You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

1043 lines
40 KiB

from collections import defaultdict
from typing import Any, Dict, List, Optional
from warnings import warn
import torch
import torch.cuda
from torch._C import _get_privateuse1_backend_name
from torch._C._profiler import _ExperimentalConfig
from torch.autograd import (
_disable_profiler,
_enable_profiler,
_kineto_step,
_prepare_profiler,
_ProfilerResult,
_supported_activities,
DeviceType,
kineto_available,
ProfilerActivity,
ProfilerConfig,
ProfilerState,
)
from torch.autograd.profiler_util import (
_filter_name,
_filter_stack_entry,
_rewrite_name,
EventList,
FunctionEvent,
MEMORY_EVENT_NAME,
MemRecordsAcc,
OUT_OF_MEMORY_EVENT_NAME,
)
from torch.futures import Future
__all__ = [
"profile",
"record_function",
"emit_itt",
"emit_nvtx",
"load_nvprof",
"EnforceUnique",
"parse_nvprof_trace",
"KinetoStepTracker",
"EventList",
"FunctionEvent",
"MemRecordsAcc",
]
try:
# Available in Python >= 3.2
from contextlib import ContextDecorator as _ContextDecorator
except ImportError:
import functools
class _ContextDecorator: # type: ignore[no-redef]
def __enter__(self):
raise NotImplementedError
def __exit__(self, exc_type, exc_val, exc_tb):
raise NotImplementedError
def __call__(self, func):
@functools.wraps(func)
def wrapped(*args, **kwargs):
with self:
return func(*args, **kwargs)
return wrapped
# global python state - whether profiler is currently enabled
# useful for fast python checks to reduce latency
_is_profiler_enabled: bool = False
def _set_is_profiler_enabled(enable: bool):
global _is_profiler_enabled
_is_profiler_enabled = enable
def _run_on_profiler_start():
_set_is_profiler_enabled(True)
def _run_on_profiler_stop():
_set_is_profiler_enabled(False)
class profile:
"""Context manager that manages autograd profiler state and holds a summary of results.
Under the hood it just records events of functions being executed in C++ and
exposes those events to Python. You can wrap any code into it and it will
only report runtime of PyTorch functions.
Note: profiler is thread local and is automatically propagated into the async tasks
Args:
enabled (bool, optional): Setting this to False makes this context manager a no-op.
use_cuda (bool, optional): Enables timing of CUDA events as well using the cudaEvent API.
Adds approximately 4us of overhead to each tensor operation.
record_shapes (bool, optional): If shapes recording is set, information
about input dimensions will be collected. This allows one to see which
dimensions have been used under the hood and further group by them
using prof.key_averages(group_by_input_shape=True). Please note that
shape recording might skew your profiling data. It is recommended to
use separate runs with and without shape recording to validate the timing.
Most likely the skew will be negligible for bottom most events (in a case
of nested function calls). But for higher level functions the total
self cpu time might be artificially increased because of the shape
collection.
with_flops (bool, optional): If with_flops is set, the profiler will estimate
the FLOPs (floating point operations) value using the operator's input shape.
This allows one to estimate the hardware performance. Currently,
this option only works for the matrix multiplication and 2D convolution operators.
profile_memory (bool, optional): track tensor memory allocation/deallocation.
with_stack (bool, optional): record source information (file and line number) for the ops.
with_modules (bool): record module hierarchy (including function names)
corresponding to the callstack of the op. e.g. If module A's forward call's
module B's forward which contains an aten::add op,
then aten::add's module hierarchy is A.B
Note that this support exist, at the moment, only for TorchScript models
and not eager mode models.
use_kineto (bool, optional): experimental, enable profiling with Kineto profiler.
use_cpu (bool, optional): profile CPU events; setting to ``False`` requires
``use_kineto=True`` and can be used to lower the overhead for GPU-only profiling.
experimental_config (_ExperimentalConfig) : A set of experimental options
used by profiler libraries like Kineto. Note, backward compatibility is not guaranteed.
