Skip to content

GitLab

Commit ed068043 authored by Jing Zhang's avatar Jing Zhang
Browse files

merged develop

parents 41852668 e823d518
Expand all Show whitespace changes
Inline Side-by-side
Showing with 7622 additions and 0 deletions
device_operation/device_gemm_xdl_instance_f16_f16_f16_mk_nk_mn.cpp 0 → 100644
View file @ ed068043
#include <stdlib.h>
#include "config.hpp"
#include "device_gemm_xdl.hpp"
#include "device_gemm_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_gemm_instance {
using F16 = ck::half_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
using device_gemm_xdl_instance_f16_f16_f16_mk_nk_mn = std::tuple<
// clang-format off
//##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| ABlockLds| BBlockLds|
//##########| Type| Type| Type| Type| | | | Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| SrcDstVectorDim| DstScalar| AddExtraM| AddExtraN|
//##########| | | | | | | | | | | | | | | Wave| Wave| Lengths_K0_N_K1| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| Lengths_K0_N_K1| Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerVector| | |
//##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 2, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 2, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 4, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 8>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 4, 8>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 2, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 2, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 2, 8>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 2, 8>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 4, 8>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 4, 8>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 2, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 1, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 1, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 2, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 1, 8>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 1, 8>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 4, 8>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 2, 8>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>,
DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 2, 8>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 4, 8>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>
// clang-format on
>;
template <>
void add_device_gemm_instance<F16, F16, F16, Row, Col, Row>(
std::vector<DeviceGemmPtr>& device_op_instances)
{
using DeviceGemms = device_gemm_instance::device_gemm_xdl_instance_f16_f16_f16_mk_nk_mn;
const auto device_gemms = DeviceGemms{};
ck::static_for<0, std::tuple_size_v<DeviceGemms>, 1>{}([&](auto i) {
using Gemm = remove_cvref_t<decltype(std::get<i>(device_gemms))>;
auto gemm = Gemm{};
device_op_instances.push_back(std::make_unique<Gemm>(gemm));
});
}
} // namespace device_gemm_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
device_operation/device_gemm_xdl_instance_f32_f32_f32_km_kn_mn.cpp 0 → 100644
View file @ ed068043
#include <stdlib.h>
#include "config.hpp"
#include "device_gemm_xdl.hpp"
#include "device_gemm_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_gemm_instance {
using F16 = ck::half_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
using device_gemm_xdl_instance_f32_f32_f32_km_kn_mn = std::tuple<
// clang-format off
//##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| ABlockLds| BBlockLds|
//##########| Type| Type| Type| Type| | | | Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| SrcDstVectorDim| DstScalar| AddExtraM| AddExtraN|
//##########| | | | | | | | | | | | | | | Wave| Wave| Lengths_K0_N_K1| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| Lengths_K0_N_K1| Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerVector| | |
//##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmXdl< F32, F32, F32, F32, Col, Row, Row, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Row, Row, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, S<1, 4, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Row, Row, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, S<1, 4, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Row, Row, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Row, Row, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, S<1, 2, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Row, Row, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<1, 2, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, S<1, 4, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Row, Row, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, S<1, 1, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Row, Row, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<1, 1, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 7, 1, true, true>
