Merge remote-tracking branch 'qatar/master'

* qatar/master: (22 commits)
  g723.1: do not pass large structs by value
  g723.1: do not bounce intermediate values via memory
  g723.1: declare a variable in the block it is used
  g723.1: avoid saving/restoring excitation
  g723.1: avoid unnecessary memcpy() in residual_interp()
  g723.1: make postfilter write directly to output buffer
  g723.1: drop unnecessary variable buf_ptr in formant_postfilter()
  g723.1: make scale_vector() output to a separate buffer
  g723.1: make autocorr_max() work on an arbitrary buffer
  g723.1: do not needlessly use int64_t
  g723.1: use saturating addition functions
  g723.1: optimise scale_vector()
  g723.1: remove useless uses of MUL64()
  g723.1: remove unnecessary argument 'shift' from dot_product()
  g723.1: deobfuscate "(x << 4) - x" to "15 * x"
  celp: optimise ff_celp_lp_synthesis_filter()
  libavutil: add saturating addition functions
  cllc: Implement ARGB support
  cllc: Add support for QRGB
  cllc: Rename some funcs to represent what they actually do
  ...

Conflicts:
	LICENSE
	libavcodec/g723_1.c
	libavcodec/x86/Makefile

Merged-by: Michael Niedermayer <michaelni@gmx.at>

9 files changed, 203 insertions, 154 deletions

diff --git a/LICENSE b/LICENSE
index 1607742b0c..7dff535a72 100644
--- a/LICENSE
+++ b/LICENSE
@@ -1,5 +1,4 @@
 FFmpeg:
--------
 
 Most files in FFmpeg are under the GNU Lesser General Public License version 2.1
 or later (LGPL v2.1+). Read the file COPYING.LGPLv2.1 for details. Some other
@@ -51,18 +50,29 @@ for you. Read the file COPYING.LGPLv3 or, if you have enabled GPL parts,
 COPYING.GPLv3 to learn the exact legal terms that apply in this case.
 
 
-external libraries:
--------------------
+external libraries
+==================
 
-Some external libraries, e.g. libx264, are under GPL and can be used in
-conjunction with FFmpeg. They require --enable-gpl to be passed to configure
-as well.
+FFmpeg can be combined with a number of external libraries, which sometimes
+affect the licensing of binaries resulting from the combination.
 
-The OpenCORE external libraries are under the Apache License 2.0. That license
-is incompatible with the LGPL v2.1 and the GPL v2, but not with version 3 of
-those licenses. So to combine the OpenCORE libraries with FFmpeg, the license
-version needs to be upgraded by passing --enable-version3 to configure.
+compatible libraries
+--------------------
 
-The nonfree external libraries libfaac and libaacplus can be hooked up in FFmpeg.
-You need to pass --enable-nonfree to configure to enable it. Employ this option
-with care as FFmpeg then becomes nonfree and unredistributable.
+The libcdio, libx264, libxavs and libxvid libraries are under GPL. When
+combining them with FFmpeg, FFmpeg needs to be licensed as GPL as well by
+passing --enable-gpl to configure.
+
+The OpenCORE and VisualOn libraries are under the Apache License 2.0. That
+license is incompatible with the LGPL v2.1 and the GPL v2, but not with
+version 3 of those licenses. So to combine these libraries with FFmpeg, the
+license version needs to be upgraded by passing --enable-version3 to configure.
+
+incompatible libraries
+----------------------
+
+The Fraunhofer AAC library, FAAC and aacplus are under licenses incompatible
+with all (L)GPL versions. Thus, unfortunately, since both licenses cannot be
+satisfied simultaneously, binaries resulting from the combination of FFmpeg
+with these libraries are nonfree und unredistributable. If you wish to enable
+any of these libraries nonetheless, pass --enable-nonfree to configure.
diff --git a/libavcodec/celp_filters.c b/libavcodec/celp_filters.c
index 8047a78452..cf2198d325 100644
--- a/libavcodec/celp_filters.c
+++ b/libavcodec/celp_filters.c
@@ -63,17 +63,16 @@ int ff_celp_lp_synthesis_filter(int16_t *out, const int16_t *filter_coeffs,
     int i,n;
 
     for (n = 0; n < buffer_length; n++) {
-        int sum = rounder;
+        int sum = -rounder, sum1;
         for (i = 1; i <= filter_length; i++)
-            sum -= filter_coeffs[i-1] * out[n-i];
+            sum += filter_coeffs[i-1] * out[n-i];
 
-        sum = ((sum >> 12) + in[n]) >> shift;
+        sum1 = ((-sum >> 12) + in[n]) >> shift;
+        sum  = av_clip_int16(sum1);
+
+        if (stop_on_overflow && sum != sum1)
+            return 1;
 
-        if (sum + 0x8000 > 0xFFFFU) {
-            if (stop_on_overflow)
-                return 1;
-            sum = (sum >> 31) ^ 32767;
-        }
         out[n] = sum;
     }
 
diff --git a/libavcodec/g723_1.c b/libavcodec/g723_1.c
index c4016f8112..c1e7d2bb49 100644
--- a/libavcodec/g723_1.c
+++ b/libavcodec/g723_1.c
@@ -65,7 +65,7 @@ typedef struct g723_1_context {
     int pf_gain;                 ///< formant postfilter
                                  ///< gain scaling unit memory
     int postfilter;
-    int16_t audio[FRAME_LEN + LPC_ORDER];
+    int16_t audio[FRAME_LEN + LPC_ORDER + PITCH_MAX];
     int16_t prev_data[HALF_FRAME_LEN];
     int16_t prev_weight_sig[PITCH_MAX];
 
@@ -245,32 +245,27 @@ static int normalize_bits(int num, int width)
 