.. warning:
Enabling memory profiling or source attribution incurs additional profiler
overhead
.. warning:
This context managers should not be called recursively, i.e. no nested
instances are allowed
.. warning:
Due to some CUDA multiprocessing limitations (multiprocessing-cuda-note_),
one cannot use the profiler with ``use_cuda = True`` to benchmark
DataLoaders with ``num_workers > 0``. If you wish to benchmark data loading,
please use ``use_cuda = False`` or ``num_workers = 0``.
Example:
>>> # xdoctest: +SKIP
>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_AUTOGRAD_PROFILER)
>>> x = torch.randn((1, 1), requires_grad=True)
>>> with torch.autograd.profiler.profile() as prof:
>>> for _ in range(100): # any normal python code, really!
>>> y = x ** 2
>>> y.backward()
>>> # NOTE: some columns were removed for brevity
>>> print(prof.key_averages().table(sort_by="self_cpu_time_total"))
----------------------------------- --------------- --------------- ---------------
Name Self CPU total CPU time avg Number of Calls
----------------------------------- --------------- --------------- ---------------
mul 32.048ms 32.048ms 200
pow 27.041ms 27.041ms 200
PowBackward0 9.727ms 55.483ms 100
torch::autograd::AccumulateGrad 9.148ms 9.148ms 100
torch::autograd::GraphRoot 691.816us 691.816us 100
----------------------------------- --------------- --------------- ---------------
"""
def __init__(
self,
enabled=True,
*,
use_cuda=False,
use_device=None,
record_shapes=False,
with_flops=False,
profile_memory=False,
with_stack=False,
with_modules=False,
use_kineto=False,
use_cpu=True,
use_mtia=False,
experimental_config=None,
):
self.enabled: bool = enabled
if not self.enabled:
return
self.use_cuda = use_cuda
self.use_device: Optional[str] = (
use_device if use_device != "privateuseone" else None
)
self.function_events: Optional[EventList] = None
self.entered = False
self.record_shapes = record_shapes
self.with_flops = with_flops
self.record_shapes |= self.with_flops
self.profile_memory = profile_memory
self.with_stack = with_stack
self.with_modules = with_modules
self.use_cpu = use_cpu
self.use_mtia = use_mtia
if experimental_config is None:
experimental_config = _ExperimentalConfig()
self.experimental_config = experimental_config
self.kineto_results: Optional[_ProfilerResult] = None
if not self.use_cpu:
assert (
use_kineto
), "Device-only events supported only with Kineto (use_kineto=True)"
if self.use_device == "cuda":
self.use_device = None
self.use_cuda = True
if self.use_device and self.use_device != _get_privateuse1_backend_name():
warn(f"{self.use_device} doesn't support profile.")
self.use_device = None
if self.use_cuda and not torch.cuda.is_available():
warn("CUDA is not available, disabling CUDA profiling")
self.use_cuda = False
self.kineto_activities = set()
if self.use_cpu:
self.kineto_activities.add(ProfilerActivity.CPU)
if self.use_mtia:
self.kineto_activities.add(ProfilerActivity.MTIA)
self.profiler_kind = ProfilerState.KINETO
if self.use_cuda:
if not use_kineto or ProfilerActivity.CUDA not in _supported_activities():
assert self.use_cpu, "Legacy CUDA profiling requires use_cpu=True"
self.profiler_kind = ProfilerState.KINETO_GPU_FALLBACK
else:
self.kineto_activities.add(ProfilerActivity.CUDA)
if self.use_device:
if (
not use_kineto
or ProfilerActivity.PrivateUse1 not in _supported_activities()
):
assert (
self.use_cpu
), "Legacy custombackend profiling requires use_cpu=True"
self.profiler_kind = ProfilerState.KINETO_PRIVATEUSE1_FALLBACK
else:
self.kineto_activities.add(ProfilerActivity.PrivateUse1)
self.profiler_kind = ProfilerState.KINETO_PRIVATEUSE1
assert (
len(self.kineto_activities) > 0
), "No activities specified for the profiler"
def config(self):
return ProfilerConfig(
self.profiler_kind,
self.record_shapes,
self.profile_memory,
self.with_stack,
self.with_flops,
self.with_modules,
self.experimental_config,
)
def __enter__(self):
if not self.enabled:
return
if self.entered:
raise RuntimeError("Profiler context manager is not reentrant")
self._prepare_trace()
