Skip to content

LATX: skip SHBR for translation blocks using YMM operands - #446

Closed
luzeng87 wants to merge 1 commit into
lat-opensource:masterfrom
luzeng87:shbr-skip-ymm
Closed

LATX: skip SHBR for translation blocks using YMM operands#446
luzeng87 wants to merge 1 commit into
lat-opensource:masterfrom
luzeng87:shbr-skip-ymm

Conversation

@luzeng87

@luzeng87 luzeng87 commented Sep 2, 2026

Copy link
Copy Markdown
Contributor

Summary

  • detect translation blocks containing YMM operands before SHBR analysis
  • keep all vector registers live for mixed-width blocks
  • retain existing SHBR behavior for XMM-only blocks

SHBR tracks low 128-bit XMM values. Applying that model to a block that also uses YMM operands can lose upper-half state at TU-internal entries. This change conservatively bypasses SHBR for such blocks.

Validation

  • clean LoongArch release build with GCC -O2 and LATX optimize O1
  • mixed XMM/YMM cold AOT execution: pass
  • five non-empty AOT files generated and stabilized
  • mixed XMM/YMM hot AOT execution: pass

Signed-off-by: Lu Zeng <luzeng87@gmail.com>
@luzeng87
luzeng87 marked this pull request as draft September 3, 2026 03:06
@luzeng87

luzeng87 commented Sep 3, 2026

Copy link
Copy Markdown
Contributor Author

Superseded by #437. The SHBR exclusion is now kept with the YMM upper-half state changes that require it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant