avfilter/af_surround: speed up slice threading for stereo upmixing

And add stereo->octagonal/8.0 upmix.

Paul B Mahol committed 4 years ago

Commit e0fbf891 parent 104b5161 file · +831 −472Browse files

Unified · Split

@@ -85,6 +85,7 @@ typedef struct AudioSurroundContext { · 84 unmodified lines
85	85	float *input_levels;
86	86	float *output_levels;
87	87	int output_lfe;
	88	+ int have_lfe;
88	89	int lowcutf;
89	90	int highcutf;
90	91
@@ -103,20 +104,23 @@ typedef struct AudioSurroundContext { · 12 unmodified lines
103	104	AVFrame *overlap_buffer;
104	105	AVFrame *window;
105	106
	107	+ float *x_pos;
	108	+ float *y_pos;
	109	+ float *l_phase;
	110	+ float *r_phase;
	111	+ float *c_phase;
	112	+ float *lfe_mag;
	113	+ float *mag_total;
	114	+
106	115	int buf_size;
	116	+ int rdft_size;
107	117	int hop_size;
108	118	AVTXContext rdft, irdft;
109	119	av_tx_fn tx_fn, itx_fn;
110	120	float *window_func_lut;
111	121
112	122	void (filter)(AVFilterContext ctx);
113		− void (upmix_stereo)(AVFilterContext ctx,
114		− float l_phase,
115		− float r_phase,
116		− float c_phase,
117		− float mag_total,
118		− float x, float y,
119		− int n);
	123	+ int (upmix)(AVFilterContext ctx, void *arg, int ch, int nb_jobs);
120	124	void (upmix_2_1)(AVFilterContext ctx,
121	125	float l_phase,
122	126	float r_phase,
@@ -279,6 +283,7 @@ static int config_output(AVFilterLink *outlink) · 156 unmodified lines
279	283	if (ret < 0)
280	284	return ret;
281	285	}
	286	+
282	287	s->output_levels = av_malloc_array(s->nb_out_channels, sizeof(*s->output_levels));
283	288	if (!s->output_levels)
284	289	return AVERROR(ENOMEM);
@@ -318,6 +323,19 @@ static int config_output(AVFilterLink *outlink) · 33 unmodified lines
318	323	if (!s->overlap_buffer \|\| !s->output \|\| !s->output_out)
319	324	return AVERROR(ENOMEM);
320	325
	326	+ s->rdft_size = s->buf_size / 2 + 1;
	327	+
	328	+ s->x_pos = av_calloc(s->rdft_size, sizeof(*s->x_pos));
	329	+ s->y_pos = av_calloc(s->rdft_size, sizeof(*s->y_pos));
	330	+ s->l_phase = av_calloc(s->rdft_size, sizeof(*s->l_phase));
	331	+ s->r_phase = av_calloc(s->rdft_size, sizeof(*s->r_phase));
	332	+ s->c_phase = av_calloc(s->rdft_size, sizeof(*s->c_phase));
	333	+ s->mag_total = av_calloc(s->rdft_size, sizeof(*s->mag_total));
	334	+ s->lfe_mag = av_calloc(s->rdft_size, sizeof(*s->lfe_mag));
	335	+ if (!s->x_pos \|\| !s->y_pos \|\| !s->l_phase \|\| !s->r_phase \|\|
	336	+ !s->c_phase \|\| !s->lfe_mag)
	337	+ return AVERROR(ENOMEM);
	338	+
321	339	return 0;
322	340	}
323	341
@@ -362,141 +380,132 @@ static inline void get_lfe(int output_lfe, int n, float lowcut, float highcut, · 38 unmodified lines
362	380	}
363	381	}
364	382
365		−static void upmix_1_0(AVFilterContext *ctx,
366		− float l_phase,
367		− float r_phase,
368		− float c_phase,
369		− float mag_total,
370		− float x, float y,
371		− int n)
	383	+#define TRANSFORM \
	384	+ dst[2 * n ] = mag * cosf(ph); \
	385	+ dst[2 * n + 1] = mag * sinf(ph);
	386	+
	387	+static int upmix_1_0(AVFilterContext ctx, void arg, int ch, int nb_jobs)
372	388	{
373	389	AudioSurroundContext *s = ctx->priv;
374		− float mag, *dst;
375		−
376		− dst = (float *)s->output->extended_data[0];
377		−
378		− mag = powf(1.f - fabsf(x), s->fc_x) * powf((y + 1.f) * .5f, s->fc_y) * mag_total;
	390	+ float dst = (float )s->output->extended_data[0];
	391	+ const float *mag_total = s->mag_total;
	392	+ const int rdft_size = s->rdft_size;
	393	+ const float *phase = s->c_phase;
	394	+ const float *x = s->x_pos;
	395	+ const float *y = s->y_pos;
	396	+ const float fc_x = s->fc_x;
	397	+ const float fc_y = s->fc_y;
	398	+
	399	+ for (int n = 0; n < rdft_size; n++) {
	400	+ float mag = powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n];
	401	+ float ph = phase[n];
	402	+
	403	+ TRANSFORM
	404	+ }
379	405
380		− dst[2 * n ] = mag * cosf(c_phase);
381		− dst[2 * n + 1] = mag * sinf(c_phase);
	406	+ return 0;
382	407	}
383	408
384		−static void upmix_stereo(AVFilterContext *ctx,
385		− float l_phase,
386		− float r_phase,
387		− float c_phase,
388		− float mag_total,
389		− float x, float y,
390		− int n)
	409	+static int upmix_stereo(AVFilterContext ctx, void arg, int ch, int nb_jobs)
391	410	{
392	411	AudioSurroundContext *s = ctx->priv;
393		− float l_mag, r_mag, dstl, dstr;
394		−
395		− dstl = (float *)s->output->extended_data[0];
396		− dstr = (float *)s->output->extended_data[1];
397		−
398		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
399		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
400		−
401		− dstl[2 * n ] = l_mag * cosf(l_phase);
402		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
	412	+ float dst = (float )s->output->extended_data[ch];
	413	+ const float *phase = ch ? s->r_phase : s->l_phase;
	414	+ const float f_x = ch ? s->fr_x : s->fl_x;
	415	+ const float f_y = ch ? s->fr_y : s->fl_y;
	416	+ const float *mag_total = s->mag_total;
	417	+ const int rdft_size = s->rdft_size;
	418	+ const float neg = ch ? -1.f : 1.f;
	419	+ const float *x = s->x_pos;
	420	+ const float *y = s->y_pos;
	421	+
	422	+ for (int n = 0; n < rdft_size; n++) {
	423	+ float mag = powf(.5f * (neg * x[n] + 1.f), f_x) * powf((y[n] + 1.f) * .5f, f_y) * mag_total[n];
	424	+ float ph = phase[n];
	425	+
	426	+ TRANSFORM
	427	+ }
403	428
404		− dstr[2 * n ] = r_mag * cosf(r_phase);
405		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
	429	+ return 0;
406	430	}
407	431
408		−static void upmix_2_1(AVFilterContext *ctx,
409		− float l_phase,
410		− float r_phase,
411		− float c_phase,
412		− float mag_total,
413		− float x, float y,
414		− int n)
	432	+static int upmix_2_1(AVFilterContext ctx, void arg, int ch, int nb_jobs)
415	433	{
416	434	AudioSurroundContext *s = ctx->priv;
417		− float lfe_mag, l_mag, r_mag, dstl, dstr, *dstlfe;
418		−
419		− dstl = (float *)s->output->extended_data[0];
420		− dstr = (float *)s->output->extended_data[1];
421		− dstlfe = (float *)s->output->extended_data[2];
422		−
423		− get_lfe(s->output_lfe, n, s->lowcut, s->highcut, &lfe_mag, &mag_total, s->lfe_mode);
424		−
425		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
426		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
427		−
428		− dstl[2 * n ] = l_mag * cosf(l_phase);
429		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
430		−
431		− dstr[2 * n ] = r_mag * cosf(r_phase);
432		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
	435	+ float dst = (float )s->output->extended_data[ch];
	436	+ const float *mag_total = s->mag_total;
	437	+ const int rdft_size = s->rdft_size;
	438	+ const float *phase = ch == 2 ? s->c_phase : ch == 1 ? s->r_phase : s->l_phase;
	439	+ const float *lfe_mag = s->lfe_mag;
	440	+ const float f_x = ch ? s->fr_x : s->fl_x;
	441	+ const float f_y = ch ? s->fr_y : s->fl_y;
	442	+ const float neg = ch ? -1.f : 1.f;
	443	+ const float *x = s->x_pos;
	444	+ const float *y = s->y_pos;
	445	+
	446	+ for (int n = 0; n < rdft_size; n++) {
	447	+ float mag = ch == 2 ? lfe_mag[n] : powf(.5f * (neg * x[n] + 1.f), f_x) * powf((y[n] + 1.f) * .5f, f_y) * mag_total[n];
	448	+ float ph = phase[n];
	449	+
	450	+ TRANSFORM
	451	+ }
433	452
434		− dstlfe[2 * n ] = lfe_mag * cosf(c_phase);
435		− dstlfe[2 * n + 1] = lfe_mag * sinf(c_phase);
	453	+ return 0;
436	454	}
437	455
438		−static void upmix_3_0(AVFilterContext *ctx,
439		− float l_phase,
440		− float r_phase,
441		− float c_phase,
442		− float mag_total,
443		− float x, float y,
444		− int n)
	456	+static int upmix_3_0(AVFilterContext ctx, void arg, int ch, int nb_jobs)
445	457	{
446	458	AudioSurroundContext *s = ctx->priv;
447		− float l_mag, r_mag, c_mag, dstc, dstl, *dstr;
448		−
449		− dstl = (float *)s->output->extended_data[0];
450		− dstr = (float *)s->output->extended_data[1];
451		− dstc = (float *)s->output->extended_data[2];
452		−
453		− c_mag = powf(1.f - fabsf(x), s->fc_x) * powf((y + 1.f) * .5f, s->fc_y) * mag_total;
454		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
455		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
456		−
457		− dstl[2 * n ] = l_mag * cosf(l_phase);
458		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
459		−
460		− dstr[2 * n ] = r_mag * cosf(r_phase);
461		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
	459	+ float dst = (float )s->output->extended_data[ch];
	460	+ const float *phase = ch == 2 ? s->c_phase : ch == 1 ? s->r_phase : s->l_phase;
	461	+ const float f_x = ch ? s->fr_x : s->fl_x;
	462	+ const float f_y = ch ? s->fr_y : s->fl_y;
	463	+ const float *mag_total = s->mag_total;
	464	+ const int rdft_size = s->rdft_size;
	465	+ const float neg = ch ? -1.f : 1.f;
	466	+ const float fc_x = s->fc_x;
	467	+ const float fc_y = s->fc_y;
	468	+ const float *x = s->x_pos;
	469	+ const float *y = s->y_pos;
	470	+
	471	+ for (int n = 0; n < rdft_size; n++) {
	472	+ float mag = ch == 2 ?
