bool inherited from number

[Ma-Jian1]

Fixes #125003

cc @EikanWang @jgong5 @wenzhe-nrv @sanchitintel @gujinghui @PenghuiCheng @XiaobingSuper @jianyuh @mingfeima @ashokei @jingxu10 @min-jean-cho @yanbing-j @Guobing-Chen @Xia-Weiwen @snadampal

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/125577

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

✅ You can merge normally! (1 Unrelated Failure)

As of commit 5512c87 with merge base ead97ee ():

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• trunk / macos-py3-arm64-mps / test (mps, 1, 1, macos-m1-13) (gh) (similar failure)
  'test/test_mps.py::TestMPS::test_adaptive_avg_pool2d_simple'

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[colesbury]

@EikanWang - would you please review this PR?

[EikanWang]

LGTM.

[EikanWang]

Please fix UT failures.

[Ma-Jian1]

@ezyang Could you help to take a review?

[ezyang]

You are still failing a bunch of tests

[Ma-Jian1]

hi all,
torch/optim/asgd.py is not part of this patch. what should I do next?
@EikanWang @ezyang

[ezyang]

@pytorchbot merge -r

[pytorchmergebot]

@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here

[pytorchmergebot]

Successfully rebased boolean-in-cpp-api onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout boolean-in-cpp-api && git pull --rebase)

[pytorchmergebot]

Merge failed

Reason: Approvers from one of the following sets are needed:

• superuser (pytorch/metamates)
• Core Reviewers (mruberry, lezcano, Skylion007, ngimel, peterbell10)
• Core Maintainers (soumith, gchanan, ezyang, dzhulgakov, malfet)

Details for Dev Infra team

Raised by workflow job

Failing merge rule: Core Maintainers

[ezyang]

I am not convinced I want to deal with the fallout from this in TorchScript. @davidberard98 for a second opinion

[facebook-github-bot]

@davidberard98 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[davidberard98]

I'm seeing internal failures - not sure if I'll be able to get to this for a few days, but I'll need to take a closer look to understand what's failing. I'll report back once I have a better understanding.

Actually, the failures are from torchvision - specifically from test_convert_image_dtype. I'm guessing this will repro in OSS but I haven't tried.

[Ma-Jian1]

hi @davidberard98 ，does it still failed on your test？

[davidberard98]

@Ma-Jian1 yes - do you think you can check using torchvision to better understand the failures?

[Ma-Jian1]

hi @davidberard98 ，I clone torchvision into local pytorch fold，and run python setup.py develop in torchvision subfolder, and then python -m pytest -sv test/test_transforms_tensor.py -k test_convert_image_dtype, I can reproduce 3 failed cases.
but after I reverted my patch, the failure still existed.
seems not related to my patch?

investigated further:
I found the max_int literal 2147483647.0 will be captured as 2147483648. in torch.script, so weird.

