fallensigh/ciciec2026
Archived
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix samples ntt bugs & set PLL & optimize matmul

Browse Source
This commit is contained in:
FallenSigh
2026-07-15 10:36:38 +08:00
parent fc91511950
commit c4d36515cc
7 changed files with 274 additions and 120 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
rtl/ip/mlkem/sample_ntt/sample_ntt_sync_shared.v
View File

@@ -43,15 +43,16 @@
`include "defines.vh"
/* verilator lint_off UNUSED */
/* verilator lint_off UNUSEDPARAM */
(* use_dsp = "no" *)
module sample_ntt_sync_shared #(parameter K = 4) (
/* verilator lint_on UNUSED */
/* verilator lint_on UNUSEDPARAM */
input clk,
input rst_n,
input [255:0] rho_i,
/* verilator lint_off UNUSED */
/* verilator lint_off UNUSEDSIGNAL */
input [2:0] k_i,
/* verilator lint_on UNUSED */
/* verilator lint_on UNUSEDSIGNAL */
input [1:0] i_idx,
input [1:0] j_idx,
input valid_i,
@@ -67,9 +68,9 @@ module sample_ntt_sync_shared #(parameter K = 4) (
output kc_ready_i,
output [1599:0] kc_state_i,
output kc_valid_i,
/* verilator lint_off UNUSED */
/* verilator lint_off UNUSEDSIGNAL */
input kc_ready_o
/* verilator lint_on UNUSED */
/* verilator lint_on UNUSEDSIGNAL */
);
// ================================================================
@@ -95,12 +96,12 @@ module sample_ntt_sync_shared #(parameter K = 4) (
// ================================================================
// Registered inputs (captured on valid_i)
// ================================================================
/* verilator lint_off UNUSED */
/* verilator lint_off UNUSEDSIGNAL */
reg [255:0] rho_r;
reg [2:0] k_r;
reg [1:0] i_r;
reg [1:0] j_r;
/* verilator lint_on UNUSED */
/* verilator lint_on UNUSEDSIGNAL */
// ================================================================
// Coefficient counter (0..256). 9 bits to avoid overflow when
@@ -223,6 +224,10 @@ module sample_ntt_sync_shared #(parameter K = 4) (
// d2 rejected, no longer need coefficients, or d2 was just accepted out.
wire grp_done = (state_r == ST_SQUEEZE) && (sq_phase_r == 2'd2) &&
(!d2_acc_r || !need_more || (valid_o_r && ready_i));
// `need_more` observes the registered counter before the final d2
// handshake increments it. Detect that boundary explicitly so a full
// rate block does not launch an unnecessary final re-permutation.
wire completes_output = valid_o_r && ready_i && (coeff_cnt_r == 9'd255);
// ================================================================
// FSM: state_next (combinational)
@@ -248,7 +253,7 @@ module sample_ntt_sync_shared #(parameter K = 4) (
// next group in this block, re-permute (block exhausted), or
// finish.
if (grp_done) begin
if (!need_more)
if (!need_more || completes_output)
state_next = ST_DONE;
else if (grp_ptr_r < GRP_MAX)
state_next = ST_SQUEEZE; // next group, no re-permute