Source code level attribution in profiler #43898

ilia-cher · 2020-08-31T19:50:58Z

Stack from ghstack:

Source code level attribution in profiler #43898 Source code level attribution in profiler

Summary:
Adding with_source parameter to enable tracking source code
(filename and line) in profiler for eager, torchscript and autograd
modes

Test Plan:
python test/test_profiler.py

Output: https://gist.github.com/ilia-cher/bd13a3ab0c4905ce58b0079df66711fa

Benchmark:
updated benchmark: profiler https://gist.github.com/ilia-cher/ff7064a311790590ffa8445c95d4d45b
as expected, no change when no stacks are used, slower when stacks are used (in particular python is slower than TS callstack)

overhead: https://gist.github.com/ilia-cher/8a2861863d3033b56181d23ba4cfbda1
overall as expected:
small increase in cpu instructions (0.7%), small increase in tls accesses expected (bc of function.h change)

https://gist.github.com/ilia-cher/9c7901670fb48082af92386b1889e2fe
https://gist.github.com/ilia-cher/69a01a6ef294511a797e8cf4a1e6db75

Differential Revision: D23432664

Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Name Self CPU total % Self CPU total CPU total % CPU total CPU time avg Number of Calls Source Location ----------------------------------- --------------- --------------- --------------- --------------- --------------- --------------- -------------------------------------------- ts_method_1 10.43% 235.364us 36.46% 822.920us 822.920us 1 test/test_profiler.py(70): test_source aten::add 7.52% 169.833us 8.88% 200.439us 200.439us 1 test/test_profiler.py(69): test_source aten::normal_ 6.26% 141.380us 6.26% 141.380us 141.380us 1 test/test_profiler.py(67): test_source aten::add 5.80% 130.830us 8.41% 189.800us 63.267us 3 test/test_profiler.py(72): test_source aten::sum 5.02% 113.340us 8.39% 189.475us 189.475us 1 test/test_profiler.py(64): ts_method_1 aten::add 4.58% 103.346us 6.33% 142.847us 142.847us 1 test/test_profiler.py(62): ts_method_1 aten::mul 4.05% 91.498us 9.62% 217.113us 217.113us 1 test/test_profiler.py(71): test_source aten::add 4.03% 90.880us 5.60% 126.405us 126.405us 1 test/test_profiler.py(58): ts_method_2 aten::empty 3.49% 78.735us 3.49% 78.735us 19.684us 4 test/test_profiler.py(72): test_source Reviewers:

dr-ci · 2020-08-31T20:00:56Z

💊 CI failures summary and remediations

As of commit 196ba66 (more details on the Dr. CI page):

1/2 failures possibly* introduced in this PR
- 1/1 non-CircleCI failure(s)
1/2 tentatively recognized as flaky ❄️
- Click here to rerun these jobs

❄️ 1 failure tentatively classified as flaky

but reruns have not yet been triggered to confirm:

pytorch_macos_10_13_py3_test (1/1)

Step: "Update Homebrew" (full log | diagnosis details | 🔁 rerun) ❄️

fatal: Could not read from remote repository.

Receiving objects:  98% (175/178) Receiving objects:  99% (177/178) Receiving objects: 100% (178/178) Receiving objects: 100% (178/178), 63.90 KiB | 10.65 MiB/s, done. 
Resolving deltas:  96% (89/92) Resolving deltas:  97% (90/92) Resolving deltas: 100% (92/92) Resolving deltas: 100% (92/92), completed with 85 local objects. 
From ssh://github.com/Homebrew/homebrew-cask-versions 
 + 15f6b44...90ed6b8 master     -> origin/master  (forced update) 
+ git reset --hard origin/master 
HEAD is now at 90ed6b8 Update microsoft-edge-beta from 86.0.622.19 to 86.0.622.28 (#9686) 
+ for path in '$(find /usr/local/Homebrew -type d -name .git)' 
+ cd /usr/local/Homebrew/Library/Taps/homebrew/homebrew-core/.git/.. 
+ git fetch --depth=1 origin 
Connection to github.com closed by remote host.  
fatal: Could not read from remote repository. 
 
Please make sure you have the correct access rights 
and the repository exists.

ci.pytorch.org: 1 failed

Failed: pr/pytorch-linux-bionic-rocm3.7-py3.6

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group.

See how this bot performed.

