Move audio/video converter out of reader/decoder

Makefile

@@ -11,11 +11,11 @@ XAV_READER_SO = $(PRIV_DIR)/libxavreader.so
 # uncomment to compile with debug logs
 XAV_DEBUG_LOGS = -DXAV_DEBUG=1
 
-DECODER_HEADERS = $(XAV_DIR)/xav_decoder.h $(XAV_DIR)/decoder.h $(XAV_DIR)/converter.h $(XAV_DIR)/utils.h $(XAV_DIR)/channel_layout.h
-DECODER_SOURCES = $(XAV_DIR)/xav_decoder.c $(XAV_DIR)/decoder.c $(XAV_DIR)/converter.c $(XAV_DIR)/utils.c
+DECODER_HEADERS = $(XAV_DIR)/xav_decoder.h $(XAV_DIR)/decoder.h $(XAV_DIR)/video_converter.h $(XAV_DIR)/audio_converter.h $(XAV_DIR)/utils.h $(XAV_DIR)/channel_layout.h
+DECODER_SOURCES = $(XAV_DIR)/xav_decoder.c $(XAV_DIR)/decoder.c $(XAV_DIR)/video_converter.c $(XAV_DIR)/audio_converter.c $(XAV_DIR)/utils.c
 
-READER_HEADERS = $(XAV_DIR)/xav_reader.h $(XAV_DIR)/reader.h $(XAV_DIR)/converter.h $(XAV_DIR)/utils.h $(XAV_DIR)/channel_layout.h
-READER_SOURCES = $(XAV_DIR)/xav_reader.c $(XAV_DIR)/reader.c $(XAV_DIR)/converter.c $(XAV_DIR)/utils.c
+READER_HEADERS = $(XAV_DIR)/xav_reader.h $(XAV_DIR)/reader.h $(XAV_DIR)/video_converter.h $(XAV_DIR)/audio_converter.h $(XAV_DIR)/utils.h $(XAV_DIR)/channel_layout.h
+READER_SOURCES = $(XAV_DIR)/xav_reader.c $(XAV_DIR)/reader.c $(XAV_DIR)/video_converter.c $(XAV_DIR)/audio_converter.c $(XAV_DIR)/utils.c
 
 CFLAGS = $(XAV_DEBUG_LOGS) -fPIC -shared
 IFLAGS = -I$(ERTS_INCLUDE_DIR) -I$(XAV_DIR)

c_src/xav/converter.c → c_src/xav/audio_converter.c

@@ -4,8 +4,8 @@
 #include <libswresample/swresample.h>
 #include <stdint.h>
 
+#include "audio_converter.h"
 #include "channel_layout.h"
-#include "converter.h"
 #include "utils.h"
 
 struct Converter *converter_alloc() {
@@ -65,20 +65,21 @@ int converter_convert(struct Converter *c, AVFrame *src_frame, uint8_t ***out_da
     return ret;
   }
 
-  *out_data = out_data_tmp;
-
   *out_samples = swr_convert(c->swr_ctx, out_data_tmp, max_out_nb_samples,
                              (const uint8_t **)src_frame->data, src_frame->nb_samples);
 
   if (*out_samples < 0) {
     XAV_LOG_DEBUG("Couldn't convert samples: %d", *out_samples);
+    av_freep(&out_data_tmp[0]);
     return -1;
   }
 
   XAV_LOG_DEBUG("Converted %d samples per channel", *out_samples);
 
   *out_size = *out_samples * out_bytes_per_sample * out_nb_channels;
 
+  *out_data = out_data_tmp;
+
   return 0;
 }
 

c_src/xav/decoder.c

@@ -1,5 +1,6 @@
 #include "decoder.h"
 #include "utils.h"
+#include "video_converter.h"
 
 static int init_converter(struct Decoder *decoder);
 
@@ -8,16 +9,6 @@ struct Decoder *decoder_alloc() {
 
   decoder->codec = NULL;
   decoder->c = NULL;
-  decoder->out_format_name = NULL;
-
-  for (int i = 0; i < 4; i++) {
-    decoder->rgb_dst_data[i] = NULL;
-  }
-
-  decoder->frame_data = NULL;
-  decoder->frame_linesize = NULL;
-  decoder->converter = NULL;
-  decoder->out_data = NULL;
 
