You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Bench #478] Qualify performance on Linux x86-64 and macOS arm64 without cross-hardware false regressions #495
Current canonical evidence is environment-keyed, but optimization claims need to survive both Tier-1 Flow platforms. BLAS/LAPACK implementation, vector width, compiler and allocator differences can reverse small wins.
Run canonical operation-level and selected scaled benchmarks on Linux x86-64/OpenBLAS and macOS arm64/Accelerate with pinned threads. Maintain independent baselines per environment; never compare absolute timings across machines.
Acceptance: every canonical loss/win has per-platform status; regressions are evaluated only within matching environment IDs; platform-specific holes become explicit child issues; Pages can filter by platform; target #478 eventually has zero material losses on both supported performance platforms.
Activity