static inline double getpix(void *priv, double x, double y, int plane)
{
int xi, yi;
GEQContext *geq = priv;
AVFrame *picref = geq->picref;
const uint8_t *src = picref->data[plane];
int linesize = picref->linesize[plane];
const int w = (plane == 1 || plane == 2) ? AV_CEIL_RSHIFT(picref->width, geq->hsub) : picref->width;
const int h = (plane == 1 || plane == 2) ? AV_CEIL_RSHIFT(picref->height, geq->vsub) : picref->height;
if (!src)
return 0;
xi = x = av_clipf(x, 0, w - 2);
yi = y = av_clipf(y, 0, h - 2);
x -= xi;
y -= yi;
if (geq->bps > 8) {
const uint16_t *src16 = (const uint16_t*)src;
linesize /= 2;
return (1-y)*((1-x)*src16[xi + yi * linesize] + x*src16[xi + 1 + yi * linesize])
+ y *((1-x)*src16[xi + (yi+1) * linesize] + x*src16[xi + 1 + (yi+1) * linesize]);
} else {
return (1-y)*((1-x)*src[xi + yi * linesize] + x*src[xi + 1 + yi * linesize])
+ y *((1-x)*src[xi + (yi+1) * linesize] + x*src[xi + 1 + (yi+1) * linesize]);
}
}
static int control(struct vo *vo, uint32_t request, void *data)
{
struct xvctx *ctx = vo->priv;
switch (request) {
case VOCTRL_GET_PANSCAN:
return VO_TRUE;
case VOCTRL_SET_PANSCAN:
resize(vo);
return VO_TRUE;
case VOCTRL_SET_EQUALIZER: {
vo->want_redraw = true;
struct voctrl_set_equalizer_args *args = data;
return xv_set_eq(vo, ctx->xv_port, args->name, args->value);
}
case VOCTRL_GET_EQUALIZER: {
struct voctrl_get_equalizer_args *args = data;
return xv_get_eq(vo, ctx->xv_port, args->name, args->valueptr);
}
case VOCTRL_SET_YUV_COLORSPACE:;
struct mp_csp_details* given_cspc = data;
int is_709 = given_cspc->format == MP_CSP_BT_709;
xv_set_eq(vo, ctx->xv_port, "bt_709", is_709 * 200 - 100);
read_xv_csp(vo);
vo->want_redraw = true;
return true;
case VOCTRL_GET_YUV_COLORSPACE:;
struct mp_csp_details* cspc = data;
read_xv_csp(vo);
*cspc = ctx->cached_csp;
return true;
case VOCTRL_REDRAW_FRAME:
redraw_frame(vo);
return true;
case VOCTRL_SCREENSHOT: {
struct voctrl_screenshot_args *args = data;
args->out_image = get_screenshot(vo);
return true;
}
case VOCTRL_WINDOW_TO_OSD_COORDS: {
float *c = data;
struct mp_rect *src = &ctx->src_rect;
struct mp_rect *dst = &ctx->dst_rect;
c[0] = av_clipf(c[0], dst->x0, dst->x1) - dst->x0;
c[1] = av_clipf(c[1], dst->y0, dst->y1) - dst->y0;
c[0] = c[0] / (dst->x1 - dst->x0) * (src->x1 - src->x0) + src->x0;
c[1] = c[1] / (dst->y1 - dst->y0) * (src->y1 - src->y0) + src->y0;
return VO_TRUE;
}
}
int events = 0;
int r = vo_x11_control(vo, &events, request, data);
if (events & (VO_EVENT_EXPOSE | VO_EVENT_RESIZE))
resize(vo);
return r;
}
int avfilter_transform(const uint8_t *src, uint8_t *dst,
int src_stride, int dst_stride,
int width, int height, const float *matrix,
enum InterpolateMethod interpolate,
enum FillMethod fill)
{
int x, y;
float x_s, y_s;
uint8_t def = 0;
uint8_t (*func)(float, float, const uint8_t *, int, int, int, uint8_t) = NULL;
switch(interpolate) {
case INTERPOLATE_NEAREST:
func = interpolate_nearest;
break;
case INTERPOLATE_BILINEAR:
func = interpolate_bilinear;
break;
case INTERPOLATE_BIQUADRATIC:
func = interpolate_biquadratic;
break;
default:
return AVERROR(EINVAL);
}
for (y = 0; y < height; y++) {
for(x = 0; x < width; x++) {
x_s = x * matrix[0] + y * matrix[1] + matrix[2];
y_s = x * matrix[3] + y * matrix[4] + matrix[5];
switch(fill) {
case FILL_ORIGINAL:
def = src[y * src_stride + x];
break;
case FILL_CLAMP:
y_s = av_clipf(y_s, 0, height - 1);
x_s = av_clipf(x_s, 0, width - 1);
def = src[(int)y_s * src_stride + (int)x_s];
break;
case FILL_MIRROR:
x_s = mirror(x_s, width-1);
y_s = mirror(y_s, height-1);
av_assert2(x_s >= 0 && y_s >= 0);
av_assert2(x_s < width && y_s < height);
def = src[(int)y_s * src_stride + (int)x_s];
}
dst[y * dst_stride + x] = func(x_s, y_s, src, width, height, src_stride, def);
}
}
return 0;
}
static double get_scene_score(AVFilterContext *ctx, AVFilterBufferRef *picref)
{
double ret = 0;
SelectContext *select = ctx->priv;
AVFilterBufferRef *prev_picref = select->prev_picref;
if (prev_picref &&
picref->video->h == prev_picref->video->h &&
picref->video->w == prev_picref->video->w &&
picref->linesize[0] == prev_picref->linesize[0]) {
int x, y;
int64_t sad;
double mafd, diff;
uint8_t *p1 = picref->data[0];
uint8_t *p2 = prev_picref->data[0];
const int linesize = picref->linesize[0];
for (sad = y = 0; y < picref->video->h; y += 8)
for (x = 0; x < linesize; x += 8)
sad += select->c.sad[1](select,
p1 + y * linesize + x,
p2 + y * linesize + x,
linesize, 8);
emms_c();
mafd = sad / (picref->video->h * picref->video->w * 3);
diff = fabs(mafd - select->prev_mafd);
ret = av_clipf(FFMIN(mafd, diff) / 100., 0, 1);
select->prev_mafd = mafd;
avfilter_unref_buffer(prev_picref);
}
select->prev_picref = avfilter_ref_buffer(picref, ~0);
return ret;
}
static double get_scene_score(AVFilterContext *ctx, AVFrame *frame)
{
double ret = 0;
SelectContext *select = ctx->priv;
AVFrame *prev_picref = select->prev_picref;
if (prev_picref &&
frame->height == prev_picref->height &&
frame->width == prev_picref->width) {
int x, y, nb_sad = 0;
int64_t sad = 0;
double mafd, diff;
uint8_t *p1 = frame->data[0];
uint8_t *p2 = prev_picref->data[0];
const int p1_linesize = frame->linesize[0];
const int p2_linesize = prev_picref->linesize[0];
for (y = 0; y < frame->height - 7; y += 8) {
for (x = 0; x < frame->width*3 - 7; x += 8) {
sad += select->sad(p1 + x, p1_linesize, p2 + x, p2_linesize);
nb_sad += 8 * 8;
}
p1 += 8 * p1_linesize;
p2 += 8 * p2_linesize;
}
emms_c();
mafd = nb_sad ? (double)sad / nb_sad : 0;
diff = fabs(mafd - select->prev_mafd);
ret = av_clipf(FFMIN(mafd, diff) / 100., 0, 1);
select->prev_mafd = mafd;
av_frame_free(&prev_picref);
}
select->prev_picref = av_frame_clone(frame);
return ret;
}
int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
if(pts == INT64_MIN)
return s->outpts;
if (s->firstpts == AV_NOPTS_VALUE)