	473	+ powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n] :
	474	+ powf(.5f * (neg * x[n] + 1.f), f_x) * powf((y[n] + 1.f) * .5f, f_y) * mag_total[n];
	475	+ float ph = phase[n];
	476	+
	477	+ TRANSFORM
	478	+ }
462	479
463		− dstc[2 * n ] = c_mag * cosf(c_phase);
464		− dstc[2 * n + 1] = c_mag * sinf(c_phase);
	480	+ return 0;
465	481	}
466	482
467		−static void upmix_3_1(AVFilterContext *ctx,
468		− float l_phase,
469		− float r_phase,
470		− float c_phase,
471		− float mag_total,
472		− float x, float y,
473		− int n)
	483	+static int upmix_3_1(AVFilterContext ctx, void arg, int ch, int nb_jobs)
474	484	{
475	485	AudioSurroundContext *s = ctx->priv;
476		− float lfe_mag, l_mag, r_mag, c_mag, dstc, dstl, dstr, dstlfe;
477		−
478		− dstl = (float *)s->output->extended_data[0];
479		− dstr = (float *)s->output->extended_data[1];
480		− dstc = (float *)s->output->extended_data[2];
481		− dstlfe = (float *)s->output->extended_data[3];
482		−
483		− get_lfe(s->output_lfe, n, s->lowcut, s->highcut, &lfe_mag, &mag_total, s->lfe_mode);
484		−
485		− c_mag = powf(1.f - fabsf(x), s->fc_x) * powf((y + 1.f) * .5f, s->fc_y) * mag_total;
486		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
487		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
488		−
489		− dstl[2 * n ] = l_mag * cosf(l_phase);
490		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
491		−
492		− dstr[2 * n ] = r_mag * cosf(r_phase);
493		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
494		−
495		− dstc[2 * n ] = c_mag * cosf(c_phase);
496		− dstc[2 * n + 1] = c_mag * sinf(c_phase);
	486	+ float dst = (float )s->output->extended_data[ch];
	487	+ const float *phase = ch >= 2 ? s->c_phase : ch == 1 ? s->r_phase : s->l_phase;
	488	+ const float f_x = ch ? s->fr_x : s->fl_x;
	489	+ const float f_y = ch ? s->fr_y : s->fl_y;
	490	+ const float *mag_total = s->mag_total;
	491	+ const int rdft_size = s->rdft_size;
	492	+ const float neg = ch ? -1.f : 1.f;
	493	+ const float *lfe_mag = s->lfe_mag;
	494	+ const float fc_x = s->fc_x;
	495	+ const float fc_y = s->fc_y;
	496	+ const float *x = s->x_pos;
	497	+ const float *y = s->y_pos;
	498	+
	499	+ for (int n = 0; n < rdft_size; n++) {
	500	+ float mag = ch == 3 ? lfe_mag[n] : ch == 2 ?
	501	+ powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n] :
	502	+ powf(.5f * (neg * x[n] + 1.f), f_x) * powf((y[n] + 1.f) * .5f, f_y) * mag_total[n];
	503	+ float ph = phase[n];
	504	+
	505	+ TRANSFORM
	506	+ }
497	507
498		− dstlfe[2 * n ] = lfe_mag * cosf(c_phase);
499		− dstlfe[2 * n + 1] = lfe_mag * sinf(c_phase);
	508	+ return 0;
500	509	}
501	510
502	511	static void upmix_3_1_surround(AVFilterContext *ctx,
@@ -534,256 +543,385 @@ static void upmix_3_1_surround(AVFilterContext *ctx, · 31 unmodified lines
534	543	dstlfe[2 * n + 1] = lfe_mag * sinf(c_phase);
535	544	}
536	545
537		−static void upmix_4_0(AVFilterContext *ctx,
538		− float l_phase,
539		− float r_phase,
540		− float c_phase,
541		− float mag_total,
542		− float x, float y,
543		− int n)
	546	+static int upmix_4_0(AVFilterContext ctx, void arg, int ch, int nb_jobs)
544	547	{
545	548	AudioSurroundContext *s = ctx->priv;
546		− float b_mag, l_mag, r_mag, c_mag, dstc, dstl, dstr, dstb;
547		−
548		− dstl = (float *)s->output->extended_data[0];
549		− dstr = (float *)s->output->extended_data[1];
550		− dstc = (float *)s->output->extended_data[2];
551		− dstb = (float *)s->output->extended_data[3];
552		−
553		− c_mag = powf(1.f - fabsf(x), s->fc_x) * powf((y + 1.f) * .5f, s->fc_y) * mag_total;
554		− b_mag = powf(1.f - fabsf(x), s->bc_x) * powf((1.f - y) * .5f, s->bc_y) * mag_total;
555		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
556		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
557		−
558		− dstl[2 * n ] = l_mag * cosf(l_phase);
559		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
560		−
561		− dstr[2 * n ] = r_mag * cosf(r_phase);
562		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
	549	+ float dst = (float )s->output->extended_data[ch];
	550	+ const float *mag_total = s->mag_total;
	551	+ const int rdft_size = s->rdft_size;
	552	+ const float *l_phase = s->l_phase;
	553	+ const float *r_phase = s->r_phase;
	554	+ const float *c_phase = s->c_phase;
	555	+ const float fl_x = s->fl_x;
	556	+ const float fl_y = s->fl_y;
	557	+ const float fr_x = s->fr_x;
	558	+ const float fr_y = s->fr_y;
	559	+ const float fc_x = s->fc_x;
	560	+ const float fc_y = s->fc_y;
	561	+ const float bc_x = s->bc_x;
	562	+ const float bc_y = s->bc_y;
	563	+ const float *x = s->x_pos;
	564	+ const float *y = s->y_pos;
	565	+
	566	+ for (int n = 0; n < rdft_size; n++) {
	567	+ float mag, ph;
	568	+
	569	+ switch (ch) {
	570	+ case 0:
	571	+ mag = powf(.5f * ( x[n] + 1.f), fl_x) * powf((y[n] + 1.f) * .5f, fl_y) * mag_total[n];
	572	+ ph = l_phase[n];
	573	+ break;
	574	+ case 1:
	575	+ mag = powf(.5f * (-x[n] + 1.f), fr_x) * powf((y[n] + 1.f) * .5f, fr_y) * mag_total[n];
	576	+ ph = r_phase[n];
	577	+ break;
	578	+ case 2:
	579	+ mag = powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n];
	580	+ ph = c_phase[n];
	581	+ break;
	582	+ case 3:
	583	+ mag = powf(1.f - fabsf(x[n]), bc_x) * powf((1.f - y[n]) * .5f, bc_y) * mag_total[n];
	584	+ ph = c_phase[n];
	585	+ break;
	586	+ }
563	587
564		− dstc[2 * n ] = c_mag * cosf(c_phase);
565		− dstc[2 * n + 1] = c_mag * sinf(c_phase);
	588	+ TRANSFORM
	589	+ }
566	590
567		− dstb[2 * n ] = b_mag * cosf(c_phase);
568		− dstb[2 * n + 1] = b_mag * sinf(c_phase);
	591	+ return 0;
569	592	}
570	593
571		−static void upmix_4_1(AVFilterContext *ctx,
572		− float l_phase,
573		− float r_phase,
574		− float c_phase,
575		− float mag_total,
576		− float x, float y,
577		− int n)
	594	+static int upmix_4_1(AVFilterContext ctx, void arg, int ch, int nb_jobs)
578	595	{
579	596	AudioSurroundContext *s = ctx->priv;
580		− float lfe_mag, b_mag, l_mag, r_mag, c_mag, dstc, dstl, dstr, dstb, *dstlfe;
581		−
582		− dstl = (float *)s->output->extended_data[0];
583		− dstr = (float *)s->output->extended_data[1];
584		− dstc = (float *)s->output->extended_data[2];
585		− dstlfe = (float *)s->output->extended_data[3];
586		− dstb = (float *)s->output->extended_data[4];
587		−
588		− get_lfe(s->output_lfe, n, s->lowcut, s->highcut, &lfe_mag, &mag_total, s->lfe_mode);
589		−
590		− dstlfe[2 * n ] = lfe_mag * cosf(c_phase);
591		− dstlfe[2 * n + 1] = lfe_mag * sinf(c_phase);
592		−
593		− c_mag = powf(1.f - fabsf(x), s->fc_x) * powf((y + 1.f) * .5f, s->fc_y) * mag_total;
594		− b_mag = powf(1.f - fabsf(x), s->bc_x) * powf((1.f - y) * .5f, s->bc_y) * mag_total;
595		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
596		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
597		−
598		− dstl[2 * n ] = l_mag * cosf(l_phase);
599		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
600		−
601		− dstr[2 * n ] = r_mag * cosf(r_phase);
602		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
	597	+ float dst = (float )s->output->extended_data[ch];
	598	+ const float *mag_total = s->mag_total;
