Files
pytorch/torch/csrc/Module.cpp
T
Grayson Derossi 86f2fb2c1e Add cublasLt as a backend for grouped GEMM (bf16/fp16 support) (#177037)
This PR adds a cublasLt backend to grouped GEMM. The cublasLt backend is default for fp16 on Blackwell for CUDA 13.2+ and Hopper for CUDA 13.3+, and can also be enabled for bf16 on those device/version combos with `torch.backends.cuda.matmul.prefer_cublaslt_grouped_gemm = True`. The cublasLt grouped GEMM path is fully compatible with `torch.compile` and CUDA Graphs.

There are some constraints (see [docs](https://docs.nvidia.com/cuda/cublas/index.html#id106)) but these are mostly handed internally- the only one that users need to be aware of is that matrices and leading dimensions must be 16-byte aligned, so matrices with some dimensions not divisible by 16 may require padding + slicing.

Here's an initial [benchmarking results spreadsheet](https://docs.google.com/spreadsheets/d/1Av3RwwmLERLKlZ2826rPAduAribjF8KQLKTHxn5Wb2Y/edit?usp=sharing). Note the two tabs. For uniform groups, the cublasLt backend is faster than the fallback (fp16) but slower than the CUTLASS kernel (bf16) except for the largest size tested. For MOE-like cases where groups are not uniform, cublasLt appears to be on average faster across the board.
[Benchmarking script](https://gist.github.com/gderossi/515581a165b9b8eb031ba7a1a5a01c55)

Scaled grouped GEMM for FP8 types is also working with all currently supported scaling types (tensorwise, per-group tensorwise, MXFP8) at #177038, to be stacked on top of this.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/177037
Approved by: https://github.com/eqy, https://github.com/drisspg, https://github.com/ngimel
2026-07-08 18:14:34 +00:00

3403 lines
104 KiB
C++

#include <ATen/DeviceAccelerator.h>
#include <fmt/core.h>
#include <sys/types.h>
#include <torch/csrc/python_headers.h>
#include <torch/csrc/utils/pythoncapi_compat.h>
#include <csignal>
#include <optional>
#ifndef _MSC_VER
#include <sys/socket.h>
#endif
#include <ATen/ATen.h>
#include <ATen/BlasBackend.h>
#include <ATen/CachedTensorUtils.h>
#include <ATen/DLConvertor.h>
#include <ATen/ExpandUtils.h>
#include <ATen/FakeTensorDispatchTables.h>
#include <ATen/LegacyVmapMode.h>
#include <ATen/LinalgBackend.h>
#include <ATen/Parallel.h>
#include <ATen/Utils.h>
#include <ATen/dlpack.h>
#include <ATen/native/ConvUtils.h>
#include <ATen/native/ForeachUtils.h>
#include <ATen/native/Normalization.h>
#include <c10/core/Device.h>
#include <c10/core/DispatchKeySet.h>
#include <c10/core/impl/COW.h>
#include <c10/core/impl/DeviceGuardImplInterface.h>
#include <c10/util/AbortHandler.h>
#include <c10/util/Backtrace.h>
#include <c10/util/Logging.h>
#include <c10/util/irange.h>
#include <c10/util/thread_name.h>
#include <libshm.h>
#include <pybind11/pybind11.h>
#include <pybind11/stl.h>
#include <torch/csrc/THConcat.h>
#include <torch/csrc/utils/pybind.h>
#include <cstdlib>
#include <iostream>
#include <unordered_map>
#include <ATen/ThreadLocalPythonObjects.h>
#include <torch/csrc/DataLoader.h>
#include <torch/csrc/Device.h>
#include <torch/csrc/DeviceAccelerator.h>
#include <torch/csrc/Dtype.h>
#include <torch/csrc/DynamicTypes.h>
#include <torch/csrc/Event.h>
#include <torch/csrc/Export.h>
#include <torch/csrc/Generator.h>
#include <torch/csrc/Layout.h>
#include <torch/csrc/MemoryFormat.h>
#include <torch/csrc/QScheme.h>
#include <torch/csrc/Stream.h>
#include <torch/csrc/THP.h>
#include <torch/csrc/TensorIterator.h>
#include <torch/csrc/TypeInfo.h>
#include <torch/csrc/acc/Module.h>
#include <torch/csrc/api/include/torch/python/init.h>
#include <torch/csrc/autograd/generated/python_return_types.h>
#include <torch/csrc/autograd/python_cpp_function.h>
#include <torch/csrc/autograd/python_enum_tag.h>
#include <torch/csrc/autograd/python_fft_functions.h>
#include <torch/csrc/autograd/python_function.h>
#include <torch/csrc/autograd/python_legacy_variable.h>
#include <torch/csrc/autograd/python_linalg_functions.h>
#include <torch/csrc/autograd/python_nested_functions.h>
#include <torch/csrc/autograd/python_nn_functions.h>
#include <torch/csrc/autograd/python_sparse_functions.h>
#include <torch/csrc/autograd/python_special_functions.h>
#include <torch/csrc/autograd/python_variable.h>
#include <torch/csrc/cpu/Module.h>
#include <torch/csrc/distributed/python_placement.h>
#include <torch/csrc/dynamo/init.h>
#include <torch/csrc/export/pybind.h>
#include <torch/csrc/functionalization/Module.h>
#include <torch/csrc/functorch/init.h>
#include <torch/csrc/fx/node.h>
#include <torch/csrc/inductor/aoti_package/pybind.h>
#include <torch/csrc/inductor/aoti_runner/pybind.h>
#include <torch/csrc/instruction_counter/Module.h>
#include <torch/csrc/jit/python/init.h>
#include <torch/csrc/jit/python/python_ir.h>
#include <torch/csrc/jit/python/python_tracer.h>
#include <torch/csrc/jit/serialization/pickler.h>
#include <torch/csrc/lazy/python/init.h>
#include <torch/csrc/monitor/python_init.h>
#include <torch/csrc/mps/Module.h>
#include <torch/csrc/mtia/Module.h>
#include <torch/csrc/multiprocessing/init.h>
#include <torch/csrc/onnx/init.h>
#include <torch/csrc/profiler/python/init.h>
#include <torch/csrc/tensor/python_tensor.h>
#include <torch/csrc/utils/disable_torch_function.h>
#include <torch/csrc/utils/init.h>
#include <torch/csrc/utils/pycfunction_helpers.h>
#include <torch/csrc/utils/python_arg_parser.h>
#include <torch/csrc/utils/python_compat.h>
#include <torch/csrc/utils/python_dispatch.h>
#include <torch/csrc/utils/python_strings.h>
#include <torch/csrc/utils/tensor_dtypes.h>
#include <torch/csrc/utils/tensor_layouts.h>
#include <torch/csrc/utils/tensor_memoryformats.h>
#include <torch/csrc/utils/tensor_new.h>
#include <torch/csrc/utils/tensor_numpy.h>
#include <torch/csrc/utils/tensor_qschemes.h>
#include <torch/csrc/utils/verbose.h>
#include <ATen/native/transformers/sdp_utils_cpp.h>
#include <torch/csrc/profiler/combined_traceback.h>
#include <torch/csrc/profiler/kineto_client_interface.h>
#include <sstream>
#ifdef USE_CUDA
#include <ATen/ROCmFABackend.h>
#include <ATen/cuda/CUDABlas.h>
#include <ATen/cuda/CUDAConfig.h>
#include <ATen/native/transformers/cuda/sdp_utils.h>
#include <torch/csrc/inductor/static_launcher/cuda.h>
#ifdef __HIP_PLATFORM_AMD__
#include <ATen/native/cudnn/hip/BatchNorm.h>
#else
#include <ATen/native/cudnn/BatchNorm.h>
#endif
#endif
#ifdef USE_XPU
#include <ATen/native/transformers/xpu/sdp_utils.h>
#ifndef _WIN32
#include <torch/csrc/inductor/static_launcher/xpu.h>
#endif
#endif
#ifdef USE_DISTRIBUTED
#ifdef USE_C10D
#include <torch/csrc/distributed/autograd/python_autograd.h>
#include <torch/csrc/distributed/c10d/c10d.h>
#include <torch/csrc/distributed/rpc/rpc.h>
#include <torch/csrc/distributed/rpc/testing/testing.h>
#endif
#endif
#if defined(USE_VALGRIND)
#include <callgrind.h>
#endif
#ifdef USE_ITT
#include <torch/csrc/itt.h>
#endif
#include <torch/nativert/python/Bindings.h>
namespace py = pybind11;
static PyObject* module;
static THPGenerator* THPDefaultCPUGenerator = nullptr;
////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
static PyObject* THPModule_initNames(PyObject* self, PyObject* arg) {
HANDLE_TH_ERRORS
static std::vector<std::string> names;
THPObjectPtr types(PySequence_Fast(arg, "expected a sequence"));
if (!types)
return nullptr;
// NOLINTNEXTLINE(bugprone-branch-clone)
auto num_classes = PySequence_Fast_GET_SIZE(types.get());
names.reserve(names.size() + num_classes);
for (Py_ssize_t i = 0; i < num_classes; i++) {
PyObject* obj = PySequence_Fast_GET_ITEM(types.get(), i);
TORCH_CHECK(PyType_Check(obj), "expected a PyTypeObject");
PyTypeObject* type = reinterpret_cast<PyTypeObject*>(obj);
THPObjectPtr module_name(PyObject_GetAttrString(obj, "__module__"));
if (!module_name)
return nullptr;
TORCH_CHECK(
THPUtils_checkString(module_name.get()),
"expected __module__ to be a string");
std::string name = THPUtils_unpackString(module_name.get());
names.emplace_back(name + "." + type->tp_name);
type->tp_name = names.back().c_str();
}
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
//
// Callback for python part. Used for additional initialization of python
// classes
static PyObject* THPModule_initExtension(
PyObject* _unused,
PyObject* shm_manager_path) {
HANDLE_TH_ERRORS
#if !defined(FBCODE_CAFFE2) && !defined(__aarch64__)
if (torch::get_cpp_stacktraces_enabled()) {
c10::SetStackTraceFetcher([]() -> std::string {
auto tb = torch::CapturedTraceback::gather(false, false, true);
if (torch::get_symbolize_mode() == torch::unwind::Mode::addr2line) {
LOG(WARNING)
<< "symbolizing C++ stack trace for exception; if this hangs, rerun with TORCH_DISABLE_ADDR2LINE=1..."
<< '\n';
}
auto s_tbs = torch::symbolize({tb.get()});
std::stringstream oss;
oss << "C++ CapturedTraceback:" << '\n';
const auto& s_tb = s_tbs.tracebacks.at(0);
for (auto idx : c10::irange(s_tb.size())) {
// Skip the first few frames:
// #1 torch::CapturedTraceback::gather(bool, bool, bool)
// #2 THPModule_initExtension
// #3 THPModule_initExtension(_object*, _object*)::{lambda()#1}
if (idx <= 3) {
continue;
}
auto frame_id = s_tb[idx];
const auto& frame = s_tbs.all_frames.at(frame_id);
oss << '#' << idx << ' ' << frame.funcname << " from " << frame.filename
<< ':' << frame.lineno << '\n';
}
return std::move(oss).str();
});
}
#endif
if (!THPUtils_checkString(shm_manager_path)) {
THPUtils_setError(
"initialization error - expected bytes/string object as shm_manager_path!");
return nullptr;
}
torch::utils::initializeLayouts();
torch::utils::initializeMemoryFormats();
torch::utils::initializeQSchemes();
torch::utils::initializeDtypes();
torch::tensors::initialize_python_bindings();
std::string path = THPUtils_unpackString(shm_manager_path);
libshm_init(path.c_str());
auto module = THPObjectPtr(PyImport_ImportModule("torch"));
if (!module)
throw python_error(); // @allow-raw-throw
THPStorage_postInit(module);
THPAutograd_initFunctions();
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
// The idea behind these functions is to make it easy to test if we are
// built with ASAN: they're designed not to crash if ASAN is not enabled, but
// to trigger ASAN if it is enabled. This lets us run a "canary" tests which
// checks if our build environment is misconfigured.
static PyObject* THPModule_crashIfCsrcASAN(PyObject* module, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
THPUtils_checkLong(arg),
"crash_if_csrc_asan expects an int, but got ",
THPUtils_typename(arg));
// NOLINTNEXTLINE(cppcoreguidelines-avoid-c-arrays, modernize-avoid-c-arrays)
volatile char x[3];
x[THPUtils_unpackInt(arg)] = 0;
// NOLINTNEXTLINE(clang-analyzer-core.CallAndMessage)
return THPUtils_packInt32(x[0]);
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_crashIfCsrcUBSAN(PyObject* module, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
THPUtils_checkLong(arg),
"crash_if_csrc_ubsan expects an int, but got ",
THPUtils_typename(arg));
int32_t x = THPUtils_unpackInt(arg);
double y = 1.0 / x;
return THPUtils_packInt32(static_cast<int>(y));
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_crashIfvptrUBSAN(PyObject* module, PyObject* noarg) {
// This code should work perfectly fine, as vtables are identical for Foo and
// Baz unless rtti and ubsan are enabled
struct Foo {
virtual int bar() = 0;
virtual ~Foo() = default;
};
struct Baz {
virtual int bar() {
return 17;
}
virtual ~Baz() = default;
};
Baz x{};
// Purposely cast through `void*` so there's no fixups applied.
