static int decode_packet(struct dec_audio *da, struct demux_packet *mpkt,
struct mp_audio **out)
{
struct spdifContext *spdif_ctx = da->priv;
spdif_ctx->out_buffer_len = 0;
if (!mpkt)
return 0;
double pts = mpkt->pts;
AVPacket pkt;
mp_set_av_packet(&pkt, mpkt, NULL);
mpkt->len = 0; // will be fully consumed
pkt.pts = pkt.dts = 0;
if (!spdif_ctx->lavf_ctx) {
if (init_filter(da, &pkt) < 0)
return -1;
}
int ret = av_write_frame(spdif_ctx->lavf_ctx, &pkt);
avio_flush(spdif_ctx->lavf_ctx->pb);
if (ret < 0)
return -1;
int samples = spdif_ctx->out_buffer_len / spdif_ctx->fmt.sstride;
*out = mp_audio_pool_get(spdif_ctx->pool, &spdif_ctx->fmt, samples);
if (!*out)
return -1;
memcpy((*out)->planes[0], spdif_ctx->out_buffer, spdif_ctx->out_buffer_len);
(*out)->pts = pts;
return 0;
}
static int filter_out(struct af_instance *af)
{
struct priv *p = af->priv;
while (rubberband_available(p->rubber) <= 0) {
const float *dummy[MP_NUM_CHANNELS] = {0};
const float **in_data = dummy;
size_t in_samples = 0;
if (p->pending) {
if (!p->pending->samples)
break;
// recover from previous EOF
if (p->needs_reset) {
rubberband_reset(p->rubber);
p->rubber_delay = 0;
}
p->needs_reset = false;
size_t needs = rubberband_get_samples_required(p->rubber);
in_data = (void *)&p->pending->planes;
in_samples = MPMIN(p->pending->samples, needs);
}
if (p->needs_reset)
break; // previous EOF
p->needs_reset = !p->pending; // EOF
rubberband_process(p->rubber, in_data, in_samples, p->needs_reset);
p->rubber_delay += in_samples;
if (!p->pending)
break;
mp_audio_skip_samples(p->pending, in_samples);
}
int out_samples = rubberband_available(p->rubber);
if (out_samples > 0) {
struct mp_audio *out =
mp_audio_pool_get(af->out_pool, af->data, out_samples);
if (!out)
return -1;
if (p->pending)
mp_audio_copy_config(out, p->pending);
float **out_data = (void *)&out->planes;
out->samples = rubberband_retrieve(p->rubber, out_data, out->samples);
p->rubber_delay -= out->samples * p->speed;
af_add_output_frame(af, out);
}
int delay_samples = p->rubber_delay;
if (p->pending)
delay_samples += p->pending->samples;
af->delay = delay_samples / (af->data->rate * p->speed);
return 0;
}
static int filter_frame(struct af_instance *af, struct mp_audio *c)
{
if (!c)
return 0;
struct mp_audio *l = mp_audio_pool_get(af->out_pool, &af->fmt_out, c->samples);
if (!l) {
talloc_free(c);
return -1;
}
mp_audio_copy_attributes(l, c);
af_pan_t* s = af->priv; // Setup for this instance
float* in = c->planes[0]; // Input audio data
float* out = NULL; // Output audio data
float* end = in+c->samples*c->nch; // End of loop
int nchi = c->nch; // Number of input channels
int ncho = l->nch; // Number of output channels
register int j,k;
out = l->planes[0];
// Execute panning
// FIXME: Too slow
while(in < end) {
for(j=0; j<ncho; j++) {
register float x = 0.0;
register float* tin = in;
for(k=0; k<nchi; k++)
x += tin[k] * s->level[j][k];
out[j] = x;
}
out+= ncho;
in+= nchi;
}
talloc_free(c);
af_add_output_frame(af, l);
return 0;
}
static int filter(struct af_instance *af, struct mp_audio *data)
{
if (!data)
return 0;
struct mp_audio *out =
mp_audio_pool_get(af->out_pool, af->data, data->samples);
if (!out) {
talloc_free(data);
return -1;
}
mp_audio_copy_attributes(out, data);
size_t len = mp_audio_psize(data) / data->bps;
if (data->bps == 4) {
