* Follow of the new ac3_decoder ;)

  - New ac3_imdct
  - New ac3_downmix

Makefile

@@ -70,6 +70,8 @@ AC3_DECODER =	src/ac3_decoder/ac3_decoder_thread.o \
 		src/ac3_decoder/ac3_mantissa.o \
 		src/ac3_decoder/ac3_rematrix.o \
 		src/ac3_decoder/ac3_imdct.o \
+		src/ac3_decoder/ac3_imdct_c.o \
+		src/ac3_decoder/ac3_srfft.o \
 		src/ac3_decoder/ac3_downmix.o \
 		src/ac3_decoder/ac3_downmix_c.o
 

src/ac3_decoder/ac3_decoder.c

@@ -2,7 +2,7 @@
  * ac3_decoder.c: core ac3 decoder
  *****************************************************************************
  * Copyright (C) 1999, 2000 VideoLAN
- * $Id: ac3_decoder.c,v 1.29 2001/04/20 12:14:34 reno Exp $
+ * $Id: ac3_decoder.c,v 1.30 2001/04/30 21:04:20 reno Exp $
  *
  * Authors: Michel Kaempf <maxx@via.ecp.fr>
  *          Michel Lespinasse <walken@zoy.org>
@@ -45,6 +45,10 @@
 #include <stdio.h>
 
 void imdct_init (imdct_t * p_imdct);
+void downmix_init (downmix_t * p_downmix);
+
+static float cmixlev_lut[4] = { 0.707, 0.595, 0.500, 0.707 };
+static float smixlev_lut[4] = { 0.707, 0.500, 0.0  , 0.500 };
 
 int ac3_init (ac3dec_t * p_ac3dec)
 {
@@ -52,6 +56,7 @@ int ac3_init (ac3dec_t * p_ac3dec)
 //    p_ac3dec->bit_stream.i_available = 0;
     p_ac3dec->mantissa.lfsr_state = 1;       /* dither_gen initialization */
     imdct_init(&p_ac3dec->imdct);
+    downmix_init(&p_ac3dec->downmix);
     
     return 0;
 }
@@ -68,7 +73,27 @@ int ac3_decode_frame (ac3dec_t * p_ac3dec, s16 * buffer)
         return 1;
     }
     
+	/* compute downmix parameters
+	 * downmix to tow channels for now */
+	p_ac3dec->dm_par.clev = 0.0;
+    p_ac3dec->dm_par.slev = 0.0; 
+    p_ac3dec->dm_par.unit = 1.0;
+	if (p_ac3dec->bsi.acmod & 0x1)	/* have center */
+	    p_ac3dec->dm_par.clev = cmixlev_lut[p_ac3dec->bsi.cmixlev];
+
+	if (p_ac3dec->bsi.acmod & 0x4)	/* have surround channels */
+		p_ac3dec->dm_par.slev = smixlev_lut[p_ac3dec->bsi.surmixlev];
+
+    p_ac3dec->dm_par.unit /= 1.0 + p_ac3dec->dm_par.clev + p_ac3dec->dm_par.slev;
+	p_ac3dec->dm_par.clev *= p_ac3dec->dm_par.unit;
+	p_ac3dec->dm_par.slev *= p_ac3dec->dm_par.unit;
+
     for (i = 0; i < 6; i++) {
+        /* Initialize freq/time sample storage */
+        memset(p_ac3dec->samples, 0, sizeof(float) * 256 * 
+                (p_ac3dec->bsi.nfchans + p_ac3dec->bsi.lfeon));
+
+
         if ((p_ac3dec_t->p_fifo->b_die) && (p_ac3dec_t->p_fifo->b_error))
         {        
             return 1;

src/ac3_decoder/ac3_decoder.h

@@ -2,7 +2,7 @@
  * ac3_decoder.h : ac3 decoder interface
  *****************************************************************************
  * Copyright (C) 1999, 2000 VideoLAN
- * $Id: ac3_decoder.h,v 1.6 2001/04/26 00:12:19 reno Exp $
+ * $Id: ac3_decoder.h,v 1.7 2001/04/30 21:04:20 reno Exp $
  *
  * Authors: Michel Kaempf <maxx@via.ecp.fr>
  *          Renaud Dartus <reno@videolan.org>
@@ -308,6 +308,7 @@ typedef struct audblk_s {
 } audblk_t;
 
