asyncts: fix the asyncts behavior when using the first_pts option

Currently it will do padding, but it does not properly handle
start-of-stream trimming as documented.

libavfilter/af_asyncts.c

@@ -33,6 +33,8 @@ typedef struct ASyncContext {
     AVAudioResampleContext *avr;
     int64_t pts;            ///< timestamp in samples of the first sample in fifo
     int min_delta;          ///< pad/trim min threshold in samples
+    int first_frame;        ///< 1 until filter_frame() has processed at least 1 frame with a pts != AV_NOPTS_VALUE
+    int64_t first_pts;      ///< user-specified first expected pts, in samples
 
     /* options */
     int resample;
@@ -50,7 +52,7 @@ static const AVOption options[] = {
     { "min_delta",  "Minimum difference between timestamps and audio data "
                     "(in seconds) to trigger padding/trimmin the data.",        OFFSET(min_delta_sec), AV_OPT_TYPE_FLOAT, { .dbl = 0.1 }, 0, INT_MAX, A },
     { "max_comp",   "Maximum compensation in samples per second.",              OFFSET(max_comp),      AV_OPT_TYPE_INT,   { .i64 = 500 }, 0, INT_MAX, A },
-    { "first_pts",  "Assume the first pts should be this value.",               OFFSET(pts),           AV_OPT_TYPE_INT64, { .i64 = AV_NOPTS_VALUE }, INT64_MIN, INT64_MAX, A },
+    { "first_pts",  "Assume the first pts should be this value.",               OFFSET(first_pts),     AV_OPT_TYPE_INT64, { .i64 = AV_NOPTS_VALUE }, INT64_MIN, INT64_MAX, A },
     { NULL },
 };
 
@@ -75,6 +77,9 @@ static int init(AVFilterContext *ctx, const char *args)
     }
     av_opt_free(s);
 
+    s->pts         = AV_NOPTS_VALUE;
+    s->first_frame = 1;
+
     return 0;
 }
 
@@ -122,6 +127,20 @@ static int64_t get_delay(ASyncContext *s)
     return avresample_available(s->avr) + avresample_get_delay(s->avr);
 }
 
+static void handle_trimming(AVFilterContext *ctx)
+{
+    ASyncContext *s = ctx->priv;
+
+    if (s->pts < s->first_pts) {
+        int delta = FFMIN(s->first_pts - s->pts, avresample_available(s->avr));
+        av_log(ctx, AV_LOG_VERBOSE, "Trimming %d samples from start\n",
+               delta);
+        avresample_read(s->avr, NULL, delta);
+        s->pts += delta;
+    } else if (s->first_frame)
+        s->pts = s->first_pts;
+}
+
 static int request_frame(AVFilterLink *link)
 {
     AVFilterContext *ctx = link->src;
@@ -134,7 +153,11 @@ static int request_frame(AVFilterLink *link)
         ret = ff_request_frame(ctx->inputs[0]);
 
     /* flush the fifo */
-    if (ret == AVERROR_EOF && (nb_samples = get_delay(s))) {
+    if (ret == AVERROR_EOF) {
+        if (s->first_pts != AV_NOPTS_VALUE)
+            handle_trimming(ctx);
+
+        if (nb_samples = get_delay(s)) {
         AVFilterBufferRef *buf = ff_get_audio_buffer(link, AV_PERM_WRITE,
                                                      nb_samples);
         if (!buf)
@@ -148,6 +171,7 @@ static int request_frame(AVFilterLink *link)
 
         buf->pts = s->pts;
         return ff_filter_frame(link, buf);
+        }
     }
 
     return ret;
@@ -185,12 +209,18 @@ static int filter_frame(AVFilterLink *inlink, AVFilterBufferRef *buf)
         return write_to_fifo(s, buf);
     }
 
+    if (s->first_pts != AV_NOPTS_VALUE) {
+        handle_trimming(ctx);
+        if (!avresample_available(s->avr))
+            return write_to_fifo(s, buf);
+    }
+
     /* when we have two timestamps, compute how many samples would we have
      * to add/remove to get proper sync between data and timestamps */
     delta    = pts - s->pts - get_delay(s);
     out_size = avresample_available(s->avr);
 
-    if (labs(delta) > s->min_delta) {
+    if (labs(delta) > s->min_delta || (s->first_frame && delta)) {
         av_log(ctx, AV_LOG_VERBOSE, "Discontinuity - %"PRId64" samples.\n", delta);
         out_size = av_clipl_int32((int64_t)out_size + delta);
     } else {
@@ -210,18 +240,33 @@ static int filter_frame(AVFilterLink *inlink, AVFilterBufferRef *buf)
             goto fail;
         }
 
-        avresample_read(s->avr, buf_out->extended_data, out_size);
-        buf_out->pts = s->pts;
+        if (s->first_frame && delta > 0) {
+            int ch;
 
-        if (delta > 0) {
-            av_samples_set_silence(buf_out->extended_data, out_size - delta,
-                                   delta, nb_channels, buf->format);
+            av_samples_set_silence(buf_out->extended_data, 0, delta,
+                                   nb_channels, buf->format);
+
+            for (ch = 0; ch < nb_channels; ch++)
+                buf_out->extended_data[ch] += delta;
+
+            avresample_read(s->avr, buf_out->extended_data, out_size);
+
+            for (ch = 0; ch < nb_channels; ch++)
+                buf_out->extended_data[ch] -= delta;
+        } else {
+            avresample_read(s->avr, buf_out->extended_data, out_size);
+
+            if (delta > 0) {
+                av_samples_set_silence(buf_out->extended_data, out_size - delta,
+                                       delta, nb_channels, buf->format);
+            }
         }
+        buf_out->pts = s->pts;
         ret = ff_filter_frame(outlink, buf_out);
         if (ret < 0)
             goto fail;
         s->got_output = 1;
-    } else {
+    } else if (avresample_available(s->avr)) {
         av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
                "whole buffer.\n");
     }
@@ -233,6 +278,7 @@ static int filter_frame(AVFilterLink *inlink, AVFilterBufferRef *buf)
     ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
                              buf->linesize[0], buf->audio->nb_samples);
 
+    s->first_frame = 0;
 fail:
     avfilter_unref_buffer(buf);