self._start_trace()
return self
def _prepare_trace(self):
self.entered = True
_prepare_profiler(self.config(), self.kineto_activities)
def _start_trace(self):
self.entered = True
_run_on_profiler_start()
_enable_profiler(self.config(), self.kineto_activities)
def __exit__(self, exc_type, exc_val, exc_tb):
if not self.enabled:
return
if self.use_cuda:
torch.cuda.synchronize()
self.kineto_results = _disable_profiler()
_run_on_profiler_stop()
parsed_results = self._parse_kineto_results(self.kineto_results)
self.function_events = EventList(
parsed_results,
use_cuda=self.use_cuda,
use_device=self.use_device,
profile_memory=self.profile_memory,
with_flops=self.with_flops,
)
self.function_events._build_tree()
return False
def __repr__(self):
if self.function_events is None:
return "<unfinished torch.autograd.profile>"
return repr(self.function_events)
def __str__(self):
if self.function_events is None:
return "<unfinished torch.autograd.profile>"
return str(self.function_events)
def _check_finish(self):
if self.function_events is None:
raise RuntimeError("Profiler didn't finish running")
def table(
self,
sort_by=None,
row_limit=100,
max_src_column_width=75,
max_name_column_width=55,
max_shapes_column_width=80,
header=None,
top_level_events_only=False,
):
self._check_finish()
assert self.function_events is not None
return self.function_events.table(
sort_by=sort_by,
row_limit=row_limit,
max_src_column_width=max_src_column_width,
max_name_column_width=max_name_column_width,
max_shapes_column_width=max_shapes_column_width,
header=header,
top_level_events_only=top_level_events_only,
)
table.__doc__ = EventList.table.__doc__
def export_chrome_trace(self, path):
self._check_finish()
if kineto_available():
self.kineto_results.save(path) # type: ignore[union-attr]
else:
return self.function_events.export_chrome_trace(path) # type: ignore[union-attr]
export_chrome_trace.__doc__ = EventList.export_chrome_trace.__doc__
def export_stacks(self, path: str, metric: str = "self_cpu_time_total"):
self._check_finish()
assert self.function_events is not None, "Expected profiling results"
assert self.with_stack, "export_stacks() requires with_stack=True"
return self.function_events.export_stacks(path, metric)
def key_averages(self, group_by_input_shape=False, group_by_stack_n=0):
self._check_finish()
assert self.function_events is not None, "Expected profiling results"
return self.function_events.key_averages(group_by_input_shape, group_by_stack_n)
key_averages.__doc__ = EventList.key_averages.__doc__
def total_average(self):
self._check_finish()
assert self.function_events is not None, "Expected profiling results"
return self.function_events.total_average()
total_average.__doc__ = EventList.total_average.__doc__
@property
def self_cpu_time_total(self):
"""Returns total time spent on CPU.
The total time is a sum of all self times across all the events.
"""
self._check_finish()
assert self.function_events is not None
return self.function_events.self_cpu_time_total
def _parse_kineto_results(self, result: _ProfilerResult):
# result.events() has most of the events - PyTorch op-level and device-level events
trace_start_us = result.trace_start_us()
mem_records = [
[evt, False] for evt in result.events() if evt.name() == MEMORY_EVENT_NAME
]
oom_records = [
evt for evt in result.events() if evt.name() == OUT_OF_MEMORY_EVENT_NAME
]
mem_records_acc = MemRecordsAcc(mem_records)
def _cpu_memory_usage(mem_record):
return (
mem_record.nbytes()
if mem_record.device_type()
in [DeviceType.CPU, DeviceType.MKLDNN, DeviceType.IDEEP]
else 0
)
def _cuda_memory_usage(mem_record):
return (
mem_record.nbytes()
if mem_record.device_type() in [DeviceType.CUDA, DeviceType.HIP]
else 0
)
def _privateuse1_memory_usage(mem_record):
return (
mem_record.nbytes()
if mem_record.device_type() in [DeviceType.PrivateUse1]
else 0
)
# Create and return FunctionEvent list
function_events = []
device_corr_map: Dict[int, List[FunctionEvent]] = {}
max_evt_id = 0
for kineto_event in result.events():
if _filter_name(kineto_event.name()):
