132 #define OFFSET(x) offsetof(ShowCWTContext, x)
133 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
207 for (
int n = 0; n <
s->nb_threads; n++)
213 for (
int n = 0; n <
s->nb_threads; n++)
219 for (
int n = 0; n <
s->frequency_band_count; n++)
257 int frequency_band_count,
258 float frequency_range,
259 float frequency_offset,
260 int frequency_scale,
float deviation)
265 for (
int y = 0; y < frequency_band_count; y++) {
266 float frequency = frequency_range * (1.f - (
float)y / frequency_band_count) + frequency_offset;
267 float frequency_derivative = frequency_range / frequency_band_count;
269 switch (frequency_scale) {
271 frequency =
powf(2.
f, frequency);
272 frequency_derivative *= logf(2.
f) * frequency;
275 frequency = 600.f * sinhf(frequency / 6.
f);
276 frequency_derivative *=
sqrtf(frequency * frequency + 360000.
f) / 6.f;
279 frequency = 700.f * (
powf(10.
f, frequency / 2595.
f) - 1.f);
280 frequency_derivative *= (frequency + 700.f) * logf(10.
f) / 2595.f;
283 frequency = 676170.4f / (47.06538f -
expf(frequency * 0.08950404
f)) - 14678.49
f;
284 frequency_derivative *= (frequency * frequency + 14990.4f * frequency + 4577850.f) / 160514.
f;
287 frequency = frequency * frequency;
288 frequency_derivative *= 2.f *
sqrtf(frequency);
291 frequency = frequency * frequency * frequency;
292 frequency_derivative *= 3.f *
powf(frequency, 2.
f / 3.
f);
295 frequency = frequency * frequency * frequency * frequency;
296 frequency_derivative *= 4.f *
powf(frequency, 3.
f / 4.
f);
299 frequency = 2.f *
powf(frequency, 3.
f / 2.
f) / 3.f;
300 frequency_derivative *=
sqrtf(frequency);
307 ret = 1.f / (frequency_derivative * deviation);
315 const float max =
s->maximum_intensity;
316 const float min =
s->minimum_intensity;
349 const int hop_size =
s->hop_size;
351 float *cache = (
float *)
s->cache->extended_data[ch];
354 const int offset = (
s->input_padding_size - hop_size) >> 1;
364 if (fin &&
s->hop_index + fin->
nb_samples < hop_size)
367 memset(
src, 0,
sizeof(
float) *
s->fft_size);
368 for (
int n = 0; n < hop_size; n++)
371 s->tx_fn(
s->fft[jobnr], dst,
src,
sizeof(*
src));
376 #define DRAW_BAR_COLOR(x) \
383 float mul = (Y - ht) * bh[0]; \
384 dstY[x] = av_clip_uint8(lrintf(Y * mul * 255.f)); \
385 dstU[x] = av_clip_uint8(lrintf((U-0.5f) * 128.f + 128)); \
386 dstV[x] = av_clip_uint8(lrintf((V-0.5f) * 128.f + 128)); \
391 float Y,
float U,
float V)
393 float *bh = ((
float *)
s->bh_out->extended_data[0]) + y;
394 const ptrdiff_t ylinesize =
s->outpicref->linesize[0];
395 const ptrdiff_t ulinesize =
s->outpicref->linesize[1];
396 const ptrdiff_t vlinesize =
s->outpicref->linesize[2];
397 const int direction =
s->direction;
398 const int sono_size =
s->sono_size;
399 const int bar_size =
s->bar_size;
400 const float rcp_bar_h = 1.f / bar_size;
401 uint8_t *dstY, *dstU, *dstV;
404 bh[0] = 1.f / (
Y + 0.0001f);
407 dstY =
s->outpicref->data[0] + y * ylinesize;
408 dstU =
