1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
|
#include <pybind11/pybind11.h>
#include <torch/csrc/cuda/Event.h>
#include <torch/csrc/cuda/Module.h>
#include <torch/csrc/cuda/Stream.h>
#include <torch/csrc/Device.h>
#include <torch/csrc/THP.h>
#include <torch/csrc/utils/python_arg_parser.h>
#include <c10/cuda/CUDAGuard.h>
#include <structmember.h>
#include <cuda_runtime_api.h>
PyObject *THCPEventClass = nullptr;
static PyObject * THCPEvent_pynew(
PyTypeObject *type, PyObject *args, PyObject *kwargs) {
HANDLE_TH_ERRORS
unsigned char enable_timing = 0;
unsigned char blocking = 0;
unsigned char interprocess = 0;
static char *kwlist[] =
{"enable_timing", "blocking", "interprocess", nullptr};
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|bbb", kwlist,
&enable_timing, &blocking, &interprocess)) {
return nullptr;
}
THPObjectPtr ptr(type->tp_alloc(type, 0));
if (!ptr) {
return nullptr;
}
THCPEvent* self = (THCPEvent *)ptr.get();
unsigned int flags =
(blocking ? cudaEventBlockingSync : cudaEventDefault) |
(enable_timing ? cudaEventDefault : cudaEventDisableTiming) |
(interprocess ? cudaEventInterprocess : cudaEventDefault);
new (&self->cuda_event) at::cuda::CUDAEvent(flags);
return (PyObject *)ptr.release();
END_HANDLE_TH_ERRORS
}
static PyObject * THCPEvent_from_ipc_handle(
PyTypeObject *type, PyObject *args, PyObject *kwargs) {
HANDLE_TH_ERRORS
static torch::PythonArgParser parser({
"from_ipc_handle(Device device, std::string ipc_handle)",
});
torch::ParsedArgs<2> parsed_args;
auto r = parser.parse(args, kwargs, parsed_args);
at::Device device = r.device(0);
std::string handle_string = r.string(1);
TORCH_CHECK(handle_string.size() == sizeof(cudaIpcEventHandle_t),
"cudaIpcEventHandle_t expects byte-like object of size ",
sizeof(cudaIpcEventHandle_t), ", but got ", handle_string.size());
TORCH_CHECK(device.type() == at::kCUDA, "Event can only be created on "
"CUDA devices, but got device type ", device.type())
THPObjectPtr ptr(type->tp_alloc(type, 0));
if (!ptr) {
return nullptr;
}
THCPEvent* self = (THCPEvent *)ptr.get();
cudaIpcEventHandle_t handle;
std::memcpy(&handle, handle_string.c_str(), handle_string.size());
new (&self->cuda_event) at::cuda::CUDAEvent(device.index(), &handle);
return (PyObject *)ptr.release();
END_HANDLE_TH_ERRORS
}
static void THCPEvent_dealloc(THCPEvent *self) {
self->cuda_event.~CUDAEvent();
Py_TYPE(self)->tp_free((PyObject*)self);
}
static PyObject * THCPEvent_get_cuda_event(THCPEvent *self, void *unused) {
HANDLE_TH_ERRORS
return PyLong_FromVoidPtr(self->cuda_event.event());
END_HANDLE_TH_ERRORS
}
static PyObject * THCPEvent_get_device(THCPEvent *self, void *unused) {
HANDLE_TH_ERRORS
at::optional<at::Device> device = self->cuda_event.device();
if (!device) {
Py_RETURN_NONE;
}
return THPDevice_New(device.value());
END_HANDLE_TH_ERRORS
}
static PyObject * THCPEvent_record(THCPEvent *self, THCPStream *stream) {
HANDLE_TH_ERRORS
self->cuda_event.record(stream->cuda_stream);
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject * THCPEvent_wait(THCPEvent *self, THCPStream *stream) {