continue
rel_start_us = kineto_event.start_us() - trace_start_us
rel_end_us = rel_start_us + kineto_event.duration_us()
abs_end_us = kineto_event.start_us() + kineto_event.duration_us()
cpu_memory_usage = 0
cuda_memory_usage = 0
privateuse1_memory_usage = 0
if kineto_event.device_type() == DeviceType.CPU:
# find the corresponding memory allocation events
for mem_record in mem_records_acc.in_interval(
kineto_event.start_us(), abs_end_us
):
cpu_memory_usage += _cpu_memory_usage(mem_record[0])
cuda_memory_usage += _cuda_memory_usage(mem_record[0])
privateuse1_memory_usage += _privateuse1_memory_usage(mem_record[0])
mem_record[1] = True
is_async = kineto_event.is_async() or (
kineto_event.start_thread_id() != kineto_event.end_thread_id()
)
fe = FunctionEvent(
id=kineto_event.correlation_id(),
name=_rewrite_name(name=kineto_event.name(), with_wildcard=True),
trace_name=_rewrite_name(name=kineto_event.name(), with_wildcard=False),
thread=kineto_event.start_thread_id(),
start_us=rel_start_us,
end_us=rel_end_us,
fwd_thread=kineto_event.fwd_thread_id(),
input_shapes=kineto_event.shapes(),
concrete_inputs=kineto_event.concrete_inputs(),
stack=[
entry
for entry in kineto_event.stack()
if _filter_stack_entry(entry)
],
scope=kineto_event.scope(),
use_device=self.use_device,
cpu_memory_usage=cpu_memory_usage,
cuda_memory_usage=cuda_memory_usage,
privateuse1_memory_usage=privateuse1_memory_usage,
is_async=is_async,
sequence_nr=kineto_event.sequence_nr(),
device_type=kineto_event.device_type(),
device_index=kineto_event.device_index(),
flops=kineto_event.flops(),
)
max_evt_id = max(max_evt_id, fe.id)
if fe.device_type == DeviceType.CPU and not fe.is_async:
if self.use_device:
privateuse1_time = kineto_event.privateuse1_elapsed_us()
if privateuse1_time > 0:
fe.append_kernel(fe.name, fe.device_index, privateuse1_time)
fe.is_legacy = True
else:
# Check if we have CUDA time as a fallback
cuda_time = kineto_event.cuda_elapsed_us()
if cuda_time > 0:
fe.append_kernel(fe.name, fe.device_index, cuda_time)
fe.is_legacy = True
function_events.append(fe)
corr_id = kineto_event.linked_correlation_id()
if corr_id > 0:
if corr_id not in device_corr_map:
device_corr_map[corr_id] = []
device_corr_map[corr_id].append(fe)
# associate CUDA kernels and CUDA runtime (CPU) with CPU events
for fe in function_events:
if (
fe.device_type == DeviceType.CPU
and not fe.is_async
and fe.id in device_corr_map
):
for f_evt in device_corr_map[fe.id]:
if f_evt.device_type == DeviceType.CUDA:
fe.append_kernel(
f_evt.name,
f_evt.device_index,
f_evt.time_range.end - f_evt.time_range.start,
)
elif f_evt.device_type == DeviceType.CPU:
# make sure that 'thread' of a CPU Kineto (e.g. CUDA Runtime) event is associated
# with the 'thread' of the corresponding linked PyTorch event to properly track
# parents and children
f_evt.thread = fe.thread
def createFunctionEventForMemoryEvents(evt):
rel_start_us = evt.start_us() - trace_start_us
fe = FunctionEvent(
id=max_evt_id,
name=evt.name(),
trace_name=None, # not outputting in the trace
thread=evt.start_thread_id(),
start_us=rel_start_us,
end_us=rel_start_us, # no duration
fwd_thread=evt.start_thread_id(),
input_shapes=[],
stack=[],
scope=0, # RecordScope::FUNCTION
use_device=self.use_device,
cpu_memory_usage=_cpu_memory_usage(evt),
cuda_memory_usage=_cuda_memory_usage(evt),
privateuse1_memory_usage=_privateuse1_memory_usage(evt),
is_async=False,
sequence_nr=-1,
device_type=DeviceType.CPU,
device_index=0,
)
return fe
# output top-level memory events
for mem_record in mem_records:
if not mem_record[1]:
max_evt_id += 1
fe = createFunctionEventForMemoryEvents(mem_record[0])
function_events.append(fe)
for oom_record in oom_records:
max_evt_id += 1
fe = createFunctionEventForMemoryEvents(oom_record)
function_events.append(fe)
function_events.sort(
key=lambda evt: [evt.time_range.start, -evt.time_range.end]
)
return function_events
class record_function(_ContextDecorator):
"""Context manager/function decorator that adds a label to a code block/function when running autograd profiler.
