improve audio processing pipeline

2 files changed, 201 insertions, 8 deletions

diff --git a/src/acvt.rs b/src/acvt.rs
new file mode 100644
index 0000000..5343039
--- /dev/null
+++ b/src/acvt.rs
@@ -0,0 +1,177 @@
+use nihav_core::frame::*;
+use nihav_core::soundcvt::*;
+
+struct AudioQueue<T> {
+    start:      usize,
+    end:        usize,
+    stride:     usize,
+    channels:   usize,
+    data:       Vec<T>,
+    ileaved:    bool,
+}
+
+impl<T:Clone+Copy+From<u8>> AudioQueue<T> {
+    fn new(channels: usize, rec_size: usize, ileaved: bool) -> Self {
+        Self {
+            start:      0,
+            end:        0,
+            stride:     if ileaved { rec_size * channels } else { rec_size },
+            channels,
+            ileaved,
+            data:       vec![0.into(); rec_size * channels],
+        }
+    }
+    fn get_cur_size(&self) -> usize { self.end - self.start }
+    fn get_cur_avail(&self) -> usize { self.stride - self.end }
+    fn get_potentially_avail(&self) -> usize { self.stride - self.get_cur_size() }
+    fn read(&mut self, src: &NAAudioBuffer<T>) {
+        let mut to_copy = src.get_length();
+        if self.ileaved {
+            to_copy *= self.channels;
+        }
+        if self.get_cur_avail() < to_copy {
+            if self.get_potentially_avail() >= to_copy {
+                self.renorm();
+            } else {
+                let new_len = (self.stride * 2).max(self.get_cur_size() + src.get_length());
+                let mut new_buf = vec![0.into(); new_len * self.channels];
+                let new_stride = if !self.ileaved { new_len } else { new_len * self.channels };
+
+                let old_len = self.get_cur_size();
+                let new_len = src.get_length();
+                for (dst, (old, new)) in new_buf.chunks_exact_mut(new_stride).zip(
+                        self.data.chunks_exact(self.stride).zip(
+                            src.get_data().chunks(src.get_stride()))) {
+                    dst[..old_len].copy_from_slice(&old[self.start..self.end]);
+                    dst[old_len..][..new_len].copy_from_slice(&new[..new_len]);
+                }
+                self.data = new_buf;
+                self.stride = new_stride;
+                self.start = 0;
+                self.end = old_len + new_len;
+                return;
+            }
+        }
+        for (dst, src) in self.data.chunks_exact_mut(self.stride).zip(src.get_data().chunks_exact(src.get_stride())) {
+            dst[self.end..][..to_copy].copy_from_slice(&src[..to_copy]);
+        }
+        self.end += to_copy;
+    }
+    fn write(&mut self, dbuf: &mut NAAudioBuffer<T>) {
+        let dst_len = dbuf.get_length();
+        let dst_stride = dbuf.get_stride();
+        let dst = dbuf.get_data_mut().unwrap();
+
+        for (dst, src) in dst.chunks_mut(dst_stride).zip(self.data.chunks_exact(self.stride)) {
+            dst[..dst_len].copy_from_slice(&src[self.start..][..dst_len]);
+        }
+        self.start += dst_len;
+    }
+    fn renorm(&mut self) {
+        if self.start == 0 {
+            return;
+        }
+
+        let move_size = self.end - self.start;
+        if move_size > 0 {
+            for chan in self.data.chunks_exact_mut(self.stride) {
+                for i in 0..move_size {
+                    chan[i] = chan[self.start + i];
+                }
+            }
+        }
+        self.end -= self.start;
+        self.start = 0;
+    }
+}
+
+enum AudioDataType {
+    U8(AudioQueue<u8>),
+    I16(AudioQueue<i16>),
+    I32(AudioQueue<i32>),
+    F32(AudioQueue<f32>),
+    Packed(AudioQueue<u8>),
+}
+
+impl AudioDataType {
+    fn get_length(&self) -> usize {
+        match self {
+            AudioDataType::U8(ref queue) => queue.get_cur_size(),
+            AudioDataType::I16(ref queue) => queue.get_cur_size(),
+            AudioDataType::I32(ref queue) => queue.get_cur_size(),
+            AudioDataType::F32(ref queue) => queue.get_cur_size(),
+            AudioDataType::Packed(ref queue) => queue.get_cur_size(),
+        }
+    }
+}