// clang-format on
>;
template <>
void add_device_gemm_instance<F32, F32, F32, Col, Row, Row>(
std::vector<DeviceGemmPtr>& device_op_instances)
{
using DeviceGemms = device_gemm_instance::device_gemm_xdl_instance_f32_f32_f32_km_kn_mn;
const auto device_gemms = DeviceGemms{};
ck::static_for<0, std::tuple_size_v<DeviceGemms>, 1>{}([&](auto i) {
using Gemm = remove_cvref_t<decltype(std::get<i>(device_gemms))>;
auto gemm = Gemm{};
device_op_instances.push_back(std::make_unique<Gemm>(gemm));
});
}
} // namespace device_gemm_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
device_operation/device_gemm_xdl_instance_f32_f32_f32_km_nk_mn.cpp 0 → 100644
View file @ ed068043
#include <stdlib.h>
#include "config.hpp"
#include "device_gemm_xdl.hpp"
#include "device_gemm_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_gemm_instance {
using F16 = ck::half_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
using device_gemm_xdl_instance_f32_f32_f32_km_nk_mn = std::tuple<
// clang-format off
//##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| ABlockLds| BBlockLds|
//##########| Type| Type| Type| Type| | | | Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| SrcDstVectorDim| DstScalar| AddExtraM| AddExtraN|
//##########| | | | | | | | | | | | | | | Wave| Wave| Lengths_K0_N_K1| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| Lengths_K0_N_K1| Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerVector| | |
//##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmXdl< F32, F32, F32, F32, Col, Col, Row, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Col, Row, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, S<1, 4, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Col, Row, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, S<1, 4, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Col, Row, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Col, Row, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, S<1, 2, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Col, Row, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<1, 2, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, S<1, 4, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Col, Row, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, S<1, 1, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Col, Col, Row, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<1, 1, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>
// clang-format on
>;
template <>
void add_device_gemm_instance<F32, F32, F32, Col, Col, Row>(
std::vector<DeviceGemmPtr>& device_op_instances)
{
using DeviceGemms = device_gemm_instance::device_gemm_xdl_instance_f32_f32_f32_km_nk_mn;
const auto device_gemms = DeviceGemms{};
ck::static_for<0, std::tuple_size_v<DeviceGemms>, 1>{}([&](auto i) {
using Gemm = remove_cvref_t<decltype(std::get<i>(device_gemms))>;
auto gemm = Gemm{};
device_op_instances.push_back(std::make_unique<Gemm>(gemm));
});
}
} // namespace device_gemm_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
device_operation/device_gemm_xdl_instance_f32_f32_f32_mk_kn_mn.cpp 0 → 100644
View file @ ed068043
#include <stdlib.h>
#include "config.hpp"
#include "device_gemm_xdl.hpp"
#include "device_gemm_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_gemm_instance {
using F16 = ck::half_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
using device_gemm_xdl_instance_f32_f32_f32_mk_kn_mn = std::tuple<
// clang-format off
//##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| ABlockLds| BBlockLds|
//##########| Type| Type| Type| Type| | | | Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| SrcDstVectorDim| DstScalar| AddExtraM| AddExtraN|
//##########| | | | | | | | | | | | | | | Wave| Wave| Lengths_K0_N_K1| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| Lengths_K0_N_K1| Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerVector| | |
//##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmXdl< F32, F32, F32, F32, Row, Row, Row, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Row, Row, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 4, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Row, Row, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 4, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Row, Row, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Row, Row, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 2, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Row, Row, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<1, 2, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 4, 4>, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Row, Row, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 1, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Row, Row, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<1, 1, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 2, 4>, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 7, 1, true, true>
