[FFmpeg-devel,v6,2/3] libavcodec/flacenc: Implement encoding of 32 bit-per-sample PCM

Add encoding of 32 bit-per-sample PCM to FLAC files to libavcodec.
Coding to this format is at this point considered experimental and
-strict experimental is needed to get ffmpeg to encode such files.
---
 libavcodec/flacenc.c    | 518 ++++++++++++++++++++++++++++++++--------
 libavcodec/put_bits.h   |   7 +
 libavcodec/put_golomb.h |  14 --
 3 files changed, 423 insertions(+), 116 deletions(-)

Martijn van Beurden:
> Add encoding of 32 bit-per-sample PCM to FLAC files to libavcodec.
> Coding to this format is at this point considered experimental and
> -strict experimental is needed to get ffmpeg to encode such files.
> ---
>  libavcodec/flacenc.c    | 518 ++++++++++++++++++++++++++++++++--------
>  libavcodec/put_bits.h   |   7 +
>  libavcodec/put_golomb.h |  14 --
>  3 files changed, 423 insertions(+), 116 deletions(-)
> 
> diff --git a/libavcodec/flacenc.c b/libavcodec/flacenc.c
> index 5d8c3f82be..b69adaf4ba 100644
> --- a/libavcodec/flacenc.c
> +++ b/libavcodec/flacenc.c
> @@ -31,7 +31,6 @@
>  #include "codec_internal.h"
>  #include "encode.h"
>  #include "put_bits.h"
> -#include "put_golomb.h"
>  #include "lpc.h"
>  #include "flac.h"
>  #include "flacdata.h"
> @@ -95,6 +94,7 @@ typedef struct FlacSubframe {
>  
>  typedef struct FlacFrame {
>      FlacSubframe subframes[FLAC_MAX_CHANNELS];
> +    int64_t samples_33bps[FLAC_MAX_BLOCKSIZE];
>      int blocksize;
>      int bs_code[2];
>      uint8_t crc8;
> @@ -282,10 +282,22 @@ static av_cold int flac_encode_init(AVCodecContext *avctx)
>          s->bps_code                = 4;
>          break;
>      case AV_SAMPLE_FMT_S32:
> -        if (avctx->bits_per_raw_sample != 24)
> -            av_log(avctx, AV_LOG_WARNING, "encoding as 24 bits-per-sample\n");
> -        avctx->bits_per_raw_sample = 24;
> -        s->bps_code                = 6;
> +        if (avctx->bits_per_raw_sample <= 24) {
> +            if (avctx->bits_per_raw_sample < 24)
> +                av_log(avctx, AV_LOG_WARNING, "encoding as 24 bits-per-sample\n");
> +            avctx->bits_per_raw_sample = 24;
> +            s->bps_code                = 6;
> +        } else if (avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL) {
> +            av_log(avctx, AV_LOG_WARNING,
> +                   "encoding as 24 bits-per-sample, more is considered "
> +                   "experimental. Add -strict experimental if you want "
> +                   "to encode more than 24 bits-per-sample\n");
> +            avctx->bits_per_raw_sample = 24;
> +            s->bps_code                = 6;
> +        } else {
> +            avctx->bits_per_raw_sample = 32;
> +            s->bps_code = 7;
> +        }
>          break;
>      }
>  
> @@ -533,8 +545,7 @@ static uint64_t rice_count_exact(const int32_t *res, int n, int k)
>      uint64_t count = 0;
>  
>      for (i = 0; i < n; i++) {
> -        int32_t v = -2 * res[i] - 1;
> -        v ^= v >> 31;
> +        unsigned v = ((unsigned)(res[i]) << 1) ^ (res[i] >> 31);
>          count += (v >> k) + 1 + k;
>      }
>      return count;
> @@ -713,8 +724,8 @@ static uint64_t calc_rice_params(RiceContext *rc,
>  
>      tmp_rc.coding_mode = rc->coding_mode;
>  
> -    for (i = 0; i < n; i++)
> -        udata[i] = (2 * data[i]) ^ (data[i] >> 31);
> +    for (i = pred_order; i < n; i++)
> +        udata[i] = ((unsigned)(data[i]) << 1) ^ (data[i] >> 31);
>  
>      calc_sum_top(pmax, exact ? kmax : 0, udata, n, pred_order, sums);
>  
> @@ -812,6 +823,180 @@ static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
>  }
>  
>  
> +static int encode_residual_fixed_with_residual_limit(int32_t *res, const int32_t *smp,
> +                                                      int n, int order)
> +{
> +    /* This function checks for every residual whether it can be
> +     * contained in <INT32_MIN,INT32_MAX]. In case it doesn't, the
