aicodix___code/ldpc_decoder.hh
2019-11-16 10:59:19 +01:00

281 lines
6.3 KiB
C++

/*
LDPC SISO layered decoder
Copyright 2018 Ahmet Inan <inan@aicodix.de>
*/
#ifndef LDPC_DECODER_HH
#define LDPC_DECODER_HH
#include <algorithm>
#include "simd.hh"
#include "rotate.hh"
namespace CODE {
template <typename TABLE, int BETA>
class LDPCDecoder
{
#ifdef __AVX2__
static const int SIMD_SIZE = 32;
// M = 360 = 30 * 12
static const int WORD_SIZE = 30;
#else
static const int SIMD_SIZE = 16;
// M = 360 = 15 * 24
static const int WORD_SIZE = 15;
#endif
static_assert(TABLE::M % WORD_SIZE == 0, "M must be multiple of word size");
static_assert(WORD_SIZE <= SIMD_SIZE, "SIMD size must be bigger or equal word size");
static const int M = TABLE::M;
static const int N = TABLE::N;
static const int K = TABLE::K;
static const int R = N-K;
static const int q = R/M;
static const int D = WORD_SIZE;
static const int W = M/D;
static const int PTY = R/D;
static const int MSG = K/D;
static const int VAR = N/D;
static const int CNC = TABLE::LINKS_MAX_CN - 2;
static const int BNL = (TABLE::LINKS_TOTAL + 1) / D;
typedef SIMD<int8_t, SIMD_SIZE> TYPE;
typedef struct { uint16_t off; uint16_t shi; } Loc;
Rotate<TYPE, D> rotate;
TYPE bnl[BNL];
TYPE var[VAR];
Loc loc[BNL];
bool wds[BNL];
int16_t csh[VAR];
uint8_t cnt[PTY];
bool start;
static TYPE eor(TYPE a, TYPE b)
{
return vreinterpret<TYPE>(veor(vmask(a), vmask(b)));
}
static TYPE orr(TYPE a, TYPE b)
{
return vreinterpret<TYPE>(vorr(vmask(a), vmask(b)));
}
static TYPE other(TYPE a, TYPE b, TYPE c)
{
return vreinterpret<TYPE>(vbsl(vceq(a, b), vmask(c), vmask(b)));
}
static TYPE mine(TYPE a, TYPE b)
{
return orr(eor(a, b), vdup<TYPE>(127));
}
static TYPE selfcorr(TYPE a, TYPE b)
{
return vreinterpret<TYPE>(vand(vmask(b), vorr(vceqz(a), veor(vcgtz(a), vcltz(b)))));
}
bool update()
{
TYPE *bl = bnl;
Loc *lo = loc;
bool *wd = wds;
auto bad = vmask(vzero<TYPE>());
for (int i = 0; i < PTY; ++i) {
int deg = cnt[i];
TYPE mags[deg], inps[deg];
TYPE min0 = vdup<TYPE>(127);
TYPE min1 = vdup<TYPE>(127);
TYPE signs = vdup<TYPE>(127);
TYPE cnv = vdup<TYPE>(127);
bool first_wd;
int last_offset = 0;
int8_t prev_val = 0;
for (int k = 0; k < deg; ++k) {
int offset = lo[k].off;
int shift = lo[k].shi;
int dshift = (shift - csh[offset]) % D;
TYPE tmp = rotate(var[offset], dshift);
if (offset == VAR-1 && shift == 1) {
prev_val = tmp.v[0];
tmp.v[0] = 127;
}
TYPE inp = vqsub(tmp, bl[k]);
if (start)
inp = tmp;
TYPE mag = vqabs(inp);
if (BETA) {
auto beta = vunsigned(vdup<TYPE>(BETA));
mag = vsigned(vqsub(vunsigned(mag), beta));
}
min1 = vmin(min1, vmax(min0, mag));
min0 = vmin(min0, mag);
signs = eor(signs, inp);
inps[k] = inp;
mags[k] = mag;
}
for (int k = 0; k < deg; ++k) {
TYPE mag = mags[k];
TYPE inp = inps[k];
TYPE out = vsign(other(mag, min0, min1), mine(signs, inp));
out = vclamp(out, -32, 31);
if (!start)
out = selfcorr(bl[k], out);
TYPE tmp = vqadd(inp, out);
cnv = vsign(cnv, tmp);
