Skip to content
New issue

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

Sign up for GitHub

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

Already on GitHub? Sign in to your account

Jump to bottom

[profiler] Support top-level memory events #51421

Closed
wants to merge 5 commits into from
Closed

[profiler] Support top-level memory events #51421

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
289a14b
[profiler] Support top-level memory events
Jan 31, 2021
6e974a3
Update on "[profiler] Support top-level memory events"
Feb 2, 2021
7d1d418
Update on "[profiler] Support top-level memory events"
Feb 3, 2021
479a7e0
Update on "[profiler] Support top-level memory events"
Feb 3, 2021
326f80c
Update on "[profiler] Support top-level memory events"
Feb 4, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
4 changes: 4 additions & 0 deletions c10/core/CPUAllocator.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -292,6 +292,10 @@ void ProfiledCPUMemoryReporter::Delete(void* ptr) {
allocated = allocated_;
nbytes = it->second;
size_table_.erase(it);
} else {
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

are there any changes required in CUDAAllocator?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

CUDACachingAllocator already saves block sizes

C10_LOG_EVERY_MS(WARNING, 1000)
<< "Memory block of unknown size was allocated before the profiling started, "
<< "profiler results will not include the deallocation event";
}
}
if (nbytes == 0) {
Expand Down
109 changes: 0 additions & 109 deletions test/test_autograd.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3258,115 +3258,6 @@ def test_profiler_aggregation_lstm(self):
with tempfile.NamedTemporaryFile() as trace_file:
prof.export_chrome_trace(trace_file.name)

def test_memory_profiler(self):
def run_profiler(tensor_creation_fn, metric):
# collecting allocs / deallocs
with profile(profile_memory=True, record_shapes=True, use_kineto=kineto_available()) as prof:
x = None
with record_function("test_user_scope_alloc"):
x = tensor_creation_fn()
with record_function("test_user_scope_dealloc"):
del x
stats = prof.key_averages(group_by_input_shape=True)
print(stats.table(sort_by=metric))
return stats

def check_metrics(stats, metric, allocs=None, deallocs=None):
stat_metrics = {}
for stat in stats:
stat_metrics[stat.key] = getattr(stat, metric)
if allocs is not None:
for alloc_fn in allocs:
self.assertTrue(alloc_fn in stat_metrics)
self.assertTrue(stat_metrics[alloc_fn] > 0)
if deallocs is not None:
for dealloc_fn in deallocs:
self.assertTrue(dealloc_fn in stat_metrics)
self.assertTrue(stat_metrics[dealloc_fn] < 0)

def create_cpu_tensor():
return torch.rand(10, 10)

def create_cuda_tensor():
return torch.rand(10, 10).cuda()

def create_mkldnn_tensor():
return torch.rand(10, 10, dtype=torch.float32).to_mkldnn()

print("Running CPU test")
stats = run_profiler(create_cpu_tensor, "cpu_memory_usage")
check_metrics(
stats,
"cpu_memory_usage",
allocs=[
"aten::empty",
"aten::rand",
"test_user_scope_alloc",
],
deallocs=[
"test_user_scope_dealloc",
]
)

if torch.cuda.is_available():
create_cuda_tensor()
print("Running CUDA test")
stats = run_profiler(create_cuda_tensor, "cuda_memory_usage")
check_metrics(
stats,
"cuda_memory_usage",
allocs=[
"test_user_scope_alloc",
"aten::to",
"aten::empty_strided",
],
deallocs=[
"test_user_scope_dealloc",
]
)
check_metrics(
stats,
"cpu_memory_usage",
allocs=[
"aten::rand",
"aten::empty",
]
)

if torch._C.has_mkldnn:
create_mkldnn_tensor()
print("Running MKLDNN test")
stats = run_profiler(create_mkldnn_tensor, "cpu_memory_usage")
check_metrics(
stats,
"cpu_memory_usage",
allocs=[
"test_user_scope_alloc",
"aten::rand",
"aten::empty",
"aten::to_mkldnn",
],
deallocs=[
"test_user_scope_dealloc",
]
)

# check partial overlap of tensor allocation with memory profiler
x = torch.rand(10, 10)
with profile(profile_memory=True, record_shapes=True, use_kineto=kineto_available()) as prof:
del x
x = torch.rand(10, 10)
del x
stats = prof.key_averages(group_by_input_shape=True)
check_metrics(
stats,
"cpu_memory_usage",
allocs=[
"aten::rand",
"aten::empty",
]
)

def test_record_function(self):
x = torch.randn(10, 10)

