X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fadpcm.c;h=ec3fe6f6ecf28bb328775b87a438d9582ad7ba62;hb=ccd3228e54cc01baaf019a4b3842e378340cbdc9;hp=de66daf7dbe801c410fa79cb084e4f381fef16c2;hpb=2433f24f60bc4087907fee57b3b1a8f04b2b37b2;p=ffmpeg

diff --git a/libavcodec/adpcm.c b/libavcodec/adpcm.c
index de66daf7dbe..ec3fe6f6ecf 100644
--- a/libavcodec/adpcm.c
+++ b/libavcodec/adpcm.c
@@ -2,18 +2,20 @@
  * ADPCM codecs
  * Copyright (c) 2001-2003 The ffmpeg Project
  *
- * This library is free software; you can redistribute it and/or
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
+ * version 2.1 of the License, or (at your option) any later version.
  *
- * This library is distributed in the hope that it will be useful,
+ * FFmpeg is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
+ * License along with FFmpeg; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 #include "avcodec.h"
@@ -203,49 +205,11 @@ static int adpcm_encode_close(AVCodecContext *avctx)
 
 static inline unsigned char adpcm_ima_compress_sample(ADPCMChannelStatus *c, short sample)
 {
-    int step_index;
-    unsigned char nibble;
-
-    int sign = 0; /* sign bit of the nibble (MSB) */
-    int delta, predicted_delta;
-
-    delta = sample - c->prev_sample;
-
-    if (delta < 0) {
-        sign = 1;
-        delta = -delta;
-    }
-
-    step_index = c->step_index;
-
-    /* nibble = 4 * delta / step_table[step_index]; */
-    nibble = (delta << 2) / step_table[step_index];
-
-    if (nibble > 7)
-        nibble = 7;
-
-    step_index += index_table[nibble];
-    if (step_index < 0)
-        step_index = 0;
-    if (step_index > 88)
-        step_index = 88;
-
-    /* what the decoder will find */
-    predicted_delta = ((step_table[step_index] * nibble) / 4) + (step_table[step_index] / 8);
-
-    if (sign)
-        c->prev_sample -= predicted_delta;
-    else
-        c->prev_sample += predicted_delta;
-
+    int delta = sample - c->prev_sample;
+    int nibble = FFMIN(7, abs(delta)*4/step_table[c->step_index]) + (delta<0)*8;
+    c->prev_sample = c->prev_sample + ((step_table[c->step_index] * yamaha_difflookup[nibble]) / 8);
     CLAMP_TO_SHORT(c->prev_sample);
-
-
-    nibble += sign << 3; /* sign * 8 */
-
-    /* save back */
-    c->step_index = step_index;
-
+    c->step_index = clip(c->step_index + index_table[nibble], 0, 88);
     return nibble;
 }
 
@@ -276,27 +240,194 @@ static inline unsigned char adpcm_ms_compress_sample(ADPCMChannelStatus *c, shor
 
 static inline unsigned char adpcm_yamaha_compress_sample(ADPCMChannelStatus *c, short sample)
 {
-    int i1 = 0, j1;
+    int nibble, delta;
 
     if(!c->step) {
         c->predictor = 0;
         c->step = 127;
     }
-    j1 = sample - c->predictor;
 
-    j1 = (j1 * 8) / c->step;
-    i1 = abs(j1) / 2;
-    if (i1 > 7)
-        i1 = 7;
-    if (j1 < 0)
-        i1 += 8;
+    delta = sample - c->predictor;
+
+    nibble = FFMIN(7, abs(delta)*4/c->step) + (delta<0)*8;
 
-    c->predictor = c->predictor + ((c->step * yamaha_difflookup[i1]) / 8);
+    c->predictor = c->predictor + ((c->step * yamaha_difflookup[nibble]) / 8);
     CLAMP_TO_SHORT(c->predictor);
-    c->step = (c->step * yamaha_indexscale[i1]) >> 8;
+    c->step = (c->step * yamaha_indexscale[nibble]) >> 8;
     c->step = clip(c->step, 127, 24567);
 
