 marfrit
|
8e6d9e6966
|
Iteration 5 close — A+G+B+E all GREEN
Heavyweight four-track iteration. All Phase 1 success criteria met:
- Track A (DEBUG sweep): ~339 lines of iter1/iter3/iter4 instrumentation
removed across 7 fork commits. Driver builds clean; per-frame log
noise zero (1 v4l2-request line per 2000-frame stress).
- Track G (PGO-disabled Firefox rebuild): firefox 150.0.1-1.1 built
on boltzmann (single-pass non-PGO, ~2h27m). 68.7 MB pkg, 169 MB
libxul (21× smaller than iter3 PGO-instrumented). 2.7× faster
decode through firefox-fourier sandbox.
- Track E (multi-context): LAST_OUTPUT_* moved from process-global
static to per-driver_data. Two concurrent mpv with 2s stagger
both decode clean.
- Track B (libplacebo segfault): 35s mpv --vo=gpu, 0 segfaults
(mpv falls through to GLES via Panfrost gracefully).
Phase 5 sonnet review came back YELLOW with 4 caveats; 3 resolved
in code (additional 107-line sweep, readback_warned removed),
1 documented as iter6+ candidate (cap_pool resolution-change race
latent under untested consumer probe patterns).
iter5-end driver sha256: 4bed52ec5d44b389. firefox-fourier 1.1
sha256: aa94c7290ee7be76. README iteration table updated.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
2026-05-05 17:39:35 +00:00 |
|