Skip to content

feat(ccu): add four-rank alltoall mesh - #79

Draft
Kur0x wants to merge 2 commits into
codex/ccu-stack-07-alltoall-2rankfrom
codex/ccu-stack-08-alltoall-mesh
Draft

feat(ccu): add four-rank alltoall mesh#79
Kur0x wants to merge 2 commits into
codex/ccu-stack-07-alltoall-2rankfrom
codex/ccu-stack-08-alltoall-mesh

Conversation

@Kur0x

@Kur0x Kur0x commented Jul 28, 2026

Copy link
Copy Markdown
Member

Summary

  • add four-rank mesh mission construction and resource packaging
  • add per-peer endpoint planning, topology selection, and multi-rank smoke support
  • retain the two-rank long-mission path

Stack

8 of 9. Depends on codex/ccu-stack-07-alltoall-2rank.

Validation

  • git diff --check
  • focused mesh suite exercised 192 tests: 165 passed, 14 skipped, 1 outdated source assertion was corrected and rerun successfully
  • 12 runner dry-run cases require Bash 4 associative arrays and are unavailable with macOS Bash 3.2; they remain enabled for Linux CI

Scope

Prior four-rank hardware evidence belongs to #71 and was not rerun from this host.

@Kur0x

Kur0x commented Jul 28, 2026

Copy link
Copy Markdown
Member Author

CI note: Host Checks for this PR never reached checkout or repository tests. Attempt 1 failed in apt-get with Ubuntu mirror connection errors; attempts 2 and 3 remained in Install host dependencies for more than 20 minutes and were cancelled to release runner capacity. The cumulative successor #80, which contains all changes from this PR, passed Host Checks and PR Gate. Local cumulative validation also passes: 142 core tests, 19 microcode/signal/memory tests, workflow policy test, source guard, and git diff --check. Re-run this PR when the Ubuntu mirror is healthy.

}
program->push_back(instr);

if (AppendSetSourceCke(shared, TILEXR_CCU_ALLTOALL_SOURCE_CKE_INIT_MASK, program, report) != TILEXR_SUCCESS) {

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Only peers.front().route.sourceCke is initialized here, but the post phase later emits SyncCke using each peer route's own sourceCke. ValidateMeshSpec permits distinct nonzero source CKEs, so such a spec returns success while peers after the first reference uninitialized source CKEs and can stall. Please initialize every unique (sourceCke, mask) pair or require all routes to share the initialized pair.

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