int offset = lo[k].off;
int shift = lo[k].shi;
if (offset == VAR-1 && shift == 1)
tmp.v[0] = prev_val;
bool this_wd = wd[k];
if (start) {
if (k)
this_wd = offset == last_offset;
else
this_wd = false;
}
if (!this_wd) {
bl[k] = out;
var[offset] = tmp;
csh[offset] = shift;
} else if (start) {
bl[k] = vzero<TYPE>();
}
if (k) {
bool next_wd = this_wd;
if (last_offset != offset) {
next_wd = first_wd;
first_wd = this_wd;
}
wd[k-1] = next_wd;
} else {
first_wd = this_wd;
}
last_offset = offset;
}
wd[deg-1] = first_wd;
bad = vorr(bad, vclez(cnv));
lo += deg;
bl += deg;
wd += deg;
}
//assert(bl <= bnl + BNL);
//std::cerr << BNL - (bl - bnl) << std::endl;
for (int n = 0; n < D; ++n)
if (bad.v[n])
return true;
return false;
}
public:
LDPCDecoder()
{
uint16_t pos[q * CNC];
uint8_t cnc[q];
for (int i = 0; i < q; ++i)
cnc[i] = 0;
int bit_pos = 0;
const int *row_ptr = TABLE::POS;
for (int g = 0; TABLE::LEN[g]; ++g) {
int bit_deg = TABLE::DEG[g];
for (int r = 0; r < TABLE::LEN[g]; ++r) {
for (int d = 0; d < bit_deg; ++d) {
int n = row_ptr[d] % q;
int m = row_ptr[d] / q;
pos[CNC*n+cnc[n]++] = bit_pos + (M - m) % M;
}
row_ptr += bit_deg;
bit_pos += M;
}
}
for (int i = 0; i < q; ++i)
for (int j = 0; j < W; ++j)
cnt[W*i+j] = cnc[i] + 2;
Loc *lo = loc;
for (int i = 0; i < q; ++i) {
int cnt = cnc[i];
int deg = cnt + 2;
int offset[cnt], shift[cnt];
for (int c = 0; c < cnt; ++c) {
shift[c] = pos[CNC*i+c] % M;
offset[c] = pos[CNC*i+c] - shift[c];
}
for (int j = 0; j < W; ++j) {
for (int c = 0; c < cnt; ++c) {
lo[c].off = offset[c] / D + shift[c] % W;
lo[c].shi = (D - shift[c] / W) % D;
shift[c] = (shift[c] + 1) % M;
}
if (i) {
lo[cnt].off = MSG+W*(i-1)+j;
lo[cnt].shi = 0;
} else if (j) {
lo[cnt].off = MSG+W*(q-1)+j-1;
lo[cnt].shi = 0;
} else {
lo[cnt].off = VAR-1;
lo[cnt].shi = 1;
}
lo[cnt+1].off = MSG+W*i+j;
lo[cnt+1].shi = 0;
std::sort(lo, lo + deg, [](const Loc &a, const Loc &b){ return a.off < b.off; });
#if 0
std::cout << deg;
for (int d = 0; d < deg; ++d)
std::cout << '\t' << (int)lo[d].off << ':' << (int)lo[d].shi;
std::cout << std::endl;
#endif
lo += deg;
}
}
//assert(lo <= loc + BNL);
//std::cerr << BNL - (lo - loc) << std::endl;
}
int operator()(int8_t *message, int8_t *parity, int trials = 25)
{
for (int i = 0; i < VAR; ++i)
csh[i] = 0;
for (int i = 0; i < K/M; ++i)
for (int j = 0; j < W; ++j)
for (int n = 0; n < D; ++n)
var[W*i+j].v[n] = message[M*i+W*n+j];
for (int i = 0; i < q; ++i)
for (int j = 0; j < W; ++j)
for (int n = 0; n < D; ++n)
var[MSG+W*i+j].v[n] = parity[q*(W*n+j)+i];
start = true;
while (--trials >= 0 && update())
start = false;
for (int i = 0; i < VAR; ++i)
var[i] = rotate(var[i], -csh[i]);
for (int i = 0; i < K/M; ++i)
for (int j = 0; j < W; ++j)
for (int n = 0; n < D; ++n)
message[M*i+W*n+j] = var[W*i+j].v[n];
for (int i = 0; i < q; ++i)
for (int j = 0; j < W; ++j)
for (int n = 0; n < D; ++n)
parity[q*(W*n+j)+i] = var[MSG+W*i+j].v[n];
return trials;
}
};
}
#endif