Don't default to -fno-exceptions -fno-rtti

[swolchok]

We need exceptions/RTTI for non-core-only use cases (see #7736).
Sending first version of this without patching any specific targets to restore these flags because I want to see if anything actually breaks.

[swolchok]

Stack from ghstack (oldest at bottom):

• -> Don't default to -fno-exceptions -fno-rtti #7746

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/7746

• 📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit 137da0f with merge base 7bc06d1 ():
💚 Looks good so far! There are no failures yet. 💚

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

[swolchok]

phi-3-mini and emformer-transcribe failures are from trunk, so it looks likely that this doesn't break anything.

[swolchok]

Specific question for reviewers: which CI builds, if any, do we want/need to disable exceptions/rtti? I could go either way on the size test; for the case where we include non-portable ops we need to leave them enabled, but it's up to us what we want to do for the core-only case.

[mergennachin]

https://github.com/pytorch/executorch/actions/runs/12919450651/job/36029994766?pr=7746 -- interesting, it is still under the limit

[mergennachin]

@mcremon-meta - are there places in cadence code where we rely on these flags to be present? if so, you might need to change your config explicitly

[mergennachin]

@robell , @freddan80 - FYI, see #7736 for background and context

[swolchok]

https://github.com/pytorch/executorch/actions/runs/12919450651/job/36029994766?pr=7746 -- interesting, it is still under the limit

#7742 lowered it from 51768 to 47664, probably because it caused us to -DNDEBUG in release builds. Mildly surprised we are only up to 47672 after this diff; I wonder why these flags were so ineffective at reducing size.

[mcremon-meta]

@mcremon-meta - are there places in cadence code where we rely on these flags to be present? if so, you might need to change your config explicitly

cc @hsharma35 @zonglinpeng in case we need to adjust something

[swolchok]

perhaps add a no-except, no-rtti flavor as well.

patched the build script in the latest update.

Can you adjust the current CI test for size_test and size_test_all_ops to use --no-exceptions, --no-rtti flags.

done! here is the effect. before restoring -fno-exceptions -fno-rtti:

ExecuTorch with no ops binary size, unstripped:
-rwxr-xr-x  1 swolchok  staff  87192 Jan 27 12:12 cmake-out/test/size_test
ExecuTorch with portable ops binary size, unstripped:
-rwxr-xr-x  1 swolchok  staff  1373168 Jan 27 12:12 cmake-out/test/size_test_all_ops

stripped:
~/src/executorch> ls -al cmake-out/test/size_test
-rwxr-xr-x  1 swolchok  staff  73672 Jan 27 12:39 cmake-out/test/size_test
~/src/executorch> ls -al cmake-out/test/size_test_all_ops
-rwxr-xr-x  1 swolchok  staff  1172688 Jan 27 12:39 cmake-out/test/size_test_all_ops

after:

ExecuTorch with no ops binary size, unstripped:
-rwxr-xr-x  1 swolchok  staff  68584 Jan 27 12:10 cmake-out/test/size_test
ExecuTorch with portable ops binary size, unstripped:
-rwxr-xr-x  1 swolchok  staff  1370064 Jan 27 12:10 cmake-out/test/size_test_all_ops

stripped:
~/src/executorch> ls -al cmake-out/test/size_test
-rwxr-xr-x  1 swolchok  staff  56552 Jan 27 12:41 cmake-out/test/size_test
~/src/executorch> ls -al cmake-out/test/size_test_all_ops
-rwxr-xr-x  1 swolchok  staff  1172064 Jan 27 12:41 cmake-out/test/size_test_all_ops

I am surprised that core is affected so much more than core + portable ops.

[swolchok]

should probably also patch the cadence build scripts to set -fno-exceptions -fno-rtti so as to preserve the example being good (EDIT: done)