+
+pub struct AudioConverter {
+    queue:      AudioDataType,
+    dst_fmt:    NAAudioInfo,
+    dst_chmap:  NAChannelMap,
+    apts:       Option<u64>,
+}
+
+impl AudioConverter {
+    pub fn new(_sinfo: &NAAudioInfo, dinfo: &NAAudioInfo, dst_chmap: NAChannelMap) -> Self {
+        let ch = usize::from(dinfo.channels);
+        let size = dinfo.block_len * 2;
+        let il = !dinfo.format.planar;
+        let queue = match (dinfo.format.bits, dinfo.format.float, dinfo.format.signed) {
+                ( 8, false, false) => AudioDataType::U8(AudioQueue::new(ch, size, il)),
+                (16, false, true)  => AudioDataType::I16(AudioQueue::new(ch, size, il)),
+                (32, false, true)  => AudioDataType::I32(AudioQueue::new(ch, size, il)),
+                (32, true, _)      => AudioDataType::F32(AudioQueue::new(ch, size, il)),
+                _ => AudioDataType::Packed(AudioQueue::new(ch, size, il)),
+            };
+        Self {
+            queue,
+            dst_fmt:    *dinfo,
+            dst_chmap,
+            apts:       None,
+        }
+    }
+    pub fn queue_frame(&mut self, buf: NABufferType, tinfo: NATimeInfo) -> bool {
+        let ret = convert_audio_frame(&buf, &self.dst_fmt, &self.dst_chmap);
+        if let Ok(dbuf) = ret {
+            if self.apts.is_none() && tinfo.get_pts().is_some() {
+                self.apts = tinfo.get_pts();
+            }
+            match (&mut self.queue, dbuf) {
+                (AudioDataType::U8(ref mut queue),  NABufferType::AudioU8(ref buf))  => queue.read(buf),
+                (AudioDataType::I16(ref mut queue), NABufferType::AudioI16(ref buf)) => queue.read(buf),
+                (AudioDataType::I32(ref mut queue), NABufferType::AudioI32(ref buf)) => queue.read(buf),
+                (AudioDataType::F32(ref mut queue), NABufferType::AudioF32(ref buf)) => queue.read(buf),
+                (AudioDataType::Packed(ref mut queue), NABufferType::AudioPacked(ref buf)) => queue.read(buf),
+                _ => unimplemented!(),
+            };
+            true
+        } else {
+            false
+        }
+    }
+    pub fn get_frame(&mut self, info: NACodecInfoRef) -> Option<NAFrame> {
+        if self.queue.get_length() >= self.dst_fmt.block_len {
+            if let Ok(mut abuf) = alloc_audio_buffer(self.dst_fmt, self.dst_fmt.block_len, self.dst_chmap.clone()) {
+                match (&mut self.queue, &mut abuf) {
+                    (AudioDataType::U8(ref mut queue),  NABufferType::AudioU8(ref mut buf))  => queue.write(buf),
+                    (AudioDataType::I16(ref mut queue), NABufferType::AudioI16(ref mut buf)) => queue.write(buf),
+                    (AudioDataType::I32(ref mut queue), NABufferType::AudioI32(ref mut buf)) => queue.write(buf),
+                    (AudioDataType::F32(ref mut queue), NABufferType::AudioF32(ref mut buf)) => queue.write(buf),
+                    (AudioDataType::Packed(ref mut queue), NABufferType::AudioPacked(ref mut buf)) => queue.write(buf),
+                    _ => unimplemented!(),
+                };
+                let tinfo = NATimeInfo::new(self.apts, None, Some(self.dst_fmt.block_len as u64), 1, self.dst_fmt.sample_rate);
+                if let Some(ref mut val) = self.apts {
+                    *val += self.dst_fmt.block_len as u64;
+                }
+                Some(NAFrame::new(tinfo, FrameType::I, true, info, abuf))
+            } else {
+                println!(" failed to allocate audio frame");
+                None
+            }
+        } else {
+            None
+        }
+    }
+}
diff --git a/src/main.rs b/src/main.rs
index a2889a8..ff19ae1 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -13,7 +13,6 @@ use nihav_core::demuxers::*;
 use nihav_core::muxers::*;
 use nihav_core::reorder::*;
 use nihav_core::scale::*;
-use nihav_core::soundcvt::*;
 use nihav_registry::detect;
 use nihav_registry::register;
 use std::env;
@@ -22,6 +21,8 @@ use std::time::{Duration, Instant};
 mod demux;
 use crate::demux::*;
 mod null;
+mod acvt;
+use crate::acvt::*;
 