 /* Everything you wanted to know about band structure */
+
 /*
  * The entire frequency domain is represented by 256 real
  * floating point fourier coefficients. Only the lower 253
@@ -326,17 +327,6 @@ typedef struct audblk_s {
  * approximate a 1/6 octave scale.
  */
 
-typedef struct stream_coeffs_s
-{
-    float fbw[5][256];
-    float lfe[256];
-} stream_coeffs_t;
-
-typedef struct stream_samples_s
-{
-    float channel[6][256];
-} stream_samples_t;
-
 typedef struct bit_allocate_s
 {
     s16 psd[256];
@@ -375,6 +365,7 @@ typedef struct imdct_s
 
     /* Delay buffer for time domain interleaving */
     float delay[6][256];
+    float delay1[6][256];
 
     /* Twiddle factors for IMDCT */
     float xcos1[N/4];
@@ -392,8 +383,32 @@ typedef struct imdct_s
     complex_t w_32[32];
     complex_t w_64[64];
 
+    float xcos_sin_sse[128 * 4] __attribute__((aligned(16)));
+    
+    /* Functions */
+    void (*fft_64p) (complex_t *a);
+    
+    void (*imdct_do_512)(struct imdct_s * p_imdct, float data[], float delay[]);
+    void (*imdct_do_512_nol)(struct imdct_s * p_imdct, float data[], float delay[]);
+
 } imdct_t;
 
+typedef struct dm_par_s {
+    float unit;
+    float clev;
+    float slev;
+} dm_par_t;
+
+typedef struct downmix_s {
+    void (*downmix_3f_2r_to_2ch)(float *samples, dm_par_t * dm_par);
+    void (*downmix_3f_1r_to_2ch)(float *samples, dm_par_t * dm_par);
+    void (*downmix_2f_2r_to_2ch)(float *samples, dm_par_t * dm_par);
+    void (*downmix_2f_1r_to_2ch)(float *samples, dm_par_t * dm_par);
+    void (*downmix_3f_0r_to_2ch)(float *samples, dm_par_t * dm_par);
+    void (*stream_sample_2ch_to_s16)(s16 *s16_samples, float *left, float *right);
+    void (*stream_sample_1ch_to_s16)(s16 *s16_samples, float *center);
+} downmix_t;
+
 struct ac3dec_s
 {
     /*
@@ -412,10 +427,12 @@ struct ac3dec_s
     bsi_t               bsi;
     audblk_t            audblk;
 
-    stream_coeffs_t     coeffs;
-    stream_samples_t    samples;
+    float               samples[6][256];
+    dm_par_t            dm_par;
 
     bit_allocate_t      bit_allocate;
     mantissa_t          mantissa;
     imdct_t             imdct;
+    downmix_t           downmix;
+
 };

src/ac3_decoder/ac3_decoder_thread.c

@@ -2,7 +2,7 @@
  * ac3_decoder_thread.c: ac3 decoder thread
  *****************************************************************************
  * Copyright (C) 1999, 2000 VideoLAN
- * $Id: ac3_decoder_thread.c,v 1.29 2001/04/25 10:22:32 massiot Exp $
+ * $Id: ac3_decoder_thread.c,v 1.30 2001/04/30 21:04:20 reno Exp $
  *
  * Authors: Michel Lespinasse <walken@zoy.org>
  *
@@ -170,6 +170,7 @@ static void RunThread (ac3dec_thread_t * p_ac3dec_t)
     {
         s16 * buffer;
         ac3_sync_info_t sync_info;
+        int ptr;
 
         if (!sync) {
             do {
@@ -177,6 +178,14 @@ static void RunThread (ac3dec_thread_t * p_ac3dec_t)
             } while ((!p_ac3dec_t->sync_ptr) && (!p_ac3dec_t->p_fifo->b_die)
                     && (!p_ac3dec_t->p_fifo->b_error));
             
