Add docs/benchmarks/baseline-gemmini-rocket.md skeleton for Phase 2 baseline #98

New issue

Closed

opened 2026-05-26 09:04:20 -03:00 by navigator · 2 comments

navigator commented

2026-05-26 09:04:20 -03:00

Owner

Goal

Create the baseline benchmark report skeleton at docs/benchmarks/baseline-gemmini-rocket.md, referenced by PLAN.md Section 7 as the landing spot for upstream-Gemmini-on-Rocket measurements (cycles, scratchpad utilization, branch predictor stats). The skeleton anchors Phase 2 work and gives the head-to-head report (report-v1.md) a stable baseline reference to point at.

Sections to include:

Overview (scope: baseline upstream CONFIG=GemminiRocketConfig only; pre-FluidPop changes)
Build environment (Chipyard hash from rtl/chipyard.hash — TBD until Week 4; Verilator version; toolchain commit)
Workloads (bare-metal Gemmini tests; ResNet-50 inference)
Metrics collected (cycles, scratchpad utilization, branch predictor hit/miss, IPC — placeholder tables)
Methodology notes (single-threaded sim, warm-up policy, reset state)
Caveats (sim-only; no power numbers; Sky130/IHP130 area numbers belong in synthesis report not here)
Reproduction recipe (placeholder: command lines TBD until Phase 2 lands)

Acceptance criteria

docs/benchmarks/baseline-gemmini-rocket.md exists with the sections above
Status: Draft skeleton + Owner: TBD header
Each section has TODO markers and 2-3 line intent description
Placeholder tables present with header rows but no fabricated data
Explicitly notes Chipyard hash is TBD until Section 6 Week 4 lands
References PLAN.md Sections 7, 6 (Week 4), and the head-to-head report deliverable
docs/benchmarks/README.md created or updated to list this baseline doc alongside report-v1.md

Plan refs

Section 7 (Phase 2 baseline measurements), Section 6 (Chipyard pin), Section 10 (deliverable #5 references this baseline)

Notes

Skeleton only — actual numbers land during Phase 2 (weeks 5-6). Do NOT fabricate cycle counts or any measurements. The reproduction recipe stays as a placeholder until the Chipyard install issue completes and Phase 2 measurement work begins.

## Goal Create the baseline benchmark report skeleton at `docs/benchmarks/baseline-gemmini-rocket.md`, referenced by PLAN.md Section 7 as the landing spot for upstream-Gemmini-on-Rocket measurements (cycles, scratchpad utilization, branch predictor stats). The skeleton anchors Phase 2 work and gives the head-to-head report (`report-v1.md`) a stable baseline reference to point at. Sections to include: - Overview (scope: baseline upstream `CONFIG=GemminiRocketConfig` only; pre-FluidPop changes) - Build environment (Chipyard hash from `rtl/chipyard.hash` — TBD until Week 4; Verilator version; toolchain commit) - Workloads (bare-metal Gemmini tests; ResNet-50 inference) - Metrics collected (cycles, scratchpad utilization, branch predictor hit/miss, IPC — placeholder tables) - Methodology notes (single-threaded sim, warm-up policy, reset state) - Caveats (sim-only; no power numbers; Sky130/IHP130 area numbers belong in synthesis report not here) - Reproduction recipe (placeholder: command lines TBD until Phase 2 lands) ## Acceptance criteria - [ ] `docs/benchmarks/baseline-gemmini-rocket.md` exists with the sections above - [ ] `Status: Draft skeleton` + `Owner: TBD` header - [ ] Each section has `TODO` markers and 2-3 line intent description - [ ] Placeholder tables present with header rows but no fabricated data - [ ] Explicitly notes Chipyard hash is TBD until Section 6 Week 4 lands - [ ] References PLAN.md Sections 7, 6 (Week 4), and the head-to-head report deliverable - [ ] `docs/benchmarks/README.md` created or updated to list this baseline doc alongside `report-v1.md` ## Plan refs Section 7 (Phase 2 baseline measurements), Section 6 (Chipyard pin), Section 10 (deliverable #5 references this baseline) ## Notes Skeleton only — actual numbers land during Phase 2 (weeks 5-6). Do NOT fabricate cycle counts or any measurements. The reproduction recipe stays as a placeholder until the Chipyard install issue completes and Phase 2 measurement work begins.