   return decoder;
 }
@@ -26,12 +17,9 @@ int decoder_init(struct Decoder *decoder, const char *codec) {
   if (strcmp(codec, "opus") == 0) {
     decoder->media_type = AVMEDIA_TYPE_AUDIO;
     decoder->codec = avcodec_find_decoder(AV_CODEC_ID_OPUS);
-    // we will initialize out_format_name with the first frame
-    decoder->out_format_name = NULL;
   } else if (strcmp(codec, "vp8") == 0) {
     decoder->media_type = AVMEDIA_TYPE_VIDEO;
     decoder->codec = avcodec_find_decoder(AV_CODEC_ID_VP8);
-    decoder->out_format_name = "rgb";
   } else {
     return -1;
   }
@@ -69,57 +57,13 @@ int decoder_decode(struct Decoder *decoder, AVPacket *pkt, AVFrame *frame) {
     return -2;
   }
 
-  ret = avcodec_receive_frame(decoder->c, frame);
-  if (ret != 0) {
-    return -1;
-  }
-
-  if (decoder->media_type == AVMEDIA_TYPE_AUDIO && decoder->out_format_name == NULL) {
-    enum AVSampleFormat out_sample_fmt = av_get_alt_sample_fmt(frame->format, 0);
-    decoder->out_format_name = av_get_sample_fmt_name(out_sample_fmt);
-  }
-
-  if (decoder->media_type == AVMEDIA_TYPE_VIDEO) {
-    if (frame->format != AV_PIX_FMT_RGB24) {
-      convert_to_rgb(frame, decoder->rgb_dst_data, decoder->rgb_dst_linesize);
-      decoder->frame_data = decoder->rgb_dst_data;
-      decoder->frame_linesize = decoder->rgb_dst_linesize;
-    } else {
-      decoder->frame_data = frame->data;
-      decoder->frame_linesize = frame->linesize;
-    }
-  } else if (decoder->media_type == AVMEDIA_TYPE_AUDIO) {
-
-    if (decoder->converter == NULL) {
-      ret = init_converter(decoder);
-      if (ret < 0) {
-        return ret;
-      }
-    }
-
-    return converter_convert(decoder->converter, frame, &decoder->out_data, &decoder->out_samples,
-                             &decoder->out_size);
-  }
-
-  return 0;
+  return avcodec_receive_frame(decoder->c, frame);
 }
 
 void decoder_free_frame(struct Decoder *decoder) {
   // TODO revisit this
   av_frame_unref(decoder->frame);
   av_packet_unref(decoder->pkt);
-
-  if (decoder->media_type == AVMEDIA_TYPE_AUDIO && decoder->frame_data == decoder->rgb_dst_data) {
-    av_freep(&decoder->frame_data[0]);
-  } else if (decoder->media_type == AVMEDIA_TYPE_VIDEO &&
-             decoder->frame_data == decoder->rgb_dst_data) {
-    av_freep(&decoder->frame_data[0]);
-  }
-
-  if (decoder->out_data != NULL) {
-    // av_freep sets pointer to NULL
-    av_freep(&decoder->out_data);
-  }
 }
 
 void decoder_free(struct Decoder **decoder) {
@@ -143,29 +87,3 @@ void decoder_free(struct Decoder **decoder) {
     *decoder = NULL;
   }
 }
-
-static int init_converter(struct Decoder *decoder) {
-  decoder->converter = converter_alloc();
-
-  if (decoder->converter == NULL) {
-    XAV_LOG_DEBUG("Couldn't allocate converter");
-    return -1;
-  }
-
-  int out_sample_rate = decoder->c->sample_rate;
-  enum AVSampleFormat out_sample_fmt = AV_SAMPLE_FMT_FLT;
-
-  struct ChannelLayout in_chlayout, out_chlayout;
-#if LIBAVUTIL_VERSION_MAJOR >= 58
-  in_chlayout.layout = decoder->c->ch_layout;
-  out_chlayout.layout = decoder->c->ch_layout;
-#else
-  in_chlayout.layout = decoder->c->channel_layout;
-  out_chlayout.layout = decoder->c->channel_layout;
-  XAV_LOG_DEBUG("in_chlayout %ld", in_chlayout.layout);
-  XAV_LOG_DEBUG("in nb_channels %d", av_get_channel_layout_nb_channels(in_chlayout.layout));
-#endif
-
-  return converter_init(decoder->converter, in_chlayout, decoder->c->sample_rate,
-                        decoder->c->sample_fmt, out_chlayout, out_sample_rate, out_sample_fmt);
-}

c_src/xav/decoder.h

@@ -1,7 +1,7 @@
 #include <libavcodec/avcodec.h>
 #include <libswresample/swresample.h>
 
