Skip to content

Merge upstream main into dev - #131

Merged
zcnrex merged 2 commits into
devfrom
chunan/sync-upstream-261102
Nov 2, 2026
Merged

zcnrex merged 2 commits into
devfrom
chunan/sync-upstream-261102

Conversation

@zcnrex

@zcnrex zcnrex commented Nov 2, 2026

Copy link
Copy Markdown
Collaborator

Bring dev up to upstream main (5081f85). This is a single two-parent merge onto dev faa20ac, in the same shape as #103. It merged without conflicts.

The upstream side is one commit, which touches only the mega_mhc JIT headers:

  • impls/sm100_mega_mhc.cuh
  • layout/mega_mhc.cuh
  • scheduler/mega_mhc.cuh

All three are byte-identical to upstream after the merge. There are no host-side, binding or API changes, so _C.so and sglang callers are unaffected.

Not GPU-validated on this branch.

🤖 Generated with Claude Code

lightseek-bot and others added 2 commits October 20, 2026 13:07
A CUDA graph kernel node keeps the same %gridid on every replay. The
mega_mhc split barriers tag each execution with %gridid, so after a replay
a barrier line still holds encode_state(grid_idx, kNumSplits), which is
exactly the state the next replay waits for. A Mix or Norm waiter whose
m-block init CTA has not started yet proceeds immediately and reads the
previous replay's split partials.

Tag executions with an epoch instead. The epoch and an exit count live in
the unused tail of the first barrier line, so the host-side buffer is
unchanged. Every CTA reads the epoch after cudaGridDependencySynchronize,
and the last CTA to retire resets the exit count and advances the epoch.

Signed-off-by: lightseek-bot <243258330+lightseek-bot@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@zcnrex
zcnrex merged commit 7ae89b8 into dev Nov 2, 2026
2 of 3 checks passed
@zcnrex
zcnrex deleted the chunan/sync-upstream-261102 branch November 2, 2026 19:30
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.

2 participants