_m_n_container_{ 175232, 256}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 4, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {2738, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
arg.a_grid_desc_k0_m_k1_container_{64, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{64, 256, 8}
arg.c_grid_desc_m_n_container_{ 175232, 256}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 4, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {2738, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
arg.a_grid_desc_k0_m_k1_container_{32, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{32, 256, 8}
arg.c_grid_desc_m_n_container_{ 175232, 256}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 4, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {2738, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
Perf: 2.45966 ms, 79.5597 TFlops, 169.325 GB/s
```
example/10_conv2d_bwd_data/conv2d_bwd_data_xdl.cpp deleted 100644 → 0
View file @ 85978e02
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/conv_util.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp"
using InDataType = ck::half_t;
using WeiDataType = ck::half_t;
using OutDataType = ck::half_t;
using AccDataType = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
using DeviceConvBwdDataInstance = ck::tensor_operation::device::
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
InDataType, // InDataType
WeiDataType, // WeiDataType
OutDataType, // OutDataType
AccDataType, // AccDataType
InElementOp, // InElementwiseOperation
WeiElementOp, // WeiElementwiseOperation
OutElementOp, // OutElementwiseOperation
ConvBwdDefault, // ConvolutionBackwardDataSpecialization
256, // BlockSize
128, // MPerBlock
128, // NPerBlock
4, // K0PerBlock
8, // K1
32, // MPerXdl
32, // NPerXdl
2, // MXdlPerWave
2, // NXdlPerWave
S<4, 64, 1>, // ABlockTransferThreadClusterLengths_K0_M_K1
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // ABlockTransferSrcAccessOrder
2, // ABlockTransferSrcVectorDim
8, // ABlockTransferSrcScalarPerVector
8, // ABlockTransferDstScalarPerVector_K1
true, // ABlockLdsAddExtraM
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1
S<2, 0, 1>, // BBlockTransferThreadClusterArrangeOrder
S<0, 2, 1>, // BBlockTransferSrcAccessOrder
1, // BBlockTransferSrcVectorDim
2, // BBlockTransferSrcScalarPerVector
8, // BBlockTransferDstScalarPerVector_K1
true, // BBlockLdsAddExtraN
7,
1>; // GemmCThreadTransferDstScalarPerVector
using ReferenceConvBwdInstance = ck::tensor_operation::host::ReferenceConvBwdData<InDataType,
WeiDataType,
OutDataType,
AccDataType,
InElementOp,
WeiElementOp,
OutElementOp>;
int main(int argc, char* argv[])
{
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
// Conv shape
ck::index_t N = 128;
ck::index_t K = 256;
ck::index_t C = 256;
ck::index_t Y = 3;
ck::index_t X = 3;
ck::index_t Hi = 71;
ck::index_t Wi = 71;
ck::index_t conv_stride_h = 2;
ck::index_t conv_stride_w = 2;
ck::index_t conv_dilation_h = 1;
ck::index_t conv_dilation_w = 1;
ck::index_t in_left_pad_h = 1;
ck::index_t in_left_pad_w = 1;
ck::index_t in_right_pad_h = 1;
ck::index_t in_right_pad_w = 1;
if(argc == 4)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
}
else if(argc == 19)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
N = std::stoi(argv[4]);
K = std::stoi(argv[5]);
C = std::stoi(argv[6]);
Y = std::stoi(argv[7]);
X = std::stoi(argv[8]);