It is useful when tracing the code profile.
Args:
name (str): Label assigned to the block of code.
node_id (int): ID of node, for distributed profiling. Unset in
non-distributed cases.
Example:
>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_AUTOGRAD_PROFILER)
>>> x = torch.randn((1, 1), requires_grad=True)
>>> with torch.autograd.profiler.profile() as prof:
... y = x ** 2
... with torch.autograd.profiler.record_function("label-z"): # label the block
... z = y ** 3
... y.backward()
...
>>> # xdoctest: +IGNORE_WANT
>>> # NOTE: some columns were removed for brevity
>>> print(prof.key_averages().table(sort_by="self_cpu_time_total"))
----------------------------------- --------------- --------------- ---------------
Name Self CPU total % CPU time avg Number of Calls
----------------------------------- --------------- --------------- ---------------
pow 60.77% 47.470us 3
mul 21.73% 25.465us 2
PowBackward0 12.03% 121.891us 1
torch::autograd::AccumulateGrad 2.70% 6.324us 1
label-z 2.13% 12.421us 1
torch::autograd::GraphRoot 0.64% 1.503us 1
----------------------------------- --------------- --------------- ---------------
Self CPU time total: 234.344us
CUDA time total: 0.000us
"""
def __init__(self, name: str, args: Optional[str] = None):
self.name: str = name
self.args: Optional[str] = args
# Whether or not we should run record function's end callbacks when exiting.
self.run_callbacks_on_exit: bool = True
# TODO: TorchScript ignores standard type annotation here
# self.record: Optional["torch.classes.profiler._RecordFunction"] = None
self.record = torch.jit.annotate(
Optional["torch.classes.profiler._RecordFunction"], None
)
def __enter__(self):
self.record = torch.ops.profiler._record_function_enter_new(
self.name, self.args
)
return self
def __exit__(self, exc_type: Any, exc_value: Any, traceback: Any):
if not self.run_callbacks_on_exit:
return
# Local variable is needed by TorchScript to refine Optional[T] to T
record = self.record
assert record is not None
# TODO: Too slow with __torch_function__ handling enabled
# See https://github.com/pytorch/pytorch/issues/76410
if not torch.jit.is_scripting():
with torch._C.DisableTorchFunctionSubclass():
torch.ops.profiler._record_function_exit._RecordFunction(record)
else:
torch.ops.profiler._record_function_exit(record)
def _call_end_callbacks_on_future(self, fut: Future[Any]) -> Future[Any]:
"""Use for profiling async calls that return a future.
Calling this function will extend recording beyond this scope, until the future is
satisfied. It is useful for profiling the end to end time of asynchronous calls.
This function should only be called once to attach the callback onto the future, and
will throw if called multiple times.
Args:
fut: (torch._C.Future): future for which to schedule
callback for.
Returns:
A future that completes with the value of the passed in future when
the profiling callbacks have ran.
"""
# Throw if we have already attached a callback onto the future.
if not self.run_callbacks_on_exit:
raise RuntimeError("_call_end_callbacks_on_future can only be called once.")
# We are scheduling to run this RecordFunction's end callbacks when the
# passed in future completes, so don't run end callbacks on exit.
self.run_callbacks_on_exit = False
# Local variable is needed by TorchScript to refine Optional[T] to T
record = self.record
assert record is not None
# TODO: Too slow with __torch_function__ handling enabled
# See https://github.com/pytorch/pytorch/issues/76410
if not torch.jit.is_scripting():
with torch._C.DisableTorchFunctionSubclass():
profiled_future = (
torch.ops.profiler._call_end_callbacks_on_jit_fut._RecordFunction(
record, fut
)
)
else:
profiled_future = torch.ops.profiler._call_end_callbacks_on_jit_fut(
record, fut
)
return profiled_future
class emit_itt:
"""Context manager that makes every autograd operation emit an ITT range.