s->outpts = s->firstpts = pts;
if(s->min_compensation >= FLT_MAX) {
return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
} else {
int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts + s->drop_output*(int64_t)s->in_sample_rate;
double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
if(fabs(fdelta) > s->min_compensation) {
if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
int ret;
if(delta > 0) ret = swr_inject_silence(s, delta / s->out_sample_rate);
else ret = swr_drop_output (s, -delta / s-> in_sample_rate);
if(ret<0){
av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
}
} else if(s->soft_compensation_duration && s->max_soft_compensation) {
int duration = s->out_sample_rate * s->soft_compensation_duration;
double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
swr_set_compensation(s, comp, duration);
}
}
return s->outpts;
}
}
static double get_scene_score(AVFilterContext *ctx, AVFrame *crnt, AVFrame *next)
{
FrameRateContext *s = ctx->priv;
double ret = 0;
ff_dlog(ctx, "get_scene_score()\n");
if (crnt->height == next->height &&
crnt->width == next->width) {
int64_t sad;
double mafd, diff;
ff_dlog(ctx, "get_scene_score() process\n");
if (s->bitdepth == 8)
sad = scene_sad8(s, crnt->data[0], crnt->linesize[0], next->data[0], next->linesize[0], crnt->width, crnt->height);
else
sad = scene_sad16(s, (const uint16_t*)crnt->data[0], crnt->linesize[0] / 2, (const uint16_t*)next->data[0], next->linesize[0] / 2, crnt->width, crnt->height);
mafd = (double)sad * 100.0 / FFMAX(1, (crnt->height & ~7) * (crnt->width & ~7)) / (1 << s->bitdepth);
diff = fabs(mafd - s->prev_mafd);
ret = av_clipf(FFMIN(mafd, diff), 0, 100.0);
s->prev_mafd = mafd;
}
ff_dlog(ctx, "get_scene_score() result is:%f\n", ret);
return ret;
}
static int filter_frame(AVFilterLink *inlink, AVFrame *insamples)
{
AVFilterContext *ctx = inlink->dst;
AVFilterLink *outlink = ctx->outputs[0];
ShowVolumeContext *s = ctx->priv;
int c, i, j, k;
double values[VAR_VARS_NB];
if (!s->out || s->out->width != outlink->w ||
s->out->height != outlink->h) {
av_frame_free(&s->out);
s->out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
if (!s->out) {
av_frame_free(&insamples);
return AVERROR(ENOMEM);
}
for (i = 0; i < outlink->h; i++)
memset(s->out->data[0] + i * s->out->linesize[0], 0, outlink->w * 4);
}
s->out->pts = insamples->pts;
for (j = 0; j < outlink->h; j++) {
uint8_t *dst = s->out->data[0] + j * s->out->linesize[0];
for (k = 0; k < s->w; k++) {
dst[k * 4 + 0] = FFMAX(dst[k * 4 + 0] - s->f, 0);
dst[k * 4 + 1] = FFMAX(dst[k * 4 + 1] - s->f, 0);
dst[k * 4 + 2] = FFMAX(dst[k * 4 + 2] - s->f, 0);
dst[k * 4 + 3] = FFMAX(dst[k * 4 + 3] - s->f, 0);
}
}
for (c = 0; c < inlink->channels; c++) {
float *src = (float *)insamples->extended_data[c];
float max = 0;
int color;
for (i = 0; i < insamples->nb_samples; i++)
max = FFMAX(max, src[i]);
max = av_clipf(max, 0, 1);
values[VAR_VOLUME] = 20.0 * log(max) / M_LN10;
color = av_expr_eval(s->c_expr, values, NULL);
for (j = 0; j < s->h; j++) {
uint8_t *dst = s->out->data[0] + (c * s->h + c * s->b + j) * s->out->linesize[0];
for (k = 0; k < s->w * max; k++)
AV_WN32A(dst + k * 4, color);
}
if (s->h >= 8 && s->draw_text)
drawtext(s->out, 2, c * (s->h + s->b) + (s->h - 8) / 2,
av_get_channel_name(av_channel_layout_extract_channel(insamples->channel_layout, c)));
}
av_frame_free(&insamples);
return ff_filter_frame(outlink, av_frame_clone(s->out));
}
void avfilter_transform(const uint8_t *src, uint8_t *dst,
int src_stride, int dst_stride,
int width, int height, const float *matrix,
enum InterpolateMethod interpolate,
enum FillMethod fill)
{
int x, y;
float x_s, y_s;
uint8_t def = 0;
uint8_t (*func)(float, float, const uint8_t *, int, int, int, uint8_t) = NULL;
switch(interpolate) {
case INTERPOLATE_NEAREST:
func = interpolate_nearest;
break;
case INTERPOLATE_BILINEAR:
func = interpolate_bilinear;
break;
case INTERPOLATE_BIQUADRATIC:
func = interpolate_biquadratic;
break;
}
for (y = 0; y < height; y++) {
for(x = 0; x < width; x++) {
x_s = x * matrix[0] + y * matrix[1] + matrix[2];
y_s = x * matrix[3] + y * matrix[4] + matrix[5];
switch(fill) {
case FILL_ORIGINAL:
def = src[y * src_stride + x];
break;
case FILL_CLAMP:
y_s = av_clipf(y_s, 0, height - 1);
x_s = av_clipf(x_s, 0, width - 1);
def = src[(int)y_s * src_stride + (int)x_s];
break;
case FILL_MIRROR:
y_s = (y_s < 0) ? -y_s : (y_s >= height) ? (height + height - y_s) : y_s;
x_s = (x_s < 0) ? -x_s : (x_s >= width) ? (width + width - x_s) : x_s;
def = src[(int)y_s * src_stride + (int)x_s];
}
dst[y * dst_stride + x] = func(x_s, y_s, src, width, height, src_stride, def);
}
}
}
static void decode(RA288Context *ractx, float gain, int cb_coef)
{
int i, j;
double sumsum;
float sum, buffer[5];
float *block = ractx->sp_block + 36; // Current block
memmove(ractx->sp_block, ractx->sp_block + 5, 36*sizeof(*ractx->sp_block));
for (i=0; i < 5; i++) {
block[i] = 0.;
for (j=0; j < 36; j++)
block[i] -= block[i-1-j]*ractx->sp_lpc[j];
}
/* block 46 of G.728 spec */
sum = 32.;
for (i=0; i < 10; i++)
sum -= ractx->gain_block[9-i] * ractx->gain_lpc[i];
/* block 47 of G.728 spec */
sum = av_clipf(sum, 0, 60);
/* block 48 of G.728 spec */
sumsum = exp(sum * 0.1151292546497) * gain; /* pow(10.0,sum/20)*gain */
for (i=0; i < 5; i++)
buffer[i] = codetable[cb_coef][i] * sumsum;
sum = scalar_product_float(buffer, buffer, 5) / 5;
sum = FFMAX(sum, 1);
/* shift and store */
memmove(ractx->gain_block, ractx->gain_block + 1,
9 * sizeof(*ractx->gain_block));
ractx->gain_block[9] = 10 * log10(sum) - 32;
for (i=1; i < 5; i++)
for (j=i-1; j >= 0; j--)
buffer[i] -= ractx->sp_lpc[i-j-1] * buffer[j];
/* output */
for (i=0; i < 5; i++)
block[i] = av_clipf(block[i] + buffer[i], -4095, 4095);
}
/**
* Saturate the output signal and interleave.