for (int s = 0; s < len; s++) {
uint32_t val = *((uint32_t *)data->planes[0] + s);
uint8_t *ptr = (uint8_t *)out->planes[0] + s * 3;
ptr[0] = val >> SHIFT(0);
ptr[1] = val >> SHIFT(1);
ptr[2] = val >> SHIFT(2);
}
} else {
for (int s = 0; s < len; s++) {
static int decode_packet(struct dec_audio *da, struct mp_audio **out)
{
struct spdifContext *spdif_ctx = da->priv;
AVFormatContext *lavf_ctx = spdif_ctx->lavf_ctx;
spdif_ctx->out_buffer_len = 0;
struct demux_packet *mpkt;
if (demux_read_packet_async(da->header, &mpkt) == 0)
return AD_WAIT;
if (!mpkt)
return AD_EOF;
AVPacket pkt;
mp_set_av_packet(&pkt, mpkt, NULL);
pkt.pts = pkt.dts = 0;
if (mpkt->pts != MP_NOPTS_VALUE) {
da->pts = mpkt->pts;
da->pts_offset = 0;
}
int ret = av_write_frame(lavf_ctx, &pkt);
talloc_free(mpkt);
avio_flush(lavf_ctx->pb);
if (ret < 0)
return AD_ERR;
int samples = spdif_ctx->out_buffer_len / spdif_ctx->fmt.sstride;
*out = mp_audio_pool_get(da->pool, &spdif_ctx->fmt, samples);
if (!*out)
return AD_ERR;
memcpy((*out)->planes[0], spdif_ctx->out_buffer, spdif_ctx->out_buffer_len);
return 0;
}
static int filter_out(struct af_instance *af)
{
af_ac3enc_t *s = af->priv;
if (!fill_buffer(af))
return 0; // need more input
AVFrame *frame = av_frame_alloc();
if (!frame) {
MP_FATAL(af, "Could not allocate memory \n");
return -1;
}
frame->nb_samples = s->in_samples;
frame->format = s->lavc_actx->sample_fmt;
frame->channel_layout = s->lavc_actx->channel_layout;
assert(s->input->num_planes <= AV_NUM_DATA_POINTERS);
frame->extended_data = frame->data;
for (int n = 0; n < s->input->num_planes; n++)
frame->data[n] = s->input->planes[n];
frame->linesize[0] = s->input->samples * s->input->sstride;
int ok;
int lavc_ret = avcodec_encode_audio2(s->lavc_actx, &s->pkt, frame, &ok);
av_frame_free(&frame);
s->input->samples = 0;
if (lavc_ret < 0 || !ok) {
MP_FATAL(af, "Encode failed.\n");
return -1;
}
MP_DBG(af, "avcodec_encode_audio got %d, pending %d.\n",
s->pkt.size, s->pending->samples);
struct mp_audio *out =
mp_audio_pool_get(af->out_pool, af->data, s->out_samples);
if (!out)
return -1;
mp_audio_copy_attributes(out, s->pending);
int frame_size = s->pkt.size;
int header_len = 0;
char hdr[8];
if (s->cfg_add_iec61937_header && s->pkt.size > 5) {
int bsmod = s->pkt.data[5] & 0x7;
int len = frame_size;
frame_size = AC3_FRAME_SIZE * 2 * 2;
header_len = 8;
AV_WL16(hdr, 0xF872); // iec 61937 syncword 1
AV_WL16(hdr + 2, 0x4E1F); // iec 61937 syncword 2
hdr[5] = bsmod; // bsmod
hdr[4] = 0x01; // data-type ac3
AV_WL16(hdr + 6, len << 3); // number of bits in payload
}
if (frame_size > out->samples * out->sstride)
abort();
char *buf = (char *)out->planes[0];
memcpy(buf, hdr, header_len);
memcpy(buf + header_len, s->pkt.data, s->pkt.size);
memset(buf + header_len + s->pkt.size, 0,
frame_size - (header_len + s->pkt.size));
swap_16((uint16_t *)(buf + header_len), s->pkt.size / 2);
out->samples = frame_size / out->sstride;
af_add_output_frame(af, out);
update_delay(af);
return 0;
}
static int filter_out(struct af_instance *af)
{
af_ac3enc_t *s = af->priv;
if (!s->pending)
return 0;
AVFrame *frame = av_frame_alloc();
if (!frame) {
MP_FATAL(af, "Could not allocate memory \n");
return -1;
}
int err = -1;
AVPacket pkt = {0};
av_init_packet(&pkt);
#if HAVE_AVCODEC_NEW_CODEC_API
// Send input as long as it wants.