test/test_transforms_tensor.py::test_convert_image_dtype[out_dtype0-in_dtype0-cpu] graph(%image.1 : Tensor,
      %dtype.1 : int):
  %382 : int = prim::Constant[value=4294967294]()
  %377 : float = prim::Constant[value=9.2233720368547758e+18]()
  %376 : float = prim::Constant[value=2147483648.]()
  %333 : Function = prim::Constant[name="_max_value"]()
  %332 : str = prim::Constant[value="builtins.RuntimeError"]() # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:79:18
  %329 : str = prim::Constant[value="The cast from {} to {} cannot be performed safely."]() # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:78:18
  %321 : int = prim::Constant[value=7]() # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:76:27
  %319 : bool = prim::Constant[value=1]() # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:75:12
  %312 : int = prim::Constant[value=4]() # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:75:68
  %311 : int = prim::Constant[value=3]() # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:75:55
  %307 : int = prim::Constant[value=6]() # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:75:27
  %206 : bool = prim::Constant[value=0]()
  %205 : NoneType = prim::Constant()
  %12 : int = prim::Constant[value=0]() # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:71:24
  %64 : float = prim::Constant[value=0.001]() # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:86:14
  %72 : float = prim::Constant[value=1.]() # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:88:37
  %3 : int = prim::dtype(%image.1)
  %5 : bool = aten::eq(%3, %dtype.1) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:65:7
  %305 : Tensor = prim::If(%5) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:65:4
    block0():
      -> (%image.1)
    block1():
      %198 : bool = aten::is_floating_point(%image.1) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:68:7
      %201 : Tensor = prim::If(%198) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:68:4
        block0():
          %207 : Tensor = aten::tensor(%12, %dtype.1, %205, %206) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:71:11
          %208 : bool = aten::is_floating_point(%207) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:71:11
          %345 : Tensor = prim::If(%208) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:71:8
            block0():
              %215 : Tensor = aten::to(%image.1, %dtype.1, %206, %206, %205) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:72:19
              -> (%215)
            block1():
              %306 : int = prim::dtype(%image.1)
              %308 : bool = aten::eq(%306, %307) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:75:12
              %310 : bool = prim::If(%308) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:75:12
                block0():
                  %314 : int[] = prim::ListConstruct(%311, %312)
                  %315 : bool = aten::__contains__(%314, %dtype.1) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:75:45
                  -> (%315)
                block1():
                  -> (%206)
              %318 : bool = prim::If(%310) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:75:12
                block0():
                  -> (%319)
                block1():
                  %320 : int = prim::dtype(%image.1)
                  %322 : bool = aten::eq(%320, %321) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:76:12
                  %324 : bool = prim::If(%322) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:76:12
                    block0():
                      %326 : bool = aten::eq(%dtype.1, %312) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:76:45
                      -> (%326)
                    block1():
                      -> (%206)
                  -> (%324)
               = prim::If(%318) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:75:8
                block0():
                  %330 : int = prim::dtype(%image.1)
                  %msg.7 : str = aten::format(%329, %330, %dtype.1) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:78:18
                   = prim::RaiseException(%msg.7, %332) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:79:12
                  -> ()
                block1():
                  -> ()
              %334 : int = prim::CallFunction(%333, %dtype.1) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:87:24
              %max_val.7 : float = aten::Float(%334) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:87:18
              %337 : float = aten::add(%max_val.7, %72) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:88:27
              %338 : float = aten::sub(%337, %64) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:88:27
              %result.7 : Tensor = aten::mul(%image.1, %338) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:88:17
              %343 : Tensor = aten::to(%result.7, %dtype.1, %206, %206, %205) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:89:15
              -> (%343)
          -> (%345)
        block1():
           = prim::Print(%376) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:93:8
          %263 : Tensor = aten::tensor(%12, %dtype.1, %205, %206) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:97:11
          %264 : bool = aten::is_floating_point(%263) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:97:11
          %375 : Tensor = prim::If(%264) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:97:8
            block0():
              %image.65 : Tensor = aten::to(%image.1, %dtype.1, %206, %206, %205) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:98:20
              %273 : Tensor = aten::div(%image.65, %376) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:99:19
              -> (%273)
            block1():
               = prim::Print(%377) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:103:8
               = prim::Print(%377, %376, %382) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:116:12
              %image.91 : Tensor = aten::to(%image.1, %dtype.1, %206, %206, %205) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:117:20
              %373 : Tensor = aten::mul(%image.91, %382) # /home/jianma/repo/pytorch/vision/torchvision/transforms/_functional_tensor.py:118:19
              -> (%373)
          -> (%375)
      -> (%201)
  return (%305)

[Ma-Jian1]

I have to run "pip uninstall torch" several times to clean the virtual env.
now verified that the test passed without this patch.

[Ma-Jian1]

hi @davidberard98, the failed case in torchvision is related to cast int to float, which has an accuracy issue.
fixed it and add an extra unit test.

[Ma-Jian1]

hi @davidberard98 ,
could you help to take a review?
I have no mac device, and it shouldn't affect this unit test.