[FFmpeg-devel] [PATCH] ALAC Encoder
Ramiro Polla
ramiro.polla
Mon Aug 18 00:48:15 CEST 2008
On Sun, Aug 17, 2008 at 7:16 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Mon, Aug 18, 2008 at 02:38:24AM +0530, Jai Menon wrote:
>> Hi,
>>
>> On Sunday 17 Aug 2008 5:17:52 pm Michael Niedermayer wrote:
>> > On Sun, Aug 17, 2008 at 11:17:10AM +0530, Jai Menon wrote:
> [...]
>
> [...]
>
>> Index: libavcodec/alacenc.c
>> ===================================================================
>> --- libavcodec/alacenc.c (revision 14818)
>> +++ libavcodec/alacenc.c (working copy)
>> @@ -33,15 +33,58 @@
>>
>> #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
>
> ok
>
>
>>
>> +#define ALAC_CHMODE_LEFT_RIGHT 1
>> +#define ALAC_CHMODE_LEFT_SIDE 8
>> +#define ALAC_CHMODE_RIGHT_SIDE 9
>> +#define ALAC_CHMODE_MID_SIDE 10
>> +
>
>> +typedef struct RiceContext {
>> + int history_mult;
>> + int initial_history;
>> + int k_modifier;
>> + int rice_modifier;
>> +} RiceContext;
>> +
>> +typedef struct LPCContext {
>> + int lpc_order;
>> + int lpc_coeff[ALAC_MAX_LPC_ORDER+1];
>> + int lpc_quant;
>> +} LPCContext;
>> +
>> +typedef struct AlacEncodeContext {
>> + int compression_level;
>> + int max_coded_frame_size;
>> + int write_sample_size;
>> + int32_t sample_buf[MAX_CHANNELS][DEFAULT_FRAME_SIZE];
>
> ok
>
>> + int32_t predictor_buf[DEFAULT_FRAME_SIZE];
>> int interlacing_shift;
>> int interlacing_leftweight;
>> PutBitContext pbctx;
>
>> + RiceContext rc;
>> + LPCContext lpc[MAX_CHANNELS];
>
> ok
>
>> DSPContext dspctx;
>> AVCodecContext *avctx;
>> } AlacEncodeContext;
>>
>
>>
>> +static void init_sample_buffers(AlacEncodeContext *s, int16_t *input_samples)
>> +{
>> + int ch, i;
>> +
>> + for(ch=0;ch<s->avctx->channels;ch++) {
>> + 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;
>
> ok
>
>
>> @@ -71,7 +114,7 @@
>>
>> 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
>
> ok
>
>
>> @@ -79,6 +122,205 @@
>> put_bits(&s->pbctx, 32, 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;
>> +
>> + 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));
>> +}
>> +
>
> I think this should be using AVCodecContext.min/max_prediction_order
>
>
>
>> +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;
>> + }
>> + }
>
> ok
>
>
>> + if(best == 0) {
>> + return ALAC_CHMODE_LEFT_RIGHT;
>> + } else if(best == 1) {
>> + return ALAC_CHMODE_LEFT_SIDE;
>> + } else if(best == 2) {
>> + return ALAC_CHMODE_RIGHT_SIDE;
>> + } else {
>> + return ALAC_CHMODE_MID_SIDE;
>> + }
>> +}
>
> i think best could simply be returned
>
>
>> +
>> +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;
>> +
>> + mode = estimate_stereo_mode(left, right, n);
>> +
>> + if(mode == ALAC_CHMODE_LEFT_RIGHT) {
>> + s->interlacing_leftweight = 0;
>> + s->interlacing_shift = 0;
>> + return;
>> + }
>> +
>> + if(mode == ALAC_CHMODE_LEFT_SIDE) {
>> + for(i=0; i<n; i++) {
>> + right[i] = left[i] - right[i];
>> + }
>> + s->interlacing_leftweight = 1;
>> + s->interlacing_shift = 0;
>> +
>> + } else {
>> + int32_t tmp;
>> + 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;
>> + }
>
>
> i think 1 mode is missing
> also it could be
> if
> else if
> else if
> else
>
> instead of
> if
> return
> if
> else
>
> which would be cleaner IMHO
>
>
>
>> +}
>> +
>> +static void alac_linear_predictor(AlacEncodeContext *s, int ch)
>> +{
>> + int i;
