Merge remote-tracking branch 'qatar/master'

* qatar/master: (28 commits)
  dfa: use more meaningful return codes
  eatgv: check vector_bits
  eatgv: check motion vectors
  Mark a number of variables only used in av_dlog() calls as av_unused.
  dvdec: drop const qualifier from variable to eliminate a warning
  avcodec: Improve comment for thread_safe_callbacks to avoid misinterpretation.
  tests/utils: don't ignore the return value of fwrite()
  lavfi/formats: use sizeof(var) instead of sizeof(type).
  lavfi: remove avfilter_default_config_input_link() declaration
  lavfi: always enable the scale filter and depend on sws.
  vf_split: support user-specifiable number of outputs.
  avconv: remove stray useless comment.
  mpegmux: add stuffing to avoid incomplete PCM frames
  rtsp: avoid const warnings from strtol() call
  avserver: check return value of ftruncate()
  lagarith: make offset array type unsigned
  dfa: add some checks to ensure that decoder won't write past frame end
  aacps: NEON optimisations
  aacps: align some arrays
  aacps: move some loops to function pointers
  ...

Conflicts:
	configure
	doc/filters.texi
	libavcodec/dfa.c
	libavcodec/eatgv.c
	libavfilter/Makefile
	libavfilter/allfilters.c
	libavfilter/avfilter.h
	libavfilter/formats.c
	libavfilter/vf_split.c

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

16 files changed, 755 insertions, 236 deletions

diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 891e29fb0a..a39ec88e06 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -57,7 +57,7 @@ OBJS-$(CONFIG_A64MULTI_ENCODER)        += a64multienc.o elbg.o
 OBJS-$(CONFIG_A64MULTI5_ENCODER)       += a64multienc.o elbg.o
 OBJS-$(CONFIG_AAC_DECODER)             += aacdec.o aactab.o aacsbr.o aacps.o \
                                           aacadtsdec.o mpeg4audio.o kbdwin.o \
-                                          sbrdsp.o
+                                          sbrdsp.o aacpsdsp.o
 OBJS-$(CONFIG_AAC_ENCODER)             += aacenc.o aaccoder.o    \
                                           aacpsy.o aactab.o      \
                                           psymodel.o iirfilter.o \
diff --git a/libavcodec/aacps.c b/libavcodec/aacps.c
index c2afec0af5..569b44eb39 100644
--- a/libavcodec/aacps.c
+++ b/libavcodec/aacps.c
@@ -27,6 +27,7 @@
 #include "aacps.h"
 #include "aacps_tablegen.h"
 #include "aacpsdata.c"
+#include "dsputil.h"
 
 #define PS_BASELINE 0  ///< Operate in Baseline PS mode
                        ///< Baseline implies 10 or 20 stereo bands,
@@ -284,7 +285,7 @@ err:
 
 /** Split one subband into 2 subsubbands with a symmetric real filter.
  * The filter must have its non-center even coefficients equal to zero. */
-static void hybrid2_re(float (*in)[2], float (*out)[32][2], const float filter[7], int len, int reverse)
+static void hybrid2_re(float (*in)[2], float (*out)[32][2], const float filter[8], int len, int reverse)
 {
     int i, j;
     for (i = 0; i < len; i++, in++) {
@@ -304,26 +305,14 @@ static void hybrid2_re(float (*in)[2], float (*out)[32][2], const float filter[7
 }
 
 /** Split one subband into 6 subsubbands with a complex filter */
-static void hybrid6_cx(float (*in)[2], float (*out)[32][2], const float (*filter)[7][2], int len)
+static void hybrid6_cx(PSDSPContext *dsp, float (*in)[2], float (*out)[32][2], const float (*filter)[8][2], int len)
 {
-    int i, j, ssb;
+    int i;
     int N = 8;
-    float temp[8][2];
+    LOCAL_ALIGNED_16(float, temp, [8], [2]);
 
     for (i = 0; i < len; i++, in++) {
-        for (ssb = 0; ssb < N; ssb++) {
-            float sum_re = filter[ssb][6][0] * in[6][0], sum_im = filter[ssb][6][0] * in[6][1];
-            for (j = 0; j < 6; j++) {
-                float in0_re = in[j][0];
-                float in0_im = in[j][1];
-                float in1_re = in[12-j][0];
-                float in1_im = in[12-j][1];
-                sum_re += filter[ssb][j][0] * (in0_re + in1_re) - filter[ssb][j][1] * (in0_im - in1_im);
-                sum_im += filter[ssb][j][0] * (in0_im + in1_im) + filter[ssb][j][1] * (in0_re - in1_re);
-            }
-            temp[ssb][0] = sum_re;
-            temp[ssb][1] = sum_im;
-        }
+        dsp->hybrid_analysis(temp, in, filter, 1, N);
         out[0][i][0] = temp[6][0];
         out[0][i][1] = temp[6][1];
         out[1][i][0] = temp[7][0];
@@ -339,28 +328,18 @@ static void hybrid6_cx(float (*in)[2], float (*out)[32][2], const float (*filter
     }
 }
 
-static void hybrid4_8_12_cx(float (*in)[2], float (*out)[32][2], const float (*filter)[7][2], int N, int len)
+static void hybrid4_8_12_cx(PSDSPContext *dsp, float (*in)[2], float (*out)[32][2], const float (*filter)[8][2], int N, int len)
 {
-    int i, j, ssb;
+    int i;
 
     for (i = 0; i < len; i++, in++) {
-        for (ssb = 0; ssb < N; ssb++) {
-            float sum_re = filter[ssb][6][0] * in[6][0], sum_im = filter[ssb][6][0] * in[6][1];
-            for (j = 0; j < 6; j++) {
-                float in0_re = in[j][0];
-                float in0_im = in[j][1];
-                float in1_re = in[12-j][0];
-                float in1_im = in[12-j][1];
-                sum_re += filter[ssb][j][0] * (in0_re + in1_re) - filter[ssb][j][1] * (in0_im - in1_im);
-                sum_im += filter[ssb][j][0] * (in0_im + in1_im) + filter[ssb][j][1] * (in0_re - in1_re);
-            }
-            out[ssb][i][0] = sum_re;
-            out[ssb][i][1] = sum_im;
-        }
+        dsp->hybrid_analysis(out[0] + i, in, filter, 32, N);
     }
 }
 
-static void hybrid_analysis(float out[91][32][2], float in[5][44][2], float L[2][38][64], int is34, int len)
+static void hybrid_analysis(PSDSPContext *dsp, float out[91][32][2],
+                            float in[5][44][2], float L[2][38][64],
+                            int is34, int len)
 {
     int i, j;
     for (i = 0; i < 5; i++) {
@@ -370,27 +349,17 @@ static void hybrid_analysis(float out[91][32][2], float in[5][44][2], float L[2]
         }
     }
     if (is34) {
-        hybrid4_8_12_cx(in[0], out,    f34_0_12, 12, len);
-        hybrid4_8_12_cx(in[1], out+12, f34_1_8,   8, len);
-        hybrid4_8_12_cx(in[2], out+20, f34_2_4,   4, len);
-        hybrid4_8_12_cx(in[3], out+24, f34_2_4,   4, len);
-        hybrid4_8_12_cx(in[4], out+28, f34_2_4,   4, len);
-        for (i = 0; i < 59; i++) {
-            for (j = 0; j < len; j++) {
-                out[i+32][j][0] = L[0][j][i+5];
-                out[i+32][j][1] = L[1][j][i+5];
-            }
-        }
+        hybrid4_8_12_cx(dsp, in[0], out,    f34_0_12, 12, len);
+        hybrid4_8_12_cx(dsp, in[1], out+12, f34_1_8,   8, len);
+        hybrid4_8_12_cx(dsp, in[2], out+20, f34_2_4,   4, len);
+        hybrid4_8_12_cx(dsp, in[3], out+24, f34_2_4,   4, len);
+        hybrid4_8_12_cx(dsp, in[4], out+28, f34_2_4,   4, len);
+        dsp->hybrid_analysis_ileave(out + 27, L, 5, len);
     } else {
-        hybrid6_cx(in[0], out, f20_0_8, len);
+        hybrid6_cx(dsp, in[0], out, f20_0_8, len);
         hybrid2_re(in[1], out+6, g1_Q2, len, 1);
         hybrid2_re(in[2], out+8, g1_Q2, len, 0);
-        for (i = 0; i < 61; i++) {
-            for (j = 0; j < len; j++) {
-                out[i+10][j][0] = L[0][j][i+3];
-                out[i+10][j][1] = L[1][j][i+3];
-            }
-        }
+        dsp->hybrid_analysis_ileave(out + 7, L, 3, len);
     }
     //update in_buf
     for (i = 0; i < 5; i++) {
@@ -398,7 +367,8 @@ static void hybrid_analysis(float out[91][32][2], float in[5][44][2], float L[2]
     }
 }
 