	599	+ const int rdft_size = s->rdft_size;
	600	+ const float *lfe_mag = s->lfe_mag;
	601	+ const float *l_phase = s->l_phase;
	602	+ const float *r_phase = s->r_phase;
	603	+ const float *c_phase = s->c_phase;
	604	+ const float fl_x = s->fl_x;
	605	+ const float fl_y = s->fl_y;
	606	+ const float fr_x = s->fr_x;
	607	+ const float fr_y = s->fr_y;
	608	+ const float fc_x = s->fc_x;
	609	+ const float fc_y = s->fc_y;
	610	+ const float bc_x = s->bc_x;
	611	+ const float bc_y = s->bc_y;
	612	+ const float *x = s->x_pos;
	613	+ const float *y = s->y_pos;
	614	+
	615	+ for (int n = 0; n < rdft_size; n++) {
	616	+ float mag, ph;
	617	+
	618	+ switch (ch) {
	619	+ case 0:
	620	+ mag = powf(.5f * ( x[n] + 1.f), fl_x) * powf((y[n] + 1.f) * .5f, fl_y) * mag_total[n];
	621	+ ph = l_phase[n];
	622	+ break;
	623	+ case 1:
	624	+ mag = powf(.5f * (-x[n] + 1.f), fr_x) * powf((y[n] + 1.f) * .5f, fr_y) * mag_total[n];
	625	+ ph = r_phase[n];
	626	+ break;
	627	+ case 2:
	628	+ mag = powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n];
	629	+ ph = c_phase[n];
	630	+ break;
	631	+ case 3:
	632	+ mag = lfe_mag[n];
	633	+ ph = c_phase[n];
	634	+ break;
	635	+ case 4:
	636	+ mag = powf(1.f - fabsf(x[n]), bc_x) * powf((1.f - y[n]) * .5f, bc_y) * mag_total[n];
	637	+ ph = c_phase[n];
	638	+ break;
	639	+ }
603	640
604		− dstc[2 * n ] = c_mag * cosf(c_phase);
605		− dstc[2 * n + 1] = c_mag * sinf(c_phase);
	641	+ TRANSFORM
	642	+ }
606	643
607		− dstb[2 * n ] = b_mag * cosf(c_phase);
608		− dstb[2 * n + 1] = b_mag * sinf(c_phase);
	644	+ return 0;
609	645	}
610	646
611		−static void upmix_5_0_back(AVFilterContext *ctx,
612		− float l_phase,
613		− float r_phase,
614		− float c_phase,
615		− float mag_total,
616		− float x, float y,
617		− int n)
	647	+static int upmix_5_0_back(AVFilterContext ctx, void arg, int ch, int nb_jobs)
618	648	{
619	649	AudioSurroundContext *s = ctx->priv;
620		− float l_mag, r_mag, ls_mag, rs_mag, c_mag, dstc, dstl, dstr, dstls, *dstrs;
621		−
622		− dstl = (float *)s->output->extended_data[0];
623		− dstr = (float *)s->output->extended_data[1];
624		− dstc = (float *)s->output->extended_data[2];
625		− dstls = (float *)s->output->extended_data[3];
626		− dstrs = (float *)s->output->extended_data[4];
627		−
628		− c_mag = powf(1.f - fabsf(x), s->fc_x) * powf((y + 1.f) * .5f, s->fc_y) * mag_total;
629		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
630		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
631		− ls_mag = powf(.5f * ( x + 1.f), s->bl_x) * powf(1.f - ((y + 1.f) * .5f), s->bl_y) * mag_total;
632		− rs_mag = powf(.5f * (-x + 1.f), s->br_x) * powf(1.f - ((y + 1.f) * .5f), s->br_y) * mag_total;
633		−
634		− dstl[2 * n ] = l_mag * cosf(l_phase);
635		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
636		−
637		− dstr[2 * n ] = r_mag * cosf(r_phase);
638		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
639		−
640		− dstc[2 * n ] = c_mag * cosf(c_phase);
641		− dstc[2 * n + 1] = c_mag * sinf(c_phase);
	650	+ float dst = (float )s->output->extended_data[ch];
	651	+ const float *mag_total = s->mag_total;
	652	+ const int rdft_size = s->rdft_size;
	653	+ const float *l_phase = s->l_phase;
	654	+ const float *r_phase = s->r_phase;
	655	+ const float *c_phase = s->c_phase;
	656	+ const float fl_x = s->fl_x;
	657	+ const float fl_y = s->fl_y;
	658	+ const float fr_x = s->fr_x;
	659	+ const float fr_y = s->fr_y;
	660	+ const float bl_x = s->bl_x;
	661	+ const float bl_y = s->bl_y;
	662	+ const float br_x = s->br_x;
	663	+ const float br_y = s->br_y;
	664	+ const float fc_x = s->fc_x;
	665	+ const float fc_y = s->fc_y;
	666	+ const float *x = s->x_pos;
	667	+ const float *y = s->y_pos;
	668	+
	669	+ for (int n = 0; n < rdft_size; n++) {
	670	+ float mag, ph;
	671	+
	672	+ switch (ch) {
	673	+ case 2:
	674	+ mag = powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n];
	675	+ ph = c_phase[n];
	676	+ break;
	677	+ case 0:
	678	+ mag = powf(.5f * ( x[n] + 1.f), fl_x) * powf((y[n] + 1.f) * .5f, fl_y) * mag_total[n];
	679	+ ph = l_phase[n];
	680	+ break;
	681	+ case 1:
	682	+ mag = powf(.5f * (-x[n] + 1.f), fr_x) * powf((y[n] + 1.f) * .5f, fr_y) * mag_total[n];
	683	+ ph = r_phase[n];
	684	+ break;
	685	+ case 3:
	686	+ mag = powf(.5f * ( x[n] + 1.f), bl_x) * powf(1.f - ((y[n] + 1.f) * .5f), bl_y) * mag_total[n];
	687	+ ph = l_phase[n];
	688	+ break;
	689	+ case 4:
	690	+ mag = powf(.5f * (-x[n] + 1.f), br_x) * powf(1.f - ((y[n] + 1.f) * .5f), br_y) * mag_total[n];
	691	+ ph = r_phase[n];
	692	+ break;
	693	+ }
642	694
643		− dstls[2 * n ] = ls_mag * cosf(l_phase);
644		− dstls[2 * n + 1] = ls_mag * sinf(l_phase);
	695	+ TRANSFORM
	696	+ }
645	697
646		− dstrs[2 * n ] = rs_mag * cosf(r_phase);
647		− dstrs[2 * n + 1] = rs_mag * sinf(r_phase);
	698	+ return 0;
648	699	}
649	700
650		−static void upmix_5_1_back(AVFilterContext *ctx,
651		− float l_phase,
652		− float r_phase,
653		− float c_phase,
654		− float mag_total,
655		− float x, float y,
656		− int n)
	701	+static int upmix_5_1_back(AVFilterContext ctx, void arg, int ch, int nb_jobs)
657	702	{
658	703	AudioSurroundContext *s = ctx->priv;
659		− float lfe_mag, l_mag, r_mag, ls_mag, rs_mag, c_mag, dstc, dstl, dstr, dstls, dstrs, dstlfe;
660		−
661		− dstl = (float *)s->output->extended_data[0];
662		− dstr = (float *)s->output->extended_data[1];
663		− dstc = (float *)s->output->extended_data[2];
664		− dstlfe = (float *)s->output->extended_data[3];
665		− dstls = (float *)s->output->extended_data[4];
666		− dstrs = (float *)s->output->extended_data[5];
667		−
668		− get_lfe(s->output_lfe, n, s->lowcut, s->highcut, &lfe_mag, &mag_total, s->lfe_mode);
669		−
670		− c_mag = powf(1.f - fabsf(x), s->fc_x) * powf((y + 1.f) * .5f, s->fc_y) * mag_total;
671		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
672		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
673		− ls_mag = powf(.5f * ( x + 1.f), s->bl_x) * powf(1.f - ((y + 1.f) * .5f), s->bl_y) * mag_total;
674		− rs_mag = powf(.5f * (-x + 1.f), s->br_x) * powf(1.f - ((y + 1.f) * .5f), s->br_y) * mag_total;
675		−
676		− dstl[2 * n ] = l_mag * cosf(l_phase);
677		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
678		−
679		− dstr[2 * n ] = r_mag * cosf(r_phase);
680		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
681		−
682		− dstc[2 * n ] = c_mag * cosf(c_phase);
683		− dstc[2 * n + 1] = c_mag * sinf(c_phase);
684		−
685		− dstlfe[2 * n ] = lfe_mag * cosf(c_phase);
686		− dstlfe[2 * n + 1] = lfe_mag * sinf(c_phase);
687		−
688		− dstls[2 * n ] = ls_mag * cosf(l_phase);
689		− dstls[2 * n + 1] = ls_mag * sinf(l_phase);
	704	+ float dst = (float )s->output->extended_data[ch];
	705	+ const float *mag_total = s->mag_total;
	706	+ const float *lfe_mag = s->lfe_mag;
	707	+ const int rdft_size = s->rdft_size;
	708	+ const float *l_phase = s->l_phase;
	709	+ const float *r_phase = s->r_phase;
	710	+ const float *c_phase = s->c_phase;
	711	+ const float fl_x = s->fl_x;
	712	+ const float fl_y = s->fl_y;
	713	+ const float fr_x = s->fr_x;