+            ptr = p_ac3dec_t->sync_ptr;
+
+            while(ptr-- && (!p_ac3dec_t->p_fifo->b_die)
+                && (!p_ac3dec_t->p_fifo->b_error))
+            {
+                p_ac3dec_t->ac3_decoder.bit_stream.p_byte++;
+            }
+                        
             /* we are in sync now */
             sync = 1;
         }

src/ac3_decoder/ac3_downmix.c

@@ -2,7 +2,7 @@
  * ac3_downmix.c: ac3 downmix functions
  *****************************************************************************
  * Copyright (C) 1999, 2000 VideoLAN
- * $Id: ac3_downmix.c,v 1.20 2001/04/20 12:14:34 reno Exp $
+ * $Id: ac3_downmix.c,v 1.21 2001/04/30 21:04:20 reno Exp $
  *
  * Authors: Michel Kaempf <maxx@via.ecp.fr>
  *          Aaron Holtzman <aholtzma@engr.uvic.ca>
@@ -29,72 +29,36 @@
 #include "threads.h"
 #include "mtime.h"
 
+#include "tests.h"
+
 #include "stream_control.h"
 #include "input_ext-dec.h"
 #include "ac3_decoder.h"
 #include "ac3_internal.h"
 #include "ac3_downmix.h"
 
-/* Pre-scaled downmix coefficients */
-static const float cmixlev_lut[4] = { 0.2928, 0.2468, 0.2071, 0.2468 };
-static const float smixlev_lut[4] = { 0.2928, 0.2071, 0.0   , 0.2071 };
-
-/* Downmix into _two_ channels...other downmix modes aren't implemented
- * to reduce complexity. Realistically, there aren't many machines around
- * with > 2 channel output anyways */
-
-int __inline__ downmix (ac3dec_t * p_ac3dec, float * channel, s16 * out_buf)
+void downmix_init (downmix_t * p_downmix)
 {
-
-    dm_par_t    dm_par;
-    
-    dm_par.clev = 0.0;
-    dm_par.slev = 0.0;
-    dm_par.unit = 1.0;
-
-    if (p_ac3dec->bsi.acmod & 0x1) /* have center */
-        dm_par.clev = cmixlev_lut[p_ac3dec->bsi.cmixlev];
-
-    if (p_ac3dec->bsi.acmod & 0x4) /* have surround channels */
-        dm_par.slev = smixlev_lut[p_ac3dec->bsi.surmixlev];
-
-    dm_par.unit /= 1.0 + dm_par.clev + dm_par.slev;
-    dm_par.clev *= dm_par.unit;
-    dm_par.slev *= dm_par.unit;
-
-
-    /*
-    if (p_ac3dec->bsi.acmod > 7)
-        intf_ErrMsg( "ac3dec: (downmix) invalid acmod number" );
-    */
-    
-    switch(p_ac3dec->bsi.acmod)
+#if 0
+    if ( TestCPU (CPU_CAPABILITY_MMX) )
+    {
+		fprintf(stderr,"Using MMX for downmix\n");
+		p_downmix->downmix_3f_2r_to_2ch = downmix_3f_2r_to_2ch_kni;
+		p_downmix->downmix_2f_2r_to_2ch = downmix_2f_2r_to_2ch_kni;
+		p_downmix->downmix_3f_1r_to_2ch = downmix_3f_1r_to_2ch_kni;
+		p_downmix->downmix_2f_1r_to_2ch = downmix_2f_1r_to_2ch_kni;
+		p_downmix->downmix_3f_0r_to_2ch = downmix_3f_0r_to_2ch_kni;
+		p_downmix->stream_sample_2ch_to_s16 = stream_sample_2ch_to_s16_kni;
+    	p_downmix->stream_sample_1ch_to_s16 = stream_sample_1ch_to_s16_kni;
+    } else 
+#endif
     {
-        case 7: // 3/2
-            downmix_3f_2r_to_2ch_c (channel, &dm_par);
-            break;
-        case 6: // 2/2
-            downmix_2f_2r_to_2ch_c (channel, &dm_par);
-            break;
-        case 5: // 3/1
-            downmix_3f_1r_to_2ch_c (channel, &dm_par);
-            break;
-        case 4: // 2/1
-            downmix_2f_1r_to_2ch_c (channel, &dm_par);
-            break;
-        case 3: // 3/0
-            downmix_3f_0r_to_2ch_c (channel, &dm_par);
-            break;
-        case 2:
-            break;
-        default: // 1/0
-            /* FIXME
-            if (p_ac3dec->bsi.acmod == 1)
-                center = p_ac3dec->samples.channel[0];
-            else if (p_ac3dec->bsi.acmod == 0)
-                center = p_ac3dec->samples.channel[0]; */
-            return 1;
+		p_downmix->downmix_3f_2r_to_2ch = downmix_3f_2r_to_2ch_c;
+		p_downmix->downmix_2f_2r_to_2ch = downmix_2f_2r_to_2ch_c;
+		p_downmix->downmix_3f_1r_to_2ch = downmix_3f_1r_to_2ch_c;
+		p_downmix->downmix_2f_1r_to_2ch = downmix_2f_1r_to_2ch_c;
+		p_downmix->downmix_3f_0r_to_2ch = downmix_3f_0r_to_2ch_c;
+		p_downmix->stream_sample_2ch_to_s16 = stream_sample_2ch_to_s16_c;
+		p_downmix->stream_sample_1ch_to_s16 = stream_sample_1ch_to_s16_c;
     }
-    return 0;
 }
-