> +     * function that called this function has to try something else */
> +    int i;
> +    int64_t res64;
> +
> +    for (i = 0; i < order; i++)
> +        res[i] = smp[i];
> +
> +    if (order == 0) {
> +        for (i = order; i < n; i++) {
> +            if (smp[i] == INT32_MIN)
> +                return 1;
> +            res[i] = smp[i];
> +        }
> +    } else if (order == 1) {
> +        for (i = order; i < n; i++) {
> +            res64 = (int64_t)smp[i] - smp[i-1];
> +            if (res64 <= INT32_MIN || res64 > INT32_MAX)
> +                return 1;
> +            res[i] = res64;
> +        }
> +    } else if (order == 2) {
> +        for (i = order; i < n; i++) {
> +            res64 = (int64_t)smp[i] - 2*(int64_t)smp[i-1] + smp[i-2];
> +            if (res64 <= INT32_MIN || res64 > INT32_MAX)
> +                return 1;
> +            res[i] = res64;
> +        }
> +    } else if (order == 3) {
> +        for (i = order; i < n; i++) {
> +            res64 = (int64_t)smp[i] - 3*(int64_t)smp[i-1] + 3*(int64_t)smp[i-2] - smp[i-3];
> +            if (res64 <= INT32_MIN || res64 > INT32_MAX)
> +                return 1;
> +            res[i] = res64;
> +        }
> +    } else {
> +        for (i = order; i < n; i++) {
> +            res64 = (int64_t)smp[i] - 4*(int64_t)smp[i-1] + 6*(int64_t)smp[i-2] - 4*(int64_t)smp[i-3] + smp[i-4];
> +            if (res64 <= INT32_MIN || res64 > INT32_MAX)
> +                return 1;
> +            res[i] = res64;
> +        }
> +    }
> +    return 0;
> +}
> +
> +
> +static int encode_residual_fixed_with_residual_limit_33bps(int32_t *res, const int64_t *smp,
> +                                                           int n, int order)
> +{
> +    /* This function checks for every residual whether it can be
> +     * contained in <INT32_MIN,INT32_MAX]. In case it doesn't, the
> +     * function that called this function has to try something else */
> +    int i;
> +    int64_t res64;
> +
> +    if (order == 0) {
> +        for (i = order; i < n; i++) {
> +            if (smp[i] <= INT32_MIN || smp[i] > INT32_MAX)
> +                return 1;
> +            res[i] = smp[i];
> +        }
> +    } else if (order == 1) {
> +        for (i = order; i < n; i++) {
> +            res64 = smp[i] - smp[i-1];
> +            if (res64 <= INT32_MIN || res64 > INT32_MAX)
> +                return 1;
> +            res[i] = res64;
> +        }
> +    } else if (order == 2) {
> +        for (i = order; i < n; i++) {
> +            res64 = smp[i] - 2*smp[i-1] + smp[i-2];
> +            if (res64 <= INT32_MIN || res64 > INT32_MAX)
> +                return 1;
> +            res[i] = res64;
> +        }
> +    } else if (order == 3) {
> +        for (i = order; i < n; i++) {
> +            res64 = smp[i] - 3*smp[i-1] + 3*smp[i-2] - smp[i-3];
> +            if (res64 <= INT32_MIN || res64 > INT32_MAX)
> +                return 1;
> +            res[i] = res64;
> +        }
> +    } else {
> +        for (i = order; i < n; i++) {
> +            res64 = smp[i] - 4*smp[i-1] + 6*smp[i-2] - 4*smp[i-3] + smp[i-4];
> +            if (res64 <= INT32_MIN || res64 > INT32_MAX)
> +                return 1;
> +            res[i] = res64;
> +        }
> +    }
> +    return 0;
> +}
> +
> +
> +static int lpc_encode_with_residual_limit(int32_t *res, const int32_t *smp, int len,
> +                                               int order, int32_t *coefs, int shift)
> +{
> +    /* This function checks for every residual whether it can be
> +     * contained in <INT32_MIN,INT32_MAX]. In case it doesn't, the
> +     * function that called this function has to try something else */
> +    for (int i = 0; i < order; i++)
> +        res[i] = smp[i];
> +    for (int i = order; i < len; i++) {
> +        int64_t p = 0, tmp;
> +        for (int j = 0; j < order; j++)
> +            p += (int64_t)coefs[j]*smp[(i-1)-j];
> +        p >>= shift;
