aacsbr: prevent multiple ff_aac_sbr_ctx_init()
[ffmpeg.git] / libavcodec / alacenc.c
index 98a3451..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
 
 #define ALAC_ESCAPE_CODE          0x1FF
 #define ALAC_MAX_LPC_ORDER        30
-
+#define DEFAULT_MAX_PRED_ORDER    6
+#define DEFAULT_MIN_PRED_ORDER    4
+#define ALAC_MAX_LPC_PRECISION    9
+#define ALAC_MAX_LPC_SHIFT        9
+
+#define ALAC_CHMODE_LEFT_RIGHT    0
+#define ALAC_CHMODE_LEFT_SIDE     1
+#define ALAC_CHMODE_RIGHT_SIDE    2
+#define ALAC_CHMODE_MID_SIDE      3
+
+typedef struct RiceContext {
+    int history_mult;
+    int initial_history;
+    int k_modifier;
+    int rice_modifier;
+} RiceContext;
+
+typedef struct AlacLPCContext {
+    int lpc_order;
+    int lpc_coeff[ALAC_MAX_LPC_ORDER+1];
+    int lpc_quant;
+} AlacLPCContext;
+
+typedef struct AlacEncodeContext {
+    int compression_level;
+    int min_prediction_order;
+    int max_prediction_order;
+    int max_coded_frame_size;
+    int write_sample_size;
+    int32_t sample_buf[MAX_CHANNELS][DEFAULT_FRAME_SIZE];
+    int32_t predictor_buf[DEFAULT_FRAME_SIZE];
     int interlacing_shift;
     int interlacing_leftweight;
     PutBitContext pbctx;
-    DSPContext dspctx;
+    RiceContext rc;
+    AlacLPCContext lpc[MAX_CHANNELS];
+    LPCContext lpc_ctx;
     AVCodecContext *avctx;
 } AlacEncodeContext;
 
 
+static void init_sample_buffers(AlacEncodeContext *s, const int16_t *input_samples)
+{
+    int ch, i;
+
+    for(ch=0;ch<s->avctx->channels;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;
+        }
+    }
+}
+
 static void encode_scalar(AlacEncodeContext *s, int x, int k, int write_sample_size)
 {
     int divisor, q, r;
@@ -71,24 +117,242 @@ static void encode_scalar(AlacEncodeContext *s, int x, int k, int write_sample_s
 
 static void write_frame_header(AlacEncodeContext *s, int is_verbatim)
 {
-    put_bits(&s->pbctx, 3,  s->channels-1);                 // No. of channels -1
+    put_bits(&s->pbctx, 3,  s->avctx->channels-1);          // No. of channels -1
     put_bits(&s->pbctx, 16, 0);                             // Seems to be zero
     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)
+{
+    int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
+    int shift[MAX_LPC_ORDER];
+    int opt_order;
+
+    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)
+{
+    int i, best;
+    int32_t lt, rt;
+    uint64_t sum[4];
+    uint64_t score[4];
+
+    /* calculate sum of 2nd order residual for each channel */
+    sum[0] = sum[1] = sum[2] = sum[3] = 0;
+    for(i=2; i<n; i++) {
+        lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
+        rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
+        sum[2] += FFABS((lt + rt) >> 1);
+        sum[3] += FFABS(lt - rt);
+        sum[0] += FFABS(lt);
+        sum[1] += FFABS(rt);
+    }
+
+    /* calculate score for each mode */
+    score[0] = sum[0] + sum[1];
+    score[1] = sum[0] + sum[3];
+    score[2] = sum[1] + sum[3];
+    score[3] = sum[2] + sum[3];
+
+    /* return mode with lowest score */
+    best = 0;
+    for(i=1; i<4; i++) {
+        if(score[i] < score[best]) {
+            best = i;
+        }
+    }
+    return best;
+}
+
+static void alac_stereo_decorrelation(AlacEncodeContext *s)
+{
+    int32_t *left = s->sample_buf[0], *right = s->sample_buf[1];
+    int i, mode, n = s->avctx->frame_size;
+    int32_t tmp;
+
+    mode = estimate_stereo_mode(left, right, n);
+
+    switch(mode)
+    {
+        case ALAC_CHMODE_LEFT_RIGHT:
+            s->interlacing_leftweight = 0;
+            s->interlacing_shift = 0;
+            break;
+
+        case ALAC_CHMODE_LEFT_SIDE:
+            for(i=0; i<n; i++) {
+                right[i] = left[i] - right[i];
+            }
+            s->interlacing_leftweight = 1;
+            s->interlacing_shift = 0;
+            break;
+
+        case ALAC_CHMODE_RIGHT_SIDE:
+            for(i=0; i<n; i++) {
+                tmp = right[i];
+                right[i] = left[i] - right[i];
+                left[i] = tmp + (right[i] >> 31);
+            }
+            s->interlacing_leftweight = 1;
+            s->interlacing_shift = 31;
+            break;
+
+        default:
+            for(i=0; i<n; i++) {
+                tmp = left[i];
+                left[i] = (tmp + right[i]) >> 1;
+                right[i] = tmp - right[i];
+            }
+            s->interlacing_leftweight = 1;
+            s->interlacing_shift = 1;
+            break;
+    }
+}
+
+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);
 