-static void hybrid_synthesis(float out[2][38][64], float in[91][32][2], int is34, int len)
+static void hybrid_synthesis(PSDSPContext *dsp, float out[2][38][64],
+                             float in[91][32][2], int is34, int len)
 {
     int i, n;
     if (is34) {
@@ -422,12 +392,7 @@ static void hybrid_synthesis(float out[2][38][64], float in[91][32][2], int is34
                 out[1][n][4] += in[28+i][n][1];
             }
         }
-        for (i = 0; i < 59; i++) {
-            for (n = 0; n < len; n++) {
-                out[0][n][i+5] = in[i+32][n][0];
-                out[1][n][i+5] = in[i+32][n][1];
-            }
-        }
+        dsp->hybrid_synthesis_deint(out, in + 27, 5, len);
     } else {
         for (n = 0; n < len; n++) {
             out[0][n][0] = in[0][n][0] + in[1][n][0] + in[2][n][0] +
@@ -439,12 +404,7 @@ static void hybrid_synthesis(float out[2][38][64], float in[91][32][2], int is34
             out[0][n][2] = in[8][n][0] + in[9][n][0];
             out[1][n][2] = in[8][n][1] + in[9][n][1];
         }
-        for (i = 0; i < 61; i++) {
-            for (n = 0; n < len; n++) {
-                out[0][n][i+3] = in[i+10][n][0];
-                out[1][n][i+3] = in[i+10][n][1];
-            }
-        }
+        dsp->hybrid_synthesis_deint(out, in + 7, 3, len);
     }
 }
 
@@ -648,8 +608,8 @@ static void map_val_20_to_34(float par[PS_MAX_NR_IIDICC])
 
 static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[32][2], int is34)
 {
-    float power[34][PS_QMF_TIME_SLOTS] = {{0}};
-    float transient_gain[34][PS_QMF_TIME_SLOTS];
+    LOCAL_ALIGNED_16(float, power, [34], [PS_QMF_TIME_SLOTS]);
+    LOCAL_ALIGNED_16(float, transient_gain, [34], [PS_QMF_TIME_SLOTS]);
     float *peak_decay_nrg = ps->peak_decay_nrg;
     float *power_smooth = ps->power_smooth;
     float *peak_decay_diff_smooth = ps->peak_decay_diff_smooth;
@@ -661,10 +621,8 @@ static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[3
     const float a_smooth          = 0.25f; ///< Smoothing coefficient
     int i, k, m, n;
     int n0 = 0, nL = 32;
-    static const int link_delay[] = { 3, 4, 5 };
-    static const float a[] = { 0.65143905753106f,
-                               0.56471812200776f,
-                               0.48954165955695f };
+
+    memset(power, 0, 34 * sizeof(*power));
 
     if (is34 != ps->is34bands_old) {
         memset(ps->peak_decay_nrg,         0, sizeof(ps->peak_decay_nrg));
@@ -674,11 +632,9 @@ static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[3
         memset(ps->ap_delay,               0, sizeof(ps->ap_delay));
     }
 
-    for (n = n0; n < nL; n++) {
-        for (k = 0; k < NR_BANDS[is34]; k++) {
-            int i = k_to_i[k];
-            power[i][n] += s[k][n][0] * s[k][n][0] + s[k][n][1] * s[k][n][1];
-        }
+    for (k = 0; k < NR_BANDS[is34]; k++) {
+        int i = k_to_i[k];
+        ps->dsp.add_squares(power[i], s[k], nL - n0);
     }
 
     //Transient detection
@@ -706,54 +662,31 @@ static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[3
     for (k = 0; k < NR_ALLPASS_BANDS[is34]; k++) {
         int b = k_to_i[k];
         float g_decay_slope = 1.f - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]);
-        float ag[PS_AP_LINKS];
         g_decay_slope = av_clipf(g_decay_slope, 0.f, 1.f);
         memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
         memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
         for (m = 0; m < PS_AP_LINKS; m++) {
             memcpy(ap_delay[k][m],   ap_delay[k][m]+numQMFSlots,           5*sizeof(ap_delay[k][m][0]));
-            ag[m] = a[m] * g_decay_slope;
-        }
-        for (n = n0; n < nL; n++) {
-            float in_re = delay[k][n+PS_MAX_DELAY-2][0] * phi_fract[is34][k][0] -
-                          delay[k][n+PS_MAX_DELAY-2][1] * phi_fract[is34][k][1];
-            float in_im = delay[k][n+PS_MAX_DELAY-2][0] * phi_fract[is34][k][1] +
-                          delay[k][n+PS_MAX_DELAY-2][1] * phi_fract[is34][k][0];
-            for (m = 0; m < PS_AP_LINKS; m++) {
-                float a_re                = ag[m] * in_re;
-                float a_im                = ag[m] * in_im;
-                float link_delay_re       = ap_delay[k][m][n+5-link_delay[m]][0];
-                float link_delay_im       = ap_delay[k][m][n+5-link_delay[m]][1];
-                float fractional_delay_re = Q_fract_allpass[is34][k][m][0];
-                float fractional_delay_im = Q_fract_allpass[is34][k][m][1];
-                ap_delay[k][m][n+5][0] = in_re;
-                ap_delay[k][m][n+5][1] = in_im;
-                in_re = link_delay_re * fractional_delay_re - link_delay_im * fractional_delay_im - a_re;
-                in_im = link_delay_re * fractional_delay_im + link_delay_im * fractional_delay_re - a_im;
-                ap_delay[k][m][n+5][0] += ag[m] * in_re;
-                ap_delay[k][m][n+5][1] += ag[m] * in_im;
-            }
-            out[k][n][0] = transient_gain[b][n] * in_re;
-            out[k][n][1] = transient_gain[b][n] * in_im;
         }
+        ps->dsp.decorrelate(out[k], delay[k] + PS_MAX_DELAY - 2, ap_delay[k],
+                            phi_fract[is34][k], Q_fract_allpass[is34][k],
+                            transient_gain[b], g_decay_slope, nL - n0);
     }
     for (; k < SHORT_DELAY_BAND[is34]; k++) {
+        int i = k_to_i[k];
         memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
         memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
-        for (n = n0; n < nL; n++) {
-            //H = delay 14
-            out[k][n][0] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-14][0];
-            out[k][n][1] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-14][1];
-        }
+        //H = delay 14
+        ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 14,
+                                transient_gain[i], nL - n0);
     }
     for (; k < NR_BANDS[is34]; k++) {
+        int i = k_to_i[k];
         memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
         memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
-        for (n = n0; n < nL; n++) {
-            //H = delay 1
-            out[k][n][0] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-1][0];
-            out[k][n][1] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-1][1];
-        }
+        //H = delay 1
+        ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 1,
+                                transient_gain[i], nL - n0);
     }
 }
 
