Remove the add bias hack for the C version of DSPContext.float_to_int16_*().

(cherry picked from commit 9d06d7bce3babb82ed650c13ed13a57f6f626a71)

12 files changed, 54 insertions, 148 deletions

diff --git a/libavcodec/aac.h b/libavcodec/aac.h
index b40d2c0003..714e314cba 100644
--- a/libavcodec/aac.h
+++ b/libavcodec/aac.h
@@ -276,7 +276,6 @@ typedef struct {
      * @{
      */
     float *output_data[MAX_CHANNELS];                 ///< Points to each element's 'ret' buffer (PCM output).
-    float add_bias;                                   ///< offset for dsp.float_to_int16
     float sf_scale;                                   ///< Pre-scale for correct IMDCT and dsp.float_to_int16.
     int sf_offset;                                    ///< offset into pow2sf_tab as appropriate for dsp.float_to_int16
     /** @} */
diff --git a/libavcodec/aacdec.c b/libavcodec/aacdec.c
index fd65b1c3ba..fddec17fcc 100644
--- a/libavcodec/aacdec.c
+++ b/libavcodec/aacdec.c
@@ -566,18 +566,10 @@ static av_cold int aac_decode_init(AVCodecContext *avctx)
     ac->random_state = 0x1f2e3d4c;
 
     // -1024 - Compensate wrong IMDCT method.
-    // 32768 - Required to scale values to the correct range for the bias method
-    //         for float to int16 conversion.
-
-    if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
-        ac->add_bias  = 385.0f;
-        ac->sf_scale  = 1. / (-1024. * 32768.);
-        ac->sf_offset = 0;
-    } else {
-        ac->add_bias  = 0.0f;
+    // 60    - Required to scale values to the correct range [-32768,32767]
+    //         for float to int16 conversion. (1 << (60 / 4)) == 32768
         ac->sf_scale  = 1. / -1024.;
         ac->sf_offset = 60;
-    }
 
     ff_aac_tableinit();
 
@@ -1701,7 +1693,7 @@ static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
 /**
  * Conduct IMDCT and windowing.
  */
-static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce, float bias)
+static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
 {
     IndividualChannelStream *ics = &sce->ics;
     float *in    = sce->coeffs;
@@ -1729,29 +1721,29 @@ static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce, float
      */
     if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
             (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
-        ac->dsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, bias, 512);
+        ac->dsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, 0, 512);
     } else {
         for (i = 0; i < 448; i++)
-            out[i] = saved[i] + bias;
+            out[i] = saved[i];
 
         if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
-            ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448,      buf + 0*128, swindow_prev, bias, 64);
-            ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow,      bias, 64);
-            ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow,      bias, 64);
-            ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow,      bias, 64);
-            ac->dsp.vector_fmul_window(temp,              buf + 3*128 + 64, buf + 4*128, swindow,      bias, 64);
+            ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448,      buf + 0*128, swindow_prev, 0, 64);
+            ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow,      0, 64);
+            ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow,      0, 64);
+            ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow,      0, 64);
+            ac->dsp.vector_fmul_window(temp,              buf + 3*128 + 64, buf + 4*128, swindow,      0, 64);
             memcpy(                    out + 448 + 4*128, temp, 64 * sizeof(float));
         } else {
-            ac->dsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, bias, 64);
+            ac->dsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, 0, 64);
             for (i = 576; i < 1024; i++)
-                out[i] = buf[i-512] + bias;
+                out[i] = buf[i-512];
         }
     }
 
     // buffer update
     if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
         for (i = 0; i < 64; i++)
-            saved[i] = temp[64 + i] - bias;
+            saved[i] = temp[64 + i];
         ac->dsp.vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
         ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
         ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
@@ -1811,13 +1803,12 @@ static void apply_independent_coupling(AACContext *ac,
 {
     int i;
     const float gain = cce->coup.gain[index][0];
-    const float bias = ac->add_bias;
     const float *src = cce->ch[0].ret;
     float *dest = target->ret;
     const int len = 1024 << (ac->m4ac.sbr == 1);
 
     for (i = 0; i < len; i++)
-        dest[i] += gain * (src[i] - bias);
+        dest[i] += gain * src[i];
 }
 