*
* @param q pointer to the COOKContext
* @param chan channel to saturate
* @param out pointer to the output vector
*/
static void saturate_output_float(COOKContext *q, int chan, float *out)
{
int j;
float *output = q->mono_mdct_output + q->samples_per_channel;
for (j = 0; j < q->samples_per_channel; j++) {
out[chan + q->nb_channels * j] = av_clipf(output[j], -1.0, 1.0);
}
}
static int filter_frame(AVFilterLink *inlink, AVFrame *in)
{
AVFilterContext *ctx = inlink->dst;
AVFilterLink *outlink = ctx->outputs[0];
ExtraStereoContext *s = ctx->priv;
const float *src = (const float *)in->data[0];
const float mult = s->mult;
AVFrame *out;
float *dst;
int n;
if (av_frame_is_writable(in)) {
out = in;
} else {
out = ff_get_audio_buffer(inlink, in->nb_samples);
if (!out) {
av_frame_free(&in);
return AVERROR(ENOMEM);
}
av_frame_copy_props(out, in);
}
dst = (float *)out->data[0];
for (n = 0; n < in->nb_samples; n++) {
float average, left, right;
left = src[n * 2 ];
right = src[n * 2 + 1];
average = (left + right) / 2.;
left = average + mult * (left - average);
right = average + mult * (right - average);
if (s->clip) {
left = av_clipf(left, -1, 1);
right = av_clipf(right, -1, 1);
}
dst[n * 2 ] = left;
dst[n * 2 + 1] = right;
}
if (out != in)
av_frame_free(&in);
return ff_filter_frame(outlink, out);
}
static void decode(RA288Context *ractx, float gain, int cb_coef)
{
int i;
double sumsum;
float sum, buffer[5];
float *block = ractx->sp_hist + 70 + 36; // current block
float *gain_block = ractx->gain_hist + 28;
memmove(ractx->sp_hist + 70, ractx->sp_hist + 75, 36*sizeof(*block));
/* block 46 of G.728 spec */
sum = 32.;
for (i=0; i < 10; i++)
sum -= gain_block[9-i] * ractx->gain_lpc[i];
/* block 47 of G.728 spec */
sum = av_clipf(sum, 0, 60);
/* block 48 of G.728 spec */
/* exp(sum * 0.1151292546497) == pow(10.0,sum/20) */
sumsum = exp(sum * 0.1151292546497) * gain * (1.0/(1<<23));
for (i=0; i < 5; i++)
buffer[i] = codetable[cb_coef][i] * sumsum;
sum = ff_dot_productf(buffer, buffer, 5) * ((1<<24)/5.);
sum = FFMAX(sum, 1);
/* shift and store */
memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));
gain_block[9] = 10 * log10(sum) - 32;
ff_celp_lp_synthesis_filterf(block, ractx->sp_lpc, buffer, 5, 36);
/* output */
for (i=0; i < 5; i++)
block[i] = av_clipf(block[i], -4095./4096., 4095./4096.);
}
static int filter_frame(AVFilterLink *inlink, AVFrame *in)
{
AVFilterContext *ctx = inlink->dst;
AVFilterLink *outlink = ctx->outputs[0];
CrystalizerContext *s = ctx->priv;
const float *src = (const float *)in->data[0];
const float mult = s->mult;
AVFrame *out;
float *dst, *prv;
int n, c;
if (!s->prev) {
s->prev = ff_get_audio_buffer(inlink, 1);
if (!s->prev) {
av_frame_free(&in);
return AVERROR(ENOMEM);
}
}
if (av_frame_is_writable(in)) {
out = in;
} else {
out = ff_get_audio_buffer(inlink, in->nb_samples);
if (!out) {
av_frame_free(&in);
return AVERROR(ENOMEM);
}
av_frame_copy_props(out, in);
}
dst = (float *)out->data[0];
prv = (float *)s->prev->data[0];
for (n = 0; n < in->nb_samples; n++) {
for (c = 0; c < in->channels; c++) {
float current = src[c];
dst[c] = current + (current - prv[c]) * mult;
prv[c] = current;
if (s->clip) {
dst[c] = av_clipf(dst[c], -1, 1);
}
}
dst += c;
src += c;
}
if (out != in)
av_frame_free(&in);
return ff_filter_frame(outlink, out);
}
static void method2_int16(af_volnorm_t *s, af_data_t *c)
{
register int i = 0;
int16_t *data = (int16_t*)c->audio; // Audio data
int len = c->len/2; // Number of samples
float curavg = 0.0, newavg, avg = 0.0;
int tmp, totallen = 0;
for (i = 0; i < len; i++)
{
tmp = data[i];
curavg += tmp * tmp;
}
curavg = sqrt(curavg / (float) len);
// Evaluate an adequate 'mul' coefficient based on previous state, current
// samples level, etc
for (i = 0; i < NSAMPLES; i++)
{
avg += s->mem[i].avg * (float)s->mem[i].len;
totallen += s->mem[i].len;
}
if (totallen > MIN_SAMPLE_SIZE)
{
avg /= (float)totallen;
if (avg >= SIL_S16)
{
s->mul = s->mid_s16 / avg;
s->mul = av_clipf(s->mul, MUL_MIN, MUL_MAX);
}
}
// Scale & clamp the samples
for (i = 0; i < len; i++)
{
tmp = s->mul * data[i];
tmp = av_clip_int16(tmp);
data[i] = tmp;
}
// Evaulation of newavg (not 100% accurate because of values clamping)
newavg = s->mul * curavg;
// Stores computed values for future smoothing
s->mem[s->idx].len = len;
s->mem[s->idx].avg = newavg;
s->idx = (s->idx + 1) % NSAMPLES;
}
static int libopus_decode(AVCodecContext *avc, void *data,
int *got_frame_ptr, AVPacket *pkt)
{
struct libopus_context *opus = avc->priv_data;
AVFrame *frame = data;
int ret, nb_samples;
frame->nb_samples = MAX_FRAME_SIZE;
ret = ff_get_buffer(avc, frame);
if (ret < 0) {
av_log(avc, AV_LOG_ERROR, "get_buffer() failed\n");
return ret;
}
if (avc->sample_fmt == AV_SAMPLE_FMT_S16)
nb_samples = opus_multistream_decode(opus->dec, pkt->data, pkt->size,
(opus_int16 *)frame->data[0],
frame->nb_samples, 0);
else
nb_samples = opus_multistream_decode_float(opus->dec, pkt->data, pkt->size,
(float *)frame->data[0],
frame->nb_samples, 0);
if (nb_samples < 0) {
av_log(avc, AV_LOG_ERROR, "Decoding error: %s\n",
opus_strerror(nb_samples));
return ff_opus_error_to_averror(nb_samples);
}
#ifndef OPUS_SET_GAIN
{
int i = avc->channels * nb_samples;
if (avc->sample_fmt == AV_SAMPLE_FMT_FLT) {
float *pcm = (float *)frame->data[0];
for (; i > 0; i--, pcm++)
*pcm = av_clipf(*pcm * opus->gain.d, -1, 1);
} else {
int16_t *pcm = (int16_t *)frame->data[0];
for (; i > 0; i--, pcm++)
*pcm = av_clip_int16(((int64_t)opus->gain.i * *pcm) >> 16);
}
}
#endif
frame->nb_samples = nb_samples;
*got_frame_ptr = 1;
return pkt->size;
}
/*
* Calculate the ReplayGain value from the specified loudness histogram;
* clip to -24 / +64 dB.
*/
static float calc_replaygain(uint32_t *histogram)
{
uint32_t loud_count = 0, total_windows = 0;
float gain;
int i;
for (i = 0; i < HISTOGRAM_SLOTS; i++)
total_windows += histogram [i];
while (i--)
if ((loud_count += histogram [i]) * 20 >= total_windows)
break;
gain = (float)(64.54 - i / 100.0);
return av_clipf(gain, -24.0, 64.0);
}
static void vector_clipf_c(float *dst, const float *src,
float min, float max, int len)
{
int i;
if (min < 0 && max > 0) {
vector_clipf_c_opposite_sign(dst, src, &min, &max, len);
} else {
for (i = 0; i < len; i += 8) {
dst[i] = av_clipf(src[i], min, max);
dst[i + 1] = av_clipf(src[i + 1], min, max);
dst[i + 2] = av_clipf(src[i + 2], min, max);
dst[i + 3] = av_clipf(src[i + 3], min, max);
dst[i + 4] = av_clipf(src[i + 4], min, max);
dst[i + 5] = av_clipf(src[i + 5], min, max);
dst[i + 6] = av_clipf(src[i + 6], min, max);
dst[i + 7] = av_clipf(src[i + 7], min, max);
}
}
}
static void method1_int16(af_volnorm_t *s, af_data_t *c)
{
register int i = 0;
int16_t *data = (int16_t*)c->audio; // Audio data
int len = c->len/2; // Number of samples
float curavg = 0.0, newavg, neededmul;
int tmp;
for (i = 0; i < len; i++)
{
tmp = data[i];
curavg += tmp * tmp;
}
curavg = sqrt(curavg / (float) len);
// Evaluate an adequate 'mul' coefficient based on previous state, current
// samples level, etc
if (curavg > SIL_S16)
{
neededmul = s->mid_s16 / (curavg * s->mul);