@@ -797,7 +730,7 @@ static void remap20(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC],
 
 static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2], int is34)
 {
-    int e, b, k, n;
+    int e, b, k;
 
     float (*H11)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H11;
     float (*H12)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H12;
@@ -909,78 +842,52 @@ static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2
             H22[0][e+1][b] = h22;
         }
         for (k = 0; k < NR_BANDS[is34]; k++) {
-            float h11r, h12r, h21r, h22r;
-            float h11i, h12i, h21i, h22i;
-            float h11r_step, h12r_step, h21r_step, h22r_step;
-            float h11i_step, h12i_step, h21i_step, h22i_step;
+            float h[2][4];
+            float h_step[2][4];
             int start = ps->border_position[e];
             int stop  = ps->border_position[e+1];
             float width = 1.f / (stop - start);
             b = k_to_i[k];
-            h11r = H11[0][e][b];
-            h12r = H12[0][e][b];
-            h21r = H21[0][e][b];
-            h22r = H22[0][e][b];
+            h[0][0] = H11[0][e][b];
+            h[0][1] = H12[0][e][b];
+            h[0][2] = H21[0][e][b];
+            h[0][3] = H22[0][e][b];
             if (!PS_BASELINE && ps->enable_ipdopd) {
             //Is this necessary? ps_04_new seems unchanged
             if ((is34 && k <= 13 && k >= 9) || (!is34 && k <= 1)) {
-                h11i = -H11[1][e][b];
-                h12i = -H12[1][e][b];
-                h21i = -H21[1][e][b];
-                h22i = -H22[1][e][b];
+                h[1][0] = -H11[1][e][b];
+                h[1][1] = -H12[1][e][b];
+                h[1][2] = -H21[1][e][b];
+                h[1][3] = -H22[1][e][b];
             } else {
-                h11i = H11[1][e][b];
-                h12i = H12[1][e][b];
-                h21i = H21[1][e][b];
-                h22i = H22[1][e][b];
+                h[1][0] = H11[1][e][b];
+                h[1][1] = H12[1][e][b];
+                h[1][2] = H21[1][e][b];
+                h[1][3] = H22[1][e][b];
             }
             }
             //Interpolation
-            h11r_step = (H11[0][e+1][b] - h11r) * width;
-            h12r_step = (H12[0][e+1][b] - h12r) * width;
-            h21r_step = (H21[0][e+1][b] - h21r) * width;
-            h22r_step = (H22[0][e+1][b] - h22r) * width;
+            h_step[0][0] = (H11[0][e+1][b] - h[0][0]) * width;
+            h_step[0][1] = (H12[0][e+1][b] - h[0][1]) * width;
+            h_step[0][2] = (H21[0][e+1][b] - h[0][2]) * width;
+            h_step[0][3] = (H22[0][e+1][b] - h[0][3]) * width;
             if (!PS_BASELINE && ps->enable_ipdopd) {
-                h11i_step = (H11[1][e+1][b] - h11i) * width;
-                h12i_step = (H12[1][e+1][b] - h12i) * width;
-                h21i_step = (H21[1][e+1][b] - h21i) * width;
-                h22i_step = (H22[1][e+1][b] - h22i) * width;
-            }
-            for (n = start + 1; n <= stop; n++) {
-                //l is s, r is d
-                float l_re = l[k][n][0];
-                float l_im = l[k][n][1];
-                float r_re = r[k][n][0];
-                float r_im = r[k][n][1];
-                h11r += h11r_step;
-                h12r += h12r_step;
-                h21r += h21r_step;
-                h22r += h22r_step;
-                if (!PS_BASELINE && ps->enable_ipdopd) {
-                    h11i += h11i_step;
-                    h12i += h12i_step;
-                    h21i += h21i_step;
-                    h22i += h22i_step;
-
-                    l[k][n][0] = h11r*l_re + h21r*r_re - h11i*l_im - h21i*r_im;
-                    l[k][n][1] = h11r*l_im + h21r*r_im + h11i*l_re + h21i*r_re;
-                    r[k][n][0] = h12r*l_re + h22r*r_re - h12i*l_im - h22i*r_im;
-                    r[k][n][1] = h12r*l_im + h22r*r_im + h12i*l_re + h22i*r_re;
-                } else {
-                    l[k][n][0] = h11r*l_re + h21r*r_re;
-                    l[k][n][1] = h11r*l_im + h21r*r_im;
-                    r[k][n][0] = h12r*l_re + h22r*r_re;
-                    r[k][n][1] = h12r*l_im + h22r*r_im;
-                }
+                h_step[1][0] = (H11[1][e+1][b] - h[1][0]) * width;
+                h_step[1][1] = (H12[1][e+1][b] - h[1][1]) * width;
+                h_step[1][2] = (H21[1][e+1][b] - h[1][2]) * width;
+                h_step[1][3] = (H22[1][e+1][b] - h[1][3]) * width;
             }
+            ps->dsp.stereo_interpolate[!PS_BASELINE && ps->enable_ipdopd](
+                l[k] + start + 1, r[k] + start + 1,
+                h, h_step, stop - start);
         }
     }
 }
 
 int ff_ps_apply(AVCodecContext *avctx, PSContext *ps, float L[2][38][64], float R[2][38][64], int top)
 {
-    float Lbuf[91][32][2];
-    float Rbuf[91][32][2];
+    LOCAL_ALIGNED_16(float, Lbuf, [91], [32][2]);
+    LOCAL_ALIGNED_16(float, Rbuf, [91], [32][2]);
     const int len = 32;
     int is34 = ps->is34bands;
 
@@ -989,11 +896,11 @@ int ff_ps_apply(AVCodecContext *avctx, PSContext *ps, float L[2][38][64], float
     if (top < NR_ALLPASS_BANDS[is34])
         memset(ps->ap_delay + top, 0, (NR_ALLPASS_BANDS[is34] - top)*sizeof(ps->ap_delay[0]));
 
-    hybrid_analysis(Lbuf, ps->in_buf, L, is34, len);
+    hybrid_analysis(&ps->dsp, Lbuf, ps->in_buf, L, is34, len);
     decorrelation(ps, Rbuf, Lbuf, is34);
     stereo_processing(ps, Lbuf, Rbuf, is34);
-    hybrid_synthesis(L, Lbuf, is34, len);
-    hybrid_synthesis(R, Rbuf, is34, len);
+    hybrid_synthesis(&ps->dsp, L, Lbuf, is34, len);
+    hybrid_synthesis(&ps->dsp, R, Rbuf, is34, len);
 
     return 0;
 }
@@ -1041,4 +948,5 @@ av_cold void ff_ps_init(void) {
 
 av_cold void ff_ps_ctx_init(PSContext *ps)
 {
+    ff_psdsp_init(&ps->dsp);
 }
diff --git a/libavcodec/aacps.h b/libavcodec/aacps.h
index 757a44366b..d310c91b95 100644
--- a/libavcodec/aacps.h
+++ b/libavcodec/aacps.h
@@ -24,6 +24,7 @@
 
 #include <stdint.h>
 
+#include "aacpsdsp.h"
 #include "avcodec.h"
 #include "get_bits.h"
 
@@ -60,18 +61,19 @@ typedef struct {
     int    is34bands;
     int    is34bands_old;
 
-    float  in_buf[5][44][2];
-    float  delay[PS_MAX_SSB][PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2];
-    float  ap_delay[PS_MAX_AP_BANDS][PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2];
-    float  peak_decay_nrg[34];
-    float  power_smooth[34];
-    float  peak_decay_diff_smooth[34];
-    float  H11[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC];
-    float  H12[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC];
-    float  H21[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC];
-    float  H22[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC];
+    DECLARE_ALIGNED(16, float, in_buf)[5][44][2];
+    DECLARE_ALIGNED(16, float, delay)[PS_MAX_SSB][PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2];
+    DECLARE_ALIGNED(16, float, ap_delay)[PS_MAX_AP_BANDS][PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2];
+    DECLARE_ALIGNED(16, float, peak_decay_nrg)[34];
+    DECLARE_ALIGNED(16, float, power_smooth)[34];
+    DECLARE_ALIGNED(16, float, peak_decay_diff_smooth)[34];
+    DECLARE_ALIGNED(16, float, H11)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC];
+    DECLARE_ALIGNED(16, float, H12)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC];
+    DECLARE_ALIGNED(16, float, H21)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC];
+    DECLARE_ALIGNED(16, float, H22)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC];
     int8_t opd_hist[PS_MAX_NR_IIDICC];
     int8_t ipd_hist[PS_MAX_NR_IIDICC];
+    PSDSPContext dsp;
 } PSContext;
 
 void ff_ps_init(void);
diff --git a/libavcodec/aacps_tablegen.c b/libavcodec/aacps_tablegen.c
index dc7797f6b8..f56930b958 100644
--- a/libavcodec/aacps_tablegen.c
+++ b/libavcodec/aacps_tablegen.c
@@ -69,23 +69,23 @@ int main(void)
     write_float_3d_array(HB, 46, 8, 4);
     printf("};\n");
 
