update to the 2.0 release of faad, patch by adland

git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@12528 b3059339-0415-0410-9bf9-f77b7e298cf2

1 files changed, 779 insertions, 116 deletions

diff --git a/libfaad2/specrec.c b/libfaad2/specrec.c
index 177050bc9b..c5abd984a7 100644
--- a/libfaad2/specrec.c
+++ b/libfaad2/specrec.c
@@ -1,6 +1,6 @@
 /*
 ** FAAD2 - Freeware Advanced Audio (AAC) Decoder including SBR decoding
-** Copyright (C) 2003 M. Bakker, Ahead Software AG, http://www.nero.com
+** Copyright (C) 2003-2004 M. Bakker, Ahead Software AG, http://www.nero.com
 **  
 ** This program is free software; you can redistribute it and/or modify
 ** it under the terms of the GNU General Public License as published by
@@ -22,7 +22,7 @@
 ** Commercial non-GPL licensing of this software is possible.
 ** For more info contact Ahead Software through Mpeg4AAClicense@nero.com.
 **
-** $Id: specrec.c,v 1.27 2003/09/30 12:43:05 menno Exp $
+** $Id: specrec.c,v 1.2 2003/10/03 22:22:27 alex Exp $
 **/
 
 /*
@@ -36,49 +36,67 @@
 #include "structs.h"
 
 #include <string.h>
+#include <stdlib.h>
 #include "specrec.h"
 #include "syntax.h"
 #include "iq_table.h"
+#include "ms.h"
+#include "is.h"
+#include "pns.h"
+#include "tns.h"
+#include "drc.h"
+#include "lt_predict.h"
+#include "ic_predict.h"
+#ifdef SSR_DEC
+#include "ssr.h"
+#include "ssr_fb.h"
+#endif
+
+
+/* static function declarations */
+static void quant_to_spec(ic_stream *ics, real_t *spec_data, uint16_t frame_len);
+static uint8_t inverse_quantization(real_t *x_invquant, const int16_t *x_quant, const uint16_t frame_len);
+
 
 #ifdef LD_DEC
-static uint8_t num_swb_512_window[] =
+ALIGN static const uint8_t num_swb_512_window[] =
 {
     0, 0, 0, 36, 36, 37, 31, 31, 0, 0, 0, 0
 };
-static uint8_t num_swb_480_window[] =
+ALIGN static const uint8_t num_swb_480_window[] =
 {
     0, 0, 0, 35, 35, 37, 30, 30, 0, 0, 0, 0
 };
 #endif
 
-static uint8_t num_swb_960_window[] =
+ALIGN static const uint8_t num_swb_960_window[] =
 {
     40, 40, 45, 49, 49, 49, 46, 46, 42, 42, 42, 40
 };
 
-static uint8_t num_swb_1024_window[] =
+ALIGN static const uint8_t num_swb_1024_window[] =
 {
     41, 41, 47, 49, 49, 51, 47, 47, 43, 43, 43, 40
 };
 
-static uint8_t num_swb_128_window[] =
+ALIGN static const uint8_t num_swb_128_window[] =
 {
     12, 12, 12, 14, 14, 14, 15, 15, 15, 15, 15, 15
 };
 
-static uint16_t swb_offset_1024_96[] =
+ALIGN static const uint16_t swb_offset_1024_96[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56,
     64, 72, 80, 88, 96, 108, 120, 132, 144, 156, 172, 188, 212, 240,
     276, 320, 384, 448, 512, 576, 640, 704, 768, 832, 896, 960, 1024
 };
 
-static uint16_t swb_offset_128_96[] =
+ALIGN static const uint16_t swb_offset_128_96[] =
 {
     0, 4, 8, 12, 16, 20, 24, 32, 40, 48, 64, 92, 128
 };
 
-static uint16_t swb_offset_1024_64[] =
+ALIGN static const uint16_t swb_offset_1024_64[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56,
     64, 72, 80, 88, 100, 112, 124, 140, 156, 172, 192, 216, 240, 268,
@@ -86,13 +104,12 @@ static uint16_t swb_offset_1024_64[] =
     864, 904, 944, 984, 1024
 };
 
-static uint16_t swb_offset_128_64[] =
+ALIGN static const uint16_t swb_offset_128_64[] =
 {
     0, 4, 8, 12, 16, 20, 24, 32, 40, 48, 64, 92, 128
 };
 
-
-static uint16_t swb_offset_1024_48[] =
+ALIGN static const uint16_t swb_offset_1024_48[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 48, 56, 64, 72,
     80, 88, 96, 108, 120, 132, 144, 160, 176, 196, 216, 240, 264, 292,
@@ -101,14 +118,14 @@ static uint16_t swb_offset_1024_48[] =
 };
 
 #ifdef LD_DEC
-static uint16_t swb_offset_512_48[] =
+ALIGN static const uint16_t swb_offset_512_48[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60, 68, 76, 84,
     92, 100, 112, 124, 136, 148, 164, 184, 208, 236, 268, 300, 332, 364, 396,
     428, 460, 512
 };
 
-static uint16_t swb_offset_480_48[] =
+ALIGN static const uint16_t swb_offset_480_48[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 64, 72 ,80 ,88,
     96, 108, 120, 132, 144, 156, 172, 188, 212, 240, 272, 304, 336, 368, 400,
@@ -116,12 +133,12 @@ static uint16_t swb_offset_480_48[] =
 };
 #endif
 