Hi = std::stoi(argv[9]);
Wi = std::stoi(argv[10]);
conv_stride_h = std::stoi(argv[11]);
conv_stride_w = std::stoi(argv[12]);
conv_dilation_h = std::stoi(argv[13]);
conv_dilation_w = std::stoi(argv[14]);
in_left_pad_h = std::stoi(argv[15]);
in_left_pad_w = std::stoi(argv[16]);
in_right_pad_h = std::stoi(argv[17]);
in_right_pad_w = std::stoi(argv[18]);
}
else
{
printf("arg1: verification (0=no, 1=yes)\n");
printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
printf("arg3: time kernel (0=n0, 1=yes)\n");
printf("arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
"RightPx\n");
exit(0);
}
const ck::index_t YEff = (Y - 1) * conv_dilation_h + 1;
const ck::index_t XEff = (X - 1) * conv_dilation_w + 1;
const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - YEff) / conv_stride_h + 1;
const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
const std::vector<ck::index_t> conv_filter_strides{{conv_stride_h, conv_stride_w}};
const std::vector<ck::index_t> conv_filter_dilations{{conv_dilation_h, conv_dilation_w}};
const std::vector<ck::index_t> input_left_pads{{in_left_pad_h, in_left_pad_w}};
const std::vector<ck::index_t> input_right_pads{{in_right_pad_h, in_right_pad_w}};
// tensor layout
auto f_host_tensor_descriptor =
[](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W) {
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
std::vector<std::size_t>({C_ * H * W, 1, W * C_, C_}));
};
Tensor<OutDataType> out_n_k_ho_wo(f_host_tensor_descriptor(N, K, Ho, Wo));
Tensor<WeiDataType> wei_k_c_y_x(f_host_tensor_descriptor(K, C, Y, X));
Tensor<InDataType> in_n_c_hi_wi_host_result(f_host_tensor_descriptor(N, C, Hi, Wi));
Tensor<InDataType> in_n_c_hi_wi_device_result(f_host_tensor_descriptor(N, C, Hi, Wi));
std::cout << "in_n_c_hi_wi: " << in_n_c_hi_wi_host_result.mDesc << std::endl;
std::cout << "wei_k_c_y_x: " << wei_k_c_y_x.mDesc << std::endl;
std::cout << "out_n_k_ho_wo: " << out_n_k_ho_wo.mDesc << std::endl;
switch(init_method)
{
case 0: break;
case 1:
out_n_k_ho_wo.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
break;
default:
out_n_k_ho_wo.GenerateTensorValue(GeneratorTensor_1<OutDataType>{1});
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{1});
}
DeviceMem in_device_buf(sizeof(InDataType) *
in_n_c_hi_wi_device_result.mDesc.GetElementSpace());
DeviceMem wei_device_buf(sizeof(WeiDataType) * wei_k_c_y_x.mDesc.GetElementSpace());
DeviceMem out_device_buf(sizeof(OutDataType) * out_n_k_ho_wo.mDesc.GetElementSpace());
out_device_buf.ToDevice(out_n_k_ho_wo.mData.data());
wei_device_buf.ToDevice(wei_k_c_y_x.mData.data());
// reset input to zero
in_n_c_hi_wi_device_result.GenerateTensorValue(GeneratorTensor_1<InDataType>{0});
in_device_buf.ToDevice(in_n_c_hi_wi_device_result.mData.data());
// do GEMM
auto conv = DeviceConvBwdDataInstance{};
auto invoker = conv.MakeInvoker();
auto argument = conv.MakeArgument(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
N,
K,
C,
std::vector<ck::index_t>{{Hi, Wi}},
std::vector<ck::index_t>{{Y, X}},
std::vector<ck::index_t>{{Ho, Wo}},
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads,
InElementOp{},
WeiElementOp{},
OutElementOp{});
if(!conv.IsSupportedArgument(argument))
{
throw std::runtime_error(
"wrong! device_conv with the specified compilation parameters does "
"not support this Conv problem");
}
