dangernoodle-io / TaipanMiner / 25290666924

90%

push

github

perf(mining): TA-320f overlap SHA_TEXT writes into busy-wait windows (#317)

- pre-stage pass2 padding (TEXT[8], TEXT[15]) during pass1's busy-wait
- restore next pass1 padding (TEXT[4..8, 15]) during pass2's busy-wait
- prime pass1's persistent slots in sha256_hw_pipeline_prep so first nonce
  after a job swap finds them initialised
- add lockstep self-test: 8 nonces vs sw sha-256d reference catches
  silent regressions in the hot loop (TA-271/TA-322 bug class)
- add sha256_hw_profile_hotloop boot diagnostic: per-phase cycle counts
  so future changes show up in the boot log alongside the microbench
- tdongle-s3 measured: 266 -> 309.7 kH/s sustained (+16.4%), 48/48 shares
  accepted on hmpool with the new layout
- canaries (sha256_hw_overlap_canary) already proved overlap is silicon-safe

Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

707 of 852 branches covered (82.98%)

Branch coverage included in aggregate %.

1609 of 1656 relevant lines covered (97.16%)

249281.88 hits per line