mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-04-19 22:39:03 +00:00
40b59a63cc
* start adding navi21 GEMM * navi_gemm_km_kn_mn_fp32 compiles and passes one test. * rename variables and functions in gridwise_gemm_dlops_v1r3 * add other 3 layouts; format instance * adding more tuning parameters add tuning parameters for other 3 layouts * add gemm_dlops_f16 * tmp * add dependence of DeviceGemm::IsSupportedArg() on arch * minor changes * minor changes * minor changes * minor changes * minor changes * minor changes * minor changes * push gemm_dlops into profiler * minor changes * if using xdl or dlops is moved into profiler_gemm_impl * minor changes * minor changes * remove is_xdl from profile_gemm_impl * make IsSupportedArg dependent on arch for other device_gemm * minor changes * minor changes * fix a bug in f_generate_tensor_value * add 64x64x64 for gemm_dlops_int8 * add 64x64x64 for gemm_dlops_int8 * comment out 3 layouts in gemm_dlops_int8; add 32x32x32 for gemm_dlops_int8; init A values to 1 * fix * start fixing tuning parameters * monir * minor changes * minor changes * minor changes * fixing * adding example * adding example * adding example * add gemm fp32 example * clean up * use 128x128x16 as MNK tile in navi21 gemm example * bug fix * fix test * use new block c tile * clean * fix build Co-authored-by: Chao Liu <chao.liu2@amd.com> Co-authored-by: shaojiewang <wsjmessi@163.com>
211 lines
9.5 KiB
C++
211 lines
9.5 KiB
C++
#include <iostream>
|
|
#include <numeric>
|
|
#include <initializer_list>
|
|
#include <cstdlib>
|
|
#include <stdlib.h>
|
|
#include <half.hpp>
|
|
|
|
#include "check_err.hpp"
|
|
#include "config.hpp"
|
|
#include "device.hpp"
|
|
#include "host_tensor.hpp"
|
|
#include "host_tensor_generator.hpp"
|
|
#include "device_tensor.hpp"
|
|
#include "device_gemm_dl.hpp"
|
|
#include "element_wise_operation.hpp"
|
|
#include "reference_gemm.hpp"
|
|
#include "gemm_specialization.hpp"
|
|
|
|
template <ck::index_t... Is>
|
|
using S = ck::Sequence<Is...>;
|
|
|
|
using F32 = float;
|
|
|
|
using Row = ck::tensor_layout::gemm::RowMajor;
|
|
using Col = ck::tensor_layout::gemm::ColumnMajor;
|
|
|
|
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
|
|
|
using ADataType = float;
|
|
using BDataType = float;
|
|
using CDataType = float;
|
|
using AccDataType = float;
|
|
|
|
using ALayout = Col;
|
|
using BLayout = Row;
|
|
using CLayout = Row;
|
|
|
|
using AElementOp = ck::tensor_operation::element_wise::PassThrough;
|
|
using BElementOp = ck::tensor_operation::element_wise::PassThrough;
|
|
using CElementOp = ck::tensor_operation::element_wise::PassThrough;
|
|
|
|
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
|
|
|
|
// clang-format off
|
|
using DeviceGemmInstance = ck::tensor_operation::device::
|
|
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
|
|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
|
|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
|
|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
|
DeviceGemmDl< F32, F32, F32, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 16, 1, 4, 4, 1, S<8, 2>, S<8, 2>, S<2, 1, 4, 1>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<2, 1, 4, 1>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 1, 2, 3, 4, 5>, 5, 4>;
|
|
// clang-format on
|
|
|
|
using ReferenceGemmInstance = ck::tensor_operation::host::
|
|
ReferenceGemm<ADataType, BDataType, CDataType, AElementOp, BElementOp, CElementOp>;
|
|
|
|
int main(int argc, char* argv[])
|
|
{
|
|
bool do_verification = true;
|
|
int init_method = 1;
|
|
bool time_kernel = false;
|
|
|
|
// GEMM shape
|
|
ck::index_t M = 3840;
|
|
ck::index_t N = 4096;
|
|
ck::index_t K = 4096;
|
|
|
|
ck::index_t StrideA = 4096;
|
|
ck::index_t StrideB = 4096;
|
|
ck::index_t StrideC = 4096;
|
|
|
|
if(argc == 1)
|
|
{
|
|
// do nothing
|
|