-    printf("static const float f20_0_8[8][7][2] = {\n");
-    write_float_3d_array(f20_0_8, 8, 7, 2);
+    printf("static const DECLARE_ALIGNED(16, float, f20_0_8)[8][8][2] = {\n");
+    write_float_3d_array(f20_0_8, 8, 8, 2);
     printf("};\n");
-    printf("static const float f34_0_12[12][7][2] = {\n");
-    write_float_3d_array(f34_0_12, 12, 7, 2);
+    printf("static const DECLARE_ALIGNED(16, float, f34_0_12)[12][8][2] = {\n");
+    write_float_3d_array(f34_0_12, 12, 8, 2);
     printf("};\n");
-    printf("static const float f34_1_8[8][7][2] = {\n");
-    write_float_3d_array(f34_1_8, 8, 7, 2);
+    printf("static const DECLARE_ALIGNED(16, float, f34_1_8)[8][8][2] = {\n");
+    write_float_3d_array(f34_1_8, 8, 8, 2);
     printf("};\n");
-    printf("static const float f34_2_4[4][7][2] = {\n");
-    write_float_3d_array(f34_2_4, 4, 7, 2);
+    printf("static const DECLARE_ALIGNED(16, float, f34_2_4)[4][8][2] = {\n");
+    write_float_3d_array(f34_2_4, 4, 8, 2);
     printf("};\n");
 
-    printf("static const float Q_fract_allpass[2][50][3][2] = {\n");
+    printf("static const DECLARE_ALIGNED(16, float, Q_fract_allpass)[2][50][3][2] = {\n");
     write_float_4d_array(Q_fract_allpass, 2, 50, 3, 2);
     printf("};\n");
-    printf("static const float phi_fract[2][50][2] = {\n");
+    printf("static const DECLARE_ALIGNED(16, float, phi_fract)[2][50][2] = {\n");
     write_float_3d_array(phi_fract, 2, 50, 2);
     printf("};\n");
 
diff --git a/libavcodec/aacps_tablegen.h b/libavcodec/aacps_tablegen.h
index 79ae644a87..70cc96c9d7 100644
--- a/libavcodec/aacps_tablegen.h
+++ b/libavcodec/aacps_tablegen.h
@@ -31,6 +31,7 @@
 #else
 #include "libavutil/common.h"
 #include "libavutil/mathematics.h"
+#include "libavutil/mem.h"
 #define NR_ALLPASS_BANDS20 30
 #define NR_ALLPASS_BANDS34 50
 #define PS_AP_LINKS 3
@@ -38,12 +39,12 @@ static float pd_re_smooth[8*8*8];
 static float pd_im_smooth[8*8*8];
 static float HA[46][8][4];
 static float HB[46][8][4];
-static float f20_0_8 [ 8][7][2];
-static float f34_0_12[12][7][2];
-static float f34_1_8 [ 8][7][2];
-static float f34_2_4 [ 4][7][2];
-static float Q_fract_allpass[2][50][3][2];
-static float phi_fract[2][50][2];
+static DECLARE_ALIGNED(16, float, f20_0_8) [ 8][8][2];
+static DECLARE_ALIGNED(16, float, f34_0_12)[12][8][2];
+static DECLARE_ALIGNED(16, float, f34_1_8) [ 8][8][2];
+static DECLARE_ALIGNED(16, float, f34_2_4) [ 4][8][2];
+static DECLARE_ALIGNED(16, float, Q_fract_allpass)[2][50][3][2];
+static DECLARE_ALIGNED(16, float, phi_fract)[2][50][2];
 
 static const float g0_Q8[] = {
     0.00746082949812f, 0.02270420949825f, 0.04546865930473f, 0.07266113929591f,
@@ -65,7 +66,7 @@ static const float g2_Q4[] = {
      0.16486303567403f,  0.23279856662996f, 0.25f
 };
 