-static uint16_t swb_offset_128_48[] =
+ALIGN static const uint16_t swb_offset_128_48[] =
 {
     0, 4, 8, 12, 16, 20, 28, 36, 44, 56, 68, 80, 96, 112, 128
 };
 
-static uint16_t swb_offset_1024_32[] =
+ALIGN static const uint16_t swb_offset_1024_32[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 48, 56, 64, 72,
     80, 88, 96, 108, 120, 132, 144, 160, 176, 196, 216, 240, 264, 292,
@@ -130,14 +147,14 @@ static uint16_t swb_offset_1024_32[] =
 };
 
 #ifdef LD_DEC
-static uint16_t swb_offset_512_32[] =
+ALIGN static const uint16_t swb_offset_512_32[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 64, 72, 80,
     88, 96, 108, 120, 132, 144, 160, 176, 192, 212, 236, 260, 288, 320, 352,
     384, 416, 448, 480, 512
 };
 
-static uint16_t swb_offset_480_32[] =
+ALIGN static const uint16_t swb_offset_480_32[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60, 64, 72, 80,
     88, 96, 104, 112, 124, 136, 148, 164, 180, 200, 224, 256, 288, 320, 352,
@@ -145,7 +162,7 @@ static uint16_t swb_offset_480_32[] =
 };
 #endif
 
-static uint16_t swb_offset_1024_24[] =
+ALIGN static const uint16_t swb_offset_1024_24[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 52, 60, 68,
     76, 84, 92, 100, 108, 116, 124, 136, 148, 160, 172, 188, 204, 220,
@@ -154,50 +171,50 @@ static uint16_t swb_offset_1024_24[] =
 };
 
 #ifdef LD_DEC
-static uint16_t swb_offset_512_24[] =
+ALIGN static const uint16_t swb_offset_512_24[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 52, 60, 68,
     80, 92, 104, 120, 140, 164, 192, 224, 256, 288, 320, 352, 384, 416,
     448, 480, 512
 };
 
-static uint16_t swb_offset_480_24[] =
+ALIGN static const uint16_t swb_offset_480_24[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 52, 60, 68, 80, 92, 104, 120,
     140, 164, 192, 224, 256, 288, 320, 352, 384, 416, 448, 480
 };
 #endif
 
-static uint16_t swb_offset_128_24[] =
+ALIGN static const uint16_t swb_offset_128_24[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 36, 44, 52, 64, 76, 92, 108, 128
 };
 
-static uint16_t swb_offset_1024_16[] =
+ALIGN static const uint16_t swb_offset_1024_16[] =
 {
     0, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 100, 112, 124,
     136, 148, 160, 172, 184, 196, 212, 228, 244, 260, 280, 300, 320, 344,
     368, 396, 424, 456, 492, 532, 572, 616, 664, 716, 772, 832, 896, 960, 1024
 };
 
-static uint16_t swb_offset_128_16[] =
+ALIGN static const uint16_t swb_offset_128_16[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 32, 40, 48, 60, 72, 88, 108, 128
 };
 
-static uint16_t swb_offset_1024_8[] =
+ALIGN static const uint16_t swb_offset_1024_8[] =
 {
     0, 12, 24, 36, 48, 60, 72, 84, 96, 108, 120, 132, 144, 156, 172,
     188, 204, 220, 236, 252, 268, 288, 308, 328, 348, 372, 396, 420, 448,
     476, 508, 544, 580, 620, 664, 712, 764, 820, 880, 944, 1024
 };
 
-static uint16_t swb_offset_128_8[] =
+ALIGN static const uint16_t swb_offset_128_8[] =
 {
     0, 4, 8, 12, 16, 20, 24, 28, 36, 44, 52, 60, 72, 88, 108, 128
 };
 
-static uint16_t *swb_offset_1024_window[] =
+ALIGN static const uint16_t *swb_offset_1024_window[] =
 {
     swb_offset_1024_96,      /* 96000 */
     swb_offset_1024_96,      /* 88200 */
@@ -214,7 +231,7 @@ static uint16_t *swb_offset_1024_window[] =
 };
 
 #ifdef LD_DEC
-static uint16_t *swb_offset_512_window[] =
+ALIGN static const uint16_t *swb_offset_512_window[] =
 {
     0,                       /* 96000 */
     0,                       /* 88200 */
@@ -230,7 +247,7 @@ static uint16_t *swb_offset_512_window[] =
     0                        /* 8000  */
 };
 
-static uint16_t *swb_offset_480_window[] =
+ALIGN static const uint16_t *swb_offset_480_window[] =
 {
     0,                       /* 96000 */
     0,                       /* 88200 */
@@ -247,7 +264,7 @@ static uint16_t *swb_offset_480_window[] =
 };
 #endif
 
-static uint16_t *swb_offset_128_window[] =
+ALIGN static const  uint16_t *swb_offset_128_window[] =
 {
     swb_offset_128_96,       /* 96000 */
     swb_offset_128_96,       /* 88200 */
@@ -410,14 +427,12 @@ uint8_t window_grouping_info(faacDecHandle hDecoder, ic_stream *ics)
   - Within a scalefactor window band, the coefficients are in ascending
     spectral order.
 */
-void quant_to_spec(ic_stream *ics, real_t *spec_data, uint16_t frame_len)
+static void quant_to_spec(ic_stream *ics, real_t *spec_data, uint16_t frame_len)
 {
     uint8_t g, sfb, win;
     uint16_t width, bin, k, gindex;
 