// NOLINTNEXTLINE(bugprone-casting-through-void,-warnings-as-errors)
auto y = static_cast<Foo*>(static_cast<void*>(&x));
auto rc = y->bar();
return THPUtils_packInt32(rc);
}
static PyObject* THPModule_crashIfATenASAN(PyObject* module, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
THPUtils_checkLong(arg),
"crash_if_aten_asan expects an int, "
"but got ",
THPUtils_typename(arg));
return THPUtils_packInt32(at::_crash_if_asan(THPUtils_unpackInt(arg)));
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_abort(PyObject* module, PyObject* noargs) {
std::terminate();
Py_RETURN_NONE;
}
static PyObject* THPModule_crashIfDebugAssertsFail(
PyObject* module,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
THPUtils_checkLong(arg),
"crash_if_debug_asserts_fail expects an int, but got ",
THPUtils_typename(arg));
TORCH_INTERNAL_ASSERT_DEBUG_ONLY(
THPUtils_unpackInt(arg) != 424242,
"Expect anything but 424242 as an input for debug builds");
return THPUtils_packInt32(0);
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_getNumThreads(PyObject* module, PyObject* noargs) {
return THPUtils_packInt32(at::get_num_threads());
}
static PyObject* THPModule_setNumThreads(PyObject* module, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
THPUtils_checkLong(arg),
"set_num_threads expects an int, but got ",
THPUtils_typename(arg));
int nthreads = THPUtils_unpackInt(arg);
TORCH_CHECK(nthreads > 0, "set_num_threads expects a positive integer");
at::set_num_threads(nthreads);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_getNumInteropThreads(
PyObject* module,
PyObject* noargs) {
return THPUtils_packUInt64(at::get_num_interop_threads());
}
static PyObject* THPModule_setNumInteropThreads(
PyObject* module,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
THPUtils_checkLong(arg),
"set_num_interop_threads expects an int, "
"but got ",
THPUtils_typename(arg));
int nthreads = THPUtils_unpackInt(arg);
TORCH_CHECK(
nthreads > 0, "set_num_interop_threads expects a positive integer");
at::set_num_interop_threads(nthreads);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setDefaultTensorType(
PyObject* _unused,
PyObject* type) {
HANDLE_TH_ERRORS
torch::tensors::py_set_default_tensor_type(type);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setDefaultDtype(PyObject* _unused, PyObject* dtype) {
HANDLE_TH_ERRORS
torch::tensors::py_set_default_dtype(dtype);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_swap_tensor_impl(PyObject* _unused, PyObject* args) {
HANDLE_TH_ERRORS
PyObject* a_ = nullptr;
PyObject* b_ = nullptr;
if (!PyArg_ParseTuple(args, "OO", &a_, &b_)) {
return nullptr;
}
// Ensure we have Tensors
TORCH_CHECK(THPVariable_Check(a_));
TORCH_CHECK(THPVariable_Check(b_));
THPVariable* a = reinterpret_cast<THPVariable*>(a_);
THPVariable* b = reinterpret_cast<THPVariable*>(b_);
// weak_use_count() adds 1 if use_count is non-zero
TORCH_CHECK(
a->cdata.weak_use_count() == 1,
"Expected no weakrefs to t1's Tensor object but got ",
a->cdata.weak_use_count() - 1);
TORCH_CHECK(
b->cdata.weak_use_count() == 1,
"Expected no weakrefs to t2's Tensor object but got ",
b->cdata.weak_use_count() - 1);
// NB: Creating local copies of *both* Tensors here ensures that they each
// hold a strong reference to their PyObject. This avoids having to fix up
// reference counts when we swap the PyObject slots below.
at::Tensor tmp_a = a->cdata;
at::Tensor tmp_b = b->cdata;
// Swap the Tensor Impl
a->cdata = tmp_b;
b->cdata = tmp_a;
// Fix up the PyObjects associated with each TensorImpl
a->cdata.unsafeGetTensorImpl()->pyobj_slot()->store_pyobj(a_);
b->cdata.unsafeGetTensorImpl()->pyobj_slot()->store_pyobj(b_);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_addDocStr(PyObject* _unused, PyObject* args) {
// adds a __doc__ string to a function, similar to numpy's arr_add_docstring
static std::vector<std::string> all_docs;
PyObject* obj = nullptr;
PyObject* doc_obj = nullptr;
if (!PyArg_ParseTuple(args, "OO", &obj, &doc_obj)) {
return nullptr;
}
const char* doc_str = "<invalid string>";
if (THPUtils_checkString(doc_obj)) {
all_docs.push_back(THPUtils_unpackString(doc_obj));
doc_str = all_docs.back().c_str();
}
if (Py_TYPE(obj) == &PyCFunction_Type) {
PyCFunctionObject* f = reinterpret_cast<PyCFunctionObject*>(obj);
if (f->m_ml->ml_doc) {
return PyErr_Format(
PyExc_RuntimeError,
"function '%s' already has a docstring",
f->m_ml->ml_name);
}
f->m_ml->ml_doc = doc_str;
} else if (strcmp(Py_TYPE(obj)->tp_name, "method_descriptor") == 0) {
PyMethodDescrObject* m = reinterpret_cast<PyMethodDescrObject*>(obj);
if (m->d_method->ml_doc) {
return PyErr_Format(
PyExc_RuntimeError,
"method '%s' already has a docstring",
m->d_method->ml_name);
}
m->d_method->ml_doc = doc_str;
} else if (strcmp(Py_TYPE(obj)->tp_name, "getset_descriptor") == 0) {
PyGetSetDescrObject* m = reinterpret_cast<PyGetSetDescrObject*>(obj);
if (m->d_getset->doc) {
return PyErr_Format(
PyExc_RuntimeError,
"attribute '%s' already has a docstring",
m->d_getset->name);
}
m->d_getset->doc = doc_str;
} else if (PyType_Check(obj)) {
PyTypeObject* t = reinterpret_cast<PyTypeObject*>(obj);
if (t->tp_doc) {
return PyErr_Format(
PyExc_RuntimeError, "Type '%s' already has a docstring", t->tp_name);
}
t->tp_doc = doc_str;
} else {
return PyErr_Format(
PyExc_TypeError,
"don't know how to add docstring to type '%s'",
Py_TYPE(obj)->tp_name);
}
return Py_NewRef(obj);
}
static PyObject* THPModule_inferSize(PyObject* _unused, PyObject* args) {
HANDLE_TH_ERRORS
Py_ssize_t num_args = args ? PyTuple_Size(args) : 0;
TORCH_CHECK(num_args == 2, "expected exactly 2 arguments");
PyObject* arg1 = PyTuple_GET_ITEM(args, 0);
TORCH_CHECK(THPSize_Check(arg1), "expected a torch.Size as argument 1");
PyObject* arg2 = PyTuple_GET_ITEM(args, 1);
TORCH_CHECK(THPSize_Check(arg2), "expected a torch.Size as argument 2");
auto size1 = THPUtils_unpackLongs(arg1);
auto size2 = THPUtils_unpackLongs(arg2);
auto sizes = at::infer_size(size1, size2);
return THPSize_NewFromSizes(static_cast<int64_t>(sizes.size()), sizes.data());
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setBackcompatBroadcastWarn(
PyObject* module,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_backcompat_broadcast_warn expects a bool, "
"but got ",
THPUtils_typename(arg));
setBackCompatBroadcastWarn(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_getBackcompatBroadcastWarn(
PyObject* module,
PyObject* noargs) {
if (getBackCompatBroadcastWarn())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setBackcompatKeepdimWarn(
PyObject* module,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_backcompat_keepdim_warn expects a bool, "
"but got ",
THPUtils_typename(arg));
setBackCompatKeepdimWarn(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_getBackcompatKeepdimWarn(
PyObject* module,
PyObject* noargs) {
if (getBackCompatKeepdimWarn())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_hasDistributed(PyObject* _unused, PyObject* noargs) {
#ifdef USE_DISTRIBUTED
Py_RETURN_TRUE;
#else
Py_RETURN_FALSE;
#endif
}
static PyObject* THPModule_showConfig(PyObject* module, PyObject* noargs) {
HANDLE_TH_ERRORS
return THPUtils_packString(at::show_config());
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_cxxFlags(PyObject* module, PyObject* noargs) {
HANDLE_TH_ERRORS
return THPUtils_packString(at::get_cxx_flags());
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_parallelInfo(PyObject* module, PyObject* noargs) {
HANDLE_TH_ERRORS
return THPUtils_packString(at::get_parallel_info());
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_getCpuCapability(
PyObject* module,
PyObject* noargs) {
HANDLE_TH_ERRORS
return THPUtils_packString(at::get_cpu_capability());
END_HANDLE_TH_ERRORS
}
namespace {
template <class T>
void DLPack_Capsule_Destructor(PyObject* data) {
if (C10_LIKELY(!PyCapsule_IsValid(data, at::DLPackTraits<T>::capsule))) {
// early out, see DLPack spec: if a consuming library sets the capsule
// name to something else, they own it and we don't need to do anything
return;
}
HANDLE_TH_ERRORS
// Causes overheads for validity checks again, but this case is rare
// since consuming libraries should rename the capsule according to spec.
// Note that this cannot set a python error (we checked validity above),
// so we don't need to handle python error state here.
T* tensor = (T*)PyCapsule_GetPointer(data, at::DLPackTraits<T>::capsule);
// the dlMTensor has not been consumed, call deleter ourselves.
// DLPack spec mentions that deleter may be NULL, but deleter from
// `at::toDLPack` is never NULL, so no need for an additional check here.
tensor->deleter(tensor);
END_HANDLE_TH_ERRORS_RET()
}
template <class T>
PyObject* THPModule_toDLPackImpl(
PyObject* self,
PyObject* args,
PyObject* kwargs) {
HANDLE_TH_ERRORS
static torch::PythonArgParser parser(
{"_to_dlpack(Tensor data, *, IntArrayRef? dl_device=None, bool? copy=None, bool read_only=False)"});
torch::ParsedArgs<4> parsed_args{};
auto r = parser.parse(args, kwargs, parsed_args);
TORCH_INTERNAL_ASSERT(r.idx == 0);
auto data = r.tensor(0);
auto dl_device = r.intlist(1);
auto copy = r.toBoolOptional(2);
auto read_only = r.toBool(3);
// Parse the int list into a tuple.
std::optional<DLDevice> optional_dl_device;
if (!dl_device.empty()) {
TORCH_CHECK(
dl_device.size() == 2,
"dl_device must be either None or a tuple of ints");
optional_dl_device = DLDevice{
static_cast<DLDeviceType>(dl_device[0]),
static_cast<int32_t>(dl_device[1])};
}
auto src = at::maybeCopyTensor(data, optional_dl_device, copy);
T* tensor = nullptr;
if (read_only) {
// read_only is only representable on the versioned struct, which carries a
// flags field. The legacy DLManagedTensor cannot express it.