// clang-format on
>;
template <>
void add_device_gemm_instance<F32, F32, F32, Row, Row, Row>(
std::vector<DeviceGemmPtr>& device_op_instances)
{
using DeviceGemms = device_gemm_instance::device_gemm_xdl_instance_f32_f32_f32_mk_kn_mn;
const auto device_gemms = DeviceGemms{};
ck::static_for<0, std::tuple_size_v<DeviceGemms>, 1>{}([&](auto i) {
using Gemm = remove_cvref_t<decltype(std::get<i>(device_gemms))>;
auto gemm = Gemm{};
device_op_instances.push_back(std::make_unique<Gemm>(gemm));
});
}
} // namespace device_gemm_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
device_operation/device_gemm_xdl_instance_f32_f32_f32_mk_nk_mn.cpp 0 → 100644
View file @ ed068043
#include <stdlib.h>
#include "config.hpp"
#include "device_gemm_xdl.hpp"
#include "device_gemm_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_gemm_instance {
using F16 = ck::half_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
using device_gemm_xdl_instance_f32_f32_f32_mk_nk_mn = std::tuple<
// clang-format off
//##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| ABlockLds| BBlockLds|
//##########| Type| Type| Type| Type| | | | Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| SrcDstVectorDim| DstScalar| AddExtraM| AddExtraN|
//##########| | | | | | | | | | | | | | | Wave| Wave| Lengths_K0_N_K1| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| Lengths_K0_N_K1| Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerVector| | |
//##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 4, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 4, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 2, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<1, 2, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 4, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 4>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 4, 4>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 1, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<1, 1, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 1, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<1, 1, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 4, 4>, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 4>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 2, 4>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>,
DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<1, 2, 4>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 4, 4>, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>
// clang-format on
>;
template <>
void add_device_gemm_instance<F32, F32, F32, Row, Col, Row>(
std::vector<DeviceGemmPtr>& device_op_instances)
{
using DeviceGemms = device_gemm_instance::device_gemm_xdl_instance_f32_f32_f32_mk_nk_mn;
const auto device_gemms = DeviceGemms{};
ck::static_for<0, std::tuple_size_v<DeviceGemms>, 1>{}([&](auto i) {
using Gemm = remove_cvref_t<decltype(std::get<i>(device_gemms))>;
auto gemm = Gemm{};
device_op_instances.push_back(std::make_unique<Gemm>(gemm));
});
}
} // namespace device_gemm_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
device_operation/include/device_base.hpp 0 → 100644
View file @ ed068043
#ifndef DEVICE_BASE_HPP
#define DEVICE_BASE_HPP
namespace ck {
namespace tensor_operation {
namespace device {
struct BaseArgument
{
BaseArgument() = default;
BaseArgument(const BaseArgument&) = default;
BaseArgument& operator=(const BaseArgument&) = default;
virtual ~BaseArgument() {}
};
struct BaseInvoker
{
BaseInvoker() = default;
BaseInvoker(const BaseInvoker&) = default;
BaseInvoker& operator=(const BaseInvoker&) = default;
virtual float Run(const BaseArgument*, int = 1) = 0;
virtual ~BaseInvoker() {}
};
struct BaseOperator
{
BaseOperator() = default;
BaseOperator(const BaseOperator&) = default;
BaseOperator& operator=(const BaseOperator&) = default;
virtual bool IsSupportedArgument(const BaseArgument*) = 0;
virtual ~BaseOperator() {}
};
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
device_operation/include/device_conv.hpp 0 → 100644
View file @ ed068043
#ifndef DEVICE_CONV_HPP
#define DEVICE_CONV_HPP
#include <iostream>
#include "device_base.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
struct DeviceConvFwd : public BaseOperator
{
virtual std::unique_ptr<BaseArgument>
MakeArgumentPointer(const void* p_in,
const void* p_wei,
void* p_out,
ck::index_t N,
ck::index_t K,
ck::index_t C,
std::vector<ck::index_t> input_spatial_lengths,
std::vector<ck::index_t> filter_spatial_lengths,
std::vector<ck::index_t> output_spatial_lengths,
std::vector<ck::index_t> conv_filter_strides,
std::vector<ck::index_t> conv_filter_dilations,
std::vector<ck::index_t> input_left_pads,
std::vector<ck::index_t> input_right_pads) = 0;
virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
};