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * N * K * Ho * Wo * C * Y * X;
std::size_t num_btype = sizeof(InDataType) * (N * C * Hi * Wi) +
sizeof(WeiDataType) * (K * C * Y * X) +
sizeof(OutDataType) * (N * K * Ho * Wo);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_btype / 1.E6 / ave_time;
std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
<< std::endl;
if(do_verification)
{
auto ref_conv = ReferenceConvBwdInstance{};
auto ref_invoker = ref_conv.MakeInvoker();
auto ref_argument = ref_conv.MakeArgument(in_n_c_hi_wi_host_result,
wei_k_c_y_x,
out_n_k_ho_wo,
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads,
InElementOp{},
WeiElementOp{},
OutElementOp{});
ref_invoker.Run(ref_argument);
in_device_buf.FromDevice(in_n_c_hi_wi_device_result.mData.data());
return ck::utils::check_err(in_n_c_hi_wi_device_result.mData,
in_n_c_hi_wi_host_result.mData)
? 0
: 1;
}
return 0;
}
example/11_conv2d_bwd_weight/CMakeLists.txt deleted 100644 → 0
View file @ 85978e02
add_example_executable(example_conv2d_bwd_weight_xdl conv2d_bwd_weight_xdl.cpp)
target_link_libraries(example_conv2d_bwd_weight_xdl PRIVATE conv_util)
example/11_conv2d_bwd_weight/README.md deleted 100644 → 0
View file @ 85978e02
# Instructions for ```example_conv2d_bwd_weight_xdl``` Example
## Run ```example_conv2d_bwd_weight_xdl```
```bash
#arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1)
#arg4: is show log (0=no, 1=yes)
#arg5 to 19: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx, split-k
./bin/example_conv2d_bwd_weight_xdl 0 1 5 0 4
```
Result
```
in_n_c_hi_wi: dim 4, lengths {128, 1024, 14, 14}, strides {200704, 1, 14336, 1024}
wei_k_c_y_x: dim 4, lengths {256, 1024, 3, 3}, strides {9216, 1, 3072, 1024}
out_n_k_ho_wo: dim 4, lengths {128, 256, 6, 6}, strides {9216, 1, 1536, 256}
arg.a_grid_desc_kbatch_k0_m_k1_{4, 144, 256, 8}
arg.b_grid_desc_kbatch_k0_n_k1_{4, 144, 9216, 8}
arg.c_grid_desc_m_n_{ 256, 9216}
launch_and_time_kernel: grid_dim {576, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 5 times...
Perf: 0.401084 ms, 54.2112 TFlops, 145.75 GB/s
```
example/11_conv2d_bwd_weight/conv2d_bwd_weight_xdl.cpp deleted 100644 → 0
View file @ 85978e02
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/conv_util.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_conv_backward_weight.hpp"
using InDataType = ck::half_t;
using WeiDataType = ck::half_t;
using OutDataType = ck::half_t;
using AccDataType = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using InLayout = ck::tensor_layout::convolution::NHWC;
using WeiLayout = ck::tensor_layout::convolution::KYXC;
using OutLayout = ck::tensor_layout::convolution::NHWK;
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
// clang-format off
using DeviceConvBwdWeightInstance = ck::tensor_operation::device::
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
InDataType, // InDataType
WeiDataType, // WeiDataType
OutDataType, // OutDataType
AccDataType, // AccDataType
InElementOp, // InElementwiseOperation
WeiElementOp, // WeiElementwiseOperation
OutElementOp, // OutElementwiseOperation
256, // BlockSize
128, // MPerBlock
128, // NPerBlock
4, // K0PerBlock
8, // K1
32, // MPerXdl
32, // NPerXdl
2, // MXdlPerWave
2, // NXdlPerWave
S<1, 4, 16, 4>, // ABlockTransferThreadClusterLengths_K0_M_K1
S<0, 3, 1, 2>, // ABlockTransferThreadClusterArrangeOrder
S<0, 2, 1, 3>, // ABlockTransferSrcAccessOrder