> +        tmp = smp[i] - p;
> +        if (tmp <= INT32_MIN || tmp > INT32_MAX)
> +            return 1;
> +        res[i] = tmp;
> +    }
> +    return 0;
> +}
> +
> +static int lpc_encode_with_residual_limit_33bps(int32_t *res, const int64_t *smp, int len,
> +                                               int order, int32_t *coefs, int shift)
> +{
> +    /* This function checks for every residual whether it can be
> +     * contained in <INT32_MIN,INT32_MAX]. In case it doesn't, the
> +     * function that called this function has to try something else */
> +    for (int i = order; i < len; i++) {
> +        int64_t p = 0, tmp;
> +        for (int j = 0; j < order; j++)
> +            p += (int64_t)coefs[j]*smp[(i-1)-j];
> +        p >>= shift;
> +        tmp = smp[i] - p;
> +        if (tmp <= INT32_MIN || tmp > INT32_MAX)
> +            return 1;
> +        res[i] = tmp;
> +    }
> +    return 0;
> +}

These two functions could be generated by a common macro. Same for
encode_residual_fixed_with_residual_limit().

> +
> +static int lpc_encode_choose_datapath(FlacEncodeContext *s, int32_t bps,
> +                                      int32_t *res, const int32_t *smp,
> +                                      const int64_t *smp_33bps, int len,
> +                                      int order, int32_t *coefs, int shift)
> +{
> +    uint64_t max_residual_value = 0;
> +    int64_t max_sample_value = ((int64_t)(1) << (bps-1));
> +    /* This calculates the max size of any residual with the current
> +     * predictor, so we know whether we need to check the residual */
> +    for (int i = 0; i < order; i++)
> +        max_residual_value += FFABS(max_sample_value * coefs[i]);
> +    max_residual_value >>= shift;
> +    max_residual_value += max_sample_value;
> +    if (bps > 32) {
> +        if (lpc_encode_with_residual_limit_33bps(res, smp_33bps, len, order, coefs, shift))
> +            return 1;
> +    } else if (max_residual_value > INT32_MAX) {
> +        if (lpc_encode_with_residual_limit(res, smp, len, order, coefs, shift))
> +            return 1;
> +    } else if (bps + s->options.lpc_coeff_precision + av_log2(order) <= 32) {
> +        s->flac_dsp.lpc16_encode(res, smp, len, order, coefs, shift);
> +    } else {
> +        s->flac_dsp.lpc32_encode(res, smp, len, order, coefs, shift);
> +    }
> +    return 0;
> +}
> +
> +#define DEFAULT_TO_VERBATIM()                               \
> +{                                                           \
> +    sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;    \
> +    if (sub->obits <= 32)                                   \
> +        memcpy(res, smp, n * sizeof(int32_t));              \
> +    return subframe_count_exact(s, sub, 0);                 \
> +}
> +
>  static int encode_residual_ch(FlacEncodeContext *s, int ch)
>  {
>      int i, n;
> @@ -821,28 +1006,38 @@ static int encode_residual_ch(FlacEncodeContext *s, int ch)
>      int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
>      int shift[MAX_LPC_ORDER];
>      int32_t *res, *smp;
> +    int64_t *smp_33bps;
>  
> -    frame = &s->frame;
> -    sub   = &frame->subframes[ch];
> -    res   = sub->residual;
> -    smp   = sub->samples;
> -    n     = frame->blocksize;
> +    frame     = &s->frame;
> +    sub       = &frame->subframes[ch];
> +    res       = sub->residual;
> +    smp       = sub->samples;
> +    smp_33bps = frame->samples_33bps;
> +    n         = frame->blocksize;
>  
>      /* CONSTANT */
> -    for (i = 1; i < n; i++)
> -        if(smp[i] != smp[0])
> -            break;
> -    if (i == n) {
> -        sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
> -        res[0] = smp[0];
> -        return subframe_count_exact(s, sub, 0);
> +    if (sub->obits > 32) {
> +        for (i = 1; i < n; i++)
> +            if(smp_33bps[i] != smp_33bps[0])
> +                break;
> +        if (i == n) {