 #define normalize_bits_int16(num) normalize_bits(num, 15)
 #define normalize_bits_int32(num) normalize_bits(num, 31)
-#define dot_product(a,b,c,d) (ff_dot_product(a,b,c)<<(d))
 
 /**
  * Scale vector contents based on the largest of their absolutes.
  */
-static int scale_vector(int16_t *vector, int length)
+static int scale_vector(int16_t *dst, const int16_t *vector, int length)
 {
-    int bits, scale, max = 0;
+    int bits, max = 0;
     int i;
 
-    const int16_t shift_table[16] = {
-        0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080,
-        0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000, 0x7fff
-    };
-
     for (i = 0; i < length; i++)
-        max = FFMAX(max, FFABS(vector[i]));
+        max |= FFABS(vector[i]);
 
     max   = FFMIN(max, 0x7FFF);
     bits  = normalize_bits(max, 15);
-    scale = shift_table[bits];
 
-    for (i = 0; i < length; i++) {
-        av_assert2(av_clipl_int32(vector[i] * (int64_t)scale << 1) == vector[i] * (int64_t)scale << 1);
-        vector[i] = (vector[i] * scale) >> 3;
-    }
+    if (bits == 15)
+        for (i = 0; i < length; i++)
+            dst[i] = vector[i] * 0x7fff >> 3;
+    else
+        for (i = 0; i < length; i++)
+            dst[i] = vector[i] << bits >> 3;
 
     return bits - 3;
 }
@@ -369,11 +364,11 @@ static void lsp2lpc(int16_t *lpc)
     for (j = 0; j < LPC_ORDER; j++) {
         int index     = lpc[j] >> 7;
         int offset    = lpc[j] & 0x7f;
-        int64_t temp1 = cos_tab[index] << 16;
+        int temp1     = cos_tab[index] << 16;
         int temp2     = (cos_tab[index + 1] - cos_tab[index]) *
                           ((offset << 8) + 0x80) << 1;
 
-        lpc[j] = -(av_clipl_int32(((temp1 + temp2) << 1) + (1 << 15)) >> 16);
+        lpc[j] = -(av_sat_dadd32(1 << 15, temp1 + temp2) >> 16);
     }
 
     /*
@@ -473,7 +468,7 @@ static void gen_dirac_train(int16_t *buf, int pitch_lag)
  * @param pitch_lag closed loop pitch lag
  * @param index     current subframe index
  */
-static void gen_fcb_excitation(int16_t *vector, G723_1_Subframe subfrm,
+static void gen_fcb_excitation(int16_t *vector, G723_1_Subframe *subfrm,
                                enum Rate cur_rate, int pitch_lag, int index)
 {
     int temp, i, j;
@@ -481,34 +476,34 @@ static void gen_fcb_excitation(int16_t *vector, G723_1_Subframe subfrm,
     memset(vector, 0, SUBFRAME_LEN * sizeof(*vector));
 
     if (cur_rate == RATE_6300) {
-        if (subfrm.pulse_pos >= max_pos[index])
+        if (subfrm->pulse_pos >= max_pos[index])
             return;
 
         /* Decode amplitudes and positions */
         j = PULSE_MAX - pulses[index];
-        temp = subfrm.pulse_pos;
+        temp = subfrm->pulse_pos;
         for (i = 0; i < SUBFRAME_LEN / GRID_SIZE; i++) {
             temp -= combinatorial_table[j][i];
             if (temp >= 0)
                 continue;
             temp += combinatorial_table[j++][i];
-            if (subfrm.pulse_sign & (1 << (PULSE_MAX - j))) {
-                vector[subfrm.grid_index + GRID_SIZE * i] =
-                                        -fixed_cb_gain[subfrm.amp_index];
+            if (subfrm->pulse_sign & (1 << (PULSE_MAX - j))) {
+                vector[subfrm->grid_index + GRID_SIZE * i] =
+                                        -fixed_cb_gain[subfrm->amp_index];
             } else {
-                vector[subfrm.grid_index + GRID_SIZE * i] =
-                                         fixed_cb_gain[subfrm.amp_index];
+                vector[subfrm->grid_index + GRID_SIZE * i] =
+                                         fixed_cb_gain[subfrm->amp_index];
             }
             if (j == PULSE_MAX)
                 break;
         }
-        if (subfrm.dirac_train == 1)
+        if (subfrm->dirac_train == 1)
             gen_dirac_train(vector, pitch_lag);
     } else { /* 5300 bps */
-        int cb_gain  = fixed_cb_gain[subfrm.amp_index];
-        int cb_shift = subfrm.grid_index;
-        int cb_sign  = subfrm.pulse_sign;
-        int cb_pos   = subfrm.pulse_pos;
+        int cb_gain  = fixed_cb_gain[subfrm->amp_index];
+        int cb_shift = subfrm->grid_index;
+        int cb_sign  = subfrm->pulse_sign;
+        int cb_pos   = subfrm->pulse_pos;
         int offset, beta, lag;
 
         for (i = 0; i < 8; i += 2) {
@@ -519,9 +514,9 @@ static void gen_fcb_excitation(int16_t *vector, G723_1_Subframe subfrm,
         }
 
         /* Enhance harmonic components */
-        lag  = pitch_contrib[subfrm.ad_cb_gain << 1] + pitch_lag +
-               subfrm.ad_cb_lag - 1;
-        beta = pitch_contrib[(subfrm.ad_cb_gain << 1) + 1];
+        lag  = pitch_contrib[subfrm->ad_cb_gain << 1] + pitch_lag +
+               subfrm->ad_cb_lag - 1;
+        beta = pitch_contrib[(subfrm->ad_cb_gain << 1) + 1];
 
         if (lag < SUBFRAME_LEN - 2) {
             for (i = lag; i < SUBFRAME_LEN; i++)
@@ -546,19 +541,25 @@ static void get_residual(int16_t *residual, int16_t *prev_excitation, int lag)
         residual[i] = prev_excitation[offset + (i - 2) % lag];
 }
 