	714	+ const float fr_y = s->fr_y;
	715	+ const float bl_x = s->bl_x;
	716	+ const float bl_y = s->bl_y;
	717	+ const float br_x = s->br_x;
	718	+ const float br_y = s->br_y;
	719	+ const float fc_x = s->fc_x;
	720	+ const float fc_y = s->fc_y;
	721	+ const float *x = s->x_pos;
	722	+ const float *y = s->y_pos;
	723	+ float mag, ph;
	724	+
	725	+ switch (ch) {
	726	+ case 0:
	727	+ for (int n = 0; n < rdft_size; n++) {
	728	+ mag = powf(.5f * ( x[n] + 1.f), fl_x) * powf((y[n] + 1.f) * .5f, fl_y) * mag_total[n];
	729	+ ph = l_phase[n];
	730	+ TRANSFORM
	731	+ }
	732	+ break;
	733	+ case 1:
	734	+ for (int n = 0; n < rdft_size; n++) {
	735	+ mag = powf(.5f * (-x[n] + 1.f), fr_x) * powf((y[n] + 1.f) * .5f, fr_y) * mag_total[n];
	736	+ ph = r_phase[n];
	737	+ TRANSFORM
	738	+ }
	739	+ break;
	740	+ case 2:
	741	+ for (int n = 0; n < rdft_size; n++) {
	742	+ mag = powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n];
	743	+ ph = c_phase[n];
	744	+ TRANSFORM
	745	+ }
	746	+ break;
	747	+ case 3:
	748	+ for (int n = 0; n < rdft_size; n++) {
	749	+ mag = lfe_mag[n];
	750	+ ph = c_phase[n];
	751	+ TRANSFORM
	752	+ }
	753	+ break;
	754	+ case 4:
	755	+ for (int n = 0; n < rdft_size; n++) {
	756	+ mag = powf(.5f * ( x[n] + 1.f), bl_x) * powf(1.f - ((y[n] + 1.f) * .5f), bl_y) * mag_total[n];
	757	+ ph = l_phase[n];
	758	+ TRANSFORM
	759	+ }
	760	+ break;
	761	+ case 5:
	762	+ for (int n = 0; n < rdft_size; n++) {
	763	+ mag = powf(.5f * (-x[n] + 1.f), br_x) * powf(1.f - ((y[n] + 1.f) * .5f), br_y) * mag_total[n];
	764	+ ph = r_phase[n];
	765	+ TRANSFORM
	766	+ }
	767	+ break;
	768	+ }
690	769
691		− dstrs[2 * n ] = rs_mag * cosf(r_phase);
692		− dstrs[2 * n + 1] = rs_mag * sinf(r_phase);
	770	+ return 0;
693	771	}
694	772
695		−static void upmix_6_0(AVFilterContext *ctx,
696		− float l_phase,
697		− float r_phase,
698		− float c_phase,
699		− float mag_total,
700		− float x, float y,
701		− int n)
	773	+static int upmix_6_0(AVFilterContext ctx, void arg, int ch, int nb_jobs)
702	774	{
703	775	AudioSurroundContext *s = ctx->priv;
704		− float l_mag, r_mag, ls_mag, rs_mag, c_mag, b_mag, dstc, dstb, dstl, dstr, dstls, dstrs;
705		−
706		− dstl = (float *)s->output->extended_data[0];
707		− dstr = (float *)s->output->extended_data[1];
708		− dstc = (float *)s->output->extended_data[2];
709		− dstb = (float *)s->output->extended_data[3];
710		− dstls = (float *)s->output->extended_data[4];
711		− dstrs = (float *)s->output->extended_data[5];
712		−
713		− c_mag = powf(1.f - fabsf(x), s->fc_x) * powf((y + 1.f) * .5f, s->fc_y) * mag_total;
714		− b_mag = powf(1.f - fabsf(x), s->bc_x) * powf((1.f - y) * .5f, s->bc_y) * mag_total;
715		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
716		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
717		− ls_mag = powf(.5f * ( x + 1.f), s->bl_x) * powf(1.f - ((y + 1.f) * .5f), s->bl_y) * mag_total;
718		− rs_mag = powf(.5f * (-x + 1.f), s->br_x) * powf(1.f - ((y + 1.f) * .5f), s->br_y) * mag_total;
719		−
720		− dstl[2 * n ] = l_mag * cosf(l_phase);
721		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
722		−
723		− dstr[2 * n ] = r_mag * cosf(r_phase);
724		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
725		−
726		− dstc[2 * n ] = c_mag * cosf(c_phase);
727		− dstc[2 * n + 1] = c_mag * sinf(c_phase);
728		−
729		− dstls[2 * n ] = ls_mag * cosf(l_phase);
730		− dstls[2 * n + 1] = ls_mag * sinf(l_phase);
731		−
732		− dstrs[2 * n ] = rs_mag * cosf(r_phase);
733		− dstrs[2 * n + 1] = rs_mag * sinf(r_phase);
	776	+ float dst = (float )s->output->extended_data[ch];
	777	+ const float *mag_total = s->mag_total;
	778	+ const int rdft_size = s->rdft_size;
	779	+ const float *l_phase = s->l_phase;
	780	+ const float *r_phase = s->r_phase;
	781	+ const float *c_phase = s->c_phase;
	782	+ const float fl_x = s->fl_x;
	783	+ const float fl_y = s->fl_y;
	784	+ const float fr_x = s->fr_x;
	785	+ const float fr_y = s->fr_y;
	786	+ const float bl_x = s->bl_x;
	787	+ const float bl_y = s->bl_y;
	788	+ const float br_x = s->br_x;
	789	+ const float br_y = s->br_y;
	790	+ const float bc_x = s->bc_x;
	791	+ const float bc_y = s->bc_y;
	792	+ const float fc_x = s->fc_x;
	793	+ const float fc_y = s->fc_y;
	794	+ const float *x = s->x_pos;
	795	+ const float *y = s->y_pos;
	796	+ float mag, ph;
	797	+
	798	+ switch (ch) {
	799	+ case 0:
	800	+ for (int n = 0; n < rdft_size; n++) {
	801	+ mag = powf(.5f * ( x[n] + 1.f), fl_x) * powf((y[n] + 1.f) * .5f, fl_y) * mag_total[n];
	802	+ ph = l_phase[n];
	803	+ TRANSFORM
	804	+ }
	805	+ break;
	806	+ case 1:
	807	+ for (int n = 0; n < rdft_size; n++) {
	808	+ mag = powf(.5f * (-x[n] + 1.f), fr_x) * powf((y[n] + 1.f) * .5f, fr_y) * mag_total[n];
	809	+ ph = r_phase[n];
	810	+ TRANSFORM
	811	+ }
	812	+ break;
	813	+ case 2:
	814	+ for (int n = 0; n < rdft_size; n++) {
	815	+ mag = powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n];
	816	+ ph = c_phase[n];
	817	+ TRANSFORM
	818	+ }
	819	+ break;
	820	+ case 3:
	821	+ for (int n = 0; n < rdft_size; n++) {
	822	+ mag = powf(1.f - fabsf(x[n]), bc_x) * powf((1.f - y[n]) * .5f, bc_y) * mag_total[n];
	823	+ ph = c_phase[n];
	824	+ TRANSFORM
	825	+ }
	826	+ break;
	827	+ case 4:
	828	+ for (int n = 0; n < rdft_size; n++) {
	829	+ mag = powf(.5f * ( x[n] + 1.f), bl_x) * powf(1.f - ((y[n] + 1.f) * .5f), bl_y) * mag_total[n];
	830	+ ph = l_phase[n];
	831	+ TRANSFORM
	832	+ }
	833	+ break;
	834	+ case 5:
	835	+ for (int n = 0; n < rdft_size; n++) {
	836	+ mag = powf(.5f * (-x[n] + 1.f), br_x) * powf(1.f - ((y[n] + 1.f) * .5f), br_y) * mag_total[n];
	837	+ ph = r_phase[n];
	838	+ TRANSFORM
	839	+ }
	840	+ break;
	841	+ }
734	842
735		− dstb[2 * n ] = b_mag * cosf(c_phase);
736		− dstb[2 * n + 1] = b_mag * sinf(c_phase);
	843	+ return 0;
737	844	}
738	845
739		−static void upmix_6_1(AVFilterContext *ctx,
740		− float l_phase,
741		− float r_phase,
742		− float c_phase,
743		− float mag_total,
744		− float x, float y,
745		− int n)
	846	+static int upmix_6_1(AVFilterContext ctx, void arg, int ch, int nb_jobs)
746	847	{
747	848	AudioSurroundContext *s = ctx->priv;
748		− float lfe_mag, l_mag, r_mag, ls_mag, rs_mag, c_mag, b_mag, dstc, dstb, dstl, dstr, dstls, dstrs, *dstlfe;
749		−
750		− dstl = (float *)s->output->extended_data[0];
751		− dstr = (float *)s->output->extended_data[1];
752		− dstc = (float *)s->output->extended_data[2];
753		− dstlfe = (float *)s->output->extended_data[3];
754		− dstb = (float *)s->output->extended_data[4];
755		− dstls = (float *)s->output->extended_data[5];
756		− dstrs = (float *)s->output->extended_data[6];
757		−
758		− get_lfe(s->output_lfe, n, s->lowcut, s->highcut, &lfe_mag, &mag_total, s->lfe_mode);
759		−
760		− c_mag = powf(1.f - fabsf(x), s->fc_x) * powf((y + 1.f) * .5f, s->fc_y) * mag_total;
761		− b_mag = powf(1.f - fabsf(x), s->bc_x) * powf((1.f - y) * .5f, s->bc_y) * mag_total;
762		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
763		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
764		− ls_mag = powf(.5f * ( x + 1.f), s->bl_x) * powf(1.f - ((y + 1.f) * .5f), s->bl_y) * mag_total;