-    return i1;
+    return nibble;
+}
+
+typedef struct TrellisPath {
+    int nibble;
+    int prev;
+} TrellisPath;
+
+typedef struct TrellisNode {
+    uint32_t ssd;
+    int path;
+    int sample1;
+    int sample2;
+    int step;
+} TrellisNode;
+
+static void adpcm_compress_trellis(AVCodecContext *avctx, const short *samples,
+                                   uint8_t *dst, ADPCMChannelStatus *c, int n)
+{
+#define FREEZE_INTERVAL 128
+    //FIXME 6% faster if frontier is a compile-time constant
+    const int frontier = 1 << avctx->trellis;
+    const int stride = avctx->channels;
+    const int version = avctx->codec->id;
+    const int max_paths = frontier*FREEZE_INTERVAL;
+    TrellisPath paths[max_paths], *p;
+    TrellisNode node_buf[2][frontier];
+    TrellisNode *nodep_buf[2][frontier];
+    TrellisNode **nodes = nodep_buf[0]; // nodes[] is always sorted by .ssd
+    TrellisNode **nodes_next = nodep_buf[1];
+    int pathn = 0, froze = -1, i, j, k;
+
+    assert(!(max_paths&(max_paths-1)));
+
+    memset(nodep_buf, 0, sizeof(nodep_buf));
+    nodes[0] = &node_buf[1][0];
+    nodes[0]->ssd = 0;
+    nodes[0]->path = 0;
+    nodes[0]->step = c->step_index;
+    nodes[0]->sample1 = c->sample1;
+    nodes[0]->sample2 = c->sample2;
+    if(version == CODEC_ID_ADPCM_IMA_WAV)
+        nodes[0]->sample1 = c->prev_sample;
+    if(version == CODEC_ID_ADPCM_MS)
+        nodes[0]->step = c->idelta;
+    if(version == CODEC_ID_ADPCM_YAMAHA) {
+        if(c->step == 0) {
+            nodes[0]->step = 127;
+            nodes[0]->sample1 = 0;
+        } else {
+            nodes[0]->step = c->step;
+            nodes[0]->sample1 = c->predictor;
+        }
+    }
+
+    for(i=0; i<n; i++) {
+        TrellisNode *t = node_buf[i&1];
+        TrellisNode **u;
+        int sample = samples[i*stride];
+        memset(nodes_next, 0, frontier*sizeof(TrellisNode*));
+        for(j=0; j<frontier && nodes[j]; j++) {
+            // higher j have higher ssd already, so they're unlikely to use a suboptimal next sample too
+            const int range = (j < frontier/2) ? 1 : 0;
+            const int step = nodes[j]->step;
+            int nidx;
+            if(version == CODEC_ID_ADPCM_MS) {
+                const int predictor = ((nodes[j]->sample1 * c->coeff1) + (nodes[j]->sample2 * c->coeff2)) / 256;
+                const int div = (sample - predictor) / step;
+                const int nmin = clip(div-range, -8, 6);
+                const int nmax = clip(div+range, -7, 7);
+                for(nidx=nmin; nidx<=nmax; nidx++) {
+                    const int nibble = nidx & 0xf;
+                    int dec_sample = predictor + nidx * step;
+#define STORE_NODE(NAME, STEP_INDEX)\
+                    int d;\
+                    uint32_t ssd;\
+                    CLAMP_TO_SHORT(dec_sample);\
+                    d = sample - dec_sample;\
+                    ssd = nodes[j]->ssd + d*d;\
+                    if(nodes_next[frontier-1] && ssd >= nodes_next[frontier-1]->ssd)\
+                        continue;\
+                    /* Collapse any two states with the same previous sample value. \
+                     * One could also distinguish states by step and by 2nd to last
+                     * sample, but the effects of that are negligible. */\
+                    for(k=0; k<frontier && nodes_next[k]; k++) {\
+                        if(dec_sample == nodes_next[k]->sample1) {\
+                            assert(ssd >= nodes_next[k]->ssd);\
+                            goto next_##NAME;\