s->mul = (1.0 - SMOOTH_MUL) * s->mul + SMOOTH_MUL * neededmul;
// clamp the mul coefficient
s->mul = av_clipf(s->mul, MUL_MIN, MUL_MAX);
}
// Scale & clamp the samples
for (i = 0; i < len; i++)
{
tmp = s->mul * data[i];
tmp = av_clip_int16(tmp);
data[i] = tmp;
}
// Evaulation of newavg (not 100% accurate because of values clamping)
newavg = s->mul * curavg;
// Stores computed values for future smoothing
s->lastavg = (1.0 - SMOOTH_LASTAVG) * s->lastavg + SMOOTH_LASTAVG * newavg;
}
static void filter_fltp(void **d, void **p, const void **s,
int nb_samples, int channels,
float mult, int clip)
{
int n, c;
for (c = 0; c < channels; c++) {
const float *src = s[c];
float *dst = d[c];
float *prv = p[c];
for (n = 0; n < nb_samples; n++) {
float current = src[n];
dst[n] = current + (current - prv[0]) * mult;
prv[0] = current;
if (clip) {
dst[n] = av_clipf(dst[n], -1, 1);
}
}
}
}
static void update_context(VignetteContext *s, AVFilterLink *inlink, AVFrame *frame)
{
int x, y;
float *dst = s->fmap;
int dst_linesize = s->fmap_linesize;
if (frame) {
s->var_values[VAR_N] = inlink->frame_count_out;
s->var_values[VAR_T] = TS2T(frame->pts, inlink->time_base);
s->var_values[VAR_PTS] = TS2D(frame->pts);
} else {
s->var_values[VAR_N] = NAN;
s->var_values[VAR_T] = NAN;
s->var_values[VAR_PTS] = NAN;
}
s->angle = av_expr_eval(s->angle_pexpr, s->var_values, NULL);
s->x0 = av_expr_eval(s->x0_pexpr, s->var_values, NULL);
s->y0 = av_expr_eval(s->y0_pexpr, s->var_values, NULL);
if (isnan(s->x0) || isnan(s->y0) || isnan(s->angle))
s->eval_mode = EVAL_MODE_FRAME;
s->angle = av_clipf(s->angle, 0, M_PI_2);
if (s->backward) {
for (y = 0; y < inlink->h; y++) {
for (x = 0; x < inlink->w; x++)
dst[x] = 1. / get_natural_factor(s, x, y);
dst += dst_linesize;
}
} else {
for (y = 0; y < inlink->h; y++) {
for (x = 0; x < inlink->w; x++)
dst[x] = get_natural_factor(s, x, y);
dst += dst_linesize;
}
}
}
static void quantize_triangular_ns(DitherContext *c, DitherState *state,
int16_t *dst, const float *src,
int nb_samples)
{
int i, j;
float *dither = &state->noise_buf[state->noise_buf_ptr];
if (state->mute > c->mute_reset_threshold)
memset(state->dither_a, 0, sizeof(state->dither_a));
for (i = 0; i < nb_samples; i++) {
float err = 0;
float sample = src[i] * S16_SCALE;
for (j = 0; j < 4; j++) {
err += c->ns_coef_b[j] * state->dither_b[j] -
c->ns_coef_a[j] * state->dither_a[j];
}
for (j = 3; j > 0; j--) {
state->dither_a[j] = state->dither_a[j - 1];
state->dither_b[j] = state->dither_b[j - 1];
}
state->dither_a[0] = err;
sample -= err;
if (state->mute > c->mute_dither_threshold) {
dst[i] = av_clip_int16(lrintf(sample));
state->dither_b[0] = 0;
} else {
dst[i] = av_clip_int16(lrintf(sample + dither[i]));
state->dither_b[0] = av_clipf(dst[i] - sample, -1.5f, 1.5f);
}
state->mute++;
if (src[i])
state->mute = 0;
}
}
static double get_scene_score(AVFilterContext *ctx, AVFrame *crnt, AVFrame *next)
{
FrameRateContext *s = ctx->priv;
double ret = 0;
ff_dlog(ctx, "get_scene_score()\n");
if (crnt &&
crnt->height == next->height &&
crnt->width == next->width) {
int x, y;
int64_t sad;
double mafd, diff;
uint8_t *p1 = crnt->data[0];
uint8_t *p2 = next->data[0];
const int p1_linesize = crnt->linesize[0];
const int p2_linesize = next->linesize[0];
ff_dlog(ctx, "get_scene_score() process\n");
for (sad = y = 0; y < crnt->height; y += 8) {
for (x = 0; x < p1_linesize; x += 8) {
sad += s->sad(p1 + y * p1_linesize + x,
p1_linesize,
p2 + y * p2_linesize + x,
p2_linesize);
}
}
emms_c();
mafd = sad / (crnt->height * crnt->width * 3);
diff = fabs(mafd - s->prev_mafd);
ret = av_clipf(FFMIN(mafd, diff), 0, 100.0);
s->prev_mafd = mafd;
}
ff_dlog(ctx, "get_scene_score() result is:%f\n", ret);
return ret;
}
static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
const AVFrame *frame, int *got_packet_ptr)
{
AACEncContext *s = avctx->priv_data;
float **samples = s->planar_samples, *samples2, *la, *overlap;
ChannelElement *cpe;
SingleChannelElement *sce;
IndividualChannelStream *ics;
int i, its, ch, w, chans, tag, start_ch, ret, frame_bits;
int target_bits, rate_bits, too_many_bits, too_few_bits;
int ms_mode = 0, is_mode = 0, tns_mode = 0, pred_mode = 0;
int chan_el_counter[4];
FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
if (s->last_frame == 2)
return 0;
/* add current frame to queue */
if (frame) {
if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
return ret;
}
copy_input_samples(s, frame);
if (s->psypp)
ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
if (!avctx->frame_number)
return 0;
start_ch = 0;
for (i = 0; i < s->chan_map[0]; i++) {
FFPsyWindowInfo* wi = windows + start_ch;
tag = s->chan_map[i+1];
chans = tag == TYPE_CPE ? 2 : 1;
cpe = &s->cpe[i];
for (ch = 0; ch < chans; ch++) {
float clip_avoidance_factor;
sce = &cpe->ch[ch];
ics = &sce->ics;
s->cur_channel = start_ch + ch;
overlap = &samples[s->cur_channel][0];
samples2 = overlap + 1024;
la = samples2 + (448+64);
if (!frame)
la = NULL;
if (tag == TYPE_LFE) {
wi[ch].window_type[0] = ONLY_LONG_SEQUENCE;
wi[ch].window_shape = 0;
wi[ch].num_windows = 1;
wi[ch].grouping[0] = 1;
/* Only the lowest 12 coefficients are used in a LFE channel.
* The expression below results in only the bottom 8 coefficients
* being used for 11.025kHz to 16kHz sample rates.
*/
ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
} else {
wi[ch] = s->psy.model->window(&s->psy, samples2, la, s->cur_channel,
ics->window_sequence[0]);
}
ics->window_sequence[1] = ics->window_sequence[0];
ics->window_sequence[0] = wi[ch].window_type[0];
ics->use_kb_window[1] = ics->use_kb_window[0];
ics->use_kb_window[0] = wi[ch].window_shape;
ics->num_windows = wi[ch].num_windows;
ics->swb_sizes = s->psy.bands [ics->num_windows == 8];
ics->num_swb = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
ics->max_sfb = FFMIN(ics->max_sfb, ics->num_swb);
ics->swb_offset = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
ff_swb_offset_128 [s->samplerate_index]:
ff_swb_offset_1024[s->samplerate_index];
ics->tns_max_bands = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
ff_tns_max_bands_128 [s->samplerate_index]:
ff_tns_max_bands_1024[s->samplerate_index];
clip_avoidance_factor = 0.0f;
for (w = 0; w < ics->num_windows; w++)
ics->group_len[w] = wi[ch].grouping[w];
for (w = 0; w < ics->num_windows; w++) {
if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
ics->window_clipping[w] = 1;
clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
} else {
ics->window_clipping[w] = 0;
}
}
if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
} else {
ics->clip_avoidance_factor = 1.0f;
}
apply_window_and_mdct(s, sce, overlap);
if (s->options.ltp && s->coder->update_ltp) {
s->coder->update_ltp(s, sce);
apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, &sce->ltp_state[0]);
s->mdct1024.mdct_calc(&s->mdct1024, sce->lcoeffs, sce->ret_buf);
}
if (!(isfinite(cpe->ch[ch].coeffs[ 0]) &&
isfinite(cpe->ch[ch].coeffs[ 128]) &&
isfinite(cpe->ch[ch].coeffs[2*128]) &&
isfinite(cpe->ch[ch].coeffs[3*128]) &&
isfinite(cpe->ch[ch].coeffs[4*128]) &&
