arm_neon.h: Enable additional native aliases

[mr-c]

Fixes: #26042

[sbc100]

Is there a test for this header that we can update ?

[mr-c]

@sbc100 @tlively There is a test case in #26042 (comment) that could be added, but I've never done that for this repo before so I might take me longer than either of you

[mr-c]

@sbc100 @tlively I adapted a test for vdotq_s32 from SIMDe

[sbc100]

lgtm, but I'll leave the final review to @tlively since I'm not really familiar with this function of the test code.

[mr-c]

@tlively @sbc100 Any advice about this error from the test I added?

/root/project/test/neon/test_neon_wasm_simd.cpp:354:5: error: loop not vectorized: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering [-Werror,-Wpass-failed=transform-warning]
  354 | int main() {
      |     ^

https://app.circleci.com/pipelines/github/emscripten-core/emscripten/48300/workflows/ce123da9-47f2-4dc0-a2a0-27ee22720ec3/jobs/1103937

[sbc100]

No idea I'm afraid. Maybe if you add -fno-inline-functions the error would at least be more precise?

[sbc100]

Maybe we could disable the auto-vectorizer for this change, along with a TODO linked to a new LLVM bug regarding error: loop not vectorized?

[mr-c]

Maybe we could disable the auto-vectorizer for this change, along with a TODO linked to a new LLVM bug regarding error: loop not vectorized?

A good plan, but I don't know how to do that; can you help?

[mr-c]

@sbc100 I fixed the test by turning off SIMDe level function inlining for this test only.

For submitting the llvm optimization bug, the following command can be used as a reproducer:

em++ test/neon/test_neon_wasm_simd.cpp -o test_neon_wasm_simd.js -sNO_DEFAULT_TO_CXX -Wclosure -Werror -Wno-limited-postlink-optimizations -O1 -msimd128 -fno-lax-vector-conversions -Wno-c++11-narrowing -mfpu=neon -msimd128

Fails for -O1, -O2, and -O3 at pass 734 transform-warning on main:

error: loop not vectorized: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering [-Werror,-Wpass-failed=transform-warning]

From our side, we run the entire SIMDe test suite (with complete coverage of the NEON intrinsics) using emscripten tot at -O2, and we have no issues.

[mr-c]

I think this is ready for merging @sbc100 @tlively ; the test failures are unrelated. Thanks 🙏 !

[sbc100]

Can you rebase one more them? I'd like the codesize tests at least to pass

[mr-c]

@sbc100 Done

1. 🔍 Codesize Checks almost passed (1 byte too few?)
2. ✔️ test-bun passed when before it had a transient failure before
3. ✔️ test-windows passes when it did not before.
4. ✔️ everything else passes

[sbc100]

Sadly we do want those codesize tests to pass before landing. We could update it in a followup but its better to update it as part of this PR.

Step to update:

$ emsdk install tot
$ ./test/runner codesize --rebase

(The above command should produce zero changes on main and the one-byte change on your branch. You can also try ./emcc --clear-cache between runs).

[sbc100]

Its the codesize stuff is not working as expected for you let me know and I can push those change to this PR (maybe on monday).

[mr-c]

@sbc100 I pushed the one byte change, but the Codesize Checks failed for main before it even tests the codesizes for this PR: https://github.com/emscripten-core/emscripten/actions/runs/21330588322/job/61394711670?pr=26068#step:5:867 (same failure from before: https://github.com/emscripten-core/emscripten/actions/runs/21318802004/job/61365560492#step:5:867 )

I can open a PR with just the Codesize fixes, but the CI test will still fail as the target branch (main) is out of date.

Since I don't have permission to dispatch gh workflow run rebaseline-tests.yml, perhaps you would do so? Or you can force-merge this PR :-D