struct DeviceConvBwd : public BaseOperator
{
virtual std::unique_ptr<BaseArgument>
MakeArgumentPointer(void* p_in,
const void* p_wei,
const void* p_out,
ck::index_t N,
ck::index_t K,
ck::index_t C,
std::vector<ck::index_t> input_spatial_lengths,
std::vector<ck::index_t> filter_spatial_lengths,
std::vector<ck::index_t> output_spatial_lengths,
std::vector<ck::index_t> conv_filter_strides,
std::vector<ck::index_t> conv_filter_dilations,
std::vector<ck::index_t> input_left_pads,
std::vector<ck::index_t> input_right_pads) = 0;
virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
};
struct DeviceConvWrw : public BaseOperator
{
virtual std::unique_ptr<BaseArgument>
MakeArgumentPointer(const void* p_in,
void* p_wei,
const void* p_out,
ck::index_t N,
ck::index_t K,
ck::index_t C,
std::vector<ck::index_t> input_spatial_lengths,
std::vector<ck::index_t> filter_spatial_lengths,
std::vector<ck::index_t> output_spatial_lengths,
std::vector<ck::index_t> conv_filter_strides,
std::vector<ck::index_t> conv_filter_dilations,
std::vector<ck::index_t> input_left_pads,
std::vector<ck::index_t> input_right_pads) = 0;
virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
};
using DeviceConvFwdPtr = std::unique_ptr<DeviceConvFwd>;
using DeviceConvBwdPtr = std::unique_ptr<DeviceConvBwd>;
using DeviceConvWrwPtr = std::unique_ptr<DeviceConvWrw>;
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
device_operation/include/device_conv_fwd_xdl.hpp 0 → 100644
View file @ ed068043
#ifndef DEVICE_CONV_FWD_XDL_HPP
#define DEVICE_CONV_FWD_XDL_HPP
#include <iostream>
#include "device.hpp"
#include "device_base.hpp"
#include "device_conv.hpp"
#include "common_header.hpp"
#include "tensor_layout.hpp"
#include "tensor_descriptor.hpp"
#include "tensor_descriptor_helper.hpp"
#include "gridwise_gemm_xdlops_v2r3.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
template <ck::index_t NDimSpatial,
typename InDataType,
typename WeiDataType,
typename OutDataType,
typename AccDataType,
typename InLayout,
typename WeiLayout,
typename OutLayout,
ck::index_t BlockSize,
ck::index_t MPerBlock,
ck::index_t NPerBlock,
ck::index_t K0PerBlock,
ck::index_t K1,
ck::index_t MPerXDL,
ck::index_t NPerXDL,
ck::index_t MXdlPerWave,
ck::index_t NXdlPerWave,
typename ABlockTransferThreadSliceLengths_K0_M_K1,
typename ABlockTransferThreadClusterLengths_K0_M_K1,
typename ABlockTransferThreadClusterArrangeOrder,
typename ABlockTransferSrcAccessOrder,
ck::index_t ABlockTransferSrcVectorDim,
ck::index_t ABlockTransferSrcScalarPerVector,
ck::index_t ABlockTransferDstScalarPerVector_K1,
typename BBlockTransferThreadSliceLengths_K0_N_K1,
typename BBlockTransferThreadClusterLengths_K0_N_K1,
typename BBlockTransferThreadClusterArrangeOrder,
typename BBlockTransferSrcAccessOrder,
ck::index_t BBlockTransferSrcVectorDim,
ck::index_t BBlockTransferSrcScalarPerVector,
ck::index_t BBlockTransferDstScalarPerVector_K1,
ck::index_t CThreadTransferSrcDstVectorDim,
ck::index_t CThreadTransferDstScalarPerVector,
bool ABlockLdsAddExtraM,
bool BBlockLdsAddExtraN>
struct DeviceConvFwdXdl;
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
device_operation/include/device_conv_instance.hpp 0 → 100644
View file @ ed068043
#ifndef DEVICE_CONV_INSTANTCE_HPP
#define DEVICE_CONV_INSTANTCE_HPP
#include "device_conv.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv_instance {
template <ck::index_t NDimSpatial,
typename InDataType,
typename WeiDataType,
typename OutDataType,
typename InLayout,
typename WeiLayout,
typename OutLayout>
void add_device_conv_fwd_instance(std::vector<DeviceConvFwdPtr>&);
template <ck::index_t NDimSpatial,
typename InDataType,
typename WeiDataType,
typename OutDataType,
typename InLayout,
typename WeiLayout,
typename OutLayout>
void add_device_conv_bwd_instance(std::vector<DeviceConvBwdPtr>&);
template <ck::index_t NDimSpatial,
typename InDataType,
typename WeiDataType,
typename OutDataType,
typename InLayout,
typename WeiLayout,
typename OutLayout>
void add_device_conv_wrw_instance(std::vector<DeviceConvWrwPtr>&);
} // namespace device_conv_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
device_operation/include/device_gemm.hpp 0 → 100644
View file @ ed068043
#ifndef DEVICE_GEMM_HPP
#define DEVICE_GEMM_HPP
#include <iostream>
#include "device_base.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
struct DeviceGemm : public BaseOperator
{
virtual std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
const void* p_b,
void* p_c,
ck::index_t M,
ck::index_t N,
ck::index_t K,
ck::index_t StrideA,
ck::index_t StrideB,
ck::index_t StrideC) = 0;
virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
};
using DeviceGemmPtr = std::unique_ptr<DeviceGemm>;
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
device_operation/include/device_gemm_instance.hpp 0 → 100644
View file @ ed068043
#ifndef DEVICE_GEMM_INSTANTCE_HPP
#define DEVICE_GEMM_INSTANTCE_HPP
