Reorganise the wma_decode_superframe() function - split into a separate init and decode functions. Each call to the decode function now decodes a single frame (2048 samples) instead of an entire superframe (which typically contained about 7 or 8 frames and can in theory contain up to 16 frames). This allows us to replace the 256KB output buffer with a 8KB buffer, and also perform more yields in the main decoding loop.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@13833 a1c6a512-1295-4272-9138-f99709370657

3 files changed, 104 insertions, 65 deletions

diff --git a/apps/codecs/libwma/wmadec.h b/apps/codecs/libwma/wmadec.h
index e23d7c2..fd2d63c 100644
--- a/apps/codecs/libwma/wmadec.h
+++ b/apps/codecs/libwma/wmadec.h
@@ -137,6 +137,11 @@ typedef struct WMADecodeContext
     fixed64 lsp_pow_m_table1[(1 << LSP_POW_BITS)];
     fixed64 lsp_pow_m_table2[(1 << LSP_POW_BITS)];
 
+    /* State of current superframe decoding */
+    int bit_offset;
+    int nb_frames;
+    int current_frame;
+
 #ifdef TRACE
 
     int frame_count;
@@ -145,7 +150,9 @@ typedef struct WMADecodeContext
 WMADecodeContext;
 
 int wma_decode_init(WMADecodeContext* s, asf_waveformatex_t *wfx);
-int wma_decode_superframe(WMADecodeContext* s,
-                          void *data, int *data_size,
-                          uint8_t *buf, int buf_size);
+int wma_decode_superframe_init(WMADecodeContext* s,
+                               uint8_t *buf, int buf_size);
+int wma_decode_superframe_frame(WMADecodeContext* s,
+                                int16_t *samples,
+                                uint8_t *buf, int buf_size);
 #endif
diff --git a/apps/codecs/libwma/wmadeci.c b/apps/codecs/libwma/wmadeci.c
index c266316..1a7081f 100644
--- a/apps/codecs/libwma/wmadeci.c
+++ b/apps/codecs/libwma/wmadeci.c
@@ -935,6 +935,9 @@ int wma_decode_init(WMADecodeContext* s, asf_waveformatex_t *wfx)
     init_coef_vlc(&s->coef_vlc[1], &s->run_table[1], &s->level_table[1],
                   &coef_vlcs[coef_vlc_table * 2 + 1], 1);
 
+    s->last_superframe_len = 0;
+    s->last_bitoffset = 0;
+
     return 0;
 }
 
@@ -1666,42 +1669,67 @@ static int wma_decode_frame(WMADecodeContext *s, int16_t *samples)
     return 0;
 }
 
-int wma_decode_superframe(WMADecodeContext* s,
-                                 void *data,    /*output*/
-                                 int *data_size,
-                                 uint8_t *buf,  /*input*/
-                                 int buf_size)
-{
-    //WMADecodeContext *s = avctx->priv_data;
-    int nb_frames, bit_offset, i, pos, len;
-    uint8_t *q;
-    int16_t *samples;
+/* Initialise the superframe decoding */
 
+int wma_decode_superframe_init(WMADecodeContext* s,
+                               uint8_t *buf,  /*input*/
+                               int buf_size)
+{
     if (buf_size==0)
     {
         s->last_superframe_len = 0;
         return 0;
     }
 
-    samples = data;
+    s->current_frame = 0;
+
     init_get_bits(&s->gb, buf, buf_size*8);
+
     if (s->use_bit_reservoir)
     {
         /* read super frame header */
         get_bits(&s->gb, 4); /* super frame index */
-        nb_frames = get_bits(&s->gb, 4) - 1;
+        s->nb_frames = get_bits(&s->gb, 4);
+
+        if (s->last_superframe_len == 0)
+            s->nb_frames --;
+        else if (s->nb_frames == 0)
+            s->nb_frames++;
+
+        s->bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3);
+    } else {
+        s->nb_frames = 1;
+    }
 
-        bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3);
+    return 1;
+}
+
+
+/* Decode a single frame in the current superframe - return -1 if
+   there was a decoding error, or the number of samples decoded. 
+*/
+
+int wma_decode_superframe_frame(WMADecodeContext* s,
+                                int16_t* samples, /*output*/
+                                uint8_t *buf,  /*input*/
+                                int buf_size)
+{
+    int pos, len;
+    uint8_t *q;
+    int done = 0;
+
+    if ((s->use_bit_reservoir) && (s->current_frame == 0))
+    {
         if (s->last_superframe_len > 0)
         {
-            /* add bit_offset bits to last frame */
-            if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) >
+            /* add s->bit_offset bits to last frame */
+            if ((s->last_superframe_len + ((s->bit_offset + 7) >> 3)) >
                     MAX_CODED_SUPERFRAME_SIZE)
             {
                 goto fail;
             }
             q = s->last_superframe + s->last_superframe_len;
-            len = bit_offset;
+            len = s->bit_offset;
             while (len > 0)
             {
                 *q++ = (get_bits)(&s->gb, 8);
@@ -1712,39 +1740,46 @@ int wma_decode_superframe(WMADecodeContext* s,
                 *q++ = (get_bits)(&s->gb, len) << (8 - len);
             }
 
-            /* XXX: bit_offset bits into last frame */
+            /* XXX: s->bit_offset bits into last frame */
             init_get_bits(&s->gb, s->last_superframe, MAX_CODED_SUPERFRAME_SIZE*8);
             /* skip unused bits */
             if (s->last_bitoffset > 0)
                 skip_bits(&s->gb, s->last_bitoffset);
+
             /* this frame is stored in the last superframe and in the
                current one */
             if (wma_decode_frame(s, samples) < 0)
             {
                 goto fail;
             }
-            samples += s->nb_channels * s->frame_len;
+            done = 1;
         }
 