-static void make_filters_from_proto(float (*filter)[7][2], const float *proto, int bands)
+static void make_filters_from_proto(float (*filter)[8][2], const float *proto, int bands)
 {
     int q, n;
     for (q = 0; q < bands; q++) {
diff --git a/libavcodec/aacpsdsp.c b/libavcodec/aacpsdsp.c
new file mode 100644
index 0000000000..e90c50b6a1
--- /dev/null
+++ b/libavcodec/aacpsdsp.c
@@ -0,0 +1,214 @@
+/*
+ * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+#include "libavutil/attributes.h"
+#include "aacpsdsp.h"
+
+static void ps_add_squares_c(float *dst, const float (*src)[2], int n)
+{
+    int i;
+    for (i = 0; i < n; i++)
+        dst[i] += src[i][0] * src[i][0] + src[i][1] * src[i][1];
+}
+
+static void ps_mul_pair_single_c(float (*dst)[2], float (*src0)[2], float *src1,
+                                 int n)
+{
+    int i;
+    for (i = 0; i < n; i++) {
+        dst[i][0] = src0[i][0] * src1[i];
+        dst[i][1] = src0[i][1] * src1[i];
+    }
+}
+
+static void ps_hybrid_analysis_c(float (*out)[2], float (*in)[2],
+                                 const float (*filter)[8][2],
+                                 int stride, int n)
+{
+    int i, j;
+
+    for (i = 0; i < n; i++) {
+        float sum_re = filter[i][6][0] * in[6][0];
+        float sum_im = filter[i][6][0] * in[6][1];
+
+        for (j = 0; j < 6; j++) {
+            float in0_re = in[j][0];
+            float in0_im = in[j][1];
+            float in1_re = in[12-j][0];
+            float in1_im = in[12-j][1];
+            sum_re += filter[i][j][0] * (in0_re + in1_re) -
+                      filter[i][j][1] * (in0_im - in1_im);
+            sum_im += filter[i][j][0] * (in0_im + in1_im) +
+                      filter[i][j][1] * (in0_re - in1_re);
+        }
+        out[i * stride][0] = sum_re;
+        out[i * stride][1] = sum_im;
+    }
+}
+
+static void ps_hybrid_analysis_ileave_c(float (*out)[32][2], float L[2][38][64],
+                                        int i, int len)
+{
+    int j;
+
+    for (; i < 64; i++) {
+        for (j = 0; j < len; j++) {
+            out[i][j][0] = L[0][j][i];
+            out[i][j][1] = L[1][j][i];
+        }
+    }
+}
+
+static void ps_hybrid_synthesis_deint_c(float out[2][38][64],
+                                        float (*in)[32][2],
+                                        int i, int len)
+{
+    int n;
+
+    for (; i < 64; i++) {
+        for (n = 0; n < len; n++) {
+            out[0][n][i] = in[i][n][0];
+            out[1][n][i] = in[i][n][1];
+        }
+    }
+}
+
+static void ps_decorrelate_c(float (*out)[2], float (*delay)[2],
+                             float (*ap_delay)[PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2],
+                             const float phi_fract[2], float (*Q_fract)[2],
+                             const float *transient_gain,
+                             float g_decay_slope,
+                             int len)
+{
+    static const float a[] = { 0.65143905753106f,
+                               0.56471812200776f,
+                               0.48954165955695f };
+    float ag[PS_AP_LINKS];
+    int m, n;
+
+    for (m = 0; m < PS_AP_LINKS; m++)
+        ag[m] = a[m] * g_decay_slope;
+
+    for (n = 0; n < len; n++) {
+        float in_re = delay[n][0] * phi_fract[0] - delay[n][1] * phi_fract[1];
+        float in_im = delay[n][0] * phi_fract[1] + delay[n][1] * phi_fract[0];
+        for (m = 0; m < PS_AP_LINKS; m++) {
+            float a_re                = ag[m] * in_re;
+            float a_im                = ag[m] * in_im;
+            float link_delay_re       = ap_delay[m][n+2-m][0];
+            float link_delay_im       = ap_delay[m][n+2-m][1];
+            float fractional_delay_re = Q_fract[m][0];
+            float fractional_delay_im = Q_fract[m][1];
+            float apd_re = in_re;
+            float apd_im = in_im;
+            in_re = link_delay_re * fractional_delay_re -
+                    link_delay_im * fractional_delay_im - a_re;
+            in_im = link_delay_re * fractional_delay_im +
+                    link_delay_im * fractional_delay_re - a_im;
+            ap_delay[m][n+5][0] = apd_re + ag[m] * in_re;
+            ap_delay[m][n+5][1] = apd_im + ag[m] * in_im;
+        }
+        out[n][0] = transient_gain[n] * in_re;
+        out[n][1] = transient_gain[n] * in_im;
+    }
+}
+
+static void ps_stereo_interpolate_c(float (*l)[2], float (*r)[2],
+                                    float h[2][4], float h_step[2][4],
+                                    int len)
+{
+    float h0 = h[0][0];
+    float h1 = h[0][1];
+    float h2 = h[0][2];
+    float h3 = h[0][3];
+    float hs0 = h_step[0][0];
+    float hs1 = h_step[0][1];
+    float hs2 = h_step[0][2];
+    float hs3 = h_step[0][3];
+    int n;
+
+    for (n = 0; n < len; n++) {
+        //l is s, r is d
+        float l_re = l[n][0];
+        float l_im = l[n][1];
+        float r_re = r[n][0];
+        float r_im = r[n][1];
+        h0 += hs0;
+        h1 += hs1;
+        h2 += hs2;
+        h3 += hs3;
+        l[n][0] = h0 * l_re + h2 * r_re;
+        l[n][1] = h0 * l_im + h2 * r_im;
+        r[n][0] = h1 * l_re + h3 * r_re;
+        r[n][1] = h1 * l_im + h3 * r_im;
+    }
+}
+
+static void ps_stereo_interpolate_ipdopd_c(float (*l)[2], float (*r)[2],
+                                           float h[2][4], float h_step[2][4],
+                                           int len)
+{
+    float h00  = h[0][0],      h10  = h[1][0];
+    float h01  = h[0][1],      h11  = h[1][1];
+    float h02  = h[0][2],      h12  = h[1][2];
+    float h03  = h[0][3],      h13  = h[1][3];
+    float hs00 = h_step[0][0], hs10 = h_step[1][0];
+    float hs01 = h_step[0][1], hs11 = h_step[1][1];
+    float hs02 = h_step[0][2], hs12 = h_step[1][2];
+    float hs03 = h_step[0][3], hs13 = h_step[1][3];
+    int n;
+
+    for (n = 0; n < len; n++) {
+        //l is s, r is d
+        float l_re = l[n][0];
+        float l_im = l[n][1];
+        float r_re = r[n][0];
+        float r_im = r[n][1];
+        h00 += hs00;
+        h01 += hs01;
+        h02 += hs02;
+        h03 += hs03;
+        h10 += hs10;
+        h11 += hs11;
+        h12 += hs12;
+        h13 += hs13;
+
+        l[n][0] = h00 * l_re + h02 * r_re - h10 * l_im - h12 * r_im;
+        l[n][1] = h00 * l_im + h02 * r_im + h10 * l_re + h12 * r_re;
+        r[n][0] = h01 * l_re + h03 * r_re - h11 * l_im - h13 * r_im;
+        r[n][1] = h01 * l_im + h03 * r_im + h11 * l_re + h13 * r_re;
+    }
+}
+
+av_cold void ff_psdsp_init(PSDSPContext *s)
+{
+    s->add_squares            = ps_add_squares_c;
+    s->mul_pair_single        = ps_mul_pair_single_c;
+    s->hybrid_analysis        = ps_hybrid_analysis_c;
+    s->hybrid_analysis_ileave = ps_hybrid_analysis_ileave_c;
+    s->hybrid_synthesis_deint = ps_hybrid_synthesis_deint_c;
+    s->decorrelate            = ps_decorrelate_c;
+    s->stereo_interpolate[0]  = ps_stereo_interpolate_c;
+    s->stereo_interpolate[1]  = ps_stereo_interpolate_ipdopd_c;
+
+    if (ARCH_ARM)
+        ff_psdsp_init_arm(s);
+}
diff --git a/libavcodec/aacpsdsp.h b/libavcodec/aacpsdsp.h
new file mode 100644
index 0000000000..93737d26d3
--- /dev/null
+++ b/libavcodec/aacpsdsp.h
@@ -0,0 +1,53 @@
+/*
+ * Copyright (c) 2012 Mans Rullgard
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef LIBAVCODEC_AACPSDSP_H
+#define LIBAVCODEC_AACPSDSP_H
+
+#define PS_QMF_TIME_SLOTS 32
+#define PS_AP_LINKS 3
+#define PS_MAX_AP_DELAY 5
+
+typedef struct PSDSPContext {
+    void (*add_squares)(float *dst, const float (*src)[2], int n);
+    void (*mul_pair_single)(float (*dst)[2], float (*src0)[2], float *src1,
+                            int n);
+    void (*hybrid_analysis)(float (*out)[2], float (*in)[2],
+                            const float (*filter)[8][2],
+                            int stride, int n);
+    void (*hybrid_analysis_ileave)(float (*out)[32][2], float L[2][38][64],
+                                   int i, int len);
+    void (*hybrid_synthesis_deint)(float out[2][38][64], float (*in)[32][2],
+                                   int i, int len);
+    void (*decorrelate)(float (*out)[2], float (*delay)[2],
+                        float (*ap_delay)[PS_QMF_TIME_SLOTS+PS_MAX_AP_DELAY][2],
+                        const float phi_fract[2], float (*Q_fract)[2],
+                        const float *transient_gain,
+                        float g_decay_slope,
+                        int len);
+    void (*stereo_interpolate[2])(float (*l)[2], float (*r)[2],
+                                  float h[2][4], float h_step[2][4],
+                                  int len);
+} PSDSPContext;
+
+void ff_psdsp_init(PSDSPContext *s);
+void ff_psdsp_init_arm(PSDSPContext *s);
+
+#endif /* LIBAVCODEC_AACPSDSP_H */
diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile
index 3d002ccdcf..d2bdd50daa 100644
--- a/libavcodec/arm/Makefile
+++ b/libavcodec/arm/Makefile
@@ -1,7 +1,8 @@
 OBJS-$(CONFIG_AC3DSP)                  += arm/ac3dsp_init_arm.o         \
                                           arm/ac3dsp_arm.o
 
-OBJS-$(CONFIG_AAC_DECODER)             += arm/sbrdsp_init_arm.o
+OBJS-$(CONFIG_AAC_DECODER)             += arm/sbrdsp_init_arm.o         \
+                                          arm/aacpsdsp_init_arm.o
 
 OBJS-$(CONFIG_DCA_DECODER)             += arm/dcadsp_init_arm.o         \
 
@@ -59,7 +60,8 @@ NEON-OBJS-$(CONFIG_H264PRED)           += arm/h264pred_neon.o           \
 
 NEON-OBJS-$(CONFIG_AC3DSP)             += arm/ac3dsp_neon.o
 
-NEON-OBJS-$(CONFIG_AAC_DECODER)        += arm/sbrdsp_neon.o
+NEON-OBJS-$(CONFIG_AAC_DECODER)        += arm/sbrdsp_neon.o             \
+                                          arm/aacpsdsp_neon.o
 