src/ac3_decoder/ac3_downmix.h

@@ -2,7 +2,7 @@
  * ac3_downmix.h: ac3 downmix functions
  *****************************************************************************
  * Copyright (C) 2000, 2001 VideoLAN
- * $Id: ac3_downmix.h,v 1.5 2001/03/21 13:42:34 sam Exp $
+ * $Id: ac3_downmix.h,v 1.6 2001/04/30 21:04:20 reno Exp $
  *
  * Authors: Renaud Dartus <reno@videolan.org>
  *
@@ -21,14 +21,7 @@
  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111, USA.
  *****************************************************************************/
 
-#define NORM 16384
-
-typedef struct dm_par_s {
-    float unit;
-    float clev;
-    float slev;
-} dm_par_t;
-
+/* C functions */
 void downmix_3f_2r_to_2ch_c(float *samples, dm_par_t * dm_par);
 void downmix_3f_1r_to_2ch_c(float *samples, dm_par_t * dm_par);
 void downmix_2f_2r_to_2ch_c(float *samples, dm_par_t * dm_par);
@@ -36,3 +29,14 @@ void downmix_2f_1r_to_2ch_c(float *samples, dm_par_t * dm_par);
 void downmix_3f_0r_to_2ch_c(float *samples, dm_par_t * dm_par);            
 void stream_sample_2ch_to_s16_c(s16 *s16_samples, float *left, float *right);
 void stream_sample_1ch_to_s16_c(s16 *s16_samples, float *center); 
+
+#if 0
+/* Kni functions */
+void downmix_3f_2r_to_2ch_kni(float *samples, dm_par_t * dm_par);
+void downmix_3f_1r_to_2ch_kni(float *samples, dm_par_t * dm_par);
+void downmix_2f_2r_to_2ch_kni(float *samples, dm_par_t * dm_par);
+void downmix_2f_1r_to_2ch_kni(float *samples, dm_par_t * dm_par);
+void downmix_3f_0r_to_2ch_kni(float *samples, dm_par_t * dm_par);            
+void stream_sample_2ch_to_s16_kni(s16 *s16_samples, float *left, float *right);
+void stream_sample_1ch_to_s16_kni(s16 *s16_samples, float *center);  
+#endif

src/ac3_decoder/ac3_downmix_c.c

@@ -2,7 +2,7 @@
  * ac3_downmix_c.c: ac3 downmix functions
  *****************************************************************************
  * Copyright (C) 1999, 2000, 2001 VideoLAN
- * $Id: ac3_downmix_c.c,v 1.5 2001/04/20 12:14:34 reno Exp $
+ * $Id: ac3_downmix_c.c,v 1.6 2001/04/30 21:04:20 reno Exp $
  *
  * Authors: Renaud Dartus <reno@videolan.org>
  *          Aaron Holtzman <aholtzma@engr.uvic.ca>
@@ -36,6 +36,7 @@
 #include "ac3_internal.h"
 
