feat(tb): add Vivado XSIM Verilog testbenches for all 10 sync modules

Add file-based vector testbenches ( + ) for:
- mod_add_sync, rng_sync, poly_arith_sync, comp_decomp_sync
- s_bram/sd_bram, sha3_chain_top
- ntt_core, poly_mul_sync
- sample_cbd_sync, sample_ntt_sync

Each module includes:
- tb_<module>_xsim.v: Vivado XSIM testbench
- gen_vectors.py: Python vector generator (stdlib only)
- vectors/<module>_input.hex: test input vectors
- xsim_run.tcl: compile + elaborate + simulate script
This commit is contained in:
2026-06-25 20:48:38 +08:00
parent ae5f0ca048
commit d4c3fc86fc
42 changed files with 7745 additions and 0 deletions

View File

@@ -0,0 +1,241 @@
#!/usr/bin/env python3
"""gen_vectors.py - Generate NTT test vectors for Vivado XSIM testbench.
Generates input vectors for ntt_core module in hex format compatible
with $readmemh. Each line encodes {1-bit mode, 256x12-bit coefficients}
packed as a single 3073-bit hex number (769 hex chars).
Q = 3329, N = 256, primitive root zeta = 17.
NTT uses Cooley-Tukey DIT with bit-reversed zeta ROM.
INTT uses Gentleman-Sande DIF with zeta ROM in reverse.
Usage:
python3 gen_vectors.py
Output:
vectors/ntt_core_input.hex
"""
import sys
import os
import random
# ============================================================
# Constants
# ============================================================
Q = 3329 # ML-KEM modulus
N = 256 # NTT size
LAYERS = 7 # log2(N) - 1 for Cooley-Tukey
# zeta_rom values: bit-reversed powers of the primitive root zeta=17
# These match sync_rtl/ntt/zeta_rom.v exactly
ZETA_ROM = [
1, 1729, 2580, 3289, 2642, 630, 1897, 848,
1062, 1919, 193, 797, 2786, 3260, 569, 1746,
296, 2447, 1339, 1476, 3046, 56, 2240, 1333,
1426, 2094, 535, 2882, 2393, 2879, 1974, 821,
289, 331, 3253, 1756, 1197, 2304, 2277, 2055,
650, 1977, 2513, 632, 2865, 33, 1320, 1915,
2319, 1435, 807, 452, 1438, 2868, 1534, 2402,
2647, 2617, 1481, 648, 2474, 3110, 1227, 910,
17, 2761, 583, 2649, 1637, 723, 2288, 1100,
1409, 2662, 3281, 233, 756, 2156, 3015, 3050,
1703, 1651, 2789, 1789, 1847, 952, 1461, 2687,
939, 2308, 2437, 2388, 733, 2337, 268, 641,
1584, 2298, 2037, 3220, 375, 2549, 2090, 1645,
1063, 319, 2773, 757, 2099, 561, 2466, 2594,
2804, 1092, 403, 1026, 1143, 2150, 2775, 886,
1722, 1212, 1874, 1029, 2110, 2935, 885, 2154,
]
def barrett_mul(a: int, b: int) -> int:
"""Barrett modular multiplication: (a * b) mod Q."""
return (a * b) % Q
def ntt_forward(coeffs: list) -> list:
"""Forward NTT: Cooley-Tukey DIT, matches ntt_core RTL exactly.
Processes 7 layers. At each layer:
- Pairs are (j, j+layer_len)
- Zeta for each block comes from zeta_rom (index increments per block)
- Butterfly: t = zeta*b; a_out = a+t; b_out = a-t (all mod Q)
Input: normal-order coefficients (index 0..255)
Output: bit-reversed NTT result
"""
a = list(coeffs)
layer_len = 128
zeta_idx = 1 # Forward NTT starts at zeta_rom[1]
for layer in range(LAYERS):
for start in range(0, N, 2 * layer_len):
zeta = ZETA_ROM[zeta_idx]
for j in range(start, start + layer_len):
t = barrett_mul(zeta, a[j + layer_len])
a_j_plus_len = (a[j] - t) % Q
a[j] = (a[j] + t) % Q
a[j + layer_len] = a_j_plus_len
zeta_idx += 1
layer_len >>= 1
return a
def intt_inverse(coeffs: list) -> list:
"""Inverse NTT: Gentleman-Sande DIF, matches ntt_core RTL exactly.
Processes 7 layers in reverse order (len=2,4,8,...,128).
- Zeta for each block comes from zeta_rom (index decrements per block)
- Butterfly: a_out = a+b; diff = b-a; b_out = zeta*diff (all mod Q)
After all layers, output is scaled by N (no 1/2 factors in GS butterfly).
The RTL then scales output by 3303 in the OUTPUT state for mode=1.
"""
a = list(coeffs)
layer_len = 2
zeta_idx = 127 # Inverse NTT starts at zeta_rom[127]
for layer in range(LAYERS):
for start in range(0, N, 2 * layer_len):
zeta = ZETA_ROM[zeta_idx]
for j in range(start, start + layer_len):
a_sum = (a[j] + a[j + layer_len]) % Q
diff = (a[j + layer_len] - a[j]) % Q
a[j] = a_sum
a[j + layer_len] = barrett_mul(zeta, diff)
zeta_idx -= 1
layer_len <<= 1
