Skip to content

fix the bug of global barrier and add the tests that can reproduce the previous bug. - #412

Open
symmetryyyyy wants to merge 4 commits into
masterfrom
review/gbar-response-phase-20260903
Open

fix the bug of global barrier and add the tests that can reproduce the previous bug.#412
symmetryyyyy wants to merge 4 commits into
masterfrom
review/gbar-response-phase-20260903

Conversation

@symmetryyyyy

Copy link
Copy Markdown
Collaborator

1 Add a regression test to reproduce the previous bug: introduce gbar_phase, a two-core SimX/RTLSim test that reproduces the global-barrier response-phase bug and verifies phase advancement across completed generations.
2 Enhance asynchronous stress coverage: extend the split arrive/wait loop in async_gbarrier from 32 to 1024 iterations, making delayed responses and phase reuse observable.
3 Restructure global-barrier response handling: replace the single pending-request state with per-barrier pending/phase/waiter bookkeeping and a request FIFO, so responses are matched by barrier ID and routed to the correct phase and waiting warps.

symmetryyyyy and others added 4 commits September 2, 2026 16:09
Add non-deadlocking diagnostics for cross-slot local phase corruption and missing global-barrier phase advancement. Register both tests in the aggregate regression suite and explicit SimX/xrtsim smoke coverage.

Co-Authored-By: RunjiaChen <runjia@u.nus.edu>

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Capture the phase and exact waiters for each pending global-barrier generation, queue completed local arrivals, and route each cluster response directly to its barrier slot. Keep response-only writes out of the barrier state RAM and qualify the working phase bypass by address.

Use one-bit phase toggles in SimX so wait comparisons keep matching the RTL after wraparound.
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