abrown opened PR #10949 from abrown:fix-10906 to bytecodealliance:main:
This fixes an issue introduced in #9045 where the XMM register used as an
i16x8destination forscalar_to_vectoris not properly zeroed prior to moving thei16there. This means that uninitialized bits from previous instructions are readable in the top lanes of the destination. The fix for this miscompilation is to zero the XMM register prior to moving thei16to the XMM.Fixes #10906.
abrown requested fitzgen for a review on PR #10949.
abrown requested wasmtime-compiler-reviewers for a review on PR #10949.
fitzgen submitted PR review:
Thanks!
fitzgen merged PR #10949.
Last updated: Dec 06 2025 at 06:05 UTC