This comment has been revised 140 times.

ngimel

Please benchmark the added overheads and make sure it's acceptable with disabled profiling/profiling w/o source attribution.

test/test_profiler.py

torch/csrc/autograd/profiler.cpp

dzhulgakov

@suo @zdevito on TorchScript changes. I left some comments, but better suggestions are welcomed

Perf measurements are absolutely a must. Both when source attribution is off and when it's on

test/test_profiler.py

torch/autograd/profiler.py

dzhulgakov · 2020-09-11T03:46:27Z

torch/csrc/autograd/function.h

      uint64_t sequence_nr,
      edge_list&& next_edges = edge_list())
      : sequence_nr_(sequence_nr),
+      thread_id_(at::RecordFunction::currentThreadId()),


cc @albanD - is it a concern that we increase size of Node? Maybe we could do some packing of both indices in a single int64_t? I'm not sure how in general sequence_nrs work - are they thread local?

It's not something I like to see for sure. Though I never actually did a full benchmark of the impact on memory and speed.

Do you actually always need this thread id or just when the profiler is running?
If so, I think a nice way to guard this to only run when the profiler runs and store it in the metadata struct that we also use for anomaly mode to store this (you can make the python dict initialization lazy to improve perf when you just store and read the thread_id).

I'm not sure how in general sequence_nrs work - are they thread local?

Yes these are just thread local variables (not propagated by TLSState)

we then should either set or not set both of them, because seq. number without thread is not very useful;
if extra int64 is too much I'd better go with packing though - less runtime checks

The autograd engine make heavy use of the seq number (that does not require the thread). So we have to set this one.

torch/csrc/autograd/profiler.cpp

torch/csrc/autograd/profiler.h

torch/csrc/jit/runtime/interpreter.cpp

[wip] Source code level attribution in profiler Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Name Self CPU total % Self CPU total CPU total % CPU total CPU time avg Number of Calls Source Location ----------------------------------- --------------- --------------- --------------- --------------- --------------- --------------- -------------------------------------------- ts_method_1 10.43% 235.364us 36.46% 822.920us 822.920us 1 test/test_profiler.py(70): test_source aten::add 7.52% 169.833us 8.88% 200.439us 200.439us 1 test/test_profiler.py(69): test_source aten::normal_ 6.26% 141.380us 6.26% 141.380us 141.380us 1 test/test_profiler.py(67): test_source aten::add 5.80% 130.830us 8.41% 189.800us 63.267us 3 test/test_profiler.py(72): test_source aten::sum 5.02% 113.340us 8.39% 189.475us 189.475us 1 test/test_profiler.py(64): ts_method_1 aten::add 4.58% 103.346us 6.33% 142.847us 142.847us 1 test/test_profiler.py(62): ts_method_1 aten::mul 4.05% 91.498us 9.62% 217.113us 217.113us 1 test/test_profiler.py(71): test_source aten::add 4.03% 90.880us 5.60% 126.405us 126.405us 1 test/test_profiler.py(58): ts_method_2 aten::empty 3.49% 78.735us 3.49% 78.735us 19.684us 4 test/test_profiler.py(72): test_source Reviewers: gh-metadata: pytorch pytorch 43898 gh/ilia-cher/40/head

[wip] Source code level attribution in profiler Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Output: https://gist.github.com/ilia-cher/bd13a3ab0c4905ce58b0079df66711fa Reviewers: gh-metadata: pytorch pytorch 43898 gh/ilia-cher/40/head

Source code level attribution in profiler Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Output: https://gist.github.com/ilia-cher/bd13a3ab0c4905ce58b0079df66711fa Reviewers: gh-metadata: pytorch pytorch 43898 gh/ilia-cher/40/head

ilia-cher · 2020-09-16T23:38:05Z

addressed most comments, will upd with a new round of benchmarking

dzhulgakov

These changes to TS look better, but @suo to double check

torch/autograd/profiler.py

torch/csrc/autograd/profiler.cpp

Source code level attribution in profiler Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Output: https://gist.github.com/ilia-cher/bd13a3ab0c4905ce58b0079df66711fa Reviewers: gh-metadata: pytorch pytorch 43898 gh/ilia-cher/40/head

ilia-cher · 2020-09-18T08:50:14Z

added profiler benchmark results - as expected when using with_stack=True it is slower, and the slow down is less when using TorchScript

Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Output: https://gist.github.com/ilia-cher/bd13a3ab0c4905ce58b0079df66711fa Benchmark: updated benchmark: profiler https://gist.github.com/ilia-cher/ff7064a311790590ffa8445c95d4d45b as expected, no change when no stacks are used, slower when stacks are used (in particular python is slower than TS callstack) overhead: https://gist.github.com/ilia-cher/8a2861863d3033b56181d23ba4cfbda1 overall as expected: small increase in cpu instructions (0.7%), small increase in tls accesses expected (bc of function.h change) Differential Revision: [D23432664](https://our.internmc.facebook.com/intern/diff/D23432664) [ghstack-poisoned]

ilia-cher · 2020-09-28T23:44:38Z

lint,
also cc. @ngimel @dzhulgakov @robieta @suo @zdevito @jamesr66a - ready for another look

Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Output: https://gist.github.com/ilia-cher/bd13a3ab0c4905ce58b0079df66711fa Benchmark: updated benchmark: profiler https://gist.github.com/ilia-cher/ff7064a311790590ffa8445c95d4d45b as expected, no change when no stacks are used, slower when stacks are used (in particular python is slower than TS callstack) overhead: https://gist.github.com/ilia-cher/8a2861863d3033b56181d23ba4cfbda1 overall as expected: small increase in cpu instructions (0.7%), small increase in tls accesses expected (bc of function.h change) Differential Revision: [D23432664](https://our.internmc.facebook.com/intern/diff/D23432664) [ghstack-poisoned]

dzhulgakov

Hm, your overhead measurements are actually show 6-8% slow perf with no profiling:
eager: 0.895ms -> 0.965ms
script: 0.585ms -> 0.625ms

That feels actually like a lot. Did you bisect where is it coming from? For eager mode there's even no interpreter changes - it must be from the RecordFunction itself then?

dzhulgakov · 2020-09-29T04:22:08Z

torch/csrc/jit/python/python_tracer.cpp

-  size_t source_line = 0;
-  std::stringstream stack_trace;
-
+std::vector<StackEntry> pythonCallstack() {


nit: it's a bit weird that it lives in python_tracer (kind of unrelated), maybe move to some more generic file? (not critical)

i'll check if there's a better location available, if not we'd prob. better not create a new one..

ngimel

I have small docs suggestions, other than that 6% overhead does seem like a lot, but it's not obvious where it's coming from.

torch/autograd/profiler.py

ilia-cher · 2020-09-29T08:35:32Z

Hm, your overhead measurements are actually show 6-8% slow perf with no profiling:
eager: 0.895ms -> 0.965ms
script: 0.585ms -> 0.625ms

That feels actually like a lot. Did you bisect where is it coming from? For eager mode there's even no interpreter changes - it must be from the RecordFunction itself then?

Re. runtimes - I included them mainly to show the rough difference between with stack / without stack profiling, and python vs TS callstacks, but for other purposes it is noisy, so I included valgrind based comparison for base vs trunk default mode.

Also reran benchmark (1000 times, no profiling, PR vs base, eager/script, different machine):

USE_MKLDNN=1 BLAS=MKL BUILD_BINARY=1 USE_CUDA=0 python setup.py develop install --cmake

pr_eager: (each line - 1000 runs of benchmark)
Overall: avg 0.811 ms, median: 0.773, min: 0.739, stddev: 0.136 ms
Overall: avg 0.774 ms, median: 0.768, min: 0.746, stddev: 0.030 ms
Overall: avg 0.837 ms, median: 0.742, min: 0.716, stddev: 0.226 ms
Overall: avg 0.870 ms, median: 0.736, min: 0.712, stddev: 0.266 ms
Overall: avg 0.786 ms, median: 0.779, min: 0.720, stddev: 0.031 ms

pr_script:
Overall: avg 0.649 ms, median: 0.552, min: 0.527, stddev: 0.156 ms
Overall: avg 0.546 ms, median: 0.540, min: 0.510, stddev: 0.025 ms
Overall: avg 0.575 ms, median: 0.537, min: 0.520, stddev: 0.108 ms
Overall: avg 0.542 ms, median: 0.535, min: 0.521, stddev: 0.028 ms
Overall: avg 0.585 ms, median: 0.543, min: 0.521, stddev: 0.110 ms