Expand Down
141 changes: 129 additions & 12 deletions test/test_profiler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,9 @@
from torch.testing._internal.common_utils import (
TestCase, run_tests, TEST_WITH_ASAN, IS_WINDOWS, TemporaryFileName, TemporaryDirectoryName)
from torch.autograd.profiler import profile as _profile
from torch.profiler import profile, kineto_available, DeviceType, ProfilerActivity
from torch.profiler import (
kineto_available, profile, record_function, DeviceType, ProfilerActivity
)

try:
import psutil
Expand Down Expand Up @@ -92,10 +94,6 @@ def forward(self, x):
c = b.sum()
c.backward()

print(p.key_averages(
group_by_stack_n=5).table(
sort_by="self_cpu_time_total", row_limit=-1))

for e in p.function_events:
if "aten::add" in e.name or "AddBackward" in e.name:
self.assertTrue(any(["test_profiler" in entry for entry in e.stack]))
Expand All @@ -122,8 +120,9 @@ def test_kineto(self):
# rerun to avoid initial start overhead
with _profile(use_cuda=True, use_kineto=True) as p:
self.payload()
print(p.key_averages().table(
sort_by="self_cuda_time_total", row_limit=-1))
output = p.key_averages().table(
sort_by="self_cuda_time_total", row_limit=-1)
# print(output)
found_gemm = False
found_memcpy = False
for e in p.function_events:
Expand Down Expand Up @@ -163,6 +162,123 @@ def test_kineto_multigpu(self):
self.assertTrue(found_gemm_1)
self.assertTrue(found_cuda)

def test_memory_profiler(self):
def run_profiler(tensor_creation_fn, metric):
# collecting allocs / deallocs
with _profile(profile_memory=True, record_shapes=True, use_kineto=kineto_available()) as prof:
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

we should make a follow-up item to either migrate or duplicate the tests with profiler instead of autograd.profiler.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

all our autograd.profiler tests should use use_kineto=kineto_available() now, and we have both kineto-enabled and not enabled CI builds; eventually when autograd.profiler is deprecated and we always use kineto, we will just make the tests use the new, kineto-only api

x = None
with record_function("test_user_scope_alloc"):
x = tensor_creation_fn()
with record_function("test_user_scope_dealloc"):
del x
return prof.key_averages(group_by_input_shape=True)

def check_metrics(stats, metric, allocs=None, deallocs=None):
stat_metrics = {}
for stat in stats:
stat_metrics[stat.key] = getattr(stat, metric)
if allocs is not None:
for alloc_fn in allocs:
self.assertTrue(alloc_fn in stat_metrics)
self.assertTrue(stat_metrics[alloc_fn] > 0)
if deallocs is not None:
for dealloc_fn in deallocs:
self.assertTrue(dealloc_fn in stat_metrics)
self.assertTrue(stat_metrics[dealloc_fn] < 0)

def create_cpu_tensor():
return torch.rand(10, 10)

def create_cuda_tensor():
return torch.rand(10, 10).cuda()

def create_mkldnn_tensor():
return torch.rand(10, 10, dtype=torch.float32).to_mkldnn()

stats = run_profiler(create_cpu_tensor, "cpu_memory_usage")
check_metrics(
stats,
"cpu_memory_usage",
allocs=[
"aten::empty",
"aten::rand",
"test_user_scope_alloc",
],
deallocs=[
"test_user_scope_dealloc",
]
)

if torch.cuda.is_available():
create_cuda_tensor()
stats = run_profiler(create_cuda_tensor, "cuda_memory_usage")
check_metrics(
stats,
"cuda_memory_usage",
allocs=[
"test_user_scope_alloc",
"aten::to",
"aten::empty_strided",
],
deallocs=[
"test_user_scope_dealloc",
]
)
check_metrics(
stats,
"cpu_memory_usage",
allocs=[
"aten::rand",
"aten::empty",
]
)

if torch._C.has_mkldnn:
create_mkldnn_tensor()
stats = run_profiler(create_mkldnn_tensor, "cpu_memory_usage")
check_metrics(
stats,
"cpu_memory_usage",
allocs=[
"test_user_scope_alloc",
"aten::rand",
"aten::empty",
"aten::to_mkldnn",
],
deallocs=[
"test_user_scope_dealloc",
]
)

