ttnn.signbit operation failed with low PCC [Grayskull and Wormhole]

[banekg]

Describe the bug ttnn.signbit operation breaks with "low PCC value" error in some test cases.

To Reproduce Steps to reproduce the behavior:

1. Checkout barsic/ttnn-param-fix-skipped branch. To be merged into main soon.
2. Run unit test test_signbit.py by using this command: pytest tests/ttnn/python_api_testing/non_working_unit_tests/grayskull/test_signbit.py

Expected behavior There is a test case presented in the unit test tests/ttnn/python_api_testing/non_working_unit_tests/grayskull/test_signbit.py and it is expected to fail with Equal check failed error.

Getting Additional info for the operation under test and its behavior To get additional information and results for different combinations of input shapes, types, layouts and memory configs for which this operation was tested you can also run locally sweeps for ttnn.std and check the results. To do this you should:

1. Follow the Getting Started page to setup the repo, environment variables and python-env
2. Activate source build/python_env/bin/activate
3. Run sweeps by using python tests/tt_eager/python_api_testing/sweep_tests/run_pytorch_test.py -i tests/ttnn/python_api_testing/sweep_tests/test_configs/ci_sweep_tests_broken/grayskull/ttnn_eltwise_signbit_test.yaml -o ./result-sweeps
4. After the run is completed all test sweeps results should be available inside specified output directory (in this case ./result-sweeps). There you will find signbit_sweep.csv which holds all executed sweeps, among which you can also find the ones that failed and were recreated by the unit test, which you can get by searching unique data_seed field.

[umadevimcw]

@jliangTT This is already discussed and blocked as we are not able to store the -0.0

[bharane-ab]

@jliangTT This is already discussed and blocked as we are not able to store the -0.0

As mentioned above we are not able to store-0.0 so signbit is fixed for positive 0.0 in PR link

[jliangTT]

Yes, spun up a high BW discussion with the leads here: https://tenstorrent.slack.com/archives/C05GH08C58F/p1714406021866519 (internal)

Let's continue to mark those as blocked in the spreadsheet and focus on other stuff. thanks!

[eyonland]

Please close this once the tests have been updated. See #8944

[VirdhatchaniKN]

Hi @ttmtrajkovic , @tt-aho , @banekg , @rtawfik01 We have a bug issue related to signbit and we have a few questions here :

• Negative Zero Issue #1939 :
  • We are not able to store -0.0 . So signbit was fixed for positive 0.0 in this PR. And the later we had a discussion under issue #8944. We would like to get your inputs on how we can test for -0.0 case ( #8944-comment )
  • Testing Branch : virdhtachani/signbit_test (#13947)
  • Test file : tests/tt_eager/python_api_testing/sweep_tests/pytests/tt_dnn/test_signbit_op.py
  • Unit Test command : pytest tests/tt_eager/python_api_testing/sweep_tests/pytests/tt_dnn/test_signbit_op.py::TestSignbit::test_run_signbit_op - Input passed as -0.0 gets passed into kernel as 0.0 value.
• Incorrect result Issue #7749 :
  • I ran the test file mentioned in this issue and made some modification to see what the problem here is. Here, I found that the signbit op fails for bfloat8_b data type of range (-1.0 , 0.0 ) due to precision error ,
  • Eg:
  • Input -0.0034approximates to -0.0
  • Passing -0.0 as bfloat8_b would treat the input as 0.0 and hence the result is incorrect
  • I've printed all the values causing assertion error in the below test file :
  • Testing Branch : virdhtachani/signbit_test (#13947)
  • Test file : tests/ttnn/python_api_testing/non_working_unit_tests/grayskull/test_eltwise_signbit_print_error.py

Would like your inputs on this to proceed further Thanks

[VirdhatchaniKN]

After discussions with @ttmtrajkovic , Issue Assigned to LLK Team https://github.com/tenstorrent/tt-metal/issues/13953

Moving this issue to On-Hold status