765		− rs_mag = powf(.5f * (-x + 1.f), s->br_x) * powf(1.f - ((y + 1.f) * .5f), s->br_y) * mag_total;
766		−
767		− dstl[2 * n ] = l_mag * cosf(l_phase);
768		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
769		−
770		− dstr[2 * n ] = r_mag * cosf(r_phase);
771		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
772		−
773		− dstc[2 * n ] = c_mag * cosf(c_phase);
774		− dstc[2 * n + 1] = c_mag * sinf(c_phase);
775		−
776		− dstlfe[2 * n ] = lfe_mag * cosf(c_phase);
777		− dstlfe[2 * n + 1] = lfe_mag * sinf(c_phase);
778		−
779		− dstls[2 * n ] = ls_mag * cosf(l_phase);
780		− dstls[2 * n + 1] = ls_mag * sinf(l_phase);
781		−
782		− dstrs[2 * n ] = rs_mag * cosf(r_phase);
783		− dstrs[2 * n + 1] = rs_mag * sinf(r_phase);
	849	+ float dst = (float )s->output->extended_data[ch];
	850	+ const float *mag_total = s->mag_total;
	851	+ const int rdft_size = s->rdft_size;
	852	+ const float *lfe_mag = s->lfe_mag;
	853	+ const float *l_phase = s->l_phase;
	854	+ const float *r_phase = s->r_phase;
	855	+ const float *c_phase = s->c_phase;
	856	+ const float fl_x = s->fl_x;
	857	+ const float fl_y = s->fl_y;
	858	+ const float fr_x = s->fr_x;
	859	+ const float fr_y = s->fr_y;
	860	+ const float bl_x = s->bl_x;
	861	+ const float bl_y = s->bl_y;
	862	+ const float br_x = s->br_x;
	863	+ const float br_y = s->br_y;
	864	+ const float bc_x = s->bc_x;
	865	+ const float bc_y = s->bc_y;
	866	+ const float fc_x = s->fc_x;
	867	+ const float fc_y = s->fc_y;
	868	+ const float *x = s->x_pos;
	869	+ const float *y = s->y_pos;
	870	+ float mag, ph;
	871	+
	872	+ switch (ch) {
	873	+ case 0:
	874	+ for (int n = 0; n < rdft_size; n++) {
	875	+ mag = powf(.5f * ( x[n] + 1.f), fl_x) * powf((y[n] + 1.f) * .5f, fl_y) * mag_total[n];
	876	+ ph = l_phase[n];
	877	+ TRANSFORM
	878	+ }
	879	+ break;
	880	+ case 1:
	881	+ for (int n = 0; n < rdft_size; n++) {
	882	+ mag = powf(.5f * (-x[n] + 1.f), fr_x) * powf((y[n] + 1.f) * .5f, fr_y) * mag_total[n];
	883	+ ph = r_phase[n];
	884	+ TRANSFORM
	885	+ }
	886	+ break;
	887	+ case 2:
	888	+ for (int n = 0; n < rdft_size; n++) {
	889	+ mag = powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n];
	890	+ ph = c_phase[n];
	891	+ TRANSFORM
	892	+ }
	893	+ break;
	894	+ case 3:
	895	+ for (int n = 0; n < rdft_size; n++) {
	896	+ mag = lfe_mag[n];
	897	+ ph = c_phase[n];
	898	+ TRANSFORM
	899	+ }
	900	+ break;
	901	+ case 4:
	902	+ for (int n = 0; n < rdft_size; n++) {
	903	+ mag = powf(1.f - fabsf(x[n]), bc_x) * powf((1.f - y[n]) * .5f, bc_y) * mag_total[n];
	904	+ ph = c_phase[n];
	905	+ TRANSFORM
	906	+ }
	907	+ break;
	908	+ case 5:
	909	+ for (int n = 0; n < rdft_size; n++) {
	910	+ mag = powf(.5f * ( x[n] + 1.f), bl_x) * powf(1.f - ((y[n] + 1.f) * .5f), bl_y) * mag_total[n];
	911	+ ph = l_phase[n];
	912	+ TRANSFORM
	913	+ }
	914	+ break;
	915	+ case 6:
	916	+ for (int n = 0; n < rdft_size; n++) {
	917	+ mag = powf(.5f * (-x[n] + 1.f), br_x) * powf(1.f - ((y[n] + 1.f) * .5f), br_y) * mag_total[n];
	918	+ ph = r_phase[n];
	919	+ TRANSFORM
	920	+ }
	921	+ break;
	922	+ }
784	923
785		− dstb[2 * n ] = b_mag * cosf(c_phase);
786		− dstb[2 * n + 1] = b_mag * sinf(c_phase);
	924	+ return 0;
787	925	}
788	926
789	927	static void upmix_5_1_back_surround(AVFilterContext *ctx,
@@ -878,110 +1016,176 @@ static void upmix_5_1_back_2_1(AVFilterContext *ctx, · 88 unmodified lines
878	1016	dstrs[2 * n + 1] = rs_mag * sinf(r_phase);
879	1017	}
880	1018
881		−static void upmix_7_0(AVFilterContext *ctx,
882		− float l_phase,
883		− float r_phase,
884		− float c_phase,
885		− float mag_total,
886		− float x, float y,
887		− int n)
	1019	+static int upmix_7_0(AVFilterContext ctx, void arg, int ch, int nb_jobs)
888	1020	{
889		− float l_mag, r_mag, ls_mag, rs_mag, c_mag, lb_mag, rb_mag;
890		− float dstc, dstl, dstr, dstls, dstrs, dstlb, *dstrb;
891	1021	AudioSurroundContext *s = ctx->priv;
	1022	+ float dst = (float )s->output->extended_data[ch];
	1023	+ const float *mag_total = s->mag_total;
	1024	+ const int rdft_size = s->rdft_size;
	1025	+ const float *l_phase = s->l_phase;
	1026	+ const float *r_phase = s->r_phase;
	1027	+ const float *c_phase = s->c_phase;
	1028	+ const float fl_x = s->fl_x;
	1029	+ const float fl_y = s->fl_y;
	1030	+ const float fr_x = s->fr_x;
	1031	+ const float fr_y = s->fr_y;
	1032	+ const float fc_x = s->fc_x;
	1033	+ const float fc_y = s->fc_y;
	1034	+ const float bl_x = s->bl_x;
	1035	+ const float bl_y = s->bl_y;
	1036	+ const float br_x = s->br_x;
	1037	+ const float br_y = s->br_y;
	1038	+ const float sl_x = s->sl_x;
	1039	+ const float sl_y = s->sl_y;
	1040	+ const float sr_x = s->sr_x;
	1041	+ const float sr_y = s->sr_y;
	1042	+ const float *x = s->x_pos;
	1043	+ const float *y = s->y_pos;
	1044	+ float mag, ph;
	1045	+
	1046	+ switch (ch) {
	1047	+ case 0:
	1048	+ for (int n = 0; n < rdft_size; n++) {
	1049	+ mag = powf(.5f * ( x[n] + 1.f), fl_x) * powf((y[n] + 1.f) * .5f, fl_y) * mag_total[n];
	1050	+ ph = l_phase[n];
	1051	+ TRANSFORM
	1052	+ }
	1053	+ break;
	1054	+ case 1:
	1055	+ for (int n = 0; n < rdft_size; n++) {
	1056	+ mag = powf(.5f * (-x[n] + 1.f), fr_x) * powf((y[n] + 1.f) * .5f, fr_y) * mag_total[n];
	1057	+ ph = r_phase[n];
	1058	+ TRANSFORM
	1059	+ }
	1060	+ break;
	1061	+ case 2:
	1062	+ for (int n = 0; n < rdft_size; n++) {
	1063	+ mag = powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n];
	1064	+ ph = c_phase[n];
	1065	+ TRANSFORM
	1066	+ }
	1067	+ break;
	1068	+ case 3:
	1069	+ for (int n = 0; n < rdft_size; n++) {
	1070	+ mag = powf(.5f * ( x[n] + 1.f), bl_x) * powf(1.f - ((y[n] + 1.f) * .5f), bl_y) * mag_total[n];
	1071	+ ph = l_phase[n];
	1072	+ TRANSFORM
	1073	+ }
	1074	+ break;
	1075	+ case 4:
	1076	+ for (int n = 0; n < rdft_size; n++) {
	1077	+ mag = powf(.5f * (-x[n] + 1.f), br_x) * powf(1.f - ((y[n] + 1.f) * .5f), br_y) * mag_total[n];
	1078	+ ph = r_phase[n];
	1079	+ TRANSFORM
	1080	+ }
	1081	+ break;
	1082	+ case 5:
	1083	+ for (int n = 0; n < rdft_size; n++) {
	1084	+ mag = powf(.5f * ( x[n] + 1.f), sl_x) * powf(1.f - fabsf(y[n]), sl_y) * mag_total[n];
	1085	+ ph = l_phase[n];
	1086	+ TRANSFORM
	1087	+ }
	1088	+ break;
	1089	+ case 6:
	1090	+ for (int n = 0; n < rdft_size; n++) {
	1091	+ mag = powf(.5f * (-x[n] + 1.f), sr_x) * powf(1.f - fabsf(y[n]), sr_y) * mag_total[n];
	1092	+ ph = r_phase[n];
	1093	+ TRANSFORM
	1094	+ }
	1095	+ break;
	1096	+ }
892	1097
893		− dstl = (float *)s->output->extended_data[0];
894		− dstr = (float *)s->output->extended_data[1];
895		− dstc = (float *)s->output->extended_data[2];
896		− dstlb = (float *)s->output->extended_data[3];
897		− dstrb = (float *)s->output->extended_data[4];
898		− dstls = (float *)s->output->extended_data[5];
899		− dstrs = (float *)s->output->extended_data[6];
900		−
901		− c_mag = powf(1.f - fabsf(x), s->fc_x) * powf((y + 1.f) * .5f, s->fc_y) * mag_total;
902		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
903		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
