Rebase fixups: drop helpers that moved to the parent PR, fix renames

- get_swr_output_num_samples_bound is now declared and defined by the parent
  commit, so remove this branch's copy.
- swr_convert_to_tensor keeps the parent's `const uint8_t**` parameter, which
  is what compiles against FFmpeg 4-6 as well as 7+.
- PacketDecoder became AudioPacketDecoder in #1666; update the converter tests.

Author: NicolasHug

src/torchcodec/_core/FFMPEGCommon.cpp

@@ -640,26 +640,6 @@ AVFilterContext* create_av_filter_context_with_options(
   return av_filter_context;
 }
 
-int64_t get_swr_output_num_samples_bound(
-    const UniqueSwrContext& swr_context,
-    int num_src_samples,
-    int src_sample_rate,
-    int out_sample_rate) {
-  if (src_sample_rate == out_sample_rate) {
-    return num_src_samples;
-  }
-  // `swr_convert()` will likely not produce this many samples: it buffers the
-  // last few, because those require future input. That's why callers must
-  // narrow to what it actually returned. We could also use
-  // `swr_get_out_samples()`, but empirically `av_rescale_rnd()` gives a
-  // tighter bound.
-  return av_rescale_rnd(
-      swr_get_delay(swr_context.get(), src_sample_rate) + num_src_samples,
-      out_sample_rate,
-      src_sample_rate,
-      AV_ROUND_UP);
-}
-
 UniqueAVFrame convert_audio_av_frame_samples(
     const UniqueSwrContext& swr_context,
     const AVFrame& src_av_frame,

src/torchcodec/_core/FFMPEGCommon.h

@@ -349,15 +349,6 @@ UniqueAVFrame convert_audio_av_frame_samples(
     int desired_num_channels,
     int num_samples_to_skip = 0);
 
-// Upper bound on the number of samples swr_convert() can emit for
-// num_src_samples of input, including whatever it still holds buffered. It is
-// only a bound: narrow the output down to what swr_convert() returned.
-int64_t get_swr_output_num_samples_bound(
-    const UniqueSwrContext& swr_context,
-    int num_src_samples,
-    int src_sample_rate,
-    int out_sample_rate);
-
 // Returns true if sws_scale can handle unaligned data.
 bool can_sws_scale_handle_unaligned_data();
 

test/test_decoders.py

@@ -5231,7 +5231,7 @@ def test_audio_decoder_mpeg_ps_resync_after_seek(self):
     @staticmethod
     def _convert_audio(asset, drain=True, **converter_kwargs):
         demuxer = AudioDemuxer(asset.path)
-        decoder = PacketDecoder(demuxer)
+        decoder = AudioPacketDecoder(demuxer)
         converter = AudioConverter(**converter_kwargs)
 
         chunks = []
@@ -5345,7 +5345,7 @@ def test_audio_converter_reset_allows_another_stream(self):
 
         def convert_all(asset):
             demuxer = AudioDemuxer(asset.path)
-            decoder = PacketDecoder(demuxer)
+            decoder = AudioPacketDecoder(demuxer)
             chunks = []
             for packet in demuxer:
                 chunks += [converter.convert(raw) for raw in decoder.decode(packet)]
@@ -5402,7 +5402,7 @@ def test_audio_pipeline_after_seek(self):
         ).shape[1]
 
         demuxer = AudioDemuxer(NASA_AUDIO_MP3.path)
-        decoder = PacketDecoder(demuxer)
+        decoder = AudioPacketDecoder(demuxer)
         converter = AudioConverter(sample_rate=16_000)
         demuxer.seek(seek_seconds)
         decoder.reset()