Remove unnecessary NVF_API from scheduler/utils.h

[wujingyue]

No description provided.

[wujingyue]

!test

[tfogal]

Thanks!

Please do LD_PRELOAD such a library into thunder and run e.g. test_nvfuser.py there or something, to double-check this is fine.

[wujingyue]

!test

[wujingyue]

Please do LD_PRELOAD such a library into thunder and run e.g. test_nvfuser.py there or something, to double-check this is fine.

Do you mean the following? I failed to understand how LD_PRELOAD helps to verify.

$ LD_PRELOAD=/opt/pytorch/nvfuser/bin/libnvfuser.so pytest thunder/tests/test_nvfuser.py
================================================================================================ test session starts =================================================================================================
platform linux -- Python 3.12.3, pytest-8.1.1, pluggy-1.5.0
Test order randomisation NOT enabled. Enable with --random-order or --random-order-bucket=<bucket_type>
benchmark: 5.1.0 (defaults: timer=torch.utils.benchmark.utils.timer.timer disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=True warmup_iterations=100000)
rootdir: /opt/pytorch/lightning-thunder
configfile: pyproject.toml
plugins: xdist-3.6.1, timestamper-0.0.10, cov-6.0.0, timeout-2.3.1, random-order-1.1.1, hypothesis-6.122.3, mpi-0.6, benchmark-5.1.0, shard-0.1.2, typeguard-4.3.0
timeout: 900.0s
timeout method: signal
timeout func_only: False
collected 47 items
Running 47 items in this shard

thunder/tests/test_nvfuser.py ..............................ssssssssssssssss.                                                                                                                                  [100%]

=================================================================================== 31 passed, 16 skipped, 233 warnings in 11.22s ====================================================================================

[tfogal]

Please do LD_PRELOAD such a library into thunder and run e.g. test_nvfuser.py there or something, to double-check this is fine.

Do you mean the following? I failed to understand how LD_PRELOAD helps to verify.

$ LD_PRELOAD=/opt/pytorch/nvfuser/bin/libnvfuser.so pytest thunder/tests/test_nvfuser.py
...

I did, yeah that works.
While initially adding that support, there were some issues of hidden symbols that somehow were not tickled by our tests.

Side note, is there a way to run thunder's tests on an nvFuser patch like this via the !test comments? @xwang233 ? Low priority.

[wujingyue]

!test does run thunder tests on an nvFuser patch, e.g., nvfuser-ci/jit_thunder_tests_20_A100.

[xwang233]

We do run thunder tests on Nvfuser PR's. Do you mean to add an additional thunder test with LD_PRELOAD=libnvfuser.so ?

[tfogal]

We do run thunder tests on Nvfuser PR's. Do you mean to add an additional thunder test with LD_PRELOAD=libnvfuser.so ?

Probably not. One should only need the LD_PRELOAD hack if there are multiple nvFusers installed on a system.