s->outpicref->data[1] + y * ulinesize;
409 dstV =
s->outpicref->data[2] + y * vlinesize;
410 for (
int x = 0; x < bar_size; x++) {
411 float ht = (bar_size - x) * rcp_bar_h;
416 dstY =
s->outpicref->data[0] + y * ylinesize;
417 dstU =
s->outpicref->data[1] + y * ulinesize;
418 dstV =
s->outpicref->data[2] + y * vlinesize;
419 for (
int x = 0; x < bar_size; x++) {
420 float ht = x * rcp_bar_h;
425 dstY =
s->outpicref->data[0] +
w - 1 - y;
426 dstU =
s->outpicref->data[1] +
w - 1 - y;
427 dstV =
s->outpicref->data[2] +
w - 1 - y;
428 for (
int x = 0; x < bar_size; x++) {
429 float ht = (bar_size - x) * rcp_bar_h;
437 dstY =
s->outpicref->data[0] +
w - 1 - y + ylinesize * sono_size;
438 dstU =
s->outpicref->data[1] +
w - 1 - y + ulinesize * sono_size;
439 dstV =
s->outpicref->data[2] +
w - 1 - y + vlinesize * sono_size;
440 for (
int x = 0; x < bar_size; x++) {
441 float ht = x * rcp_bar_h;
454 const ptrdiff_t ylinesize =
s->outpicref->linesize[0];
455 const ptrdiff_t ulinesize =
s->outpicref->linesize[1];
456 const ptrdiff_t vlinesize =
s->outpicref->linesize[2];
457 const ptrdiff_t alinesize =
s->outpicref->linesize[3];
458 const float log_factor = 1.f/logf(
s->logarithmic_basis);
459 const int count =
s->frequency_band_count;
460 const int start = (count * jobnr) / nb_jobs;
461 const int end = (count * (jobnr+1)) / nb_jobs;
462 const int nb_channels =
s->nb_channels;
463 const int iscale =
s->intensity_scale;
464 const int ihop_index =
s->ihop_index;
465 const int ihop_size =
s->ihop_size;
466 const float rotation =
s->rotation;
467 const int direction =
s->direction;
468 uint8_t *dstY, *dstU, *dstV, *dstA;
469 const int sono_size =
s->sono_size;
470 const int bar_size =
s->bar_size;
471 const int mode =
s->mode;
472 const int w_1 =
s->w - 1;
473 const int x =
s->pos;
476 for (
int y = start; y < end; y++) {
478 0 * ihop_size + ihop_index;
486 dstY =
s->outpicref->data[0] + y * ylinesize;
487 dstU =
s->outpicref->data[1] + y * ulinesize;
488 dstV =
s->outpicref->data[2] + y * vlinesize;
489 dstA =
s->outpicref->data[3] ?
s->outpicref->data[3] + y * alinesize :
NULL;
493 dstY =
s->outpicref->data[0] + x * ylinesize + w_1 - y;
494 dstU =
s->outpicref->data[1] + x * ulinesize + w_1 - y;
495 dstV =
s->outpicref->data[2] + x * vlinesize + w_1 - y;
496 dstA =
s->outpicref->data[3] ?
s->outpicref->data[3] + x * alinesize + w_1 - y :
NULL;
506 switch (
s->direction) {
508 memmove(dstY, dstY + 1, w_1);
509 memmove(dstU, dstU + 1, w_1);
510 memmove(dstV, dstV + 1, w_1);
512 memmove(dstA, dstA + 1, w_1);
515 memmove(dstY + 1, dstY, w_1);
516 memmove(dstU + 1, dstU, w_1);
517 memmove(dstV + 1, dstV, w_1);
519 memmove(dstA + 1, dstA, w_1);
542 u = hypotf(
src[0].re,
src[0].im);
556 U = 0.5f + 0.5f * z *
u;
557 V = 0.5f + 0.5f * z * v;
573 const int nb_channels =
s->nb_channels;
574 const float yf = 1.f / nb_channels;
578 for (
int ch = 0; ch < nb_channels; ch++) {
582 z = hypotf(srcn[0].re, srcn[0].im);
586 U += z * yf *
sinf(2.