+static int dot_product(const int16_t *a, const int16_t *b, int length)
+{
+    int sum = ff_dot_product(a,b,length);
+    return av_sat_add32(sum, sum);
+}
+
 /**
  * Generate adaptive codebook excitation.
  */
 static void gen_acb_excitation(int16_t *vector, int16_t *prev_excitation,
-                               int pitch_lag, G723_1_Subframe subfrm,
+                               int pitch_lag, G723_1_Subframe *subfrm,
                                enum Rate cur_rate)
 {
     int16_t residual[SUBFRAME_LEN + PITCH_ORDER - 1];
     const int16_t *cb_ptr;
-    int lag = pitch_lag + subfrm.ad_cb_lag - 1;
+    int lag = pitch_lag + subfrm->ad_cb_lag - 1;
 
     int i;
-    int64_t sum;
+    int sum;
 
     get_residual(residual, prev_excitation, lag);
 
@@ -569,28 +570,27 @@ static void gen_acb_excitation(int16_t *vector, int16_t *prev_excitation,
         cb_ptr = adaptive_cb_gain170;
 
     /* Calculate adaptive vector */
-    cb_ptr += subfrm.ad_cb_gain * 20;
+    cb_ptr += subfrm->ad_cb_gain * 20;
     for (i = 0; i < SUBFRAME_LEN; i++) {
         sum = ff_dot_product(residual + i, cb_ptr, PITCH_ORDER);
-        vector[i] = av_clipl_int32((sum << 2) + (1 << 15)) >> 16;
+        vector[i] = av_sat_dadd32(1 << 15, av_sat_add32(sum, sum)) >> 16;
     }
 }
 
 /**
  * Estimate maximum auto-correlation around pitch lag.
  *
- * @param p         the context
+ * @param buf       buffer with offset applied
  * @param offset    offset of the excitation vector
  * @param ccr_max   pointer to the maximum auto-correlation
  * @param pitch_lag decoded pitch lag
  * @param length    length of autocorrelation
  * @param dir       forward lag(1) / backward lag(-1)
  */
-static int autocorr_max(G723_1_Context *p, int offset, int *ccr_max,
+static int autocorr_max(const int16_t *buf, int offset, int *ccr_max,
                         int pitch_lag, int length, int dir)
 {
     int limit, ccr, lag = 0;
-    int16_t *buf = p->excitation + offset;
     int i;
 
     pitch_lag = FFMIN(PITCH_MAX - 3, pitch_lag);
@@ -600,7 +600,7 @@ static int autocorr_max(G723_1_Context *p, int offset, int *ccr_max,
         limit = pitch_lag + 3;
 
     for (i = pitch_lag - 3; i <= limit; i++) {
-        ccr = ff_dot_product(buf, buf + dir * i, length)<<1;
+        ccr = dot_product(buf, buf + dir * i, length);
 
         if (ccr > *ccr_max) {
             *ccr_max = ccr;
@@ -624,7 +624,7 @@ static void comp_ppf_gains(int lag, PPFParam *ppf, enum Rate cur_rate,
                            int tgt_eng, int ccr, int res_eng)
 {
     int pf_residual;     /* square of postfiltered residual */
-    int64_t temp1, temp2;
+    int temp1, temp2;
 
     ppf->index = lag;
 
@@ -641,7 +641,7 @@ static void comp_ppf_gains(int lag, PPFParam *ppf, enum Rate cur_rate,
         /* pf_res^2 = tgt_eng + 2*ccr*gain + res_eng*gain^2 */
         temp1       = (tgt_eng << 15) + (ccr * ppf->opt_gain << 1);
         temp2       = (ppf->opt_gain * ppf->opt_gain >> 15) * res_eng;
-        pf_residual = av_clipl_int32(temp1 + temp2 + (1 << 15)) >> 16;
+        pf_residual = av_sat_add32(temp1, temp2 + (1 << 15)) >> 16;
 
         if (tgt_eng >= pf_residual << 1) {
             temp1 = 0x7fff;
@@ -674,7 +674,7 @@ static void comp_ppf_coeff(G723_1_Context *p, int offset, int pitch_lag,
 
     int16_t scale;
     int i;
-    int64_t temp1, temp2;
+    int temp1, temp2;
 
     /*
      * 0 - target energy
@@ -684,10 +684,10 @@ static void comp_ppf_coeff(G723_1_Context *p, int offset, int pitch_lag,
      * 4 - backward residual energy
      */
     int energy[5] = {0, 0, 0, 0, 0};
-    int16_t *buf  = p->excitation + offset;
-    int fwd_lag   = autocorr_max(p, offset, &energy[1], pitch_lag,
+    int16_t *buf  = p->audio + LPC_ORDER + offset;
+    int fwd_lag   = autocorr_max(buf, offset, &energy[1], pitch_lag,
                                  SUBFRAME_LEN, 1);
-    int back_lag  = autocorr_max(p, offset, &energy[3], pitch_lag,
+    int back_lag  = autocorr_max(buf, offset, &energy[3], pitch_lag,
                                  SUBFRAME_LEN, -1);
 
     ppf->index    = 0;
@@ -699,17 +699,15 @@ static void comp_ppf_coeff(G723_1_Context *p, int offset, int pitch_lag,
         return;
 
     /* Compute target energy */
-    energy[0] = ff_dot_product(buf, buf, SUBFRAME_LEN)<<1;
+    energy[0] = dot_product(buf, buf, SUBFRAME_LEN);
 
