Module parallel 

Credit-bounded replication of the signing bottlenecks.

Two engines may finish on the same cycle. Dropping either completion would silently corrupt the signer, so each lane has a local 31-entry BRAM FIFO. Credits cover the entire accepted-to-retired interval, including FIFO time. Thus a lane cannot produce more results than its queue can hold. Retirement is round-robin, one result per cycle, with caller tags unchanged.

Structs

DualMultiComb

Two 24-context fixed-base engines, preserving the original request protocol. occupancy saturates at 31; it remains nonzero until queued results retire.

DualPointCompress

Two pair-inversion engines, allowing four point encodings to progress in parallel. occupancy saturates at 15, including buffered completions.

Merge2

Two independent BRAM completion queues and a fair one-port arbiter.

Sha512Pool8

Eight SHA lanes with one issue port and one retirement port.

Functions

dual_compress_kernel

Pair-compression dispatch and retirement.

dual_multicomb_kernel

Fixed-base dispatch and retirement.

merge_kernel

Completion merge transition. The producer must enforce 31 outstanding credits.

sha_pool8_kernel

Eight-lane single-issue SHA dispatch.