> +            sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
> +            return subframe_count_exact(s, sub, 0);
> +        }
> +    } else {
> +        for (i = 1; i < n; i++)
> +            if(smp[i] != smp[0])
> +                break;
> +        if (i == n) {
> +            sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
> +            res[0] = smp[0];
> +            return subframe_count_exact(s, sub, 0);
> +        }
>      }
>  
>      /* VERBATIM */
>      if (frame->verbatim_only || n < 5) {
> -        sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
> -        memcpy(res, smp, n * sizeof(int32_t));
> -        return subframe_count_exact(s, sub, 0);
> +        DEFAULT_TO_VERBATIM();
>      }
>  
>      min_order  = s->options.min_prediction_order;
> @@ -859,15 +1054,32 @@ static int encode_residual_ch(FlacEncodeContext *s, int ch)
>          opt_order = 0;
>          bits[0]   = UINT32_MAX;
>          for (i = min_order; i <= max_order; i++) {
> -            encode_residual_fixed(res, smp, n, i);
> +            if (sub->obits == 33) {
> +                if (encode_residual_fixed_with_residual_limit_33bps(res, smp_33bps, n, i))
> +                    continue;
> +            } else if (sub->obits + i >= 32) {
> +                if (encode_residual_fixed_with_residual_limit(res, smp, n, i))
> +                    continue;
> +            } else
> +                encode_residual_fixed(res, smp, n, i);
>              bits[i] = find_subframe_rice_params(s, sub, i);
>              if (bits[i] < bits[opt_order])
>                  opt_order = i;
>          }
> +        if (opt_order == 0 && bits[0] == UINT32_MAX) {
> +            /* No predictor found with residuals within <INT32_MIN,INT32_MAX],
> +             * so encode a verbatim subframe instead */
> +            DEFAULT_TO_VERBATIM();
> +        }
>          sub->order     = opt_order;
>          sub->type_code = sub->type | sub->order;
>          if (sub->order != max_order) {
> -            encode_residual_fixed(res, smp, n, sub->order);
> +            if (sub->obits == 33)
> +                encode_residual_fixed_with_residual_limit_33bps(res, smp_33bps, n, sub->order);
> +            else if (sub->obits + i >= 32)
> +                encode_residual_fixed_with_residual_limit(res, smp, n, sub->order);
> +            else
> +                encode_residual_fixed(res, smp, n, sub->order);
>              find_subframe_rice_params(s, sub, sub->order);
>          }
>          return subframe_count_exact(s, sub, sub->order);
> @@ -875,6 +1087,14 @@ static int encode_residual_ch(FlacEncodeContext *s, int ch)
>  
>      /* LPC */
>      sub->type = FLAC_SUBFRAME_LPC;
> +    if (sub->obits == 33)
> +        /* As ff_lpc_calc_coefs is shared with other codecs and the LSB
> +         * probably isn't predictable anyway, throw away LSB for analysis
> +         * so it fits 32 bit int and existing function can be used
> +         * unmodified */
> +        for (i = 0; i < n; i++)
> +            smp[i] = smp_33bps[i] >> 1;
> +
>      opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, smp, n, min_order, max_order,
>                                    s->options.lpc_coeff_precision, coefs, shift, s->options.lpc_type,
>                                    s->options.lpc_passes, omethod,
> @@ -895,13 +1115,8 @@ static int encode_residual_ch(FlacEncodeContext *s, int ch)
>              order = av_clip(order, min_order - 1, max_order - 1);
>              if (order == last_order)
>                  continue;
> -            if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(order) <= 32) {
> -                s->flac_dsp.lpc16_encode(res, smp, n, order+1, coefs[order],
> -                                         shift[order]);
> -            } else {
> -                s->flac_dsp.lpc32_encode(res, smp, n, order+1, coefs[order],
> -                                         shift[order]);
> -            }