     /* Compute forward residual energy */
     if (fwd_lag)
-        energy[2] = ff_dot_product(buf + fwd_lag, buf + fwd_lag,
-                                   SUBFRAME_LEN)<<1;
+        energy[2] = dot_product(buf + fwd_lag, buf + fwd_lag, SUBFRAME_LEN);
 
     /* Compute backward residual energy */
     if (back_lag)
-        energy[4] = ff_dot_product(buf - back_lag, buf - back_lag,
-                                   SUBFRAME_LEN)<<1;
+        energy[4] = dot_product(buf - back_lag, buf - back_lag, SUBFRAME_LEN);
 
     /* Normalize and shorten */
     temp1 = 0;
@@ -758,28 +756,28 @@ static int comp_interp_index(G723_1_Context *p, int pitch_lag,
                              int *exc_eng, int *scale)
 {
     int offset = PITCH_MAX + 2 * SUBFRAME_LEN;
-    int16_t *buf = p->excitation + offset;
+    int16_t *buf = p->audio + LPC_ORDER;
 
     int index, ccr, tgt_eng, best_eng, temp;
 
-    *scale = scale_vector(p->excitation, FRAME_LEN + PITCH_MAX);
+    *scale = scale_vector(buf, p->excitation, FRAME_LEN + PITCH_MAX);
+    buf   += offset;
 
     /* Compute maximum backward cross-correlation */
     ccr   = 0;
-    index = autocorr_max(p, offset, &ccr, pitch_lag, SUBFRAME_LEN * 2, -1);
-    ccr   = av_clipl_int32((int64_t)ccr + (1 << 15)) >> 16;
+    index = autocorr_max(buf, offset, &ccr, pitch_lag, SUBFRAME_LEN * 2, -1);
+    ccr   = av_sat_add32(ccr, 1 << 15) >> 16;
 
     /* Compute target energy */
-    tgt_eng  = ff_dot_product(buf, buf, SUBFRAME_LEN * 2)<<1;
-    *exc_eng = av_clipl_int32(tgt_eng + (1 << 15)) >> 16;
+    tgt_eng  = dot_product(buf, buf, SUBFRAME_LEN * 2);
+    *exc_eng = av_sat_add32(tgt_eng, 1 << 15) >> 16;
 
     if (ccr <= 0)
         return 0;
 
     /* Compute best energy */
-    best_eng = ff_dot_product(buf - index, buf - index,
-                              SUBFRAME_LEN * 2)<<1;
-    best_eng = av_clipl_int32((int64_t)best_eng + (1 << 15)) >> 16;
+    best_eng = dot_product(buf - index, buf - index, SUBFRAME_LEN * 2);
+    best_eng = av_sat_add32(best_eng, 1 << 15) >> 16;
 
     temp = best_eng * *exc_eng >> 3;
 
@@ -806,10 +804,9 @@ static void residual_interp(int16_t *buf, int16_t *out, int lag,
         int16_t *vector_ptr = buf + PITCH_MAX;
         /* Attenuate */
         for (i = 0; i < lag; i++)
-            vector_ptr[i - lag] = vector_ptr[i - lag] * 3 >> 2;
-        av_memcpy_backptr((uint8_t*)vector_ptr, lag * sizeof(*vector_ptr),
-                          FRAME_LEN * sizeof(*vector_ptr));
-        memcpy(out, vector_ptr, FRAME_LEN * sizeof(*vector_ptr));
+            out[i] = vector_ptr[i - lag] * 3 >> 2;
+        av_memcpy_backptr((uint8_t*)(out + lag), lag * sizeof(*out),
+                          (FRAME_LEN - lag) * sizeof(*out));
     } else {  /* Unvoiced */
         for (i = 0; i < FRAME_LEN; i++) {
             *rseed = *rseed * 521 + 259;
@@ -861,9 +858,9 @@ static void gain_scale(G723_1_Context *p, int16_t * buf, int energy)
     num   = energy;
     denom = 0;
     for (i = 0; i < SUBFRAME_LEN; i++) {
-        int64_t temp = buf[i] >> 2;
-        temp  = av_clipl_int32(MUL64(temp, temp) << 1);
-        denom = av_clipl_int32(denom + temp);
+        int temp = buf[i] >> 2;
+        temp *= temp;
+        denom = av_sat_dadd32(denom, temp);
     }
 
     if (num && denom) {
@@ -882,7 +879,7 @@ static void gain_scale(G723_1_Context *p, int16_t * buf, int energy)
     }
 
     for (i = 0; i < SUBFRAME_LEN; i++) {
-        p->pf_gain = ((p->pf_gain << 4) - p->pf_gain + gain + (1 << 3)) >> 4;
+        p->pf_gain = (15 * p->pf_gain + gain + (1 << 3)) >> 4;
         buf[i]     = av_clip_int16((buf[i] * (p->pf_gain + (p->pf_gain >> 4)) +
                                    (1 << 10)) >> 11);
     }
@@ -893,11 +890,13 @@ static void gain_scale(G723_1_Context *p, int16_t * buf, int energy)
  *
  * @param p   the context
  * @param lpc quantized lpc coefficients
- * @param buf output buffer
+ * @param buf input buffer
+ * @param dst output buffer
  */
-static void formant_postfilter(G723_1_Context *p, int16_t *lpc, int16_t *buf)
+static void formant_postfilter(G723_1_Context *p, int16_t *lpc,
+                               int16_t *buf, int16_t *dst)
 {
-    int16_t filter_coef[2][LPC_ORDER], *buf_ptr;
+    int16_t filter_coef[2][LPC_ORDER];
     int filter_signal[LPC_ORDER + FRAME_LEN], *signal_ptr;
     int i, j, k;
 