-    real_t tmp_spec[1024];
-
-    memset(tmp_spec, 0, frame_len*sizeof(real_t));
+    ALIGN real_t tmp_spec[1024] = {0};
 
     k = 0;
     gindex = 0;
@@ -452,106 +467,115 @@ void quant_to_spec(ic_stream *ics, real_t *spec_data, uint16_t frame_len)
     memcpy(spec_data, tmp_spec, frame_len*sizeof(real_t));
 }
 
-#ifndef FIXED_POINT
-void build_tables(real_t *pow2_table)
-{
-    uint16_t i;
-
-    /* build pow(2, 0.25*x) table for scalefactors */
-    for(i = 0; i < POW_TABLE_SIZE; i++)
-    {
-        pow2_table[i] = REAL_CONST(pow(2.0, 0.25 * (i-100)));
-    }
-}
-#endif
-
-static INLINE real_t iquant(int16_t q, real_t *tab)
+static INLINE real_t iquant(int16_t q, const real_t *tab, uint8_t *error)
 {
 #ifdef FIXED_POINT
+    static const real_t errcorr[] = {
+        REAL_CONST(0), REAL_CONST(1.0/8.0), REAL_CONST(2.0/8.0), REAL_CONST(3.0/8.0),
+        REAL_CONST(4.0/8.0),  REAL_CONST(5.0/8.0), REAL_CONST(6.0/8.0), REAL_CONST(7.0/8.0),
+        REAL_CONST(0)
+    };
+    real_t x1, x2;
     int16_t sgn = 1;
 
-    if (q == 0) return 0;
-
     if (q < 0)
     {
         q = -q;
         sgn = -1;
     }
 
-    if (q >= IQ_TABLE_SIZE)
-        return 0; /* sgn * tab[q>>3] * 16; */
+    if (q < IQ_TABLE_SIZE)
+        return sgn * tab[q];
 
-    return sgn * tab[q];
+    /* linear interpolation */
+    x1 = tab[q>>3];
+    x2 = tab[(q>>3) + 1];
+    return sgn * 16 * (MUL_R(errcorr[q&7],(x2-x1)) + x1);
 #else
-    int16_t sgn = 1;
-
-    if (q == 0) return 0;
-
     if (q < 0)
     {
-        q = -q;
-        sgn = -1;
-    }
+        /* tab contains a value for all possible q [0,8192] */
+        if (-q < IQ_TABLE_SIZE)
+            return -tab[-q];
 
-    if (q < IQ_TABLE_SIZE)
-        return sgn * tab[q];
+        *error = 17;
+        return 0;
+    } else {
+        /* tab contains a value for all possible q [0,8192] */
+        if (q < IQ_TABLE_SIZE)
+            return tab[q];
 
-    return sgn * pow(q, 4./3.);
+        *error = 17;
+        return 0;
+    }
 #endif
 }
 
-void inverse_quantization(real_t *x_invquant, int16_t *x_quant, uint16_t frame_len)
+static uint8_t inverse_quantization(real_t *x_invquant, const int16_t *x_quant, const uint16_t frame_len)
 {
     int16_t i;
-    real_t *tab = iq_table;
+    uint8_t error = 0; /* Init error flag */
+    const real_t *tab = iq_table;
 
-    for(i = 0; i < frame_len; i+=4)
+    for (i = 0; i < frame_len; i+=4)
     {
-        x_invquant[i] = iquant(x_quant[i], tab);
-        x_invquant[i+1] = iquant(x_quant[i+1], tab);
-        x_invquant[i+2] = iquant(x_quant[i+2], tab);
-        x_invquant[i+3] = iquant(x_quant[i+3], tab);
+        x_invquant[i] = iquant(x_quant[i], tab, &error);
+        x_invquant[i+1] = iquant(x_quant[i+1], tab, &error);
+        x_invquant[i+2] = iquant(x_quant[i+2], tab, &error);
+        x_invquant[i+3] = iquant(x_quant[i+3], tab, &error);
     }
+
+    return error;
 }
 