base_eager:
Overall: avg 0.764 ms, median: 0.751, min: 0.731, stddev: 0.042 ms
Overall: avg 0.808 ms, median: 0.782, min: 0.765, stddev: 0.115 ms
Overall: avg 0.884 ms, median: 0.844, min: 0.811, stddev: 0.151 ms
Overall: avg 0.802 ms, median: 0.791, min: 0.775, stddev: 0.033 ms
Overall: avg 0.815 ms, median: 0.806, min: 0.776, stddev: 0.031 ms

base_script:
Overall: avg 0.706 ms, median: 0.588, min: 0.571, stddev: 0.184 ms
Overall: avg 0.590 ms, median: 0.582, min: 0.571, stddev: 0.031 ms
Overall: avg 0.587 ms, median: 0.580, min: 0.570, stddev: 0.030 ms
Overall: avg 0.637 ms, median: 0.596, min: 0.583, stddev: 0.109 ms
Overall: avg 0.702 ms, median: 0.612, min: 0.584, stddev: 0.162 ms

https://gist.github.com/ilia-cher/a3974d77664628f0f42d27b616738013

So there's no stat. sig. information about runtime for base vs PR, though callgrind stats seem to give reproducible stats (https://gist.github.com/ilia-cher/8a2861863d3033b56181d23ba4cfbda1) e.g. they show increase in tls accesses (expected bc of if (profiler::profilerEnabled()) { thread_id_ = at::RecordFunction::currentThreadId(); } and some rel. small increase in cpu instructions.

Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Output: https://gist.github.com/ilia-cher/bd13a3ab0c4905ce58b0079df66711fa Benchmark: updated benchmark: profiler https://gist.github.com/ilia-cher/ff7064a311790590ffa8445c95d4d45b as expected, no change when no stacks are used, slower when stacks are used (in particular python is slower than TS callstack) overhead: https://gist.github.com/ilia-cher/8a2861863d3033b56181d23ba4cfbda1 overall as expected: small increase in cpu instructions (0.7%), small increase in tls accesses expected (bc of function.h change) Differential Revision: [D23432664](https://our.internmc.facebook.com/intern/diff/D23432664) [ghstack-poisoned]

ilia-cher · 2020-09-29T09:35:26Z

rest of comments + added support for @robieta Timer into profiler_bench.py

Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Output: https://gist.github.com/ilia-cher/bd13a3ab0c4905ce58b0079df66711fa Benchmark: updated benchmark: profiler https://gist.github.com/ilia-cher/ff7064a311790590ffa8445c95d4d45b as expected, no change when no stacks are used, slower when stacks are used (in particular python is slower than TS callstack) overhead: https://gist.github.com/ilia-cher/8a2861863d3033b56181d23ba4cfbda1 overall as expected: small increase in cpu instructions (0.7%), small increase in tls accesses expected (bc of function.h change) Differential Revision: [D23432664](https://our.internmc.facebook.com/intern/diff/D23432664) [ghstack-poisoned]

ilia-cher · 2020-09-29T10:06:42Z

example of benchmarking with --use_timer (also varies)

<torch.utils._benchmark.utils.common.Measurement object at 0x7f2cc961ee80>
payload()
  Median: 787.50 us
  IQR:    62.62 us (766.31 to 828.93)
  111 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7fd6fde9fe80>
payload()
  Median: 913.64 us
  IQR:    33.24 us (903.73 to 936.96)
  105 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7f721a1e0e80>
payload()
  Median: 768.92 us
  IQR:    22.46 us (759.44 to 781.90)
  127 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7f6d0490de80>
payload()
  Median: 821.17 us
  IQR:    12.48 us (815.65 to 828.13)
  122 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7f038ce2de80>
payload()
  Median: 762.76 us
  IQR:    38.57 us (752.31 to 790.88)
  122 measurements, 1000 runs per measurement, 1 thread

robieta · 2020-09-29T16:07:27Z

example of benchmarking with --use_timer (also varies)

<torch.utils._benchmark.utils.common.Measurement object at 0x7f2cc961ee80>
payload()
  Median: 787.50 us
  IQR:    62.62 us (766.31 to 828.93)
  111 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7fd6fde9fe80>
payload()
  Median: 913.64 us
  IQR:    33.24 us (903.73 to 936.96)
  105 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7f721a1e0e80>
payload()
  Median: 768.92 us
  IQR:    22.46 us (759.44 to 781.90)
  127 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7f6d0490de80>
payload()
  Median: 821.17 us
  IQR:    12.48 us (815.65 to 828.13)
  122 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7f038ce2de80>
payload()
  Median: 762.76 us
  IQR:    38.57 us (752.31 to 790.88)
  122 measurements, 1000 runs per measurement, 1 thread