#include "device_gemm.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_gemm_instance {
template <typename ADataType,
typename BDataType,
typename CDataType,
typename ALayout,
typename BLayout,
typename CLayout>
void add_device_gemm_instance(std::vector<DeviceGemmPtr>&);
} // namespace device_gemm_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
device_operation/include/device_gemm_xdl.hpp 0 → 100644
View file @ ed068043
#ifndef DEVICE_GEMM_XDL_HPP
#define DEVICE_GEMM_XDL_HPP
#include <iostream>
#include "device.hpp"
#include "gemm_common.hpp"
#include "device_base.hpp"
#include "device_gemm.hpp"
#include "common_header.hpp"
#include "tensor_layout.hpp"
#include "tensor_descriptor.hpp"
#include "tensor_descriptor_helper.hpp"
#include "gridwise_gemm_xdlops_v2r3.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
template <typename ADataType,
typename BDataType,
typename CDataType,
typename AccDataType,
typename ALayout,
typename BLayout,
typename CLayout,
ck::index_t BlockSize,
ck::index_t MPerBlock,
ck::index_t NPerBlock,
ck::index_t K0PerBlock,
ck::index_t K1,
ck::index_t MPerXDL,
ck::index_t NPerXDL,
ck::index_t MXdlPerWave,
ck::index_t NXdlPerWave,
typename ABlockTransferThreadSliceLengths_K0_M_K1,
typename ABlockTransferThreadClusterLengths_K0_M_K1,
typename ABlockTransferThreadClusterArrangeOrder,
typename ABlockTransferSrcAccessOrder,
ck::index_t ABlockTransferSrcVectorDim,
ck::index_t ABlockTransferSrcScalarPerVector,
ck::index_t ABlockTransferDstScalarPerVector_K1,
typename BBlockTransferThreadSliceLengths_K0_N_K1,
typename BBlockTransferThreadClusterLengths_K0_N_K1,
typename BBlockTransferThreadClusterArrangeOrder,
typename BBlockTransferSrcAccessOrder,
ck::index_t BBlockTransferSrcVectorDim,
ck::index_t BBlockTransferSrcScalarPerVector,
ck::index_t BBlockTransferDstScalarPerVector_K1,
ck::index_t CThreadTransferSrcDstVectorDim,
ck::index_t CThreadTransferDstScalarPerVector,
bool ABlockLdsAddExtraM,
bool BBlockLdsAddExtraN>
struct DeviceGemmXdl : public DeviceGemm
{
static constexpr auto I0 = Number<0>{};
static constexpr auto I1 = Number<1>{};
static constexpr auto I2 = Number<2>{};
static constexpr auto K1Number = Number<K1>{};
static auto MakeAGridDescriptor_K0_M_K1(index_t M, index_t K, index_t StrideA)
{
assert(K % K1 == 0);
const index_t K0 = K / K1;
const auto a_grid_desc_m_k = [&]() {
if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
{
return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
}
else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
{
return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
}
}();
const auto a_grid_desc_k0_m_k1 =
transform_tensor_descriptor(a_grid_desc_m_k,
make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
make_pass_through_transform(M)),
make_tuple(Sequence<1>{}, Sequence<0>{}),
make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
return a_grid_desc_k0_m_k1;
}
static auto MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t StrideB)
{
assert(K % K1 == 0);
const index_t K0 = K / K1;
const auto b_grid_desc_k_n = [&]() {
if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
{
return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
}
else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
{
return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
}
}();
const auto b_grid_desc_k0_n_k1 =
transform_tensor_descriptor(b_grid_desc_k_n,
make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
make_pass_through_transform(N)),
make_tuple(Sequence<0>{}, Sequence<1>{}),
make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
return b_grid_desc_k0_n_k1;
}
static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
{
if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
{
return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
}
else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
{
return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
}
}
using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
using CGridDesc_M_N = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
// TODO remove these hacks
static constexpr auto a_k0_m_k1_grid_step_hacks =
make_tuple(make_tuple(Sequence<0, 0, 0>{}, // 0+: K0
Sequence<0, 0, 0>{}, // 1+: M
Sequence<0, 0, 0>{}), // 2+: K1
make_tuple(Sequence<0, 0, 0>{}, // 0-: K0
Sequence<0, 0, 0>{}, // 1-: M
Sequence<0, 0, 0>{})); // 2-: K1
static constexpr auto b_k0_n_k1_grid_step_hacks =
make_tuple(make_tuple(Sequence<0, 0, 0>{}, // 0+: K0
Sequence<0, 0, 0>{}, // 1+: N
Sequence<0, 0, 0>{}), // 2+: K1
make_tuple(Sequence<0, 0, 0>{}, // 0-: K0
Sequence<0, 0, 0>{}, // 1-: N
Sequence<0, 0, 0>{})); // 2-: K1
static constexpr auto c_m0_n0_m1_n1_m2_m3_m4_n2_grid_step_hacks =
make_tuple(make_tuple(Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 0+: M0
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 1+: N0
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 2+: M1