+                        }\
+                    }\
+                    for(k=0; k<frontier; k++) {\
+                        if(!nodes_next[k] || ssd < nodes_next[k]->ssd) {\
+                            TrellisNode *u = nodes_next[frontier-1];\
+                            if(!u) {\
+                                assert(pathn < max_paths);\
+                                u = t++;\
+                                u->path = pathn++;\
+                            }\
+                            u->ssd = ssd;\
+                            u->step = STEP_INDEX;\
+                            u->sample2 = nodes[j]->sample1;\
+                            u->sample1 = dec_sample;\
+                            paths[u->path].nibble = nibble;\
+                            paths[u->path].prev = nodes[j]->path;\
+                            memmove(&nodes_next[k+1], &nodes_next[k], (frontier-k-1)*sizeof(TrellisNode*));\
+                            nodes_next[k] = u;\
+                            break;\
+                        }\
+                    }\
+                    next_##NAME:;
+                    STORE_NODE(ms, FFMAX(16, (AdaptationTable[nibble] * step) >> 8));
+                }
+            } else if(version == CODEC_ID_ADPCM_IMA_WAV) {
+#define LOOP_NODES(NAME, STEP_TABLE, STEP_INDEX)\
+                const int predictor = nodes[j]->sample1;\
+                const int div = (sample - predictor) * 4 / STEP_TABLE;\
+                int nmin = clip(div-range, -7, 6);\
+                int nmax = clip(div+range, -6, 7);\
+                if(nmin<=0) nmin--; /* distinguish -0 from +0 */\
+                if(nmax<0) nmax--;\
+                for(nidx=nmin; nidx<=nmax; nidx++) {\
+                    const int nibble = nidx<0 ? 7-nidx : nidx;\
+                    int dec_sample = predictor + (STEP_TABLE * yamaha_difflookup[nibble]) / 8;\
+                    STORE_NODE(NAME, STEP_INDEX);\
+                }
+                LOOP_NODES(ima, step_table[step], clip(step + index_table[nibble], 0, 88));
+            } else { //CODEC_ID_ADPCM_YAMAHA
+                LOOP_NODES(yamaha, step, clip((step * yamaha_indexscale[nibble]) >> 8, 127, 24567));
+#undef LOOP_NODES
+#undef STORE_NODE
+            }
+        }
+
+        u = nodes;
+        nodes = nodes_next;
+        nodes_next = u;
+
+        // prevent overflow
+        if(nodes[0]->ssd > (1<<28)) {
+            for(j=1; j<frontier && nodes[j]; j++)
+                nodes[j]->ssd -= nodes[0]->ssd;
+            nodes[0]->ssd = 0;
+        }
+
+        // merge old paths to save memory
+        if(i == froze + FREEZE_INTERVAL) {
+            p = &paths[nodes[0]->path];
+            for(k=i; k>froze; k--) {
+                dst[k] = p->nibble;
+                p = &paths[p->prev];
+            }
+            froze = i;
+            pathn = 0;
+            // other nodes might use paths that don't coincide with the frozen one.
+            // checking which nodes do so is too slow, so just kill them all.
+            // this also slightly improves quality, but I don't know why.
+            memset(nodes+1, 0, (frontier-1)*sizeof(TrellisNode*));
+        }
+    }
+
+    p = &paths[nodes[0]->path];
+    for(i=n-1; i>froze; i--) {
+        dst[i] = p->nibble;
+        p = &paths[p->prev];
+    }
+
+    c->predictor = nodes[0]->sample1;
+    c->sample1 = nodes[0]->sample1;
+    c->sample2 = nodes[0]->sample2;
+    c->step_index = nodes[0]->step;
+    c->step = nodes[0]->step;
+    c->idelta = nodes[0]->step;
 }
 
