Repository navigation

refactor(parallel)!: ♻️ replace the partition runtime with sharded OpenMP - #398

Open

robertodr wants to merge 67 commits into

mainfrom

refactor-parallelization

Open

refactor(parallel)!: ♻️ replace the partition runtime with sharded OpenMP#398
robertodr wants to merge 67 commits into
mainfrom
refactor-parallelization

Commits on Sep 23, 2026
- docs(parallelism): 📝 add rank-local OpenMP design and implementation plan
  robertodr
  committed
  35b33b8
- chore: spec and plan revision after grill-with-docs
  robertodr
  committed
  9bede89
Commits on Sep 24, 2026
- chore: spec and plan revision after rebase, add mini-app experiment
  robertodr
  committed
  35f359f
- chore: further refinement of spec and plan
  robertodr
  committed
  e88a436
Commits on Sep 25, 2026
- feat(index-miniapp): 🧪 add Task 0 MPI-off index mini-app
  Roberto Di Remigio Eikås
  committed
  7007658
- docs(parallelism): 📝 record Task 0 outcome and defer the ownership decision
  Roberto Di Remigio Eikås
  committed
  1ebd05f
Commits on Sep 26, 2026
- feat(bench): ✨ add rank-local OpenMP parity evidence tooling
  Roberto Di Remigio Eikås
  committed
  7962864
- docs(parallelism): 📝 freeze the Task 1 baseline campaign
  Roberto Di Remigio Eikås
  committed
  8fc2564
Commits on Sep 27, 2026
- docs(parallelism): 📝 record where Task 0 and Task 1 artifacts live
  Roberto Di Remigio Eikås
  committed
  af430da
- feat(parallel): ✨ add the OpenMP block worksharing primitive and require OpenMP
  Roberto Di Remigio Eikås
  committed
  ec59ad9
- docs(parallelism): 📝 record the Task 2 outcome
  Roberto Di Remigio Eikås
  committed
  d749787
Commits on Sep 28, 2026
- feat(parallel): ✨ plumb per-object thread budgets and distributed failure boundaries
  Roberto Di Remigio Eikås
  committed
  8cd7b62
- fix(bench): 🐛 read the process allocation before importing monoprop
  Roberto Di Remigio Eikås
  committed
  ef39488
- docs(parallelism): 📝 record the Task 3 outcome
  Roberto Di Remigio Eikås
  committed
  1f87c01
- docs(parallelism): 📝 record the baseline tool sync and add a Task 11 check
  Roberto Di Remigio Eikås
  committed
  e96af6e
- feat(parallel): ✨ thread cosine scaling and fused coefficient updates
  Roberto Di Remigio Eikås
  committed
  ea42df6
- docs(parallelism): 📝 record the Task 4 outcome and plan a std::fma evaluation
  Roberto Di Remigio Eikås
  committed
  a3631cb
- feat(parallel): ✨ thread the bitmap scan with a source-ordered merge
  Roberto Di Remigio Eikås
  committed
  29acf78
- docs(parallelism): 📝 record the Task 5 outcome
  Roberto Di Remigio Eikås
  committed
  24bb70e
Commits on Sep 29, 2026
- feat(parallel): ✨ thread frozen lookup with checked incoming decode
  Roberto Di Remigio Eikås
  committed
  43efb16
- docs(parallelism): 📝 record Task 6 and the decision to re-design with sharding
  Roberto Di Remigio Eikås
  committed
  a6a995a
- docs(parallelism): 📝 plan the sharded OpenMP replacement
  robertodr
  committed
  f31174c
- feat(parallel): ✨ add the fixed-team phase primitive for sharded OpenMP
  Roberto Di Remigio Eikås
  committed
  1cbcf2c
- docs(parallelism): 📝 record S0 acceptance and the notebook lint waiver
  Roberto Di Remigio Eikås
  committed
  4cdffac
Commits on Sep 30, 2026
- feat(parallel): ✨ add owner-initialized shard state for sharded OpenMP
  Roberto Di Remigio Eikås
  committed
  37d1ced
- docs(parallelism): 📝 record the S1 outcome in the sharded OpenMP plan
  Roberto Di Remigio Eikås
  committed
  a56daac
- feat(parallel): ✨ add direct-buffer sharded construction and propagation
  Roberto Di Remigio Eikås
  committed
  ffb3a36
- docs(parallelism): 📝 record the S2 outcome in the sharded OpenMP plan
  Roberto Di Remigio Eikås
  committed
  0c2503b
- feat(parallel): ✨ add snapshot-safe sharded evaluation
  Roberto Di Remigio Eikås
  committed
  910fde3
- docs(parallelism): 📝 record the S3 outcome in the sharded OpenMP plan
  Roberto Di Remigio Eikås
  committed
  f347e6f
Commits on Oct 1, 2026
- feat(parallel): ✨ integrate the sharded OpenMP prototype behind MonomialPropagator
  Roberto Di Remigio Eikås
  committed
  a790362
- docs(parallelism): 📝 record the S4 outcome in the sharded OpenMP plan
  Roberto Di Remigio Eikås
  committed
  7c4bc92
- style(parallel): 🎨 format the S4 prototype files and exempt the sharded probe from test naming
  Roberto Di Remigio Eikås
  committed
  b9dc2b3
- refactor(parallel): ⚡️ cut partner lookup and handoff overhead in the sharded prototype
  Roberto Di Remigio Eikås
  committed
  2f24091
- docs(parallelism): 📝 record the S4 checkpoint, the rework decision and the next step
  Roberto Di Remigio Eikås
  committed
  567da52
Commits on Oct 2, 2026