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 3+: N1
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 4+: M2
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 5+: M3
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 6+: M4
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}), // 7+: N2
make_tuple(Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 0-: M0
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 1-: N0
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 2-: M1
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 3-: N1
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 4-: M2
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 5-: M3
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{}, // 6-: M4
Sequence<0, 0, 0, 0, 0, 0, 0, 0, 0>{})); // 7-: N2
static constexpr auto a_k0_m_k1_grid_move_slice_window_step_hacks = Sequence<0, 0, 0>{};
static constexpr auto b_k0_n_k1_grid_move_slice_window_step_hacks = Sequence<0, 0, 0>{};
// GridwiseGemm
using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3<
BlockSize,
ADataType, // TODO: distinguish A/B datatype
AccDataType,
CDataType,
InMemoryDataOperationEnum_t::Set,
AGridDesc_K0_M_K1,
BGridDesc_K0_N_K1,
CGridDesc_M_N,
MPerBlock,
NPerBlock,
K0PerBlock,
MPerXDL,
NPerXDL,
K1,
MXdlPerWave,
NXdlPerWave,
ABlockTransferThreadSliceLengths_K0_M_K1,
ABlockTransferThreadClusterLengths_K0_M_K1,
ABlockTransferThreadClusterArrangeOrder,
ABlockTransferSrcAccessOrder,
ABlockTransferSrcVectorDim,
ABlockTransferSrcScalarPerVector,
ABlockTransferDstScalarPerVector_K1,
false, // AThreadTransferSrcResetCoordinateAfterRun,
BBlockTransferThreadSliceLengths_K0_N_K1,
BBlockTransferThreadClusterLengths_K0_N_K1,
BBlockTransferThreadClusterArrangeOrder,
BBlockTransferSrcAccessOrder,
BBlockTransferSrcVectorDim,
BBlockTransferSrcScalarPerVector,
BBlockTransferDstScalarPerVector_K1,
false, // BThreadTransferSrcResetCoordinateAfterRun,
Sequence<0, 2, 4, 5, 6, 1, 3, 7>, // CThreadTransferSrcDstAccessOrder,
CThreadTransferSrcDstVectorDim,
CThreadTransferDstScalarPerVector,
decltype(a_k0_m_k1_grid_step_hacks), // AGridStepHacks,
decltype(b_k0_n_k1_grid_step_hacks), // BGridStepHacks,
decltype(c_m0_n0_m1_n1_m2_m3_m4_n2_grid_step_hacks), // CGridStepHacks,
decltype(a_k0_m_k1_grid_move_slice_window_step_hacks), // AGridMoveSliceWindowStepHacks,
decltype(b_k0_n_k1_grid_move_slice_window_step_hacks), // BGridMoveSliceWindowStepHacks,
false, // CAccessOrderMRepeatNRepeat,
ABlockLdsAddExtraM,
BBlockLdsAddExtraN>;
using CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 =
decltype(GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(CGridDesc_M_N{}));
using Block2CTileMap = decltype(GridwiseGemm::MakeBlock2CTileMap(CGridDesc_M_N{}, 1, 1));
// Argument
struct Argument : public BaseArgument
{
Argument(const ADataType* p_a_grid,
const BDataType* p_b_grid,
CDataType* p_c_grid,
index_t M,
index_t N,
index_t K,
index_t StrideA,
index_t StrideB,
index_t StrideC,
index_t M01,
index_t N01)
: p_a_grid_{p_a_grid},
p_b_grid_{p_b_grid},
p_c_grid_{p_c_grid},
a_grid_desc_k0_m_k1_{},
b_grid_desc_k0_n_k1_{},
c_grid_desc_m_n_{},
c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_{},
block_2_ctile_map_{},
M01_{M01},
N01_{N01}
{
a_grid_desc_k0_m_k1_ = DeviceGemmXdl::MakeAGridDescriptor_K0_M_K1(M, K, StrideA);
b_grid_desc_k0_n_k1_ = DeviceGemmXdl::MakeBGridDescriptor_K0_N_K1(K, N, StrideB);
c_grid_desc_m_n_ = DeviceGemmXdl::MakeCGridDescriptor_M_N(M, N, StrideC);
if(GridwiseGemm::CheckValidity(
a_grid_desc_k0_m_k1_, b_grid_desc_k0_n_k1_, c_grid_desc_m_n_, M01_, N01_))
{
c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_ =
GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(c_grid_desc_m_n_);
block_2_ctile_map_ = GridwiseGemm::MakeBlock2CTileMap(c_grid_desc_m_n_, M01, N01);
}
}
// private:
const ADataType* p_a_grid_;
const BDataType* p_b_grid_;
CDataType* p_c_grid_;
AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
CGridDesc_M_N c_grid_desc_m_n_;
CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_;
Block2CTileMap block_2_ctile_map_;
index_t M01_;
index_t N01_;
};
// Invoker
struct Invoker : public BaseInvoker
{
using Argument = DeviceGemmXdl::Argument;
float Run(const Argument& arg, int nrepeat = 1)
{
{
std::cout << "arg.a_grid_desc_k0_m_k1_{" << arg.a_grid_desc_k0_m_k1_.GetLength(I0)
<< ", " << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
<< arg.a_grid_desc_k0_m_k1_.GetLength(I2) << "}" << std::endl;
std::cout << "arg.b_grid_desc_k0_n_k1_{" << arg.b_grid_desc_k0_n_k1_.GetLength(I0)
<< ", " << arg.b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
<< arg.b_grid_desc_k0_n_k1_.GetLength(I2) << "}" << std::endl;
std::cout << "arg.c_grid_desc_m_n_{ " << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
<< arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
}
if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