if constexpr (std::is_same_v<T, DLManagedTensorVersioned>) {
tensor = at::toDLPackVersionedReadOnly(src);
} else {
TORCH_CHECK(
false,
"read_only DLPack export requires the versioned DLPack protocol "
"(max_version >= 1)");
}
} else {
tensor = at::DLPackTraits<T>::toDLPack(src);
}
return PyCapsule_New(
tensor, at::DLPackTraits<T>::capsule, DLPack_Capsule_Destructor<T>);
END_HANDLE_TH_ERRORS
}
} // namespace
static PyObject* THPModule_toDLPack(
PyObject* self,
PyObject* args,
PyObject* kwargs) {
return THPModule_toDLPackImpl<DLManagedTensor>(self, args, kwargs);
}
static PyObject* THPModule_toDLPackVersioned(
PyObject* self,
PyObject* args,
PyObject* kwargs) {
return THPModule_toDLPackImpl<DLManagedTensorVersioned>(self, args, kwargs);
}
static PyObject* THPModule_fromDLPack(PyObject* _unused, PyObject* data) {
using namespace torch::autograd;
HANDLE_TH_ERRORS
auto tensor = torch::utils::tensor_fromDLPack(data);
return THPVariable_Wrap(tensor);
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_torchDeviceToDLDevice(
PyObject* _unused,
PyObject* data) {
HANDLE_TH_ERRORS
TORCH_CHECK(
THPDevice_Check(data),
"torchDeviceToDLDevice: expected torch.device argument.");
auto device = reinterpret_cast<THPDevice*>(data)->device;
auto dl_device = at::torchDeviceToDLDevice(device);
auto tuple = PyTuple_New(2);
if (!tuple) {
throw python_error(); // @allow-raw-throw
}
PyTuple_SET_ITEM(tuple, 0, THPUtils_packInt64(dl_device.device_type));
PyTuple_SET_ITEM(tuple, 1, THPUtils_packInt64(dl_device.device_id));
return tuple;
END_HANDLE_TH_ERRORS
}
struct TorchDLPackExchangeAPI : public DLPackExchangeAPI {
TorchDLPackExchangeAPI() {
header.version.major = DLPACK_MAJOR_VERSION;
header.version.minor = DLPACK_MINOR_VERSION;
header.prev_api = nullptr;
managed_tensor_allocator = ManagedTensorAllocator;
managed_tensor_from_py_object_no_sync = ManagedTensorFromPyObjectNoSync;
managed_tensor_to_py_object_no_sync = ManagedTensorToPyObjectNoSync;
dltensor_from_py_object_no_sync = DLTensorFromPyObjectNoSync;
current_work_stream = CurrentWorkStream;
}
static const DLPackExchangeAPI* Global() {
static TorchDLPackExchangeAPI inst;
return &inst;
}
private:
// Fast non-owning PyObject→DLTensor conversion
static int DLTensorFromPyObjectNoSync(void* py_obj, DLTensor* out) {
try {
// Use handle (non-owning) to avoid unnecessary refcount operations
py::handle handle(static_cast<PyObject*>(py_obj));
at::Tensor tensor = handle.cast<at::Tensor>();
at::toDLPackNonOwning(tensor, out);
return 0;
} catch (const std::exception& e) {
PyErr_SetString(PyExc_RuntimeError, e.what());
return -1;
}
}
// PyObject→DLManagedTensorVersioned conversion
static int ManagedTensorFromPyObjectNoSync(
void* py_obj,
DLManagedTensorVersioned** out) {
try {
py::handle handle(static_cast<PyObject*>(py_obj));
at::Tensor tensor = handle.cast<at::Tensor>();
*out = at::toDLPackVersioned(tensor);
return 0;
} catch (const std::exception& e) {
PyErr_SetString(PyExc_RuntimeError, e.what());
return -1;
}
}
// DLManagedTensorVersioned→PyObject conversion
static int ManagedTensorToPyObjectNoSync(
DLManagedTensorVersioned* src,
void** py_obj_out) {
try {
at::Tensor tensor = at::fromDLPackVersioned(src, nullptr);
*py_obj_out = THPVariable_Wrap(tensor);
return 0;
} catch (const std::exception& e) {
PyErr_SetString(PyExc_RuntimeError, e.what());
return -1;
}
}
// Allocate new tensor from prototype
static int ManagedTensorAllocator(
DLTensor* prototype,
DLManagedTensorVersioned** out,
void* error_ctx,
void (
*SetError)(void* error_ctx, const char* kind, const char* message)) {
try {
at::IntArrayRef shape(
prototype->shape, prototype->shape + prototype->ndim);
at::TensorOptions options =
at::TensorOptions()
.dtype(at::toScalarType(prototype->dtype))
.device(at::dlDeviceToTorchDevice(
prototype->device.device_type, prototype->device.device_id));
at::Tensor tensor = at::empty(shape, options);
*out = at::toDLPackVersioned(tensor);
return 0;
} catch (const std::exception& e) {
SetError(error_ctx, "MemoryError", e.what());
return -1;
}
}
// Get current CUDA/ROCm or XPU work stream
static int CurrentWorkStream(
DLDeviceType device_type,
int32_t device_id,
void** out_stream) {
try {
*out_stream = nullptr;
const auto acc_type = at::accelerator::getAccelerator(false);
if (!acc_type) {
return 0;
}
if (at::torchDeviceToDLDevice(*acc_type).device_type == device_type) {
*out_stream =
at::accelerator::getCurrentStream(device_id).native_handle();
}
return 0;
} catch (const std::exception& e) {
PyErr_SetString(PyExc_RuntimeError, e.what());
return -1;
}
}
};
static PyObject* THPModule_DLPackExchangeAPI(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
return PyCapsule_New(
const_cast<DLPackExchangeAPI*>(TorchDLPackExchangeAPI::Global()),
"dlpack_exchange_api",
nullptr);
END_HANDLE_TH_ERRORS
}
// Read-only exchange API: identical to TorchDLPackExchangeAPI except the two
// export slots advertise the tensor as read-only
// (DLPACK_FLAG_BITMASK_READ_ONLY) and export through const_data_ptr(). The
// import, allocator (which produces a fresh writable output) and stream slots
// are inherited unchanged.
struct TorchConstDLPackExchangeAPI : public TorchDLPackExchangeAPI {
TorchConstDLPackExchangeAPI() {
managed_tensor_from_py_object_no_sync = ManagedTensorFromPyObjectNoSync;
dltensor_from_py_object_no_sync = DLTensorFromPyObjectNoSync;
}
static const DLPackExchangeAPI* Global() {
static TorchConstDLPackExchangeAPI inst;
return &inst;
}
private:
static int DLTensorFromPyObjectNoSync(void* py_obj, DLTensor* out) {
try {
py::handle handle(static_cast<PyObject*>(py_obj));
at::Tensor tensor = handle.cast<at::Tensor>();
at::toDLPackNonOwning(tensor, out, /*read_only=*/true);
return 0;
} catch (const std::exception& e) {
PyErr_SetString(PyExc_RuntimeError, e.what());
return -1;
}
}
static int ManagedTensorFromPyObjectNoSync(
void* py_obj,
DLManagedTensorVersioned** out) {
try {
py::handle handle(static_cast<PyObject*>(py_obj));
at::Tensor tensor = handle.cast<at::Tensor>();
*out = at::toDLPackVersionedReadOnly(tensor);
return 0;
} catch (const std::exception& e) {
PyErr_SetString(PyExc_RuntimeError, e.what());
return -1;
}
}
};
static PyObject* THPModule_ConstDLPackExchangeAPI(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
return PyCapsule_New(
const_cast<DLPackExchangeAPI*>(TorchConstDLPackExchangeAPI::Global()),
"dlpack_exchange_api",
nullptr);
END_HANDLE_TH_ERRORS
}
static PyObject* THModule_getCppBacktrace(PyObject* _unused, PyObject* args) {
HANDLE_TH_ERRORS
size_t frames_to_skip = 0;
size_t maximum_number_of_frames = 0;
if (!PyArg_ParseTuple(
args, "LL", &frames_to_skip, &maximum_number_of_frames)) {
return nullptr;
}
return THPUtils_packString(
c10::get_backtrace(frames_to_skip, maximum_number_of_frames, true));
END_HANDLE_TH_ERRORS
}
static PyObject* THModule_rename_privateuse1_backend(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
THPUtils_checkString(arg),
"_rename_privateuse1_backend expects a str, but got ",
THPUtils_typename(arg));
const std::string backend_name = THPUtils_unpackString(arg);
c10::register_privateuse1_backend(backend_name);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THModule_get_privateuse1_backend_name(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
return THPUtils_packString(c10::get_privateuse1_backend());
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setAllowTF32CuDNN(PyObject* _unused, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_allow_tf32_cublas expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setAllowTF32CuDNN(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_allowTF32CuDNN(PyObject* _unused, PyObject* noargs) {
HANDLE_TH_ERRORS
if (at::globalContext().allowTF32CuDNN())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setFloat32MatmulPrecision(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
THPUtils_checkString(arg),
"set_float32_matmul_precision expects a str, "
"but got ",
THPUtils_typename(arg));
std::string s = THPUtils_unpackString(arg);
at::globalContext().setFloat32MatmulPrecision(s);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_float32MatmulPrecision(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
std::string s = "highest";
auto p = at::globalContext().float32MatmulPrecision();
if (p == at::Float32MatmulPrecision::HIGH) {
s = "high";
} else if (p == at::Float32MatmulPrecision::MEDIUM) {
s = "medium";
}
return THPUtils_packString(s);
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setSDPPriorityOrder(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
auto priority_order = THPUtils_unpackLongs(arg);
at::globalContext().setSDPPriorityOrder(priority_order);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_sDPPriorityOrder(
PyObject* _unused,
PyObject* noargs) {
auto ordervec = at::globalContext().sDPPriorityOrder();
auto order =
THPObjectPtr(PyList_New(static_cast<Py_ssize_t>(ordervec.size())));
for (const auto i : c10::irange(ordervec.size())) {
PyObject* i64 = THPUtils_packInt64(static_cast<int64_t>(ordervec[i]));
if (!i64)
return nullptr;
PyList_SET_ITEM(order.get(), i, i64);
}
return order.release();
}
static PyObject* THPModule_setSDPUseFlash(PyObject* _unused, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_sdp_use_math expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setSDPUseFlash(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_userEnabledFlashSDP(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().userEnabledFlashSDP())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setSDPUseFA3(PyObject* _unused, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_sdp_use_fa3 expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setSDPUseFA3(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_userEnabledFA3SDP(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().userEnabledFA3SDP())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setSDPUseMemEfficient(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_sdp_use_math expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setSDPUseMemEfficient(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* userEnabledMemEfficientSDP(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().userEnabledMemEfficientSDP())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setSDPUseMath(PyObject* _unused, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_sdp_use_math expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setSDPUseMath(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_userEnabledMathSDP(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().userEnabledMathSDP())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setAllowFP16BF16ReductionMathSDP(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_sdp_use_math expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setAllowFP16BF16ReductionMathSDP(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_allowFP16BF16ReductionMathSDP(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().allowFP16BF16ReductionMathSDP())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setSDPUseOverrideable(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_sdp_use_overrideable expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setSDPUseOverrideable(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_userEnabledOverrideableSDP(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().userEnabledOverrideableSDP())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setSDPUseCuDNN(PyObject* _unused, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_sdp_use_cudnn expects a bool, "
"but got %s",
THPUtils_typename(arg));
at::globalContext().setSDPUseCuDNN(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_userEnabledCuDNNSDP(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().userEnabledCuDNNSDP())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setUserEnabledCuDNN(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_enabled_cudnn expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setUserEnabledCuDNN(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_userEnabledCuDNN(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().userEnabledCuDNN())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setUserEnabledMkldnn(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_enabled_mkldnn expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setUserEnabledMkldnn(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_userEnabledMkldnn(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().userEnabledMkldnn())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setDeterministicCuDNN(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_deterministic_cudnn expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setDeterministicCuDNN(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_deterministicCuDNN(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().deterministicCuDNN())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setDeterministicMkldnn(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_deterministic_mkldnn expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setDeterministicMkldnn(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_deterministicMkldnn(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().deterministicMkldnn())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setDeterministicAlgorithms(
PyObject* _unused,
PyObject* args,
PyObject* kwargs) {
HANDLE_TH_ERRORS
static torch::PythonArgParser parser(
{"_set_deterministic_algorithms(bool mode, *, bool warn_only=False)"});
torch::ParsedArgs<2> parsed_args{};
auto r = parser.parse(args, kwargs, parsed_args);
bool mode = r.toBool(0);
bool warn_only = r.toBool(1);
at::globalContext().setDeterministicAlgorithms(mode, warn_only);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setAllowTF32OneDNN(
PyObject* _unsued,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"_set_onednn_allow_tf32 expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setAllowTF32OneDNN(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_allowTF32OneDNN(
PyObject* _unused,
PyObject* noargs) {
#ifdef USE_XPU
if (at::globalContext().allowTF32OneDNN())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
#else
Py_RETURN_NONE;
#endif
}
static PyObject* THPModule_deterministicAlgorithms(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().deterministicAlgorithms()) {
Py_RETURN_TRUE;
}
Py_RETURN_FALSE;
}
static PyObject* THPModule_deterministicAlgorithmsWarnOnly(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().deterministicAlgorithmsWarnOnly()) {
Py_RETURN_TRUE;
}
Py_RETURN_FALSE;
}
static PyObject* THPModule_setDeterministicFillUninitializedMemory(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg), "expected a bool, but got ", THPUtils_typename(arg));
at::globalContext().setDeterministicFillUninitializedMemory(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_deterministicFillUninitializedMemory(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().deterministicFillUninitializedMemory())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setUserEnabledNNPACK(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_enabled_NNPACK expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setUserEnabledNNPACK(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_userEnabledNNPACK(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().userEnabledNNPACK())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setWarnAlways(PyObject* _unused, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"setWarnOnlyOnce expects a bool, "
"but got ",
THPUtils_typename(arg));
c10::WarningUtils::set_warnAlways(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_warnAlways(PyObject* _unused, PyObject* noargs) {
if (c10::WarningUtils::get_warnAlways()) {
Py_RETURN_TRUE;
}
Py_RETURN_FALSE;
}
// Used only for testing C++ to Python warning translations.
static PyObject* THPModule_warn(PyObject* _unused, PyObject* noargs) {
HANDLE_TH_ERRORS
TORCH_WARN("Test message for TORCH_WARN");
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
// Used only for testing C++ to Python warning translations.