-#include "converter.h"
+#include "audio_converter.h"
 #include "utils.h"
 
 struct Decoder {
@@ -10,24 +10,6 @@ struct Decoder {
   AVPacket *pkt;
   const AVCodec *codec;
   AVCodecContext *c;
-
-  const char *out_format_name;
-
-  uint8_t *rgb_dst_data[4];
-  int rgb_dst_linesize[4];
-
-  uint8_t **frame_data;
-  int *frame_linesize;
-
-  struct Converter *converter;
-  // Buffer where audio samples are written after conversion.
-  // We always convet to packed format, so only out_data[0] is set.
-  uint8_t **out_data;
-  // Number of samples in out_data buffer
-  int out_samples;
-  // Size of out_data buffer.
-  // This is the same as out_samples * bytes_per_sample(out_format) * out_channels.
-  int out_size;
 };
 
 struct Decoder *decoder_alloc();

c_src/xav/reader.c

@@ -21,13 +21,6 @@ struct Reader *reader_alloc() {
   reader->options = NULL;
   reader->in_format_name = NULL;
   reader->out_format_name = NULL;
-  reader->frame_data = NULL;
-  reader->frame_linesize = NULL;
-  for (int i = 0; i < 4; i++) {
-    reader->rgb_dst_data[i] = NULL;
-  }
-  reader->converter = NULL;
-  reader->out_data = NULL;
 
   return reader;
 }
@@ -105,7 +98,7 @@ int reader_next_frame(struct Reader *reader) {
 
   if (ret == 0) {
     XAV_LOG_DEBUG("Received frame");
-    goto fin;
+    return 0;
   } else if (ret == AVERROR_EOF) {
     XAV_LOG_DEBUG("EOF");
     return ret;
@@ -188,47 +181,10 @@ int reader_next_frame(struct Reader *reader) {
     }
   }
 
-fin:
-  if (reader->media_type == AVMEDIA_TYPE_VIDEO && reader->frame->format != AV_PIX_FMT_RGB24) {
-    XAV_LOG_DEBUG("Converting video to RGB");
-    convert_to_rgb(reader->frame, reader->rgb_dst_data, reader->rgb_dst_linesize);
-    reader->frame_data = reader->rgb_dst_data;
-    reader->frame_linesize = reader->rgb_dst_linesize;
-  } else if (reader->media_type == AVMEDIA_TYPE_VIDEO) {
-    reader->frame_data = reader->frame->data;
-    reader->frame_linesize = reader->frame->linesize;
-  } else if (reader->media_type == AVMEDIA_TYPE_AUDIO) {
-    XAV_LOG_DEBUG("Converting audio to desired out format");
-
-    if (reader->converter == NULL) {
-      XAV_LOG_DEBUG("Converter not initialized. Initializing.");
-      ret = init_converter(reader);
-      if (ret < 0) {
-        return ret;
-      }
-    }
-
-    return converter_convert(reader->converter, reader->frame, &reader->out_data,
-                             &reader->out_samples, &reader->out_size);
-  }
-
   return 0;
 }
 
-void reader_free_frame(struct Reader *reader) {
-  av_frame_unref(reader->frame);
-
-  if (reader->media_type == AVMEDIA_TYPE_AUDIO && reader->frame_data == reader->rgb_dst_data) {
-    av_freep(&reader->frame_data[0]);
-  } else if (reader->media_type == AVMEDIA_TYPE_VIDEO &&
-             reader->frame_data == reader->rgb_dst_data) {
-    av_freep(&reader->frame_data[0]);
-  }
-
-  if (reader->out_data != NULL) {
-    av_freep(&reader->out_data);
-  }
-}
+void reader_free_frame(struct Reader *reader) { av_frame_unref(reader->frame); }
 
