aacsbr: prevent multiple ff_aac_sbr_ctx_init()
[ffmpeg.git] / libavcodec / alacenc.c
index f18ab26..c399471 100644 (file)
  */
 
 #include "avcodec.h"
-#include "bitstream.h"
+#include "put_bits.h"
 #include "dsputil.h"
 #include "lpc.h"
+#include "mathops.h"
 
 #define DEFAULT_FRAME_SIZE        4096
 #define DEFAULT_SAMPLE_SIZE       16
@@ -50,11 +51,11 @@ typedef struct RiceContext {
     int rice_modifier;
 } RiceContext;
 
-typedef struct LPCContext {
+typedef struct AlacLPCContext {
     int lpc_order;
     int lpc_coeff[ALAC_MAX_LPC_ORDER+1];
     int lpc_quant;
-} LPCContext;
+} AlacLPCContext;
 
 typedef struct AlacEncodeContext {
     int compression_level;
@@ -68,18 +69,18 @@ typedef struct AlacEncodeContext {
     int interlacing_leftweight;
     PutBitContext pbctx;
     RiceContext rc;
-    LPCContext lpc[MAX_CHANNELS];
-    DSPContext dspctx;
+    AlacLPCContext lpc[MAX_CHANNELS];
+    LPCContext lpc_ctx;
     AVCodecContext *avctx;
 } AlacEncodeContext;
 
 
-static void init_sample_buffers(AlacEncodeContext *s, int16_t *input_samples)
+static void init_sample_buffers(AlacEncodeContext *s, const int16_t *input_samples)
 {
     int ch, i;
 
     for(ch=0;ch<s->avctx->channels;ch++) {
-        int16_t *sptr = input_samples + ch;
+        const int16_t *sptr = input_samples + ch;
         for(i=0;i<s->avctx->frame_size;i++) {
             s->sample_buf[ch][i] = *sptr;
             sptr += s->avctx->channels;
@@ -121,7 +122,7 @@ static void write_frame_header(AlacEncodeContext *s, int is_verbatim)
     put_bits(&s->pbctx, 1,  1);                             // Sample count is in the header
     put_bits(&s->pbctx, 2,  0);                             // FIXME: Wasted bytes field
     put_bits(&s->pbctx, 1,  is_verbatim);                   // Audio block is verbatim
-    put_bits(&s->pbctx, 32, s->avctx->frame_size);          // No. of samples in the frame
+    put_bits32(&s->pbctx, s->avctx->frame_size);            // No. of samples in the frame
 }
 
 static void calc_predictor_params(AlacEncodeContext *s, int ch)
@@ -130,12 +131,28 @@ static void calc_predictor_params(AlacEncodeContext *s, int ch)
     int shift[MAX_LPC_ORDER];
     int opt_order;
 
-    opt_order = ff_lpc_calc_coefs(&s->dspctx, s->sample_buf[ch], s->avctx->frame_size, DEFAULT_MIN_PRED_ORDER, DEFAULT_MAX_PRED_ORDER,
-                                   ALAC_MAX_LPC_PRECISION, coefs, shift, 1, ORDER_METHOD_EST, ALAC_MAX_LPC_SHIFT, 1);
-
-    s->lpc[ch].lpc_order = opt_order;
-    s->lpc[ch].lpc_quant = shift[opt_order-1];
-    memcpy(s->lpc[ch].lpc_coeff, coefs[opt_order-1], opt_order*sizeof(int));
+    if (s->compression_level == 1) {
+        s->lpc[ch].lpc_order = 6;
+        s->lpc[ch].lpc_quant = 6;
+        s->lpc[ch].lpc_coeff[0] =  160;
+        s->lpc[ch].lpc_coeff[1] = -190;
+        s->lpc[ch].lpc_coeff[2] =  170;
+        s->lpc[ch].lpc_coeff[3] = -130;
+        s->lpc[ch].lpc_coeff[4] =   80;
+        s->lpc[ch].lpc_coeff[5] =  -25;
+    } else {
+        opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, s->sample_buf[ch],
+                                      s->avctx->frame_size,
+                                      s->min_prediction_order,
+                                      s->max_prediction_order,
+                                      ALAC_MAX_LPC_PRECISION, coefs, shift,
+                                      FF_LPC_TYPE_LEVINSON, 0,
+                                      ORDER_METHOD_EST, ALAC_MAX_LPC_SHIFT, 1);
+
+        s->lpc[ch].lpc_order = opt_order;
+        s->lpc[ch].lpc_quant = shift[opt_order-1];
+        memcpy(s->lpc[ch].lpc_coeff, coefs[opt_order-1], opt_order*sizeof(int));
+    }
 }
 
 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
@@ -217,12 +234,121 @@ static void alac_stereo_decorrelation(AlacEncodeContext *s)
     }
 }
 