static PyObject* THPModule_warnDeprecation(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
TORCH_WARN_DEPRECATION("Test message for TORCH_WARN_DEPRECATION");
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setBenchmarkCuDNN(PyObject* _unused, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_benchmark_cudnn expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setBenchmarkCuDNN(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_benchmarkCuDNN(PyObject* _unused, PyObject* noargs) {
if (at::globalContext().benchmarkCuDNN()) {
Py_RETURN_TRUE;
}
Py_RETURN_FALSE;
}
static PyObject* THPModule_setCuDNNDepthwiseKernel(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
THPUtils_checkString(arg),
"set_cudnn_depthwise_kernel expects a string, "
"but got ",
THPUtils_typename(arg));
std::string mode = THPUtils_unpackString(arg);
at::globalContext().setCuDNNDepthwiseKernel(at::str2cudnn_depthwise(mode));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_getCuDNNDepthwiseKernel(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
auto mode =
at::cudnn_depthwise2str(at::globalContext().cudnnDepthwiseKernel());
return THPUtils_packString(mode);
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setImmediateMiopen(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_immediate_miopen expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setImmediateMiopen(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_immediateMiopen(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().immediateMiopen()) {
Py_RETURN_TRUE;
}
Py_RETURN_FALSE;
}
static PyObject* THPModule_setAllowTF32CuBLAS(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_allow_tf32_cublas expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setAllowTF32CuBLAS(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_allowTF32CuBLAS(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
if (at::globalContext().allowTF32CuBLAS()) {
Py_RETURN_TRUE;
}
Py_RETURN_FALSE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setPreferCublasltGroupedGemm(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_prefer_cublaslt_grouped_gemm expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setPreferCublasltGroupedGemm(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_preferCublasltGroupedGemm(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
if (at::globalContext().preferCublasltGroupedGemm()) {
Py_RETURN_TRUE;
}
Py_RETURN_FALSE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setAllowFP16ReductionCuBLAS(
PyObject* _unused,
PyObject* args) {
HANDLE_TH_ERRORS
PyObject* allow_reduction_obj = nullptr;
PyObject* allow_splitk_obj = Py_None;
if (!PyArg_ParseTuple(args, "O|O", &allow_reduction_obj, &allow_splitk_obj)) {
return nullptr;
}
TORCH_CHECK(
PyBool_Check(allow_reduction_obj),
"set_allow_fp16_reduction_cublas expects a bool for allow_reduced_precision, "
"but got ",
THPUtils_typename(allow_reduction_obj));
bool allow_reduction = Py_IsTrue(allow_reduction_obj);
bool allow_splitk = true;
if (!Py_IsNone(allow_splitk_obj)) {
TORCH_CHECK(
PyBool_Check(allow_splitk_obj),
"set_allow_fp16_reduction_cublas expects a bool for allow_splitk, "
"but got ",
THPUtils_typename(allow_splitk_obj));
allow_splitk = Py_IsTrue(allow_splitk_obj);
}
at::globalContext().setAllowFP16ReductionCuBLAS(
allow_reduction, allow_splitk);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_allowFP16ReductionCuBLAS(
PyObject* _unused,
PyObject* noargs) {
auto option = at::globalContext().allowFP16ReductionCuBLAS();
bool allow_reduced_precision =
option == at::CuBLASReductionOption::AllowReducedPrecisionWithSplitK;
bool allow_splitk = option !=
at::CuBLASReductionOption::DisallowReducedPrecisionDisallowSplitK;
return PyTuple_Pack(
2,
allow_reduced_precision ? Py_True : Py_False,
allow_splitk ? Py_True : Py_False);
}
static PyObject* THPModule_setAllowBF16ReductionCuBLAS(
PyObject* _unused,
PyObject* args) {
HANDLE_TH_ERRORS
PyObject* allow_reduction_obj = nullptr;
PyObject* allow_splitk_obj = Py_None;
if (!PyArg_ParseTuple(args, "O|O", &allow_reduction_obj, &allow_splitk_obj)) {
return nullptr;
}
TORCH_CHECK(
PyBool_Check(allow_reduction_obj),
"set_allow_bf16_reduction_cublas expects a bool for allow_reduced_precision, "
"but got ",
THPUtils_typename(allow_reduction_obj));
bool allow_reduction = Py_IsTrue(allow_reduction_obj);
bool allow_splitk = true;
if (!Py_IsNone(allow_splitk_obj)) {
TORCH_CHECK(
PyBool_Check(allow_splitk_obj),
"set_allow_bf16_reduction_cublas expects a bool for allow_splitk, "
"but got ",
THPUtils_typename(allow_splitk_obj));
allow_splitk = Py_IsTrue(allow_splitk_obj);
}
at::globalContext().setAllowBF16ReductionCuBLAS(
allow_reduction, allow_splitk);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_allowBF16ReductionCuBLAS(
PyObject* _unused,
PyObject* noargs) {
auto option = at::globalContext().allowBF16ReductionCuBLAS();
bool allow_reduced_precision =
option == at::CuBLASReductionOption::AllowReducedPrecisionWithSplitK;
bool allow_splitk = option !=
at::CuBLASReductionOption::DisallowReducedPrecisionDisallowSplitK;
return PyTuple_Pack(
2,
allow_reduced_precision ? Py_True : Py_False,
allow_splitk ? Py_True : Py_False);
}
static PyObject* THPModule_setAllowFP16AccumulationCuBLAS(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_allow_fp16_accumulation_cublas expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setAllowFP16AccumulationCuBLAS(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_allowFP16AccumulationCuBLAS(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().allowFP16AccumulationCuBLAS()) {
Py_RETURN_TRUE;
}
Py_RETURN_FALSE;
}
static PyObject* THPModule_setAllowFP16ReductionCPU(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"set_allow_fp16_reduction_cpu expects a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setAllowFP16ReductionCPU(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_allowFP16ReductionCPU(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().allowFP16ReductionCPU()) {
Py_RETURN_TRUE;
}
Py_RETURN_FALSE;
}
static PyObject* THPModule_setFlushDenormal(PyObject* _unused, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"flush_denormal expects a bool, "
"but got ",
THPUtils_typename(arg));
if (!at::globalContext().setFlushDenormal(Py_IsTrue(arg))) {
Py_RETURN_FALSE;
};
Py_RETURN_TRUE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_getDefaultDtype(PyObject* _unused, PyObject* arg) {
HANDLE_TH_ERRORS
auto scalar_type = torch::tensors::get_default_scalar_type();
return Py_NewRef(torch::getTHPDtype(scalar_type));
END_HANDLE_TH_ERRORS
}
// Returns the set of dtypes that are fully realized Python singletons: all
// floating/complex/integer/bool/float8/float4/bits types. Excludes quantized
// dtypes and the placeholder sub-byte int/uint types, matching the dtypes
// tracked in torch.utils._dtype_abbrs.
static PyObject* THPModule_getAllDtypes(PyObject* _unused, PyObject* noargs) {
HANDLE_TH_ERRORS
#define DEFINE_SCALAR_TYPE(_1, n) at::ScalarType::n,
auto all_scalar_types = {
AT_FORALL_SCALAR_TYPES_WITH_COMPLEX_AND_QINTS(DEFINE_SCALAR_TYPE)};
#undef DEFINE_SCALAR_TYPE
auto is_subbyte_dummy = [](at::ScalarType t) {
switch (t) {
case at::ScalarType::UInt1:
case at::ScalarType::UInt2:
case at::ScalarType::UInt3:
case at::ScalarType::UInt4:
case at::ScalarType::UInt5:
case at::ScalarType::UInt6:
case at::ScalarType::UInt7:
case at::ScalarType::Int1:
case at::ScalarType::Int2:
case at::ScalarType::Int3:
case at::ScalarType::Int4:
case at::ScalarType::Int5:
case at::ScalarType::Int6:
case at::ScalarType::Int7:
return true;
default:
return false;
}
};
THPObjectPtr result(PyList_New(0));
if (!result)
throw python_error();
for (auto scalar_type : all_scalar_types) {
if (c10::isQIntType(scalar_type) || is_subbyte_dummy(scalar_type)) {
continue;
}
auto* dtype = reinterpret_cast<PyObject*>(torch::getTHPDtype(scalar_type));
if (PyList_Append(result.get(), dtype) < 0) {
throw python_error();
}
}
return result.release();
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_getDefaultDevice(PyObject* _unused, PyObject* arg) {
HANDLE_TH_ERRORS
return THPUtils_packString(c10::DeviceTypeName(
dispatchKeyToDeviceType(torch::tensors::get_default_dispatch_key()),
/*lower_case=*/true));
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setQEngine(PyObject* /* unused */, PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
THPUtils_checkLong(arg),
"set_qengine expects an int, "
"but got ",
THPUtils_typename(arg));
auto qengine = THPUtils_unpackLong(arg);
// NOLINTNEXTLINE(clang-analyzer-optin.core.EnumCastOutOfRange)
at::globalContext().setQEngine(static_cast<at::QEngine>(qengine));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_qEngine(PyObject* _unused, PyObject* noargs) {
return THPUtils_packInt64(
static_cast<int64_t>(at::globalContext().qEngine()));
}
static PyObject* THPModule_supportedQEngines(
PyObject* _unused,
PyObject* noargs) {
const auto& qengines = at::globalContext().supportedQEngines();
auto list =
THPObjectPtr(PyList_New(static_cast<Py_ssize_t>(qengines.size())));
if (!list)
return nullptr;
for (const auto i : c10::irange(qengines.size())) {
PyObject* i64 = THPUtils_packInt64(static_cast<int64_t>(qengines[i]));
if (!i64)
return nullptr;
PyList_SET_ITEM(list.get(), i, i64);
}
return list.release();
}
static PyObject* THPModule_isEnabledXNNPACK(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().isXNNPACKAvailable())
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_setCheckSparseTensorInvariants(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
if (Py_IsNone(arg)) {
at::globalContext().setCheckSparseTensorInvariants(std::nullopt);
} else {
TORCH_CHECK(
PyBool_Check(arg),
"set_check_sparse_tensor_invariants expects a bool or None, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setCheckSparseTensorInvariants(Py_IsTrue(arg));
}
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_checkSparseTensorInvariants(
PyObject* _unused,
PyObject* noargs) {
if (at::globalContext().checkSparseTensorInvariants().value_or(false))
Py_RETURN_TRUE;
else
Py_RETURN_FALSE;
}
static PyObject* THPModule_willEngineExecuteNode(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
bool isTHPFunction = THPFunction_Check(arg);
bool isTHPCppFunction = torch::autograd::THPCppFunction_Check(arg);
TORCH_CHECK(
isTHPFunction || isTHPCppFunction,
"_will_engine_execute_node expects an grad_fn, "
"but got ",
THPUtils_typename(arg));
const auto exec_info = torch::autograd::get_current_graph_task_exec_info();
TORCH_CHECK(
exec_info,
"_get_should_execute_nodes should only be called during the backward pass");
torch::autograd::Node* node = nullptr;
c10::intrusive_ptr<torch::autograd::Node> node_sp;
if (isTHPFunction) {
node_sp = (reinterpret_cast<THPFunction*>(arg))->cdata;
node = node_sp.get();
} else {
node =
(reinterpret_cast<torch::autograd::THPCppFunction*>(arg))->cdata.get();
}
const auto nodes_in_graph =
torch::autograd::get_current_graph_task_nodes_in_graph();
bool ret = nodes_in_graph->find(node) != nodes_in_graph->end();
if (ret && !exec_info->empty()) {
auto it = exec_info->find(node);
if (it == exec_info->end() || !it->second.should_execute()) {
ret = false;
} else {
TORCH_CHECK(
!(node->topological_nr() == 0 && it->second.captures_),
"A leaf node was passed to _will_engine_execute_node but we are "
"currently running autograd.grad(). This is currently not supported.");
}
}
if (ret) {
Py_RETURN_TRUE;
} else {
Py_RETURN_FALSE;
}
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_getCurrentGraphTaskExecutionOrder(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
std::vector<torch::autograd::Node*> nodes =
torch::autograd::get_current_graph_task_execution_order();
TORCH_CHECK(
!nodes.empty(),
"_current_graph_task_execution_order should only be called during the backward pass");
auto list = THPObjectPtr(PyList_New(static_cast<Py_ssize_t>(nodes.size())));
if (!list)
return nullptr;
for (const auto i : c10::irange(nodes.size())) {
// This node is guaranteed to be alive since the backward is still running
PyObject* pyobj_node =
torch::autograd::functionToPyObject(nodes[i]->getptr());
PyList_SET_ITEM(list.get(), i, pyobj_node);
}
return list.release();
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_getCurrentGraphTaskId(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
return THPUtils_packInt64(torch::autograd::get_current_graph_task_id());
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_getCurrentNode(PyObject* _unused, PyObject* noargs) {
HANDLE_TH_ERRORS
return torch::autograd::functionToPyObject(
torch::autograd::get_current_node());
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_isDefaultMobileCPUAllocatorSet(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
return PyBool_FromLong(at::globalContext().isDefaultMobileCPUAllocatorSet());
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_setDefaultMobileCPUAllocator(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
at::globalContext().setDefaultMobileCPUAllocator();
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_unsetDefaultMobileCPUAllocator(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
at::globalContext().unsetDefaultMobileCPUAllocator();
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_vmapmode_increment_nesting(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
return THPUtils_packInt64(at::impl::VmapMode::increment_nesting());
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_vmapmode_decrement_nesting(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
return THPUtils_packInt64(at::impl::VmapMode::decrement_nesting());
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_set_display_vmap_fallback_warnings_mode(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"enabled must be a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setDisplayVmapFallbackWarnings(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_are_vmap_fallback_warnings_enabled(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
if (at::globalContext().areVmapFallbackWarningsEnabled()) {
Py_RETURN_TRUE;
} else {
Py_RETURN_FALSE;
}
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_set_warn_on_accumulate_grad_stream_mismatch(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"enabled must be a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setWarnOnAccumulateGradStreamMismatch(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_warn_on_accumulate_grad_stream_mismatch(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
if (at::globalContext().warnOnAccumulateGradStreamMismatch()) {
Py_RETURN_TRUE;
} else {
Py_RETURN_FALSE;
}
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_set_override_stale_capture_stream(
PyObject* _unused,
PyObject* arg) {
HANDLE_TH_ERRORS
TORCH_CHECK(
PyBool_Check(arg),
"enabled must be a bool, "
"but got ",
THPUtils_typename(arg));
at::globalContext().setOverrideStaleCaptureStream(Py_IsTrue(arg));
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject* THPModule_override_stale_capture_stream(
PyObject* _unused,
PyObject* noargs) {
HANDLE_TH_ERRORS
if (at::globalContext().overrideStaleCaptureStream()) {
Py_RETURN_TRUE;
} else {
Py_RETURN_FALSE;
}
END_HANDLE_TH_ERRORS
}
static PyObject* THCPModule_ensureCUDADeviceGuardSet(
PyObject* self,
PyObject* noargs) {
HANDLE_TH_ERRORS
c10::impl::ensureCUDADeviceGuardSet();
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
struct TorchModuleState {
PyObject* log_api_usage_seen; // dict used by _log_api_usage_once
};
static int torchmodule_traverse(PyObject* mod, visitproc visit, void* arg) {
auto* state = static_cast<TorchModuleState*>(PyModule_GetState(mod));
Py_VISIT(state->log_api_usage_seen);
return 0;
}
static int torchmodule_clear(PyObject* mod) {
auto* state = static_cast<TorchModuleState*>(PyModule_GetState(mod));
Py_CLEAR(state->log_api_usage_seen);
return 0;
}
static void torchmodule_free(void* mod) {
torchmodule_clear((PyObject*)mod);
}
// Thread-safe in free-threaded Python.