 #ifndef FIXED_POINT
-static INLINE real_t get_scale_factor_gain(uint16_t scale_factor, real_t *pow2_table)
-{
-    if (scale_factor < POW_TABLE_SIZE)
-        return pow2_table[scale_factor];
-    else
-        return REAL_CONST(pow(2.0, 0.25 * (scale_factor - 100)));
-}
-#else
-static real_t pow2_table[] =
+ALIGN static const real_t pow2sf_tab[] = {
+    2.9802322387695313E-008, 5.9604644775390625E-008, 1.1920928955078125E-007,
+    2.384185791015625E-007, 4.76837158203125E-007, 9.5367431640625E-007,
+    1.9073486328125E-006, 3.814697265625E-006, 7.62939453125E-006,
+    1.52587890625E-005, 3.0517578125E-005, 6.103515625E-005,
+    0.0001220703125, 0.000244140625, 0.00048828125,
+    0.0009765625, 0.001953125, 0.00390625,
+    0.0078125, 0.015625, 0.03125,
+    0.0625, 0.125, 0.25,
+    0.5, 1.0, 2.0,
+    4.0, 8.0, 16.0, 32.0,
+    64.0, 128.0, 256.0,
+    512.0, 1024.0, 2048.0,
+    4096.0, 8192.0, 16384.0,
+    32768.0, 65536.0, 131072.0,
+    262144.0, 524288.0, 1048576.0,
+    2097152.0, 4194304.0, 8388608.0,
+    16777216.0, 33554432.0, 67108864.0,
+    134217728.0, 268435456.0, 536870912.0,
+    1073741824.0, 2147483648.0, 4294967296.0,
+    8589934592.0, 17179869184.0, 34359738368.0,
+    68719476736.0, 137438953472.0, 274877906944.0
+};
+#endif
+
+ALIGN static real_t pow2_table[] =
 {
-    COEF_CONST(0.59460355750136),
-    COEF_CONST(0.70710678118655),
-    COEF_CONST(0.84089641525371),
+#if 0
+    COEF_CONST(0.59460355750136053335874998528024), /* 2^-0.75 */
+    COEF_CONST(0.70710678118654752440084436210485), /* 2^-0.5 */
+    COEF_CONST(0.84089641525371454303112547623321), /* 2^-0.25 */
+#endif
     COEF_CONST(1.0),
-    COEF_CONST(1.18920711500272),
-    COEF_CONST(1.41421356237310),
-    COEF_CONST(1.68179283050743)
+    COEF_CONST(1.1892071150027210667174999705605), /* 2^0.25 */
+    COEF_CONST(1.4142135623730950488016887242097), /* 2^0.5 */
+    COEF_CONST(1.6817928305074290860622509524664) /* 2^0.75 */
 };
-#endif
 
-void apply_scalefactors(faacDecHandle hDecoder, ic_stream *ics, real_t *x_invquant,
-                        uint16_t frame_len)
+void apply_scalefactors(faacDecHandle hDecoder, ic_stream *ics,
+                        real_t *x_invquant, uint16_t frame_len)
 {
     uint8_t g, sfb;
     uint16_t top;
-#ifndef FIXED_POINT
-    real_t scale;
-#else
     int32_t exp, frac;
-#endif
     uint8_t groups = 0;
     uint16_t nshort = frame_len/8;
 
-    static real_t max_fp = 0;
-    static real_t max_exp = 0;
-    static real_t max_frac = 0;
-
     for (g = 0; g < ics->num_window_groups; g++)
     {
         uint16_t k = 0;
@@ -564,12 +588,20 @@ void apply_scalefactors(faacDecHandle hDecoder, ic_stream *ics, real_t *x_invqua
         {
             top = ics->sect_sfb_offset[g][sfb+1];
 
-#ifndef FIXED_POINT
-            scale = get_scale_factor_gain(ics->scale_factors[g][sfb], hDecoder->pow2_table);
-#else
-            exp = (ics->scale_factors[g][sfb] - 100) / 4;
-            frac = (ics->scale_factors[g][sfb] - 100) % 4;
+            /* this could be scalefactor for IS or PNS, those can be negative or bigger then 255 */
+            /* just ignore them */
+            if (ics->scale_factors[g][sfb] < 0 || ics->scale_factors[g][sfb] > 255)
+            {
+                exp = 0;
+                frac = 0;
+            } else {
+                /* ics->scale_factors[g][sfb] must be between 0 and 255 */
+                exp = (ics->scale_factors[g][sfb] /* - 100 */) >> 2;
+                frac = (ics->scale_factors[g][sfb] /* - 100 */) & 3;
+            }
 