It is useful when running the program under Intel(R) VTune Profiler::
vtune <--vtune-flags> <regular command here>
The Instrumentation and Tracing Technology (ITT) API enables your application to generate and
control the collection of trace data during its execution across different Intel tools.
This context manager is to annotate Intel(R) VTune Profiling trace. With help of this context manager,
you will be able to see labled ranges in Intel(R) VTune Profiler GUI.
.. warning:
This context manager should not be called recursively, i.e. at most one
instance should be enabled at any given time.
Args:
enabled (bool, optional): Setting ``enabled=False`` makes this context manager a no-op.
Default: ``True``.
record_shapes (bool, optional): If ``record_shapes=True``, the itt range wrapping
each autograd op will append information about the sizes of Tensor arguments received
by that op, in the following format:
``[[arg0.size(0), arg0.size(1), ...], [arg1.size(0), arg1.size(1), ...], ...]``
Non-tensor arguments will be represented by ``[]``.
Arguments will be listed in the order they are received by the backend op.
Please note that this order may not match the order in which those arguments were passed
on the Python side. Also note that shape recording may increase the overhead of itt range creation.
Default: ``False``
Example:
>>> # xdoctest: +SKIP("Undefined variables")
>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_AUTOGRAD_PROFILER)
>>> with torch.autograd.profiler.emit_itt():
... model(x)
"""
def __init__(self, enabled=True, record_shapes=False):
self.enabled = enabled
self.entered = False
self.record_shapes = record_shapes
def __enter__(self):
if not self.enabled:
return
if self.entered:
raise RuntimeError("ITT annotation context manager is not reentrant")
self.entered = True
_run_on_profiler_start()
_enable_profiler(
ProfilerConfig(
ProfilerState.ITT,
self.record_shapes,
False,
False,
False,
False,
_ExperimentalConfig(),
),
set(),
)
return self
def __exit__(self, exc_type, exc_val, exc_tb):
if not self.enabled:
return
_disable_profiler()
_run_on_profiler_stop()
return False
class emit_nvtx:
"""Context manager that makes every autograd operation emit an NVTX range.
It is useful when running the program under nvprof::
nvprof --profile-from-start off -o trace_name.prof -- <regular command here>
Unfortunately, there's no way to force nvprof to flush the data it collected
to disk, so for CUDA profiling one has to use this context manager to annotate
nvprof traces and wait for the process to exit before inspecting them.
Then, either NVIDIA Visual Profiler (nvvp) can be used to visualize the timeline, or
:func:`torch.autograd.profiler.load_nvprof` can load the results for inspection
e.g. in Python REPL.
.. warning:
This context manager should not be called recursively, i.e. at most one
instance should be enabled at any given time.
Args:
enabled (bool, optional): Setting ``enabled=False`` makes this context manager a no-op.
Default: ``True``.
record_shapes (bool, optional): If ``record_shapes=True``, the nvtx range wrapping
each autograd op will append information about the sizes of Tensor arguments received
by that op, in the following format:
``[[arg0.size(0), arg0.size(1), ...], [arg1.size(0), arg1.size(1), ...], ...]``
Non-tensor arguments will be represented by ``[]``.
Arguments will be listed in the order they are received by the backend op.
Please note that this order may not match the order in which those arguments were passed
on the Python side. Also note that shape recording may increase the overhead of nvtx range creation.
Default: ``False``
Example:
>>> # xdoctest: +SKIP("undefined variables")
>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_AUTOGRAD_PROFILER)
>>> with torch.cuda.profiler.profile():
... model(x) # Warmup CUDA memory allocator and profiler
... with torch.autograd.profiler.emit_nvtx():
... model(x)
**Forward-backward correlation**
When viewing a profile created using :class:`emit_nvtx` in the Nvidia Visual Profiler,
correlating each backward-pass op with the corresponding forward-pass op can be difficult.
To ease this task, :class:`emit_nvtx` appends sequence number information to the ranges it
generates.