HANDLE_TH_ERRORS
{
pybind11::gil_scoped_release no_gil;
self->cuda_event.block(stream->cuda_stream);
}
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject * THCPEvent_query(THCPEvent *self, PyObject *noargs) {
HANDLE_TH_ERRORS
return PyBool_FromLong(self->cuda_event.query());
END_HANDLE_TH_ERRORS
}
static PyObject * THCPEvent_elapsed_time(THCPEvent *self, THCPEvent *other) {
HANDLE_TH_ERRORS
return PyFloat_FromDouble(self->cuda_event.elapsed_time(other->cuda_event));
END_HANDLE_TH_ERRORS
}
static PyObject * THCPEvent_synchronize(THCPEvent *self, PyObject *noargs) {
HANDLE_TH_ERRORS
{
pybind11::gil_scoped_release no_gil;
self->cuda_event.synchronize();
}
Py_RETURN_NONE;
END_HANDLE_TH_ERRORS
}
static PyObject * THCPEvent_ipc_handle(THCPEvent *self, PyObject *noargs) {
HANDLE_TH_ERRORS
cudaIpcEventHandle_t handle;
self->cuda_event.ipc_handle(&handle);
return PyBytes_FromStringAndSize((const char *)&handle, sizeof(handle));
END_HANDLE_TH_ERRORS
}
static struct PyGetSetDef THCPEvent_properties[] = {
{"device", (getter)THCPEvent_get_device, nullptr, nullptr, nullptr},
{"cuda_event", (getter)THCPEvent_get_cuda_event, nullptr, nullptr, nullptr},
{nullptr}
};
static PyMethodDef THCPEvent_methods[] = {
{(char*)"from_ipc_handle", (PyCFunction)(void(*)(void))THCPEvent_from_ipc_handle,
METH_CLASS | METH_VARARGS | METH_KEYWORDS, nullptr},
{(char*)"record", (PyCFunction)THCPEvent_record, METH_O, nullptr},
{(char*)"wait", (PyCFunction)THCPEvent_wait, METH_O, nullptr},
{(char*)"query", (PyCFunction)THCPEvent_query, METH_NOARGS, nullptr},
{(char*)"elapsed_time", (PyCFunction)THCPEvent_elapsed_time, METH_O, nullptr},
{(char*)"synchronize", (PyCFunction)THCPEvent_synchronize,
METH_NOARGS, nullptr},
{(char*)"ipc_handle", (PyCFunction)THCPEvent_ipc_handle,
METH_NOARGS, nullptr},
{nullptr}
};
PyTypeObject THCPEventType = {
PyVarObject_HEAD_INIT(nullptr, 0)
"torch._C._CudaEventBase", /* tp_name */
sizeof(THCPEvent), /* tp_basicsize */
0, /* tp_itemsize */
(destructor)THCPEvent_dealloc, /* tp_dealloc */
0, /* tp_vectorcall_offset */
0, /* tp_getattr */
0, /* tp_setattr */
0, /* tp_reserved */
0, /* tp_repr */
0, /* tp_as_number */
0, /* tp_as_sequence */
0, /* tp_as_mapping */
0, /* tp_hash */
0, /* tp_call */
0, /* tp_str */
0, /* tp_getattro */
0, /* tp_setattro */
0, /* tp_as_buffer */
Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE, /* tp_flags */
nullptr, /* tp_doc */
0, /* tp_traverse */
0, /* tp_clear */
0, /* tp_richcompare */
0, /* tp_weaklistoffset */
0, /* tp_iter */
0, /* tp_iternext */
THCPEvent_methods, /* tp_methods */
0, /* tp_members */
THCPEvent_properties, /* tp_getset */
0, /* tp_base */
0, /* tp_dict */
0, /* tp_descr_get */
0, /* tp_descr_set */
0, /* tp_dictoffset */
0, /* tp_init */
0, /* tp_alloc */
THCPEvent_pynew, /* tp_new */
};
void THCPEvent_init(PyObject *module) {
THCPEventClass = (PyObject*)&THCPEventType;
if (PyType_Ready(&THCPEventType) < 0) {
throw python_error();
}
Py_INCREF(&THCPEventType);
if (PyModule_AddObject(
module, "_CudaEventBase", (PyObject *)&THCPEventType) < 0) {
throw python_error();
}
}
|
