Skip to content

[lapack][rocsolver] Use the 64-bit pivot API for getrf and getrs #762

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
zjin-lcf wants to merge 1 commit into
base: develop
Choose a base branch
from
Open

[lapack][rocsolver] Use the 64-bit pivot API for getrf and getrs #762

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 23 additions & 6 deletions src/lapack/backends/rocsolver/rocsolver_helper.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -255,16 +255,33 @@ struct RocmEquivalentType<std::complex<double>> {
using Type = rocblas_double_complex;
};

/* 64-bit pivot API */

#if !defined(ROCSOLVER_VERSION_MAJOR)
#define ONEMATH_ROCSOLVER_VERSION 0
#else
#define ONEMATH_ROCSOLVER_VERSION (ROCSOLVER_VERSION_MAJOR * 100 + ROCSOLVER_VERSION_MINOR)
#endif

// rocSOLVER exposes getrf and getrs entry points taking int64_t pivots, which lets oneMath hand
// the user's ipiv array straight through instead of converting it. They were added in rocSOLVER
// 3.26 (ROCm 6.2); earlier versions keep the conversion path, so the minimum supported ROCm is
// unchanged.
#define ONEMATH_ROCSOLVER_HAS_64BIT_PIVOTS (ONEMATH_ROCSOLVER_VERSION >= 326)

/* devinfo */

inline int get_rocsolver_devinfo(sycl::queue& queue, sycl::buffer<int>& devInfo) {
sycl::host_accessor<int, 1, sycl::access::mode::read> dev_info_{ devInfo };
// The 64-bit entry points report info as int64_t, the legacy ones as int.
template <typename INFO_T>
inline INFO_T get_rocsolver_devinfo(sycl::queue& queue, sycl::buffer<INFO_T>& devInfo) {
sycl::host_accessor<INFO_T, 1, sycl::access::mode::read> dev_info_{ devInfo };
return dev_info_[0];
}

inline int get_rocsolver_devinfo(sycl::queue& queue, const int* devInfo) {
int dev_info_;
queue.memcpy(&dev_info_, devInfo, sizeof(int));
template <typename INFO_T>
inline INFO_T get_rocsolver_devinfo(sycl::queue& queue, const INFO_T* devInfo) {
INFO_T dev_info_;
queue.memcpy(&dev_info_, devInfo, sizeof(INFO_T));
queue.wait();
return dev_info_;
}
Expand All @@ -273,7 +290,7 @@ template <typename DEVINFO_T>
inline void lapack_info_check(sycl::queue& queue, DEVINFO_T devinfo, const char* func_name,
const char* cufunc_name) {
queue.wait();
const int devinfo_ = get_rocsolver_devinfo(queue, devinfo);
const auto devinfo_ = get_rocsolver_devinfo(queue, devinfo);
if (devinfo_ > 0)
throw oneapi::math::lapack::computation_error(
func_name, std::string(cufunc_name) + " failed with info = " + std::to_string(devinfo_),
Expand Down
180 changes: 180 additions & 0 deletions src/lapack/backends/rocsolver/rocsolver_lapack.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -132,6 +132,49 @@ GEQRF_LAUNCHER(std::complex<double>, rocsolver_zgeqrf)

#undef GEQRF_LAUNCHER

#if ONEMATH_ROCSOLVER_HAS_64BIT_PIVOTS

template <typename Func, typename T>
void getrf(const char* func_name, Func func, sycl::queue& queue, std::int64_t m, std::int64_t n,
sycl::buffer<T>& a, std::int64_t lda, sycl::buffer<std::int64_t>& ipiv,
sycl::buffer<T>& scratchpad, std::int64_t scratchpad_size) {
using rocmDataType = typename RocmEquivalentType<T>::Type;
sycl::buffer<std::int64_t> devInfo{ 1 };

queue.submit([&](sycl::handler& cgh) {
auto a_acc = a.template get_access<sycl::access::mode::read_write>(cgh);
auto ipiv_acc = ipiv.template get_access<sycl::access::mode::write>(cgh);
auto devInfo_acc = devInfo.template get_access<sycl::access::mode::write>(cgh);
onemath_rocsolver_host_task(cgh, queue, [=](RocsolverScopedContextHandler& sc) {
auto handle = sc.get_handle(queue);
auto a_ = sc.get_mem<rocmDataType*>(a_acc);
auto ipiv_ = sc.get_mem<std::int64_t*>(ipiv_acc);
auto devInfo_ = sc.get_mem<std::int64_t*>(devInfo_acc);
rocblas_status err;
rocsolver_native_named_func(func_name, func, err, handle, m, n, a_, lda, ipiv_,
devInfo_);
});
});
lapack_info_check(queue, devInfo, __func__, func_name);
}