@@ -919,23 +918,19 @@ static void formant_postfilter(G723_1_Context *p, int16_t *lpc, int16_t *buf)
     memcpy(p->fir_mem, buf + FRAME_LEN, LPC_ORDER * sizeof(int16_t));
     memcpy(p->iir_mem, filter_signal + FRAME_LEN, LPC_ORDER * sizeof(int));
 
-    buf_ptr    = buf + LPC_ORDER;
+    buf += LPC_ORDER;
     signal_ptr = filter_signal + LPC_ORDER;
     for (i = 0; i < SUBFRAMES; i++) {
-        int16_t temp_vector[SUBFRAME_LEN];
         int temp;
         int auto_corr[2];
         int scale, energy;
 
         /* Normalize */
-        memcpy(temp_vector, buf_ptr, SUBFRAME_LEN * sizeof(*temp_vector));
-        scale = scale_vector(temp_vector, SUBFRAME_LEN);
+        scale = scale_vector(dst, buf, SUBFRAME_LEN);
 
         /* Compute auto correlation coefficients */
-        auto_corr[0] = ff_dot_product(temp_vector, temp_vector + 1,
-                                      SUBFRAME_LEN - 1)<<1;
-        auto_corr[1] = ff_dot_product(temp_vector, temp_vector,
-                                      SUBFRAME_LEN)<<1;
+        auto_corr[0] = dot_product(dst, dst + 1, SUBFRAME_LEN - 1);
+        auto_corr[1] = dot_product(dst, dst,     SUBFRAME_LEN);
 
         /* Compute reflection coefficient */
         temp = auto_corr[1] >> 16;
@@ -947,9 +942,8 @@ static void formant_postfilter(G723_1_Context *p, int16_t *lpc, int16_t *buf)
 
         /* Compensation filter */
         for (j = 0; j < SUBFRAME_LEN; j++) {
-            buf_ptr[j] = av_clipl_int32((int64_t)signal_ptr[j] +
-                                        ((signal_ptr[j - 1] >> 16) *
-                                         temp << 1)) >> 16;
+            dst[j] = av_sat_dadd32(signal_ptr[j],
+                                   (signal_ptr[j - 1] >> 16) * temp) >> 16;
         }
 
         /* Compute normalized signal energy */
@@ -959,10 +953,11 @@ static void formant_postfilter(G723_1_Context *p, int16_t *lpc, int16_t *buf)
         } else
             energy = auto_corr[1] >> temp;
 
-        gain_scale(p, buf_ptr, energy);
+        gain_scale(p, dst, energy);
 
-        buf_ptr    += SUBFRAME_LEN;
+        buf        += SUBFRAME_LEN;
         signal_ptr += SUBFRAME_LEN;
+        dst        += SUBFRAME_LEN;
     }
 }
 
@@ -978,9 +973,9 @@ static int g723_1_decode_frame(AVCodecContext *avctx, void *data,
     int16_t cur_lsp[LPC_ORDER];
     int16_t lpc[SUBFRAMES * LPC_ORDER];
     int16_t acb_vector[SUBFRAME_LEN];
-    int16_t *vector_ptr;
     int16_t *out;
     int bad_frame = 0, i, j, ret;
+    int16_t *audio = p->audio;
 
     if (buf_size < frame_size[dec_mode]) {
         if (buf_size)
@@ -1022,48 +1017,38 @@ static int g723_1_decode_frame(AVCodecContext *avctx, void *data,
         /* Generate the excitation for the frame */
         memcpy(p->excitation, p->prev_excitation,
                PITCH_MAX * sizeof(*p->excitation));
-        vector_ptr = p->excitation + PITCH_MAX;
         if (!p->erased_frames) {
+            int16_t *vector_ptr = p->excitation + PITCH_MAX;
+
             /* Update interpolation gain memory */
             p->interp_gain = fixed_cb_gain[(p->subframe[2].amp_index +
                                             p->subframe[3].amp_index) >> 1];
             for (i = 0; i < SUBFRAMES; i++) {
-                gen_fcb_excitation(vector_ptr, p->subframe[i], p->cur_rate,
+                gen_fcb_excitation(vector_ptr, &p->subframe[i], p->cur_rate,
                                    p->pitch_lag[i >> 1], i);
                 gen_acb_excitation(acb_vector, &p->excitation[SUBFRAME_LEN * i],
-                                   p->pitch_lag[i >> 1], p->subframe[i],
+                                   p->pitch_lag[i >> 1], &p->subframe[i],
                                    p->cur_rate);
                 /* Get the total excitation */
                 for (j = 0; j < SUBFRAME_LEN; j++) {
-                    vector_ptr[j] = av_clip_int16(vector_ptr[j] << 1);
-                    vector_ptr[j] = av_clip_int16(vector_ptr[j] +
-                                                  acb_vector[j]);
+                    int v = av_clip_int16(vector_ptr[j] << 1);
+                    vector_ptr[j] = av_clip_int16(v + acb_vector[j]);
                 }
                 vector_ptr += SUBFRAME_LEN;
             }
 
             vector_ptr = p->excitation + PITCH_MAX;
 
-            /* Save the excitation */
-            memcpy(p->audio + LPC_ORDER, vector_ptr, FRAME_LEN * sizeof(*p->audio));
-
             p->interp_index = comp_interp_index(p, p->pitch_lag[1],
                                                 &p->sid_gain, &p->cur_gain);
 
+            /* Peform pitch postfiltering */
             if (p->postfilter) {
                 i = PITCH_MAX;
                 for (j = 0; j < SUBFRAMES; i += SUBFRAME_LEN, j++)
                     comp_ppf_coeff(p, i, p->pitch_lag[j >> 1],
                                    ppf + j, p->cur_rate);
-            }
-
-            /* Restore the original excitation */
-            memcpy(p->excitation, p->prev_excitation,
-                   PITCH_MAX * sizeof(*p->excitation));
-            memcpy(vector_ptr, p->audio + LPC_ORDER, FRAME_LEN * sizeof(*vector_ptr));
 
