mdct wrapper function to match fft
[ffmpeg.git] / libavcodec / vorbis_dec.c
index ed13229..a178360 100644 (file)
@@ -153,9 +153,6 @@ typedef struct vorbis_context_s {
     float *channel_residues;
     float *channel_floors;
     float *saved;
-    float *ret;
-    float *buf;
-    float *buf_tmp;
     uint_fast32_t add_bias; // for float->int conversion
     uint_fast32_t exp_bias;
 } vorbis_context;
@@ -181,9 +178,6 @@ static void vorbis_free(vorbis_context *vc) {
     av_freep(&vc->channel_residues);
     av_freep(&vc->channel_floors);
     av_freep(&vc->saved);
-    av_freep(&vc->ret);
-    av_freep(&vc->buf);
-    av_freep(&vc->buf_tmp);
 
     av_freep(&vc->residues);
     av_freep(&vc->modes);
@@ -899,10 +893,7 @@ static int vorbis_parse_id_hdr(vorbis_context *vc){
 
     vc->channel_residues= av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
     vc->channel_floors  = av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
-    vc->saved           = av_mallocz((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
-    vc->ret             = av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
-    vc->buf             = av_malloc( vc->blocksize[1]                       * sizeof(float));
-    vc->buf_tmp         = av_malloc( vc->blocksize[1]                       * sizeof(float));
+    vc->saved           = av_mallocz((vc->blocksize[1]/4)*vc->audio_channels * sizeof(float));
     vc->previous_window=0;
 
     ff_mdct_init(&vc->mdct[0], bl0, 1);
@@ -935,7 +926,7 @@ static av_cold int vorbis_decode_init(AVCodecContext *avccontext) {
     vc->avccontext = avccontext;
     dsputil_init(&vc->dsp, avccontext);
 
-    if(vc->dsp.float_to_int16 == ff_float_to_int16_c) {
+    if(vc->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
         vc->add_bias = 385;
         vc->exp_bias = 0;
     } else {
@@ -980,6 +971,7 @@ static av_cold int vorbis_decode_init(AVCodecContext *avccontext) {
     avccontext->channels = vc->audio_channels;
     avccontext->sample_rate = vc->audio_samplerate;
     avccontext->frame_size  = FFMIN(vc->blocksize[0], vc->blocksize[1])>>2;
+    avccontext->sample_fmt = SAMPLE_FMT_S16;
 
     return 0 ;
 }
@@ -1330,6 +1322,14 @@ static int vorbis_residue_decode(vorbis_context *vc, vorbis_residue *vr, uint_fa
                                         vec[voffs+k     ]+=codebook.codevectors[coffs  ];  // FPMATH
                                         vec[voffs+k+vlen]+=codebook.codevectors[coffs+1];  // FPMATH
                                     }
+                                } else if(dim==4) {
+                                    for(k=0;k<step;++k, voffs+=2) {
+                                        coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * 4;
+                                        vec[voffs       ]+=codebook.codevectors[coffs  ];  // FPMATH
+                                        vec[voffs+1     ]+=codebook.codevectors[coffs+2];  // FPMATH
+                                        vec[voffs+vlen  ]+=codebook.codevectors[coffs+1];  // FPMATH
+                                        vec[voffs+vlen+1]+=codebook.codevectors[coffs+3];  // FPMATH
+                                    }
                                 } else
                                 for(k=0;k<step;++k) {
                                     coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
@@ -1415,7 +1415,7 @@ static int vorbis_parse_audio_packet(vorbis_context *vc) {
     uint_fast8_t mode_number;
     uint_fast8_t blockflag;
     uint_fast16_t blocksize;
-    int_fast32_t i,j;
+    int_fast32_t i,j,dir;
     uint_fast8_t no_residue[vc->audio_channels];
     uint_fast8_t do_not_decode[vc->audio_channels];
     vorbis_mapping *mapping;
@@ -1517,26 +1517,29 @@ static int vorbis_parse_audio_packet(vorbis_context *vc) {
 // MDCT, overlap/add, save data for next overlapping  FPMATH
 