2, // ABlockTransferSrcVectorDim
8, // ABlockTransferSrcScalarPerVector
2, // ABlockTransferDstScalarPerVector_K1
true, // ABlockLdsAddExtraM
S<1, 4, 16, 4>, // BBlockTransferThreadClusterLengths_K0_N_K1
S<0, 3, 1, 2>, // BBlockTransferThreadClusterArrangeOrder
S<0, 2, 1, 3>, // BBlockTransferSrcAccessOrder
2, // BBlockTransferSrcVectorDim
8, // BBlockTransferSrcScalarPerVector
2, // BBlockTransferDstScalarPerVector_K1
true, // BBlockLdsAddExtraN
1, // CShuffleMXdlPerWavePerShuffle
1, // CShuffleNXdlPerWavePerShuffle
S<1, 32, 1, 4>, // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
8>; // CBlockTransferScalarPerVector_NWaveNPerXdl
// clang-format on
using ReferenceConvBwdWeightInstance =
ck::tensor_operation::host::ReferenceConvBwdWeight<InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp>;
int main(int argc, char* argv[])
{
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
int do_log = 0;
int split_k = 4;
// Conv shape
ck::index_t N = 128;
ck::index_t K = 256;
ck::index_t C = 1024;
ck::index_t Y = 3;
ck::index_t X = 3;
ck::index_t Hi = 14;
ck::index_t Wi = 14;
ck::index_t conv_stride_h = 2;
ck::index_t conv_stride_w = 2;
ck::index_t conv_dilation_h = 1;
ck::index_t conv_dilation_w = 1;
ck::index_t in_left_pad_h = 0;
ck::index_t in_left_pad_w = 0;
ck::index_t in_right_pad_h = 0;
ck::index_t in_right_pad_w = 0;
if(argc == 6)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
do_log = std::stoi(argv[4]);
split_k = std::stoi(argv[5]);
}
else if(argc == 21)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
do_log = std::stoi(argv[4]);
split_k = std::stoi(argv[5]);
N = std::stoi(argv[6]);
K = std::stoi(argv[7]);
C = std::stoi(argv[8]);
Y = std::stoi(argv[9]);
X = std::stoi(argv[10]);
Hi = std::stoi(argv[11]);
Wi = std::stoi(argv[12]);
conv_stride_h = std::stoi(argv[13]);
conv_stride_w = std::stoi(argv[14]);
conv_dilation_h = std::stoi(argv[15]);
conv_dilation_w = std::stoi(argv[16]);
in_left_pad_h = std::stoi(argv[17]);
in_left_pad_w = std::stoi(argv[18]);
in_right_pad_h = std::stoi(argv[19]);
in_right_pad_w = std::stoi(argv[20]);
}
else
{
printf("arg1: verification (0=no, 1=yes)\n");
printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
printf("arg3: time kernel (0=n0, 1=yes)\n");
printf("arg4: is show log (0=no, 1=yes)\n");
printf("arg5: split-k \n");
printf("arg6 to 19: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
"RightPx\n");
exit(0);
}
const ck::index_t YEff = (Y - 1) * conv_dilation_h + 1;
const ck::index_t XEff = (X - 1) * conv_dilation_w + 1;
const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - YEff) / conv_stride_h + 1;
const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
const std::vector<ck::index_t> conv_filter_strides{{conv_stride_h, conv_stride_w}};
const std::vector<ck::index_t> conv_filter_dilations{{conv_dilation_h, conv_dilation_w}};
const std::vector<ck::index_t> input_left_pads{{in_left_pad_h, in_left_pad_w}};
const std::vector<ck::index_t> input_right_pads{{in_right_pad_h, in_right_pad_w}};
// tensor layout
auto f_host_tensor_descriptor = [](std::size_t N_,
std::size_t C_,
std::size_t H,
std::size_t W,
auto layout) {
if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCHW>::value ||
ck::is_same<decltype(layout), ck::tensor_layout::convolution::KCYX>::value ||
ck::is_same<decltype(layout), ck::tensor_layout::convolution::NKHW>::value)
{