During the forward pass, each function range is decorated with ``seq=<N>``. ``seq`` is a running
counter, incremented each time a new backward Function object is created and stashed for backward.
Thus, the ``seq=<N>`` annotation associated with each forward function range tells you that
if a backward Function object is created by this forward function,
the backward object will receive sequence number N.
During the backward pass, the top-level range wrapping each C++ backward Function's
``apply()`` call is decorated with ``stashed seq=<M>``. ``M`` is the sequence number that
the backward object was created with. By comparing ``stashed seq`` numbers in backward with ``seq``
numbers in forward, you can track down which forward op created each backward Function.
Any functions executed during the backward pass are also decorated with ``seq=<N>``. During
default backward (with ``create_graph=False``) this information is irrelevant, and in fact,
``N`` may simply be 0 for all such functions. Only the top-level ranges associated with
backward Function objects' ``apply()`` methods are useful, as a way to correlate these Function
objects with the earlier forward pass.
**Double-backward**
If, on the other hand, a backward pass with ``create_graph=True`` is underway (in other words,
if you are setting up for a double-backward), each function's execution during backward
is given a nonzero, useful ``seq=<N>``. Those functions may themselves create Function objects
to be executed later during double-backward, just as the original functions in the forward pass did.
The relationship between backward and double-backward is conceptually the same as the relationship
between forward and backward: The functions still emit current-sequence-number-tagged ranges,
the Function objects they create still stash those sequence numbers, and during the eventual
double-backward, the Function objects' ``apply()`` ranges are still tagged with ``stashed seq``
numbers, which can be compared to `seq` numbers from the backward pass.
.. warning:
The sequence number is thread-local, and some forward functions don't create an associated
backward Function object (instead delegating that to sub-functions further down the call chain).
For these reasons, the correspondence of stashed sequence numbers in
backward Function ``apply()`` ranges with `seq` numbers in forward-pass ranges is
not guaranteed to be 1 to 1. The sequence numbers alone may not be enough to fully
disambiguate which forward function created which
backward Function object. You may need to make a judgment based on analytic knowledge of what
the expected correspondence should be.
"""
def __init__(self, enabled=True, record_shapes=False):
self.enabled = enabled
self.entered = False
self.record_shapes = record_shapes
def __enter__(self):
if not self.enabled:
return
if self.entered:
raise RuntimeError("NVTX annotation context manager is not reentrant")
self.entered = True
torch.cuda.synchronize()
_run_on_profiler_start()
_enable_profiler(
ProfilerConfig(
ProfilerState.NVTX,
self.record_shapes,
False,
False,
False,
False,
_ExperimentalConfig(),
),
set(),
)
return self
def __exit__(self, exc_type, exc_val, exc_tb):
if not self.enabled:
return
torch.cuda.synchronize()
_disable_profiler()
_run_on_profiler_stop()
return False
def load_nvprof(path):
"""Open an nvprof trace file and parses autograd annotations.
Args:
path (str): path to nvprof trace
"""
return EventList(parse_nvprof_trace(path))
class EnforceUnique:
"""Raises an error if a key is seen more than once."""
def __init__(self):
self.seen = set()
def see(self, *key):
r"""
Observe a key and raise an error if it is seen multiple times.