 /**
@@ -1861,7 +1852,6 @@ static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
 static void spectral_to_sample(AACContext *ac)
 {
     int i, type;
-    float imdct_bias = (ac->m4ac.sbr <= 0) ? ac->add_bias : 0.0f;
     for (type = 3; type >= 0; type--) {
         for (i = 0; i < MAX_ELEM_ID; i++) {
             ChannelElement *che = ac->che[type][i];
@@ -1875,9 +1865,9 @@ static void spectral_to_sample(AACContext *ac)
                 if (type <= TYPE_CPE)
                     apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
                 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
-                    imdct_and_windowing(ac, &che->ch[0], imdct_bias);
+                    imdct_and_windowing(ac, &che->ch[0]);
                     if (type == TYPE_CPE) {
-                        imdct_and_windowing(ac, &che->ch[1], imdct_bias);
+                        imdct_and_windowing(ac, &che->ch[1]);
                     }
                     if (ac->m4ac.sbr > 0) {
                         ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
diff --git a/libavcodec/aacsbr.c b/libavcodec/aacsbr.c
index 050305a3fe..9b10bf25fe 100644
--- a/libavcodec/aacsbr.c
+++ b/libavcodec/aacsbr.c
@@ -1175,12 +1175,10 @@ static void sbr_qmf_analysis(DSPContext *dsp, FFTContext *mdct, const float *in,
 static void sbr_qmf_synthesis(DSPContext *dsp, FFTContext *mdct,
                               float *out, float X[2][38][64],
                               float mdct_buf[2][64],
-                              float *v0, int *v_off, const unsigned int div,
-                              float bias, float scale)
+                              float *v0, int *v_off, const unsigned int div)
 {
     int i, n;
     const float *sbr_qmf_window = div ? sbr_qmf_window_ds : sbr_qmf_window_us;
-    int scale_and_bias = scale != 1.0f || bias != 0.0f;
     float *v;
     for (i = 0; i < 32; i++) {
         if (*v_off == 0) {
@@ -1222,9 +1220,6 @@ static void sbr_qmf_synthesis(DSPContext *dsp, FFTContext *mdct,
         dsp->vector_fmul_add(out, v + ( 960 >> div), sbr_qmf_window + (448 >> div), out   , 64 >> div);
         dsp->vector_fmul_add(out, v + (1024 >> div), sbr_qmf_window + (512 >> div), out   , 64 >> div);
         dsp->vector_fmul_add(out, v + (1216 >> div), sbr_qmf_window + (576 >> div), out   , 64 >> div);
-        if (scale_and_bias)
-            for (n = 0; n < 64 >> div; n++)
-                out[n] = out[n] * scale + bias;
         out += 64 >> div;
     }
 }
@@ -1760,12 +1755,10 @@ void ff_sbr_apply(AACContext *ac, SpectralBandReplication *sbr, int id_aac,
     sbr_qmf_synthesis(&ac->dsp, &sbr->mdct, L, sbr->X[0], sbr->qmf_filter_scratch,
                       sbr->data[0].synthesis_filterbank_samples,
                       &sbr->data[0].synthesis_filterbank_samples_offset,
-                      downsampled,
-                      ac->add_bias, -1024 * ac->sf_scale);
+                      downsampled);
     if (nch == 2)
         sbr_qmf_synthesis(&ac->dsp, &sbr->mdct, R, sbr->X[1], sbr->qmf_filter_scratch,
                           sbr->data[1].synthesis_filterbank_samples,
                           &sbr->data[1].synthesis_filterbank_samples_offset,
-                          downsampled,
-                          ac->add_bias, -1024 * ac->sf_scale);
+                          downsampled);
 }
diff --git a/libavcodec/ac3dec.c b/libavcodec/ac3dec.c
index a1e731dcf9..2f9bc261bd 100644
--- a/libavcodec/ac3dec.c
+++ b/libavcodec/ac3dec.c
@@ -196,13 +196,7 @@ static av_cold int ac3_decode_init(AVCodecContext *avctx)
     av_lfg_init(&s->dith_state, 0);
 