static PyObject* LogAPIUsageOnceFromPython(PyObject* self, PyObject* event) {
auto* state = static_cast<TorchModuleState*>(PyModule_GetState(self));
PyObject* api_usage_seen = state->log_api_usage_seen;
int found = PyDict_Contains(api_usage_seen, event);
if (found < 0) {
return nullptr;
} else if (found != 0) {
Py_RETURN_NONE;
}
const char* event_str = PyUnicode_AsUTF8(event);
if (!event_str) {
return nullptr;
}
// Returns 0 if we inserted, 1 if already present, -1 on error.
int rc = PyDict_SetDefaultRef(api_usage_seen, event, Py_None, nullptr);
if (rc < 0) {
return nullptr;
}
if (rc == 0) {
c10::LogAPIUsage(event_str);
}
Py_RETURN_NONE;
}
static std::initializer_list<PyMethodDef> TorchMethods = {
{"_log_api_usage_once", LogAPIUsageOnceFromPython, METH_O, nullptr},
{"_initExtension", THPModule_initExtension, METH_O, nullptr},
{"_autograd_init", THPAutograd_initExtension, METH_NOARGS, nullptr},
{"_add_docstr", THPModule_addDocStr, METH_VARARGS, nullptr},
{"_swap_tensor_impl", THPModule_swap_tensor_impl, METH_VARARGS, nullptr},
{"_init_names", THPModule_initNames, METH_O, nullptr},
{"_has_distributed", THPModule_hasDistributed, METH_NOARGS, nullptr},
{"_set_default_tensor_type",
THPModule_setDefaultTensorType,
METH_O,
nullptr},
{"_set_default_dtype", THPModule_setDefaultDtype, METH_O, nullptr},
{"_infer_size", THPModule_inferSize, METH_VARARGS, nullptr},
{"_abort", THPModule_abort, METH_NOARGS, nullptr},
{"_crash_if_csrc_asan", THPModule_crashIfCsrcASAN, METH_O, nullptr},
{"_crash_if_csrc_ubsan", THPModule_crashIfCsrcUBSAN, METH_O, nullptr},
{"_crash_if_vptr_ubsan", THPModule_crashIfvptrUBSAN, METH_NOARGS, nullptr},
{"_crash_if_aten_asan", THPModule_crashIfATenASAN, METH_O, nullptr},
{"_crash_if_debug_asserts_fail",
THPModule_crashIfDebugAssertsFail,
METH_O,
nullptr},
{"_show_config", THPModule_showConfig, METH_NOARGS, nullptr},
{"_cxx_flags", THPModule_cxxFlags, METH_NOARGS, nullptr},
{"_parallel_info", THPModule_parallelInfo, METH_NOARGS, nullptr},
{"_get_cpu_capability", THPModule_getCpuCapability, METH_NOARGS, nullptr},
{"_set_backcompat_broadcast_warn",
THPModule_setBackcompatBroadcastWarn,
METH_O,
nullptr},
{"_get_backcompat_broadcast_warn",
THPModule_getBackcompatBroadcastWarn,
METH_NOARGS,
nullptr},
{"_set_backcompat_keepdim_warn",
THPModule_setBackcompatKeepdimWarn,
METH_O,
nullptr},
{"_get_backcompat_keepdim_warn",
THPModule_getBackcompatKeepdimWarn,
METH_NOARGS,
nullptr},
{"get_num_threads", THPModule_getNumThreads, METH_NOARGS, nullptr},
{"set_num_threads", THPModule_setNumThreads, METH_O, nullptr},
{"get_num_interop_threads",
THPModule_getNumInteropThreads,
METH_NOARGS,
nullptr},
{"set_num_interop_threads",
THPModule_setNumInteropThreads,
METH_O,
nullptr},
{"_get_flash_sdp_enabled",
THPModule_userEnabledFlashSDP,
METH_NOARGS,
nullptr},
{"_set_sdp_priority_order", THPModule_setSDPPriorityOrder, METH_O, nullptr},
{"_get_sdp_priority_order",
THPModule_sDPPriorityOrder,
METH_NOARGS,
nullptr},
{"_set_sdp_use_flash", THPModule_setSDPUseFlash, METH_O, nullptr},
{"_get_fa3_sdp_enabled", THPModule_userEnabledFA3SDP, METH_NOARGS, nullptr},
{"_set_sdp_use_fa3", THPModule_setSDPUseFA3, METH_O, nullptr},
{"_get_mem_efficient_sdp_enabled",
userEnabledMemEfficientSDP,
METH_NOARGS,
nullptr},
{"_set_sdp_use_mem_efficient",
THPModule_setSDPUseMemEfficient,
METH_O,
nullptr},
{"_get_math_sdp_enabled",
THPModule_userEnabledMathSDP,
METH_NOARGS,
nullptr},
{"_set_sdp_use_math", THPModule_setSDPUseMath, METH_O, nullptr},
{"_get_math_sdp_allow_fp16_bf16_reduction",
THPModule_allowFP16BF16ReductionMathSDP,
METH_NOARGS,
nullptr},
{"_set_math_sdp_allow_fp16_bf16_reduction",
THPModule_setAllowFP16BF16ReductionMathSDP,
METH_O,
nullptr},
{"_get_overrideable_sdp_enabled",
THPModule_userEnabledOverrideableSDP,
METH_NOARGS,
nullptr},
{"_set_sdp_use_overrideable",
THPModule_setSDPUseOverrideable,
METH_O,
nullptr},
{"_get_cudnn_sdp_enabled",
THPModule_userEnabledCuDNNSDP,
METH_NOARGS,
nullptr},
{"_set_sdp_use_cudnn", THPModule_setSDPUseCuDNN, METH_O, nullptr},
{"_get_cudnn_enabled", THPModule_userEnabledCuDNN, METH_NOARGS, nullptr},
{"_set_cudnn_enabled", THPModule_setUserEnabledCuDNN, METH_O, nullptr},
{"_get_mkldnn_enabled", THPModule_userEnabledMkldnn, METH_NOARGS, nullptr},
{"_set_mkldnn_enabled", THPModule_setUserEnabledMkldnn, METH_O, nullptr},
{"_get_cudnn_allow_tf32", THPModule_allowTF32CuDNN, METH_NOARGS, nullptr},
{"_set_cudnn_allow_tf32", THPModule_setAllowTF32CuDNN, METH_O, nullptr},
{"_get_onednn_allow_tf32", THPModule_allowTF32OneDNN, METH_NOARGS, nullptr},
{"_set_onednn_allow_tf32", THPModule_setAllowTF32OneDNN, METH_O, nullptr},
{"_get_cudnn_benchmark", THPModule_benchmarkCuDNN, METH_NOARGS, nullptr},
{"_set_cudnn_benchmark", THPModule_setBenchmarkCuDNN, METH_O, nullptr},
{"_get_cudnn_depthwise_kernel",
THPModule_getCuDNNDepthwiseKernel,
METH_NOARGS,
nullptr},
{"_set_cudnn_depthwise_kernel",
THPModule_setCuDNNDepthwiseKernel,
METH_O,
nullptr},
{"_get_miopen_immediate", THPModule_immediateMiopen, METH_NOARGS, nullptr},
{"_set_miopen_immediate", THPModule_setImmediateMiopen, METH_O, nullptr},
{"_get_cudnn_deterministic",
THPModule_deterministicCuDNN,
METH_NOARGS,
nullptr},
{"_set_cudnn_deterministic",
THPModule_setDeterministicCuDNN,
METH_O,
nullptr},
{"_get_mkldnn_deterministic",
THPModule_deterministicMkldnn,
METH_NOARGS,
nullptr},
{"_set_mkldnn_deterministic",
THPModule_setDeterministicMkldnn,
METH_O,
nullptr},
{"_get_deterministic_algorithms",
THPModule_deterministicAlgorithms,
METH_NOARGS,
nullptr},
{"_get_deterministic_algorithms_warn_only",
THPModule_deterministicAlgorithmsWarnOnly,
METH_NOARGS,
nullptr},
{"_set_deterministic_algorithms",
castPyCFunctionWithKeywords(THPModule_setDeterministicAlgorithms),
METH_VARARGS | METH_KEYWORDS,
nullptr},
{"_get_deterministic_fill_uninitialized_memory",
THPModule_deterministicFillUninitializedMemory,
METH_NOARGS,
nullptr},
{"_set_deterministic_fill_uninitialized_memory",
THPModule_setDeterministicFillUninitializedMemory,
METH_O,
nullptr},
{"_get_nnpack_enabled", THPModule_userEnabledNNPACK, METH_NOARGS, nullptr},
{"_set_nnpack_enabled", THPModule_setUserEnabledNNPACK, METH_O, nullptr},
{"_get_warnAlways", THPModule_warnAlways, METH_NOARGS, nullptr},
{"_set_warnAlways", THPModule_setWarnAlways, METH_O, nullptr},
{"_warn", THPModule_warn, METH_NOARGS, nullptr},
{"_warn_deprecation", THPModule_warnDeprecation, METH_NOARGS, nullptr},
{"_get_cublas_allow_tf32", THPModule_allowTF32CuBLAS, METH_NOARGS, nullptr},
{"_set_cublas_allow_tf32", THPModule_setAllowTF32CuBLAS, METH_O, nullptr},
{"_get_cublaslt_prefer_grouped_gemm",
THPModule_preferCublasltGroupedGemm,
METH_NOARGS,
nullptr},
{"_set_cublaslt_prefer_grouped_gemm",
THPModule_setPreferCublasltGroupedGemm,
METH_O,
nullptr},
{"_get_float32_matmul_precision",
THPModule_float32MatmulPrecision,
METH_NOARGS,
nullptr},
{"_set_float32_matmul_precision",
THPModule_setFloat32MatmulPrecision,
METH_O,
nullptr},
{"_get_cublas_allow_fp16_reduced_precision_reduction",
THPModule_allowFP16ReductionCuBLAS,
METH_NOARGS,
nullptr},
{"_set_cublas_allow_fp16_reduced_precision_reduction",
THPModule_setAllowFP16ReductionCuBLAS,
METH_VARARGS,
nullptr},
{"_get_cublas_allow_bf16_reduced_precision_reduction",
THPModule_allowBF16ReductionCuBLAS,
METH_NOARGS,
nullptr},
{"_set_cublas_allow_bf16_reduced_precision_reduction",
THPModule_setAllowBF16ReductionCuBLAS,
METH_VARARGS,
nullptr},
{"_get_cublas_allow_fp16_accumulation",
THPModule_allowFP16AccumulationCuBLAS,
METH_NOARGS,
nullptr},
{"_set_cublas_allow_fp16_accumulation",
THPModule_setAllowFP16AccumulationCuBLAS,
METH_O,
nullptr},
{"_get_cpu_allow_fp16_reduced_precision_reduction",
THPModule_allowFP16ReductionCPU,
METH_NOARGS,
nullptr},
{"_set_cpu_allow_fp16_reduced_precision_reduction",
THPModule_setAllowFP16ReductionCPU,
METH_O,
nullptr},
{"_vmapmode_increment_nesting",
THPModule_vmapmode_increment_nesting,
METH_NOARGS,
nullptr},
{"_vmapmode_decrement_nesting",
THPModule_vmapmode_decrement_nesting,
METH_NOARGS,
nullptr},
{"_debug_only_display_vmap_fallback_warnings",
THPModule_set_display_vmap_fallback_warnings_mode,
METH_O,
nullptr},
{"_debug_only_are_vmap_fallback_warnings_enabled",
THPModule_are_vmap_fallback_warnings_enabled,
METH_NOARGS,
nullptr},
{"_set_warn_on_accumulate_grad_stream_mismatch",
THPModule_set_warn_on_accumulate_grad_stream_mismatch,
METH_O,
nullptr},
{"_warn_on_accumulate_grad_stream_mismatch",
THPModule_warn_on_accumulate_grad_stream_mismatch,
METH_NOARGS,
nullptr},
{"_set_override_stale_capture_stream",
THPModule_set_override_stale_capture_stream,
METH_O,
nullptr},
{"_override_stale_capture_stream",
THPModule_override_stale_capture_stream,
METH_NOARGS,
nullptr},
{"_to_dlpack",
castPyCFunctionWithKeywords(THPModule_toDLPack),
METH_VARARGS | METH_KEYWORDS,
nullptr},
{"_to_dlpack_versioned",
castPyCFunctionWithKeywords(THPModule_toDLPackVersioned),
METH_VARARGS | METH_KEYWORDS,
nullptr},
{"_from_dlpack", THPModule_fromDLPack, METH_O, nullptr},
{"_torchDeviceToDLDevice",
THPModule_torchDeviceToDLDevice,
METH_O,
nullptr},
{"_dlpack_exchange_api", THPModule_DLPackExchangeAPI, METH_NOARGS, nullptr},
{"_const_dlpack_exchange_api",
THPModule_ConstDLPackExchangeAPI,
METH_NOARGS,
nullptr},
{"_get_cpp_backtrace", THModule_getCppBacktrace, METH_VARARGS, nullptr},
{"_rename_privateuse1_backend",
THModule_rename_privateuse1_backend,
METH_O,
nullptr},
{"_get_privateuse1_backend_name",
THModule_get_privateuse1_backend_name,