904		− lb_mag = powf(.5f * ( x + 1.f), s->bl_x) * powf(1.f - ((y + 1.f) * .5f), s->bl_y) * mag_total;
905		− rb_mag = powf(.5f * (-x + 1.f), s->br_x) * powf(1.f - ((y + 1.f) * .5f), s->br_y) * mag_total;
906		− ls_mag = powf(.5f * ( x + 1.f), s->sl_x) * powf(1.f - fabsf(y), s->sl_y) * mag_total;
907		− rs_mag = powf(.5f * (-x + 1.f), s->sr_x) * powf(1.f - fabsf(y), s->sr_y) * mag_total;
908		−
909		− dstl[2 * n ] = l_mag * cosf(l_phase);
910		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
911		−
912		− dstr[2 * n ] = r_mag * cosf(r_phase);
913		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
914		−
915		− dstc[2 * n ] = c_mag * cosf(c_phase);
916		− dstc[2 * n + 1] = c_mag * sinf(c_phase);
917		−
918		− dstlb[2 * n ] = lb_mag * cosf(l_phase);
919		− dstlb[2 * n + 1] = lb_mag * sinf(l_phase);
920		−
921		− dstrb[2 * n ] = rb_mag * cosf(r_phase);
922		− dstrb[2 * n + 1] = rb_mag * sinf(r_phase);
923		−
924		− dstls[2 * n ] = ls_mag * cosf(l_phase);
925		− dstls[2 * n + 1] = ls_mag * sinf(l_phase);
926		−
927		− dstrs[2 * n ] = rs_mag * cosf(r_phase);
928		− dstrs[2 * n + 1] = rs_mag * sinf(r_phase);
	1098	+ return 0;
929	1099	}
930	1100
931		−static void upmix_7_1(AVFilterContext *ctx,
932		− float l_phase,
933		− float r_phase,
934		− float c_phase,
935		− float mag_total,
936		− float x, float y,
937		− int n)
	1101	+static int upmix_7_1(AVFilterContext ctx, void arg, int ch, int nb_jobs)
938	1102	{
939		− float lfe_mag, l_mag, r_mag, ls_mag, rs_mag, c_mag, lb_mag, rb_mag;
940		− float dstc, dstl, dstr, dstls, dstrs, dstlb, dstrb, dstlfe;
941	1103	AudioSurroundContext *s = ctx->priv;
	1104	+ float dst = (float )s->output->extended_data[ch];
	1105	+ const float *mag_total = s->mag_total;
	1106	+ const int rdft_size = s->rdft_size;
	1107	+ const float *lfe_mag = s->lfe_mag;
	1108	+ const float *l_phase = s->l_phase;
	1109	+ const float *r_phase = s->r_phase;
	1110	+ const float *c_phase = s->c_phase;
	1111	+ const float fl_x = s->fl_x;
	1112	+ const float fl_y = s->fl_y;
	1113	+ const float fr_x = s->fr_x;
	1114	+ const float fr_y = s->fr_y;
	1115	+ const float fc_x = s->fc_x;
	1116	+ const float fc_y = s->fc_y;
	1117	+ const float bl_x = s->bl_x;
	1118	+ const float bl_y = s->bl_y;
	1119	+ const float br_x = s->br_x;
	1120	+ const float br_y = s->br_y;
	1121	+ const float sl_x = s->sl_x;
	1122	+ const float sl_y = s->sl_y;
	1123	+ const float sr_x = s->sr_x;
	1124	+ const float sr_y = s->sr_y;
	1125	+ const float *x = s->x_pos;
	1126	+ const float *y = s->y_pos;
	1127	+ float mag, ph;
	1128	+
	1129	+ switch (ch) {
	1130	+ case 0:
	1131	+ for (int n = 0; n < rdft_size; n++) {
	1132	+ mag = powf(.5f * ( x[n] + 1.f), fl_x) * powf((y[n] + 1.f) * .5f, fl_y) * mag_total[n];
	1133	+ ph = l_phase[n];
	1134	+ TRANSFORM
	1135	+ }
	1136	+ break;
	1137	+ case 1:
	1138	+ for (int n = 0; n < rdft_size; n++) {
	1139	+ mag = powf(.5f * (-x[n] + 1.f), fr_x) * powf((y[n] + 1.f) * .5f, fr_y) * mag_total[n];
	1140	+ ph = r_phase[n];
	1141	+ TRANSFORM
	1142	+ }
	1143	+ break;
	1144	+ case 2:
	1145	+ for (int n = 0; n < rdft_size; n++) {
	1146	+ mag = powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n];
	1147	+ ph = c_phase[n];
	1148	+ TRANSFORM
	1149	+ }
	1150	+ break;
	1151	+ case 3:
	1152	+ for (int n = 0; n < rdft_size; n++) {
	1153	+ mag = lfe_mag[n];
	1154	+ ph = c_phase[n];
	1155	+ TRANSFORM
	1156	+ }
	1157	+ break;
	1158	+ case 4:
	1159	+ for (int n = 0; n < rdft_size; n++) {
	1160	+ mag = powf(.5f * ( x[n] + 1.f), bl_x) * powf(1.f - ((y[n] + 1.f) * .5f), bl_y) * mag_total[n];
	1161	+ ph = l_phase[n];
	1162	+ TRANSFORM
	1163	+ }
	1164	+ break;
	1165	+ case 5:
	1166	+ for (int n = 0; n < rdft_size; n++) {
	1167	+ mag = powf(.5f * (-x[n] + 1.f), br_x) * powf(1.f - ((y[n] + 1.f) * .5f), br_y) * mag_total[n];
	1168	+ ph = r_phase[n];
	1169	+ TRANSFORM
	1170	+ }
	1171	+ break;
	1172	+ case 6:
	1173	+ for (int n = 0; n < rdft_size; n++) {
	1174	+ mag = powf(.5f * ( x[n] + 1.f), sl_x) * powf(1.f - fabsf(y[n]), sl_y) * mag_total[n];
	1175	+ ph = l_phase[n];
	1176	+ TRANSFORM
	1177	+ }
	1178	+ break;
	1179	+ case 7:
	1180	+ for (int n = 0; n < rdft_size; n++) {
	1181	+ mag = powf(.5f * (-x[n] + 1.f), sr_x) * powf(1.f - fabsf(y[n]), sr_y) * mag_total[n];
	1182	+ ph = r_phase[n];
	1183	+ TRANSFORM
	1184	+ }
	1185	+ break;
	1186	+ }
942	1187
943		− dstl = (float *)s->output->extended_data[0];
944		− dstr = (float *)s->output->extended_data[1];
945		− dstc = (float *)s->output->extended_data[2];
946		− dstlfe = (float *)s->output->extended_data[3];
947		− dstlb = (float *)s->output->extended_data[4];
948		− dstrb = (float *)s->output->extended_data[5];
949		− dstls = (float *)s->output->extended_data[6];
950		− dstrs = (float *)s->output->extended_data[7];
951		−
952		− get_lfe(s->output_lfe, n, s->lowcut, s->highcut, &lfe_mag, &mag_total, s->lfe_mode);
953		−
954		− c_mag = powf(1.f - fabsf(x), s->fc_x) * powf((y + 1.f) * .5f, s->fc_y) * mag_total;
955		− l_mag = powf(.5f * ( x + 1.f), s->fl_x) * powf((y + 1.f) * .5f, s->fl_y) * mag_total;
956		− r_mag = powf(.5f * (-x + 1.f), s->fr_x) * powf((y + 1.f) * .5f, s->fr_y) * mag_total;
957		− lb_mag = powf(.5f * ( x + 1.f), s->bl_x) * powf(1.f - ((y + 1.f) * .5f), s->bl_y) * mag_total;
958		− rb_mag = powf(.5f * (-x + 1.f), s->br_x) * powf(1.f - ((y + 1.f) * .5f), s->br_y) * mag_total;
959		− ls_mag = powf(.5f * ( x + 1.f), s->sl_x) * powf(1.f - fabsf(y), s->sl_y) * mag_total;
960		− rs_mag = powf(.5f * (-x + 1.f), s->sr_x) * powf(1.f - fabsf(y), s->sr_y) * mag_total;
961		−
962		− dstl[2 * n ] = l_mag * cosf(l_phase);
963		− dstl[2 * n + 1] = l_mag * sinf(l_phase);
964		−
965		− dstr[2 * n ] = r_mag * cosf(r_phase);
966		− dstr[2 * n + 1] = r_mag * sinf(r_phase);
967		−
968		− dstc[2 * n ] = c_mag * cosf(c_phase);
969		− dstc[2 * n + 1] = c_mag * sinf(c_phase);
970		−
971		− dstlfe[2 * n ] = lfe_mag * cosf(c_phase);
972		− dstlfe[2 * n + 1] = lfe_mag * sinf(c_phase);
973		−
974		− dstlb[2 * n ] = lb_mag * cosf(l_phase);
975		− dstlb[2 * n + 1] = lb_mag * sinf(l_phase);
976		−
977		− dstrb[2 * n ] = rb_mag * cosf(r_phase);
978		− dstrb[2 * n + 1] = rb_mag * sinf(r_phase);
979		−
980		− dstls[2 * n ] = ls_mag * cosf(l_phase);
981		− dstls[2 * n + 1] = ls_mag * sinf(l_phase);
982		−
983		− dstrs[2 * n ] = rs_mag * cosf(r_phase);
984		− dstrs[2 * n + 1] = rs_mag * sinf(r_phase);
	1188	+ return 0;
985	1189	}
986	1190
987	1191	static void upmix_7_1_5_0_side(AVFilterContext *ctx,
@@ -1100,16 +1304,116 @@ static void upmix_7_1_5_1(AVFilterContext *ctx, · 112 unmodified lines
1100	1304	dstrs[2 * n + 1] = rs_mag * sinf(sr_phase);
1101	1305	}
1102	1306
1103		−static void filter_stereo(AVFilterContext *ctx)
	1307	+static int upmix_8_0(AVFilterContext ctx, void arg, int ch, int nb_jobs)
1104	1308	{
1105	1309	AudioSurroundContext *s = ctx->priv;
1106		− float srcl, srcr;
1107		− int n;
	1310	+ float dst = (float )s->output->extended_data[ch];
	1311	+ const float *mag_total = s->mag_total;
	1312	+ const int rdft_size = s->rdft_size;
	1313	+ const float *l_phase = s->l_phase;
	1314	+ const float *r_phase = s->r_phase;