-        /* read each frame starting from bit_offset */
-        pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3;
+        /* read each frame starting from s->bit_offset */
+        pos = s->bit_offset + 4 + 4 + s->byte_offset_bits + 3;
         init_get_bits(&s->gb, buf + (pos >> 3), (MAX_CODED_SUPERFRAME_SIZE - (pos >> 3))*8);
         len = pos & 7;
         if (len > 0)
             skip_bits(&s->gb, len);
 
         s->reset_block_lengths = 1;
-        for(i=0;i<nb_frames;++i)
+    }
+
+    /* If we haven't decoded a frame yet, do it now */
+    if (!done)
+    {
+        if (wma_decode_frame(s, samples) < 0)
         {
-            if (wma_decode_frame(s, samples) < 0)
-            {
-                goto fail;
-            }
-            samples += s->nb_channels * s->frame_len;
+            goto fail;
         }
+    }
+
+    s->current_frame++;
 
+    if ((s->use_bit_reservoir) && (s->current_frame == s->nb_frames))
+    {
         /* we copy the end of the frame in the last frame buffer */
-        pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7);
+        pos = get_bits_count(&s->gb) + ((s->bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7);
         s->last_bitoffset = pos & 7;
         pos >>= 3;
         len = buf_size - pos;
@@ -1755,21 +1790,11 @@ int wma_decode_superframe(WMADecodeContext* s,
         s->last_superframe_len = len;
         memcpy(s->last_superframe, buf + pos, len);
     }
-    else
-    {
-        /* single frame decode */
-        if (wma_decode_frame(s, samples) < 0)
-        {
-            goto fail;
-        }
-        samples += s->nb_channels * s->frame_len;
-    }
-    *data_size = (int8_t *)samples - (int8_t *)data;
-    return s->block_align;
+
+    return s->frame_len;
+
 fail:
     /* when error, we reset the bit reservoir */
     s->last_superframe_len = 0;
     return -1;
 }
-
-
diff --git a/apps/codecs/wma.c b/apps/codecs/wma.c
index a8e386a..ab4a808 100644
--- a/apps/codecs/wma.c
+++ b/apps/codecs/wma.c
@@ -23,16 +23,15 @@
 
 CODEC_HEADER
 
-#define MAX_BLOCKSIZE 2048
+/* The output buffer containing the decoded samples (channels 0 and 1)
+   BLOCK_MAX_SIZE is 2048 (samples) and MAX_CHANNELS is 2.
+ */
 
-/* The output buffers containing the decoded samples (channels 0 and 1) */
+static uint16_t decoded[BLOCK_MAX_SIZE * MAX_CHANNELS];
 
 /* NOTE: WMADecodeContext is 142688 bytes (on x86) */
 static WMADecodeContext wmadec;
 
-/* TODO: Check the size of this */
-#define OUTBUF_SIZE 256*1024
-
 enum asf_error_e {
     ASF_ERROR_INTERNAL       = -1,  /* incorrect input to API calls */
     ASF_ERROR_OUTOFMEM       = -2,  /* some malloc inside program failed */
@@ -286,8 +285,8 @@ enum codec_status codec_main(void)
     unsigned char* inbuffer;
     size_t resume_offset;
     size_t n;
-    int wmares, res, padding, outbufsize;
-    uint8_t* outbuf;
+    int i;
+    int wmares, res, padding;
 
     /* Generic codec initialisation */
     ci->configure(CODEC_SET_FILEBUF_WATERMARK, 1024*512);
@@ -313,8 +312,6 @@ enum codec_status codec_main(void)
         goto exit;
     }
 
-    outbuf = codec_malloc(OUTBUF_SIZE);
-
     /* Copy the format metadata we've stored in the id3 TOC field.  This 
        saves us from parsing it again here. */
     memcpy(&wfx, ci->id3->toc, sizeof(wfx));
@@ -355,21 +352,31 @@ enum codec_status codec_main(void)
         if (res > 0) {
             inbuffer = ci->request_buffer(&n, res - padding);
 
-            wmares = wma_decode_superframe(&wmadec,
-                                        outbuf,&outbufsize,
-                                        inbuffer,res - padding);
-
-            ci->advance_buffer(res);
-
-            if (wmares > 0) {
-                ci->pcmbuf_insert(outbuf, NULL, outbufsize / (wfx.channels * 2));
-                samplesdone += (outbufsize / (wfx.channels * 2));
-                DEBUGF("Decoded %d samples\n",(outbufsize / (wfx.channels * 2)));
-                elapsedtime = (samplesdone*10)/(wfx.rate/100);
-                ci->set_elapsed(elapsedtime);
+            wma_decode_superframe_init(&wmadec,
+                                       inbuffer,res - padding);
+
+            for (i=0; i < wmadec.nb_frames; i++)
+            {
+                wmares = wma_decode_superframe_frame(&wmadec,
+                                                     decoded,
+                                                     inbuffer,res - padding);
+
+                ci->yield ();
+
+                if (wmares < 0)
+                {
+                    LOGF("WMA decode error %d\n",wmares);
+                    goto done;
+                } else if (wmares > 0) {
+                    ci->pcmbuf_insert(decoded, NULL, wmares);
+                    samplesdone += wmares;
+                    elapsedtime = (samplesdone*10)/(wfx.rate/100);
+                    ci->set_elapsed(elapsedtime);
+                }
+                ci->yield ();
             }
 
-            ci->yield ();
+            ci->advance_buffer(res);
         }
     }
     retval = CODEC_OK;