METH_NOARGS,
nullptr},
{"set_flush_denormal", THPModule_setFlushDenormal, METH_O, nullptr},
{"get_default_dtype", THPModule_getDefaultDtype, METH_NOARGS, nullptr},
{"_get_all_dtypes", THPModule_getAllDtypes, METH_NOARGS, nullptr},
{"_get_default_device", THPModule_getDefaultDevice, METH_NOARGS, nullptr},
{"_get_qengine", THPModule_qEngine, METH_NOARGS, nullptr},
{"_set_qengine", THPModule_setQEngine, METH_O, nullptr},
{"_supported_qengines", THPModule_supportedQEngines, METH_NOARGS, nullptr},
{"_is_xnnpack_enabled", THPModule_isEnabledXNNPACK, METH_NOARGS, nullptr},
{"_set_check_sparse_tensor_invariants",
THPModule_setCheckSparseTensorInvariants,
METH_O,
nullptr},
{"_check_sparse_tensor_invariants",
THPModule_checkSparseTensorInvariants,
METH_NOARGS,
nullptr},
{"_will_engine_execute_node",
THPModule_willEngineExecuteNode,
METH_O,
nullptr},
{"_current_graph_task_execution_order",
THPModule_getCurrentGraphTaskExecutionOrder,
METH_NOARGS,
nullptr},
{"_current_graph_task_id",
THPModule_getCurrentGraphTaskId,
METH_NOARGS,
nullptr},
{"_current_autograd_node", THPModule_getCurrentNode, METH_NOARGS, nullptr},
{"_is_default_mobile_cpu_allocator_set",
THPModule_isDefaultMobileCPUAllocatorSet,
METH_NOARGS,
nullptr},
{"_set_default_mobile_cpu_allocator",
THPModule_setDefaultMobileCPUAllocator,
METH_NOARGS,
nullptr},
{"_unset_default_mobile_cpu_allocator",
THPModule_unsetDefaultMobileCPUAllocator,
METH_NOARGS,
nullptr},
{"_is_torch_function_enabled",
THPModule_isEnabledTorchFunction,
METH_NOARGS,
nullptr},
{"_is_torch_function_all_disabled",
THPModule_isAllDisabledTorchFunction,
METH_NOARGS,
nullptr},
{"_disabled_torch_function_impl",
THPModule_disable_torch_function,
METH_VARARGS,
nullptr},
{"_disabled_torch_dispatch_impl",
THPModule_disable_torch_dispatch,
METH_VARARGS,
nullptr},
{"_skip_one_hop_torch_function",
THPModule_skip_one_hop_torch_function,
METH_VARARGS,
nullptr},
{"_peek_should_skip_torch_function",
THPModule_peek_should_skip_torch_function,
METH_NOARGS,
nullptr},
{"_set_skip_next_torch_function",
THPModule_set_skip_next_torch_function,
METH_O,
nullptr},
{"_has_torch_function", THPModule_has_torch_function, METH_O, nullptr},
{"_has_torch_function_unary",
THPModule_has_torch_function_unary,
METH_O,
nullptr},
{"_has_torch_function_variadic",
reinterpret_cast<PyCFunction>(
reinterpret_cast<void (*)()>(THPModule_has_torch_function_variadic)),
METH_FASTCALL,
nullptr},
{"_ensureCUDADeviceGuardSet",
THCPModule_ensureCUDADeviceGuardSet,
METH_NOARGS,
nullptr},
{nullptr, nullptr, 0, nullptr}
};
#ifdef USE_CUDA
// NOLINTBEGIN(misc-use-internal-linkage)
void THCPStream_init(PyObject* module);
void THCPEvent_init(PyObject* module);
void THCPGraph_init(PyObject* module);
void THCPMemPool_init(PyObject* module);
PyMethodDef* THCPModule_methods();
namespace torch::cuda {
void initModule(PyObject* module);
} // namespace torch::cuda
// NOLINTEND(misc-use-internal-linkage)
#endif
#ifdef USE_XPU
PyMethodDef* THXPModule_methods();
void THXPStream_init(PyObject* module);
void THXPEvent_init(PyObject* module);
void THXPMemPool_init(PyObject* module);
void THXPGraph_init(PyObject* module);
namespace torch::xpu {
void initModule(PyObject* module);
} // namespace torch::xpu
#endif
static std::vector<PyMethodDef> methods;
static void LogAPIUsageMetadataFromPython(
const std::string& event,
const std::map<std::string, std::string>& metadata_map) {
c10::LogAPIUsageMetadata(event, metadata_map);
}
// Weak reference to tensor, used to test a tensor isn't leaked
class WeakTensorRef {
c10::weak_intrusive_ptr<c10::TensorImpl> weakref_;
public:
WeakTensorRef(const at::Tensor& t) : weakref_(t.getIntrusivePtr()) {}
bool expired() {
return weakref_.expired();
}
};
namespace {
using SigHandler = void (*)(int);
SigHandler* _getOldHandler(int signum) {
#define SIG_CHECK(n) \
if (signum == (n)) { \
static SigHandler handler = nullptr; \
return &handler; \
}
SIG_CHECK(SIGSEGV);
SIG_CHECK(SIGILL);
TORCH_CHECK(false, "unexpected signal number");
#undef SIG_CHECK
}
extern "C" void _signalHandler(int signum) {
// Note that technically there's not much you're allowed to do here - but
// we're probably dying anyway so give it a try...
auto oldAction = *_getOldHandler(signum);
*_getOldHandler(signum) = nullptr;
// If we hit another signal don't run this handler again.
std::signal(signum, oldAction);
#ifdef _WIN32
const char* signame = "<unknown>";
#else
const char* signame = strsignal(signum);
#endif
fprintf(
stderr,
"Process %d crashed with signal %s (%d):\n",
getpid(),
signame,
signum);
auto bt = c10::get_backtrace();
fwrite(bt.data(), 1, bt.size(), stderr);
// Try to run the previous signal handler
if (oldAction != SIG_IGN && oldAction != SIG_DFL) {
oldAction(signum);
}
if (oldAction != SIG_IGN) {
_exit(-1);
}
}
void _initCrashHandler() {
*_getOldHandler(SIGILL) = std::signal(SIGILL, _signalHandler);
*_getOldHandler(SIGSEGV) = std::signal(SIGSEGV, _signalHandler);
}
} // anonymous namespace
extern "C" TORCH_PYTHON_API PyObject* initModule();
// separate decl and defn for msvc error C2491
PyObject* initModule() {
HANDLE_TH_ERRORS
c10::initLogging();
c10::set_terminate_handler();
at::internal::lazy_init_num_threads();
C10_LOG_API_USAGE_ONCE("torch.python.import");
#define ASSERT_TRUE(cmd) \
if (!(cmd)) \
return nullptr
THPUtils_addPyMethodDefs(methods, std::data(TorchMethods));
THPUtils_addPyMethodDefs(methods, DataLoaderMethods);
THPUtils_addPyMethodDefs(methods, torch::autograd::python_functions());
THPUtils_addPyMethodDefs(methods, torch::multiprocessing::python_functions());
THPUtils_addPyMethodDefs(methods, torch::mps::python_functions());
#ifdef USE_CUDA
THPUtils_addPyMethodDefs(methods, THCPModule_methods());
#endif
#ifdef USE_XPU
THPUtils_addPyMethodDefs(methods, THXPModule_methods());
#endif
#if defined(USE_DISTRIBUTED) && defined(USE_C10D)
THPUtils_addPyMethodDefs(
methods, torch::distributed::c10d::python_functions());
#ifndef _WIN32
THPUtils_addPyMethodDefs(
methods, torch::distributed::rpc::python_functions());
THPUtils_addPyMethodDefs(
methods, torch::distributed::autograd::python_functions());
THPUtils_addPyMethodDefs(
methods, torch::distributed::rpc::testing::python_functions());
#endif
#endif
static struct PyModuleDef torchmodule = {
PyModuleDef_HEAD_INIT,
"torch._C",
nullptr,
sizeof(TorchModuleState),
methods.data(),
nullptr, // m_slots
torchmodule_traverse,
torchmodule_clear,
torchmodule_free};
module = PyModule_Create(&torchmodule);
ASSERT_TRUE(module);
auto* mod_state = static_cast<TorchModuleState*>(PyModule_GetState(module));
mod_state->log_api_usage_seen = PyDict_New();
ASSERT_TRUE(mod_state->log_api_usage_seen);
#ifdef Py_GIL_DISABLED
PyUnstable_Module_SetGIL(module, Py_MOD_GIL_NOT_USED);
#endif
ASSERT_TRUE(THPGenerator_init(module));
ASSERT_TRUE(THPException_init(module));
THPSize_init(module);
THPDtype_init(module);
THPDTypeInfo_init(module);
THPLayout_init(module);
THPMemoryFormat_init(module);
THPQScheme_init(module);
THPDevice_init(module);
THPStream_init(module);
THPEvent_init(module);
NodeBase_init(module);
NodeIter_init(module);
ASSERT_TRUE(THPVariable_initModule(module));
ASSERT_TRUE(THPFunction_initModule(module));
ASSERT_TRUE(THPEngine_initModule(module));
// NOTE: We need to be able to access OperatorExportTypes from ONNX for use in
// the export side of JIT, so this ONNX init needs to appear before the JIT
// init.
torch::onnx::initONNXBindings(module);
torch::autograd::initEnumTag(module);
torch::jit::initJITBindings(module);
torch::monitor::initMonitorBindings(module);
torch::impl::dispatch::initDispatchBindings(module);
torch::dynamo::initDynamoBindings(module);
torch::functorch::impl::initFuncTorchBindings(module);
torch::throughput_benchmark::initThroughputBenchmarkBindings(module);
torch::autograd::initReturnTypes(module);
torch::autograd::initNNFunctions(module);
torch::autograd::initFFTFunctions(module);
torch::autograd::initLinalgFunctions(module);
torch::autograd::initNestedFunctions(module);
torch::autograd::initSparseFunctions(module);
torch::autograd::initSpecialFunctions(module);
torch::autograd::init_legacy_variable(module);
torch::profiler::initPythonBindings(module);
torch::python::init_bindings(module);
torch::lazy::initLazyBindings(module);
torch::_export::initExportBindings(module);
torch::inductor::initAOTIRunnerBindings(module);
torch::inductor::initAOTIPackageBindings(module);
#ifdef USE_ITT
torch::profiler::initIttBindings(module);
#endif
#ifdef USE_CUDA
torch::cuda::initModule(module);
#endif
#if defined(USE_CUDA)
ASSERT_TRUE(StaticCudaLauncher_init(module));
ASSERT_TRUE(FastCudaLauncher_init(module));
#endif
#if defined(USE_XPU) && !defined(_WIN32)
ASSERT_TRUE(StaticXpuLauncher_init(module));
#endif
#ifdef USE_MPS
torch::mps::initModule(module);
#endif
#ifdef USE_XPU
torch::xpu::initModule(module);
#endif
torch::mtia::initModule(module);
torch::cpu::initModule(module);
torch::accelerator::initModule(module);
torch::instruction_counter::initModule(module);
torch::acc::initModule(module);
torch::initVerboseBindings(module);
ASSERT_TRUE(THPStorage_init(module));
torch::functionalization::initModule(module);
#ifdef USE_CUDA
// This will only initialise base classes and attach them to library namespace
// They won't be ready for real usage until importing cuda module, that will
// complete the process (but it defines Python classes before calling back
// into C, so these lines have to execute first)..