	1315	+ const float *c_phase = s->c_phase;
	1316	+ const float fl_x = s->fl_x;
	1317	+ const float fl_y = s->fl_y;
	1318	+ const float fr_x = s->fr_x;
	1319	+ const float fr_y = s->fr_y;
	1320	+ const float fc_x = s->fc_x;
	1321	+ const float fc_y = s->fc_y;
	1322	+ const float bl_x = s->bl_x;
	1323	+ const float bl_y = s->bl_y;
	1324	+ const float br_x = s->br_x;
	1325	+ const float br_y = s->br_y;
	1326	+ const float sl_x = s->sl_x;
	1327	+ const float sl_y = s->sl_y;
	1328	+ const float sr_x = s->sr_x;
	1329	+ const float sr_y = s->sr_y;
	1330	+ const float bc_x = s->bc_x;
	1331	+ const float bc_y = s->bc_y;
	1332	+ const float *x = s->x_pos;
	1333	+ const float *y = s->y_pos;
	1334	+ float mag, ph;
	1335	+
	1336	+ switch (ch) {
	1337	+ case 0:
	1338	+ for (int n = 0; n < rdft_size; n++) {
	1339	+ mag = powf(.5f * ( x[n] + 1.f), fl_x) * powf((y[n] + 1.f) * .5f, fl_y) * mag_total[n];
	1340	+ ph = l_phase[n];
	1341	+ TRANSFORM
	1342	+ }
	1343	+ break;
	1344	+ case 1:
	1345	+ for (int n = 0; n < rdft_size; n++) {
	1346	+ mag = powf(.5f * (-x[n] + 1.f), fr_x) * powf((y[n] + 1.f) * .5f, fr_y) * mag_total[n];
	1347	+ ph = r_phase[n];
	1348	+ TRANSFORM
	1349	+ }
	1350	+ break;
	1351	+ case 2:
	1352	+ for (int n = 0; n < rdft_size; n++) {
	1353	+ mag = powf(1.f - fabsf(x[n]), fc_x) * powf((y[n] + 1.f) * .5f, fc_y) * mag_total[n];
	1354	+ ph = c_phase[n];
	1355	+ TRANSFORM
	1356	+ }
	1357	+ break;
	1358	+ case 3:
	1359	+ for (int n = 0; n < rdft_size; n++) {
	1360	+ mag = powf(.5f * ( x[n] + 1.f), bl_x) * powf(1.f - ((y[n] + 1.f) * .5f), bl_y) * mag_total[n];
	1361	+ ph = l_phase[n];
	1362	+ TRANSFORM
	1363	+ }
	1364	+ break;
	1365	+ case 4:
	1366	+ for (int n = 0; n < rdft_size; n++) {
	1367	+ mag = powf(.5f * (-x[n] + 1.f), br_x) * powf(1.f - ((y[n] + 1.f) * .5f), br_y) * mag_total[n];
	1368	+ ph = r_phase[n];
	1369	+ TRANSFORM
	1370	+ }
	1371	+ break;
	1372	+ case 5:
	1373	+ for (int n = 0; n < rdft_size; n++) {
	1374	+ mag = powf(1.f - fabsf(x[n]), bc_x) * powf((1.f - y[n]) * .5f, bc_y) * mag_total[n];
	1375	+ ph = l_phase[n];
	1376	+ TRANSFORM
	1377	+ }
	1378	+ break;
	1379	+ case 6:
	1380	+ for (int n = 0; n < rdft_size; n++) {
	1381	+ mag = powf(.5f * ( x[n] + 1.f), sl_x) * powf(1.f - fabsf(y[n]), sl_y) * mag_total[n];
	1382	+ ph = l_phase[n];
	1383	+ TRANSFORM
	1384	+ }
	1385	+ break;
	1386	+ case 7:
	1387	+ for (int n = 0; n < rdft_size; n++) {
	1388	+ mag = powf(.5f * (-x[n] + 1.f), sr_x) * powf(1.f - fabsf(y[n]), sr_y) * mag_total[n];
	1389	+ ph = r_phase[n];
	1390	+ TRANSFORM
	1391	+ }
	1392	+ break;
	1393	+ }
1108	1394
1109		− srcl = (float *)s->input->extended_data[0];
1110		− srcr = (float *)s->input->extended_data[1];
	1395	+ return 0;
	1396	+}
1111	1397
1112		− for (n = 0; n < s->buf_size / 2 + 1; n++) {
	1398	+static void filter_stereo(AVFilterContext *ctx)
	1399	+{
	1400	+ AudioSurroundContext *s = ctx->priv;
	1401	+ const float srcl = (const float )s->input->extended_data[0];
	1402	+ const float srcr = (const float )s->input->extended_data[1];
	1403	+ const int output_lfe = s->output_lfe && s->have_lfe;
	1404	+ const int lfe_mode = s->lfe_mode;
	1405	+ const float highcut = s->highcut;
	1406	+ const float lowcut = s->lowcut;
	1407	+ const float angle = s->angle;
	1408	+ float *magtotal = s->mag_total;
	1409	+ float *lfemag = s->lfe_mag;
	1410	+ float *lphase = s->l_phase;
	1411	+ float *rphase = s->r_phase;
	1412	+ float *cphase = s->c_phase;
	1413	+ float *xpos = s->x_pos;
	1414	+ float *ypos = s->y_pos;
	1415	+
	1416	+ for (int n = 0; n < s->rdft_size; n++) {
1113	1417	float l_re = srcl[2 * n], r_re = srcr[2 * n];
1114	1418	float l_im = srcl[2 * n + 1], r_im = srcr[2 * n + 1];
1115	1419	float c_phase = atan2f(l_im + r_im, l_re + r_re);
@@ -1127,9 +1431,15 @@ static void filter_stereo(AVFilterContext *ctx) · 11 unmodified lines
1127	1431	phase_dif = 2 * M_PI - phase_dif;
1128	1432
1129	1433	stereo_position(mag_dif, phase_dif, &x, &y);
1130		− stereo_transform(&x, &y, s->angle);
1131		−
1132		− s->upmix_stereo(ctx, l_phase, r_phase, c_phase, mag_total, x, y, n);
	1434	+ stereo_transform(&x, &y, angle);
	1435	+ get_lfe(output_lfe, n, lowcut, highcut, &lfemag[n], &mag_total, lfe_mode);
	1436	+
	1437	+ xpos[n] = x;
	1438	+ ypos[n] = y;
	1439	+ lphase[n] = l_phase;
	1440	+ rphase[n] = r_phase;
	1441	+ cphase[n] = c_phase;
	1442	+ magtotal[n] = mag_total;
1133	1443	}
1134	1444	}
1135	1445
@@ -1143,7 +1453,7 @@ static void filter_surround(AVFilterContext *ctx) · 7 unmodified lines
1143	1453	srcr = (float *)s->input->extended_data[1];
1144	1454	srcc = (float *)s->input->extended_data[2];
1145	1455
1146		− for (n = 0; n < s->buf_size / 2 + 1; n++) {
	1456	+ for (n = 0; n < s->rdft_size; n++) {
1147	1457	float l_re = srcl[2 * n], r_re = srcr[2 * n];
1148	1458	float l_im = srcl[2 * n + 1], r_im = srcr[2 * n + 1];
1149	1459	float c_re = srcc[2 * n], c_im = srcc[2 * n + 1];
@@ -1179,7 +1489,7 @@ static void filter_2_1(AVFilterContext *ctx) · 29 unmodified lines
1179	1489	srcr = (float *)s->input->extended_data[1];
1180	1490	srclfe = (float *)s->input->extended_data[2];
1181	1491
1182		− for (n = 0; n < s->buf_size / 2 + 1; n++) {
	1492	+ for (n = 0; n < s->rdft_size; n++) {
1183	1493	float l_re = srcl[2 * n], r_re = srcr[2 * n];
1184	1494	float l_im = srcl[2 * n + 1], r_im = srcr[2 * n + 1];
1185	1495	float lfe_re = srclfe[2 * n], lfe_im = srclfe[2 * n + 1];
@@ -1216,7 +1526,7 @@ static void filter_5_0_side(AVFilterContext *ctx) · 30 unmodified lines
1216	1526	srcsl = (float *)s->input->extended_data[3];
1217	1527	srcsr = (float *)s->input->extended_data[4];
1218	1528
1219		− for (n = 0; n < s->buf_size / 2 + 1; n++) {
	1529	+ for (n = 0; n < s->rdft_size; n++) {
1220	1530	float fl_re = srcl[2 * n], fr_re = srcr[2 * n];
1221	1531	float fl_im = srcl[2 * n + 1], fr_im = srcr[2 * n + 1];
1222	1532	float c_re = srcc[2 * n], c_im = srcc[2 * n + 1];
@@ -1274,7 +1584,7 @@ static void filter_5_1_side(AVFilterContext *ctx) · 51 unmodified lines
1274	1584	srcsl = (float *)s->input->extended_data[4];
1275	1585	srcsr = (float *)s->input->extended_data[5];
1276	1586
1277		− for (n = 0; n < s->buf_size / 2 + 1; n++) {
	1587	+ for (n = 0; n < s->rdft_size; n++) {
1278	1588	float fl_re = srcl[2 * n], fr_re = srcr[2 * n];
1279	1589	float fl_im = srcl[2 * n + 1], fr_im = srcr[2 * n + 1];
1280	1590	float c_re = srcc[2 * n], c_im = srcc[2 * n + 1];
@@ -1333,7 +1643,7 @@ static void filter_5_1_back(AVFilterContext *ctx) · 52 unmodified lines
1333	1643	srcbl = (float *)s->input->extended_data[4];
1334	1644	srcbr = (float *)s->input->extended_data[5];
1335	1645
1336		− for (n = 0; n < s->buf_size / 2 + 1; n++) {
	1646	+ for (n = 0; n < s->rdft_size; n++) {
1337	1647	float fl_re = srcl[2 * n], fr_re = srcr[2 * n];
1338	1648	float fl_im = srcl[2 * n + 1], fr_im = srcr[2 * n + 1];
1339	1649	float c_re = srcc[2 * n], c_im = srcc[2 * n + 1];
@@ -1409,48 +1719,59 @@ static av_cold int init(AVFilterContext *ctx) · 69 unmodified lines
1409	1719	out_channel_layout = s->out_channel_layout.order == AV_CHANNEL_ORDER_NATIVE ?