}
|
|
else if(argc == 4)
|
|
{
|
|
do_verification = std::stoi(argv[1]);
|
|
init_method = std::stoi(argv[2]);
|
|
time_kernel = std::stoi(argv[3]);
|
|
}
|
|
else if(argc == 10)
|
|
{
|
|
do_verification = std::stoi(argv[1]);
|
|
init_method = std::stoi(argv[2]);
|
|
time_kernel = std::stoi(argv[3]);
|
|
|
|
M = std::stoi(argv[4]);
|
|
N = std::stoi(argv[5]);
|
|
K = std::stoi(argv[6]);
|
|
|
|
StrideA = std::stoi(argv[7]);
|
|
StrideB = std::stoi(argv[8]);
|
|
StrideC = std::stoi(argv[9]);
|
|
}
|
|
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 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
|
|
exit(1);
|
|
}
|
|
|
|
auto f_host_tensor_descriptor =
|
|
[](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
|
|
if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
|
|
{
|
|
return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
|
|
std::vector<std::size_t>({stride, 1}));
|
|
}
|
|
else
|
|
{
|
|
return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
|
|
std::vector<std::size_t>({1, stride}));
|
|
}
|
|
};
|
|
|
|
Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
|
|
Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
|
|
Tensor<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
|
|
Tensor<CDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
|
|
|
|
std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
|
|
std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
|
|
std::cout << "c_m_n: " << c_m_n_host_result.mDesc << std::endl;
|
|
|
|
switch(init_method)
|
|
{
|
|
case 0: break;
|
|
case 1:
|
|
a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
|
|
b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
|
|
break;
|
|
case 2:
|
|
a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
|
|
b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
|
|
break;
|
|
default:
|
|
a_m_k.GenerateTensorValue(GeneratorTensor_Sequential<0>{});
|
|
b_k_n.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
|
|
}
|
|
|
|
DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
|
|
DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
|
|
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpace());
|
|
|
|
a_m_k_device_buf.ToDevice(a_m_k.mData.data());
|
|
b_k_n_device_buf.ToDevice(b_k_n.mData.data());
|
|
|
|
auto a_element_op = AElementOp{};
|
|
auto b_element_op = BElementOp{};
|
|
auto c_element_op = CElementOp{};
|
|
|
|
// do GEMM
|
|
auto gemm = DeviceGemmInstance{};
|
|
auto invoker = gemm.MakeInvoker();
|
|
auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
|
|
static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
|
|
static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
|
|
M,
|
|
N,
|
|
K,
|
|
StrideA,
|
|
StrideB,
|
|
StrideC,
|
|
a_element_op,
|
|
b_element_op,
|
|
c_element_op);
|
|
|
|
if(!gemm.IsSupportedArgument(argument))
|
|
{
|
|
std::cout << "wrong! device_gemm with the specified compilation parameters does "
|
|
"not support this GEMM problem"
|
|
<< std::endl;
|
|
|
|
return 0;
|
|
}
|
|
|
|
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
|
|
|
|
std::size_t flop = std::size_t(2) * M * N * K;
|
|
std::size_t num_btype =
|
|
sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(CDataType) * M * N;
|
|
|
|
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, "
|
|
<< gemm.GetTypeString() << std::endl;
|
|
|
|
c_m_n_device_buf.FromDevice(c_m_n_device_result.mData.data());
|
|
|
|
bool pass = true;
|
|
|
|
if(do_verification)
|
|
{
|
|
auto ref_gemm = ReferenceGemmInstance{};
|
|
auto ref_invoker = ref_gemm.MakeInvoker();
|
|
|
|
auto ref_argument = ref_gemm.MakeArgument(
|
|
a_m_k, b_k_n, c_m_n_host_result, a_element_op, b_element_op, c_element_op);
|
|
|
|
ref_invoker.Run(ref_argument);
|
|
|
|
pass = ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData);
|
|
}
|
|
|
|
return pass ? 0 : 1;
|
|
}
|