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
std::vector<std::size_t>({C_ * H * W, H * W, W, 1}));
}
else if constexpr(ck::is_same<decltype(layout),
ck::tensor_layout::convolution::NHWC>::value ||
ck::is_same<decltype(layout),
ck::tensor_layout::convolution::KYXC>::value ||
ck::is_same<decltype(layout),
ck::tensor_layout::convolution::NHWK>::value)
{
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
std::vector<std::size_t>({C_ * H * W, 1, W * C_, C_}));
}
};
Tensor<InDataType> in_n_c_hi_wi(f_host_tensor_descriptor(N, C, Hi, Wi, InLayout{}));
Tensor<WeiDataType> wei_k_c_y_x_host_result(f_host_tensor_descriptor(K, C, Y, X, WeiLayout{}));
Tensor<WeiDataType> wei_k_c_y_x_device_result(
f_host_tensor_descriptor(K, C, Y, X, WeiLayout{}));
Tensor<OutDataType> out_n_k_ho_wo(f_host_tensor_descriptor(N, K, Ho, Wo, OutLayout{}));
std::cout << "in_n_c_hi_wi: " << in_n_c_hi_wi.mDesc << std::endl;
std::cout << "wei_k_c_y_x: " << wei_k_c_y_x_host_result.mDesc << std::endl;
std::cout << "out_n_k_ho_wo: " << out_n_k_ho_wo.mDesc << std::endl;
switch(init_method)
{
case 0: break;
case 1:
in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5});
out_n_k_ho_wo.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
break;
default:
in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_1<InDataType>{1});
out_n_k_ho_wo.GenerateTensorValue(GeneratorTensor_1<OutDataType>{1});
}
wei_k_c_y_x_device_result.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{0});
DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_hi_wi.mDesc.GetElementSpace());
DeviceMem wei_device_buf(sizeof(WeiDataType) *
wei_k_c_y_x_device_result.mDesc.GetElementSpace());
DeviceMem out_device_buf(sizeof(OutDataType) * out_n_k_ho_wo.mDesc.GetElementSpace());
in_device_buf.ToDevice(in_n_c_hi_wi.mData.data());
out_device_buf.ToDevice(out_n_k_ho_wo.mData.data());
wei_device_buf.ToDevice(wei_k_c_y_x_device_result.mData.data());
// do GEMM
auto conv = DeviceConvBwdWeightInstance{};
auto invoker = conv.MakeInvoker();
auto argument = conv.MakeArgument(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
N,
K,
C,
std::vector<ck::index_t>{{Hi, Wi}},
std::vector<ck::index_t>{{Y, X}},
std::vector<ck::index_t>{{Ho, Wo}},
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads,
InElementOp{},
WeiElementOp{},
OutElementOp{},
split_k);
if(!conv.IsSupportedArgument(argument))
{
std::cout << "wrong! device_conv with the specified compilation parameters does "
"not support this Conv problem"
<< std::endl;
return 1;
}
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * N * K * Ho * Wo * C * Y * X;
std::size_t num_btype = sizeof(InDataType) * (N * C * Hi * Wi) +
sizeof(WeiDataType) * (K * C * Y * X) +
sizeof(OutDataType) * (N * K * Ho * Wo);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_btype / 1.E6 / ave_time;
std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
<< std::endl;
if(do_verification)
{
auto ref_conv = ReferenceConvBwdWeightInstance{};
auto ref_invoker = ref_conv.MakeInvoker();
auto ref_argument = ref_conv.MakeArgument(in_n_c_hi_wi,
wei_k_c_y_x_host_result,
out_n_k_ho_wo,
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads,
InElementOp{},
WeiElementOp{},
OutElementOp{});
ref_invoker.Run(ref_argument);
wei_device_buf.FromDevice(wei_k_c_y_x_device_result.mData.data());
if(do_log)
{
LogRangeAsType<float>(std::cout << "out: ", out_n_k_ho_wo.mData, ",") << std::endl;