 NEON-OBJS-$(CONFIG_DCA_DECODER)        += arm/dcadsp_neon.o             \
                                           arm/synth_filter_neon.o       \
diff --git a/libavcodec/arm/aacpsdsp_init_arm.c b/libavcodec/arm/aacpsdsp_init_arm.c
new file mode 100644
index 0000000000..6326376004
--- /dev/null
+++ b/libavcodec/arm/aacpsdsp_init_arm.c
@@ -0,0 +1,57 @@
+/*
+ * Copyright (c) 2012 Mans Rullgard
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+
+#include "libavutil/arm/cpu.h"
+#include "libavutil/attributes.h"
+#include "libavcodec/aacpsdsp.h"
+
+void ff_ps_add_squares_neon(float *dst, const float (*src)[2], int n);
+void ff_ps_mul_pair_single_neon(float (*dst)[2], float (*src0)[2],
+                                float *src1, int n);
+void ff_ps_hybrid_analysis_neon(float (*out)[2], float (*in)[2],
+                                const float (*filter)[8][2],
+                                int stride, int n);
+void ff_ps_hybrid_analysis_ileave_neon(float (*out)[32][2], float L[2][38][64],
+                                       int i, int len);
+void ff_ps_hybrid_synthesis_deint_neon(float out[2][38][64], float (*in)[32][2],
+                                       int i, int len);
+void ff_ps_decorrelate_neon(float (*out)[2], float (*delay)[2],
+                            float (*ap_delay)[PS_QMF_TIME_SLOTS+PS_MAX_AP_DELAY][2],
+                            const float phi_fract[2], float (*Q_fract)[2],
+                            const float *transient_gain, float g_decay_slope,
+                            int len);
+void ff_ps_stereo_interpolate_neon(float (*l)[2], float (*r)[2],
+                                   float h[2][4], float h_step[2][4],
+                                   int len);
+
+av_cold void ff_psdsp_init_arm(PSDSPContext *s)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (have_neon(cpu_flags)) {
+        s->add_squares            = ff_ps_add_squares_neon;
+        s->mul_pair_single        = ff_ps_mul_pair_single_neon;
+        s->hybrid_synthesis_deint = ff_ps_hybrid_synthesis_deint_neon;
+        s->hybrid_analysis        = ff_ps_hybrid_analysis_neon;
+        s->stereo_interpolate[0]  = ff_ps_stereo_interpolate_neon;
+    }
+}
diff --git a/libavcodec/arm/aacpsdsp_neon.S b/libavcodec/arm/aacpsdsp_neon.S
new file mode 100644
index 0000000000..e75760a2d6
--- /dev/null
+++ b/libavcodec/arm/aacpsdsp_neon.S
@@ -0,0 +1,272 @@
+/*
+ * Copyright (c) 2012 Mans Rullgard
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "asm.S"
+
+function ff_ps_add_squares_neon, export=1
+        mov             r3,  r0
+        sub             r2,  r2,  #4
+        vld1.32         {q0},     [r1,:128]!
+        vmul.f32        q0,  q0,  q0
+        vld1.32         {q2},     [r1,:128]!
+        vmul.f32        q2,  q2,  q2
+        vld1.32         {q1},     [r0,:128]!
+1:
+        vpadd.f32       d6,  d0,  d1
+        vld1.32         {q0},     [r1,:128]!
+        vpadd.f32       d7,  d4,  d5
+        vmul.f32        q0,  q0,  q0
+        vld1.32         {q2},     [r1,:128]!
+        vadd.f32        q3,  q1,  q3
+        vld1.32         {q1},     [r0,:128]!
+        vmul.f32        q2,  q2,  q2
+        vst1.32         {q3},     [r3,:128]!
+        subs            r2,  r2,  #4
+        bgt             1b
+        vpadd.f32       d6,  d0,  d1
+        vpadd.f32       d7,  d4,  d5
+        vadd.f32        q1,  q1,  q3
+        vst1.32         {q1},     [r3,:128]!
+        bx              lr
+endfunc
+
+function ff_ps_mul_pair_single_neon, export=1
+        sub             r3,  r3,  #4
+        tst             r1,  #8
+        bne             2f
+        vld1.32         {q0},     [r1,:128]!
+1:
+        vld1.32         {q3},     [r2,:128]!
+        vmul.f32        d4,  d0,  d6[0]
+        vmul.f32        d5,  d1,  d6[1]
+        vld1.32         {q1},     [r1,:128]!
+        vmul.f32        d6,  d2,  d7[0]
+        vmul.f32        d7,  d3,  d7[1]
+        vld1.32         {q0},     [r1,:128]!
+        vst1.32         {q2,q3},  [r0,:128]!
+        subs            r3,  r3,  #4
+        bgt             1b
+        vld1.32         {q3},     [r2,:128]!
+        vmul.f32        d4,  d0,  d6[0]
+        vmul.f32        d5,  d1,  d6[1]
+        vld1.32         {q1},     [r1,:128]!
+        vmul.f32        d6,  d2,  d7[0]
+        vmul.f32        d7,  d3,  d7[1]
+        vst1.32         {q2,q3},  [r0,:128]!
+        bx              lr
+2:
+        vld1.32         {d0},     [r1,:64]!
+        vld1.32         {d1,d2},  [r1,:128]!
+1:
+        vld1.32         {q3},     [r2,:128]!
+        vmul.f32        d4,  d0,  d6[0]
+        vmul.f32        d5,  d1,  d6[1]
+        vld1.32         {d0,d1},  [r1,:128]!
+        vmul.f32        d6,  d2,  d7[0]
+        vmul.f32        d7,  d0,  d7[1]
+        vmov            d0,  d1
+        vld1.32         {d1,d2},  [r1,:128]!
+        vst1.32         {q2,q3},  [r0,:128]!
+        subs            r3,  r3,  #4
+        bgt             1b
+        vld1.32         {q3},     [r2,:128]!
+        vmul.f32        d4,  d0,  d6[0]
+        vmul.f32        d5,  d1,  d6[1]
+        vld1.32         {d0},     [r1,:64]!
+        vmul.f32        d6,  d2,  d7[0]
+        vmul.f32        d7,  d0,  d7[1]
+        vst1.32         {q2,q3},  [r0,:128]!
+        bx              lr
+endfunc
+
+function ff_ps_hybrid_synthesis_deint_neon, export=1
+        push            {r4-r8,lr}
+        add             r0,  r0,  r2,  lsl #2
+        add             r1,  r1,  r2,  lsl #5+1+2
+        rsb             r2,  r2,  #64
+        mov             r5,  #64*4
+        mov             lr,  r0
+        add             r4,  r0,  #38*64*4
+        mov             r12, r3
+2:
+        vld1.32         {d0,d1},  [r1,:128]!
+        vst1.32         {d0[0]},  [lr,:32], r5
+        vst1.32         {d0[1]},  [r4,:32], r5
+        vst1.32         {d1[0]},  [lr,:32], r5
+        vst1.32         {d1[1]},  [r4,:32], r5
+        subs            r12, r12, #2
+        bgt             2b
+        add             r0,  r0,  #4
+        sub             r2,  r2,  #1
+        tst             r2,  #2
+        bne             6f
+1:
+        mov             lr,  r0
+        add             r4,  r0,  #38*64*4
+        add             r6,  r1,  #  32*2*4
+        add             r7,  r1,  #2*32*2*4
+        add             r8,  r1,  #3*32*2*4
+        mov             r12, r3
+2:
+        vld1.32         {d0,d1},  [r1,:128]!
+        vld1.32         {d2,d3},  [r6,:128]!
+        vld1.32         {d4,d5},  [r7,:128]!
+        vld1.32         {d6,d7},  [r8,:128]!
+        vst4.32         {d0[0],d2[0],d4[0],d6[0]}, [lr,:128], r5
+        vst4.32         {d0[1],d2[1],d4[1],d6[1]}, [r4,:128], r5
+        vst4.32         {d1[0],d3[0],d5[0],d7[0]}, [lr,:128], r5
+        vst4.32         {d1[1],d3[1],d5[1],d7[1]}, [r4,:128], r5
+        subs            r12, r12, #2
+        bgt             2b
+        add             r0,  r0,  #16
+        add             r1,  r1,  #3*32*2*4
+        subs            r2,  r2,  #4
+        bgt             1b
+        pop             {r4-r8,pc}
+6:
+        mov             lr,  r0
+        add             r4,  r0,  #38*64*4
+        add             r6,  r1,  #32*2*4
+        mov             r12, r3
+2:
+        vld1.32         {d0,d1},  [r1,:128]!
+        vld1.32         {d2,d3},  [r6,:128]!
+        vst2.32         {d0[0],d2[0]}, [lr,:64], r5
+        vst2.32         {d0[1],d2[1]}, [r4,:64], r5
+        vst2.32         {d1[0],d3[0]}, [lr,:64], r5
+        vst2.32         {d1[1],d3[1]}, [r4,:64], r5
+        subs            r12, r12, #2
+        bgt             2b
+        add             r0,  r0,  #8
+        add             r1,  r1,  #32*2*4
+        sub             r2,  r2,  #2
+        b               1b
+endfunc
+
+function ff_ps_hybrid_analysis_neon, export=1
+        vldm            r1,  {d19-d31}
+        ldr             r12, [sp]
+        lsl             r3,  r3,  #3
+        vadd.f32        d16, d19, d31
+        vadd.f32        d17, d20, d30
+        vsub.f32        d18, d19, d31
+        vsub.f32        d19, d20, d30
+        vsub.f32        d0,  d21, d29
+        vsub.f32        d1,  d22, d28
+        vadd.f32        d2,  d21, d29
+        vadd.f32        d3,  d22, d28
+        vadd.f32        d20, d23, d27
+        vadd.f32        d21, d24, d26
+        vsub.f32        d22, d23, d27
+        vsub.f32        d23, d24, d26
+        vmov.i32        d6,  #1<<31
+        vmov.i32        d7,  #0
+        vmov.f32        q14, #0.0
+        vmov.f32        q15, #0.0
+        vtrn.32         d6,  d7
+        vrev64.32       q9,  q9
+        vrev64.32       q0,  q0
+        vrev64.32       q11, q11
+        veor            q9,  q9,  q3
+        veor            q0,  q0,  q3
+        veor            q11, q11, q3
+        vld1.32         {q13},    [r2,:128]!
+        vtrn.32         q8,  q9
+        vtrn.32         q1,  q0
+        vtrn.32         q10, q11
+        sub             r12, r12, #1
+        vmla.f32        q14, q8,  q13
+        vld1.32         {q2},     [r2,:128]!
+        vmla.f32        q15, q9,  q13
+1:
+        vmla.f32        q14, q1,  q2
+        vld1.32         {q13},    [r2,:128]!
+        vmla.f32        q15, q0,  q2
+        vmla.f32        q14, q10, q13
+        vld1.32         {q2},     [r2,:128]!
+        vmla.f32        q15, q11, q13
+        vld1.32         {q13},    [r2,:128]!
+        vadd.f32        d6,  d28, d29
+        vadd.f32        d7,  d30, d31
+        vmov.f32        q14, #0.0
+        vmov.f32        q15, #0.0
+        vmla.f32        q14, q8,  q13
+        vpadd.f32       d6,  d6,  d7
+        vmla.f32        q15, q9,  q13
+        vmla.f32        d6,  d25, d4[0]
+        vld1.32         {q2},     [r2,:128]!
+        vst1.32         {d6},     [r0,:64], r3
+        subs            r12, r12, #1
+        bgt             1b
+        vmla.f32        q14, q1,  q2
+        vld1.32         {q13},    [r2,:128]!
+        vmla.f32        q15, q0,  q2
+        vmla.f32        q14, q10, q13
+        vld1.32         {q2},     [r2,:128]!
+        vmla.f32        q15, q11, q13
+        vadd.f32        d6,  d28, d29
+        vadd.f32        d7,  d30, d31
+        vpadd.f32       d6,  d6,  d7
+        vmla.f32        d6,  d25, d4[0]
+        vst1.32         {d6},     [r0,:64], r3
+        bx              lr
+endfunc
+
+function ff_ps_stereo_interpolate_neon, export=1
+        vld1.32         {q0},     [r2]
+        vld1.32         {q14},    [r3]
+        vadd.f32        q15, q14, q14
+        mov             r2,  r0
+        mov             r3,  r1
+        ldr             r12, [sp]
+        vadd.f32        q1,  q0,  q14
+        vadd.f32        q0,  q0,  q15
+        vld1.32         {q2},     [r0,:64]!
+        vld1.32         {q3},     [r1,:64]!
+        subs            r12, r12, #1
+        beq             2f
+1:
+        vmul.f32        d16, d4,  d2[0]
+        vmul.f32        d17, d5,  d0[0]
+        vmul.f32        d18, d4,  d2[1]
+        vmul.f32        d19, d5,  d0[1]
+        vmla.f32        d16, d6,  d3[0]
+        vmla.f32        d17, d7,  d1[0]
+        vmla.f32        d18, d6,  d3[1]
+        vmla.f32        d19, d7,  d1[1]
+        vadd.f32        q1,  q1,  q15
+        vadd.f32        q0,  q0,  q15
+        vld1.32         {q2},     [r0,:64]!
+        vld1.32         {q3},     [r1,:64]!
+        vst1.32         {q8},     [r2,:64]!
+        vst1.32         {q9},     [r3,:64]!
+        subs            r12, r12, #2
+        bgt             1b
+        it              lt
+        bxlt            lr
+2:
+        vmul.f32        d16, d4,  d2[0]
+        vmul.f32        d18, d4,  d2[1]
+        vmla.f32        d16, d6,  d3[0]
+        vmla.f32        d18, d6,  d3[1]
+        vst1.32         {d16},    [r2,:64]!
+        vst1.32         {d18},    [r3,:64]!
+        bx              lr
+endfunc
diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h
index b1c6d474b1..db17917b8a 100644
--- a/libavcodec/avcodec.h
+++ b/libavcodec/avcodec.h
@@ -2722,7 +2722,7 @@ typedef struct AVCodecContext {
 