> +            if(lpc_encode_choose_datapath(s, sub->obits, res, smp, smp_33bps, n, order+1, coefs[order], shift[order]))
> +                continue;
>              bits[i] = find_subframe_rice_params(s, sub, order+1);
>              if (bits[i] < bits[opt_index]) {
>                  opt_index = i;
> @@ -915,11 +1130,8 @@ static int encode_residual_ch(FlacEncodeContext *s, int ch)
>          opt_order = 0;
>          bits[0]   = UINT32_MAX;
>          for (i = min_order-1; i < max_order; i++) {
> -            if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(i) <= 32) {
> -                s->flac_dsp.lpc16_encode(res, smp, n, i+1, coefs[i], shift[i]);
> -            } else {
> -                s->flac_dsp.lpc32_encode(res, smp, n, i+1, coefs[i], shift[i]);
> -            }
> +            if(lpc_encode_choose_datapath(s, sub->obits, res, smp, smp_33bps, n, i+1, coefs[i], shift[i]))
> +                continue;
>              bits[i] = find_subframe_rice_params(s, sub, i+1);
>              if (bits[i] < bits[opt_order])
>                  opt_order = i;
> @@ -937,11 +1149,8 @@ static int encode_residual_ch(FlacEncodeContext *s, int ch)
>              for (i = last-step; i <= last+step; i += step) {
>                  if (i < min_order-1 || i >= max_order || bits[i] < UINT32_MAX)
>                      continue;
> -                if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(i) <= 32) {
> -                    s->flac_dsp.lpc32_encode(res, smp, n, i+1, coefs[i], shift[i]);
> -                } else {
> -                    s->flac_dsp.lpc16_encode(res, smp, n, i+1, coefs[i], shift[i]);
> -                }
> +                if(lpc_encode_choose_datapath(s, sub->obits, res, smp, smp_33bps, n, i+1, coefs[i], shift[i]))
> +                    continue;
>                  bits[i] = find_subframe_rice_params(s, sub, i+1);
>                  if (bits[i] < bits[opt_order])
>                      opt_order = i;
> @@ -978,11 +1187,8 @@ static int encode_residual_ch(FlacEncodeContext *s, int ch)
>                  if (diffsum >8)
>                      continue;
>  
> -                if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(opt_order - 1) <= 32) {
> -                    s->flac_dsp.lpc16_encode(res, smp, n, opt_order, lpc_try, shift[opt_order-1]);
> -                } else {
> -                    s->flac_dsp.lpc32_encode(res, smp, n, opt_order, lpc_try, shift[opt_order-1]);
> -                }
> +                if(lpc_encode_choose_datapath(s, sub->obits, res, smp, smp_33bps, n, opt_order, lpc_try, shift[opt_order-1]))
> +                    continue;
>                  score = find_subframe_rice_params(s, sub, opt_order);
>                  if (score < best_score) {
>                      best_score = score;
> @@ -999,10 +1205,10 @@ static int encode_residual_ch(FlacEncodeContext *s, int ch)
>      for (i = 0; i < sub->order; i++)
>          sub->coefs[i] = coefs[sub->order-1][i];
>  
> -    if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(opt_order) <= 32) {
> -        s->flac_dsp.lpc16_encode(res, smp, n, sub->order, sub->coefs, sub->shift);
> -    } else {
> -        s->flac_dsp.lpc32_encode(res, smp, n, sub->order, sub->coefs, sub->shift);
> +    if(lpc_encode_choose_datapath(s, sub->obits, res, smp, smp_33bps, n, sub->order, sub->coefs, sub->shift)) {
> +        /* No predictor found with residuals within <INT32_MIN,INT32_MAX],
> +         * so encode a verbatim subframe instead */
> +        DEFAULT_TO_VERBATIM();
>      }
>  
>      find_subframe_rice_params(s, sub, sub->order);
> @@ -1069,54 +1275,88 @@ static int encode_frame(FlacEncodeContext *s)
>  
>  static void remove_wasted_bits(FlacEncodeContext *s)
>  {
> -    int ch, i;
> +    int ch, i, wasted_bits;
>  
>      for (ch = 0; ch < s->channels; ch++) {
>          FlacSubframe *sub = &s->frame.subframes[ch];
> -        int32_t v         = 0;
>  
> -        for (i = 0; i < s->frame.blocksize; i++) {