isfinite(cpe->ch[ch].coeffs[5*128]) &&
isfinite(cpe->ch[ch].coeffs[6*128]) &&
isfinite(cpe->ch[ch].coeffs[7*128]))
) {
av_log(avctx, AV_LOG_ERROR, "Input contains NaN/+-Inf\n");
return AVERROR(EINVAL);
}
avoid_clipping(s, sce);
}
start_ch += chans;
}
if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)
return ret;
frame_bits = its = 0;
do {
init_put_bits(&s->pb, avpkt->data, avpkt->size);
if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
put_bitstream_info(s, LIBAVCODEC_IDENT);
start_ch = 0;
target_bits = 0;
memset(chan_el_counter, 0, sizeof(chan_el_counter));
for (i = 0; i < s->chan_map[0]; i++) {
FFPsyWindowInfo* wi = windows + start_ch;
const float *coeffs[2];
tag = s->chan_map[i+1];
chans = tag == TYPE_CPE ? 2 : 1;
cpe = &s->cpe[i];
cpe->common_window = 0;
memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
put_bits(&s->pb, 3, tag);
put_bits(&s->pb, 4, chan_el_counter[tag]++);
for (ch = 0; ch < chans; ch++) {
sce = &cpe->ch[ch];
coeffs[ch] = sce->coeffs;
sce->ics.predictor_present = 0;
sce->ics.ltp.present = 0;
memset(sce->ics.ltp.used, 0, sizeof(sce->ics.ltp.used));
memset(sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));
memset(&sce->tns, 0, sizeof(TemporalNoiseShaping));
for (w = 0; w < 128; w++)
if (sce->band_type[w] > RESERVED_BT)
sce->band_type[w] = 0;
}
s->psy.bitres.alloc = -1;
s->psy.bitres.bits = s->last_frame_pb_count / s->channels;
s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
if (s->psy.bitres.alloc > 0) {
/* Lambda unused here on purpose, we need to take psy's unscaled allocation */
target_bits += s->psy.bitres.alloc
* (s->lambda / (avctx->global_quality ? avctx->global_quality : 120));
s->psy.bitres.alloc /= chans;
}
s->cur_type = tag;
for (ch = 0; ch < chans; ch++) {
s->cur_channel = start_ch + ch;
if (s->options.pns && s->coder->mark_pns)
s->coder->mark_pns(s, avctx, &cpe->ch[ch]);
s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
}
if (chans > 1
&& wi[0].window_type[0] == wi[1].window_type[0]
&& wi[0].window_shape == wi[1].window_shape) {
cpe->common_window = 1;
for (w = 0; w < wi[0].num_windows; w++) {
if (wi[0].grouping[w] != wi[1].grouping[w]) {
cpe->common_window = 0;
break;
}
}
}
for (ch = 0; ch < chans; ch++) { /* TNS and PNS */
sce = &cpe->ch[ch];
s->cur_channel = start_ch + ch;
if (s->options.tns && s->coder->search_for_tns)
s->coder->search_for_tns(s, sce);
if (s->options.tns && s->coder->apply_tns_filt)
s->coder->apply_tns_filt(s, sce);
if (sce->tns.present)
tns_mode = 1;
if (s->options.pns && s->coder->search_for_pns)
s->coder->search_for_pns(s, avctx, sce);
}
s->cur_channel = start_ch;
if (s->options.intensity_stereo) { /* Intensity Stereo */
if (s->coder->search_for_is)
s->coder->search_for_is(s, avctx, cpe);
if (cpe->is_mode) is_mode = 1;
apply_intensity_stereo(cpe);
}
if (s->options.pred) { /* Prediction */
for (ch = 0; ch < chans; ch++) {
sce = &cpe->ch[ch];
s->cur_channel = start_ch + ch;
if (s->options.pred && s->coder->search_for_pred)
s->coder->search_for_pred(s, sce);
if (cpe->ch[ch].ics.predictor_present) pred_mode = 1;
}
if (s->coder->adjust_common_pred)
s->coder->adjust_common_pred(s, cpe);
for (ch = 0; ch < chans; ch++) {
sce = &cpe->ch[ch];
s->cur_channel = start_ch + ch;
if (s->options.pred && s->coder->apply_main_pred)
s->coder->apply_main_pred(s, sce);
}
s->cur_channel = start_ch;
}
if (s->options.mid_side) { /* Mid/Side stereo */
if (s->options.mid_side == -1 && s->coder->search_for_ms)
s->coder->search_for_ms(s, cpe);
else if (cpe->common_window)
memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask));
apply_mid_side_stereo(cpe);
}
adjust_frame_information(cpe, chans);
if (s->options.ltp) { /* LTP */
for (ch = 0; ch < chans; ch++) {
sce = &cpe->ch[ch];
s->cur_channel = start_ch + ch;
if (s->coder->search_for_ltp)
s->coder->search_for_ltp(s, sce, cpe->common_window);
if (sce->ics.ltp.present) pred_mode = 1;
}
s->cur_channel = start_ch;
if (s->coder->adjust_common_ltp)
s->coder->adjust_common_ltp(s, cpe);
}
if (chans == 2) {
put_bits(&s->pb, 1, cpe->common_window);
if (cpe->common_window) {
put_ics_info(s, &cpe->ch[0].ics);
if (s->coder->encode_main_pred)
s->coder->encode_main_pred(s, &cpe->ch[0]);
if (s->coder->encode_ltp_info)
s->coder->encode_ltp_info(s, &cpe->ch[0], 1);
encode_ms_info(&s->pb, cpe);
if (cpe->ms_mode) ms_mode = 1;
}
}
for (ch = 0; ch < chans; ch++) {
s->cur_channel = start_ch + ch;
encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
}
start_ch += chans;
}
if (avctx->flags & CODEC_FLAG_QSCALE) {
/* When using a constant Q-scale, don't mess with lambda */
break;
}
/* rate control stuff
* allow between the nominal bitrate, and what psy's bit reservoir says to target
* but drift towards the nominal bitrate always
*/
frame_bits = put_bits_count(&s->pb);
rate_bits = avctx->bit_rate * 1024 / avctx->sample_rate;
rate_bits = FFMIN(rate_bits, 6144 * s->channels - 3);
too_many_bits = FFMAX(target_bits, rate_bits);
too_many_bits = FFMIN(too_many_bits, 6144 * s->channels - 3);
too_few_bits = FFMIN(FFMAX(rate_bits - rate_bits/4, target_bits), too_many_bits);
/* When using ABR, be strict (but only for increasing) */
too_few_bits = too_few_bits - too_few_bits/8;
too_many_bits = too_many_bits + too_many_bits/2;
if ( its == 0 /* for steady-state Q-scale tracking */
|| (its < 5 && (frame_bits < too_few_bits || frame_bits > too_many_bits))
|| frame_bits >= 6144 * s->channels - 3 )
{
float ratio = ((float)rate_bits) / frame_bits;
if (frame_bits >= too_few_bits && frame_bits <= too_many_bits) {
/*
* This path is for steady-state Q-scale tracking
* When frame bits fall within the stable range, we still need to adjust
* lambda to maintain it like so in a stable fashion (large jumps in lambda
* create artifacts and should be avoided), but slowly
*/
ratio = sqrtf(sqrtf(ratio));
ratio = av_clipf(ratio, 0.9f, 1.1f);
} else {
/* Not so fast though */
ratio = sqrtf(ratio);
}
s->lambda = FFMIN(s->lambda * ratio, 65536.f);
/* Keep iterating if we must reduce and lambda is in the sky */
if (ratio > 0.9f && ratio < 1.1f) {
break;
} else {
if (is_mode || ms_mode || tns_mode || pred_mode) {
for (i = 0; i < s->chan_map[0]; i++) {
// Must restore coeffs
chans = tag == TYPE_CPE ? 2 : 1;
cpe = &s->cpe[i];
for (ch = 0; ch < chans; ch++)
memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
}
}
its++;
}
} else {
break;
}
} while (1);
if (s->options.ltp && s->coder->ltp_insert_new_frame)
s->coder->ltp_insert_new_frame(s);
put_bits(&s->pb, 3, TYPE_END);
flush_put_bits(&s->pb);
s->last_frame_pb_count = put_bits_count(&s->pb);
s->lambda_sum += s->lambda;
s->lambda_count++;
if (!frame)
s->last_frame++;
ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
&avpkt->duration);
avpkt->size = put_bits_count(&s->pb) >> 3;
*got_packet_ptr = 1;
return 0;
}
/**
* Find the estimated global motion for a scene given the most likely shift
* for each block in the frame. The global motion is estimated to be the
* same as the motion from most blocks in the frame, so if most blocks
* move one pixel to the right and two pixels down, this would yield a
* motion vector (1, -2).