     /**
      * Set by the client if its custom get_buffer() callback can be called
-     * from another thread, which allows faster multithreaded decoding.
+     * synchronously from another thread, which allows faster multithreaded decoding.
      * draw_horiz_band() will be called from other threads regardless of this setting.
      * Ignored if the default get_buffer() is used.
      * - encoding: Set by user.
diff --git a/libavcodec/dfa.c b/libavcodec/dfa.c
index 9664b47d4f..2f8e4f122d 100644
--- a/libavcodec/dfa.c
+++ b/libavcodec/dfa.c
@@ -49,7 +49,7 @@ static int decode_copy(GetByteContext *gb, uint8_t *frame, int width, int height
     const int size = width * height;
 
     if (bytestream2_get_buffer(gb, frame, size) != size)
-        return -1;
+        return AVERROR_INVALIDDATA;
     return 0;
 }
 
@@ -66,23 +66,23 @@ static int decode_tsw1(GetByteContext *gb, uint8_t *frame, int width, int height
     if (segments == 0 && offset == frame_end - frame)
         return 0; // skip frame
     if (frame_end - frame <= offset)
-        return -1;
+        return AVERROR_INVALIDDATA;
     frame += offset;
     while (segments--) {
         if (bytestream2_get_bytes_left(gb) < 2)
-            return -1;
+            return AVERROR_INVALIDDATA;
         if (mask == 0x10000) {
             bitbuf = bytestream2_get_le16u(gb);
             mask = 1;
         }
         if (frame_end - frame < 2)
-            return -1;
+            return AVERROR_INVALIDDATA;
         if (bitbuf & mask) {
             v = bytestream2_get_le16(gb);
             offset = (v & 0x1FFF) << 1;
             count = ((v >> 13) + 2) << 1;
             if (frame - frame_start < offset || frame_end - frame < count)
-                return -1;
+                return AVERROR_INVALIDDATA;
             av_memcpy_backptr(frame, offset, count);
             frame += count;
         } else {
@@ -105,19 +105,19 @@ static int decode_dsw1(GetByteContext *gb, uint8_t *frame, int width, int height
     segments = bytestream2_get_le16(gb);
     while (segments--) {
         if (bytestream2_get_bytes_left(gb) < 2)
-            return -1;
+            return AVERROR_INVALIDDATA;
         if (mask == 0x10000) {
             bitbuf = bytestream2_get_le16u(gb);
             mask = 1;
         }
         if (frame_end - frame < 2)
-            return -1;
+            return AVERROR_INVALIDDATA;
         if (bitbuf & mask) {
             v = bytestream2_get_le16(gb);
             offset = (v & 0x1FFF) << 1;
             count = ((v >> 13) + 2) << 1;
             if (frame - frame_start < offset || frame_end - frame < count)
-                return -1;
+                return AVERROR_INVALIDDATA;
             // can't use av_memcpy_backptr() since it can overwrite following pixels
             for (v = 0; v < count; v++)
                 frame[v] = frame[v - offset];
@@ -144,19 +144,19 @@ static int decode_dds1(GetByteContext *gb, uint8_t *frame, int width, int height
     segments = bytestream2_get_le16(gb);
     while (segments--) {
         if (bytestream2_get_bytes_left(gb) < 2)
-            return -1;
+            return AVERROR_INVALIDDATA;
         if (mask == 0x10000) {
             bitbuf = bytestream2_get_le16u(gb);
             mask = 1;
         }
         if (frame_end - frame < width + 2)
-            return -1;
+            return AVERROR_INVALIDDATA;
         if (bitbuf & mask) {
             v = bytestream2_get_le16(gb);
             offset = (v & 0x1FFF) << 2;
             count = ((v >> 13) + 2) << 1;
             if (frame - frame_start < offset || frame_end - frame < count*2 + width)
-                return -1;
+                return AVERROR_INVALIDDATA;
             for (i = 0; i < count; i++) {
                 frame[0] = frame[1] =
                 frame[width] = frame[width + 1] = frame[-offset];
@@ -166,6 +166,8 @@ static int decode_dds1(GetByteContext *gb, uint8_t *frame, int width, int height
         } else if (bitbuf & (mask << 1)) {
             frame += bytestream2_get_le16(gb) * 2;
         } else {
+            if (frame_end - frame < width + 2)
+                return AVERROR_INVALIDDATA;
             frame[0] = frame[1] =
             frame[width] = frame[width + 1] =  bytestream2_get_byte(gb);
             frame += 2;
@@ -186,32 +188,32 @@ static int decode_bdlt(GetByteContext *gb, uint8_t *frame, int width, int height
 
     count = bytestream2_get_le16(gb);
     if (count >= height)
-        return -1;
+        return AVERROR_INVALIDDATA;
     frame += width * count;
     lines = bytestream2_get_le16(gb);
     if (count + lines > height)
-        return -1;
+        return AVERROR_INVALIDDATA;
 
