Skip to content

[codex] Lower grouped integer widening to unpack - #1250

Draft
Zhendong404 wants to merge 2 commits into
hw-native-sys:mainfrom
mouliangyu:codex/vmi-integer-unpack
Draft

[codex] Lower grouped integer widening to unpack#1250
Zhendong404 wants to merge 2 commits into
hw-native-sys:mainfrom
mouliangyu:codex/vmi-integer-unpack

Conversation

@Zhendong404

Copy link
Copy Markdown
Collaborator

Summary

  • Lower grouped integer widening for gs(2)/gs(4)/gs(8), lane_stride=1, to vsunpack/vzunpack.
  • Chain two unpack operations for i8 to i32.
  • Add VMI lit coverage for signed/unsigned group layouts.

Validation

  • cmake --build build -j8 --target pto-test-opt
  • Focused VMI/FileCheck regression checks passed.

Draft note: TileLang codegen probing is being validated separately; the current generated PTODSL still shows a direct BF16-to-F32 vmi.vcvt in the end-to-end per-token case.

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