# Apply output scaling (multiply by 3303) as the RTL does in mode=1
for i in range(N):
a[i] = barrett_mul(a[i], 3303)
return a
def coeffs_to_hex(coeffs: list) -> str:
"""Convert 256 12-bit coefficients to a 768-char hex string.
coeffs[0] is the MSB of the hex output, coeffs[255] is the LSB.
Each coefficient is 3 hex chars (12 bits).
"""
result = 0
for c in coeffs:
result = (result << 12) | (c & 0xFFF)
return f"{result:0768X}"
def write_vector(f, mode: int, coeffs: list, label: str):
"""Write a single test vector to the hex file.
Format: {mode_hex_digit}{768 hex chars for 256 coeffs}
Total: 769 hex chars per line.
mode=0 -> hex digit '0', mode=1 -> hex digit '1'.
"""
mode_hex = "0" if mode == 0 else "1"
coeffs_hex = coeffs_to_hex(coeffs)
line = mode_hex + coeffs_hex
f.write(f"// {label}\n")
f.write(line + "\n")
def hex_char_to_int(c: str) -> int:
"""Convert single hex char to integer."""
return int(c, 16)
def generate_vectors():
"""Generate test vectors for ntt_core."""
os.makedirs("vectors", exist_ok=True)
hex_path = os.path.join("vectors", "ntt_core_input.hex")
# All tests are listed here with labels
tests = []
# --- Test 0: Forward NTT on all zeros ---
zeros = [0] * N
expected = ntt_forward(zeros)
tests.append((0, zeros, expected, "FWD: all zeros"))
# --- Test 1: Forward NTT on impulse at index 0 ---
imp0 = [0] * N
imp0[0] = 1
expected = ntt_forward(imp0)
tests.append((0, imp0, expected, "FWD: impulse at [0]"))
# --- Test 2: Forward NTT on impulse at index 1 ---
imp1 = [0] * N
imp1[1] = 1
expected = ntt_forward(imp1)
tests.append((0, imp1, expected, "FWD: impulse at [1]"))
# --- Test 3: Forward NTT on ramp [0,1,2,...,255] ---
ramp = [i % Q for i in range(N)]
expected = ntt_forward(ramp)
tests.append((0, ramp, expected, "FWD: ramp 0..255"))
# --- Test 4: Forward NTT on all ones ---
ones = [1] * N
expected = ntt_forward(ones)
tests.append((0, ones, expected, "FWD: all ones"))
# --- Test 5: Inverse NTT on all zeros ---
expected = intt_inverse(zeros)
tests.append((1, zeros, expected, "INV: all zeros"))
# --- Test 6: Inverse NTT on impulse at index 0 ---
expected = intt_inverse(imp0)
tests.append((1, imp0, expected, "INV: impulse at [0]"))
# --- Test 7: Inverse NTT on NTT(impulse) → should recover impulse*256 ---
ntt_imp0 = ntt_forward(imp0)
expected = intt_inverse(ntt_imp0)
tests.append((1, ntt_imp0, expected, "INV(NTT(imp[0])) → imp[0]*256"))
# --- Test 8: Inverse NTT on NTT(ramp) → should recover ramp*256 ---
ntt_ramp = ntt_forward(ramp)
expected = intt_inverse(ntt_ramp)
tests.append((1, ntt_ramp, expected, "INV(NTT(ramp)) → ramp*256"))
# --- Tests 9-12: Forward NTT on random vectors ---
random.seed(0x5EED)
for i in range(4):
rand_vec = [random.randrange(0, Q) for _ in range(N)]
expected = ntt_forward(rand_vec)
tests.append((0, rand_vec, expected, f"FWD: random {i}"))
# --- Tests 13-16: Inverse NTT on random vectors ---
for i in range(4):
rand_vec = [random.randrange(0, Q) for _ in range(N)]
expected = intt_inverse(rand_vec)
tests.append((1, rand_vec, expected, f"INV: random {i}"))
# --- Tests 17-18: Inverse NTT on NTT(random) → roundtrip ---
for i in range(2):
rand_vec = [random.randrange(0, Q) for _ in range(N)]
ntt_vec = ntt_forward(rand_vec)
recovered = intt_inverse(ntt_vec)
tests.append((1, ntt_vec, recovered, f"INV(NTT(random {i})) → random*256"))
# Write input hex file
with open(hex_path, "w") as f:
f.write("// ntt_core test vectors\n")
f.write("// Format: {1 hex digit mode}{768 hex chars coeffs}\n")
f.write("// mode: 0=forward NTT, 1=inverse NTT\n")
f.write("// coeffs: 256 x 12-bit values, coeff[0] at MSB position\n")
f.write("\n")
for mode, coeffs, expected, label in tests:
write_vector(f, mode, coeffs, label)
print(f"Generated {len(tests)} test vectors → {hex_path}")
# Print expected results for reference (for manual verification)
print("\nExpected output summary (first 4 coeffs of each test):")
for mode, coeffs, expected, label in tests:
first4 = expected[:4]
print(f" {label:45s} → [{first4[0]:4d}, {first4[1]:4d}, {first4[2]:4d}, {first4[3]:4d}, ...]")
if __name__ == "__main__":
generate_vectors()