+#ifdef FIXED_POINT
+            exp -= 25;
             /* IMDCT pre-scaling */
             if (hDecoder->object_type == LD)
             {
@@ -585,12 +617,7 @@ void apply_scalefactors(faacDecHandle hDecoder, ic_stream *ics, real_t *x_invqua
             /* minimum size of a sf band is 4 and always a multiple of 4 */
             for ( ; k < top; k += 4)
             {
-#ifndef FIXED_POINT
-                x_invquant[k+(groups*nshort)]   = x_invquant[k+(groups*nshort)]   * scale;
-                x_invquant[k+(groups*nshort)+1] = x_invquant[k+(groups*nshort)+1] * scale;
-                x_invquant[k+(groups*nshort)+2] = x_invquant[k+(groups*nshort)+2] * scale;
-                x_invquant[k+(groups*nshort)+3] = x_invquant[k+(groups*nshort)+3] * scale;
-#else
+#ifdef FIXED_POINT
                 if (exp < 0)
                 {
                     x_invquant[k+(groups*nshort)] >>= -exp;
@@ -603,17 +630,653 @@ void apply_scalefactors(faacDecHandle hDecoder, ic_stream *ics, real_t *x_invqua
                     x_invquant[k+(groups*nshort)+2] <<= exp;
                     x_invquant[k+(groups*nshort)+3] <<= exp;
                 }
-
-                if (frac)
-                {
-                    x_invquant[k+(groups*nshort)]   = MUL_R_C(x_invquant[k+(groups*nshort)],pow2_table[frac + 3]);
-                    x_invquant[k+(groups*nshort)+1] = MUL_R_C(x_invquant[k+(groups*nshort)+1],pow2_table[frac + 3]);
-                    x_invquant[k+(groups*nshort)+2] = MUL_R_C(x_invquant[k+(groups*nshort)+2],pow2_table[frac + 3]);
-                    x_invquant[k+(groups*nshort)+3] = MUL_R_C(x_invquant[k+(groups*nshort)+3],pow2_table[frac + 3]);
-                }
+#else
+                x_invquant[k+(groups*nshort)]   = x_invquant[k+(groups*nshort)]   * pow2sf_tab[exp/*+25*/];
+                x_invquant[k+(groups*nshort)+1] = x_invquant[k+(groups*nshort)+1] * pow2sf_tab[exp/*+25*/];
+                x_invquant[k+(groups*nshort)+2] = x_invquant[k+(groups*nshort)+2] * pow2sf_tab[exp/*+25*/];
+                x_invquant[k+(groups*nshort)+3] = x_invquant[k+(groups*nshort)+3] * pow2sf_tab[exp/*+25*/];
 #endif
+
+                x_invquant[k+(groups*nshort)]   = MUL_C(x_invquant[k+(groups*nshort)],pow2_table[frac /* + 3*/]);
+                x_invquant[k+(groups*nshort)+1] = MUL_C(x_invquant[k+(groups*nshort)+1],pow2_table[frac /* + 3*/]);
+                x_invquant[k+(groups*nshort)+2] = MUL_C(x_invquant[k+(groups*nshort)+2],pow2_table[frac /* + 3*/]);
+                x_invquant[k+(groups*nshort)+3] = MUL_C(x_invquant[k+(groups*nshort)+3],pow2_table[frac /* + 3*/]);
             }
         }
         groups += ics->window_group_length[g];
     }
 }
+
+#ifdef USE_SSE
+void apply_scalefactors_sse(faacDecHandle hDecoder, ic_stream *ics,
+                            real_t *x_invquant, uint16_t frame_len)
+{
+    uint8_t g, sfb;
+    uint16_t top;
+    int32_t exp, frac;
+    uint8_t groups = 0;
+    uint16_t nshort = frame_len/8;
+
+    for (g = 0; g < ics->num_window_groups; g++)
+    {
+        uint16_t k = 0;
+
+        /* using this nshort*groups doesn't hurt long blocks, because
+           long blocks only have 1 group, so that means 'groups' is
+           always 0 for long blocks
+        */
+        for (sfb = 0; sfb < ics->max_sfb; sfb++)
+        {
+            top = ics->sect_sfb_offset[g][sfb+1];
+
+            exp = (ics->scale_factors[g][sfb] /* - 100 */) >> 2;
+            frac = (ics->scale_factors[g][sfb] /* - 100 */) & 3;
+
+            /* minimum size of a sf band is 4 and always a multiple of 4 */
+            for ( ; k < top; k += 4)
+            {
+                __m128 m1 = _mm_load_ps(&x_invquant[k+(groups*nshort)]);
+                __m128 m2 = _mm_load_ps1(&pow2sf_tab[exp /*+25*/]);
+                __m128 m3 = _mm_load_ps1(&pow2_table[frac /* + 3*/]);
+                __m128 m4 = _mm_mul_ps(m1, m2);
+                __m128 m5 = _mm_mul_ps(m3, m4);
+                _mm_store_ps(&x_invquant[k+(groups*nshort)], m5);
+            }
+        }
+        groups += ics->window_group_length[g];
+    }
+}
+#endif
+
+static uint8_t allocate_single_channel(faacDecHandle hDecoder, uint8_t channel,
+                                       uint8_t output_channels)
+{
+    uint8_t mul = 1;
+
+#ifdef MAIN_DEC
+    /* MAIN object type prediction */
+    if (hDecoder->object_type == MAIN)
+    {
+        /* allocate the state only when needed */
+        if (hDecoder->pred_stat[channel] == NULL)
+        {
+            hDecoder->pred_stat[channel] = (pred_state*)faad_malloc(hDecoder->frameLength * sizeof(pred_state));
+            reset_all_predictors(hDecoder->pred_stat[channel], hDecoder->frameLength);
+        }
+    }
+#endif
+
+#ifdef LTP_DEC
+    if (is_ltp_ot(hDecoder->object_type))