-    for(i=0;i<s->channels;i++) {
+    for(i=0;i<s->avctx->channels;i++) {
 
         calc_predictor_params(s, i);
 
@@ -105,7 +369,7 @@ static void write_compressed_frame(AlacEncodeContext *s)
 
     // apply lpc and entropy coding to audio samples
 
-    for(i=0;i<s->channels;i++) {
+    for(i=0;i<s->avctx->channels;i++) {
         alac_linear_predictor(s, i);
         alac_entropy_coder(s);
     }
@@ -114,23 +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;
-    s->channels            = avctx->channels;
-    s->samplerate          = avctx->sample_rate;
+    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;
@@ -138,19 +406,18 @@ 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*s->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 + s->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+21, s->channels);
+    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, s->samplerate*s->channels*avctx->bits_per_sample); // average bitrate
-    AV_WB32(alac_extradata+32, s->samplerate);
+    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
     if(s->compression_level > 0) {
@@ -159,6 +426,34 @@ 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 > ALAC_MAX_LPC_ORDER) {
+            av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", avctx->min_prediction_order);
+                return -1;
+        }
+
+        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 > ALAC_MAX_LPC_ORDER) {
+            av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", avctx->max_prediction_order);
+                return -1;
+        }
+
+        s->max_prediction_order = avctx->max_prediction_order;
+    }
+
+    if(s->max_prediction_order < s->min_prediction_order) {
+        av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
+               s->min_prediction_order, s->max_prediction_order);
+        return -1;
+    }
+
     avctx->extradata = alac_extradata;
     avctx->extradata_size = ALAC_EXTRADATA_SIZE;
 
@@ -166,32 +461,82 @@ 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);
 
-    allocate_sample_buffers(s);
+    return ret;
+}
 
-    return 0;
+static int alac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
+                             int buf_size, void *data)
+{
+    AlacEncodeContext *s = avctx->priv_data;
+    PutBitContext *pb = &s->pbctx;
+    int i, out_bytes, verbatim_flag = 0;
+
+    if(avctx->frame_size > DEFAULT_FRAME_SIZE) {
+        av_log(avctx, AV_LOG_ERROR, "input frame size exceeded\n");
+        return -1;
+    }
+
+    if(buf_size < 2*s->max_coded_frame_size) {
+        av_log(avctx, AV_LOG_ERROR, "buffer size is too small\n");
+        return -1;
+    }
+
+verbatim:
+    init_put_bits(pb, frame, buf_size);
+
+    if((s->compression_level == 0) || verbatim_flag) {
+        // Verbatim mode
+        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++);
+        }
+    } else {
+        init_sample_buffers(s, data);
+        write_frame_header(s, 0);
+        write_compressed_frame(s);
+    }
+
+    put_bits(pb, 3, 7);
+    flush_put_bits(pb);
+    out_bytes = put_bits_count(pb) >> 3;
+
+    if(out_bytes > s->max_coded_frame_size) {
+        /* frame too large. use verbatim mode */
+        if(verbatim_flag || (s->compression_level == 0)) {
+            /* still too large. must be an error. */
+            av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
+            return -1;
+        }
+        verbatim_flag = 1;
+        goto verbatim;
+    }
+
+    return out_bytes;
 }
 
 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);
-    free_sample_buffers(s);
     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)"),
 };