+static void alac_linear_predictor(AlacEncodeContext *s, int ch)
+{
+    int i;
+    AlacLPCContext lpc = s->lpc[ch];
+
+    if(lpc.lpc_order == 31) {
+        s->predictor_buf[0] = s->sample_buf[ch][0];
+
+        for(i=1; i<s->avctx->frame_size; i++)
+            s->predictor_buf[i] = s->sample_buf[ch][i] - s->sample_buf[ch][i-1];
+
+        return;
+    }
+
+    // generalised linear predictor
+
+    if(lpc.lpc_order > 0) {
+        int32_t *samples  = s->sample_buf[ch];
+        int32_t *residual = s->predictor_buf;
+
+        // generate warm-up samples
+        residual[0] = samples[0];
+        for(i=1;i<=lpc.lpc_order;i++)
+            residual[i] = samples[i] - samples[i-1];
+
+        // perform lpc on remaining samples
+        for(i = lpc.lpc_order + 1; i < s->avctx->frame_size; i++) {
+            int sum = 1 << (lpc.lpc_quant - 1), res_val, j;
+
+            for (j = 0; j < lpc.lpc_order; j++) {
+                sum += (samples[lpc.lpc_order-j] - samples[0]) *
+                        lpc.lpc_coeff[j];
+            }
+
+            sum >>= lpc.lpc_quant;
+            sum += samples[0];
+            residual[i] = sign_extend(samples[lpc.lpc_order+1] - sum,
+                                      s->write_sample_size);
+            res_val = residual[i];
+
+            if(res_val) {
+                int index = lpc.lpc_order - 1;
+                int neg = (res_val < 0);
+
+                while(index >= 0 && (neg ? (res_val < 0):(res_val > 0))) {
+                    int val = samples[0] - samples[lpc.lpc_order - index];
+                    int sign = (val ? FFSIGN(val) : 0);
+
+                    if(neg)
+                        sign*=-1;
+
+                    lpc.lpc_coeff[index] -= sign;
+                    val *= sign;
+                    res_val -= ((val >> lpc.lpc_quant) *
+                            (lpc.lpc_order - index));
+                    index--;
+                }
+            }
+            samples++;
+        }
+    }
+}
+
+static void alac_entropy_coder(AlacEncodeContext *s)
+{
+    unsigned int history = s->rc.initial_history;
+    int sign_modifier = 0, i, k;
+    int32_t *samples = s->predictor_buf;
+
+    for(i=0;i < s->avctx->frame_size;) {
+        int x;
+
+        k = av_log2((history >> 9) + 3);
+
+        x = -2*(*samples)-1;
+        x ^= (x>>31);
+
+        samples++;
+        i++;
+
+        encode_scalar(s, x - sign_modifier, k, s->write_sample_size);
+
+        history += x * s->rc.history_mult
+                   - ((history * s->rc.history_mult) >> 9);
+
+        sign_modifier = 0;
+        if(x > 0xFFFF)
+            history = 0xFFFF;
+
+        if((history < 128) && (i < s->avctx->frame_size)) {
+            unsigned int block_size = 0;
+
+            k = 7 - av_log2(history) + ((history + 16) >> 6);
+
+            while((*samples == 0) && (i < s->avctx->frame_size)) {
+                samples++;
+                i++;
+                block_size++;
+            }
+            encode_scalar(s, block_size, k, 16);
+
+            sign_modifier = (block_size <= 0xFFFF);
+
+            history = 0;
+        }
+
+    }
+}
+
 static void write_compressed_frame(AlacEncodeContext *s)
 {
     int i, j;
 
-    /* only simple mid/side decorrelation supported as of now */
-    alac_stereo_decorrelation(s);
+    if(s->avctx->channels == 2)
+        alac_stereo_decorrelation(s);
     put_bits(&s->pbctx, 8, s->interlacing_shift);
     put_bits(&s->pbctx, 8, s->interlacing_leftweight);
 
@@ -252,21 +378,27 @@ static void write_compressed_frame(AlacEncodeContext *s)
 static av_cold int alac_encode_init(AVCodecContext *avctx)
 {
     AlacEncodeContext *s    = avctx->priv_data;
+    int ret;
     uint8_t *alac_extradata = av_mallocz(ALAC_EXTRADATA_SIZE+1);
 
     avctx->frame_size      = DEFAULT_FRAME_SIZE;
-    avctx->bits_per_sample = DEFAULT_SAMPLE_SIZE;
+    avctx->bits_per_coded_sample = DEFAULT_SAMPLE_SIZE;
 
-    if(avctx->sample_fmt != SAMPLE_FMT_S16) {
+    if(avctx->sample_fmt != AV_SAMPLE_FMT_S16) {
         av_log(avctx, AV_LOG_ERROR, "only pcm_s16 input samples are supported\n");
         return -1;
     }
 
+    if(avctx->channels > 2) {
+        av_log(avctx, AV_LOG_ERROR, "channels > 2 not supported\n");
+        return AVERROR_PATCHWELCOME;
+    }
+
     // Set default compression level
     if(avctx->compression_level == FF_COMPRESSION_DEFAULT)
-        s->compression_level = 1;
+        s->compression_level = 2;
     else
-        s->compression_level = av_clip(avctx->compression_level, 0, 1);
+        s->compression_level = av_clip(avctx->compression_level, 0, 2);
 