And these are 100 second runs each? If so, that's a noisy system.

ilia-cher · 2020-09-29T17:09:46Z

example of benchmarking with --use_timer (also varies)

<torch.utils._benchmark.utils.common.Measurement object at 0x7f2cc961ee80>
payload()
  Median: 787.50 us
  IQR:    62.62 us (766.31 to 828.93)
  111 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7fd6fde9fe80>
payload()
  Median: 913.64 us
  IQR:    33.24 us (903.73 to 936.96)
  105 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7f721a1e0e80>
payload()
  Median: 768.92 us
  IQR:    22.46 us (759.44 to 781.90)
  127 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7f6d0490de80>
payload()
  Median: 821.17 us
  IQR:    12.48 us (815.65 to 828.13)
  122 measurements, 1000 runs per measurement, 1 thread
<torch.utils._benchmark.utils.common.Measurement object at 0x7f038ce2de80>
payload()
  Median: 762.76 us
  IQR:    38.57 us (752.31 to 790.88)
  122 measurements, 1000 runs per measurement, 1 thread

And these are 100 second runs each? If so, that's a noisy system.

yes, it is 100s

Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Output: https://gist.github.com/ilia-cher/bd13a3ab0c4905ce58b0079df66711fa Benchmark: updated benchmark: profiler https://gist.github.com/ilia-cher/ff7064a311790590ffa8445c95d4d45b as expected, no change when no stacks are used, slower when stacks are used (in particular python is slower than TS callstack) overhead: https://gist.github.com/ilia-cher/8a2861863d3033b56181d23ba4cfbda1 overall as expected: small increase in cpu instructions (0.7%), small increase in tls accesses expected (bc of function.h change) https://gist.github.com/ilia-cher/9c7901670fb48082af92386b1889e2fe Differential Revision: [D23432664](https://our.internmc.facebook.com/intern/diff/D23432664) [ghstack-poisoned]

robieta

Perf looks good; variation seems to just be noise. I'll defer to others about the rest of the changes.

Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Output: https://gist.github.com/ilia-cher/bd13a3ab0c4905ce58b0079df66711fa Benchmark: updated benchmark: profiler https://gist.github.com/ilia-cher/ff7064a311790590ffa8445c95d4d45b as expected, no change when no stacks are used, slower when stacks are used (in particular python is slower than TS callstack) overhead: https://gist.github.com/ilia-cher/8a2861863d3033b56181d23ba4cfbda1 overall as expected: small increase in cpu instructions (0.7%), small increase in tls accesses expected (bc of function.h change) https://gist.github.com/ilia-cher/9c7901670fb48082af92386b1889e2fe Differential Revision: [D23432664](https://our.internmc.facebook.com/intern/diff/D23432664) [ghstack-poisoned]

codecov · 2020-09-30T02:19:55Z

Codecov Report

❗ No coverage uploaded for pull request base (gh/ilia-cher/40/base@eea4a04). Click here to learn what that means.
The diff coverage is n/a.

@@                   Coverage Diff                   @@
##             gh/ilia-cher/40/base   #43898   +/-   ##
=======================================================
  Coverage                        ?   68.60%           
=======================================================
  Files                           ?      405           
  Lines                           ?    52021           
  Branches                        ?        0           
=======================================================
  Hits                            ?    35687           
  Misses                          ?    16334           
  Partials                        ?        0

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update eea4a04...196ba66. Read the comment docs.

Summary: Adding with_source parameter to enable tracking source code (filename and line) in profiler for eager, torchscript and autograd modes Test Plan: python test/test_profiler.py Output: https://gist.github.com/ilia-cher/bd13a3ab0c4905ce58b0079df66711fa Benchmark: updated benchmark: profiler https://gist.github.com/ilia-cher/ff7064a311790590ffa8445c95d4d45b as expected, no change when no stacks are used, slower when stacks are used (in particular python is slower than TS callstack) overhead: https://gist.github.com/ilia-cher/8a2861863d3033b56181d23ba4cfbda1 overall as expected: small increase in cpu instructions (0.7%), small increase in tls accesses expected (bc of function.h change) https://gist.github.com/ilia-cher/9c7901670fb48082af92386b1889e2fe Differential Revision: [D23432664](https://our.internmc.facebook.com/intern/diff/D23432664) [ghstack-poisoned]

facebook-github-bot · 2020-09-30T08:19:28Z

@ilia-cher merged this pull request in f5c95d5.