f *
M_PI * (ch * yf + rotation));
587 V += z * yf *
cosf(2.
f *
M_PI * (ch * yf + rotation));
603 Y = hypotf(
src[0].re,
src[0].im);
606 U = 0.5f + 0.5f *
U *
Y /
M_PI;
622 Y = 0.5f + 0.5f *
Y /
M_PI;
634 Y = hypotf(
src[0].re,
src[0].im);
655 const int ch = *(
int *)
arg;
659 const int output_padding_size =
s->output_padding_size;
660 const int input_padding_size =
s->input_padding_size;
661 const float scale = 1.f / input_padding_size;
662 const int ihop_size =
s->ihop_size;
663 const int count =
s->frequency_band_count;
664 const int start = (count * jobnr) / nb_jobs;
665 const int end = (count * (jobnr+1)) / nb_jobs;
667 for (
int y = start; y < end; y++) {
673 const unsigned *
index = (
const unsigned *)
s->index;
674 const int kernel_start =
s->kernel_start[y];
675 const int kernel_stop =
s->kernel_stop[y];
676 const int kernel_range = kernel_stop - kernel_start + 1;
679 if (kernel_start >= 0) {
681 memcpy(srcx, fft_out + kernel_start,
sizeof(*fft_out) * kernel_range);
684 memcpy(srcx+
offset, fft_out,
sizeof(*fft_out) * (kernel_range-
offset));
685 memcpy(srcx, fft_out+input_padding_size-
offset,
sizeof(*fft_out)*
offset);
688 s->fdsp->vector_fmul_scalar((
float *)srcx, (
const float *)srcx,
scale,
FFALIGN(kernel_range * 2, 4));
689 s->fdsp->vector_fmul((
float *)dstx, (
const float *)srcx,
690 (
const float *)kernel,
FFALIGN(kernel_range * 2, 16));
692 memset(isrc, 0,
sizeof(*isrc) * output_padding_size);
694 const unsigned *kindex =
index + kernel_start;
695 for (
int i = 0;
i < kernel_range;
i++) {
696 const unsigned n = kindex[
i];
698 isrc[n].re += dstx[
i].re;
699 isrc[n].im += dstx[
i].im;
702 for (
int i = 0;
i < kernel_range;
i++) {
703 const unsigned n = (
i-kernel_start) & (output_padding_size-1);
705 isrc[n].re += dstx[
i].re;
706 isrc[n].im += dstx[
i].im;
710 s->itx_fn(
s->ifft[jobnr], idst, isrc,
sizeof(*isrc));
712 memcpy(chout, idst,
sizeof(*chout) * ihop_size);
713 for (
int n = 0; n < ihop_size; n++) {
714 chout[n].
re += over[n].
re;
715 chout[n].
im += over[n].
im;
717 memcpy(over, idst + ihop_size,
sizeof(*over) * ihop_size);
726 const int size =
s->input_padding_size;
727 const int output_sample_count =
s->output_sample_count;
728 const int fsize =
s->frequency_band_count;
729 int *kernel_start =
s->kernel_start;
730 int *kernel_stop =
s->kernel_stop;
731 unsigned *
index =
s->index;
732 int range_min = INT_MAX;
733 int range_max = 0,
ret = 0;
740 for (
int y = 0; y <
fsize; y++) {
742 int start = INT_MIN, stop = INT_MAX;
743 const float frequency =
s->frequency_band[y*2];
744 const float deviation = 1.f / (
s->frequency_band[y*2+1] *
745 output_sample_count);
750 memset(tkernel, 0,
size *
sizeof(*tkernel));
751 for (
int n =
a; n <
b; n++) {
752 float ff,
f = n+0.5f-frequency;
754 ff =
expf(-
f*
f*deviation);
755 tkernel[n+
range] = ff;
758 for (
int n =
a; n <
b; n++) {
759 if (tkernel[n+
range] != 0.
