aacps: move some loops to function pointers

Signed-off-by: Mans Rullgard <mans@mansr.com>

5 files changed, 331 insertions, 161 deletions

diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 657f6a0708..367dc1d646 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -53,7 +53,7 @@ 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 6c9dcf2f1b..7578034f82 100644
--- a/libavcodec/aacps.c
+++ b/libavcodec/aacps.c
@@ -304,26 +304,14 @@ 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)[7][2], int len)
 {
-    int i, j, ssb;
+    int i;
     int N = 8;
     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 +327,18 @@ 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)[7][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 +348,17 @@ 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 +366,8 @@ 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 +391,7 @@ 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 +403,7 @@ 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);
     }
 }
 
@@ -661,10 +620,6 @@ 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 };
 
     if (is34 != ps->is34bands_old) {
         memset(ps->peak_decay_nrg,         0, sizeof(ps->peak_decay_nrg));
@@ -674,11 +629,9 @@ 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 +659,31 @@ 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 +727,7 @@ 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;
@@ -909,70 +839,44 @@ 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);
         }
     }
 }
@@ -989,11 +893,11 @@ 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 +945,5 @@ 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 124fbee221..d5c355d5bd 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"
 
@@ -72,6 +73,7 @@ typedef struct {
     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/aacpsdsp.c b/libavcodec/aacpsdsp.c
new file mode 100644
index 0000000000..3d9eb61fd6
--- /dev/null
+++ b/libavcodec/aacpsdsp.c
@@ -0,0 +1,211 @@
+/*
+ * 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)[7][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;
+}
diff --git a/libavcodec/aacpsdsp.h b/libavcodec/aacpsdsp.h
new file mode 100644
index 0000000000..909d3418f2
--- /dev/null
+++ b/libavcodec/aacpsdsp.h
@@ -0,0 +1,52 @@
+/*
+ * 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)[7][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);
+
+#endif /* LIBAVCODEC_AACPSDSP_H */