+    {
+        /* allocate the state only when needed */
+        if (hDecoder->lt_pred_stat[channel] == NULL)
+        {
+            hDecoder->lt_pred_stat[channel] = (int16_t*)faad_malloc(hDecoder->frameLength*4 * sizeof(int16_t));
+            memset(hDecoder->lt_pred_stat[channel], 0, hDecoder->frameLength*4 * sizeof(int16_t));
+        }
+    }
+#endif
+
+    if (hDecoder->time_out[channel] == NULL)
+    {
+        mul = 1;
+#ifdef SBR_DEC
+        hDecoder->sbr_alloced[hDecoder->fr_ch_ele] = 0;
+        if ((hDecoder->sbr_present_flag == 1) || (hDecoder->forceUpSampling == 1))
+        {
+            /* SBR requires 2 times as much output data */
+            mul = 2;
+            hDecoder->sbr_alloced[hDecoder->fr_ch_ele] = 1;
+        }
+#endif
+        hDecoder->time_out[channel] = (real_t*)faad_malloc(mul*hDecoder->frameLength*sizeof(real_t));
+        memset(hDecoder->time_out[channel], 0, mul*hDecoder->frameLength*sizeof(real_t));
+    }
+#if (defined(PS_DEC) || defined(DRM_PS))
+    if (output_channels == 2)
+    {
+        if (hDecoder->time_out[channel+1] == NULL)
+        {
+            hDecoder->time_out[channel+1] = (real_t*)faad_malloc(mul*hDecoder->frameLength*sizeof(real_t));
+            memset(hDecoder->time_out[channel+1], 0, mul*hDecoder->frameLength*sizeof(real_t));
+        }
+    }
+#endif
+
+    if (hDecoder->fb_intermed[channel] == NULL)
+    {
+        hDecoder->fb_intermed[channel] = (real_t*)faad_malloc(hDecoder->frameLength*sizeof(real_t));
+        memset(hDecoder->fb_intermed[channel], 0, hDecoder->frameLength*sizeof(real_t));
+    }
+
+#ifdef SSR_DEC
+    if (hDecoder->object_type == SSR)
+    {
+        if (hDecoder->ssr_overlap[channel] == NULL)
+        {
+            hDecoder->ssr_overlap[channel] = (real_t*)faad_malloc(2*hDecoder->frameLength*sizeof(real_t));
+            memset(hDecoder->ssr_overlap[channel], 0, 2*hDecoder->frameLength*sizeof(real_t));
+        }
+        if (hDecoder->prev_fmd[channel] == NULL)
+        {
+            uint16_t k;
+            hDecoder->prev_fmd[channel] = (real_t*)faad_malloc(2*hDecoder->frameLength*sizeof(real_t));
+            for (k = 0; k < 2*hDecoder->frameLength; k++)
+                hDecoder->prev_fmd[channel][k] = REAL_CONST(-1);
+        }
+    }
+#endif
+
+    return 0;
+}
+
+static uint8_t allocate_channel_pair(faacDecHandle hDecoder,
+                                     uint8_t channel, uint8_t paired_channel)
+{
+    uint8_t mul = 1;
+
+#ifdef MAIN_DEC
+    /* MAIN object type prediction */
+    if (hDecoder->object_type == MAIN)
+    {
+        /* allocate the state only when needed */
+        if (hDecoder->pred_stat[channel] == NULL)
+        {
+            hDecoder->pred_stat[channel] = (pred_state*)faad_malloc(hDecoder->frameLength * sizeof(pred_state));
+            reset_all_predictors(hDecoder->pred_stat[channel], hDecoder->frameLength);
+        }
+        if (hDecoder->pred_stat[paired_channel] == NULL)
+        {
+            hDecoder->pred_stat[paired_channel] = (pred_state*)faad_malloc(hDecoder->frameLength * sizeof(pred_state));
+            reset_all_predictors(hDecoder->pred_stat[paired_channel], hDecoder->frameLength);
+        }
+    }
+#endif
+
+#ifdef LTP_DEC
+    if (is_ltp_ot(hDecoder->object_type))
+    {
+        /* allocate the state only when needed */
+        if (hDecoder->lt_pred_stat[channel] == NULL)
+        {
+            hDecoder->lt_pred_stat[channel] = (int16_t*)faad_malloc(hDecoder->frameLength*4 * sizeof(int16_t));
+            memset(hDecoder->lt_pred_stat[channel], 0, hDecoder->frameLength*4 * sizeof(int16_t));
+        }
+        if (hDecoder->lt_pred_stat[paired_channel] == NULL)
+        {
+            hDecoder->lt_pred_stat[paired_channel] = (int16_t*)faad_malloc(hDecoder->frameLength*4 * sizeof(int16_t));
+            memset(hDecoder->lt_pred_stat[paired_channel], 0, hDecoder->frameLength*4 * sizeof(int16_t));
+        }
+    }
+#endif
+
+    if (hDecoder->time_out[channel] == NULL)
+    {
+        mul = 1;
+#ifdef SBR_DEC
+        hDecoder->sbr_alloced[hDecoder->fr_ch_ele] = 0;
+        if ((hDecoder->sbr_present_flag == 1) || (hDecoder->forceUpSampling == 1))
+        {
+            /* SBR requires 2 times as much output data */
+            mul = 2;
+            hDecoder->sbr_alloced[hDecoder->fr_ch_ele] = 1;
+        }
+#endif
+        hDecoder->time_out[channel] = (real_t*)faad_malloc(mul*hDecoder->frameLength*sizeof(real_t));
+        memset(hDecoder->time_out[channel], 0, mul*hDecoder->frameLength*sizeof(real_t));
+    }
+    if (hDecoder->time_out[paired_channel] == NULL)
+    {
+        hDecoder->time_out[paired_channel] = (real_t*)faad_malloc(mul*hDecoder->frameLength*sizeof(real_t));
+        memset(hDecoder->time_out[paired_channel], 0, mul*hDecoder->frameLength*sizeof(real_t));
