static guint
fill_queue (GstScaletempo * scaletempo, GstBuffer * buf_in, guint offset)
{
GstScaletempoPrivate *p = GST_SCALETEMPO_GET_PRIVATE (scaletempo);
guint bytes_in = GST_BUFFER_SIZE (buf_in) - offset;
guint offset_unchanged = offset;
if (p->bytes_to_slide > 0) {
if (p->bytes_to_slide < p->bytes_queued) {
guint bytes_in_move = p->bytes_queued - p->bytes_to_slide;
memmove (p->buf_queue, p->buf_queue + p->bytes_to_slide, bytes_in_move);
p->bytes_to_slide = 0;
p->bytes_queued = bytes_in_move;
} else {
guint bytes_in_skip;
p->bytes_to_slide -= p->bytes_queued;
bytes_in_skip = MIN (p->bytes_to_slide, bytes_in);
p->bytes_queued = 0;
p->bytes_to_slide -= bytes_in_skip;
offset += bytes_in_skip;
bytes_in -= bytes_in_skip;
}
}
if (bytes_in > 0) {
guint bytes_in_copy = MIN (p->bytes_queue_max - p->bytes_queued, bytes_in);
memcpy (p->buf_queue + p->bytes_queued,
GST_BUFFER_DATA (buf_in) + offset, bytes_in_copy);
p->bytes_queued += bytes_in_copy;
offset += bytes_in_copy;
}
return offset - offset_unchanged;
}
static guint
best_overlap_offset_float (GstScaletempo * scaletempo)
{
GstScaletempoPrivate *p = GST_SCALETEMPO_GET_PRIVATE (scaletempo);
gfloat *pw, *po, *ppc, *search_start;
gfloat best_corr = G_MININT;
guint best_off = 0;
gint i, off;
pw = p->table_window;
po = p->buf_overlap;
po += p->samples_per_frame;
ppc = p->buf_pre_corr;
for (i = p->samples_per_frame; i < p->samples_overlap; i++) {
*ppc++ = *pw++ * *po++;
}
search_start = (gfloat *) p->buf_queue + p->samples_per_frame;
for (off = 0; off < p->frames_search; off++) {
gfloat corr = 0;
gfloat *ps = search_start;
ppc = p->buf_pre_corr;
for (i = p->samples_per_frame; i < p->samples_overlap; i++) {
corr += *ppc++ * *ps++;
}
if (corr > best_corr) {
best_corr = corr;
best_off = off;
}
search_start += p->samples_per_frame;
}
return best_off * p->bytes_per_frame;
}
static void
output_overlap_s16 (GstScaletempo * scaletempo,
gpointer buf_out, guint bytes_off)
{
GstScaletempoPrivate *p = GST_SCALETEMPO_GET_PRIVATE (scaletempo);
gint16 *pout = buf_out;
gint32 *pb = p->table_blend;
gint16 *po = p->buf_overlap;
gint16 *pin = (gint16 *) (p->buf_queue + bytes_off);
gint i;
for (i = 0; i < p->samples_overlap; i++) {
*pout++ = *po - ((*pb++ * (*po - *pin++)) >> 16);
po++;
}
}
static guint
best_overlap_offset_s16 (GstScaletempo * scaletempo)
{
GstScaletempoPrivate *p = GST_SCALETEMPO_GET_PRIVATE (scaletempo);
gint32 *pw, *ppc;
gint16 *po, *search_start;
gint64 best_corr = G_MININT64;
guint best_off = 0;
guint off;
glong i;
pw = p->table_window;
po = p->buf_overlap;
po += p->samples_per_frame;
ppc = p->buf_pre_corr;
for (i = p->samples_per_frame; i < p->samples_overlap; i++) {
*ppc++ = (*pw++ * *po++) >> 15;
}
search_start = (gint16 *) p->buf_queue + p->samples_per_frame;
for (off = 0; off < p->frames_search; off++) {
gint64 corr = 0;
gint16 *ps = search_start;
ppc = p->buf_pre_corr;
ppc += p->samples_overlap - p->samples_per_frame;
ps += p->samples_overlap - p->samples_per_frame;
i = -((glong) p->samples_overlap - (glong) p->samples_per_frame);
do {
corr += ppc[i + 0] * ps[i + 0];
corr += ppc[i + 1] * ps[i + 1];
corr += ppc[i + 2] * ps[i + 2];
corr += ppc[i + 3] * ps[i + 3];
i += 4;
} while (i < 0);
if (corr > best_corr) {
best_corr = corr;
best_off = off;
}
search_start += p->samples_per_frame;
}
return best_off * p->bytes_per_frame;
}
static guint
fill_queue (GstScaletempo * scaletempo, GstBuffer * buf_in, guint offset)
{
