dangernoodle-io / TaipanMiner / 26553245725

90%

push

github

jae/mining sw filter byteorder cleanup (#440)

* fix(mining): correct SW-path target byte order + filter regression tests

pack_target_word0 packed target[28] as the high byte, but target is
little-endian (target[31]=MSB). the SW early-reject then compared raw
state[7] (= bswap of the true PoW MSB word) against that byte-reversed
target word — i.e. bswap(hash_msw) <= bswap(target_msw). byte-swapping
both sides is not monotonic, so the comparison was meaningless and the
filter passed nearly every nonce (same class of bug as the D0 kernel
filter fixed in #439).

this was invisible at the mine_nonce_range level because meets_target is
the final arbiter there — the early-reject is purely an optimization, so a
no-op filter still yields correct found/not-found. only correctness was
unaffected; the SW path (host tests + native builds) just did redundant
work. fix: pack the true MSB word (target[31]<<24|...|target[28]) and
compare bswap32(state[7]) against it.

add regression tests that catch this class directly:
- test_pack_target_word0_exact_byte_order: asserts exact true-MSB packing
- test_sw_hash_nonce_rejects_over_target: calls sw_hash_nonce directly so
  HASH_CHECK/HASH_MISS selectivity is observable (a no-op filter wrongly
  returns HASH_CHECK for an over-target hash). both fail on the old code.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* refactor(mining): remove dead two-pass AHB mining helpers

sha256_hw_mine_first / sha256_hw_mine_second were the old two-call S3
mining path, superseded by the inline zero-bswap pipeline (Phase 3
hw_hot_loop_kernel). neither has any caller (only their own decl +
def). mine_second additionally carried a latent byte-reversed target
compare (same bug class as the SW path), unreachable but misleading.
remove both and their stale header comments.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

---------

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

2549 of 3217 branches covered (79.24%)

Branch coverage included in aggregate %.

5 of 5 new or added lines in 1 file covered. (100.0%)

5158 of 5332 relevant lines covered (96.74%)

154885.46 hits per line