     /* set bias values for float to int16 conversion */
-    if(s->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
-        s->add_bias = 385.0f;
-        s->mul_bias = 1.0f;
-    } else {
-        s->add_bias = 0.0f;
         s->mul_bias = 32767.0f;
-    }
 
     /* allow downmixing to stereo or mono */
     if (avctx->channels > 0 && avctx->request_channels > 0 &&
@@ -626,9 +620,6 @@ static void do_rematrixing(AC3DecodeContext *s)
 static inline void do_imdct(AC3DecodeContext *s, int channels)
 {
     int ch;
-    float add_bias = s->add_bias;
-    if(s->out_channels==1 && channels>1)
-        add_bias *= LEVEL_MINUS_3DB; // compensate for the gain in downmix
 
     for (ch=1; ch<=channels; ch++) {
         if (s->block_switch[ch]) {
@@ -637,13 +628,13 @@ static inline void do_imdct(AC3DecodeContext *s, int channels)
             for(i=0; i<128; i++)
                 x[i] = s->transform_coeffs[ch][2*i];
             ff_imdct_half(&s->imdct_256, s->tmp_output, x);
-            s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
+            s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, 0, 128);
             for(i=0; i<128; i++)
                 x[i] = s->transform_coeffs[ch][2*i+1];
             ff_imdct_half(&s->imdct_256, s->delay[ch-1], x);
         } else {
             ff_imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
-            s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
+            s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, 0, 128);
             memcpy(s->delay[ch-1], s->tmp_output+128, 128*sizeof(float));
         }
     }
diff --git a/libavcodec/ac3dec.h b/libavcodec/ac3dec.h
index 0707769478..55520cdcee 100644
--- a/libavcodec/ac3dec.h
+++ b/libavcodec/ac3dec.h
@@ -190,7 +190,6 @@ typedef struct {
 
 ///@defgroup opt optimization
     DSPContext dsp;                         ///< for optimization
-    float add_bias;                         ///< offset for float_to_int16 conversion
     float mul_bias;                         ///< scaling for float_to_int16 conversion
 ///@}
 
diff --git a/libavcodec/binkaudio.c b/libavcodec/binkaudio.c
index 7f579822db..ae2f6c88b0 100644
--- a/libavcodec/binkaudio.c
+++ b/libavcodec/binkaudio.c
@@ -222,11 +222,6 @@ static void decode_block(BinkAudioContext *s, short *out, int use_dct)
             ff_rdft_calc(&s->trans.rdft, coeffs);
     }
 
-    if (s->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
-        for (i = 0; i < s->channels; i++)
-            for (j = 0; j < s->frame_len; j++)
-                s->coeffs_ptr[i][j] = 385.0 + s->coeffs_ptr[i][j]*(1.0/32767.0);
-    }
     s->dsp.float_to_int16_interleave(out, (const float **)s->coeffs_ptr, s->frame_len, s->channels);
 
     if (!s->first) {
diff --git a/libavcodec/dca.c b/libavcodec/dca.c
index c1c70cde1b..aa71411efa 100644
--- a/libavcodec/dca.c
+++ b/libavcodec/dca.c
@@ -311,7 +311,6 @@ typedef struct {
     DECLARE_ALIGNED(16, float, raXin)[32];
 
     int output;                 ///< type of output
-    float add_bias;             ///< output bias
     float scale_bias;           ///< output scale
 