>> + LPCContext 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
>
>> + i = lpc.lpc_order;
>> + residual[0] = samples[0];
>> + while(i > 0) {
>> + residual[i] = samples[i] - samples[i-1];
>> + i--;
>> + }
>
> this also can be changed to a for() loop
> alternatively residual could be droped and the stuff could all be done
> in place in samples but this may be tricky
>
>
>> + // perform lpc on remaining samples
>> + for(i = lpc.lpc_order + 1; i < s->avctx->frame_size; i++) {
>> + int sum = 0, res_val, j;
>> +
>> + for (j = 0; j < lpc.lpc_order; j++) {
>> + sum += (samples[lpc.lpc_order-j] - samples[0]) *
>> + lpc.lpc_coeff[j];
>> + }
>> + sum += (1 << (lpc.lpc_quant - 1));
>> + sum >>= lpc.lpc_quant;
>> + sum += samples[0];
>> + residual[i] = samples[lpc.lpc_order+1] - sum;
>> + 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);
>
> not sure if its worth but this could be simplified to:
>
> history -= (((history - (x<<9)) * s->rc.history_mult) >> 9);
> (assuming things dont overflow)
>
>
>> +
>> + sign_modifier = 0;
>> + if(x > 0xFFFF)
>> + history = 0xFFFF;
>> +
>> + if((history < 128) && (i < s->avctx->frame_size)) {
>> + unsigned int block_size = 0;
>> +
>
>> + sign_modifier = 1;
>
> unused
>
>
>> + 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;
>
>> @@ -88,7 +330,7 @@
>> 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 +347,7 @@
>>
>> // 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);
>> }
>> @@ -118,8 +360,6 @@
>>
>> avctx->frame_size = DEFAULT_FRAME_SIZE;
>> avctx->bits_per_sample = DEFAULT_SAMPLE_SIZE;
>> - s->channels = avctx->channels;
>> - s->samplerate = avctx->sample_rate;
>>
>> if(avctx->sample_fmt != SAMPLE_FMT_S16) {
>> av_log(avctx, AV_LOG_ERROR, "only pcm_s16 input samples are supported\n");
>> @@ -139,18 +379,18 @@
>> 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;
>> + avctx->frame_size*avctx->channels*avctx->bits_per_sample)>>3;
>>
>> - s->write_sample_size = avctx->bits_per_sample + s->channels - 1; // FIXME: consider wasted_bytes
>> + s->write_sample_size = avctx->bits_per_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+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_sample); // average bitrate
>> + AV_WB32(alac_extradata+32, avctx->sample_rate);
>>
>> // Set relevant extradata fields
>> if(s->compression_level > 0) {
>> @@ -168,19 +408,66 @@
>> s->avctx = avctx;
>> dsputil_init(&s->dspctx, avctx);
>>
>> - allocate_sample_buffers(s);
>> -
>> return 0;
>> }
>>
>> -static av_cold int alac_encode_close(AVCodecContext *avctx)
>> +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;
>> + }
>
> ok
>
>
>> +
>> + init_put_bits(pb, frame, buf_size);
>> +
>> +verbatim:
>
> the label can be moved before the init_put_bits() which makes the second
> call to it before the goto uneeded
>
>
>> + if((s->compression_level == 0) || verbatim_flag) {
>> + // Verbatim mode
>> + 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;
>> + }
>
> ok
>
>> + init_put_bits(pb, frame, buf_size);
>
>> + verbatim_flag = 1;
>> + goto verbatim;
>> + }
>> +
>> + return out_bytes;
>> +}
>> +
>> +static av_cold int alac_encode_close(AVCodecContext *avctx)
>> +{
>> av_freep(&avctx->extradata);
>> avctx->extradata_size = 0;
>> av_freep(&avctx->coded_frame);
>> - free_sample_buffers(s);
>> return 0;
>> }
>>
>
> ok
Applied more ok'd parts.
More information about the ffmpeg-devel
mailing list