THCPStream_init(module);
THCPEvent_init(module);
THCPGraph_init(module);
THCPMemPool_init(module);
#endif
#ifdef USE_XPU
THXPStream_init(module);
THXPEvent_init(module);
THXPMemPool_init(module);
THXPGraph_init(module);
#endif
torch::distributed::initPlacementBindings(module);
torch::initTensorIteratorBindings(module);
auto set_module_attr =
[&](const char* name, PyObject* v, bool incref = true) {
// PyModule_AddObject steals reference
if (incref) {
Py_INCREF(v);
}
int ret = PyModule_AddObject(module, name, v);
if (ret != 0) {
Py_DECREF(v);
}
return ret == 0;
};
#if defined(USE_CUDNN) || defined(USE_ROCM)
PyObject* has_cudnn = Py_True;
#else
PyObject* has_cudnn = Py_False;
#endif
ASSERT_TRUE(set_module_attr("_has_cudnn", has_cudnn));
#if defined(USE_CUSPARSELT) || defined(USE_ROCM)
PyObject* has_cusparselt = Py_True;
#else
PyObject* has_cusparselt = Py_False;
#endif
ASSERT_TRUE(set_module_attr("_has_cusparselt", has_cusparselt));
#if AT_MKL_ENABLED() || AT_POCKETFFT_ENABLED()
PyObject* has_spectral = Py_True;
#else
PyObject* has_spectral = Py_False;
#endif
ASSERT_TRUE(set_module_attr("has_spectral", has_spectral));
// force ATen to initialize because it handles
// setting up TH Errors so that they throw C++ exceptions
at::init();
// Automatically translate errors thrown from pybind11 functions
// NOLINTNEXTLINE(performance-unnecessary-value-param)
py::register_exception_translator([](std::exception_ptr e) {
try {
if (e) {
std::rethrow_exception(e);
}
}
CATCH_TH_ERRORS()
});
auto py_module = py::reinterpret_borrow<py::module>(module);
py_module.def("_initCrashHandler", &_initCrashHandler);
py_module.def("_demangle", &c10::demangle);
{
using at::impl::FakeDispatchCategory;
auto add_for = [](FakeDispatchCategory category) {
return [category](const std::string& name, const std::string& overload) {
at::impl::fakeDispatchTableAdd(
category, c10::OperatorName(name, overload));
};
};
py_module.def(
"_fake_dispatch_register_decomp",
add_for(FakeDispatchCategory::Decomp));
py_module.def(
"_fake_dispatch_register_meta", add_for(FakeDispatchCategory::Meta));
py_module.def(
"_fake_dispatch_register_op_impl",
add_for(FakeDispatchCategory::OpImpl));
py_module.def(
"_fake_dispatch_register_prim_meta",
add_for(FakeDispatchCategory::PrimMeta));
py_module.def(
"_fake_dispatch_deregister_op_impl",
[](const std::string& name, const std::string& overload) {
at::impl::fakeDispatchTableRemove(
FakeDispatchCategory::OpImpl, c10::OperatorName(name, overload));
});
}
py_module.def("_log_api_usage_metadata", &LogAPIUsageMetadataFromPython);
py_module.def(
"init_num_threads",
torch::wrap_pybind_function(at::init_num_threads),
R"(
init_num_threads()
Initializes the number of parallel threads used on the current thread.
Call this whenever a new thread is created in order to propagate values from
:func:`torch.set_num_threads` onto the new thread.
)");
py_module.def("_set_cached_tensors_enabled", [](bool enabled) {
at::caching::set_cached_tensors_enabled(enabled);
});
py_module.def("_add_cached_tensor", [](const at::Tensor& t) {
at::caching::add_cached_tensor(t);
});
py_module.def("_remove_cached_tensor", [](const at::Tensor& t) {
at::caching::remove_cached_tensor(t);
});
py_module.def("_is_cached_tensor", [](const at::Tensor& t) {
return at::caching::is_cached_tensor(t);
});
py_module.def("_storage_Use_Count", [](size_t storage_impl_ptr) {
// NOLINTNEXTLINE(performance-no-int-to-ptr)
c10::StorageImpl* storage_impl = (c10::StorageImpl*)storage_impl_ptr;
return c10::raw::intrusive_ptr::use_count(storage_impl);
});
ASSERT_TRUE(
set_module_attr("has_openmp", at::hasOpenMP() ? Py_True : Py_False));
ASSERT_TRUE(set_module_attr("has_mkl", at::hasMKL() ? Py_True : Py_False));
ASSERT_TRUE(
set_module_attr("_has_kleidiai", at::hasKleidiAI() ? Py_True : Py_False));
ASSERT_TRUE(
set_module_attr("has_lapack", at::hasLAPACK() ? Py_True : Py_False));
ASSERT_TRUE(set_module_attr(
"_has_eigen_sparse", at::hasEigenSparse() ? Py_True : Py_False));
py_module.def("_valgrind_supported_platform", []() {
#if defined(USE_VALGRIND)
return true;
#else
return false;
#endif
});
py_module.def("_valgrind_toggle", []() {
#if defined(USE_VALGRIND)
CALLGRIND_TOGGLE_COLLECT;
#else
TORCH_CHECK(false, "Valgrind is not supported.");
#endif
});
py_module.def("_valgrind_toggle_and_dump_stats", []() {
#if defined(USE_VALGRIND)
// NB: If we don't toggle collect around dump stats, callgrind_annotate
// won't process the results correctly. Specifically,
// `callgrind_annotate --inclusive=no` will be almost completely empty.
CALLGRIND_TOGGLE_COLLECT;
CALLGRIND_DUMP_STATS;
#else
TORCH_CHECK(false, "Valgrind is not supported.");
#endif
});
py::class_<WeakTensorRef>(py_module, "_WeakTensorRef")
.def(py::init([](py::object tensor) {
return WeakTensorRef(THPVariable_Unpack(tensor.ptr()));
}))
.def("expired", &WeakTensorRef::expired);
py::enum_<at::native::ConvBackend>(py_module, "_ConvBackend")
.value("CudaDepthwise2d", at::native::ConvBackend::CudaDepthwise2d)
.value("CudaDepthwise3d", at::native::ConvBackend::CudaDepthwise3d)
.value("Cudnn", at::native::ConvBackend::Cudnn)
.value("CudnnTranspose", at::native::ConvBackend::CudnnTranspose)
.value("Empty", at::native::ConvBackend::Empty)
.value("Miopen", at::native::ConvBackend::Miopen)
.value("MiopenDepthwise", at::native::ConvBackend::MiopenDepthwise)
.value("MiopenTranspose", at::native::ConvBackend::MiopenTranspose)
.value("Mkldnn", at::native::ConvBackend::Mkldnn)
.value("MkldnnEmpty", at::native::ConvBackend::MkldnnEmpty)
.value("NnpackSpatial", at::native::ConvBackend::NnpackSpatial)
.value("Overrideable", at::native::ConvBackend::Overrideable)
.value("Slow2d", at::native::ConvBackend::Slow2d)
.value("Slow3d", at::native::ConvBackend::Slow3d)
.value("SlowDilated2d", at::native::ConvBackend::SlowDilated2d)
.value("SlowDilated3d", at::native::ConvBackend::SlowDilated3d)
.value("SlowTranspose2d", at::native::ConvBackend::SlowTranspose2d)
.value("SlowTranspose3d", at::native::ConvBackend::SlowTranspose3d)
.value(
"Winograd3x3Depthwise", at::native::ConvBackend::Winograd3x3Depthwise)
.value("Xnnpack2d", at::native::ConvBackend::Xnnpack2d)
.value("Mps", at::native::ConvBackend::Mps)
.value("MpsTranspose,", at::native::ConvBackend::MpsTranspose);
py_module.def(
"_select_conv_backend",
[](const at::Tensor& input,
const at::Tensor& weight,
const std::optional<at::Tensor>& bias_opt,
at::SymIntArrayRef stride_,
at::SymIntArrayRef padding_,
at::SymIntArrayRef dilation_,
bool transposed_,
at::SymIntArrayRef output_padding_,
c10::SymInt groups_) {
return at::native::select_conv_backend(
input,
weight,
bias_opt,
stride_,
padding_,
dilation_,
transposed_,
output_padding_,
std::move(groups_),
std::nullopt);
},
py::arg("input"),
py::arg("weight"),
py::arg("bias"),
py::arg("stride"),
py::arg("padding"),
py::arg("dilation"),
py::arg("transposed"),
py::arg("output_padding"),
py::arg("groups"));
// overload for bias_sizes_opt/backward TODO: figure out default value
py_module.def(
"_select_conv_backend",
[](const at::Tensor& input,
const at::Tensor& weight,
const std::optional<at::Tensor>& bias,
at::SymIntArrayRef stride_,
at::SymIntArrayRef padding_,
at::SymIntArrayRef dilation_,
bool transposed_,
at::SymIntArrayRef output_padding_,
c10::SymInt groups_,
std::optional<std::vector<c10::SymInt>> bias_sizes_opt) {
c10::OptionalArrayRef<c10::SymInt> ref = std::nullopt;
if (bias_sizes_opt) {
ref = (*bias_sizes_opt);
}
return at::native::select_conv_backend(
input,
weight,
bias,
stride_,
padding_,
dilation_,
transposed_,
output_padding_,
std::move(groups_),
ref);
},
py::arg("input"),
py::arg("weight"),
py::arg("bias"),
py::arg("stride"),
py::arg("padding"),
py::arg("dilation"),
py::arg("transposed"),
py::arg("output_padding"),
py::arg("groups"),
py::arg("bias_sizes"));
py_module.def(
"_conv_determine_backend_memory_format",
at::native::_determine_backend_memory_format);
////////////////////////////////////////////////////////////////////////////////
// Scaled Dot Product Attention utilities
////////////////////////////////////////////////////////////////////////////////
py::class_<sdp::sdp_params>(py_module, "_SDPAParams")
.def(py::init([](at::Tensor const& query,
at::Tensor const& key,
at::Tensor const& value,
std::optional<at::Tensor> attn_mask,
double dropout,
bool is_causal,
bool enable_gqa) {
return sdp::sdp_params{
query,
key,
value,
std::move(attn_mask),
dropout,
is_causal,
enable_gqa};
}))
.def_readonly("query", &sdp::sdp_params::query)
.def_readonly("key", &sdp::sdp_params::key)
.def_readonly("value", &sdp::sdp_params::value)
.def_readonly("attn_mask", &sdp::sdp_params::attn_mask)
.def_readonly("dropout", &sdp::sdp_params::dropout)
.def_readonly("is_causal", &sdp::sdp_params::is_causal)
.def_readonly("enable_gqa", &sdp::sdp_params::enable_gqa);
py::enum_<sdp::SDPBackend>(
py_module,
"_SDPBackend",
"An enum-like class that contains the different backends for scaled dot product attention.\n\n... warning:: This class is in beta and subject to change.\n\n"
"This backend class is designed to be used with the sdpa_kernel context manager."
"See :func: torch.nn.attention.sdpa_kernel for more details.")