while (1) {
err = read_input_frame(af, frame);
if (err < 0)
goto done;
if (err == 0)
break;
err = -1;
int lavc_ret = avcodec_send_frame(s->lavc_actx, frame);
// On EAGAIN, we're supposed to read remaining output.
if (lavc_ret == AVERROR(EAGAIN))
break;
if (lavc_ret < 0) {
MP_FATAL(af, "Encode failed.\n");
goto done;
}
s->encoder_buffered += s->input->samples;
s->input->samples = 0;
}
int lavc_ret = avcodec_receive_packet(s->lavc_actx, &pkt);
if (lavc_ret == AVERROR(EAGAIN)) {
// Need to buffer more input.
err = 0;
goto done;
}
if (lavc_ret < 0) {
MP_FATAL(af, "Encode failed.\n");
goto done;
}
#else
err = read_input_frame(af, frame);
if (err < 0)
goto done;
if (err == 0)
goto done;
err = -1;
int ok;
int lavc_ret = avcodec_encode_audio2(s->lavc_actx, &pkt, frame, &ok);
s->input->samples = 0;
if (lavc_ret < 0 || !ok) {
MP_FATAL(af, "Encode failed.\n");
goto done;
}
#endif
MP_DBG(af, "avcodec_encode_audio got %d, pending %d.\n",
pkt.size, s->pending->samples + s->input->samples);
s->encoder_buffered -= AC3_FRAME_SIZE;
struct mp_audio *out =
mp_audio_pool_get(af->out_pool, af->data, s->out_samples);
if (!out)
goto done;
mp_audio_copy_attributes(out, s->pending);
int frame_size = pkt.size;
int header_len = 0;
char hdr[8];
if (s->cfg_add_iec61937_header && pkt.size > 5) {
int bsmod = pkt.data[5] & 0x7;
int len = frame_size;
frame_size = AC3_FRAME_SIZE * 2 * 2;
header_len = 8;
AV_WL16(hdr, 0xF872); // iec 61937 syncword 1
AV_WL16(hdr + 2, 0x4E1F); // iec 61937 syncword 2
hdr[5] = bsmod; // bsmod
hdr[4] = 0x01; // data-type ac3
AV_WL16(hdr + 6, len << 3); // number of bits in payload
}
if (frame_size > out->samples * out->sstride)
abort();
char *buf = (char *)out->planes[0];
memcpy(buf, hdr, header_len);
memcpy(buf + header_len, pkt.data, pkt.size);
memset(buf + header_len + pkt.size, 0,
frame_size - (header_len + pkt.size));
swap_16((uint16_t *)(buf + header_len), pkt.size / 2);
out->samples = frame_size / out->sstride;
af_add_output_frame(af, out);
err = 0;
done:
av_packet_unref(&pkt);
av_frame_free(&frame);
update_delay(af);
return err;
}
static int filter_frame(struct af_instance *af, struct mp_audio *data)
{
if (!data)
return 0;
struct mp_audio *outframe =
mp_audio_pool_get(af->out_pool, &af->fmt_out, data->samples);
if (!outframe) {
talloc_free(data);
return -1;
}
mp_audio_copy_attributes(outframe, data);
af_surround_t* s = (af_surround_t*)af->priv;
const float* m = steering_matrix[0];
float* in = data->planes[0]; // Input audio data
float* out = outframe->planes[0]; // Output audio data
float* end = in + data->samples * data->nch;
int i = s->i; // Filter queue index
int ri = s->ri; // Read index for delay queue
int wi = s->wi; // Write index for delay queue
while(in < end){
/* Dominance:
abs(in[0]) abs(in[1]);
abs(in[0]+in[1]) abs(in[0]-in[1]);
10 * log( abs(in[0]) / (abs(in[1])|1) );
10 * log( abs(in[0]+in[1]) / (abs(in[0]-in[1])|1) ); */
/* About volume balancing...
