hpca-workflow

HPCA Workflow

Use this as the project-management skill for an HPCA submission. Replace every date

with the current official timetable and plan backward from the late-July gate.

HPCA is a conference, not a journal: it has no standing editor-in-chief and no

article-processing charge. It is run by the IEEE-CS Technical Committee on Computer

Architecture (TCCA); the rotating leadership is the per-edition General and Program

Chairs (HPCA 2027 program chairs were reported as Aamer Jaleel and Mattan Erez —

待核实), and the cost model is registration fees. Chairs rotate yearly, so re-check

the current organization page rather than carrying a name forward.

The shape of the HPCA year

HPCA sits in the winter seat: a late-July deadline feeds an autumn review

and a February/March conference. It is the last major architecture deadline of

the summer, so a project that misses it waits for MICRO's spring gate or ISCA's

November gate — plan the fallback before you need it.

Milestones

• Venue fit: confirm a measured microarchitectural cost meets a buildable

mechanism whose behavior is the contribution.

• Evidence lock: freeze the mechanism, the simulator/silicon configuration,

workloads, sampled regions, baselines, and the artifact mirror.

• Registration step: submit real title, abstract, authors, topics, and complete

conflicts about a week before the paper.

• Paper step: upload the anonymous IEEE-template PDF and all HotCRP fields.
• Review wait: triage reviews for correctness, methodology, novelty, and

reproducibility concerns as they land.

• Rebuttal/revision window: respond and revise in one combined window (recent

cycles run a single window, not two rounds).

• Notification → acceptance: archive reviews, the response, and the submitted

versions; then IEEE camera-ready, artifact evaluation on the AE HotCRP, and the

winter talk.

Backward plan from the paper deadline

| Weeks out (heuristic) | Hardware-evaluated milestone |

|---|---|

| 8+ | Mechanism implemented; fidelity contract (tool, config, workloads) fixed |

| 6 | Baseline sweeps complete; per-workload results, not just means |

| 4 | Real-silicon or validated-simulator numbers logged with machine state |

| 3 | Full draft in the IEEE two-column template |

| 2 | Internal mock review by an architecture-minded reader |

| 1 | Reference author-list check, anonymity sweep, artifact mirror assembled |

| 0 | Registration then paper on HotCRP |

These offsets are planning heuristics — anchor every one to the current official

timetable, never to a previous cycle.

Failure modes by stage

• Infrastructure still churning at week 4 forces last-minute configuration changes

no one has validated — the classic HPCA methodology reject in the making.

• Leaving template conversion to the final week surfaces overflow late; two-column

architecture figures are tighter than a working draft's.

• Skipping the mock review forfeits the only chance to hear "your baseline is

under-tuned" from a friend rather than a reviewer.

• Building the artifact mirror after the deadline under pressure is how anonymity

leaks and un-pinned simulators ship.

Coordination notes

• Assign one named owner for the fidelity contract and another for the anonymity

sweep; shared ownership is how both slip.

• Archive the exact submitted PDF, the artifact mirror, and the configuration files,

since rebuttal replies must quote them precisely.

Output format

[Current stage] idea / evaluation / writing / registration / paper / window / accepted
[Next official deadline] <date and source, or unknown>
[Critical path] <three tasks that determine readiness>
[Risk register] <page/anonymity/instrument/reproducibility/artifact/presentation>
[Fallback venue] <MICRO / ISCA / ASPLOS branch if rejected>
[Owner map] <task -> person or role>