     retlen = (blocksize + vc->blocksize[previous_window])/4;
-    for(j=0;j<vc->audio_channels;++j) {
+    dir = retlen <= blocksize/2; // pick an order so that ret[] can reuse floors[] without stepping on any data we need
+    for(j=dir?0:vc->audio_channels-1; (unsigned)j<vc->audio_channels; j+=dir*2-1) {
         uint_fast16_t bs0=vc->blocksize[0];
         uint_fast16_t bs1=vc->blocksize[1];
-        float *saved=vc->saved+j*bs1/2;
-        float *ret=vc->ret+j*retlen;
-        float *buf=vc->buf;
+        float *residue=vc->channel_residues+res_chan[j]*blocksize/2;
+        float *floor=vc->channel_floors+j*blocksize/2;
+        float *saved=vc->saved+j*bs1/4;
+        float *ret=vc->channel_floors+j*retlen;
+        float *buf=residue;
         const float *win=vc->win[blockflag&previous_window];
 
-        vc->mdct[0].fft.imdct_calc(&vc->mdct[blockflag], buf, vc->channel_floors+j*blocksize/2, vc->buf_tmp);
+        ff_imdct_half(&vc->mdct[blockflag], buf, floor);
 
         if(blockflag == previous_window) {
-            vc->dsp.vector_fmul_window(ret, saved, buf, win, fadd_bias, blocksize/2);
+            vc->dsp.vector_fmul_window(ret, saved, buf, win, fadd_bias, blocksize/4);
         } else if(blockflag > previous_window) {
-            vc->dsp.vector_fmul_window(ret, saved, buf+(bs1-bs0)/4, win, fadd_bias, bs0/2);
-            copy_normalize(ret+bs0/2, buf+(bs1+bs0)/4, (bs1-bs0)/4, vc->exp_bias, fadd_bias);
+            vc->dsp.vector_fmul_window(ret, saved, buf, win, fadd_bias, bs0/4);
+            copy_normalize(ret+bs0/2, buf+bs0/4, (bs1-bs0)/4, vc->exp_bias, fadd_bias);
         } else {
             copy_normalize(ret, saved, (bs1-bs0)/4, vc->exp_bias, fadd_bias);
-            vc->dsp.vector_fmul_window(ret+(bs1-bs0)/4, saved+(bs1-bs0)/4, buf, win, fadd_bias, bs0/2);
+            vc->dsp.vector_fmul_window(ret+(bs1-bs0)/4, saved+(bs1-bs0)/4, buf, win, fadd_bias, bs0/4);
         }
-        memcpy(saved, buf+blocksize/2, blocksize/2*sizeof(float));
+        memcpy(saved, buf+blocksize/4, blocksize/4*sizeof(float));
     }
 
     vc->previous_window = blockflag;
@@ -1551,6 +1554,8 @@ static int vorbis_decode_frame(AVCodecContext *avccontext,
 {
     vorbis_context *vc = avccontext->priv_data ;
     GetBitContext *gb = &(vc->gb);
+    const float *channel_ptrs[vc->audio_channels];
+    int i;
 
     int_fast16_t len;
 
@@ -1577,7 +1582,9 @@ static int vorbis_decode_frame(AVCodecContext *avccontext,
 
     AV_DEBUG("parsed %d bytes %d bits, returned %d samples (*ch*bits) \n", get_bits_count(gb)/8, get_bits_count(gb)%8, len);
 
-    vc->dsp.float_to_int16_interleave(data, vc->ret, len, vc->audio_channels);
+    for(i=0; i<vc->audio_channels; i++)
+        channel_ptrs[i] = vc->channel_floors+i*len;
+    vc->dsp.float_to_int16_interleave(data, channel_ptrs, len, vc->audio_channels);
     *data_size=len*2*vc->audio_channels;
 
     return buf_size ;