*/
static void find_motion(DeshakeContext *deshake, uint8_t *src1, uint8_t *src2,
int width, int height, int stride, Transform *t)
{
int x, y;
IntMotionVector mv = {0, 0};
int counts[2*MAX_R+1][2*MAX_R+1];
int count_max_value = 0;
int contrast;
int pos;
double *angles = av_malloc(sizeof(*angles) * width * height / (16 * deshake->blocksize));
int center_x = 0, center_y = 0;
double p_x, p_y;
// Reset counts to zero
for (x = 0; x < deshake->rx * 2 + 1; x++) {
for (y = 0; y < deshake->ry * 2 + 1; y++) {
counts[x][y] = 0;
}
}
pos = 0;
// Find motion for every block and store the motion vector in the counts
for (y = deshake->ry; y < height - deshake->ry - (deshake->blocksize * 2); y += deshake->blocksize * 2) {
// We use a width of 16 here to match the libavcodec sad functions
for (x = deshake->rx; x < width - deshake->rx - 16; x += 16) {
// If the contrast is too low, just skip this block as it probably
// won't be very useful to us.
contrast = block_contrast(src2, x, y, stride, deshake->blocksize);
if (contrast > deshake->contrast) {
//av_log(NULL, AV_LOG_ERROR, "%d\n", contrast);
find_block_motion(deshake, src1, src2, x, y, stride, &mv);
if (mv.x != -1 && mv.y != -1) {
counts[mv.x + deshake->rx][mv.y + deshake->ry] += 1;
if (x > deshake->rx && y > deshake->ry)
angles[pos++] = block_angle(x, y, 0, 0, &mv);
center_x += mv.x;
center_y += mv.y;
}
}
}
}
if (pos) {
center_x /= pos;
center_y /= pos;
t->angle = clean_mean(angles, pos);
if (t->angle < 0.001)
t->angle = 0;
} else {
t->angle = 0;
}
// Find the most common motion vector in the frame and use it as the gmv
for (y = deshake->ry * 2; y >= 0; y--) {
for (x = 0; x < deshake->rx * 2 + 1; x++) {
//av_log(NULL, AV_LOG_ERROR, "%5d ", counts[x][y]);
if (counts[x][y] > count_max_value) {
t->vector.x = x - deshake->rx;
t->vector.y = y - deshake->ry;
count_max_value = counts[x][y];
}
}
//av_log(NULL, AV_LOG_ERROR, "\n");
}
p_x = (center_x - width / 2.0);
p_y = (center_y - height / 2.0);
t->vector.x += (cos(t->angle)-1)*p_x - sin(t->angle)*p_y;
t->vector.y += sin(t->angle)*p_x + (cos(t->angle)-1)*p_y;
// Clamp max shift & rotation?
t->vector.x = av_clipf(t->vector.x, -deshake->rx * 2, deshake->rx * 2);
t->vector.y = av_clipf(t->vector.y, -deshake->ry * 2, deshake->ry * 2);
t->angle = av_clipf(t->angle, -0.1, 0.1);
//av_log(NULL, AV_LOG_ERROR, "%d x %d\n", avg->x, avg->y);
av_free(angles);
}
static void stereo_position(float a, float p, float *x, float *y)
{
*x = av_clipf(a+FFMAX(0, sinf(p-M_PI_2))*FFDIFFSIGN(a,0), -1, 1);
*y = av_clipf(cosf(a*M_PI_2+M_PI)*cosf(M_PI_2-p/M_PI)*M_LN10+1, -1, 1);
}
av_cold int ff_rate_control_init(MpegEncContext *s)
{
RateControlContext *rcc = &s->rc_context;
int i, res;
static const char * const const_names[] = {
"PI",
"E",
"iTex",
"pTex",
"tex",
"mv",
"fCode",
"iCount",
"mcVar",
"var",
"isI",
"isP",
"isB",
"avgQP",
"qComp",
#if 0
"lastIQP",
"lastPQP",
"lastBQP",
"nextNonBQP",
#endif
"avgIITex",
"avgPITex",
"avgPPTex",
"avgBPTex",
"avgTex",
NULL
};
static double (* const func1[])(void *, double) = {
(void *)bits2qp,
(void *)qp2bits,
NULL
};
static const char * const func1_names[] = {
"bits2qp",
"qp2bits",
NULL
};
emms_c();
if (!s->avctx->rc_max_available_vbv_use && s->avctx->rc_buffer_size) {
if (s->avctx->rc_max_rate) {
s->avctx->rc_max_available_vbv_use = av_clipf(s->avctx->rc_max_rate/(s->avctx->rc_buffer_size*get_fps(s->avctx)), 1.0/3, 1.0);
} else
s->avctx->rc_max_available_vbv_use = 1.0;
}
res = av_expr_parse(&rcc->rc_eq_eval,
s->avctx->rc_eq ? s->avctx->rc_eq : "tex^qComp",
const_names, func1_names, func1,
NULL, NULL, 0, s->avctx);
if (res < 0) {
av_log(s->avctx, AV_LOG_ERROR, "Error parsing rc_eq \"%s\"\n", s->avctx->rc_eq);
return res;
}
for (i = 0; i < 5; i++) {
rcc->pred[i].coeff = FF_QP2LAMBDA * 7.0;
rcc->pred[i].count = 1.0;
rcc->pred[i].decay = 0.4;
rcc->i_cplx_sum [i] =
rcc->p_cplx_sum [i] =
rcc->mv_bits_sum[i] =
rcc->qscale_sum [i] =
rcc->frame_count[i] = 1; // 1 is better because of 1/0 and such
rcc->last_qscale_for[i] = FF_QP2LAMBDA * 5;
}
rcc->buffer_index = s->avctx->rc_initial_buffer_occupancy;
if (!rcc->buffer_index)
rcc->buffer_index = s->avctx->rc_buffer_size * 3 / 4;
if (s->flags & CODEC_FLAG_PASS2) {
int i;
char *p;
/* find number of pics */
p = s->avctx->stats_in;
for (i = -1; p; i++)
p = strchr(p + 1, ';');
i += s->max_b_frames;
if (i <= 0 || i >= INT_MAX / sizeof(RateControlEntry))
return -1;
rcc->entry = av_mallocz(i * sizeof(RateControlEntry));
rcc->num_entries = i;
/* init all to skipped p frames
* (with b frames we might have a not encoded frame at the end FIXME) */
for (i = 0; i < rcc->num_entries; i++) {
RateControlEntry *rce = &rcc->entry[i];
rce->pict_type = rce->new_pict_type = AV_PICTURE_TYPE_P;
rce->qscale = rce->new_qscale = FF_QP2LAMBDA * 2;
rce->misc_bits = s->mb_num + 10;
rce->mb_var_sum = s->mb_num * 100;
}
/* read stats */
p = s->avctx->stats_in;
for (i = 0; i < rcc->num_entries - s->max_b_frames; i++) {
RateControlEntry *rce;
int picture_number;
int e;
char *next;
next = strchr(p, ';');
if (next) {
(*next) = 0; // sscanf in unbelievably slow on looong strings // FIXME copy / do not write
next++;
}
e = sscanf(p, " in:%d ", &picture_number);
assert(picture_number >= 0);
assert(picture_number < rcc->num_entries);
rce = &rcc->entry[picture_number];
e += sscanf(p, " in:%*d out:%*d type:%d q:%f itex:%d ptex:%d mv:%d misc:%d fcode:%d bcode:%d mc-var:%"SCNd64" var:%"SCNd64" icount:%d skipcount:%d hbits:%d",
&rce->pict_type, &rce->qscale, &rce->i_tex_bits, &rce->p_tex_bits,
&rce->mv_bits, &rce->misc_bits,
&rce->f_code, &rce->b_code,
&rce->mc_mb_var_sum, &rce->mb_var_sum,
&rce->i_count, &rce->skip_count, &rce->header_bits);
if (e != 14) {
av_log(s->avctx, AV_LOG_ERROR,
"statistics are damaged at line %d, parser out=%d\n",
i, e);
return -1;
}
p = next;
}
if (init_pass2(s) < 0)
return -1;
// FIXME maybe move to end
if ((s->flags & CODEC_FLAG_PASS2) && s->avctx->rc_strategy == FF_RC_STRATEGY_XVID) {
#if CONFIG_LIBXVID
return ff_xvid_rate_control_init(s);
#else
av_log(s->avctx, AV_LOG_ERROR,
"Xvid ratecontrol requires libavcodec compiled with Xvid support.\n");
return -1;
#endif
}
}
if (!(s->flags & CODEC_FLAG_PASS2)) {
rcc->short_term_qsum = 0.001;
rcc->short_term_qcount = 0.001;
rcc->pass1_rc_eq_output_sum = 0.001;
rcc->pass1_wanted_bits = 0.001;
if (s->avctx->qblur > 1.0) {
av_log(s->avctx, AV_LOG_ERROR, "qblur too large\n");
return -1;
}
/* init stuff with the user specified complexity */
if (s->avctx->rc_initial_cplx) {
for (i = 0; i < 60 * 30; i++) {
double bits = s->avctx->rc_initial_cplx * (i / 10000.0 + 1.0) * s->mb_num;
RateControlEntry rce;
if (i % ((s->gop_size + 3) / 4) == 0)
rce.pict_type = AV_PICTURE_TYPE_I;
else if (i % (s->max_b_frames + 1))
rce.pict_type = AV_PICTURE_TYPE_B;
else
rce.pict_type = AV_PICTURE_TYPE_P;
rce.new_pict_type = rce.pict_type;
rce.mc_mb_var_sum = bits * s->mb_num / 100000;