#define GETRF_LAUNCHER(TYPE, ROCSOLVER_ROUTINE) \
void getrf(sycl::queue& queue, std::int64_t m, std::int64_t n, sycl::buffer<TYPE>& a, \
std::int64_t lda, sycl::buffer<std::int64_t>& ipiv, sycl::buffer<TYPE>& scratchpad, \
std::int64_t scratchpad_size) { \
getrf(#ROCSOLVER_ROUTINE, ROCSOLVER_ROUTINE, queue, m, n, a, lda, ipiv, scratchpad, \
scratchpad_size); \
}

GETRF_LAUNCHER(float, rocsolver_sgetrf_64)
GETRF_LAUNCHER(double, rocsolver_dgetrf_64)
GETRF_LAUNCHER(std::complex<float>, rocsolver_cgetrf_64)
GETRF_LAUNCHER(std::complex<double>, rocsolver_zgetrf_64)

#undef GETRF_LAUNCHER

#else // no 64-bit pivot API: factorise into a temporary 32-bit array and widen it

template <typename Func, typename T>
void getrf(const char* func_name, Func func, sycl::queue& queue, std::int64_t m, std::int64_t n,
sycl::buffer<T>& a, std::int64_t lda, sycl::buffer<std::int64_t>& ipiv,
Expand Down Expand Up @@ -188,6 +231,8 @@ GETRF_LAUNCHER(std::complex<double>, rocsolver_zgetrf)

#undef GETRF_LAUNCHER

#endif // ONEMATH_ROCSOLVER_HAS_64BIT_PIVOTS

void getri(sycl::queue& queue, std::int64_t n, sycl::buffer<std::complex<float>>& a,
std::int64_t lda, sycl::buffer<std::int64_t>& ipiv,
sycl::buffer<std::complex<float>>& scratchpad, std::int64_t scratchpad_size) {
Expand All @@ -209,6 +254,50 @@ void getri(sycl::queue& queue, std::int64_t n, sycl::buffer<std::complex<double>
throw unimplemented("lapack", "getri");
}

#if ONEMATH_ROCSOLVER_HAS_64BIT_PIVOTS

template <typename Func, typename T>
inline void getrs(const char* func_name, Func func, sycl::queue& queue,
oneapi::math::transpose trans, std::int64_t n, std::int64_t nrhs,
sycl::buffer<T>& a, std::int64_t lda, sycl::buffer<std::int64_t>& ipiv,
sycl::buffer<T>& b, std::int64_t ldb, sycl::buffer<T>& scratchpad,
std::int64_t scratchpad_size) {
using rocmDataType = typename RocmEquivalentType<T>::Type;

queue.submit([&](sycl::handler& cgh) {
auto a_acc = a.template get_access<sycl::access::mode::read>(cgh);
auto ipiv_acc = ipiv.template get_access<sycl::access::mode::read>(cgh);
auto b_acc = b.template get_access<sycl::access::mode::write>(cgh);
onemath_rocsolver_host_task(cgh, queue, [=](RocsolverScopedContextHandler& sc) {
auto handle = sc.get_handle(queue);
auto a_ = sc.get_mem<rocmDataType*>(a_acc);
auto ipiv_ = sc.get_mem<std::int64_t*>(ipiv_acc);
auto b_ = sc.get_mem<rocmDataType*>(b_acc);
rocblas_status err;
rocsolver_native_named_func(func_name, func, err, handle, get_rocblas_operation(trans),
n, nrhs, a_, lda, ipiv_, b_, ldb);
});
});
}