     DECLARE_ALIGNED(16, float, subband_samples)[DCA_BLOCKS_MAX][DCA_PRIM_CHANNELS_MAX][DCA_SUBBANDS][8];
@@ -868,7 +867,7 @@ static int dca_subframe_header(DCAContext * s, int base_channel, int block_index
 
 static void qmf_32_subbands(DCAContext * s, int chans,
                             float samples_in[32][8], float *samples_out,
-                            float scale, float bias)
+                            float scale)
 {
     const float *prCoeff;
     int i;
@@ -897,7 +896,7 @@ static void qmf_32_subbands(DCAContext * s, int chans,
         s->synth.synth_filter_float(&s->imdct,
                               s->subband_fir_hist[chans], &s->hist_index[chans],
                               s->subband_fir_noidea[chans], prCoeff,
-                              samples_out, s->raXin, scale, bias);
+                              samples_out, s->raXin, scale, 0);
         samples_out+= 32;
 
     }
@@ -905,8 +904,7 @@ static void qmf_32_subbands(DCAContext * s, int chans,
 
 static void lfe_interpolation_fir(DCAContext *s, int decimation_select,
                                   int num_deci_sample, float *samples_in,
-                                  float *samples_out, float scale,
-                                  float bias)
+                                  float *samples_out, float scale)
 {
     /* samples_in: An array holding decimated samples.
      *   Samples in current subframe starts from samples_in[0],
@@ -931,7 +929,7 @@ static void lfe_interpolation_fir(DCAContext *s, int decimation_select,
     /* Interpolation */
     for (deciindex = 0; deciindex < num_deci_sample; deciindex++) {
         s->dcadsp.lfe_fir(samples_out, samples_in, prCoeff, decifactor,
-                          scale, bias);
+                          scale, 0);
         samples_in++;
         samples_out += 2 * decifactor;
     }
@@ -939,19 +937,19 @@ static void lfe_interpolation_fir(DCAContext *s, int decimation_select,
 
 /* downmixing routines */
 #define MIX_REAR1(samples, si1, rs, coef) \
-     samples[i]     += (samples[si1] - add_bias) * coef[rs][0];  \
-     samples[i+256] += (samples[si1] - add_bias) * coef[rs][1];
+     samples[i]     += samples[si1] * coef[rs][0];  \
+     samples[i+256] += samples[si1] * coef[rs][1];
 
 #define MIX_REAR2(samples, si1, si2, rs, coef) \
-     samples[i]     += (samples[si1] - add_bias) * coef[rs][0] + (samples[si2] - add_bias) * coef[rs+1][0]; \
-     samples[i+256] += (samples[si1] - add_bias) * coef[rs][1] + (samples[si2] - add_bias) * coef[rs+1][1];
+     samples[i]     += samples[si1] * coef[rs][0] + samples[si2] * coef[rs+1][0]; \
+     samples[i+256] += samples[si1] * coef[rs][1] + samples[si2] * coef[rs+1][1];
 
 #define MIX_FRONT3(samples, coef) \
-    t = samples[i+c] - add_bias; \
-    u = samples[i+l] - add_bias; \
-    v = samples[i+r] - add_bias; \
-    samples[i]     = t * coef[0][0] + u * coef[1][0] + v * coef[2][0] + add_bias; \
-    samples[i+256] = t * coef[0][1] + u * coef[1][1] + v * coef[2][1] + add_bias;
+    t = samples[i+c]; \
+    u = samples[i+l]; \
+    v = samples[i+r]; \
+    samples[i]     = t * coef[0][0] + u * coef[1][0] + v * coef[2][0]; \
+    samples[i+256] = t * coef[0][1] + u * coef[1][1] + v * coef[2][1];
 
 #define DOWNMIX_TO_STEREO(op1, op2) \
     for (i = 0; i < 256; i++){ \
@@ -961,7 +959,7 @@ static void lfe_interpolation_fir(DCAContext *s, int decimation_select,
 
 static void dca_downmix(float *samples, int srcfmt,
                         int downmix_coef[DCA_PRIM_CHANNELS_MAX][2],
-                        const int8_t *channel_mapping, float add_bias)
+                        const int8_t *channel_mapping)
 {
     int c,l,r,sl,sr,s;
     int i;
@@ -1193,13 +1191,12 @@ static int dca_filter_channels(DCAContext * s, int block_index)
 /*        static float pcm_to_double[8] =
             {32768.0, 32768.0, 524288.0, 524288.0, 0, 8388608.0, 8388608.0};*/
          qmf_32_subbands(s, k, subband_samples[k], &s->samples[256 * s->channel_order_tab[k]],
-                            M_SQRT1_2*s->scale_bias /*pcm_to_double[s->source_pcm_res] */ ,
-                            s->add_bias );
+                         M_SQRT1_2*s->scale_bias /*pcm_to_double[s->source_pcm_res] */ );
     }
 