1410	1720	s->out_channel_layout.u.mask : 0;
1411	1721
	1722	+ s->have_lfe = 0;
	1723	+
1412	1724	switch (in_channel_layout) {
1413	1725	case AV_CH_LAYOUT_STEREO:
1414	1726	s->filter = filter_stereo;
1415	1727	switch (out_channel_layout) {
1416	1728	case AV_CH_LAYOUT_MONO:
1417		− s->upmix_stereo = upmix_1_0;
	1729	+ s->upmix = upmix_1_0;
1418	1730	break;
1419	1731	case AV_CH_LAYOUT_STEREO:
1420		− s->upmix_stereo = upmix_stereo;
	1732	+ s->upmix = upmix_stereo;
1421	1733	break;
1422	1734	case AV_CH_LAYOUT_2POINT1:
1423		− s->upmix_stereo = upmix_2_1;
	1735	+ s->upmix = upmix_2_1;
	1736	+ s->have_lfe = 1;
1424	1737	break;
1425	1738	case AV_CH_LAYOUT_SURROUND:
1426		− s->upmix_stereo = upmix_3_0;
	1739	+ s->upmix = upmix_3_0;
1427	1740	break;
1428	1741	case AV_CH_LAYOUT_3POINT1:
1429		− s->upmix_stereo = upmix_3_1;
	1742	+ s->upmix = upmix_3_1;
	1743	+ s->have_lfe = 1;
1430	1744	break;
1431	1745	case AV_CH_LAYOUT_4POINT0:
1432		− s->upmix_stereo = upmix_4_0;
	1746	+ s->upmix = upmix_4_0;
1433	1747	break;
1434	1748	case AV_CH_LAYOUT_4POINT1:
1435		− s->upmix_stereo = upmix_4_1;
	1749	+ s->upmix = upmix_4_1;
	1750	+ s->have_lfe = 1;
1436	1751	break;
1437	1752	case AV_CH_LAYOUT_5POINT0_BACK:
1438		− s->upmix_stereo = upmix_5_0_back;
	1753	+ s->upmix = upmix_5_0_back;
1439	1754	break;
1440	1755	case AV_CH_LAYOUT_5POINT1_BACK:
1441		− s->upmix_stereo = upmix_5_1_back;
	1756	+ s->upmix = upmix_5_1_back;
	1757	+ s->have_lfe = 1;
1442	1758	break;
1443	1759	case AV_CH_LAYOUT_6POINT0:
1444		− s->upmix_stereo = upmix_6_0;
	1760	+ s->upmix = upmix_6_0;
1445	1761	break;
1446	1762	case AV_CH_LAYOUT_6POINT1:
1447		− s->upmix_stereo = upmix_6_1;
	1763	+ s->upmix = upmix_6_1;
	1764	+ s->have_lfe = 1;
1448	1765	break;
1449	1766	case AV_CH_LAYOUT_7POINT0:
1450		− s->upmix_stereo = upmix_7_0;
	1767	+ s->upmix = upmix_7_0;
1451	1768	break;
1452	1769	case AV_CH_LAYOUT_7POINT1:
1453		− s->upmix_stereo = upmix_7_1;
	1770	+ s->upmix = upmix_7_1;
	1771	+ s->have_lfe = 1;
	1772	+ break;
	1773	+ case AV_CH_LAYOUT_OCTAGONAL:
	1774	+ s->upmix = upmix_8_0;
1454	1775	break;
1455	1776	default:
1456	1777	goto fail;
@@ -1461,6 +1782,7 @@ static av_cold int init(AVFilterContext *ctx) · 4 unmodified lines
1461	1782	switch (out_channel_layout) {
1462	1783	case AV_CH_LAYOUT_5POINT1_BACK:
1463	1784	s->upmix_2_1 = upmix_5_1_back_2_1;
	1785	+ s->have_lfe = 1;
1464	1786	break;
1465	1787	default:
1466	1788	goto fail;
@@ -1471,9 +1793,11 @@ static av_cold int init(AVFilterContext *ctx) · 4 unmodified lines
1471	1793	switch (out_channel_layout) {
1472	1794	case AV_CH_LAYOUT_3POINT1:
1473	1795	s->upmix_3_0 = upmix_3_1_surround;
	1796	+ s->have_lfe = 1;
1474	1797	break;
1475	1798	case AV_CH_LAYOUT_5POINT1_BACK:
1476	1799	s->upmix_3_0 = upmix_5_1_back_surround;
	1800	+ s->have_lfe = 1;
1477	1801	break;
1478	1802	default:
1479	1803	goto fail;
@@ -1484,6 +1808,7 @@ static av_cold int init(AVFilterContext *ctx) · 4 unmodified lines
1484	1808	switch (out_channel_layout) {
1485	1809	case AV_CH_LAYOUT_7POINT1:
1486	1810	s->upmix_5_0 = upmix_7_1_5_0_side;
	1811	+ s->have_lfe = 1;
1487	1812	break;
1488	1813	default:
1489	1814	goto fail;
@@ -1494,6 +1819,7 @@ static av_cold int init(AVFilterContext *ctx) · 4 unmodified lines
1494	1819	switch (out_channel_layout) {
1495	1820	case AV_CH_LAYOUT_7POINT1:
1496	1821	s->upmix_5_1 = upmix_7_1_5_1;
	1822	+ s->have_lfe = 1;
1497	1823	break;
1498	1824	default:
1499	1825	goto fail;
@@ -1504,6 +1830,7 @@ static av_cold int init(AVFilterContext *ctx) · 4 unmodified lines
1504	1830	switch (out_channel_layout) {
1505	1831	case AV_CH_LAYOUT_7POINT1:
1506	1832	s->upmix_5_1 = upmix_7_1_5_1;
	1833	+ s->have_lfe = 1;
1507	1834	break;
1508	1835	default:
1509	1836	goto fail;
@@ -1540,35 +1867,43 @@ fail: · 30 unmodified lines
1540	1867	return 0;
1541	1868	}
1542	1869
1543		−static int fft_channel(AVFilterContext ctx, void arg, int ch, int nb_jobs)
	1870	+static int fft_channel(AVFilterContext ctx, AVFrame in, int ch)
1544	1871	{
1545	1872	AudioSurroundContext *s = ctx->priv;
1546	1873	float src = (float )s->input_in->extended_data[ch];
1547	1874	float win = (float )s->window->extended_data[ch];
1548	1875	const int offset = s->buf_size - s->hop_size;
1549	1876	const float level_in = s->input_levels[ch];
1550		− AVFrame *in = arg;
1551	1877
1552	1878	memmove(src, &src[s->hop_size], offset * sizeof(float));
1553	1879	memcpy(&src[offset], in->extended_data[ch], in->nb_samples * sizeof(float));
1554	1880	memset(&src[offset + in->nb_samples], 0, (s->hop_size - in->nb_samples) * sizeof(float));
1555	1881
1556		− for (int n = 0; n < s->buf_size; n++) {
	1882	+ for (int n = 0; n < s->buf_size; n++)
1557	1883	win[n] = src[n] * s->window_func_lut[n] * level_in;
1558		− }
1559	1884
1560	1885	s->tx_fn(s->rdft[ch], (float *)s->input->extended_data[ch], win, sizeof(float));
1561	1886
1562	1887	return 0;
1563	1888	}
1564	1889
1565		−static int ifft_channel(AVFilterContext ctx, void arg, int ch, int nb_jobs)
	1890	+static int fft_channels(AVFilterContext ctx, void arg, int jobnr, int nb_jobs)
	1891	+{
	1892	+ AVFrame *in = arg;
	1893	+ const int start = (in->ch_layout.nb_channels * jobnr) / nb_jobs;
	1894	+ const int end = (in->ch_layout.nb_channels * (jobnr+1)) / nb_jobs;
	1895	+
	1896	+ for (int ch = start; ch < end; ch++)
	1897	+ fft_channel(ctx, in, ch);
	1898	+
	1899	+ return 0;
	1900	+}
	1901	+
	1902	+static int ifft_channel(AVFilterContext ctx, AVFrame out, int ch)
1566	1903	{
1567	1904	AudioSurroundContext *s = ctx->priv;
1568	1905	const float level_out = s->output_levels[ch];
1569		− AVFrame *out = arg;
1570	1906	float dst, ptr;
1571		− int n;
1572	1907
1573	1908	dst = (float *)s->output_out->extended_data[ch];
1574	1909	ptr = (float *)s->overlap_buffer->extended_data[ch];
@@ -1580,9 +1915,8 @@ static int ifft_channel(AVFilterContext ctx, void arg, int ch, int nb_jobs) · 5 unmodified lines
1580	1915	memset(s->overlap_buffer->extended_data[ch] + s->buf_size * sizeof(float),
1581	1916	0, s->hop_size * sizeof(float));
1582	1917
1583		− for (n = 0; n < s->buf_size; n++) {
	1918	+ for (int n = 0; n < s->buf_size; n++)
1584	1919	ptr[n] += dst[n] * s->window_func_lut[n] * level_out;
1585		− }
1586	1920
1587	1921	ptr = (float *)s->overlap_buffer->extended_data[ch];
1588	1922	dst = (float *)out->extended_data[ch];
@@ -1591,6 +1925,18 @@ static int ifft_channel(AVFilterContext ctx, void arg, int ch, int nb_jobs) · 2 unmodified lines
1591	1925	return 0;
1592	1926	}
1593	1927
	1928	+static int ifft_channels(AVFilterContext ctx, void arg, int jobnr, int nb_jobs)
	1929	+{
	1930	+ AVFrame *out = arg;
	1931	+ const int start = (out->ch_layout.nb_channels * jobnr) / nb_jobs;
	1932	+ const int end = (out->ch_layout.nb_channels * (jobnr+1)) / nb_jobs;
	1933	+
	1934	+ for (int ch = start; ch < end; ch++)
	1935	+ ifft_channel(ctx, out, ch);
	1936	+
	1937	+ return 0;
	1938	+}
	1939	+
1594	1940	static int filter_frame(AVFilterLink inlink, AVFrame in)
1595	1941	{
1596	1942	AVFilterContext *ctx = inlink->dst;
@@ -1598,15 +1944,20 @@ static int filter_frame(AVFilterLink inlink, AVFrame in) · 1 unmodified line
1598	1944	AudioSurroundContext *s = ctx->priv;
1599	1945	AVFrame *out;
1600	1946
1601		− ff_filter_execute(ctx, fft_channel, in, NULL, inlink->ch_layout.nb_channels);
	1947	+ ff_filter_execute(ctx, fft_channels, in, NULL,
	1948	+ FFMIN(inlink->ch_layout.nb_channels,
	1949	+ ff_filter_get_nb_threads(ctx)));
1602	1950
1603	1951	s->filter(ctx);
	1952	+ ff_filter_execute(ctx, s->upmix, NULL, NULL, outlink->ch_layout.nb_channels);
1604	1953
1605	1954	out = ff_get_audio_buffer(outlink, s->hop_size);
1606	1955	if (!out)
1607	1956	return AVERROR(ENOMEM);
1608	1957
1609		− ff_filter_execute(ctx, ifft_channel, out, NULL, outlink->ch_layout.nb_channels);
	1958	+ ff_filter_execute(ctx, ifft_channels, out, NULL,
	1959	+ FFMIN(outlink->ch_layout.nb_channels,
	1960	+ ff_filter_get_nb_threads(ctx)));
1610	1961
1611	1962	out->pts = in->pts;
1612	1963	out->nb_samples = in->nb_samples;
@@ -1670,6 +2021,14 @@ static av_cold void uninit(AVFilterContext *ctx) · 57 unmodified lines
1670	2021	av_freep(&s->rdft);
1671	2022	av_freep(&s->irdft);
1672	2023	av_freep(&s->window_func_lut);
	2024	+
	2025	+ av_freep(&s->x_pos);
	2026	+ av_freep(&s->y_pos);
	2027	+ av_freep(&s->l_phase);
	2028	+ av_freep(&s->r_phase);
	2029	+ av_freep(&s->c_phase);
	2030	+ av_freep(&s->mag_total);
	2031	+ av_freep(&s->lfe_mag);
1673	2032	}
1674	2033
1675	2034	#define OFFSET(x) offsetof(AudioSurroundContext, x)

avfilter/af_surround: speed up slice threading for stereo upmixing

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code