#define GETRS_LAUNCHER(TYPE, ROCSOLVER_ROUTINE) \
void getrs(sycl::queue& queue, oneapi::math::transpose trans, std::int64_t n, \
std::int64_t nrhs, sycl::buffer<TYPE>& a, std::int64_t lda, \
sycl::buffer<std::int64_t>& ipiv, sycl::buffer<TYPE>& b, std::int64_t ldb, \
sycl::buffer<TYPE>& scratchpad, std::int64_t scratchpad_size) { \
getrs(#ROCSOLVER_ROUTINE, ROCSOLVER_ROUTINE, queue, trans, n, nrhs, a, lda, ipiv, b, ldb, \
scratchpad, scratchpad_size); \
}

GETRS_LAUNCHER(float, rocsolver_sgetrs_64)
GETRS_LAUNCHER(double, rocsolver_dgetrs_64)
GETRS_LAUNCHER(std::complex<float>, rocsolver_cgetrs_64)
GETRS_LAUNCHER(std::complex<double>, rocsolver_zgetrs_64)

#undef GETRS_LAUNCHER

#else // no 64-bit pivot API: narrow the pivots into a temporary 32-bit array

template <typename Func, typename T>
inline void getrs(const char* func_name, Func func, sycl::queue& queue,
oneapi::math::transpose trans, std::int64_t n, std::int64_t nrhs,
Expand Down Expand Up @@ -264,6 +353,8 @@ GETRS_LAUNCHER(std::complex<double>, rocsolver_zgetrs)

#undef GETRS_LAUNCHER

#endif // ONEMATH_ROCSOLVER_HAS_64BIT_PIVOTS

template <typename Func, typename T_A, typename T_B>
inline void gesvd(const char* func_name, Func func, sycl::queue& queue, oneapi::math::jobsvd jobu,
oneapi::math::jobsvd jobvt, std::int64_t m, std::int64_t n, sycl::buffer<T_A>& a,
Expand Down Expand Up @@ -1259,6 +1350,50 @@ GEQRF_LAUNCHER_USM(std::complex<double>, rocsolver_zgeqrf)

#undef GEQRF_LAUNCHER_USM

#if ONEMATH_ROCSOLVER_HAS_64BIT_PIVOTS

template <typename Func, typename T>
inline sycl::event getrf(const char* func_name, Func func, sycl::queue& queue, std::int64_t m,
std::int64_t n, T* a, std::int64_t lda, std::int64_t* ipiv, T* scratchpad,
std::int64_t scratchpad_size,
const std::vector<sycl::event>& dependencies) {
using rocmDataType = typename RocmEquivalentType<T>::Type;

std::int64_t* devInfo = (std::int64_t*)malloc_device(sizeof(std::int64_t), queue);
auto done = queue.submit([&](sycl::handler& cgh) {
cgh.depends_on(dependencies);
onemath_rocsolver_host_task(cgh, queue, [=](RocsolverScopedContextHandler& sc) {
auto handle = sc.get_handle(queue);
auto a_ = reinterpret_cast<rocmDataType*>(a);
rocblas_status err;
rocsolver_native_named_func(func_name, func, err, handle, m, n, a_, lda, ipiv, devInfo);
});
});

// lapack_info_check calls queue.wait()
lapack_info_check(queue, devInfo, __func__, func_name);
free(devInfo, queue);
return done;
}

#define GETRF_LAUNCHER_USM(TYPE, ROCSOLVER_ROUTINE) \
sycl::event getrf(sycl::queue& queue, std::int64_t m, std::int64_t n, TYPE* a, \
std::int64_t lda, std::int64_t* ipiv, TYPE* scratchpad, \
std::int64_t scratchpad_size, \
const std::vector<sycl::event>& dependencies) { \
return getrf(#ROCSOLVER_ROUTINE, ROCSOLVER_ROUTINE, queue, m, n, a, lda, ipiv, scratchpad, \
scratchpad_size, dependencies); \
}

GETRF_LAUNCHER_USM(float, rocsolver_sgetrf_64)
GETRF_LAUNCHER_USM(double, rocsolver_dgetrf_64)
GETRF_LAUNCHER_USM(std::complex<float>, rocsolver_cgetrf_64)
GETRF_LAUNCHER_USM(std::complex<double>, rocsolver_zgetrf_64)