     /* Down mixing */
     if (s->avctx->request_channels == 2 && s->prim_channels > 2) {
-        dca_downmix(s->samples, s->amode, s->downmix_coef, s->channel_order_tab, s->add_bias);
+        dca_downmix(s->samples, s->amode, s->downmix_coef, s->channel_order_tab);
     }
 
     /* Generate LFE samples for this subsubframe FIXME!!! */
@@ -1207,7 +1204,7 @@ static int dca_filter_channels(DCAContext * s, int block_index)
         lfe_interpolation_fir(s, s->lfe, 2 * s->lfe,
                               s->lfe_data + 2 * s->lfe * (block_index + 4),
                               &s->samples[256 * dca_lfe_index[s->amode]],
-                              (1.0/256.0)*s->scale_bias,  s->add_bias);
+                              (1.0/256.0)*s->scale_bias);
         /* Outputs 20bits pcm samples */
     }
 
@@ -1798,8 +1795,8 @@ static int dca_decode_frame(AVCodecContext * avctx,
             float* rt_chan   = s->samples + s->channel_order_tab[s->xch_base_channel - 1] * 256;
             int j;
             for(j = 0; j < 256; ++j) {
-                lt_chan[j] -= (back_chan[j] - s->add_bias) * M_SQRT1_2;
-                rt_chan[j] -= (back_chan[j] - s->add_bias) * M_SQRT1_2;
+                lt_chan[j] -= back_chan[j] * M_SQRT1_2;
+                rt_chan[j] -= back_chan[j] * M_SQRT1_2;
             }
         }
 
@@ -1841,11 +1838,6 @@ static av_cold int dca_decode_init(AVCodecContext * avctx)
         s->samples_chanptr[i] = s->samples + i * 256;
     avctx->sample_fmt = AV_SAMPLE_FMT_S16;
 
-    if (s->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
-        s->add_bias = 385.0f;
-        s->scale_bias = 1.0 / 32768.0;
-    } else {
-        s->add_bias = 0.0f;
         s->scale_bias = 1.0;
 
         /* allow downmixing to stereo */
@@ -1853,7 +1845,6 @@ static av_cold int dca_decode_init(AVCodecContext * avctx)
                 avctx->request_channels == 2) {
             avctx->channels = avctx->request_channels;
         }
-    }
 
 
     return 0;
diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index 2ed0052977..03a5eeea45 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -3910,14 +3910,7 @@ static void vector_clipf_c(float *dst, const float *src, float min, float max, i
 }
 
 static av_always_inline int float_to_int16_one(const float *src){
-    int_fast32_t tmp = *(const int32_t*)src;
-    if(tmp & 0xf0000){
-        tmp = (0x43c0ffff - tmp)>>31;
-        // is this faster on some gcc/cpu combinations?
-//      if(tmp > 0x43c0ffff) tmp = 0xFFFF;
-//      else                 tmp = 0;
-    }
-    return tmp - 0x8000;
+    return av_clip_int16(lrintf(*src));
 }
 
 void ff_float_to_int16_c(int16_t *dst, const float *src, long len){
diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index baa68bebe7..eabecf0d48 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -435,8 +435,7 @@ typedef struct DSPContext {
      */
     void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
 