facebook-github-bot · 2020-09-30T08:19:36Z

@ilia-cher merged this pull request in f5c95d5.

ilia-cher requested review from albanD, apaszke, mrshenli, pritamdamania87 and zhaojuanmao as code owners August 31, 2020 19:50

facebook-github-bot added the oncall: jit Add this issue/PR to JIT oncall triage queue label Aug 31, 2020

ilia-cher requested review from dzhulgakov, jamesr66a, ngimel, robieta and suo August 31, 2020 21:46

ngimel reviewed Sep 2, 2020

View reviewed changes

test/test_profiler.py Show resolved Hide resolved

torch/csrc/autograd/profiler.cpp Outdated Show resolved Hide resolved

dzhulgakov reviewed Sep 11, 2020

View reviewed changes

ilia-cher added 4 commits September 15, 2020 17:12

ilia-cher changed the title ~~[wip] Source code level attribution in profiler~~ Source code level attribution in profiler Sep 16, 2020

ilia-cher added 2 commits September 16, 2020 02:35

dzhulgakov reviewed Sep 18, 2020

View reviewed changes

torch/autograd/profiler.py Show resolved Hide resolved

torch/csrc/autograd/profiler.cpp Outdated Show resolved Hide resolved

ilia-cher requested review from pietern and rohan-varma as code owners September 18, 2020 08:20

ilia-cher added 2 commits September 18, 2020 01:39

dzhulgakov reviewed Sep 29, 2020

View reviewed changes

ngimel reviewed Sep 29, 2020

View reviewed changes

torch/autograd/profiler.py Outdated Show resolved Hide resolved

torch/autograd/profiler.py Outdated Show resolved Hide resolved

torch/autograd/profiler.py Outdated Show resolved Hide resolved

ilia-cher mentioned this pull request Sep 29, 2020

Debug #45495

Closed

ilia-cher added 2 commits September 29, 2020 02:43

ilia-cher requested review from dzhulgakov and ngimel September 29, 2020 10:15

ilia-cher added 2 commits September 29, 2020 13:54

robieta reviewed Sep 30, 2020

View reviewed changes

facebook-github-bot closed this in f5c95d5 Sep 30, 2020

facebook-github-bot added the merged label Sep 30, 2020

facebook-github-bot deleted the gh/ilia-cher/40/head branch October 3, 2020 14:18

mruberry added the Merged label Oct 28, 2020

Source code level attribution in profiler #43898

Source code level attribution in profiler #43898

Uh oh!

Conversation

ilia-cher commented Aug 31, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dr-ci bot commented Aug 31, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

💊 CI failures summary and remediations

❄️ 1 failure tentatively classified as flaky

pytorch_macos_10_13_py3_test (1/1)

ci.pytorch.org: 1 failed

Uh oh!

ngimel left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

dzhulgakov left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

dzhulgakov Sep 11, 2020

Choose a reason for hiding this comment

Uh oh!

albanD Sep 11, 2020

Choose a reason for hiding this comment

Uh oh!

ilia-cher Sep 14, 2020

Choose a reason for hiding this comment

Uh oh!

albanD Sep 14, 2020

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ilia-cher commented Sep 16, 2020

Uh oh!

dzhulgakov left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

ilia-cher commented Sep 18, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ilia-cher commented Sep 28, 2020

Uh oh!

dzhulgakov left a comment

Choose a reason for hiding this comment

Uh oh!

dzhulgakov Sep 29, 2020

Choose a reason for hiding this comment

Uh oh!

ilia-cher Sep 29, 2020

Choose a reason for hiding this comment

Uh oh!

ngimel left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ilia-cher commented Sep 29, 2020

Uh oh!

ilia-cher commented Sep 29, 2020

Uh oh!

ilia-cher commented Sep 29, 2020

Uh oh!

robieta commented Sep 29, 2020

Uh oh!

ilia-cher commented Sep 29, 2020

Uh oh!

ilia-cher commented Aug 31, 2020 •

edited

Loading

dr-ci bot commented Aug 31, 2020 •

edited

Loading

ilia-cher commented Sep 18, 2020 •

edited

Loading

codecov bot commented Sep 30, 2020 •

edited

Loading