+    }
+
+    if (hDecoder->fb_intermed[channel] == NULL)
+    {
+        hDecoder->fb_intermed[channel] = (real_t*)faad_malloc(hDecoder->frameLength*sizeof(real_t));
+        memset(hDecoder->fb_intermed[channel], 0, hDecoder->frameLength*sizeof(real_t));
+    }
+    if (hDecoder->fb_intermed[paired_channel] == NULL)
+    {
+        hDecoder->fb_intermed[paired_channel] = (real_t*)faad_malloc(hDecoder->frameLength*sizeof(real_t));
+        memset(hDecoder->fb_intermed[paired_channel], 0, hDecoder->frameLength*sizeof(real_t));
+    }
+
+#ifdef SSR_DEC
+    if (hDecoder->object_type == SSR)
+    {
+        if (hDecoder->ssr_overlap[cpe->channel] == NULL)
+        {
+            hDecoder->ssr_overlap[cpe->channel] = (real_t*)faad_malloc(2*hDecoder->frameLength*sizeof(real_t));
+            memset(hDecoder->ssr_overlap[cpe->channel], 0, 2*hDecoder->frameLength*sizeof(real_t));
+        }
+        if (hDecoder->ssr_overlap[cpe->paired_channel] == NULL)
+        {
+            hDecoder->ssr_overlap[cpe->paired_channel] = (real_t*)faad_malloc(2*hDecoder->frameLength*sizeof(real_t));
+            memset(hDecoder->ssr_overlap[cpe->paired_channel], 0, 2*hDecoder->frameLength*sizeof(real_t));
+        }
+        if (hDecoder->prev_fmd[cpe->channel] == NULL)
+        {
+            uint16_t k;
+            hDecoder->prev_fmd[cpe->channel] = (real_t*)faad_malloc(2*hDecoder->frameLength*sizeof(real_t));
+            for (k = 0; k < 2*hDecoder->frameLength; k++)
+                hDecoder->prev_fmd[cpe->channel][k] = REAL_CONST(-1);
+        }
+        if (hDecoder->prev_fmd[cpe->paired_channel] == NULL)
+        {
+            uint16_t k;
+            hDecoder->prev_fmd[cpe->paired_channel] = (real_t*)faad_malloc(2*hDecoder->frameLength*sizeof(real_t));
+            for (k = 0; k < 2*hDecoder->frameLength; k++)
+                hDecoder->prev_fmd[cpe->paired_channel][k] = REAL_CONST(-1);
+        }
+    }
+#endif
+
+    return 0;
+}
+
+uint8_t reconstruct_single_channel(faacDecHandle hDecoder, ic_stream *ics,
+                                   element *sce, int16_t *spec_data)
+{
+    uint8_t retval, output_channels;
+    ALIGN real_t spec_coef[1024];
+
+#ifdef PROFILE
+    int64_t count = faad_get_ts();
+#endif
+
+
+    /* determine whether some mono->stereo tool is used */
+#if (defined(PS_DEC) || defined(DRM_PS))
+    output_channels = hDecoder->ps_used[hDecoder->fr_ch_ele] ? 2 : 1;
+#else
+    output_channels = 1;
+#endif
+    if (hDecoder->element_output_channels[hDecoder->fr_ch_ele] == 0)
+    {
+        /* element_output_channels not set yet */
+        hDecoder->element_output_channels[hDecoder->fr_ch_ele] = output_channels;
+    } else if (hDecoder->element_output_channels[hDecoder->fr_ch_ele] != output_channels) {
+        /* element inconsistency */
+        return 21;
+    }
+
+
+    if (hDecoder->element_alloced[hDecoder->fr_ch_ele] == 0)
+    {
+        retval = allocate_single_channel(hDecoder, sce->channel, output_channels);
+        if (retval > 0)
+            return retval;
+
+        hDecoder->element_alloced[hDecoder->fr_ch_ele] = 1;
+    }
+
+
+    /* inverse quantization */
+    retval = inverse_quantization(spec_coef, spec_data, hDecoder->frameLength);
+    if (retval > 0)
+        return retval;
+
+    /* apply scalefactors */
+#ifndef USE_SSE
+    apply_scalefactors(hDecoder, ics, spec_coef, hDecoder->frameLength);
+#else
+    hDecoder->apply_sf_func(hDecoder, ics, spec_coef, hDecoder->frameLength);
+#endif
+
+    /* deinterleave short block grouping */
+    if (ics->window_sequence == EIGHT_SHORT_SEQUENCE)
+        quant_to_spec(ics, spec_coef, hDecoder->frameLength);
+
+#ifdef PROFILE
+    count = faad_get_ts() - count;
+    hDecoder->requant_cycles += count;
+#endif
+
+
+    /* pns decoding */
+    pns_decode(ics, NULL, spec_coef, NULL, hDecoder->frameLength, 0, hDecoder->object_type);
+
+#ifdef MAIN_DEC
+    /* MAIN object type prediction */
+    if (hDecoder->object_type == MAIN)
+    {
+        /* intra channel prediction */
+        ic_prediction(ics, spec_coef, hDecoder->pred_stat[sce->channel], hDecoder->frameLength,
+            hDecoder->sf_index);
+
+        /* In addition, for scalefactor bands coded by perceptual
+           noise substitution the predictors belonging to the
+           corresponding spectral coefficients are reset.
+        */
+        pns_reset_pred_state(ics, hDecoder->pred_stat[sce->channel]);
+    }
+#endif
+
+#ifdef LTP_DEC
+    if (is_ltp_ot(hDecoder->object_type))
+    {
+#ifdef LD_DEC
+        if (hDecoder->object_type == LD)
+        {
+            if (ics->ltp.data_present)
+            {
+                if (ics->ltp.lag_update)
+                    hDecoder->ltp_lag[sce->channel] = ics->ltp.lag;
