From: Florent Kermarrec Date: Sat, 13 Jul 2019 09:44:29 +0000 (+0200) Subject: soc/cores: add ECC (Error Correcting Code) X-Git-Tag: 24jan2021_ls180~1104 X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=ee8fec10ff1f76aa93504a210861d93d22c553b3;p=litex.git soc/cores: add ECC (Error Correcting Code) Hamming codes with additional parity (SECDED): - Single Error Correction - Double Error Detection --- diff --git a/litex/soc/cores/ecc.py b/litex/soc/cores/ecc.py new file mode 100644 index 00000000..0a63c086 --- /dev/null +++ b/litex/soc/cores/ecc.py @@ -0,0 +1,155 @@ +# This file is Copyright (c) 2018-2019 Florent Kermarrec +# License: BSD + +""" +Error Correcting Code + +Hamming codes with additional parity (SECDED): +- Single Error Correction +- Double Error Detection +""" + +from functools import reduce +from operator import xor + +from migen import * + + +def compute_m_n(k): + m = 1 + while (2**m < (m + k + 1)): + m = m + 1; + n = m + k + return m, n + + +def compute_syndrome_positions(m): + r = [] + i = 1 + while i <= m: + r.append(i) + i = i << 1 + return r + + +def compute_data_positions(m): + r = [] + e = compute_syndrome_positions(m) + for i in range(1, m + 1): + if not i in e: + r.append(i) + return r + + +def compute_cover_positions(m, p): + r = [] + i = p + while i <= m: + for j in range(min(p, m - i + 1)): + r.append(i + j) + i += 2*p + return r + + +class SECDED: + def place_data(self, data, codeword): + d_pos = compute_data_positions(len(codeword)) + for i, d in enumerate(d_pos): + self.comb += codeword[d-1].eq(data[i]) + + def extract_data(self, codeword, data): + d_pos = compute_data_positions(len(codeword)) + for i, d in enumerate(d_pos): + self.comb += data[i].eq(codeword[d-1]) + + def compute_syndrome(self, codeword, syndrome): + p_pos = compute_syndrome_positions(len(codeword)) + for i, p in enumerate(p_pos): + pn = Signal() + c_pos = compute_cover_positions(len(codeword), 2**i) + for c in c_pos: + new_pn = Signal() + self.comb += new_pn.eq(pn ^ codeword[c-1]) + pn = new_pn + self.comb += syndrome[i].eq(pn) + + def place_syndrome(self, syndrome, codeword): + p_pos = compute_syndrome_positions(len(codeword)) + for i, p in enumerate(p_pos): + self.comb += codeword[p-1].eq(syndrome[i]) + + def compute_parity(self, codeword, parity): + self.comb += parity.eq(reduce(xor, + [codeword[i] for i in range(len(codeword))])) + + +class ECCEncoder(SECDED, Module): + def __init__(self, k): + m, n = compute_m_n(k) + + self.i = i = Signal(k) + self.o = o = Signal(n + 1) + + # # # + + syndrome = Signal(m) + parity = Signal() + codeword_d = Signal(n) + codeword_d_p = Signal(n) + codeword = Signal(n + 1) + + # place data bits in codeword + self.place_data(i, codeword_d) + # compute and place syndrome bits + self.compute_syndrome(codeword_d, syndrome) + self.comb += codeword_d_p.eq(codeword_d) + self.place_syndrome(syndrome, codeword_d_p) + # compute parity + self.compute_parity(codeword_d_p, parity) + # output codeword + parity + self.comb += o.eq(Cat(parity, codeword_d_p)) + + +class ECCDecoder(SECDED, Module): + def __init__(self, k): + m, n = compute_m_n(k) + + self.enable = Signal() + self.i = i = Signal(n + 1) + self.o = o = Signal(k) + + self.sec = sec = Signal() + self.ded = ded = Signal() + + # # # + + syndrome = Signal(m) + parity = Signal() + codeword = Signal(n) + codeword_c = Signal(n) + + # input codeword + parity + self.compute_parity(i, parity) + self.comb += codeword.eq(i[1:]) + # compute_syndrome + self.compute_syndrome(codeword, syndrome) + self.comb += If(~self.enable, syndrome.eq(0)) + # locate/correct codeword error bit if any and flip it + cases = {} + cases["default"] = codeword_c.eq(codeword) + for i in range(1, 2**len(syndrome)): + cases[i] = codeword_c.eq(codeword ^ (1<<(i-1))) + self.comb += Case(syndrome, cases) + # extract data / status + self.extract_data(codeword_c, o) + self.comb += [ + If(syndrome != 0, + # double error detected + If(~parity, + ded.eq(1) + # single error corrected + ).Else( + sec.eq(1) + ) + ) + ] diff --git a/test/test_ecc.py b/test/test_ecc.py new file mode 100644 index 00000000..ca7929fb --- /dev/null +++ b/test/test_ecc.py @@ -0,0 +1,100 @@ +# This file is Copyright (c) 2018-2019 Florent Kermarrec +# License: BSD + +import unittest +import random + +from migen import * + +from litedram.common import * +from litedram.frontend.ecc import * + +from litex.gen.sim import * + + +class TestECC(unittest.TestCase): + def test_m_n(self): + m, n = compute_m_n(15) + self.assertEqual(m, 5) + self.assertEqual(n, 20) + + def test_syndrome_positions(self): + p_pos = compute_syndrome_positions(20) + p_pos_ref = [1, 2, 4, 8, 16] + self.assertEqual(p_pos, p_pos_ref) + + def test_data_positions(self): + d_pos = compute_data_positions(20) + d_pos_ref = [3, 5, 6, 7, 9, 10, 11, 12, 13, 14, 15, 17, 18, 19, 20] + self.assertEqual(d_pos, d_pos_ref) + + def test_cover_positions(self): + c_pos_ref = { + 0 : [1, 3, 5, 7, 9, 11, 13, 15, 17, 19], + 1 : [2, 3, 6, 7, 10, 11, 14, 15, 18, 19], + 2 : [4, 5, 6, 7, 12, 13, 14, 15, 20], + 3 : [8, 9, 10, 11, 12, 13, 14, 15], + 4 : [16, 17, 18, 19, 20] + } + for i in range(5): + c_pos = compute_cover_positions(20, 2**i) + self.assertEqual(c_pos, c_pos_ref[i]) + + def test_ecc(self, k=15): + class DUT(Module): + def __init__(self, k): + m, n = compute_m_n(k) + self.flip = Signal(n + 1) + + # # # + + self.submodules.encoder = ECCEncoder(k) + self.submodules.decoder = ECCDecoder(k) + + self.comb += self.decoder.i.eq(self.encoder.o ^ self.flip) + + def generator(dut, k, nvalues, nerrors): + dut.errors = 0 + prng = random.Random(42) + yield dut.decoder.enable.eq(1) + for i in range(nvalues): + data = prng.randrange(2**k-1) + yield dut.encoder.i.eq(data) + # FIXME: error when fliping parity bit + if nerrors == 1: + flip_bit1 = (prng.randrange(len(dut.flip)-2) + 1) + yield dut.flip.eq(1<