#undef GETRF_LAUNCHER_USM

#else // no 64-bit pivot API: factorise into a temporary 32-bit array and widen it

template <typename Func, typename T>
inline sycl::event getrf(const char* func_name, Func func, sycl::queue& queue, std::int64_t m,
std::int64_t n, T* a, std::int64_t lda, std::int64_t* ipiv, T* scratchpad,
Expand Down Expand Up @@ -1320,6 +1455,8 @@ GETRF_LAUNCHER_USM(std::complex<double>, rocsolver_zgetrf)

#undef GETRF_LAUNCHER_USM

#endif // ONEMATH_ROCSOLVER_HAS_64BIT_PIVOTS

sycl::event getri(sycl::queue& queue, std::int64_t n, std::complex<float>* a, std::int64_t lda,
std::int64_t* ipiv, std::complex<float>* scratchpad, std::int64_t scratchpad_size,
const std::vector<sycl::event>& dependencies) {
Expand All @@ -1341,6 +1478,47 @@ sycl::event getri(sycl::queue& queue, std::int64_t n, std::complex<double>* a, s
throw unimplemented("lapack", "getri");
}

#if ONEMATH_ROCSOLVER_HAS_64BIT_PIVOTS

template <typename Func, typename T>
inline sycl::event getrs(const char* func_name, Func func, sycl::queue& queue,
oneapi::math::transpose trans, std::int64_t n, std::int64_t nrhs, T* a,
std::int64_t lda, std::int64_t* ipiv, T* b, std::int64_t ldb,
T* scratchpad, std::int64_t scratchpad_size,
const std::vector<sycl::event>& dependencies) {
using rocmDataType = typename RocmEquivalentType<T>::Type;

return queue.submit([&](sycl::handler& cgh) {
cgh.depends_on(dependencies);
onemath_rocsolver_host_task(cgh, queue, [=](RocsolverScopedContextHandler& sc) {
auto handle = sc.get_handle(queue);
auto a_ = reinterpret_cast<rocmDataType*>(a);
auto b_ = reinterpret_cast<rocmDataType*>(b);
rocblas_status err;
rocsolver_native_named_func(func_name, func, err, handle, get_rocblas_operation(trans),
n, nrhs, a_, lda, ipiv, b_, ldb);
});
});
}

#define GETRS_LAUNCHER_USM(TYPE, ROCSOLVER_ROUTINE) \
sycl::event getrs(sycl::queue& queue, oneapi::math::transpose trans, std::int64_t n, \
std::int64_t nrhs, TYPE* a, std::int64_t lda, std::int64_t* ipiv, TYPE* b, \
std::int64_t ldb, TYPE* scratchpad, std::int64_t scratchpad_size, \
const std::vector<sycl::event>& dependencies) { \
return getrs(#ROCSOLVER_ROUTINE, ROCSOLVER_ROUTINE, queue, trans, n, nrhs, a, lda, ipiv, \
b, ldb, scratchpad, scratchpad_size, dependencies); \
}

GETRS_LAUNCHER_USM(float, rocsolver_sgetrs_64)
GETRS_LAUNCHER_USM(double, rocsolver_dgetrs_64)
GETRS_LAUNCHER_USM(std::complex<float>, rocsolver_cgetrs_64)
GETRS_LAUNCHER_USM(std::complex<double>, rocsolver_zgetrs_64)

#undef GETRS_LAUNCHER_USM

#else // no 64-bit pivot API: narrow the pivots into a temporary 32-bit array

template <typename Func, typename T>
inline sycl::event getrs(const char* func_name, Func func, sycl::queue& queue,
oneapi::math::transpose trans, std::int64_t n, std::int64_t nrhs, T* a,
Expand Down Expand Up @@ -1402,6 +1580,8 @@ GETRS_LAUNCHER_USM(std::complex<double>, rocsolver_zgetrs)

#undef GETRS_LAUNCHER_USM

#endif // ONEMATH_ROCSOLVER_HAS_64BIT_PIVOTS

template <typename Func, typename T_A, typename T_B>
inline sycl::event gesvd(const char* func_name, Func func, sycl::queue& queue,
oneapi::math::jobsvd jobu, oneapi::math::jobsvd jobvt, std::int64_t m,
Expand Down
Loading