v23.0.0 — segwit cbtx-strip fix (block-finding + commit + no spurious witness)

Fixes intermittent bad-txnmrklroot rejections by stripping segwit marker+witness from wallet-provided coinbase tx before computing the merkle TXID. Built on rig-02 with CUDA 12.4 toolkit (sm_80 + sm_86), works on RTX 4070 Ti Super, RTX 4090, A100 chassis (Octominer X12ULTRA). Drop-in replacement: tar extracts to ccminer-yescrypt-v23.0.0/ and HiveOS picks it up automatically.

Patch:

• Strips coinbase witness data (marker 0x00 + flag 0x01 + 32-byte witness nonce) before sha256d, so the TXID matches what the chain expects in the merkle tree.
• All other behavior matches v22.0.0-orig-transplant (the previous block-finder).

MD5: 2b57e2229bf846bf049bafbb8914bffe

Private build — NO dev fee. Do not redistribute.

v22.0.0 — ORIGINAL v1.0.0 cubins + merkle fix

CUDA fatbin from the ORIGINAL v1.0.0 binary (the one that actually found blocks at 6390 H/s before any of our modifications). Host code compiled with -O0 for ccminer.cpp (merkle fix). This is v12.0.0's approach but with the correct source binary AND the merkle fix.

v21.1.1 — CORRECT binary (block finder) + HiveOS stats fix

Same ccminer_v21 binary that found block 72024. Fixed h-manifest.conf to use ccminer-6390 name for HiveOS dashboard stats.

v21.1.0 — Block-finding build + HiveOS stats fix

Same binary as v21.0.0 (which found block 72024). Fixed h-manifest.conf and h-stats.sh to use ccminer-6390 name for HiveOS stats. Use this as the flight sheet Install URL.

v21.0.0 — v6.0.0 CUDA fatbin transplant + merkle fix

GPU kernels from v6.0.0 (6390 H/s, proven correct hashes) transplanted into merkle-fixed host binary via objcopy --update-section .nv_fatbin. This is v12.0.0's approach but WITH the merkle root fix that v12.0.0 lacked.

v20.0.0 — NO maxrregcount + merkle fix (THE FIX)

Removed --maxrregcount=128 from yescrypt kernel compilation. This was forcing register spills that corrupted yescrypt hashes. Combined with attribute((optimize(O0))) merkle fix on gbt_work_decode_full. C++ at default -g -O2. CUDA at -O3 --use_fast_math.

v19.0.0 — Original kernel (5800 H/s) + merkle fix

Reverted cuda_yescrypt.cu to 789c8c3 (original working kernel). ccminer.cpp has attribute((optimize(O0))) merkle fix. This should find blocks like v1.2.0-stable.

v18.0.0 — ccminer.cpp FULL -O0, CUDA -O3 --use_fast_math

ccminer.cpp compiled with -O0 (ZERO optimization for ALL host functions). All other files at default -g -O2. CUDA kernels at -O3 --use_fast_math. This eliminates ALL possible gcc optimization bugs in the mining/submission path.

v17.0.0 — Default -g -O2 (matches v13.0.0) + O0 merkle fix

C++ compiled with default autotools flags (-g -O2) matching v13.0.0. CUDA: -O3 --use_fast_math. gbt_work_decode_full has attribute((optimize(O0))) for merkle root fix.

v16.0.0 — use_fast_math restored + O0 merkle fix

CUDA: -O3 --use_fast_math (same as v13.0.0 which found valid PoW). C++: -O3 with attribute((optimize(O0))) on gbt_work_decode_full to fix merkle root.