LogRangeAsType<float>(std::cout << "in : ", in_n_c_hi_wi.mData, ",") << std::endl;
LogRangeAsType<float>(
std::cout << "wei_device(after): ", wei_k_c_y_x_device_result.mData, ",")
<< std::endl;
LogRangeAsType<float>(std::cout << "wei_host : ", wei_k_c_y_x_host_result.mData, ",")
<< std::endl;
}
return ck::utils::check_err(wei_k_c_y_x_device_result.mData, wei_k_c_y_x_host_result.mData)
? 0
: 1;
}
return 0;
}
example/12_reduce/reduce_blockwise.cpp
View file @ 500fa995
...@@ -13,11 +13,11 @@ ...@@ -13,11 +13,11 @@
#include "ck/tensor_operation/gpu/device/device_reduce_multiblock.hpp" #include "ck/tensor_operation/gpu/device/device_reduce_multiblock.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/host_tensor/host_common_util.hpp" #include "ck/library/utility/host_common_util.hpp"
#include "ck/library/host_tensor/host_reduction.hpp" #include "ck/library/utility/host_reduction.hpp"
using namespace ck; using namespace ck;
using namespace ck::tensor_operation::device; using namespace ck::tensor_operation::device;
...@@ -230,13 +230,13 @@ int main(int argc, char* argv[]) ...@@ -230,13 +230,13 @@ int main(int argc, char* argv[])
} }
if(beta != 0.0f) if(beta != 0.0f)
for(size_t i = 0; i < out_ref.mDesc.GetElementSpace(); i++) for(size_t i = 0; i < out_ref.mDesc.GetElementSpaceSize(); i++)
out.mData[i] = out_ref.mData[i]; out.mData[i] = out_ref.mData[i];
}; };
// these buffers are usually provided by the user application // these buffers are usually provided by the user application
DeviceMem in_dev(sizeof(InDataType) * in.mDesc.GetElementSpace()); DeviceMem in_dev(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
DeviceMem out_dev(sizeof(OutDataType) * out.mDesc.GetElementSpace()); DeviceMem out_dev(sizeof(OutDataType) * out.mDesc.GetElementSpaceSize());
in_dev.ToDevice(in.mData.data()); in_dev.ToDevice(in.mData.data());
......
example/12_reduce/reduce_blockwise_two_call.cpp
View file @ 500fa995
...@@ -14,11 +14,11 @@ ...@@ -14,11 +14,11 @@
#include "ck/tensor_operation/gpu/device/device_reduce_multiblock.hpp" #include "ck/tensor_operation/gpu/device/device_reduce_multiblock.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/host_tensor/host_common_util.hpp" #include "ck/library/utility/host_common_util.hpp"
#include "ck/library/host_tensor/host_reduction.hpp" #include "ck/library/utility/host_reduction.hpp"
using namespace ck; using namespace ck;
using namespace ck::tensor_operation::device; using namespace ck::tensor_operation::device;
...@@ -171,13 +171,13 @@ int main(int argc, char* argv[]) ...@@ -171,13 +171,13 @@ int main(int argc, char* argv[])
} }
if(beta != 0.0f) if(beta != 0.0f)
for(size_t i = 0; i < out_ref.mDesc.GetElementSpace(); i++) for(size_t i = 0; i < out_ref.mDesc.GetElementSpaceSize(); i++)
out.mData[i] = out_ref.mData[i]; out.mData[i] = out_ref.mData[i];
}; };
DeviceMem in_1_dev(sizeof(InOutDataType) * in_1.mDesc.GetElementSpace()); DeviceMem in_1_dev(sizeof(InOutDataType) * in_1.mDesc.GetElementSpaceSize());
DeviceMem in_2_dev(sizeof(InOutDataType) * in_2.mDesc.GetElementSpace()); DeviceMem in_2_dev(sizeof(InOutDataType) * in_2.mDesc.GetElementSpaceSize());
DeviceMem out_dev(sizeof(InOutDataType) * out.mDesc.GetElementSpace()); DeviceMem out_dev(sizeof(InOutDataType) * out.mDesc.GetElementSpaceSize());
in_1_dev.ToDevice(in_1.mData.data()); in_1_dev.ToDevice(in_1.mData.data());
......