 void reader_free(struct Reader **reader) {
   XAV_LOG_DEBUG("Freeing Reader object");
@@ -259,42 +215,3 @@ void reader_free(struct Reader **reader) {
     *reader = NULL;
   }
 }
-
-static int init_converter(struct Reader *reader) {
-  reader->converter = converter_alloc();
-
-  if (reader->converter == NULL) {
-    XAV_LOG_DEBUG("Couldn't allocate converter");
-    return -1;
-  }
-
-  int out_sample_rate = 16000;
-  enum AVSampleFormat out_sample_fmt = AV_SAMPLE_FMT_FLT;
-
-  struct ChannelLayout in_chlayout, out_chlayout;
-#if LIBAVUTIL_VERSION_MAJOR >= 58
-  in_chlayout.layout = reader->c->ch_layout;
-  av_channel_layout_from_mask(&out_chlayout.layout, AV_CH_LAYOUT_MONO);
-#else
-  in_chlayout.layout = reader->frame->channel_layout;
-  out_chlayout.layout = AV_CH_LAYOUT_MONO;
-
-  if (reader->frame->channel_layout == 0 && reader->frame->channels > 0) {
-    // In newer FFmpeg versions, 0 means that the order of channels is
-    // unspecified but there still might be information about channels number.
-    // Let's check againts it and take default channel order for the given channels number.
-    // This is also what newer FFmpeg versions do under the hood when passing
-    // unspecified channel order.
-    XAV_LOG_DEBUG("Channel layout unset. Setting to default for channels number: %d",
-                  reader->frame->channels);
-    in_chlayout.layout = av_get_default_channel_layout(reader->frame->channels);
-  } else if (reader->frame->channel_layout == 0) {
-    XAV_LOG_DEBUG("Both channel layout and channels are unset. Cannot init converter.");
-    return -1;
-  }
-
-#endif
-
-  return converter_init(reader->converter, in_chlayout, reader->c->sample_rate,
-                        reader->c->sample_fmt, out_chlayout, out_sample_rate, out_sample_fmt);
-}

c_src/xav/reader.h

@@ -9,7 +9,7 @@
 #include <stdlib.h>
 #include <string.h>
 
-#include "converter.h"
+#include "audio_converter.h"
 #include "utils.h"
 
 struct Reader {
@@ -26,28 +26,6 @@ struct Reader {
 
   const char *in_format_name;
   const char *out_format_name;
-
-  // used for converting decoded frame
-  // to rgb pixel format
-  uint8_t *rgb_dst_data[4];
-  int rgb_dst_linesize[4];
-
-  // points either to frame->data
-  // frame->linesize or rgb_dst_data
-  // rgb_dst_linesize depending on
-  // whether convertion to rgb was needed
-  uint8_t **frame_data;
-  int *frame_linesize;
-
-  struct Converter *converter;
-  // Buffer where audio samples are written after conversion.
-  // We always convet to packed format, so only out_data[0] is set.
-  uint8_t **out_data;
-  // Number of samples in out_data buffer
-  int out_samples;
-  // Size of out_data buffer.
-  // This is the same as out_samples * bytes_per_sample(out_format) * out_channels.
-  int out_size;
 };
 
 struct Reader *reader_alloc();

c_src/xav/utils.c

@@ -55,8 +55,8 @@ ERL_NIF_TERM xav_nif_audio_frame_to_term(ErlNifEnv *env, uint8_t **out_data, int
   return enif_make_tuple(env, 4, data_term, format_term, samples_term, pts_term);
 }
 
-ERL_NIF_TERM xav_nif_video_frame_to_term(ErlNifEnv *env, AVFrame *frame, unsigned char *data[],
-                                         int *linesize, const char *format_name) {
+ERL_NIF_TERM xav_nif_video_frame_to_term(ErlNifEnv *env, AVFrame *frame, uint8_t *data[4],
+                                         int linesize[4], const char *format_name) {
   ERL_NIF_TERM data_term;
   unsigned char *ptr = enif_make_new_binary(env, linesize[0] * frame->height, &data_term);
   memcpy(ptr, data[0], linesize[0] * frame->height);

c_src/xav/utils.h

@@ -23,7 +23,7 @@ void convert_to_rgb(AVFrame *src_frame, uint8_t *dst_data[], int dst_linesize[])
 ERL_NIF_TERM xav_nif_ok(ErlNifEnv *env, ERL_NIF_TERM data_term);
 ERL_NIF_TERM xav_nif_error(ErlNifEnv *env, char *reason);
 ERL_NIF_TERM xav_nif_raise(ErlNifEnv *env, char *msg);
-ERL_NIF_TERM xav_nif_video_frame_to_term(ErlNifEnv *env, AVFrame *frame, unsigned char *data[],
-                                         int *linesize, const char *out_format_name);
+ERL_NIF_TERM xav_nif_video_frame_to_term(ErlNifEnv *env, AVFrame *frame, uint8_t *out_data[4],
+                                         int out_linesize[4], const char *out_format);
 ERL_NIF_TERM xav_nif_audio_frame_to_term(ErlNifEnv *env, uint8_t **out_data, int out_samples,
                                          int out_size, const char *out_format, int pts);