f) {
760 if (tkernel[n+
range] > FLT_MIN)
767 for (
int n =
b; n >=
a; n--) {
768 if (tkernel[n+
range] != 0.
f) {
769 if (tkernel[n+
range] > FLT_MIN)
776 if (start == INT_MIN || stop == INT_MAX) {
781 kernel_start[y] = start;
782 kernel_stop[y] = stop;
790 for (
int n = 0; n <= stop - start; n++) {
791 kernel[n].
re = tkernel[n+
range+start];
792 kernel[n].
im = tkernel[n+
range+start];
795 range_min =
FFMIN(range_min, stop+1-start);
796 range_max =
FFMAX(range_max, stop+1-start);
798 s->kernel[y] = kernel;
801 for (
int n = 0; n <
size; n++)
802 index[n] = n & (
s->output_padding_size - 1);
817 const float limit_frequency =
inlink->sample_rate * 0.5f;
818 float maximum_frequency =
fminf(
s->maximum_frequency, limit_frequency);
819 float minimum_frequency =
s->minimum_frequency;
823 if (minimum_frequency >= maximum_frequency) {
825 minimum_frequency, maximum_frequency);
835 switch (
s->direction) {
838 s->bar_size =
s->w *
s->bar_ratio;
839 s->sono_size =
s->w -
s->bar_size;
840 s->frequency_band_count =
s->h;
844 s->bar_size =
s->h *
s->bar_ratio;
845 s->sono_size =
s->h -
s->bar_size;
846 s->frequency_band_count =
s->w;
850 switch (
s->frequency_scale) {
852 minimum_frequency = logf(minimum_frequency) / logf(2.
f);
853 maximum_frequency = logf(maximum_frequency) / logf(2.
f);
856 minimum_frequency = 6.f * asinhf(minimum_frequency / 600.
f);
857 maximum_frequency = 6.f * asinhf(maximum_frequency / 600.
f);
860 minimum_frequency = 2595.f *
log10f(1.
f + minimum_frequency / 700.
f);
861 maximum_frequency = 2595.f *
log10f(1.
f + maximum_frequency / 700.
f);
864 minimum_frequency = 11.17268f * logf(1.
f + (46.06538
f * minimum_frequency) / (minimum_frequency + 14678.49
f));
865 maximum_frequency = 11.17268f * logf(1.
f + (46.06538
f * maximum_frequency) / (maximum_frequency + 14678.49
f));
868 minimum_frequency =
sqrtf(minimum_frequency);
869 maximum_frequency =
sqrtf(maximum_frequency);
872 minimum_frequency =
cbrtf(minimum_frequency);
873 maximum_frequency =
cbrtf(maximum_frequency);
876 minimum_frequency =
powf(minimum_frequency, 0.25
f);
877 maximum_frequency =
powf(maximum_frequency, 0.25
f);
880 minimum_frequency =
powf(9.
f * (minimum_frequency * minimum_frequency) / 4.
f, 1.
f / 3.
f);
881 maximum_frequency =
powf(9.
f * (maximum_frequency * maximum_frequency) / 4.
f, 1.
f / 3.
f);
885 s->frequency_band =
av_calloc(
s->frequency_band_count,
886 sizeof(*
s->frequency_band) * 2);
887 if (!