-            /* Peform pitch postfiltering */
-            if (p->postfilter)
                 for (i = 0, j = 0; j < SUBFRAMES; i += SUBFRAME_LEN, j++)
                     ff_acelp_weighted_vector_sum(p->audio + LPC_ORDER + i,
                                                  vector_ptr + i,
@@ -1071,24 +1056,35 @@ static int g723_1_decode_frame(AVCodecContext *avctx, void *data,
                                                  ppf[j].sc_gain,
                                                  ppf[j].opt_gain,
                                                  1 << 14, 15, SUBFRAME_LEN);
+            } else {
+                audio = vector_ptr - LPC_ORDER;
+            }
 
+            /* Save the excitation for the next frame */
+            memcpy(p->prev_excitation, p->excitation + FRAME_LEN,
+                   PITCH_MAX * sizeof(*p->excitation));
         } else {
             p->interp_gain = (p->interp_gain * 3 + 2) >> 2;
             if (p->erased_frames == 3) {
                 /* Mute output */
                 memset(p->excitation, 0,
                        (FRAME_LEN + PITCH_MAX) * sizeof(*p->excitation));
+                memset(p->prev_excitation, 0,
+                       PITCH_MAX * sizeof(*p->excitation));
                 memset(p->frame.data[0], 0,
                        (FRAME_LEN + LPC_ORDER) * sizeof(int16_t));
             } else {
+                int16_t *buf = p->audio + LPC_ORDER;
+
                 /* Regenerate frame */
-                residual_interp(p->excitation, p->audio + LPC_ORDER, p->interp_index,
+                residual_interp(p->excitation, buf, p->interp_index,
                                 p->interp_gain, &p->random_seed);
+
+                /* Save the excitation for the next frame */
+                memcpy(p->prev_excitation, buf + (FRAME_LEN - PITCH_MAX),
+                       PITCH_MAX * sizeof(*p->excitation));
             }
         }
-        /* Save the excitation for the next frame */
-        memcpy(p->prev_excitation, p->excitation + FRAME_LEN,
-               PITCH_MAX * sizeof(*p->excitation));
     } else {
         memset(out, 0, FRAME_LEN * 2);
         av_log(avctx, AV_LOG_WARNING,
@@ -1104,13 +1100,12 @@ static int g723_1_decode_frame(AVCodecContext *avctx, void *data,
     memcpy(p->audio, p->synth_mem, LPC_ORDER * sizeof(*p->audio));
     for (i = LPC_ORDER, j = 0; j < SUBFRAMES; i += SUBFRAME_LEN, j++)
         ff_celp_lp_synthesis_filter(p->audio + i, &lpc[j * LPC_ORDER],
-                                    p->audio + i, SUBFRAME_LEN, LPC_ORDER,
+                                    audio + i, SUBFRAME_LEN, LPC_ORDER,
                                     0, 1, 1 << 12);
     memcpy(p->synth_mem, p->audio + FRAME_LEN, LPC_ORDER * sizeof(*p->audio));
 
     if (p->postfilter) {
-        formant_postfilter(p, lpc, p->audio);
-        memcpy(p->frame.data[0], p->audio + LPC_ORDER, FRAME_LEN * 2);
+        formant_postfilter(p, lpc, p->audio, out);
     } else { // if output is not postfiltered it should be scaled by 2
         for (i = 0; i < FRAME_LEN; i++)
             out[i] = av_clip_int16(p->audio[LPC_ORDER + i] << 1);
@@ -1214,14 +1209,14 @@ static void comp_autocorr(int16_t *buf, int16_t *autocorr)
     int16_t vector[LPC_FRAME];
 
     memcpy(vector, buf, LPC_FRAME * sizeof(int16_t));
-    scale_vector(vector, LPC_FRAME);
+    scale_vector(vector, vector, LPC_FRAME);
 
     /* Apply the Hamming window */
     for (i = 0; i < LPC_FRAME; i++)
         vector[i] = (vector[i] * hamming_window[i] + (1 << 14)) >> 15;
 
     /* Compute the first autocorrelation coefficient */
-    temp = dot_product(vector, vector, LPC_FRAME, 0);
+    temp = ff_dot_product(vector, vector, LPC_FRAME);
 
     /* Apply a white noise correlation factor of (1025/1024) */
     temp += temp >> 10;
@@ -1236,7 +1231,7 @@ static void comp_autocorr(int16_t *buf, int16_t *autocorr)
         memset(autocorr + 1, 0, LPC_ORDER * sizeof(int16_t));
     } else {
         for (i = 1; i <= LPC_ORDER; i++) {
-           temp = dot_product(vector, vector + i, LPC_FRAME - i, 0);
+           temp = ff_dot_product(vector, vector + i, LPC_FRAME - i);
            temp = MULL2((temp << scale), binomial_window[i - 1]);
            autocorr[i] = av_clipl_int32((int64_t)temp + (1 << 15)) >> 16;
         }
@@ -1416,8 +1411,8 @@ static void lpc2lsp(int16_t *lpc, int16_t *prev_lsp, int16_t *lsp)
             temp[j] = (weight[j + (offset)] * lsp_band##num[i][j] +\
                       (1 << 14)) >> 15;\
         }\
-        error =  dot_product(lsp + (offset), temp, size, 1) << 1;\
-        error -= dot_product(lsp_band##num[i], temp, size, 1);\
+        error =  dot_product(lsp + (offset), temp, size) << 1;\
+        error -= dot_product(lsp_band##num[i], temp, size);\
         if (error > max) {\
             max = error;\
             lsp_index[num] = i;\
@@ -1522,7 +1517,7 @@ static int estimate_pitch(int16_t *buf, int start)
 
     int i;
 