     while (lines--) {
         if (bytestream2_get_bytes_left(gb) < 1)
-            return -1;
+            return AVERROR_INVALIDDATA;
         line_ptr = frame;
         frame += width;
         segments = bytestream2_get_byteu(gb);
         while (segments--) {
             if (frame - line_ptr <= bytestream2_peek_byte(gb))
-                return -1;
+                return AVERROR_INVALIDDATA;
             line_ptr += bytestream2_get_byte(gb);
             count = (int8_t)bytestream2_get_byte(gb);
             if (count >= 0) {
                 if (frame - line_ptr < count)
-                    return -1;
+                    return AVERROR_INVALIDDATA;
                 if (bytestream2_get_buffer(gb, line_ptr, count) != count)
-                    return -1;
+                    return AVERROR_INVALIDDATA;
             } else {
                 count = -count;
                 if (frame - line_ptr < count)
-                    return -1;
+                    return AVERROR_INVALIDDATA;
                 memset(line_ptr, bytestream2_get_byte(gb), count);
             }
             line_ptr += count;
@@ -226,20 +228,23 @@ static int decode_wdlt(GetByteContext *gb, uint8_t *frame, int width, int height
     const uint8_t *frame_end   = frame + width * height;
     uint8_t *line_ptr;
     int count, i, v, lines, segments;
+    int y = 0;
 
     lines = bytestream2_get_le16(gb);
     if (lines > height)
-        return -1;
+        return AVERROR_INVALIDDATA;
 
     while (lines--) {
         if (bytestream2_get_bytes_left(gb) < 2)
-            return -1;
+            return AVERROR_INVALIDDATA;
         segments = bytestream2_get_le16u(gb);
         while ((segments & 0xC000) == 0xC000) {
+            unsigned skip_lines = -(int16_t)segments;
             unsigned delta = -((int16_t)segments * width);
-            if (frame_end - frame <= delta)
-                return -1;
+            if (frame_end - frame <= delta || y + lines + skip_lines > height)
+                return AVERROR_INVALIDDATA;
             frame    += delta;
+            y        += skip_lines;
             segments = bytestream2_get_le16(gb);
         }
         if (frame_end <= frame)
@@ -250,21 +255,22 @@ static int decode_wdlt(GetByteContext *gb, uint8_t *frame, int width, int height
         }
         line_ptr = frame;
         frame += width;
+        y++;
         while (segments--) {
             if (frame - line_ptr <= bytestream2_peek_byte(gb))
-                return -1;
+                return AVERROR_INVALIDDATA;
             line_ptr += bytestream2_get_byte(gb);
             count = (int8_t)bytestream2_get_byte(gb);
             if (count >= 0) {
                 if (frame - line_ptr < count * 2)
-                    return -1;
+                    return AVERROR_INVALIDDATA;
                 if (bytestream2_get_buffer(gb, line_ptr, count * 2) != count * 2)
-                    return -1;
+                    return AVERROR_INVALIDDATA;
                 line_ptr += count * 2;
             } else {
                 count = -count;
                 if (frame - line_ptr < count * 2)
-                    return -1;
+                    return AVERROR_INVALIDDATA;
                 v = bytestream2_get_le16(gb);
                 for (i = 0; i < count; i++)
                     bytestream_put_le16(&line_ptr, v);
@@ -277,7 +283,7 @@ static int decode_wdlt(GetByteContext *gb, uint8_t *frame, int width, int height
 
 static int decode_unk6(GetByteContext *gb, uint8_t *frame, int width, int height)
 {
-    return -1;
+    return AVERROR_PATCHWELCOME;
 }
 
 static int decode_blck(GetByteContext *gb, uint8_t *frame, int width, int height)
@@ -336,7 +342,7 @@ static int dfa_decode_frame(AVCodecContext *avctx,
             if (decoder[chunk_type - 2](&gb, s->frame_buf, avctx->width, avctx->height)) {
                 av_log(avctx, AV_LOG_ERROR, "Error decoding %s chunk\n",
                        chunk_name[chunk_type - 2]);
-                return -1;
+                return AVERROR_INVALIDDATA;
             }
         } else {
             av_log(avctx, AV_LOG_WARNING, "Ignoring unknown chunk type %d\n",
diff --git a/libavcodec/dvdec.c b/libavcodec/dvdec.c
index 4e09d0dd9a..9178524251 100644
--- a/libavcodec/dvdec.c
+++ b/libavcodec/dvdec.c
@@ -313,7 +313,7 @@ static int dvvideo_decode_frame(AVCodecContext *avctx,
                                  void *data, int *data_size,
                                  AVPacket *avpkt)
 {
-    const uint8_t *buf = avpkt->data;
+    uint8_t *buf = avpkt->data;
     int buf_size = avpkt->size;
     DVVideoContext *s = avctx->priv_data;
     const uint8_t* vsc_pack;
diff --git a/libavcodec/eatgv.c b/libavcodec/eatgv.c
index 01d0489c4c..3b19d703f9 100644
--- a/libavcodec/eatgv.c
+++ b/libavcodec/eatgv.c
@@ -138,7 +138,6 @@ static int unpack(const uint8_t *src, const uint8_t *src_end, unsigned char *dst
  * @return 0 on success, -1 on critical buffer underflow
  */
 static int tgv_decode_inter(TgvContext * s, const uint8_t *buf, const uint8_t *buf_end){
-    unsigned last_frame_size = s->avctx->height*s->last_frame.linesize[0];
     int num_mvs;
     int num_blocks_raw;
     int num_blocks_packed;
@@ -158,7 +157,8 @@ static int tgv_decode_inter(TgvContext * s, const uint8_t *buf, const uint8_t *b
     buf += 12;
 
     if (vector_bits > MIN_CACHE_BITS || !vector_bits) {
-        av_log(s->avctx, AV_LOG_ERROR, "vector_bits %d invalid\n", vector_bits);
+        av_log(s->avctx, AV_LOG_ERROR,
+               "Invalid value for motion vector bits: %d\n", vector_bits);
         return AVERROR_INVALIDDATA;
     }
 
@@ -212,14 +212,17 @@ static int tgv_decode_inter(TgvContext * s, const uint8_t *buf, const uint8_t *b
         int src_stride;
 
         if (vector < num_mvs) {
-            unsigned offset =
-                (y*4 + s->mv_codebook[vector][1])*s->last_frame.linesize[0] +
-                 x*4 + s->mv_codebook[vector][0];
+            int mx = x * 4 + s->mv_codebook[vector][0];
+            int my = y * 4 + s->mv_codebook[vector][1];
 
-            src_stride = s->last_frame.linesize[0];
-            if (offset >= last_frame_size - (3*src_stride+3))
+            if (   mx < 0 || mx + 4 > s->avctx->width
+                || my < 0 || my + 4 > s->avctx->height) {
+                av_log(s->avctx, AV_LOG_ERROR, "MV %d %d out of picture\n", mx, my);
                 continue;
-            src = s->last_frame.data[0] + offset;
+            }
+
+            src = s->last_frame.data[0] + mx + my * s->last_frame.linesize[0];
+            src_stride = s->last_frame.linesize[0];
         }else{
             int offset = vector - num_mvs;
             if (offset<num_blocks_raw)
diff --git a/libavcodec/lagarith.c b/libavcodec/lagarith.c
index 3df5f2600a..c2a8ed18e4 100644
--- a/libavcodec/lagarith.c
+++ b/libavcodec/lagarith.c
@@ -467,7 +467,7 @@ static int lag_decode_frame(AVCodecContext *avctx,
     AVFrame *const p = &l->picture;
     uint8_t frametype = 0;
     uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
-    int offs[4];
+    uint32_t offs[4];
     uint8_t *srcs[4], *dst;
     int i, j, planes = 3;
 
diff --git a/libavcodec/wmalosslessdec.c b/libavcodec/wmalosslessdec.c
index 0b386ceedc..dabd093bf4 100644
--- a/libavcodec/wmalosslessdec.c
+++ b/libavcodec/wmalosslessdec.c
@@ -22,6 +22,7 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+#include "libavutil/attributes.h"
 #include "avcodec.h"
 #include "internal.h"
 #include "get_bits.h"
@@ -1045,7 +1046,7 @@ static int decode_frame(WmallDecodeCtx *s)
     /* no idea what these are for, might be the number of samples
        that need to be skipped at the beginning or end of a stream */
     if (get_bits1(gb)) {
-        int skip;
+        int av_unused skip;
 
         /* usually true for the first frame */
         if (get_bits1(gb)) {