> -            v |= sub->samples[i];
> -            if (v & 1)
> -                break;
> -        }
> +        if (sub->obits > 32) {
> +            int64_t v = 0;
> +            for (i = 0; i < s->frame.blocksize; i++) {
> +                v |= s->frame.samples_33bps[i];
> +                if (v & 1)
> +                    break;
> +            }
> +
> +            if (!v || (v & 1))
> +                return;
>  
> -        if (v && !(v & 1)) {
>              v = ff_ctz(v);
>  
> +            /* If any wasted bits are found, samples are moved
> +             * from frame.samples_33bps to frame.subframes[ch] */
>              for (i = 0; i < s->frame.blocksize; i++)
> -                sub->samples[i] >>= v;
> +                sub->samples[i] = s->frame.samples_33bps[i] >> v;
> +            wasted_bits = v;
> +        } else {
> +            int32_t v = 0;
> +            for (i = 0; i < s->frame.blocksize; i++) {
> +                v |= sub->samples[i];
> +                if (v & 1)
> +                    break;
> +            }
> +
> +            if (!v || (v & 1))
> +                return;
>  
> -            sub->wasted = v;
> -            sub->obits -= v;
> +            v = ff_ctz(v);

The return value of ff_ctz() is undefined if its argument is zero, i.e.
if (int)v == 0. You are checking for !v above. Can it happen that (int)v
is zero even if v is nonzero?

>  
> -            /* for 24-bit, check if removing wasted bits makes the range better
> -               suited for using RICE instead of RICE2 for entropy coding */
> -            if (sub->obits <= 17)
> -                sub->rc.coding_mode = CODING_MODE_RICE;
> +            for (i = 0; i < s->frame.blocksize; i++)
> +                sub->samples[i] >>= v;
> +            wasted_bits = v;
>          }
> +
> +        sub->wasted = wasted_bits;
> +        sub->obits -= wasted_bits;
> +
> +        /* for 24-bit, check if removing wasted bits makes the range better
> +         * suited for using RICE instead of RICE2 for entropy coding */
> +        if (sub->obits <= 17)
> +            sub->rc.coding_mode = CODING_MODE_RICE;
>      }
>  }
>  
>  
>  static int estimate_stereo_mode(const int32_t *left_ch, const int32_t *right_ch, int n,
> -                                int max_rice_param)
> +                                int max_rice_param, int bps)
>  {
>      int i, best;
> -    int32_t lt, rt;
>      uint64_t sum[4];
>      uint64_t score[4];
>      int k;
>  
>      /* 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);
> +    if(bps < 30) {
> +        int32_t lt, rt;
> +        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);
> +        }
> +    } else {
> +        int64_t lt, rt;

These variables could be declared in the loop body.

> +        for (i = 2; i < n; i++) {

We allow "for (int i = 2;". This goes for the whole patch.

> +            lt = (int64_t)left_ch[i]  - 2*(int64_t)left_ch[i-1]  + left_ch[i-2];
> +            rt = (int64_t)right_ch[i] - 2*(int64_t)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);
> +        }
>      }
>      /* estimate bit counts */
>      for (i = 0; i < 4; i++) {
> @@ -1147,12 +1387,14 @@ static void channel_decorrelation(FlacEncodeContext *s)
>  {
>      FlacFrame *frame;
>      int32_t *left, *right;
> +    int64_t *side_33bps;
>      int i, n;
>  
> -    frame = &s->frame;
> -    n     = frame->blocksize;
> -    left  = frame->subframes[0].samples;
> -    right = frame->subframes[1].samples;
> +    frame      = &s->frame;
> +    n          = frame->blocksize;
> +    left       = frame->subframes[0].samples;
> +    right      = frame->subframes[1].samples;
> +    side_33bps = frame->samples_33bps;
>  
>      if (s->channels != 2) {
>          frame->ch_mode = FLAC_CHMODE_INDEPENDENT;
> @@ -1161,29 +1403,49 @@ static void channel_decorrelation(FlacEncodeContext *s)
>  
>      if (s->options.ch_mode < 0) {