-    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
-     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
+    /* convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
     void (*float_to_int16)(int16_t *dst, const float *src, long len);
     void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
 
diff --git a/libavcodec/nellymoserdec.c b/libavcodec/nellymoserdec.c
index 89dde91248..e70d0622da 100644
--- a/libavcodec/nellymoserdec.c
+++ b/libavcodec/nellymoserdec.c
@@ -49,7 +49,6 @@ typedef struct NellyMoserDecodeContext {
     float           state[128];
     AVLFG           random_state;
     GetBitContext   gb;
-    int             add_bias;
     float           scale_bias;
     DSPContext      dsp;
     FFTContext      imdct_ctx;
@@ -65,7 +64,7 @@ static void overlap_and_window(NellyMoserDecodeContext *s, float *state, float *
 
     while (bot < NELLY_BUF_LEN) {
         audio[bot] = a_in [bot]*ff_sine_128[bot]
-                    +state[bot]*ff_sine_128[top] + s->add_bias;
+                    +state[bot]*ff_sine_128[top];
 
         bot++;
         top--;
@@ -136,13 +135,7 @@ static av_cold int decode_init(AVCodecContext * avctx) {
 
     dsputil_init(&s->dsp, avctx);
 
-    if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
-        s->add_bias = 385;
-        s->scale_bias = 1.0/(8*32768);
-    } else {
-        s->add_bias = 0;
         s->scale_bias = 1.0/(1*8);
-    }
 
     /* Generate overlap window */
     if (!ff_sine_128[127])
diff --git a/libavcodec/vorbis_dec.c b/libavcodec/vorbis_dec.c
index cdb485a9c9..69b784c668 100644
--- a/libavcodec/vorbis_dec.c
+++ b/libavcodec/vorbis_dec.c
@@ -153,8 +153,7 @@ typedef struct vorbis_context_s {
     float        *channel_residues;
     float        *channel_floors;
     float        *saved;
-    uint_fast32_t add_bias; // for float->int conversion
-    uint_fast32_t exp_bias;
+    float         scale_bias; // for float->int conversion
 } vorbis_context;
 
 /* Helper functions */
@@ -932,8 +931,8 @@ static int vorbis_parse_id_hdr(vorbis_context *vc)
     vc->saved            =  av_mallocz((vc->blocksize[1] / 4) * vc->audio_channels * sizeof(float));
     vc->previous_window  = 0;
 
-    ff_mdct_init(&vc->mdct[0], bl0, 1, vc->exp_bias ? -(1 << 15) : -1.0);
-    ff_mdct_init(&vc->mdct[1], bl1, 1, vc->exp_bias ? -(1 << 15) : -1.0);
+    ff_mdct_init(&vc->mdct[0], bl0, 1, -vc->scale_bias);
+    ff_mdct_init(&vc->mdct[1], bl1, 1, -vc->scale_bias);
 
     AV_DEBUG(" vorbis version %d \n audio_channels %d \n audio_samplerate %d \n bitrate_max %d \n bitrate_nom %d \n bitrate_min %d \n blk_0 %d blk_1 %d \n ",
             vc->version, vc->audio_channels, vc->audio_samplerate, vc->bitrate_maximum, vc->bitrate_nominal, vc->bitrate_minimum, vc->blocksize[0], vc->blocksize[1]);
@@ -963,13 +962,7 @@ static av_cold int vorbis_decode_init(AVCodecContext *avccontext)
     vc->avccontext = avccontext;
     dsputil_init(&vc->dsp, avccontext);
 
-    if (vc->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
-        vc->add_bias = 385;
-        vc->exp_bias = 0;
-    } else {
-        vc->add_bias = 0;
-        vc->exp_bias = 15 << 23;
-    }
+    vc->scale_bias = 32768.0f;
 
     if (!headers_len) {
         av_log(avccontext, AV_LOG_ERROR, "Extradata missing.\n");
@@ -1453,18 +1446,6 @@ void vorbis_inverse_coupling(float *mag, float *ang, int blocksize)
     }
 }
 