Surround encoding does the following:
Lt=L+.707*C+.707*S, Rt=R+.707*C-.707*S
So S should be extracted as:
(Lt-Rt)
But we are splitting the S to two output channels, so we
must take 3dB off as we split it:
Ls=Rs=.707*(Lt-Rt)
Trouble is, Lt could be +1, Rt -1, so possibility that S will
overflow. So to avoid that, we cut L/R by 3dB (*.707), and S by
6dB (/2). This keeps the overall balance, but guarantees no
overflow. */
// Output front left and right
out[0] = m[0]*in[0] + m[1]*in[1];
out[1] = m[2]*in[0] + m[3]*in[1];
// Low-pass output @ 7kHz
FIR((&s->lq[i]), s->w, s->dl[wi]);
// Delay output by d ms
out[2] = s->dl[ri];
#ifdef SPLITREAR
// Low-pass output @ 7kHz
FIR((&s->rq[i]), s->w, s->dr[wi]);
// Delay output by d ms
out[3] = s->dr[ri];
#else
out[3] = -out[2];
#endif
// Update delay queues indexes
UPDATEQI(ri);
UPDATEQI(wi);
// Calculate and save surround in circular queue
#ifdef SPLITREAR
ADDQUE(i, s->rq, s->lq, m[6]*in[0]+m[7]*in[1], m[8]*in[0]+m[9]*in[1]);
#else
ADDQUE(i, s->lq, m[4]*in[0]+m[5]*in[1]);
#endif
// Next sample...
in = &in[data->nch];
out = &out[af->data->nch];
}
// Save indexes
s->i = i; s->ri = ri; s->wi = wi;
talloc_free(data);
af_add_output_frame(af, outframe);
return 0;
}
static int filter(struct af_instance *af, struct mp_audio *data)
{
af_scaletempo_t *s = af->priv;
if (s->scale == 1.0) {
af->delay = 0;
af_add_output_frame(af, data);
return 0;
}
int in_samples = data ? data->samples : 0;
struct mp_audio *out = mp_audio_pool_get(af->out_pool, af->data,
((int)(in_samples / s->frames_stride_scaled) + 1) * s->frames_stride);
if (!out) {
talloc_free(data);
return -1;
}
if (data)
mp_audio_copy_attributes(out, data);
int offset_in = fill_queue(af, data, 0);
int8_t *pout = out->planes[0];
while (s->bytes_queued >= s->bytes_queue) {
int ti;
float tf;
int bytes_off = 0;
// output stride
if (s->output_overlap) {
if (s->best_overlap_offset)
bytes_off = s->best_overlap_offset(s);
s->output_overlap(s, pout, bytes_off);
}
memcpy(pout + s->bytes_overlap,
s->buf_queue + bytes_off + s->bytes_overlap,
s->bytes_standing);
pout += s->bytes_stride;
// input stride
memcpy(s->buf_overlap,
s->buf_queue + bytes_off + s->bytes_stride,
s->bytes_overlap);
tf = s->frames_stride_scaled + s->frames_stride_error;
ti = (int)tf;
s->frames_stride_error = tf - ti;
s->bytes_to_slide = ti * s->bytes_per_frame;
offset_in += fill_queue(af, data, offset_in);
}
// This filter can have a negative delay when scale > 1:
// output corresponding to some length of input can be decided and written
// after receiving only a part of that input.