-    orig_eng = dot_product(buf + offset, buf + offset, HALF_FRAME_LEN, 0);
+    orig_eng = ff_dot_product(buf + offset, buf + offset, HALF_FRAME_LEN);
 
     for (i = PITCH_MIN; i <= PITCH_MAX - 3; i++) {
         offset--;
@@ -1530,7 +1525,7 @@ static int estimate_pitch(int16_t *buf, int start)
         /* Update energy and compute correlation */
         orig_eng += buf[offset] * buf[offset] -
                     buf[offset + HALF_FRAME_LEN] * buf[offset + HALF_FRAME_LEN];
-        ccr      =  dot_product(buf + start, buf + offset, HALF_FRAME_LEN, 0);
+        ccr      =  ff_dot_product(buf + start, buf + offset, HALF_FRAME_LEN);
         if (ccr <= 0)
             continue;
 
@@ -1591,13 +1586,13 @@ static void comp_harmonic_coeff(int16_t *buf, int16_t pitch_lag, HFParam *hf)
 
     for (i = 0, j = pitch_lag - 3; j <= pitch_lag + 3; i++, j++) {
         /* Compute residual energy */
-        energy[i << 1] = dot_product(buf - j, buf - j, SUBFRAME_LEN, 0);
+        energy[i << 1] = ff_dot_product(buf - j, buf - j, SUBFRAME_LEN);
         /* Compute correlation */
-        energy[(i << 1) + 1] = dot_product(buf, buf - j, SUBFRAME_LEN, 0);
+        energy[(i << 1) + 1] = ff_dot_product(buf, buf - j, SUBFRAME_LEN);
     }
 
     /* Compute target energy */
-    energy[14] = dot_product(buf, buf, SUBFRAME_LEN, 0);
+    energy[14] = ff_dot_product(buf, buf, SUBFRAME_LEN);
 
     /* Normalize */
     max = 0;
@@ -1778,19 +1773,19 @@ static void acb_search(G723_1_Context *p, int16_t *residual,
 
         /* Compute crosscorrelation with the signal */
         for (j = 0; j < PITCH_ORDER; j++) {
-            temp = dot_product(buf, flt_buf[j], SUBFRAME_LEN, 0);
+            temp = ff_dot_product(buf, flt_buf[j], SUBFRAME_LEN);
             ccr_buf[count++] = av_clipl_int32(temp << 1);
         }
 
         /* Compute energies */
         for (j = 0; j < PITCH_ORDER; j++) {
             ccr_buf[count++] = dot_product(flt_buf[j], flt_buf[j],
-                                           SUBFRAME_LEN, 1);
+                                           SUBFRAME_LEN);
         }
 
         for (j = 1; j < PITCH_ORDER; j++) {
             for (k = 0; k < j; k++) {
-                temp = dot_product(flt_buf[j], flt_buf[k], SUBFRAME_LEN, 0);
+                temp = ff_dot_product(flt_buf[j], flt_buf[k], SUBFRAME_LEN);
                 ccr_buf[count++] = av_clipl_int32(temp<<2);
             }
         }
@@ -1893,20 +1888,20 @@ static void get_fcb_param(FCBParam *optim, int16_t *impulse_resp,
         temp_corr[i] = impulse_r[i] >> 1;
 
     /* Compute impulse response autocorrelation */
-    temp = dot_product(temp_corr, temp_corr, SUBFRAME_LEN, 1);
+    temp = dot_product(temp_corr, temp_corr, SUBFRAME_LEN);
 
     scale = normalize_bits_int32(temp);
     impulse_corr[0] = av_clipl_int32((temp << scale) + (1 << 15)) >> 16;
 
     for (i = 1; i < SUBFRAME_LEN; i++) {
-        temp = dot_product(temp_corr + i, temp_corr, SUBFRAME_LEN - i, 1);
+        temp = dot_product(temp_corr + i, temp_corr, SUBFRAME_LEN - i);
         impulse_corr[i] = av_clipl_int32((temp << scale) + (1 << 15)) >> 16;
     }
 
     /* Compute crosscorrelation of impulse response with residual signal */
     scale -= 4;
     for (i = 0; i < SUBFRAME_LEN; i++){
-        temp = dot_product(buf + i, impulse_r, SUBFRAME_LEN - i, 1);
+        temp = dot_product(buf + i, impulse_r, SUBFRAME_LEN - i);
         if (scale < 0)
             ccr1[i] = temp >> -scale;
         else
@@ -2185,7 +2180,7 @@ static int g723_1_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
     memcpy(vector, p->prev_weight_sig, sizeof(int16_t) * PITCH_MAX);
     memcpy(vector + PITCH_MAX, in, sizeof(int16_t) * FRAME_LEN);
 
-    scale_vector(vector, FRAME_LEN + PITCH_MAX);
+    scale_vector(vector, vector, FRAME_LEN + PITCH_MAX);
 
     p->pitch_lag[0] = estimate_pitch(vector, PITCH_MAX);
     p->pitch_lag[1] = estimate_pitch(vector, PITCH_MAX + HALF_FRAME_LEN);
@@ -2237,14 +2232,14 @@ static int g723_1_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
 
         acb_search(p, residual, impulse_resp, in, i);
         gen_acb_excitation(residual, p->prev_excitation,p->pitch_lag[i >> 1],
-                           p->subframe[i], p->cur_rate);
+                           &p->subframe[i], p->cur_rate);
         sub_acb_contrib(residual, impulse_resp, in);
 
         fcb_search(p, impulse_resp, in, i);
 
