We don't need an X sample cache anymore, and we also can simplify the access macro slightly.
Unified · Split
| @@ -115,7 +115,6 @@ typedef struct EBUR128Context { · 114 unmodified lines | ||
| 115 | 115 | |
| 116 | 116 | /* Filter caches. |
| 117 | 117 | * The mult by 3 in the following is for X[i], X[i-1] and X[i-2] */ |
| 118 | − double *x; ///< 3 input samples cache for each channel | |
| 119 | 118 | double *y; ///< 3 pre-filter samples cache for each channel |
| 120 | 119 | double *z; ///< 3 RLB-filter samples cache for each channel |
| 121 | 120 | double pre_b[3]; ///< pre-filter numerator coefficients |
| @@ -446,11 +445,10 @@ static int config_audio_output(AVFilterLink *outlink) · 324 unmodified lines | ||
| 446 | 445 | AV_CH_SURROUND_DIRECT_LEFT |AV_CH_SURROUND_DIRECT_RIGHT) |
| 447 | 446 | |
| 448 | 447 | ebur128->nb_channels = nb_channels; |
| 449 | − ebur128->x = av_calloc(nb_channels, 3 * sizeof(*ebur128->x)); | |
| 450 | 448 | ebur128->y = av_calloc(nb_channels, 3 * sizeof(*ebur128->y)); |
| 451 | 449 | ebur128->z = av_calloc(nb_channels, 3 * sizeof(*ebur128->z)); |
| 452 | 450 | ebur128->ch_weighting = av_calloc(nb_channels, sizeof(*ebur128->ch_weighting)); |
| 453 | − if (!ebur128->ch_weighting || !ebur128->x || !ebur128->y || !ebur128->z) | |
| 451 | + if (!ebur128->ch_weighting || !ebur128->y || !ebur128->z) | |
| 454 | 452 | return AVERROR(ENOMEM); |
| 455 | 453 | |
| 456 | 454 | #define I400_BINS(x) ((x) * 4 / 10) |
| @@ -673,34 +671,30 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *insamples) · 216 unmodified lines | ||
| 673 | 671 | MOVE_TO_NEXT_CACHED_ENTRY(3000); |
| 674 | 672 | |
| 675 | 673 | for (ch = 0; ch < nb_channels; ch++) { |
| 676 | − double bin; | |
| 677 | − | |
| 678 | 674 | if (ebur128->peak_mode & PEAK_MODE_SAMPLES_PEAKS) |
| 679 | 675 | ebur128->sample_peaks[ch] = FFMAX(ebur128->sample_peaks[ch], fabs(samples[idx_insample * nb_channels + ch])); |
| 680 | 676 | |
| 681 | − ebur128->x[ch * 3] = samples[idx_insample * nb_channels + ch]; // set X[i] | |
| 682 | − | |
| 683 | 677 | if (!ebur128->ch_weighting[ch]) |
| 684 | 678 | continue; |
| 685 | 679 | |
| 686 | 680 | /* Y[i] = X[i]*b0 + X[i-1]*b1 + X[i-2]*b2 - Y[i-1]*a1 - Y[i-2]*a2 */ |
| 687 | −#define FILTER(Y, X, NUM, DEN) do { \ | |
| 688 | − double *dst = ebur128->Y + ch*3; \ | |
| 689 | − double src = ebur128->X[ch*3] ; \ | |
| 690 | − double dst0 = NUM[0] * src + dst[1]; \ | |
| 691 | − dst[1] = NUM[1] * src + dst[2] - DEN[1] * dst0; \ | |
| 692 | − dst[2] = NUM[2] * src - DEN[2] * dst0; \ | |
| 693 | − dst[0] = dst0; \ | |
| 681 | +#define FILTER(DST, SRC, NUM, DEN) do { \ | |
| 682 | + const double tmp = DST[0] = NUM[0] * SRC + DST[1]; \ | |
| 683 | + DST[1] = NUM[1] * SRC + DST[2] - DEN[1] * tmp; \ | |
| 684 | + DST[2] = NUM[2] * SRC - DEN[2] * tmp; \ | |
| 694 | 685 | } while (0) |
| 695 | 686 | |
| 687 | + const double x = samples[idx_insample * nb_channels + ch]; | |
| 688 | + double *restrict y = &ebur128->y[3 * ch]; | |
| 689 | + double *restrict z = &ebur128->z[3 * ch]; | |
| 690 | + | |
| 696 | 691 | // TODO: merge both filters in one? |
| 697 | 692 | FILTER(y, x, ebur128->pre_b, ebur128->pre_a); // apply pre-filter |
| 698 | − FILTER(z, y, ebur128->rlb_b, ebur128->rlb_a); // apply RLB-filter | |
| 699 | − | |
| 700 | − bin = ebur128->z[ch * 3] * ebur128->z[ch * 3]; | |
| 693 | + FILTER(z, *y, ebur128->rlb_b, ebur128->rlb_a); // apply RLB-filter | |
| 701 | 694 | |
| 702 | 695 | /* add the new value, and limit the sum to the cache size (400ms or 3s) |
| 703 | 696 | * by removing the oldest one */ |
| 697 | + double bin = *z * *z; | |
| 704 | 698 | ebur128->i400.sum [ch] = ebur128->i400.sum [ch] + bin - ebur128->i400.cache [ch][bin_id_400]; |
| 705 | 699 | ebur128->i3000.sum[ch] = ebur128->i3000.sum[ch] + bin - ebur128->i3000.cache[ch][bin_id_3000]; |
| 706 | 700 | |
| @@ -1073,7 +1067,6 @@ static av_cold void uninit(AVFilterContext *ctx) · 366 unmodified lines | ||
| 1073 | 1067 | } |
| 1074 | 1068 | |
| 1075 | 1069 | av_freep(&ebur128->y_line_ref); |
| 1076 | − av_freep(&ebur128->x); | |
| 1077 | 1070 | av_freep(&ebur128->y); |
| 1078 | 1071 | av_freep(&ebur128->z); |
| 1079 | 1072 | av_freep(&ebur128->ch_weighting); |