GstScaletempoPrivate *p = GST_SCALETEMPO_GET_PRIVATE (scaletempo);
guint bytes_in = gst_buffer_get_size (buf_in) - offset;
guint offset_unchanged = offset;
GstMapInfo map;
gst_buffer_map (buf_in, &map, GST_MAP_READ);
if (p->bytes_to_slide > 0) {
if (p->bytes_to_slide < p->bytes_queued) {
guint bytes_in_move = p->bytes_queued - p->bytes_to_slide;
memmove (p->buf_queue, p->buf_queue + p->bytes_to_slide, bytes_in_move);
p->bytes_to_slide = 0;
p->bytes_queued = bytes_in_move;
} else {
guint bytes_in_skip;
p->bytes_to_slide -= p->bytes_queued;
bytes_in_skip = MIN (p->bytes_to_slide, bytes_in);
p->bytes_queued = 0;
p->bytes_to_slide -= bytes_in_skip;
offset += bytes_in_skip;
bytes_in -= bytes_in_skip;
}
}
if (bytes_in > 0) {
guint bytes_in_copy = MIN (p->bytes_queue_max - p->bytes_queued, bytes_in);
memcpy (p->buf_queue + p->bytes_queued, map.data + offset, bytes_in_copy);
p->bytes_queued += bytes_in_copy;
offset += bytes_in_copy;
}
gst_buffer_unmap (buf_in, &map);
return offset - offset_unchanged;
}
static void
reinit_buffers (GstScaletempo * scaletempo)
{
GstScaletempoPrivate *p = GST_SCALETEMPO_GET_PRIVATE (scaletempo);
gint i, j;
guint frames_overlap;
guint new_size;
guint frames_stride = p->ms_stride * p->sample_rate / 1000.0;
p->bytes_stride = frames_stride * p->bytes_per_frame;
/* overlap */
frames_overlap = frames_stride * p->percent_overlap;
if (frames_overlap < 1) { /* if no overlap */
p->bytes_overlap = 0;
p->bytes_standing = p->bytes_stride;
p->samples_standing = p->bytes_standing / p->bytes_per_sample;
p->output_overlap = NULL;
} else {
guint prev_overlap = p->bytes_overlap;
p->bytes_overlap = frames_overlap * p->bytes_per_frame;
p->samples_overlap = frames_overlap * p->samples_per_frame;
p->bytes_standing = p->bytes_stride - p->bytes_overlap;
p->samples_standing = p->bytes_standing / p->bytes_per_sample;
p->buf_overlap = g_realloc (p->buf_overlap, p->bytes_overlap);
p->table_blend = g_realloc (p->table_blend, p->samples_overlap * 4); /* sizeof (gint32|gfloat) */
if (p->bytes_overlap > prev_overlap) {
memset ((guint8 *) p->buf_overlap + prev_overlap, 0,
p->bytes_overlap - prev_overlap);
}
if (p->use_int) {
gint32 *pb = p->table_blend;
gint64 blend = 0;
for (i = 0; i < frames_overlap; i++) {
gint32 v = blend / frames_overlap;
for (j = 0; j < p->samples_per_frame; j++) {
*pb++ = v;
}
blend += 65535; /* 2^16 */
}
p->output_overlap = output_overlap_s16;
} else {
gfloat *pb = p->table_blend;
gfloat t = (gfloat) frames_overlap;
for (i = 0; i < frames_overlap; i++) {
gfloat v = i / t;
for (j = 0; j < p->samples_per_frame; j++) {
*pb++ = v;
}
}
p->output_overlap = output_overlap_float;
}
}
/* best overlap */
p->frames_search =
(frames_overlap <= 1) ? 0 : p->ms_search * p->sample_rate / 1000.0;
if (p->frames_search < 1) { /* if no search */
p->best_overlap_offset = NULL;
} else {
guint bytes_pre_corr = (p->samples_overlap - p->samples_per_frame) * 4; /* sizeof (gint32|gfloat) */
p->buf_pre_corr =
g_realloc (p->buf_pre_corr, bytes_pre_corr + UNROLL_PADDING);
p->table_window = g_realloc (p->table_window, bytes_pre_corr);
if (p->use_int) {
gint64 t = frames_overlap;
gint32 n = 8589934588LL / (t * t); /* 4 * (2^31 - 1) / t^2 */
gint32 *pw;
memset ((guint8 *) p->buf_pre_corr + bytes_pre_corr, 0, UNROLL_PADDING);
pw = p->table_window;
for (i = 1; i < frames_overlap; i++) {
gint32 v = (i * (t - i) * n) >> 15;
for (j = 0; j < p->samples_per_frame; j++) {
*pw++ = v;
}
}
p->best_overlap_offset = best_overlap_offset_s16;
} else {