af->delay = (s->bytes_queued - s->bytes_to_slide) / s->scale
/ out->sstride / out->rate;
out->samples = (pout - (int8_t *)out->planes[0]) / out->sstride;
talloc_free(data);
if (out->samples) {
af_add_output_frame(af, out);
} else {
talloc_free(out);
}
return 0;
}
/* Filter data through filter
Two "tricks" are used to compensate the "color" of the KEMAR data:
1. The KEMAR data is refiltered to ensure that the front L, R channels
on the same side of the ear are equalized (especially in the high
frequencies).
2. A bass compensation is introduced to ensure that 0-200 Hz are not
damped (without any real 3D acoustical image, however).
*/
static int filter(struct af_instance *af, struct mp_audio *data)
{
af_hrtf_t *s = af->priv;
if (!data)
return 0;
struct mp_audio *outframe =
mp_audio_pool_get(af->out_pool, &af->fmt_out, data->samples);
if (!outframe) {
talloc_free(data);
return -1;
}
mp_audio_copy_attributes(outframe, data);
short *in = data->planes[0]; // Input audio data
short *out = outframe->planes[0]; // Output audio data
short *end = in + data->samples * data->nch; // Loop end
float common, left, right, diff, left_b, right_b;
const int dblen = s->dlbuflen, hlen = s->hrflen, blen = s->basslen;
if(s->print_flag) {
s->print_flag = 0;
switch (s->decode_mode) {
case HRTF_MIX_51:
MP_INFO(af, "Using HRTF to mix %s discrete surround into "
"L, R channels\n", s->matrix_mode ? "5+1" : "5");
break;
case HRTF_MIX_STEREO:
MP_INFO(af, "Using HRTF to mix stereo into "
"L, R channels\n");
break;
case HRTF_MIX_MATRIX2CH:
MP_INFO(af, "Using active matrix to decode 2 channel "
"input, HRTF to mix %s matrix surround into "
"L, R channels\n", "3/2");
break;
default:
MP_WARN(af, "bogus decode_mode: %d\n", s->decode_mode);
break;
}
if(s->matrix_mode)
MP_INFO(af, "Using active matrix to decode rear center "
"channel\n");
}
/* MPlayer's 5 channel layout (notation for the variable):
*
* 0: L (LF), 1: R (RF), 2: Ls (LR), 3: Rs (RR), 4: C (CF), matrix
* encoded: Cs (CR)
*
* or: L = left, C = center, R = right, F = front, R = rear
*
* Filter notation:
*
* CF
* OF AF
* Ear->
* OR AR
* CR
*
* or: C = center, A = same side, O = opposite, F = front, R = rear
*/
while(in < end) {
const int k = s->cyc_pos;
update_ch(s, in, k);
/* Simulate a 7.5 ms -20 dB echo of the center channel in the
front channels (like reflection from a room wall) - a kind of
psycho-acoustically "cheating" to focus the center front
channel, which is normally hard to be perceived as front */
s->lf[k] += CFECHOAMPL * s->cf[(k + CFECHODELAY) % s->dlbuflen];
s->rf[k] += CFECHOAMPL * s->cf[(k + CFECHODELAY) % s->dlbuflen];
switch (s->decode_mode) {
case HRTF_MIX_51:
case HRTF_MIX_MATRIX2CH:
/* Mixer filter matrix */
common = conv(dblen, hlen, s->cf, s->cf_ir, k + s->cf_o);
if(s->matrix_mode) {
/* In matrix decoding mode, the rear channel gain must be
renormalized, as there is an additional channel. */
matrix_decode(in, k, 2, 3, 0, s->dlbuflen,
s->lr_fwr, s->rr_fwr,
s->lrprr_fwr, s->lrmrr_fwr,