example/13_pool2d_fwd/pool2d_fwd_common.hpp
View file @ 500fa995
...@@ -13,9 +13,9 @@ ...@@ -13,9 +13,9 @@
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
template <typename InDataType, template <typename InDataType,
typename OutDataType, typename OutDataType,
...@@ -204,10 +204,11 @@ bool pool_test(bool do_verification, ...@@ -204,10 +204,11 @@ bool pool_test(bool do_verification,
default: in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-5.0, 5.0}); default: in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-5.0, 5.0});
} }
DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_hi_wi.mDesc.GetElementSpace()); DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_hi_wi.mDesc.GetElementSpaceSize());
DeviceMem out_device_buf(sizeof(OutDataType) * out_n_c_ho_wo_device.mDesc.GetElementSpace()); DeviceMem out_device_buf(sizeof(OutDataType) *
out_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
DeviceMem out_indices_device_buf(sizeof(IndexDataType) * DeviceMem out_indices_device_buf(sizeof(IndexDataType) *
out_indices_n_c_ho_wo_device.mDesc.GetElementSpace()); out_indices_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
in_device_buf.ToDevice(in_n_c_hi_wi.mData.data()); in_device_buf.ToDevice(in_n_c_hi_wi.mData.data());
......
example/14_gemm_xdl_requant_relu_requant/gemm_xdl_requant_relu_requant_int8.cpp
View file @ 500fa995
...@@ -12,9 +12,9 @@ ...@@ -12,9 +12,9 @@
#include "ck/tensor_operation/gpu/device/device_gemm_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/device_gemm_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
...@@ -190,9 +190,9 @@ int main(int argc, char* argv[]) ...@@ -190,9 +190,9 @@ int main(int argc, char* argv[])
b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5}); b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
} }
DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace()); DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace()); DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpace()); DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
a_m_k_device_buf.ToDevice(a_m_k.mData.data()); a_m_k_device_buf.ToDevice(a_m_k.mData.data());
b_k_n_device_buf.ToDevice(b_k_n.mData.data()); b_k_n_device_buf.ToDevice(b_k_n.mData.data());
......
example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp
View file @ 500fa995
...@@ -13,9 +13,9 @@ ...@@ -13,9 +13,9 @@
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
...@@ -38,6 +38,7 @@ using EDataType = F16; ...@@ -38,6 +38,7 @@ using EDataType = F16;
using ALayout = Row; using ALayout = Row;
using BLayout = Col; using BLayout = Col;
using DsLayout = ck::Tuple<>;
using ELayout = Row; using ELayout = Row;
using AElementOp = PassThrough; using AElementOp = PassThrough;
...@@ -46,13 +47,13 @@ using CDEElementOp = PassThrough; ...@@ -46,13 +47,13 @@ using CDEElementOp = PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemmXdl using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemm_Xdl
// clang-format off // clang-format off
//######| ALayout| BLayout| ELayout| 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| //######| ALayout| BLayout| DsLayout| ELayout| 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|
//######| | | | 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| //######| | | | | 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| //######| | | | | | | | | | | 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|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ALayout, BLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>; < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>;
// clang-format on // clang-format on
using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType, using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