     // Initialize default Rice parameters
     s->rc.history_mult    = 40;
@@ -274,18 +406,17 @@ static av_cold int alac_encode_init(AVCodecContext *avctx)
     s->rc.k_modifier      = 14;
     s->rc.rice_modifier   = 4;
 
-    s->max_coded_frame_size = (ALAC_FRAME_HEADER_SIZE + ALAC_FRAME_FOOTER_SIZE +
-                               avctx->frame_size*avctx->channels*avctx->bits_per_sample)>>3;
+    s->max_coded_frame_size = 8 + (avctx->frame_size*avctx->channels*avctx->bits_per_coded_sample>>3);
 
-    s->write_sample_size  = avctx->bits_per_sample + avctx->channels - 1; // FIXME: consider wasted_bytes
+    s->write_sample_size  = avctx->bits_per_coded_sample + avctx->channels - 1; // FIXME: consider wasted_bytes
 
     AV_WB32(alac_extradata,    ALAC_EXTRADATA_SIZE);
     AV_WB32(alac_extradata+4,  MKBETAG('a','l','a','c'));
     AV_WB32(alac_extradata+12, avctx->frame_size);
-    AV_WB8 (alac_extradata+17, avctx->bits_per_sample);
+    AV_WB8 (alac_extradata+17, avctx->bits_per_coded_sample);
     AV_WB8 (alac_extradata+21, avctx->channels);
     AV_WB32(alac_extradata+24, s->max_coded_frame_size);
-    AV_WB32(alac_extradata+28, avctx->sample_rate*avctx->channels*avctx->bits_per_sample); // average bitrate
+    AV_WB32(alac_extradata+28, avctx->sample_rate*avctx->channels*avctx->bits_per_coded_sample); // average bitrate
     AV_WB32(alac_extradata+32, avctx->sample_rate);
 
     // Set relevant extradata fields
@@ -295,9 +426,10 @@ static av_cold int alac_encode_init(AVCodecContext *avctx)
         AV_WB8(alac_extradata+20, s->rc.k_modifier);
     }
 
+    s->min_prediction_order = DEFAULT_MIN_PRED_ORDER;
     if(avctx->min_prediction_order >= 0) {
         if(avctx->min_prediction_order < MIN_LPC_ORDER ||
-            avctx->min_prediction_order > MAX_LPC_ORDER) {
+           avctx->min_prediction_order > ALAC_MAX_LPC_ORDER) {
             av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", avctx->min_prediction_order);
                 return -1;
         }
@@ -305,9 +437,10 @@ static av_cold int alac_encode_init(AVCodecContext *avctx)
         s->min_prediction_order = avctx->min_prediction_order;
     }
 
+    s->max_prediction_order = DEFAULT_MAX_PRED_ORDER;
     if(avctx->max_prediction_order >= 0) {
         if(avctx->max_prediction_order < MIN_LPC_ORDER ||
-           avctx->max_prediction_order > MAX_LPC_ORDER) {
+           avctx->max_prediction_order > ALAC_MAX_LPC_ORDER) {
             av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", avctx->max_prediction_order);
                 return -1;
         }
@@ -328,9 +461,10 @@ static av_cold int alac_encode_init(AVCodecContext *avctx)
     avctx->coded_frame->key_frame = 1;
 
     s->avctx = avctx;
-    dsputil_init(&s->dspctx, avctx);
+    ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size, s->max_prediction_order,
+                      FF_LPC_TYPE_LEVINSON);
 
-    return 0;
+    return ret;
 }
 
 static int alac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
@@ -355,7 +489,7 @@ verbatim:
 
     if((s->compression_level == 0) || verbatim_flag) {
         // Verbatim mode
-        int16_t *samples = data;
+        const int16_t *samples = data;
         write_frame_header(s, 1);
         for(i=0; i<avctx->frame_size*avctx->channels; i++) {
             put_sbits(pb, 16, *samples++);
@@ -386,20 +520,23 @@ verbatim:
 
 static av_cold int alac_encode_close(AVCodecContext *avctx)
 {
+    AlacEncodeContext *s = avctx->priv_data;
+    ff_lpc_end(&s->lpc_ctx);
     av_freep(&avctx->extradata);
     avctx->extradata_size = 0;
     av_freep(&avctx->coded_frame);
     return 0;
 }
 
-AVCodec alac_encoder = {
-    "alac",
-    CODEC_TYPE_AUDIO,
-    CODEC_ID_ALAC,
-    sizeof(AlacEncodeContext),
-    alac_encode_init,
-    alac_encode_frame,
-    alac_encode_close,
+AVCodec ff_alac_encoder = {
+    .name           = "alac",
+    .type           = AVMEDIA_TYPE_AUDIO,
+    .id             = CODEC_ID_ALAC,
+    .priv_data_size = sizeof(AlacEncodeContext),
+    .init           = alac_encode_init,
+    .encode         = alac_encode_frame,
+    .close          = alac_encode_close,
     .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
+    .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE},
     .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
 };