rce.mb_var_sum = s->mb_num;
rce.qscale = FF_QP2LAMBDA * 2;
rce.f_code = 2;
rce.b_code = 1;
rce.misc_bits = 1;
if (s->pict_type == AV_PICTURE_TYPE_I) {
rce.i_count = s->mb_num;
rce.i_tex_bits = bits;
rce.p_tex_bits = 0;
rce.mv_bits = 0;
} else {
rce.i_count = 0; // FIXME we do know this approx
rce.i_tex_bits = 0;
rce.p_tex_bits = bits * 0.9;
rce.mv_bits = bits * 0.1;
}
rcc->i_cplx_sum[rce.pict_type] += rce.i_tex_bits * rce.qscale;
rcc->p_cplx_sum[rce.pict_type] += rce.p_tex_bits * rce.qscale;
rcc->mv_bits_sum[rce.pict_type] += rce.mv_bits;
rcc->frame_count[rce.pict_type]++;
get_qscale(s, &rce, rcc->pass1_wanted_bits / rcc->pass1_rc_eq_output_sum, i);
// FIXME misbehaves a little for variable fps
rcc->pass1_wanted_bits += s->bit_rate / get_fps(s->avctx);
}
}
}
return 0;
}
static void search_for_pns(AACEncContext *s, AVCodecContext *avctx, SingleChannelElement *sce)
{
FFPsyBand *band;
int w, g, w2, i;
int wlen = 1024 / sce->ics.num_windows;
int bandwidth, cutoff;
float *PNS = &s->scoefs[0*128], *PNS34 = &s->scoefs[1*128];
float *NOR34 = &s->scoefs[3*128];
uint8_t nextband[128];
const float lambda = s->lambda;
const float freq_mult = avctx->sample_rate*0.5f/wlen;
const float thr_mult = NOISE_LAMBDA_REPLACE*(100.0f/lambda);
const float spread_threshold = FFMIN(0.75f, NOISE_SPREAD_THRESHOLD*FFMAX(0.5f, lambda/100.f));
const float dist_bias = av_clipf(4.f * 120 / lambda, 0.25f, 4.0f);
const float pns_transient_energy_r = FFMIN(0.7f, lambda / 140.f);
int refbits = avctx->bit_rate * 1024.0 / avctx->sample_rate
/ ((avctx->flags & CODEC_FLAG_QSCALE) ? 2.0f : avctx->channels)
* (lambda / 120.f);
/** Keep this in sync with twoloop's cutoff selection */
float rate_bandwidth_multiplier = 1.5f;
int prev = -1000, prev_sf = -1;
int frame_bit_rate = (avctx->flags & CODEC_FLAG_QSCALE)
? (refbits * rate_bandwidth_multiplier * avctx->sample_rate / 1024)
: (avctx->bit_rate / avctx->channels);
frame_bit_rate *= 1.15f;
if (avctx->cutoff > 0) {
bandwidth = avctx->cutoff;
} else {
bandwidth = FFMAX(3000, AAC_CUTOFF_FROM_BITRATE(frame_bit_rate, 1, avctx->sample_rate));
}
cutoff = bandwidth * 2 * wlen / avctx->sample_rate;
memcpy(sce->band_alt, sce->band_type, sizeof(sce->band_type));
ff_init_nextband_map(sce, nextband);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
int wstart = w*128;
for (g = 0; g < sce->ics.num_swb; g++) {
int noise_sfi;
float dist1 = 0.0f, dist2 = 0.0f, noise_amp;
float pns_energy = 0.0f, pns_tgt_energy, energy_ratio, dist_thresh;
float sfb_energy = 0.0f, threshold = 0.0f, spread = 2.0f;
float min_energy = -1.0f, max_energy = 0.0f;
const int start = wstart+sce->ics.swb_offset[g];
const float freq = (start-wstart)*freq_mult;
const float freq_boost = FFMAX(0.88f*freq/NOISE_LOW_LIMIT, 1.0f);
if (freq < NOISE_LOW_LIMIT || (start-wstart) >= cutoff) {
if (!sce->zeroes[w*16+g])
prev_sf = sce->sf_idx[w*16+g];
continue;
}
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
sfb_energy += band->energy;
spread = FFMIN(spread, band->spread);
threshold += band->threshold;
if (!w2) {
min_energy = max_energy = band->energy;
} else {
min_energy = FFMIN(min_energy, band->energy);
max_energy = FFMAX(max_energy, band->energy);
}
}
/* Ramps down at ~8000Hz and loosens the dist threshold */
dist_thresh = av_clipf(2.5f*NOISE_LOW_LIMIT/freq, 0.5f, 2.5f) * dist_bias;
/* PNS is acceptable when all of these are true:
* 1. high spread energy (noise-like band)
* 2. near-threshold energy (high PE means the random nature of PNS content will be noticed)
* 3. on short window groups, all windows have similar energy (variations in energy would be destroyed by PNS)
*
* At this stage, point 2 is relaxed for zeroed bands near the noise threshold (hole avoidance is more important)
*/
if ((!sce->zeroes[w*16+g] && !ff_sfdelta_can_remove_band(sce, nextband, prev_sf, w*16+g)) ||
((sce->zeroes[w*16+g] || !sce->band_alt[w*16+g]) && sfb_energy < threshold*sqrtf(1.0f/freq_boost)) || spread < spread_threshold ||
(!sce->zeroes[w*16+g] && sce->band_alt[w*16+g] && sfb_energy > threshold*thr_mult*freq_boost) ||
min_energy < pns_transient_energy_r * max_energy ) {
sce->pns_ener[w*16+g] = sfb_energy;
if (!sce->zeroes[w*16+g])
prev_sf = sce->sf_idx[w*16+g];
continue;
}
pns_tgt_energy = sfb_energy*FFMIN(1.0f, spread*spread);
noise_sfi = av_clip(roundf(log2f(pns_tgt_energy)*2), -100, 155); /* Quantize */
noise_amp = -ff_aac_pow2sf_tab[noise_sfi + POW_SF2_ZERO]; /* Dequantize */
if (prev != -1000) {
int noise_sfdiff = noise_sfi - prev + SCALE_DIFF_ZERO;
if (noise_sfdiff < 0 || noise_sfdiff > 2*SCALE_MAX_DIFF) {
if (!sce->zeroes[w*16+g])
prev_sf = sce->sf_idx[w*16+g];
continue;
}
}
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
float band_energy, scale, pns_senergy;
const int start_c = (w+w2)*128+sce->ics.swb_offset[g];
band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
for (i = 0; i < sce->ics.swb_sizes[g]; i+=2) {
double rnd[2];
av_bmg_get(&s->lfg, rnd);
PNS[i+0] = (float)rnd[0];
PNS[i+1] = (float)rnd[1];
}
band_energy = s->fdsp->scalarproduct_float(PNS, PNS, sce->ics.swb_sizes[g]);
scale = noise_amp/sqrtf(band_energy);
s->fdsp->vector_fmul_scalar(PNS, PNS, scale, sce->ics.swb_sizes[g]);
pns_senergy = s->fdsp->scalarproduct_float(PNS, PNS, sce->ics.swb_sizes[g]);
pns_energy += pns_senergy;
abs_pow34_v(NOR34, &sce->coeffs[start_c], sce->ics.swb_sizes[g]);
abs_pow34_v(PNS34, PNS, sce->ics.swb_sizes[g]);
dist1 += quantize_band_cost(s, &sce->coeffs[start_c],
NOR34,
sce->ics.swb_sizes[g],
sce->sf_idx[(w+w2)*16+g],
sce->band_alt[(w+w2)*16+g],
lambda/band->threshold, INFINITY, NULL, NULL, 0);
/* Estimate rd on average as 5 bits for SF, 4 for the CB, plus spread energy * lambda/thr */
dist2 += band->energy/(band->spread*band->spread)*lambda*dist_thresh/band->threshold;
}
if (g && sce->band_type[w*16+g-1] == NOISE_BT) {
dist2 += 5;
} else {
dist2 += 9;
}
energy_ratio = pns_tgt_energy/pns_energy; /* Compensates for quantization error */
sce->pns_ener[w*16+g] = energy_ratio*pns_tgt_energy;
if (sce->zeroes[w*16+g] || !sce->band_alt[w*16+g] || (energy_ratio > 0.85f && energy_ratio < 1.25f && dist2 < dist1)) {
sce->band_type[w*16+g] = NOISE_BT;
sce->zeroes[w*16+g] = 0;
prev = noise_sfi;
} else {
if (!sce->zeroes[w*16+g])
prev_sf = sce->sf_idx[w*16+g];
}
}
}
}
static av_cold int encode_init(AVCodecContext* avc_context)
{
th_info t_info;
th_comment t_comment;
ogg_packet o_packet;
unsigned int offset;
TheoraContext *h = avc_context->priv_data;
uint32_t gop_size = avc_context->gop_size;
/* Set up the theora_info struct */
th_info_init(&t_info);
t_info.frame_width = FFALIGN(avc_context->width, 16);
t_info.frame_height = FFALIGN(avc_context->height, 16);
t_info.pic_width = avc_context->width;
t_info.pic_height = avc_context->height;
t_info.pic_x = 0;
t_info.pic_y = 0;
/* Swap numerator and denominator as time_base in AVCodecContext gives the