- refactor(parallel): ⚡️ single-barrier checkpoints, staged partner reads and exact record reservation
  Roberto Di Remigio Eikås
  committed
  554c3c8
- docs(parallelism): 📝 record the second S4 rework, its checkpoint and the proceed decision
  Roberto Di Remigio Eikås
  committed
  dc2e102
- feat(parallel): ✨ run the sharded OpenMP prototype over P MPI ranks
  Roberto Di Remigio Eikås
  committed
  1a22a8f
- docs(parallelism): 📝 record S5, its MPI+OpenMP checkpoint and the flagged optimizations
  Roberto Di Remigio Eikås
  committed
  ef103e9
Commits on Oct 3, 2026
- feat(parallel)!: ✨ final sharded OpenMP interface and test migration
  Roberto Di Remigio Eikås
  committed
  454830f
- docs(parallelism): 📝 record S6 and its acceptance
  Roberto Di Remigio Eikås
  committed
  c91eeae
Commits on Oct 4, 2026
- perf(parallel): ⚡️ lay replay rounds out owner-parallel
  Roberto Di Remigio Eikås
  committed
  bb71b9a
- docs(parallelism): 📝 record the post-S6 re-check, B-replay and B-memory
  Roberto Di Remigio Eikås
  committed
  2ab9b5e
Commits on Oct 7, 2026
- perf(parallel): ⚡️ cheaper exchange rounds, persistent staging and a T = 1 fast path
  Roberto Di Remigio Eikås
  committed
  e877419
- docs(parallelism): 📝 record S7 and the post-S7 remedy rounds
  Roberto Di Remigio Eikås
  committed
  b69152d
- docs(parallelism): 📝 note that the post-S7 remedy artifacts were deleted
  Roberto Di Remigio Eikås
  committed
  61ebbb1
- perf(parallel): ⚡️ stagger the final shrink, reuse paring buffers, 8-byte exchange offsets
  Roberto Di Remigio Eikås
  committed
  50c56a1
- docs(parallelism): 📝 record round A and the open parity points
  Roberto Di Remigio Eikås
  committed
  62173a5
- docs(parallelism): 📝 schedule the open parity optimizations at the end of S8
  Roberto Di Remigio Eikås
  committed
  42ee0e4
- docs(parallelism): 📝 record the owner's waiver of the S7-pass prerequisite for S8
  Roberto Di Remigio Eikås
  committed
  d023e67
Commits on Oct 8, 2026
- refactor(parallel)!: 🔥 remove the legacy partition runtime and require only MPI_THREAD_FUNNELED
  Roberto Di Remigio Eikås
  committed
  1fcbae0
- perf(sharded): ⚡️ close the remaining parity misses after the cutover
  Roberto Di Remigio Eikås
  committed
  d359c15
- docs(parallelism): 📝 record S8's optimization rounds and the campaign protocol
  Roberto Di Remigio Eikås
  committed
  747b562
Commits on Oct 9, 2026
- docs(parallelism): 📝 record S8's final parity campaign
  Roberto Di Remigio Eikås
  committed
  5ce60df
Commits on Oct 10, 2026
- perf(sharded): ⚡️ cut the per-process memory the sharded runtime adds at T > 1
  Roberto Di Remigio Eikås
  committed
  4fdddbc
- docs(parallelism): 📝 record S8's optimization round 5
  Roberto Di Remigio Eikås
  committed
  393f172
Commits on Oct 11, 2026
- chore(merge): 🔀 merge origin/main into the sharded OpenMP refactor
  Roberto Di Remigio Eikås
  committed
  94cbf92
- perf(sharded): ⚡️ keep main's fold off the per-thread memory and the small cosine calls
  Roberto Di Remigio Eikås
  committed
  1a4573d
Commits on Oct 12, 2026
- docs(parallelism): 📝 record the merge of main and step 6
  Roberto Di Remigio Eikås
  committed
  a28bc43
Commits on Oct 15, 2026
- chore(merge): 🔀 merge origin/main's test refactors into the sharded OpenMP refactor
  Roberto Di Remigio Eikås
  committed
  1eaebd4
- docs(parallelism): 📝 record the open optimization points and the items kept for later
  Roberto Di Remigio Eikås
  committed
  4e54b2e
Commits on Oct 16, 2026
- fix(tests): 🐛 build the stack-residency probe on macOS
  Roberto Di Remigio Eikås
  committed
  3cec097
- fix(build): 🐛 limit GCC's floating-point contraction to one expression
  Roberto Di Remigio Eikås
  committed
  7daa511
- fix(tests): 🐛 let the multi-rank probes oversubscribe small runners
  Roberto Di Remigio Eikås
  committed
  1c2aa8e
- test(ci): ✅ widen the TSan leg and close the review's CI gaps
  Roberto Di Remigio Eikås
  committed
  96756d4
- fix(sharded): 🐛 keep argument errors recoverable and tighten failure paths
  Roberto Di Remigio Eikås
  committed
  43dce9c
Commits on Oct 23, 2026
- chore(merge): 🔀 merge origin/main's per-term expectation values into the sharded OpenMP refactor
  Roberto Di Remigio Eikås
  committed
  c9cfa84

Commits

Commits on Sep 23, 2026

Commits on Sep 24, 2026

Commits on Sep 25, 2026

Commits on Sep 26, 2026

Commits on Sep 27, 2026

Commits on Sep 28, 2026

Commits on Sep 29, 2026

Commits on Sep 30, 2026

Commits on Oct 1, 2026

Commits on Oct 2, 2026

Commits on Oct 3, 2026

Commits on Oct 4, 2026

Commits on Oct 7, 2026

Commits on Oct 8, 2026

Commits on Oct 9, 2026

Commits on Oct 10, 2026

Commits on Oct 11, 2026

Commits on Oct 12, 2026

Commits on Oct 15, 2026

Commits on Oct 16, 2026

Commits on Oct 23, 2026