.value("ERROR", sdp::SDPBackend::error)
.value("MATH", sdp::SDPBackend::math)
.value("FLASH_ATTENTION", sdp::SDPBackend::flash_attention)
.value("EFFICIENT_ATTENTION", sdp::SDPBackend::efficient_attention)
.value("CUDNN_ATTENTION", sdp::SDPBackend::cudnn_attention)
.value("OVERRIDEABLE", sdp::SDPBackend::overrideable);
py_module.def("_is_flash_attention_available", []() {
#if defined(USE_CUDA) || defined(USE_XPU)
return sdp::is_flash_attention_available();
#else
return false;
#endif
});
py_module.def(
"_can_use_flash_attention",
[](const sdp::sdp_params& params, bool debug) {
#if defined(USE_CUDA) || defined(USE_XPU)
return sdp::can_use_flash_attention(params, debug);
#else
return false;
#endif
});
py_module.def(
"_can_use_mem_efficient_attention",
[](const sdp::sdp_params& params, bool debug) {
#ifdef USE_CUDA
return sdp::can_use_mem_efficient_attention(params, debug);
#else
return false;
#endif
});
py_module.def(
"_can_use_cudnn_attention",
[](const sdp::sdp_params& params, bool debug) {
#ifdef USE_CUDA
return sdp::can_use_cudnn_attention(params, debug);
#else
return false;
#endif
});
py::enum_<at::LinalgBackend>(py_module, "_LinalgBackend")
.value("Default", at::LinalgBackend::Default)
.value("Cusolver", at::LinalgBackend::Cusolver)
.value("Magma", at::LinalgBackend::Magma);
py_module.def("_set_linalg_preferred_backend", [](at::LinalgBackend b) {
at::globalContext().setLinalgPreferredBackend(b);
});
py_module.def("_get_linalg_preferred_backend", []() {
return at::globalContext().linalgPreferredBackend();
});
py::enum_<at::BlasBackend>(py_module, "_BlasBackend")
.value("Default", at::BlasBackend::Default)
.value("Cublas", at::BlasBackend::Cublas)
.value("Cublaslt", at::BlasBackend::Cublaslt)
.value("Ck", at::BlasBackend::Ck);
py_module.def("_set_blas_preferred_backend", [](at::BlasBackend b) {
at::globalContext().setBlasPreferredBackend(b);
});
py_module.def("_get_blas_preferred_backend", []() {
return at::globalContext().blasPreferredBackend();
});
py_module.def("_get_blas_default_backend", []() {
return at::globalContext().blasDefaultBackend();
});
py::enum_<at::blas::ScalingType>(
py_module, "_ScalingType", "Supported Tensor scaling types")
.value(
"TensorWise",
at::blas::ScalingType::TensorWise,
"Single scale per-tensor")
.value(
"RowWise", at::blas::ScalingType::RowWise, "Scale per-row of tensor")
.value(
"BlockWise1x16",
at::blas::ScalingType::BlockWise1x16,
"Scale per 16 contiguous values")
.value(
"BlockWise1x32",
at::blas::ScalingType::BlockWise1x32,
"Scale per 32 contiguous values")
.value(
"BlockWise1x128",
at::blas::ScalingType::BlockWise1x128,
"Scale per 128 contiguous values")
.value(
"BlockWise128x128",
at::blas::ScalingType::BlockWise128x128,
"Scale per 128x128 tile");
py::enum_<at::blas::SwizzleType>(
py_module, "_SwizzleType", "Supported scale swizzle types")
.value("NO_SWIZZLE", at::blas::SwizzleType::NO_SWIZZLE, "No swizzling")
.value(
"SWIZZLE_32_4_4",
at::blas::SwizzleType::SWIZZLE_32_4_4,
"Blackwell-style 32x4x4 swizzle");
py::enum_<at::ROCmFABackend>(py_module, "_ROCmFABackend")
.value("Default", at::ROCmFABackend::Default)
.value("AOTriton", at::ROCmFABackend::AOTriton)
.value("Ck", at::ROCmFABackend::Ck);
py_module.def("_set_rocm_fa_preferred_backend", [](at::ROCmFABackend b) {
at::globalContext().setROCmFAPreferredBackend(b);
});
py_module.def("_get_rocm_fa_preferred_backend", []() {
return at::globalContext().getROCmFAPreferredBackend();
});
py_module.def("_is_ck_sdpa_available", []() {
#ifdef USE_ROCM
return at::globalContext().ckSDPASupported() &&
at::globalContext().hasCKSDPA();
#else
return false;
#endif
});
py_module.def(
"_set_sm_carveout_experimental", [](std::optional<int32_t> val) {
at::globalContext()._setSMCarveout_EXPERIMENTAL(val);
});
py_module.def("_get_sm_carveout_experimental", []() {
return at::globalContext()._SMCarveout_EXPERIMENTAL();
});
py_module.def(
"_construct_storage_from_data_pointer",
[](int64_t data_ptr, c10::Device device, size_t size_bytes) {
return c10::Storage(
c10::Storage::use_byte_size_t(),
size_bytes,
// NOLINTNEXTLINE(performance-no-int-to-ptr)
at::DataPtr(reinterpret_cast<void*>(data_ptr), device));
});
py_module.def(
"_get_fp32_precision_getter",
[](const std::string& backend, const std::string& op) {
return at::precision2str(at::globalContext().float32Precision(
at::str2backend(backend), at::str2op(op)));
});
py_module.def(
"_set_fp32_precision_setter",
[](const std::string& backend,
const std::string& op,
const std::string& precision) {
at::globalContext().setFloat32Precision(
at::str2backend(backend),
at::str2op(op),
at::str2precision(precision));
return precision;
});
py_module.def(
"_stash_obj_in_tls", [](const std::string& key, py::handle arg) {
Py_INCREF(arg.ptr());
at::impl::ThreadLocalPythonObjects::get_state().set(
key,
std::make_shared<c10::SafePyObject>(arg.ptr(), getPyInterpreter()));
});
py_module.def("_get_obj_in_tls", [](const std::string& key) -> py::handle {
auto safe_pyobject =
at::impl::ThreadLocalPythonObjects::get_state().get(key);
auto obj = safe_pyobject->ptr(getPyInterpreter());
return py::handle(obj);
});
py_module.def("_is_key_in_tls", [](const std::string& key) -> bool {
return at::impl::ThreadLocalPythonObjects::get_state().contains(key);
});
py_module.def("_remove_obj_from_tls", [](const std::string& key) {
at::impl::ThreadLocalPythonObjects::erase(key);
});
py_module.def("_accelerator_hooks_device_count", []() {
auto device_type = at::getAccelerator();
if (device_type.has_value()) {
return at::globalContext()
.getAcceleratorHooksInterface(device_type)
.deviceCount();
}
return static_cast<c10::DeviceIndex>(-1);
});
py_module.def(
"_accelerator_hooks_set_current_device",
[](c10::DeviceIndex device_index) {
auto device_type = at::getAccelerator();
if (device_type.has_value()) {
at::globalContext()
.getAcceleratorHooksInterface(device_type)
.setCurrentDevice(device_index);
}
});
py_module.def("_accelerator_hooks_get_current_device", []() {
auto device_type = at::getAccelerator();
if (device_type.has_value()) {
return at::globalContext()
.getAcceleratorHooksInterface(device_type)
.getCurrentDevice();
}
return static_cast<c10::DeviceIndex>(-1);
});
py_module.def(
"_accelerator_hooks_exchange_device", [](c10::DeviceIndex device_index) {
auto device_type = at::getAccelerator();
if (device_type.has_value()) {
return at::globalContext()
.getAcceleratorHooksInterface(device_type)
.exchangeDevice(device_index);
}
return static_cast<c10::DeviceIndex>(-1);
});
py_module.def(
"_accelerator_hooks_maybe_exchange_device",
[](c10::DeviceIndex device_index) {
auto device_type = at::getAccelerator();
if (device_type.has_value()) {
return at::globalContext()
.getAcceleratorHooksInterface(device_type)
.maybeExchangeDevice(device_index);
}
return static_cast<c10::DeviceIndex>(-1);
});
py_module.def(
"_get_accelerator",
[](std::optional<bool> check = std::nullopt) {
auto acc = at::getAccelerator(check.value_or(false));
if (acc.has_value()) {
bool is_available = at::globalContext()
.getAcceleratorHooksInterface(acc)
.isAvailable();
if (!is_available) {
acc = std::nullopt;
}
}
return c10::Device(acc.value_or(c10::DeviceType::CPU), -1);
},
py::arg("check") = nullptr);
#ifdef USE_CUDA
PyObject* has_cuda = Py_True;
#else
PyObject* has_cuda = Py_False;
#endif
#ifdef USE_MPS
PyObject* has_mps = Py_True;
#else
PyObject* has_mps = Py_False;
#endif
#ifdef USE_XPU
PyObject* has_xpu = Py_True;
#else
PyObject* has_xpu = Py_False;
#endif
ASSERT_TRUE(set_module_attr("_has_cuda", has_cuda));
ASSERT_TRUE(
set_module_attr("_has_magma", at::hasMAGMA() ? Py_True : Py_False));
ASSERT_TRUE(set_module_attr("_has_mps", has_mps));
ASSERT_TRUE(set_module_attr("_has_xpu", has_xpu));
ASSERT_TRUE(
set_module_attr("_has_mkldnn", at::hasMKLDNN() ? Py_True : Py_False));
ASSERT_TRUE(set_module_attr(
"_has_mkldnn_acl", AT_MKLDNN_ACL_ENABLED() ? Py_True : Py_False));
ASSERT_TRUE(set_module_attr("_GLIBCXX_USE_CXX11_ABI", Py_True));
py_module.def(
"_set_conj", [](const at::Tensor& x, bool conj) { x._set_conj(conj); });
py_module.def(
"_set_neg", [](const at::Tensor& x, bool neg) { x._set_neg(neg); });
py_module.def("_get_tensor_metadata", &torch::jit::getTensorMetadata);
py_module.def(
"_set_tensor_metadata",
static_cast<void (*)(
const at::Tensor&, std::unordered_map<std::string, bool>)>(
torch::jit::setTensorMetadata));
py_module.def("_dispatch_key_set", [](const at::Tensor& x) {
return toString(x.key_set());
});
py_module.def(
"_has_storage", [](const at::Tensor& x) { return x.has_storage(); });
py_module.def("_set_meta_in_tls_dispatch_include", [](bool meta_in_tls) {
auto local_keyset = c10::impl::tls_local_dispatch_key_set();
c10::DispatchKeySet key_set({at::DispatchKey::Meta});
if (meta_in_tls) {
local_keyset.included_ = local_keyset.included_ | key_set;
} else {
local_keyset.included_ =
local_keyset.included_.remove_backend(c10::BackendComponent::MetaBit);
}
c10::impl::_force_tls_local_dispatch_key_set(local_keyset);
});
py_module.def("_meta_in_tls_dispatch_include", []() {
auto local_keyset = c10::impl::tls_local_dispatch_key_set();
return local_keyset.included_.has_backend(c10::BackendComponent::MetaBit);
});
py_module.def("_dump_local_tls_set", []() {
auto local_keyset = c10::impl::tls_local_dispatch_key_set();
std::cout << "Included: " << toString(local_keyset.included_) << '\n';
std::cout << "Excluded: " << toString(local_keyset.excluded_) << '\n';
});
py_module.def(
"_should_allow_numbers_as_tensors", [](const std::string& name) {
return torch::should_allow_numbers_as_tensors(name);
});
py_module.def(
"_group_tensors_by_device_and_dtype",
[](const std::vector<std::vector<std::optional<at::Tensor>>>&
nested_tensorlist,
const bool with_indices) {
return at::native::_group_tensors_by_first_tensors_device_and_dtype(
nested_tensorlist, with_indices);
});
py_module.def(
"_storage_address",
[](const at::Tensor& tensor) {
return reinterpret_cast<std::intptr_t>(
tensor.storage().unsafeGetStorageImpl());
},
"Gets the memory address of the Tensor's StorageImpl.");
py_module.def(
"_data_address",
[](const at::Tensor& tensor) {
return reinterpret_cast<std::intptr_t>(tensor.storage().data());
},
"Gets the memory address of the Tensor's data pointer.");
py_module.def(
"_is_cow_tensor",
[](const at::Tensor& tensor) {
return c10::impl::cow::is_cow_data_ptr(tensor.storage().data_ptr());
},
"Checks if a tensor's data pointer is COW");
py_module.def(
"_get_cudnn_batch_norm_reserve_space_size",
[](const at::Tensor& input, bool training) {
#ifdef USE_CUDA
return at::native::_get_cudnn_batch_norm_reserve_space_size(
input, training);
#else
TORCH_CHECK(false, "PyTorch was not built with cuda");
#endif
},
py::arg("input"),
py::arg("training"));
py::enum_<at::native::BatchNormBackend>(py_module, "_BatchNormBackend")
.value("Native", at::native::BatchNormBackend::Native)
.value("Cudnn", at::native::BatchNormBackend::Cudnn)
.value("Miopen", at::native::BatchNormBackend::Miopen);
py_module.def(
"_select_batch_norm_backend",
[](const at::Tensor& input,
const at::Tensor& weight,
const at::Tensor& bias,
const at::Tensor& running_mean,
const at::Tensor& running_var,
bool training,
double eps) {
return at::native::_select_batch_norm_backend(
input, weight, bias, running_mean, running_var, training, eps);
},
py::arg("input"),
py::arg("weight"),
py::arg("bias"),
py::arg("running_mean"),
py::arg("running_var"),
py::arg("training"),
py::arg("eps"));
const auto& defaultGenerator = at::detail::getDefaultCPUGenerator();
THPDefaultCPUGenerator = reinterpret_cast<THPGenerator*>(
THPGenerator_initDefaultGenerator(defaultGenerator));
// This reference is meant to be given away, so no need to incref here.
ASSERT_TRUE(set_module_attr(
"default_generator",
(PyObject*)THPDefaultCPUGenerator,
/* incref= */ false));
ASSERT_TRUE(set_module_attr(
"DisableTorchFunctionSubclass",
(PyObject*)THPModule_DisableTorchFunctionSubclassType(),
/* incref= */ false));
ASSERT_TRUE(set_module_attr(
"DisableTorchFunction",
(PyObject*)THPModule_DisableTorchFunctionType(),
/* incref= */ false));
py::enum_<at::impl::TorchFunctionDisabledState>(
py_module, "_TorchFunctionState")
.value("ENABLED", at::impl::TorchFunctionDisabledState::ENABLED)
.value(
"SUBCLASSES_DISABLED",
at::impl::TorchFunctionDisabledState::SUBCLASSES_DISABLED)
.value(
"ALL_DISABLED", at::impl::TorchFunctionDisabledState::ALL_DISABLED);
py_module.def(
"_set_torch_function_state",
[](at::impl::TorchFunctionDisabledState state) {
at::impl::PythonTorchFunctionTLS::set_disabled_state(state);
});
py_module.def("_get_torch_function_state", []() {
return at::impl::PythonTorchFunctionTLS::get_disabled_state();
});
torch::set_disabled_torch_function_impl(
PyObject_GetAttrString(module, "_disabled_torch_function_impl"));
ASSERT_TRUE(torch::disabled_torch_function_impl() != nullptr);
torch::set_disabled_torch_dispatch_impl(
PyObject_GetAttrString(module, "_disabled_torch_dispatch_impl"));
ASSERT_TRUE(torch::disabled_torch_dispatch_impl() != nullptr);
// init kineto here
#ifdef USE_KINETO
torch::global_kineto_init();
#endif
auto nativert_module = py_module.def_submodule("_nativert");
torch::nativert::initModelRunnerPybind(nativert_module);
return module;
END_HANDLE_TH_ERRORS
}
// Checks that the _C shared library isn't initialized multiple times. This
// can happen if the same csrc files are compiled into multiple shared
// libraries.
static void pytorch_duplicate_guard() {
static int initialized = 0;
if (initialized) {
fmt::print(stderr, "pytorch: _C shared library re-initialized\n");
abort();
}
initialized = 1;
}
struct call_duplicate_guard {
call_duplicate_guard() {
pytorch_duplicate_guard();
}
};
static call_duplicate_guard _call_duplicate_guard;