arg.b_grid_desc_k0_n_k1_,
arg.c_grid_desc_m_n_,
arg.M01_,
arg.N01_))
{
throw std::runtime_error(
"wrong! GridwiseGemm_km_kn_m0m1n0n1_xdlops_v2r3 has invalid setting");
}
const index_t grid_size = GridwiseGemm::CalculateGridSize(arg.c_grid_desc_m_n_);
const auto K0 = arg.a_grid_desc_k0_m_k1_.GetLength(I0);
const bool has_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
float ave_time = 0;
if(has_main_k0_block_loop)
{
const auto kernel = kernel_gemm_xdlops_v2r3<
GridwiseGemm,
ADataType, // TODO: distiguish A/B datatype
CDataType,
remove_reference_t<DeviceGemmXdl::AGridDesc_K0_M_K1>,
remove_reference_t<DeviceGemmXdl::BGridDesc_K0_N_K1>,
remove_reference_t<DeviceGemmXdl::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
remove_reference_t<DeviceGemmXdl::Block2CTileMap>,
true>;
ave_time = launch_and_time_kernel(kernel,
nrepeat,
dim3(grid_size),
dim3(BlockSize),
0,
arg.p_a_grid_,
arg.p_b_grid_,
arg.p_c_grid_,
arg.a_grid_desc_k0_m_k1_,
arg.b_grid_desc_k0_n_k1_,
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
arg.block_2_ctile_map_);
}
else
{
const auto kernel = kernel_gemm_xdlops_v2r3<
GridwiseGemm,
ADataType, // TODO: distiguish A/B datatype
CDataType,
remove_reference_t<DeviceGemmXdl::AGridDesc_K0_M_K1>,
remove_reference_t<DeviceGemmXdl::BGridDesc_K0_N_K1>,
remove_reference_t<DeviceGemmXdl::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
remove_reference_t<DeviceGemmXdl::Block2CTileMap>,
false>;
ave_time = launch_and_time_kernel(kernel,
nrepeat,
dim3(grid_size),
dim3(BlockSize),
0,
arg.p_a_grid_,
arg.p_b_grid_,
arg.p_c_grid_,
arg.a_grid_desc_k0_m_k1_,
arg.b_grid_desc_k0_n_k1_,
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
arg.block_2_ctile_map_);
}
return ave_time;
}
// polymorphic
float Run(const BaseArgument* p_arg, int nrepeat = 1) override
{
return Run(*dynamic_cast<const Argument*>(p_arg), nrepeat);
}
};
static constexpr bool IsValidCompilationParameter()
{
// TODO: properly implement this check
return true;
}
static bool IsSupportedArgument(const Argument& arg)
{
return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
arg.b_grid_desc_k0_n_k1_,
arg.c_grid_desc_m_n_,
arg.M01_,
arg.N01_);
}
// polymorphic
bool IsSupportedArgument(const BaseArgument* p_arg) override
{
return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
}
static auto MakeArgument(const ADataType* p_a,
const BDataType* p_b,
CDataType* p_c,
index_t M,
index_t N,
index_t K,
index_t StrideA,
index_t StrideB,
index_t StrideC)
{
return Argument{p_a, p_b, p_c, M, N, K, StrideA, StrideB, StrideC, 1, 1};
}
static auto MakeInvoker() { return Invoker{}; }
// polymorphic
std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
const void* p_b,
void* p_c,
index_t M,
index_t N,
index_t K,
index_t StrideA,
index_t StrideB,
index_t StrideC) override
{
return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
static_cast<const BDataType*>(p_b),
static_cast<CDataType*>(p_c),
M,
N,
K,
StrideA,
StrideB,
StrideC,
1,
1);
}
// polymorphic
std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
{
return std::make_unique<Invoker>(Invoker{});
}
};
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
host/host_tensor/include/gemm_common.hpp → device_operation/include/gemm_common.hpp
View file @ ed068043
......@@ -13,4 +13,10 @@ enum GemmMatrixLayout
KM_NK_NM, // 7
};
enum GemmDataType
{
F32_F32_F32, // 0
F16_F16_F16, // 1
};
#endif
device_operation/include/tensor_layout.hpp 0 → 100644
View file @ ed068043
#ifndef TENSOR_LAYOUT_HPP
#define TENSOR_LAYOUT_HPP
namespace ck {
namespace tensor_layout {
struct BaseTensorLayout
{
};
namespace gemm {
struct RowMajor : public BaseTensorLayout
{
};
struct ColumnMajor : public BaseTensorLayout
{
};
} // namespace gemm
namespace convolution {
struct NHWC : public BaseTensorLayout
{
};
struct KYXC : public BaseTensorLayout
{
};
struct NHWK : public BaseTensorLayout
{
};
struct NCHW : public BaseTensorLayout
{
};
struct KCYX : public BaseTensorLayout
{
};
struct NKHW : public BaseTensorLayout
{
};
} // namespace convolution
} // namespace tensor_layout
} // namespace ck
#endif
example/1_gemm_xdl/README.md 0 → 100644
View file @ ed068043
# Instructions for ```gemm_xdl``` Example
## Docker script
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ``gemm_xdl```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j gemm_xdl
```
## Run ```gemm_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)
./example/gemm_xdl.sh 0 1 5
```
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
```
a_m_k: dim 2, lengths {3840, 4096}, strides {4096, 1}
b_k_n: dim 2, lengths {4096, 4096}, strides {1, 4096}
c_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
arg.a_grid_desc_k0_m_k1_{512, 3840, 8}
arg.b_grid_desc_k0_n_k1_{512, 4096, 8}
arg.c_grid_desc_m_n_{ 3840, 4096}
launch_and_time_kernel: grid_dim {480, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 5 times...