&(s->adapt_lr_gain), &(s->adapt_rr_gain),
&(s->adapt_lrprr_gain), &(s->adapt_lrmrr_gain),
s->lr, s->rr, NULL, NULL, s->cr);
common +=
conv(dblen, hlen, s->cr, s->cr_ir, k + s->cr_o) *
M1_76DB;
left =
( conv(dblen, hlen, s->lf, s->af_ir, k + s->af_o) +
conv(dblen, hlen, s->rf, s->of_ir, k + s->of_o) +
(conv(dblen, hlen, s->lr, s->ar_ir, k + s->ar_o) +
conv(dblen, hlen, s->rr, s->or_ir, k + s->or_o)) *
M1_76DB + common);
right =
( conv(dblen, hlen, s->rf, s->af_ir, k + s->af_o) +
conv(dblen, hlen, s->lf, s->of_ir, k + s->of_o) +
(conv(dblen, hlen, s->rr, s->ar_ir, k + s->ar_o) +
conv(dblen, hlen, s->lr, s->or_ir, k + s->or_o)) *
M1_76DB + common);
} else {
left =
( conv(dblen, hlen, s->lf, s->af_ir, k + s->af_o) +
conv(dblen, hlen, s->rf, s->of_ir, k + s->of_o) +
conv(dblen, hlen, s->lr, s->ar_ir, k + s->ar_o) +
conv(dblen, hlen, s->rr, s->or_ir, k + s->or_o) +
common);
right =
( conv(dblen, hlen, s->rf, s->af_ir, k + s->af_o) +
conv(dblen, hlen, s->lf, s->of_ir, k + s->of_o) +
conv(dblen, hlen, s->rr, s->ar_ir, k + s->ar_o) +
conv(dblen, hlen, s->lr, s->or_ir, k + s->or_o) +
common);
}
break;
case HRTF_MIX_STEREO:
left =
( conv(dblen, hlen, s->lf, s->af_ir, k + s->af_o) +
conv(dblen, hlen, s->rf, s->of_ir, k + s->of_o));
right =
( conv(dblen, hlen, s->rf, s->af_ir, k + s->af_o) +
conv(dblen, hlen, s->lf, s->of_ir, k + s->of_o));
break;
default:
/* make gcc happy */
left = 0.0;
right = 0.0;
break;
}
/* Bass compensation for the lower frequency cut of the HRTF. A
cross talk of the left and right channel is introduced to
match the directional characteristics of higher frequencies.
The bass will not have any real 3D perception, but that is
OK (note at 180 Hz, the wavelength is about 2 m, and any
spatial perception is impossible). */
left_b = conv(dblen, blen, s->ba_l, s->ba_ir, k);
right_b = conv(dblen, blen, s->ba_r, s->ba_ir, k);
left += (1 - BASSCROSS) * left_b + BASSCROSS * right_b;
right += (1 - BASSCROSS) * right_b + BASSCROSS * left_b;
/* Also mix the LFE channel (if available) */
if(data->nch >= 6) {
left += in[5] * M3_01DB;
right += in[5] * M3_01DB;
}
/* Amplitude renormalization. */
left *= AMPLNORM;
right *= AMPLNORM;
switch (s->decode_mode) {
case HRTF_MIX_51:
case HRTF_MIX_STEREO:
/* "Cheating": linear stereo expansion to amplify the 3D
perception. Note: Too much will destroy the acoustic space
and may even result in headaches. */
diff = STEXPAND2 * (left - right);
out[0] = av_clip_int16(left + diff);
out[1] = av_clip_int16(right - diff);
break;
case HRTF_MIX_MATRIX2CH:
/* Do attempt any stereo expansion with matrix encoded
sources. The L, R channels are already stereo expanded
by the steering, any further stereo expansion will sound
very unnatural. */
out[0] = av_clip_int16(left);
out[1] = av_clip_int16(right);
break;
}
/* Next sample... */
in = &in[data->nch];
out = &out[af->data->nch];
(s->cyc_pos)--;
if(s->cyc_pos < 0)
s->cyc_pos += dblen;
}
talloc_free(data);
af_add_output_frame(af, outframe);
return 0;
}