 #include "ac3_downmix.h"
+
 void __inline__ downmix_3f_2r_to_2ch_c (float *samples, dm_par_t *dm_par)
 {
     int i;
@@ -137,8 +138,8 @@ void __inline__ stream_sample_2ch_to_s16_c (s16 *out_buf, float *left, float *ri
 {
     int i;
     for (i=0; i < 256; i++) {
-        *out_buf++ = (s16) (*left++  * NORM);
-        *out_buf++ = (s16) (*right++ * NORM);
+        *out_buf++ = (s16) (*left++);
+        *out_buf++ = (s16) (*right++);
     }
 }
 
@@ -149,7 +150,7 @@ void __inline__ stream_sample_1ch_to_s16_c (s16 *out_buf, float *center)
     float tmp;
 
     for (i=0; i < 256; i++) {
-        *out_buf++ = tmp = (s16) (0.7071f * *center++ * NORM);
+        *out_buf++ = tmp = (s16) (0.7071f * *center++);
         *out_buf++ = tmp;
     }
 }

src/ac3_decoder/ac3_imdct_c.h

+int  imdct_init_c (imdct_t * p_imdct);
+void imdct_do_256(imdct_t * p_imdct, float data[], float delay[]);
+void imdct_do_256_nol(imdct_t * p_imdct, float data[], float delay[]);
+void imdct_do_512_c(imdct_t * p_imdct, float data[], float delay[]);
+void imdct_do_512_nol_c(imdct_t * p_imdct, float data[], float delay[]);
+

src/ac3_decoder/ac3_mantissa.c

@@ -2,7 +2,7 @@
  * ac3_mantissa.c: ac3 mantissa computation
  *****************************************************************************
  * Copyright (C) 1999, 2000, 2001 VideoLAN
- * $Id: ac3_mantissa.c,v 1.24 2001/04/26 00:12:19 reno Exp $
+ * $Id: ac3_mantissa.c,v 1.25 2001/04/30 21:04:20 reno Exp $
  *
  * Authors: Michel Kaempf <maxx@via.ecp.fr>
  *          Aaron Holtzman <aholtzma@engr.uvic.ca>
@@ -411,10 +411,10 @@ static __inline__ void uncouple_channel (ac3dec_t * p_ac3dec, u32 ch)
              * so the channels are uncorrelated */
             if (p_ac3dec->audblk.dithflag[ch] && !p_ac3dec->audblk.cpl_bap[i])
             {
-                p_ac3dec->coeffs.fbw[ch][i] = cpl_coord * dither_gen(&p_ac3dec->mantissa) *
+                p_ac3dec->samples[ch][i] = cpl_coord * dither_gen(&p_ac3dec->mantissa) *
                     scale_factor[p_ac3dec->audblk.cpl_exp[i]];
             } else {
-                p_ac3dec->coeffs.fbw[ch][i]  = cpl_coord * p_ac3dec->audblk.cpl_flt[i];
+                p_ac3dec->samples[ch][i]  = cpl_coord * p_ac3dec->audblk.cpl_flt[i];
             }
             i++;
         }
@@ -432,7 +432,7 @@ void mantissa_unpack (ac3dec_t * p_ac3dec)
 