Perf: 1.19685 ms, 107.657 TFlops, 78.8501 GB/s
```
example/1_gemm_xdl/gemm_xdl.cpp 0 → 100644
View file @ ed068043
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include <stdlib.h>
#include <half.hpp>
#include "config.hpp"
#include "print.hpp"
#include "device.hpp"
#include "host_tensor.hpp"
#include "host_tensor_generator.hpp"
#include "gemm_common.hpp"
#include "host_gemm.hpp"
#include "device_tensor.hpp"
#include "device_base.hpp"
#include "device_gemm_xdl.hpp"
template <typename ADataType,
typename BDataType,
typename CDataType,
typename ALayout,
typename BLayout,
typename CLayout>
struct DeviceGemmInstance;
template <>
struct DeviceGemmInstance<ck::half_t,
ck::half_t,
ck::half_t,
ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::RowMajor>
{
using F16 = ck::half_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
// Compilation parameters for NT problem
// clang-format off
using type =
//########################################| AData| BData| CData| AccData| ALayout| BLayout| CLayout| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| ABlockLds| BBlockLds|
//########################################| Type| Type| Type| Type| | | | Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| SrcDstVectorDim| DstScalar| AddExtraM| AddExtraN|
//########################################| | | | | | | | | | | | | | | Wave| Wave| Lengths_K0_N_K1| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| Lengths_K0_N_K1| Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerVector| | |
//########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
ck::tensor_operation::device::DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, S<1, 2, 8>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 7, 1, true, true>;
// clang-format on
};
template <>
struct DeviceGemmInstance<float,
float,
float,
ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::RowMajor>
{
using F16 = ck::half_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
// Compilation parameters for NT problem
// clang-format off
using type =
//########################################| AData| BData| CData| AccData| ALayout| BLayout| CLayout| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| ABlockLds| BBlockLds|
//########################################| Type| Type| Type| Type| | | | Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ThreadSlice| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| SrcDstVectorDim| DstScalar| AddExtraM| AddExtraN|
//########################################| | | | | | | | | | | | | | | Wave| Wave| Lengths_K0_N_K1| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| Lengths_K0_N_K1| Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerVector| | |
//########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
ck::tensor_operation::device::DeviceGemmXdl< F32, F32, F32, F32, Row, Col, Row, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, S<1, 2, 4>, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 7, 1, true, true>;
// clang-format on
};
int main(int argc, char* argv[])
{
if(argc != 4)
{
printf("arg1: verification (0=no, 1=yes)\n");
printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
printf("arg3: run kernel # of times (>1)\n");
exit(0);
}
const bool do_verification = std::stoi(argv[1]);
const int init_method = std::stoi(argv[2]);
const int nrepeat = std::stoi(argv[3]);
// 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;
// matrix data type
using ADataType = ck::half_t;
using BDataType = ck::half_t;
using CDataType = ck::half_t;
// matrix layout
using ALayout = ck::tensor_layout::gemm::RowMajor;
using BLayout = ck::tensor_layout::gemm::ColumnMajor;
using CLayout = ck::tensor_layout::gemm::RowMajor;
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<BDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
Tensor<BDataType> 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{-5, 5});
b_k_n.GenerateTensorValue(GeneratorTensor_2{-5, 5});
break;
default:
a_m_k.GenerateTensorValue(GeneratorTensor_3<float>{0.0, 1.0});
b_k_n.GenerateTensorValue(GeneratorTensor_3<float>{-0.5, 0.5});
}
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());
c_m_n_device_buf.ToDevice(c_m_n_device_result.mData.data());
// do GEMM
auto gemm =
typename DeviceGemmInstance<ADataType, BDataType, CDataType, ALayout, BLayout, CLayout>::
type{};
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);
if(!gemm.IsSupportedArgument(argument))
{
throw std::runtime_error(
"wrong! device_gemm with the specified compilation parameters does "
"not support this GEMM problem");
}
float ave_time = invoker.Run(argument, nrepeat);
std::size_t flop = std::size_t(2) * M * N * K;
std::size_t num_btype =
sizeof(ADataType) * M * K + sizeof(BDataType) * K * M + 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"
<< std::endl;
c_m_n_device_buf.FromDevice(c_m_n_device_result.mData.data());
if(do_verification)
{
host_gemm_mk_kn_mn(a_m_k, b_k_n, c_m_n_host_result);
check_error(c_m_n_host_result, c_m_n_device_result);
}
}
example/CMakeLists.txt 0 → 100644
View file @ ed068043
include_directories(BEFORE
include
${PROJECT_SOURCE_DIR}/host/host_tensor/include
${PROJECT_SOURCE_DIR}/host/device/include
${PROJECT_SOURCE_DIR}/device_operation/include
${PROJECT_SOURCE_DIR}/composable_kernel/include
${PROJECT_SOURCE_DIR}/composable_kernel/include/utility
${PROJECT_SOURCE_DIR}/composable_kernel/include/tensor_description
${PROJECT_SOURCE_DIR}/composable_kernel/include/tensor_operation
${PROJECT_SOURCE_DIR}/composable_kernel/include/problem_transform
${PROJECT_SOURCE_DIR}/external/rocm/include
)
set(GEMM_XDL_SOURCE 1_gemm_xdl/gemm_xdl.cpp)
add_executable(gemm_xdl ${GEMM_XDL_SOURCE})
target_link_libraries(gemm_xdl PRIVATE host_tensor)
external/half/include/half.hpp 0 → 100644
View file @ ed068043
This diff is collapsed.
host/driver_offline/CMakeLists.txt
View file @ ed068043
include_directories(BEFORE
include
${PROJECT_SOURCE_DIR}/host/host_tensor/include
${PROJECT_SOURCE_DIR}/host/device/include
${PROJECT_SOURCE_DIR}/host/solver/include
${PROJECT_SOURCE_DIR}/composable_kernel/include
${PROJECT_SOURCE_DIR}/composable_kernel/include/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