+            }
+            ics->ltp.lag = hDecoder->ltp_lag[sce->channel];
+        }
+#endif
+
+        /* long term prediction */
+        lt_prediction(ics, &(ics->ltp), spec_coef, hDecoder->lt_pred_stat[sce->channel], hDecoder->fb,
+            ics->window_shape, hDecoder->window_shape_prev[sce->channel],
+            hDecoder->sf_index, hDecoder->object_type, hDecoder->frameLength);
+    }
+#endif
+
+    /* tns decoding */
+    tns_decode_frame(ics, &(ics->tns), hDecoder->sf_index, hDecoder->object_type,
+        spec_coef, hDecoder->frameLength);
+
+    /* drc decoding */
+    if (hDecoder->drc->present)
+    {
+        if (!hDecoder->drc->exclude_mask[sce->channel] || !hDecoder->drc->excluded_chns_present)
+            drc_decode(hDecoder->drc, spec_coef);
+    }
+
+
+    /* filter bank */
+#ifdef SSR_DEC
+    if (hDecoder->object_type != SSR)
+    {
+#endif
+#ifdef USE_SSE
+        hDecoder->fb->if_func(hDecoder->fb, ics->window_sequence, ics->window_shape,
+            hDecoder->window_shape_prev[sce->channel], spec_coef,
+            hDecoder->time_out[sce->channel], hDecoder->object_type, hDecoder->frameLength);
+#else
+        ifilter_bank(hDecoder->fb, ics->window_sequence, ics->window_shape,
+            hDecoder->window_shape_prev[sce->channel], spec_coef,
+            hDecoder->time_out[sce->channel], hDecoder->fb_intermed[sce->channel],
+            hDecoder->object_type, hDecoder->frameLength);
+#endif
+#ifdef SSR_DEC
+    } else {
+        ssr_decode(&(ics->ssr), hDecoder->fb, ics->window_sequence, ics->window_shape,
+            hDecoder->window_shape_prev[sce->channel], spec_coef, hDecoder->time_out[sce->channel],
+            hDecoder->ssr_overlap[sce->channel], hDecoder->ipqf_buffer[sce->channel], hDecoder->prev_fmd[sce->channel],
+            hDecoder->frameLength);
+    }
+#endif
+
+    /* save window shape for next frame */
+    hDecoder->window_shape_prev[sce->channel] = ics->window_shape;
+
+#ifdef LTP_DEC
+    if (is_ltp_ot(hDecoder->object_type))
+    {
+        lt_update_state(hDecoder->lt_pred_stat[sce->channel], hDecoder->time_out[sce->channel],
+            hDecoder->fb_intermed[sce->channel], hDecoder->frameLength, hDecoder->object_type);
+    }
+#endif
+
+#ifdef SBR_DEC
+    if (((hDecoder->sbr_present_flag == 1) || (hDecoder->forceUpSampling == 1))
+        && hDecoder->sbr_alloced[hDecoder->fr_ch_ele])
+    {
+        uint8_t ele = hDecoder->fr_ch_ele;
+        uint8_t ch = sce->channel;
+
+        /* following case can happen when forceUpSampling == 1 */
+        if (hDecoder->sbr[ele] == NULL)
+        {
+            hDecoder->sbr[ele] = sbrDecodeInit(hDecoder->frameLength,
+                sce->ele_id, 2*get_sample_rate(hDecoder->sf_index)
+#ifdef DRM
+                , 0
+#endif
+                );
+        }
+
+        /* check if any of the PS tools is used */
+#if (defined(PS_DEC) || defined(DRM_PS))
+        if (output_channels == 1)
+        {
+#endif
+            retval = sbrDecodeSingleFrame(hDecoder->sbr[ele], hDecoder->time_out[ch],
+                hDecoder->postSeekResetFlag, hDecoder->forceUpSampling);
+#if (defined(PS_DEC) || defined(DRM_PS))
+        } else {
+            retval = sbrDecodeSingleFramePS(hDecoder->sbr[ele], hDecoder->time_out[ch],
+                hDecoder->time_out[ch+1], hDecoder->postSeekResetFlag,
+                hDecoder->forceUpSampling);
+        }
+#endif
+        if (retval > 0)
+            return retval;
+    } else if (((hDecoder->sbr_present_flag == 1) || (hDecoder->forceUpSampling == 1))
+        && !hDecoder->sbr_alloced[hDecoder->fr_ch_ele])
+    {
+        return 23;
+    }
+#endif
+
+    return 0;
+}
+
+uint8_t reconstruct_channel_pair(faacDecHandle hDecoder, ic_stream *ics1, ic_stream *ics2,
+                                 element *cpe, int16_t *spec_data1, int16_t *spec_data2)
+{
+    uint8_t retval;
+    ALIGN real_t spec_coef1[1024];
+    ALIGN real_t spec_coef2[1024];
+
+#ifdef PROFILE
+    int64_t count = faad_get_ts();
+#endif
+    if (hDecoder->element_alloced[hDecoder->fr_ch_ele] == 0)
+    {
+        retval = allocate_channel_pair(hDecoder, cpe->channel, cpe->paired_channel);
+        if (retval > 0)
+            return retval;
+
+        hDecoder->element_alloced[hDecoder->fr_ch_ele] = 1;
+    }
+
+    /* inverse quantization */
+    retval = inverse_quantization(spec_coef1, spec_data1, hDecoder->frameLength);
+    if (retval > 0)
+        return retval;
+
+    retval = inverse_quantization(spec_coef2, spec_data2, hDecoder->frameLength);
+    if (retval > 0)
+        return retval;
+
+    /* apply scalefactors */
+#ifndef USE_SSE
+    apply_scalefactors(hDecoder, ics1, spec_coef1, hDecoder->frameLength);
+    apply_scalefactors(hDecoder, ics2, spec_coef2, hDecoder->