s->frequency_band)
890 s->nb_consumed_samples =
inlink->sample_rate *
892 s->frequency_band_count, maximum_frequency - minimum_frequency,
893 minimum_frequency,
s->frequency_scale,
s->deviation);
894 s->nb_consumed_samples =
FFMIN(
s->nb_consumed_samples, 65536);
897 s->nb_channels =
inlink->ch_layout.nb_channels;
901 s->input_sample_count = 1 << (32 -
ff_clz(
s->nb_consumed_samples));
902 s->input_padding_size = 1 << (32 -
ff_clz(
s->input_sample_count));
904 s->output_padding_size = 1 << (32 -
ff_clz(
s->output_sample_count));
906 s->hop_size =
s->input_sample_count;
907 s->ihop_size =
s->output_padding_size >> 1;
920 for (
int n = 0; n <
s->nb_threads; n++) {
930 for (
int n = 0; n <
s->nb_threads; n++) {
941 s->kernel =
av_calloc(
s->frequency_band_count,
sizeof(*
s->kernel));
948 s->index =
av_calloc(
s->input_padding_size,
sizeof(*
s->index));
949 s->kernel_start =
av_calloc(
s->frequency_band_count,
sizeof(*
s->kernel_start));
950 s->kernel_stop =
av_calloc(
s->frequency_band_count,
sizeof(*
s->kernel_stop));
951 if (!
s->outpicref || !
s->fft_in || !
s->fft_out || !
s->src_x || !
s->dst_x || !
s->over ||
952 !
s->ifft_in || !
s->ifft_out || !
s->kernel_start || !
s->kernel_stop || !
s->ch_out ||
953 !
s->cache || !
s->index || !
s->bh_out || !
s->kernel)
956 s->ch_out->format =
inlink->format;
957 s->ch_out->nb_samples = 2 *
s->ihop_size *
inlink->ch_layout.nb_channels;
958 s->ch_out->ch_layout.nb_channels =
s->frequency_band_count;
963 s->ifft_in->format =
inlink->format;
964 s->ifft_in->nb_samples =
s->ifft_size * 2;
965 s->ifft_in->ch_layout.nb_channels =
s->nb_threads;
970 s->ifft_out->format =
inlink->format;
971 s->ifft_out->nb_samples =
s->ifft_size * 2;
972 s->ifft_out->ch_layout.nb_channels =
s->nb_threads;
977 s->src_x->format =
inlink->format;
978 s->src_x->nb_samples =
s->fft_size * 2;
979 s->src_x->ch_layout.nb_channels =
s->nb_threads;
984 s->dst_x->format =
inlink->format;
985 s->dst_x->nb_samples =
s->fft_size * 2;
986 s->dst_x->ch_layout.nb_channels =
s->nb_threads;
991 s->outpicref->sample_aspect_ratio = (
AVRational){1,1};
993 for (
int y = 0; y < outlink->
h; y++) {
994 memset(
s->outpicref->data[0] + y *
s->outpicref->linesize[0], 0, outlink->
w);
995 memset(
s->outpicref->data[1] + y *
s->outpicref->linesize[1], 128, outlink->
w);
996 memset(
s->outpicref->data[2] + y *
s->outpicref->linesize[2], 128, outlink->
w);
997 if (
s->outpicref->data[3])
998 memset(
s->outpicref->data[3] + y *
s->outpicref->linesize[3], 0, outlink->
w);
1004 for (
int n = 0; n <
s->frequency_band_count; n++) {
1005 s->frequency_band[2*n ] *=
factor;
1006 s->frequency_band[2*n+1] *=
factor;
1018 switch (
s->direction) {
1021 s->pos =
s->bar_size;
1025 s->pos =
s->sono_size;
1030 if (strcmp(
s->rate_str,
"auto")) {
1035 s->frame_rate =
s->auto_frame_rate;
1052 const int nb_planes = 3 + (
s->outpicref->data[3] !=
NULL);
1057 switch (
s->direction) {
1059 for (
int p = 0; p < nb_planes; p++) {
