/* loudness.c — EBU R128 / ITU-R BS.1770 K-weighting + 100 ms energy steps.
 *
 * The heavy per-sample part of integrated-loudness measurement, compiled to
 * WebAssembly (scripts/build-loudness-wasm.sh → frontend/loudness-wasm.js).
 * JavaScript computes the filter coefficients for the sample rate, feeds the
 * decoded audio in chunks through inbuf(), then reads the per-step energies
 * and applies the 400 ms block gating (absolute -70 LUFS, relative -10 LU).
 * No libc: only multiply-adds here. */

#define CHUNK 65536
#define MAXSTEPS 400000              /* 100 ms steps: ~11 hours */

static float in0[CHUNK], in1[CHUNK];
static double stepE[MAXSTEPS];
static int nsteps, nch, stepLen, stepPos;
static double acc;
static double k1[5], k2[5];          /* b0 b1 b2 a1 a2 for each stage */
static double st[2][8];              /* per channel: stage1 x1 x2 y1 y2, stage2 x1 x2 y1 y2 */

float *inbuf(int ch) { return ch ? in1 : in0; }
double *steps(void) { return stepE; }
int count(void) { return nsteps; }

void setup(int channels, int steplen,
           double a0, double a1, double a2, double a3, double a4,
           double b0, double b1, double b2, double b3, double b4) {
  nch = channels > 1 ? 2 : 1;
  stepLen = steplen > 0 ? steplen : 1;
  stepPos = 0; nsteps = 0; acc = 0;
  k1[0] = a0; k1[1] = a1; k1[2] = a2; k1[3] = a3; k1[4] = a4;
  k2[0] = b0; k2[1] = b1; k2[2] = b2; k2[3] = b3; k2[4] = b4;
  for (int c = 0; c < 2; c++) for (int j = 0; j < 8; j++) st[c][j] = 0;
}

void feed(int n) {
  if (n > CHUNK) n = CHUNK;
  for (int i = 0; i < n; i++) {
    double e = 0;
    for (int c = 0; c < nch; c++) {
      double *s = st[c];
      double x = c ? in1[i] : in0[i];
      double y = k1[0] * x + k1[1] * s[0] + k1[2] * s[1] - k1[3] * s[2] - k1[4] * s[3];
      s[1] = s[0]; s[0] = x; s[3] = s[2]; s[2] = y;
      double z = k2[0] * y + k2[1] * s[4] + k2[2] * s[5] - k2[3] * s[6] - k2[4] * s[7];
      s[5] = s[4]; s[4] = y; s[7] = s[6]; s[6] = z;
      e += z * z;
    }
    acc += e;
    if (++stepPos == stepLen) {
      if (nsteps < MAXSTEPS) stepE[nsteps++] = acc;
      acc = 0; stepPos = 0;
    }
  }
}