"""
if key in self.seen:
raise RuntimeError("duplicate key: " + str(key))
self.seen.add(key)
def parse_nvprof_trace(path):
import sqlite3
conn = sqlite3.connect(path)
conn.row_factory = sqlite3.Row
# Parse strings table
strings = {}
for r in conn.execute("SELECT _id_ as id, value FROM StringTable"):
strings[r["id"]] = torch._C._demangle(r["value"])
# First, find all functions and create FunctionEvents for them
marker_query = """
SELECT
start.id AS marker_id, start.name, start.timestamp AS start_time, end.timestamp AS end_time
FROM
CUPTI_ACTIVITY_KIND_MARKER AS start INNER JOIN CUPTI_ACTIVITY_KIND_MARKER AS end
ON start.id = end.id
WHERE
start.name != 0 AND end.name = 0
"""
functions = []
functions_map = {}
unique = EnforceUnique()
for row in conn.execute(marker_query):
unique.see(row["marker_id"])
evt = FunctionEvent(
id=row["marker_id"],
node_id=0, # missing a node_id when calling FunctionEvent. This is just to ensure
# that pytorch doesn't crash when creating a FunctionEvent() object
name=strings[row["name"]],
start_us=row["start_time"],
end_us=row["end_time"],
thread=0,
) # TODO: find in sqlite database
functions.append(evt)
functions_map[evt.id] = evt
# Now, correlate all kernels with FunctionEvents
kernel_query = """
SELECT
start.id AS marker_id, start.name, start.timestamp, end.timestamp,
runtime._id_ AS runtime_id, runtime.cbid, runtime.start AS runtime_start, runtime.end AS runtime_end,
kernel.start AS kernel_start, kernel.end AS kernel_end, kernel.name AS kernel_name
FROM
CUPTI_ACTIVITY_KIND_MARKER AS start
INNER JOIN CUPTI_ACTIVITY_KIND_MARKER AS end
ON start.id = end.id
INNER JOIN CUPTI_ACTIVITY_KIND_RUNTIME as runtime
ON (start.timestamp < runtime.start AND runtime.end < end.timestamp)
INNER JOIN CUPTI_ACTIVITY_KIND_CONCURRENT_KERNEL AS kernel
ON kernel.correlationId = runtime.correlationId
"""
unique = EnforceUnique()
for row in conn.execute(kernel_query):
unique.see(row["marker_id"], row["runtime_id"])
# 211 is cudaKernelLaunch for cuda >= 9.2
assert row["cbid"] == 211
evt = functions_map[row["marker_id"]]
evt.append_kernel(
row["kernel_name"], 0, row["kernel_end"] - row["kernel_start"]
)
functions.sort(key=lambda evt: evt.time_range.start)
return functions
class KinetoStepTracker:
"""Provides an abstraction for incrementing the step count globally.
Previously, we only had one place to mark that a step() has occurred
in the program via pytorch profiler step(). We will now add step hooks
in the Optimizer class https://github.com/pytorch/pytorch/issues/88446
- This could mean programs that already call profiler.step() every
iteration can end up double incrementing step count.
- If a model uses multiple optimizers we can also have double or more
counting of the step.
We fix this by adding a layer of abstraction before calling step()
to the kineto library. The idea is to maintain steps per requester in a dict:
.. code-block::
{
"ProfilerStep": 100, # triggered by profiler step() call
"Optimizer1Step": 100, # Optimizer 1 or 2 are just examples, could be SGD, Adam etc
"Optimizer2Step": 100,
}
To figure out the global step count just take the max of dict values (100).
If one of the count increments the max will go up.
.. code-block::
{
"ProfilerStep": 100,
"Optimizer1Step": 101, # Optimizer1 got incremented first say
"Optimizer2Step": 100,
}
Then global step count is 101
We only call the kineto step() function when global count increments.
NOTE: Please do not use the KinetoStepTracker in modules beside the Optimizer
for now. The result could be incorrect increments of the step count.
"""
_current_step = 0
_step_dict: Dict[str, int] = defaultdict(int)
@classmethod
def init_step_count(cls, requester: str):
r"""
Initialize for a given requester.
"""
cls._step_dict[requester] = cls._current_step
@classmethod
def erase_step_count(cls, requester: str) -> bool:
r"""
Remove a given requester.
"""
return cls._step_dict.pop(requester, None) is not None
@classmethod
def increment_step(cls, requester: str) -> int:
"""Increments the step count for the requester.
Additionally if the max over all step counts has incremented then
trigger the _kineto_step() returns global step count
"""
if requester not in cls._step_dict:
cls.init_step_count(requester)
cls._step_dict[requester] += 1
new_step = max(cls._step_dict.values())
if new_step > cls._current_step:
delta = new_step - cls._current_step
if delta > 1:
warn(
"Profiler step count has increased more than 1 - "
f"current_step = {cls._current_step} step dict = {cls._step_dict}"
)
for _ in range(0, delta):
_kineto_step()
cls._current_step = new_step
return cls._current_step
@classmethod
def current_step(cls) -> int:
r"""
Get the latest step for any requester
"""
return cls._current_step