Always warn on cpu fallback

[dvrogozh]

Fixes: #262, pytorch/pytorch#126488

5bf9e0c ("Register operator's implementation lazily") disabled warnings printout on explicit CPU fallback. I believe that users and customers will benefit from these warnings in all the cases. Note that "explicit fallback" seems to be some internal intel classification term for supported/unsupported operations unlikely known to others. Thus, non-intel users will likely care for cpu fallback in general regardless of its type. This PR adds warning back for all CPU fallback cases. We did discuss in pytorch/pytorch#126488 that maybe printout in Release build might not be needed - I am thinking otherwise and added printout for all the build modes. Let's discuss this in the PR review.

CC: @gujinghui @EikanWang @fengyuan14 @guangyey @jgong5

[EikanWang]

LGTM

[EikanWang]

@dvrogozh , do you have the privilege to add a reviewer?

[dvrogozh]

@dvrogozh , do you have the privilege to add a reviewer?

No, unfortunately I do not.

[dvrogozh]

@EikanWang @fengyuan14. Some of pull / preci-ut tests fail. Are these some known failures since I can hardly attribute them to the change in this PR?

[fengyuan14]

@EikanWang @fengyuan14. Some of pull / preci-ut tests fail. Are these some known failures since I can hardly attribute them to the change in this PR?

Retrigger the preci. Skipped cases failed due to PyTorch daily update.

[dvrogozh]

@fengyuan14 : tests are still failing. Should this be retriggered again?

[fengyuan14]

@fengyuan14 : tests are still failing. Should this be retriggered again?
Two new errors never encounter. Let me try retrigger

FAILED test_autograd_xpu.py::TestAutogradDeviceTypeXPU::test_copy_r_to_c_xpu - AssertionError: False is not true : 
FAILED test_autograd_xpu.py::TestAutogradDeviceTypeXPU::test_to_r_to_c_xpu - AssertionError: False is not true : 

[fengyuan14]

I checked the failure again. The failures should be caused by the change in the PR.

The case says there should be no warning, but we warn in CPU fallback code path.

[dvrogozh]

@fengyuan14 : Ok, so there are some tests in pytorch which don't allow warnings. However, the tests which fail actually run few basic ops - in a way someone might consider that these ops should not fail and should run on appropriate device. I.e. actually someone might consider that these tests are actually broken and incorrectly report status.

There are few ways to proceed with this PR:

1. Actually fix the tests. This means identify and implement missing xpu ops.
2. Silence warning conditional to something like release build or environment variable disabled by default.

I need your opinions on this. From my side, I will check tomorrow which ops are not implemented. As of now I checked only 1 test:

• TestAutogradDeviceTypeXPU.test_copy_r_to_c_xpu - need aten::all.all_out

[fengyuan14]

@fengyuan14 : Ok, so there are some tests in pytorch which don't allow warnings. However, the tests which fail actually run few basic ops - in a way someone might consider that these ops should not fail and should run on appropriate device. I.e. actually someone might consider that these tests are actually broken and incorrectly report status.

There are few ways to proceed with this PR:

1. Actually fix the tests. This means identify and implement missing xpu ops.
2. Silence warning conditional to something like release build or environment variable disabled by default.

I need your opinions on this. From my side, I will check tomorrow which ops are not implemented. As of now I checked only 1 test:

• TestAutogradDeviceTypeXPU.test_copy_r_to_c_xpu - need aten::all.all_out

I prefer the first option. Looks there won't be too many efforts to take. And aten::all.all_out is in our development plan. We won't take extra efforts. We can prioritize the op.

[fengyuan14]

#368

[dvrogozh]

From my side, I will check tomorrow which ops are not implemented.

As far as I can tell from the log, we have 2 tests failing, both requiring same op:

• TestAutogradDeviceTypeXPU.test_copy_r_to_c_xpu - need aten::all.all_out
• TestAutogradDeviceTypeXPU.test_to_r_to_c_xpu - need aten::all.all_out