-static void copy_normalize(float *dst, float *src, int len, int exp_bias,
-                           float add_bias)
-{
-    int i;
-    if (exp_bias) {
-        memcpy(dst, src, len * sizeof(float));
-    } else {
-        for (i = 0; i < len; i++)
-            dst[i] = src[i] + add_bias;
-    }
-}
-
 // Decode the audio packet using the functions above
 
 static int vorbis_parse_audio_packet(vorbis_context *vc)
@@ -1484,7 +1465,6 @@ static int vorbis_parse_audio_packet(vorbis_context *vc)
     uint_fast8_t res_chan[255];
     uint_fast8_t res_num = 0;
     int_fast16_t retlen  = 0;
-    float fadd_bias = vc->add_bias;
 
     if (get_bits1(gb)) {
         av_log(vc->avccontext, AV_LOG_ERROR, "Not a Vorbis I audio packet.\n");
@@ -1595,13 +1575,13 @@ static int vorbis_parse_audio_packet(vorbis_context *vc)
         const float *win  = vc->win[blockflag & previous_window];
 
         if (blockflag == previous_window) {
-            vc->dsp.vector_fmul_window(ret, saved, buf, win, fadd_bias, blocksize / 4);
+            vc->dsp.vector_fmul_window(ret, saved, buf, win, 0, blocksize / 4);
         } else if (blockflag > previous_window) {
-            vc->dsp.vector_fmul_window(ret, saved, buf, win, fadd_bias, bs0 / 4);
-            copy_normalize(ret+bs0/2, buf+bs0/4, (bs1-bs0)/4, vc->exp_bias, fadd_bias);
+            vc->dsp.vector_fmul_window(ret, saved, buf, win, 0, bs0 / 4);
+            memcpy(ret+bs0/2, buf+bs0/4, ((bs1-bs0)/4) * sizeof(float));
         } else {
-            copy_normalize(ret, saved, (bs1 - bs0) / 4, vc->exp_bias, fadd_bias);
-            vc->dsp.vector_fmul_window(ret + (bs1 - bs0) / 4, saved + (bs1 - bs0) / 4, buf, win, fadd_bias, bs0 / 4);
+            memcpy(ret, saved, ((bs1 - bs0) / 4) * sizeof(float));
+            vc->dsp.vector_fmul_window(ret + (bs1 - bs0) / 4, saved + (bs1 - bs0) / 4, buf, win, 0, bs0 / 4);
         }
         memcpy(saved, buf + blocksize / 4, blocksize / 4 * sizeof(float));
     }
diff --git a/libavcodec/wmadec.c b/libavcodec/wmadec.c
index b720ea5098..03d7bd19fc 100644
--- a/libavcodec/wmadec.c
+++ b/libavcodec/wmadec.c
@@ -768,9 +768,8 @@ next:
 /* decode a frame of frame_len samples */
 static int wma_decode_frame(WMACodecContext *s, int16_t *samples)
 {
-    int ret, i, n, ch, incr;
-    int16_t *ptr;
-    float *iptr;
+    int ret, n, ch, incr;
+    const float *output[MAX_CHANNELS];
 
 #ifdef TRACE
     tprintf(s->avctx, "***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len);
@@ -790,21 +789,6 @@ static int wma_decode_frame(WMACodecContext *s, int16_t *samples)
     /* convert frame to integer */
     n = s->frame_len;
     incr = s->nb_channels;
-    if (s->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
-        for(ch = 0; ch < s->nb_channels; ch++) {
-            ptr = samples + ch;
-            iptr = s->frame_out[ch];
-
-            for(i=0;i<n;i++) {
-                *ptr = av_clip_int16(lrintf(*iptr++));
-                ptr += incr;
-            }
-            /* prepare for next block */
-            memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len],
-                    s->frame_len * sizeof(float));
-        }
-    } else {
-        const float *output[MAX_CHANNELS];
         for (ch = 0; ch < MAX_CHANNELS; ch++)
             output[ch] = s->frame_out[ch];
         s->dsp.float_to_int16_interleave(samples, output, n, incr);
@@ -812,7 +796,6 @@ static int wma_decode_frame(WMACodecContext *s, int16_t *samples)
             /* prepare for next block */
             memmove(&s->frame_out[ch][0], &s->frame_out[ch][n], n * sizeof(float));
         }
-    }
 
 #ifdef TRACE
     dump_shorts(s, "samples", samples, n * s->nb_channels);