 fn format_time(ms: u64) -> String {
     let s = ms / 1000;
@@ -102,7 +103,7 @@ struct OutputStreamOptions {
 
 enum OutputConvert {
     Video(NAScale, NABufferType),
-    Audio(NAAudioInfo, NAChannelMap),
+    Audio(AudioConverter),
     None,
 }
 
@@ -549,8 +550,9 @@ println!("can't generate default channel map for {} channels", dainfo.channels);
                                             return false;
                                         },
                                     };
+                                let acvt = AudioConverter::new(sainfo, dainfo, dchmap);
 //todo channelmap
-                                OutputConvert::Audio(*dainfo, dchmap)
+                                OutputConvert::Audio(acvt)
                             }
                         },
                         _ => OutputConvert::None,
@@ -627,7 +629,8 @@ println!("can't generate default channel map for {} channels", dainfo.channels);
                                     },
                                 };
 //todo channelmap
-                            OutputConvert::Audio(*dainfo, dchmap)
+                            let acvt = AudioConverter::new(sainfo, dainfo, dchmap);
+                            OutputConvert::Audio(acvt)
                         }
                     },
                     _ => OutputConvert::None,
@@ -766,13 +769,12 @@ fn encode_frame(dst_id: u32, encoder: &mut Box<dyn NAEncoder>, cvt: &mut OutputC
                 }
                 dbuf.clone()
             },
-            OutputConvert::Audio(ref dinfo, ref dchmap) => {
-                let ret = convert_audio_frame(&buf, dinfo, dchmap);
-                if ret.is_err() {
+            OutputConvert::Audio(ref mut acvt) => {
+                if !acvt.queue_frame(buf, frm.get_time_information()) {
                     println!("error converting audio for stream {}", dst_id);
                     return false;
                 }
-                ret.unwrap()
+                return true;
             },
         }
     };
@@ -1202,6 +1204,7 @@ println!("stream {} - {} {}", i, s, info.get_name());
                         break;
                     }
                     let frm = ret.unwrap();
+                    let tinfo = frm.get_info();
                     reorderer.add_frame(frm);
                     while let Some(frm) = reorderer.get_frame() {
                         if !encode_frame(dst_id, encoder, cvt, frm, &transcoder.scale_opts) {
@@ -1218,6 +1221,19 @@ println!("stream {} - {} {}", i, s, info.get_name());
                             mux.mux_frame(pkt).unwrap();
                         }
                     }
+                    if let OutputConvert::Audio(ref mut acvt) = cvt {
+                        while let Some(ofrm) = acvt.get_frame(tinfo.clone()) {
+                            if encoder.encode(&ofrm).is_err() {
+                                break;
+                            }
+                            while let Ok(Some(pkt)) = encoder.get_packet() {
+                                if transcoder.end != NATimePoint::None && !pkt.ts.less_than(transcoder.end) { break 'main_loop; }
+                                let pkt_size = pkt.get_buffer().len();
+                                adata_size += pkt_size;
+                                mux.mux_frame(pkt).unwrap();
+                            }
+                        }
+                    }
                 } else {
                     println!("no decoder for stream {}", src_id);
                     break;