 static int adpcm_encode_frame(AVCodecContext *avctx,
@@ -335,6 +466,24 @@ static int adpcm_encode_frame(AVCodecContext *avctx,
             }
 
             /* stereo: 4 bytes (8 samples) for left, 4 bytes for right, 4 bytes left, ... */
+            if(avctx->trellis > 0) {
+                uint8_t buf[2][n*8];
+                adpcm_compress_trellis(avctx, samples, buf[0], &c->status[0], n*8);
+                if(avctx->channels == 2)
+                    adpcm_compress_trellis(avctx, samples+1, buf[1], &c->status[1], n*8);
+                for(i=0; i<n; i++) {
+                    *dst++ = buf[0][8*i+0] | (buf[0][8*i+1] << 4);
+                    *dst++ = buf[0][8*i+2] | (buf[0][8*i+3] << 4);
+                    *dst++ = buf[0][8*i+4] | (buf[0][8*i+5] << 4);
+                    *dst++ = buf[0][8*i+6] | (buf[0][8*i+7] << 4);
+                    if (avctx->channels == 2) {
+                        *dst++ = buf[1][8*i+0] | (buf[1][8*i+1] << 4);
+                        *dst++ = buf[1][8*i+2] | (buf[1][8*i+3] << 4);
+                        *dst++ = buf[1][8*i+4] | (buf[1][8*i+5] << 4);
+                        *dst++ = buf[1][8*i+6] | (buf[1][8*i+7] << 4);
+                    }
+                }
+            } else
             for (; n>0; n--) {
                 *dst = adpcm_ima_compress_sample(&c->status[0], samples[0]) & 0x0F;
                 *dst |= (adpcm_ima_compress_sample(&c->status[0], samples[avctx->channels]) << 4) & 0xF0;
@@ -394,6 +543,21 @@ static int adpcm_encode_frame(AVCodecContext *avctx,
             *dst++ = c->status[i].sample2 >> 8;
         }
 
+        if(avctx->trellis > 0) {
+            int n = avctx->block_align - 7*avctx->channels;
+            uint8_t buf[2][n];
+            if(avctx->channels == 1) {
+                n *= 2;
+                adpcm_compress_trellis(avctx, samples, buf[0], &c->status[0], n);
+                for(i=0; i<n; i+=2)
+                    *dst++ = (buf[0][i] << 4) | buf[0][i+1];
+            } else {
+                adpcm_compress_trellis(avctx, samples, buf[0], &c->status[0], n);
+                adpcm_compress_trellis(avctx, samples+1, buf[1], &c->status[1], n);
+                for(i=0; i<n; i++)
+                    *dst++ = (buf[0][i] << 4) | buf[1][i];
+            }
+        } else
         for(i=7*avctx->channels; i<avctx->block_align; i++) {
             int nibble;
             nibble = adpcm_ms_compress_sample(&c->status[ 0], *samples++)<<4;
@@ -403,6 +567,20 @@ static int adpcm_encode_frame(AVCodecContext *avctx,
         break;
     case CODEC_ID_ADPCM_YAMAHA:
         n = avctx->frame_size / 2;
+        if(avctx->trellis > 0) {
+            uint8_t buf[2][n*2];
+            n *= 2;
+            if(avctx->channels == 1) {
+                adpcm_compress_trellis(avctx, samples, buf[0], &c->status[0], n);
+                for(i=0; i<n; i+=2)
+                    *dst++ = buf[0][i] | (buf[0][i+1] << 4);
+            } else {
+                adpcm_compress_trellis(avctx, samples, buf[0], &c->status[0], n);
+                adpcm_compress_trellis(avctx, samples+1, buf[1], &c->status[1], n);
+                for(i=0; i<n; i++)
+                    *dst++ = buf[0][i] | (buf[1][i] << 4);
+            }
+        } else
         for (; n>0; n--) {
             for(i = 0; i < avctx->channels; i++) {
                 int nibble;
@@ -694,8 +872,10 @@ static int adpcm_decode_frame(AVCodecContext *avctx,
 
         cs->step_index = (*src++) & 0x7F;
 