>          int max_rice_param = (1 << frame->subframes[0].rc.coding_mode) - 2;
> -        frame->ch_mode = estimate_stereo_mode(left, right, n, max_rice_param);
> +        frame->ch_mode = estimate_stereo_mode(left, right, n, max_rice_param, s->avctx->bits_per_raw_sample);
>      } else
>          frame->ch_mode = s->options.ch_mode;
>  
>      /* perform decorrelation and adjust bits-per-sample */
>      if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT)
>          return;
> -    if (frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
> -        int32_t tmp;
> -        for (i = 0; i < n; i++) {
> -            tmp      = left[i];
> -            left[i]  = (tmp + right[i]) >> 1;
> -            right[i] =  tmp - right[i];
> +    if(s->avctx->bits_per_raw_sample == 32) {
> +        if (frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
> +            int64_t tmp;
> +            for (i = 0; i < n; i++) {
> +                tmp           = left[i];
> +                left[i]       = (tmp + right[i]) >> 1;
> +                side_33bps[i] =  tmp - right[i];
> +            }
> +            frame->subframes[1].obits++;
> +        } else if (frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
> +            for (i = 0; i < n; i++)
> +                side_33bps[i] = (int64_t)left[i] - right[i];
> +            frame->subframes[1].obits++;
> +        } else {
> +            for (i = 0; i < n; i++)
> +                side_33bps[i] = (int64_t)left[i] - right[i];
> +            frame->subframes[0].obits++;
>          }
> -        frame->subframes[1].obits++;
> -    } else if (frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
> -        for (i = 0; i < n; i++)
> -            right[i] = left[i] - right[i];
> -        frame->subframes[1].obits++;
>      } else {
> -        for (i = 0; i < n; i++)
> -            left[i] -= right[i];
> -        frame->subframes[0].obits++;
> +        if (frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
> +            int32_t tmp;
> +            for (i = 0; i < n; i++) {
> +                tmp      = left[i];
> +                left[i]  = (tmp + right[i]) >> 1;
> +                right[i] =  tmp - right[i];
> +            }
> +            frame->subframes[1].obits++;
> +        } else if (frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
> +            for (i = 0; i < n; i++)
> +                right[i] = left[i] - right[i];
> +            frame->subframes[1].obits++;
> +        } else {
> +            for (i = 0; i < n; i++)
> +                left[i] -= right[i];
> +            frame->subframes[0].obits++;
> +        }
>      }
>  }
>  
> @@ -1232,6 +1494,25 @@ static void write_frame_header(FlacEncodeContext *s)
>  }
>  
>  
> +static inline void set_sr_golomb_flac(PutBitContext *pb, int i, int k)
> +{
> +    unsigned v, e;
> +
> +    v = ((unsigned)(i) << 1) ^ (i >> 31);
> +
> +    e = (v >> k) + 1;
> +    while (e > 31) {
> +        put_bits(pb, 31, 0);
> +        e -= 31;
> +    }
> +    put_bits(pb, e, 1);
> +    if (k) {
> +        unsigned mask = UINT32_MAX >> (32-k);
> +        put_bits(pb, k, v & mask);
> +    }
> +}
> +
> +
>  static void write_subframes(FlacEncodeContext *s)
>  {
>      int ch;
> @@ -1252,14 +1533,38 @@ static void write_subframes(FlacEncodeContext *s)
>  
>          /* subframe */
>          if (sub->type == FLAC_SUBFRAME_CONSTANT) {
> -            put_sbits(&s->pb, sub->obits, res[0]);
> +            if(sub->obits == 33)
> +                put_sbits63(&s->pb, 33, s->frame.samples_33bps[0]);
> +            else if(sub->obits == 32)
> +                put_bits32(&s->pb, res[0]);
> +            else
> +                put_sbits(&s->pb, sub->obits, res[0]);
>          } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
> -            while (res < frame_end)
> -                put_sbits(&s->pb, sub->obits, *res++);
> +            if (sub->obits == 33) {
> +                int64_t *res64 = s->frame.samples_33bps;