...@@ -178,12 +179,12 @@ int main(int argc, char* argv[]) ...@@ -178,12 +179,12 @@ int main(int argc, char* argv[])
for(std::size_t i = 0; i < gemm_descs.size(); i++) for(std::size_t i = 0; i < gemm_descs.size(); i++)
{ {
a_tensors_device.emplace_back( a_tensors_device.emplace_back(std::make_unique<DeviceMem>(
std::make_unique<DeviceMem>(sizeof(ADataType) * a_tensors[i].mDesc.GetElementSpace())); sizeof(ADataType) * a_tensors[i].mDesc.GetElementSpaceSize()));
b_tensors_device.emplace_back( b_tensors_device.emplace_back(std::make_unique<DeviceMem>(
std::make_unique<DeviceMem>(sizeof(BDataType) * b_tensors[i].mDesc.GetElementSpace())); sizeof(BDataType) * b_tensors[i].mDesc.GetElementSpaceSize()));
c_tensors_device.emplace_back(std::make_unique<DeviceMem>( c_tensors_device.emplace_back(std::make_unique<DeviceMem>(
sizeof(EDataType) * c_device_tensors[i].mDesc.GetElementSpace())); sizeof(EDataType) * c_device_tensors[i].mDesc.GetElementSpaceSize()));
a_tensors_device[i]->ToDevice(a_tensors[i].mData.data()); a_tensors_device[i]->ToDevice(a_tensors[i].mData.data());
b_tensors_device[i]->ToDevice(b_tensors[i].mData.data()); b_tensors_device[i]->ToDevice(b_tensors[i].mData.data());
......
example/16_gemm_reduce/gemm_reduce_xdl_max_fp16.cpp
View file @ 500fa995
...@@ -12,9 +12,9 @@ ...@@ -12,9 +12,9 @@
#include "ck/tensor_operation/gpu/device/device_gemm_reduce_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/device_gemm_reduce_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
...@@ -173,11 +173,11 @@ int main(int argc, char* argv[]) ...@@ -173,11 +173,11 @@ int main(int argc, char* argv[])
break; break;
} }
DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace()); DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace()); DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
DeviceMem c_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpace()); DeviceMem c_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
DeviceMem reduce_device_buf(sizeof(ReduceDataType) * DeviceMem reduce_device_buf(sizeof(ReduceDataType) *
reduce_m_device_result.mDesc.GetElementSpace()); reduce_m_device_result.mDesc.GetElementSpaceSize());
a_device_buf.ToDevice(a_m_k.mData.data()); a_device_buf.ToDevice(a_m_k.mData.data());
b_device_buf.ToDevice(b_k_n.mData.data()); b_device_buf.ToDevice(b_k_n.mData.data());
......
example/16_gemm_reduce/gemm_reduce_xdl_mean_squaremean_fp16.cpp
View file @ 500fa995
...@@ -14,9 +14,9 @@ ...@@ -14,9 +14,9 @@
#include "ck/utility/reduction_operator.hpp" #include "ck/utility/reduction_operator.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
...@@ -188,13 +188,13 @@ int main(int argc, char* argv[]) ...@@ -188,13 +188,13 @@ int main(int argc, char* argv[])
break; break;
} }
DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace()); DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace()); DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
DeviceMem c_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpace()); DeviceMem c_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
DeviceMem reduce0_device_buf(sizeof(ReduceDataType) * DeviceMem reduce0_device_buf(sizeof(ReduceDataType) *
reduce0_m_device_result.mDesc.GetElementSpace()); reduce0_m_device_result.mDesc.GetElementSpaceSize());
DeviceMem reduce1_device_buf(sizeof(ReduceDataType) * DeviceMem reduce1_device_buf(sizeof(ReduceDataType) *
reduce1_m_device_result.mDesc.GetElementSpace()); reduce1_m_device_result.mDesc.GetElementSpaceSize());
a_device_buf.ToDevice(a_m_k.mData.data()); a_device_buf.ToDevice(a_m_k.mData.data());
b_device_buf.ToDevice(b_k_n.mData.data()); b_device_buf.ToDevice(b_k_n.mData.data());
......
example/17_convnd_bwd_data/CMakeLists.txt 0 → 100644
View file @ 500fa995
add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
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