* time period between frames, but theora_info needs the framerate. */
t_info.fps_numerator = avc_context->time_base.den;
t_info.fps_denominator = avc_context->time_base.num;
if (avc_context->sample_aspect_ratio.num) {
t_info.aspect_numerator = avc_context->sample_aspect_ratio.num;
t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
} else {
t_info.aspect_numerator = 1;
t_info.aspect_denominator = 1;
}
if (avc_context->color_primaries == AVCOL_PRI_BT470M)
t_info.colorspace = TH_CS_ITU_REC_470M;
else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
t_info.colorspace = TH_CS_ITU_REC_470BG;
else
t_info.colorspace = TH_CS_UNSPECIFIED;
if (avc_context->pix_fmt == AV_PIX_FMT_YUV420P)
t_info.pixel_fmt = TH_PF_420;
else if (avc_context->pix_fmt == AV_PIX_FMT_YUV422P)
t_info.pixel_fmt = TH_PF_422;
else if (avc_context->pix_fmt == AV_PIX_FMT_YUV444P)
t_info.pixel_fmt = TH_PF_444;
else {
av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
return -1;
}
av_pix_fmt_get_chroma_sub_sample(avc_context->pix_fmt,
&h->uv_hshift, &h->uv_vshift);
if (avc_context->flags & CODEC_FLAG_QSCALE) {
/* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
Theora accepts a quality parameter p, which is:
* 0 <= p <=63
* an int value
*/
t_info.quality = av_clipf(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
t_info.target_bitrate = 0;
} else {
t_info.target_bitrate = avc_context->bit_rate;
t_info.quality = 0;
}
/* Now initialise libtheora */
h->t_state = th_encode_alloc(&t_info);
if (!h->t_state) {
av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
return -1;
}
h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
/* Clear up theora_info struct */
th_info_clear(&t_info);
if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
&gop_size, sizeof(gop_size))) {
av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
return -1;
}
// need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
if (avc_context->flags & CODEC_FLAG_PASS1) {
if (get_stats(avc_context, 0))
return -1;
} else if (avc_context->flags & CODEC_FLAG_PASS2) {
if (submit_stats(avc_context))
return -1;
}
/*
Output first header packet consisting of theora
header, comment, and tables.
Each one is prefixed with a 16bit size, then they
are concatenated together into libavcodec's extradata.
*/
offset = 0;
/* Headers */
th_comment_init(&t_comment);
while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
if (concatenate_packet(&offset, avc_context, &o_packet))
return -1;
th_comment_clear(&t_comment);
/* Set up the output AVFrame */
avc_context->coded_frame = av_frame_alloc();
return 0;
}
static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[32][2], int is34)
{
float power[34][PS_QMF_TIME_SLOTS] = {{0}};
float transient_gain[34][PS_QMF_TIME_SLOTS];
float *peak_decay_nrg = ps->peak_decay_nrg;
float *power_smooth = ps->power_smooth;
float *peak_decay_diff_smooth = ps->peak_decay_diff_smooth;
float (*delay)[PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2] = ps->delay;
float (*ap_delay)[PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2] = ps->ap_delay;
const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20;
const float peak_decay_factor = 0.76592833836465f;
const float transient_impact = 1.5f;
const float a_smooth = 0.25f; ///< Smoothing coefficient
int i, k, m, n;
int n0 = 0, nL = 32;
static const int link_delay[] = { 3, 4, 5 };
static const float a[] = { 0.65143905753106f,
0.56471812200776f,
0.48954165955695f };
if (is34 != ps->is34bands_old) {
memset(ps->peak_decay_nrg, 0, sizeof(ps->peak_decay_nrg));
memset(ps->power_smooth, 0, sizeof(ps->power_smooth));
memset(ps->peak_decay_diff_smooth, 0, sizeof(ps->peak_decay_diff_smooth));
memset(ps->delay, 0, sizeof(ps->delay));
memset(ps->ap_delay, 0, sizeof(ps->ap_delay));
}
for (n = n0; n < nL; n++) {
for (k = 0; k < NR_BANDS[is34]; k++) {
int i = k_to_i[k];
power[i][n] += s[k][n][0] * s[k][n][0] + s[k][n][1] * s[k][n][1];
}
}
//Transient detection
for (i = 0; i < NR_PAR_BANDS[is34]; i++) {
for (n = n0; n < nL; n++) {
float decayed_peak = peak_decay_factor * peak_decay_nrg[i];
float denom;
peak_decay_nrg[i] = FFMAX(decayed_peak, power[i][n]);
power_smooth[i] += a_smooth * (power[i][n] - power_smooth[i]);
peak_decay_diff_smooth[i] += a_smooth * (peak_decay_nrg[i] - power[i][n] - peak_decay_diff_smooth[i]);
denom = transient_impact * peak_decay_diff_smooth[i];
transient_gain[i][n] = (denom > power_smooth[i]) ?
power_smooth[i] / denom : 1.0f;
}
}
//Decorrelation and transient reduction
// PS_AP_LINKS - 1
// -----
// | | Q_fract_allpass[k][m]*z^-link_delay[m] - a[m]*g_decay_slope[k]
//H[k][z] = z^-2 * phi_fract[k] * | | ----------------------------------------------------------------
// | | 1 - a[m]*g_decay_slope[k]*Q_fract_allpass[k][m]*z^-link_delay[m]
// m = 0
//d[k][z] (out) = transient_gain_mapped[k][z] * H[k][z] * s[k][z]
for (k = 0; k < NR_ALLPASS_BANDS[is34]; k++) {
int b = k_to_i[k];
float g_decay_slope = 1.f - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]);
float ag[PS_AP_LINKS];
g_decay_slope = av_clipf(g_decay_slope, 0.f, 1.f);
memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
for (m = 0; m < PS_AP_LINKS; m++) {
memcpy(ap_delay[k][m], ap_delay[k][m]+numQMFSlots, 5*sizeof(ap_delay[k][m][0]));
ag[m] = a[m] * g_decay_slope;
}
for (n = n0; n < nL; n++) {
float in_re = delay[k][n+PS_MAX_DELAY-2][0] * phi_fract[is34][k][0] -
delay[k][n+PS_MAX_DELAY-2][1] * phi_fract[is34][k][1];
float in_im = delay[k][n+PS_MAX_DELAY-2][0] * phi_fract[is34][k][1] +
delay[k][n+PS_MAX_DELAY-2][1] * phi_fract[is34][k][0];
for (m = 0; m < PS_AP_LINKS; m++) {
float a_re = ag[m] * in_re;
float a_im = ag[m] * in_im;
float link_delay_re = ap_delay[k][m][n+5-link_delay[m]][0];
float link_delay_im = ap_delay[k][m][n+5-link_delay[m]][1];
float fractional_delay_re = Q_fract_allpass[is34][k][m][0];
float fractional_delay_im = Q_fract_allpass[is34][k][m][1];
ap_delay[k][m][n+5][0] = in_re;
ap_delay[k][m][n+5][1] = in_im;
in_re = link_delay_re * fractional_delay_re - link_delay_im * fractional_delay_im - a_re;
in_im = link_delay_re * fractional_delay_im + link_delay_im * fractional_delay_re - a_im;
ap_delay[k][m][n+5][0] += ag[m] * in_re;
ap_delay[k][m][n+5][1] += ag[m] * in_im;
}
out[k][n][0] = transient_gain[b][n] * in_re;
out[k][n][1] = transient_gain[b][n] * in_im;
}
}
for (; k < SHORT_DELAY_BAND[is34]; k++) {
memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
for (n = n0; n < nL; n++) {
//H = delay 14
out[k][n][0] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-14][0];
out[k][n][1] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-14][1];
}
}
for (; k < NR_BANDS[is34]; k++) {
memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
for (n = n0; n < nL; n++) {
//H = delay 1
out[k][n][0] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-1][0];
out[k][n][1] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-1][1];
}
}
}