# check top-level memory events
with _profile(profile_memory=True, use_kineto=kineto_available()) as prof:
x = torch.rand(10, 10)
del x
if torch.cuda.is_available():
y = torch.rand(10, 10).cuda()
del y
gc.collect()
stats = prof.key_averages(group_by_input_shape=True)
check_metrics(
stats,
"cpu_memory_usage",
allocs=[
"aten::rand",
"aten::empty"
],
deallocs=[
"[memory]"
]
)
if torch.cuda.is_available():
check_metrics(
stats,
"cuda_memory_usage",
deallocs=[
"[memory]"
]
)

def test_high_level_trace(self):
"""Checks that python side high level events are recorded.
"""
Expand Down Expand Up @@ -267,7 +383,6 @@ def test_flops(self):
with _profile(record_shapes=True, with_flops=True, use_kineto=kineto_available()) as prof:
model(inputs)
profiler_output = prof.key_averages(group_by_input_shape=True).table(sort_by="cpu_time_total", row_limit=10)
print(profiler_output)
self.assertIn("FLOPS", profiler_output)

@unittest.skipIf(not kineto_available(), "Kineto is required")
Expand All @@ -279,8 +394,9 @@ def test_kineto_profiler_api(self):
self.payload()

def trace_handler(p):
print(p.key_averages().table(
sort_by="self_cuda_time_total", row_limit=-1))
output = p.key_averages().table(
sort_by="self_cuda_time_total", row_limit=-1)
# print(output)
# p.export_chrome_trace("/tmp/test_trace_" + str(called_num[0]) + ".json")
called_num[0] += 1

Expand Down Expand Up @@ -308,8 +424,9 @@ def trace_handler(p):
) as p:
self.payload()
self.payload()
print(p.key_averages().table(
sort_by="self_cuda_time_total", row_limit=-1))
output = p.key_averages().table(
sort_by="self_cuda_time_total", row_limit=-1)
# print(output)

def test_export_stacks(self):
with _profile(with_stack=True, use_kineto=kineto_available()) as p:
Expand Down
2 changes: 1 addition & 1 deletion torch/_C/__init__.pyi.in
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -654,7 +654,7 @@ class _TensorBase(object):
output_nr: _int
_version: _int
_base: Optional[Tensor]
_cdata: _int
_cdata: _int
grad_fn: Any
_grad_fn: Any
_grad: Optional[Tensor]
Expand Down
12 changes: 6 additions & 6 deletions torch/_torch_docs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5151,10 +5151,10 @@ def merge_dicts(*dicts):
add_docstr(torch.fmax, r"""
fmax(input, other, *, out=None) -> Tensor

Computes the element-wise maximum of :attr:`input` and :attr:`other`.
Computes the element-wise maximum of :attr:`input` and :attr:`other`.

This is like :func:`torch.maximum` except it handles NaNs differently:
if exactly one of the two elements being compared is a NaN then the non-NaN element is taken as the maximum.
This is like :func:`torch.maximum` except it handles NaNs differently:
if exactly one of the two elements being compared is a NaN then the non-NaN element is taken as the maximum.
Only if both elements are NaN is NaN propagated.

This function is a wrapper around C++'s ``std::fmax`` and is similar to NumPy's ``fmax`` function.
Expand Down Expand Up @@ -5626,10 +5626,10 @@ def merge_dicts(*dicts):
add_docstr(torch.fmin, r"""
fmin(input, other, *, out=None) -> Tensor

Computes the element-wise minimum of :attr:`input` and :attr:`other`.
Computes the element-wise minimum of :attr:`input` and :attr:`other`.

This is like :func:`torch.minimum` except it handles NaNs differently:
if exactly one of the two elements being compared is a NaN then the non-NaN element is taken as the minimum.
This is like :func:`torch.minimum` except it handles NaNs differently:
if exactly one of the two elements being compared is a NaN then the non-NaN element is taken as the minimum.
Only if both elements are NaN is NaN propagated.

This function is a wrapper around C++'s ``std::fmin`` and is similar to NumPy's ``fmin`` function.
Expand Down