This repository has been archived on 2026-08-21. You can view files and clone it. You cannot open issues or pull requests or push a commit.

FallenSigh e30a65bd72 perf(mlkem): pipeline CBD->bank write to cut 100 MHz critical path ...

At 100 MHz the worst path (WNS -1.58 ns) was CBD sample: squeeze_buf[buf_ptr]
extraction -> CBD add/sub -> mod-Q -> bank_t/bank_se write, all combinational
in one cycle and routing-dominated (squeeze_buf and the banks place far apart).
Register the CBD write (data/addr/enable) so the bank write commits one cycle
later; the registered write takes priority in the bank_se/bank_t mux so the
1-cycle spillover into the next state uses the correct addr/data. No consumer
reads these coeffs until the following stage (>1 cycle later), so no RAW hazard.

Verified byte-exact vs NIST KAT: Encaps 9/9, Decaps 9/9, KeyGen 11/11,
hello_world e2e -- all PASS.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

2026-07-11 07:42:35 +08:00

..

common

feat: add ML-KEM accelerator with AXI slave wrapper

2026-06-30 01:39:35 +08:00

comp_decomp

perf(mlkem): stream compress feed (B1) — ~27% fewer Encaps cycles

2026-07-11 02:00:03 +08:00

ntt

perf(mlkem): register butterfly inputs to cut the 100 MHz NTT critical path

2026-07-11 07:05:12 +08:00

poly_arith

feat: add ML-KEM accelerator with AXI slave wrapper

2026-06-30 01:39:35 +08:00

poly_mul

perf(mlkem): move poly_mul coeff storage to even/odd banked BRAM

2026-07-11 05:41:27 +08:00

sample_cbd

fix: replace UNUSEDPARAM/UNUSEDSIGNAL with UNUSED for verilator lint

2026-06-30 02:10:07 +08:00

sample_ntt

mlkem test diagnostics and allocator updates

2026-07-08 02:35:56 +08:00

sha3

fix(mlkem): verilator lint fixes — BLKLOOPINIT and UNUSEDSIGNAL

2026-07-08 03:36:52 +08:00

storage

feat: add ML-KEM accelerator with AXI slave wrapper

2026-06-30 01:39:35 +08:00

top

perf(mlkem): pipeline CBD->bank write to cut 100 MHz critical path

2026-07-11 07:42:35 +08:00