audio: make de-planarization faster

Uses the same trick as the planarization code to turn per-sample memcpy
calls into mov instructions. Makes decoding a ~25min 48000Hz 2ch floatle
audio file faster from 3.8s to 2.7s.

audio/decode/ad_lavc.c

@@ -52,8 +52,8 @@ LIBAD_EXTERN(ffmpeg)
 struct priv {
     AVCodecContext *avctx;
     AVFrame *avframe;
-    char *output;
-    char *output_packed; // used by deplanarize to store packed audio samples
+    uint8_t *output;
+    uint8_t *output_packed; // used by deplanarize to store packed audio samples
     int output_left;
     int unitsize;
     int previous_data_left;  // input demuxer packet data
@@ -363,6 +363,7 @@ static av_always_inline void deplanarize(struct sh_audio *sh)
 {
     struct priv *priv = sh->context;
 
+    uint8_t **planes  = priv->avframe->extended_data;
     size_t bps        = av_get_bytes_per_sample(priv->avctx->sample_fmt);
     size_t nb_samples = priv->avframe->nb_samples;
     size_t channels   = priv->avctx->channels;
@@ -372,17 +373,7 @@ static av_always_inline void deplanarize(struct sh_audio *sh)
         priv->output_packed =
             talloc_realloc_size(priv, priv->output_packed, size);
 
-    size_t offset = 0;
-    unsigned char *output_ptr = priv->output_packed;
-    unsigned char **src = priv->avframe->data;
-
-    for (size_t s = 0; s < nb_samples; s++) {
-        for (size_t c = 0; c < channels; c++) {
-            memcpy(output_ptr, src[c] + offset, bps);
-            output_ptr += bps;
-        }
-        offset += bps;
-    }
+    reorder_to_packed(priv->output_packed, planes, bps, channels, nb_samples);
 
     priv->output = priv->output_packed;
 }

audio/reorder_ch.c

@@ -1399,7 +1399,8 @@ int main(int argc, char *argv[]) {
 #endif
 
 static inline void reorder_to_planar_(void *restrict out, const void *restrict in,
-        size_t size, size_t nchan, size_t nmemb) {
+                                      size_t size, size_t nchan, size_t nmemb)
+{
     size_t i, c;
     char *outptr = (char *) out;
     size_t instep = nchan * size;
@@ -1413,7 +1414,7 @@ static inline void reorder_to_planar_(void *restrict out, const void *restrict i
 }
 
 void reorder_to_planar(void *restrict out, const void *restrict in,
-        size_t size, size_t nchan, size_t nmemb)
+                       size_t size, size_t nchan, size_t nmemb)
 {
     // special case for mono (nothing to do...)
     if (nchan == 1)
@@ -1432,3 +1433,35 @@ void reorder_to_planar(void *restrict out, const void *restrict in,
     else
         reorder_to_planar_(out, in, size, nchan, nmemb);
 }
+
+static inline void reorder_to_packed_(uint8_t *out, uint8_t **in,
+                                      size_t size, size_t nchan, size_t nmemb)
+{
+    size_t outstep = nchan * size;
+
+    for (size_t c = 0; c < nchan; ++c) {
+        char *outptr = out + c * size;
+        char *inptr = in[c];
+        for (size_t i = 0; i < nmemb; ++i, outptr += outstep, inptr += size) {
+            memcpy(outptr, inptr, size);
+        }
+    }
+}
+
+// out = destination array of packed samples of given size, nmemb frames
+// in[channel] = source array of samples for the given channel
+void reorder_to_packed(uint8_t *out, uint8_t **in,
+                       size_t size, size_t nchan, size_t nmemb)
+{
+    if (nchan == 1)
+        memcpy(out, in, size * nchan * nmemb);
+    // See reorder_to_planar() why this is done this way
+    else if (size == 1)
+        reorder_to_packed_(out, in, 1, nchan, nmemb);
+    else if (size == 2)
+        reorder_to_packed_(out, in, 2, nchan, nmemb);
+    else if (size == 4)
+        reorder_to_packed_(out, in, 4, nchan, nmemb);
+    else
+        reorder_to_packed_(out, in, size, nchan, nmemb);
+}

audio/reorder_ch.h

@@ -23,6 +23,8 @@
 #ifndef MPLAYER_REORDER_CH_H
 #define MPLAYER_REORDER_CH_H
 
+#include <inttypes.h>
+
 // L   - Left
 // R   - Right
 // C   - Center
@@ -132,6 +134,8 @@ void reorder_channel_nch(void *buf,
 
 /// Utility function for planar audio conversions
 void reorder_to_planar(void *restrict out, const void *restrict in,
-        size_t size, size_t nchan, size_t nmemb);
+                       size_t size, size_t nchan, size_t nmemb);
+void reorder_to_packed(uint8_t *out, uint8_t **in,
+                       size_t size, size_t nchan, size_t nmemb);
 
 #endif /* MPLAYER_REORDER_CH_H */