> +                int64_t *frame_end64 = &s->frame.samples_33bps[s->frame.blocksize];
> +                while (res64 < frame_end64)
> +                    put_sbits63(&s->pb, 33, (*res64++));
> +            } else if (sub->obits == 32) {
> +                while (res < frame_end)
> +                    put_bits32(&s->pb, *res++);
> +            } else {
> +                while (res < frame_end)
> +                    put_sbits(&s->pb, sub->obits, *res++);
> +            }
>          } else {
>              /* warm-up samples */
> -            for (i = 0; i < sub->order; i++)
> -                put_sbits(&s->pb, sub->obits, *res++);
> +            if (sub->obits == 33) {
> +                for (i = 0; i < sub->order; i++)
> +                    put_sbits63(&s->pb, 33, s->frame.samples_33bps[i]);
> +                res += sub->order;
> +            } else if (sub->obits == 32) {
> +                for (i = 0; i < sub->order; i++)
> +                    put_bits32(&s->pb, *res++);
> +            } else {
> +                for (i = 0; i < sub->order; i++)
> +                    put_sbits(&s->pb, sub->obits, *res++);
> +            }
>  
>              /* LPC coefficients */
>              if (sub->type == FLAC_SUBFRAME_LPC) {
> @@ -1284,7 +1589,7 @@ static void write_subframes(FlacEncodeContext *s)
>                  int k = sub->rc.params[p];
>                  put_bits(&s->pb, sub->rc.coding_mode, k);
>                  while (res < part_end)
> -                    set_sr_golomb_flac(&s->pb, *res++, k, INT32_MAX, 0);
> +                    set_sr_golomb_flac(&s->pb, *res++, k);
>                  part_end = FFMIN(frame_end, part_end + psize);
>              }
>          }
> @@ -1332,7 +1637,7 @@ static int update_md5_sum(FlacEncodeContext *s, const void *samples)
>                              (const uint16_t *) samples, buf_size / 2);
>          buf = s->md5_buffer;
>  #endif
> -    } else {
> +    } else if (s->avctx->bits_per_raw_sample <= 24) {
>          int i;
>          const int32_t *samples0 = samples;
>          uint8_t *tmp            = s->md5_buffer;
> @@ -1342,6 +1647,15 @@ static int update_md5_sum(FlacEncodeContext *s, const void *samples)
>              AV_WL24(tmp + 3*i, v);
>          }
>          buf = s->md5_buffer;
> +    } else {
> +        /* s->avctx->bits_per_raw_sample <= 32 */
> +        int i;
> +        const int32_t *samples0 = samples;
> +        uint8_t *tmp            = s->md5_buffer;
> +
> +        for (i = 0; i < s->frame.blocksize * s->channels; i++)
> +            AV_WL32(tmp + 4*i, samples0[i]);
> +        buf = s->md5_buffer;
>      }
>      av_md5_update(s->md5ctx, buf, buf_size);
>  
> diff --git a/libavcodec/put_bits.h b/libavcodec/put_bits.h
> index 4b4f977ad5..4561dc131a 100644
> --- a/libavcodec/put_bits.h
> +++ b/libavcodec/put_bits.h
> @@ -363,6 +363,13 @@ static inline void put_bits64(PutBitContext *s, int n, uint64_t value)
>      }
>  }
>  
> +static inline void put_sbits63(PutBitContext *pb, int n, int64_t value)
> +{
> +    av_assert2(n >= 0 && n < 64);
> +
> +    put_bits64(pb, n, (uint64_t)(value) & (~(UINT64_MAX << n)));
> +}
> +
>  /**
>   * Return the pointer to the byte where the bitstream writer will put
>   * the next bit.
> diff --git a/libavcodec/put_golomb.h b/libavcodec/put_golomb.h
> index 9ca911fc3c..df47fd2c0a 100644
> --- a/libavcodec/put_golomb.h
> +++ b/libavcodec/put_golomb.h
> @@ -151,18 +151,4 @@ static inline void set_sr_golomb(PutBitContext *pb, int i, int k, int limit,
>      set_ur_golomb(pb, v, k, limit, esc_len);
>  }
>  
> -/**
> - * write signed golomb rice code (flac).
> - */
> -static inline void set_sr_golomb_flac(PutBitContext *pb, int i, int k,
> -                                      int limit, int esc_len)
> -{
> -    int v;
> -
> -    v  = -2 * i - 1;
> -    v ^= (v >> 31);
> -
> -    set_ur_golomb_jpegls(pb, v, k, limit, esc_len);
> -}
> -
>  #endif /* AVCODEC_PUT_GOLOMB_H */