-        if (cs->step_index > 88) av_log(avctx, AV_LOG_ERROR, "ERROR: step_index = %i\n", cs->step_index);
-        if (cs->step_index > 88) cs->step_index = 88;
+        if (cs->step_index > 88){
+            av_log(avctx, AV_LOG_ERROR, "ERROR: step_index = %i\n", cs->step_index);
+            cs->step_index = 88;
+        }
 
         cs->step = step_table[cs->step_index];
 
@@ -721,35 +901,32 @@ static int adpcm_decode_frame(AVCodecContext *avctx,
         if (avctx->block_align != 0 && buf_size > avctx->block_align)
             buf_size = avctx->block_align;
 
+//        samples_per_block= (block_align-4*chanels)*8 / (bits_per_sample * chanels) + 1;
+
         for(i=0; i<avctx->channels; i++){
             cs = &(c->status[i]);
-            cs->predictor = *src++;
-            cs->predictor |= (*src++) << 8;
-            if(cs->predictor & 0x8000)
-                cs->predictor -= 0x10000;
-            CLAMP_TO_SHORT(cs->predictor);
+            cs->predictor = (int16_t)(src[0] + (src[1]<<8));
+            src+=2;
 
         // XXX: is this correct ??: *samples++ = cs->predictor;
 
             cs->step_index = *src++;
-            if (cs->step_index < 0) cs->step_index = 0;
-            if (cs->step_index > 88) cs->step_index = 88;
-            if (*src++) av_log(avctx, AV_LOG_ERROR, "unused byte should be null !!\n"); /* unused */
+            if (cs->step_index > 88){
+                av_log(avctx, AV_LOG_ERROR, "ERROR: step_index = %i\n", cs->step_index);
+                cs->step_index = 88;
+            }
+            if (*src++) av_log(avctx, AV_LOG_ERROR, "unused byte should be null but is %d!!\n", src[-1]); /* unused */
         }
 
-        for(m=4; src < (buf + buf_size);) {
-            *samples++ = adpcm_ima_expand_nibble(&c->status[0], src[0] & 0x0F, 3);
-            if (st)
-                *samples++ = adpcm_ima_expand_nibble(&c->status[1], src[4] & 0x0F, 3);
-            *samples++ = adpcm_ima_expand_nibble(&c->status[0], (src[0] >> 4) & 0x0F, 3);
-            if (st) {
-                *samples++ = adpcm_ima_expand_nibble(&c->status[1], (src[4] >> 4) & 0x0F, 3);
-                if (!--m) {
-                    m=4;
-                    src+=4;
-                }
+        while(src < buf + buf_size){
+            for(m=0; m<4; m++){
+                for(i=0; i<=st; i++)
+                    *samples++ = adpcm_ima_expand_nibble(&c->status[i], src[4*i] & 0x0F, 3);
+                for(i=0; i<=st; i++)
+                    *samples++ = adpcm_ima_expand_nibble(&c->status[i], src[4*i] >> 4  , 3);
+                src++;
             }
-            src++;
+            src += 4*st;
         }
         break;
     case CODEC_ID_ADPCM_4XM:
@@ -1182,7 +1359,6 @@ ADPCM_CODEC(CODEC_ID_ADPCM_IMA_SMJPEG, adpcm_ima_smjpeg);
 ADPCM_CODEC(CODEC_ID_ADPCM_MS, adpcm_ms);
 ADPCM_CODEC(CODEC_ID_ADPCM_4XM, adpcm_4xm);
 ADPCM_CODEC(CODEC_ID_ADPCM_XA, adpcm_xa);
-ADPCM_CODEC(CODEC_ID_ADPCM_ADX, adpcm_adx);
 ADPCM_CODEC(CODEC_ID_ADPCM_EA, adpcm_ea);
 ADPCM_CODEC(CODEC_ID_ADPCM_CT, adpcm_ct);
 ADPCM_CODEC(CODEC_ID_ADPCM_SWF, adpcm_swf);