1060 ptrdiff_t linesize =
s->outpicref->linesize[p];
1062 for (
int y =
s->h - 1; y >
s->bar_size; y--) {
1063 uint8_t *dst =
s->outpicref->data[p] + y * linesize;
1065 memmove(dst, dst - linesize,
s->w);
1070 for (
int p = 0; p < nb_planes; p++) {
1071 ptrdiff_t linesize =
s->outpicref->linesize[p];
1073 for (
int y = 0; y <
s->sono_size; y++) {
1074 uint8_t *dst =
s->outpicref->data[p] + y * linesize;
1076 memmove(dst, dst + linesize,
s->w);
1089 switch (
s->direction) {
1092 if (
s->pos >=
s->w) {
1093 s->pos =
s->bar_size;
1100 s->pos =
s->sono_size;
1106 if (
s->pos >=
s->h) {
1107 s->pos =
s->bar_size;
1114 s->pos =
s->sono_size;
1121 switch (
s->direction) {
1124 s->pos =
s->bar_size;
1128 s->pos =
s->sono_size;
1135 switch (
s->direction) {
1137 for (
int p = 0; p < nb_planes; p++) {
1138 ptrdiff_t linesize =
s->outpicref->linesize[p];
1139 const int size =
s->w -
s->pos;
1140 const int fill = p > 0 && p < 3 ? 128 : 0;
1141 const int x =
s->pos;
1143 for (
int y = 0; y <
s->h; y++) {
1144 uint8_t *dst =
s->outpicref->data[p] + y * linesize + x;
1146 memset(dst, fill,
size);
1151 for (
int p = 0; p < nb_planes; p++) {
1152 ptrdiff_t linesize =
s->outpicref->linesize[p];
1153 const int size =
s->w -
s->pos;
1154 const int fill = p > 0 && p < 3 ? 128 : 0;
1156 for (
int y = 0; y <
s->h; y++) {
1157 uint8_t *dst =
s->outpicref->data[p] + y * linesize;
1159 memset(dst, fill,
size);
1164 for (
int p = 0; p < nb_planes; p++) {
1165 ptrdiff_t linesize =
s->outpicref->linesize[p];
1166 const int fill = p > 0 && p < 3 ? 128 : 0;
1168 for (
int y =
s->pos; y < s->
h; y++) {
1169 uint8_t *dst =
s->outpicref->data[p] + y * linesize;
1171 memset(dst, fill,
s->w);
1176 for (
int p = 0; p < nb_planes; p++) {
1177 ptrdiff_t linesize =
s->outpicref->linesize[p];
1178 const int fill = p > 0 && p < 3 ? 128 : 0;
1180 for (
int y =
s->h -
s->pos; y >= 0; y--) {
1181 uint8_t *dst =
s->outpicref->data[p] + y * linesize;
1183 memset(dst, fill,
s->w);
1194 const int offset = (
s->input_padding_size -
s->hop_size) >> 1;
1198 s->outpicref->duration = 1;
1202 if (
s->ihop_index >=
s->ihop_size)
1203 s->ihop_index =
s->hop_index = 0;
1208 if (
s->old_pts <
s->outpicref->pts) {
1218 s->old_pts =
s->outpicref->pts;
1234 const int count =
s->nb_channels;
1235 const int start = (count * jobnr) / nb_jobs;
1236 const int end = (count * (jobnr+1)) / nb_jobs;
1238 for (
int ch = start; ch < end; ch++)
1257 if (
s->hop_index <
s->hop_size) {
1264 if (
ret > 0 ||
s->eof) {
1266 FFMIN(
s->nb_threads,
s->nb_channels));
1268 if (
s->hop_index == 0) {
1269 s->in_pts = fin->
pts;
1276 s->hop_index =
s->hop_size;
1281 if (
s->hop_index >=
s->hop_size ||
s->ihop_index > 0) {
1282 for (
int ch = 0; ch <
s->nb_channels &&
s->ihop_index == 0; ch++) {
1310 s->hop_index >=
s->hop_size ||
s->eof) {
1333 .description =
NULL_IF_CONFIG_SMALL(
"Convert input audio to a CWT (Continuous Wavelet Transform) spectrum video output."),
1340 .priv_class = &showcwt_class,