         /* Reconstruct the excitation */
         gen_acb_excitation(impulse_resp, p->prev_excitation, p->pitch_lag[i >> 1],
-                           p->subframe[i], RATE_6300);
+                           &p->subframe[i], RATE_6300);
 
         memmove(p->prev_excitation, p->prev_excitation + SUBFRAME_LEN,
                sizeof(int16_t) * (PITCH_MAX - SUBFRAME_LEN));
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index 3e998efafc..e01454a1b7 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -41,7 +41,7 @@ YASM-OBJS-$(CONFIG_AAC_DECODER)        += x86/sbrdsp.o
 YASM-OBJS-$(CONFIG_AC3DSP)             += x86/ac3dsp.o
 YASM-OBJS-$(CONFIG_DCT)                += x86/dct32_sse.o
 YASM-OBJS-$(CONFIG_DIRAC_DECODER)      += x86/diracdsp_mmx.o x86/diracdsp_yasm.o
-YASM-OBJS-$(CONFIG_ENCODERS)           += x86/dsputilenc_yasm.o
+YASM-OBJS-$(CONFIG_ENCODERS)           += x86/dsputilenc.o
 YASM-OBJS-$(CONFIG_FFT)                += x86/fft_mmx.o                 \
                                           $(YASM-OBJS-FFT-yes)
 
@@ -65,11 +65,11 @@ YASM-OBJS-$(CONFIG_RV30_DECODER)       += x86/rv34dsp.o
 YASM-OBJS-$(CONFIG_RV40_DECODER)       += x86/rv34dsp.o                 \
                                           x86/rv40dsp.o
 YASM-OBJS-$(CONFIG_V210_DECODER)       += x86/v210.o
-YASM-OBJS-$(CONFIG_VC1_DECODER)        += x86/vc1dsp_yasm.o
+YASM-OBJS-$(CONFIG_VC1_DECODER)        += x86/vc1dsp.o
 YASM-OBJS-$(CONFIG_VP3DSP)             += x86/vp3dsp.o
 YASM-OBJS-$(CONFIG_VP6_DECODER)        += x86/vp56dsp.o
 YASM-OBJS-$(CONFIG_VP8_DECODER)        += x86/vp8dsp.o
 
-YASM-OBJS                              += x86/dsputil_yasm.o            \
+YASM-OBJS                              += x86/dsputil.o                 \
                                           x86/deinterlace.o             \
                                           x86/fmtconvert.o              \
diff --git a/libavcodec/x86/dsputil_yasm.asm b/libavcodec/x86/dsputil.asm
index 19884a36a8..19884a36a8 100644
--- a/libavcodec/x86/dsputil_yasm.asm
+++ b/libavcodec/x86/dsputil.asm
diff --git a/libavcodec/x86/dsputilenc_yasm.asm b/libavcodec/x86/dsputilenc.asm
index 2d805c7da8..2d805c7da8 100644
--- a/libavcodec/x86/dsputilenc_yasm.asm
+++ b/libavcodec/x86/dsputilenc.asm
diff --git a/libavcodec/x86/vc1dsp_yasm.asm b/libavcodec/x86/vc1dsp.asm
index 590aa509a7..590aa509a7 100644
--- a/libavcodec/x86/vc1dsp_yasm.asm
+++ b/libavcodec/x86/vc1dsp.asm
diff --git a/libavutil/arm/intmath.h b/libavutil/arm/intmath.h
index 4af4bf32d1..cf6293525f 100644
--- a/libavutil/arm/intmath.h
+++ b/libavutil/arm/intmath.h
@@ -83,6 +83,21 @@ static av_always_inline av_const unsigned av_clip_uintp2_arm(int a, int p)
     return x;
 }
 
+#define av_sat_add32 av_sat_add32_arm
+static av_always_inline int av_sat_add32_arm(int a, int b)
+{
+    int r;
+    __asm__ ("qadd %0, %1, %2" : "=r"(r) : "r"(a), "r"(b));
+    return r;
+}
+
+#define av_sat_dadd32 av_sat_dadd32_arm
+static av_always_inline int av_sat_dadd32_arm(int a, int b)
+{
+    int r;
+    __asm__ ("qdadd %0, %1, %2" : "=r"(r) : "r"(a), "r"(b));
+    return r;
+}
 
 #else /* HAVE_ARMV6 */
 
diff --git a/libavutil/common.h b/libavutil/common.h
index a11a3251e9..3e3baab3a1 100644
--- a/libavutil/common.h
+++ b/libavutil/common.h
@@ -187,6 +187,30 @@ static av_always_inline av_const unsigned av_clip_uintp2_c(int a, int p)
 }
 
 /**
+ * Add two signed 32-bit values with saturation.
+ *
+ * @param  a one value
+ * @param  b another value
+ * @return sum with signed saturation
+ */
+static av_always_inline int av_sat_add32_c(int a, int b)
+{
+    return av_clipl_int32((int64_t)a + b);
+}
+
+/**
+ * Add a doubled value to another value with saturation at both stages.
+ *
+ * @param  a first value
+ * @param  b value doubled and added to a
+ * @return sum with signed saturation
+ */
+static av_always_inline int av_sat_dadd32_c(int a, int b)
+{
+    return av_sat_add32(a, av_sat_add32(b, b));
+}
+
+/**
  * Clip a float value into the amin-amax range.
  * @param a value to clip
  * @param amin minimum value of the clip range
@@ -392,6 +416,12 @@ static av_always_inline av_const int av_popcount64_c(uint64_t x)
 #ifndef av_clip_uintp2
 #   define av_clip_uintp2   av_clip_uintp2_c
 #endif
+#ifndef av_sat_add32
+#   define av_sat_add32     av_sat_add32_c
+#endif
+#ifndef av_sat_dadd32
+#   define av_sat_dadd32    av_sat_dadd32_c
+#endif
 #ifndef av_clipf
 #   define av_clipf         av_clipf_c
 #endif