     for (i=0; i< p_ac3dec->bsi.nfchans; i++) {
         for (j=0; j < p_ac3dec->audblk.endmant[i]; j++)
-            p_ac3dec->coeffs.fbw[i][j] = coeff_get_float(p_ac3dec, p_ac3dec->audblk.fbw_bap[i][j],
+            p_ac3dec->samples[i][j] = coeff_get_float(p_ac3dec, p_ac3dec->audblk.fbw_bap[i][j],
                     p_ac3dec->audblk.dithflag[i], p_ac3dec->audblk.fbw_exp[i][j]);
 
         if (p_ac3dec->audblk.cplinu && p_ac3dec->audblk.chincpl[i] && !(done_cpl)) {
@@ -458,7 +458,7 @@ void mantissa_unpack (ac3dec_t * p_ac3dec)
     if (p_ac3dec->bsi.lfeon) {
         /* There are always 7 mantissas for lfe, no dither for lfe */
         for (j=0; j < 7 ; j++)
-            p_ac3dec->coeffs.lfe[j] = coeff_get_float(p_ac3dec, p_ac3dec->audblk.lfe_bap[j],
+            p_ac3dec->samples[5][j] = coeff_get_float(p_ac3dec, p_ac3dec->audblk.lfe_bap[j],
                     0, p_ac3dec->audblk.lfe_exp[j]);
     }
 }

src/ac3_decoder/ac3_rematrix.c

@@ -2,7 +2,7 @@
  * ac3_rematrix.c: ac3 audio rematrixing
  *****************************************************************************
  * Copyright (C) 1999, 2000 VideoLAN
- * $Id: ac3_rematrix.c,v 1.14 2001/04/26 11:23:16 sam Exp $
+ * $Id: ac3_rematrix.c,v 1.15 2001/04/30 21:04:20 reno Exp $
  *
  * Authors: Michel Kaempf <maxx@via.ecp.fr>
  *          Aaron Holtzman <aholtzma@engr.uvic.ca>
@@ -71,10 +71,10 @@ void rematrix (ac3dec_t * p_ac3dec)
         end = min_value(rematrix_band[i].end ,12 * p_ac3dec->audblk.cplbegf + 36);
 
         for (j=start;j < end; j++) {
-            left  = 0.5f * (p_ac3dec->coeffs.fbw[0][j] + p_ac3dec->coeffs.fbw[1][j]);
-            right = 0.5f * (p_ac3dec->coeffs.fbw[0][j] - p_ac3dec->coeffs.fbw[1][j]);
-            p_ac3dec->coeffs.fbw[0][j] = left;
-            p_ac3dec->coeffs.fbw[1][j] = right;
+            left  = 0.5f * (p_ac3dec->samples[0][j] + p_ac3dec->samples[1][j]);
+            right = 0.5f * (p_ac3dec->samples[0][j] - p_ac3dec->samples[1][j]);
+            p_ac3dec->samples[0][j] = left;
+            p_ac3dec->samples[1][j] = right;
         }
     }
 }

src/ac3_decoder/ac3_srfft.c

+#include "defs.h"
+
+#include <math.h>
+#include <stdio.h>
+
+#include "config.h"
+#include "common.h"
+#include "threads.h"
+#include "mtime.h"
+
+#include "stream_control.h"
+#include "input_ext-dec.h"
+
+#include "ac3_decoder.h"
+#include "ac3_srfft.h"
+
+void fft_8 (complex_t *x);
+
+void fft_4(complex_t *x)
+{
+  /* delta_p = 1 here */
+  /* x[k] = sum_{i=0..3} x[i] * w^{i*k}, w=e^{-2*pi/4} 
+   */
+
+  register float yt_r, yt_i, yb_r, yb_i, u_r, u_i, vi_r, vi_i;
+  
+  yt_r = x[0].real;
+  yb_r = yt_r - x[2].real;
+  yt_r += x[2].real;
+
+  u_r = x[1].real;
+  vi_i = x[3].real - u_r;
+  u_r += x[3].real;
+  
+  u_i = x[1].imag;
+  vi_r = u_i - x[3].imag;
+  u_i += x[3].imag;
+
+  yt_i = yt_r;
+  yt_i += u_r;
+  x[0].real = yt_i;
+  yt_r -= u_r;
+  x[2].real = yt_r;
+  yt_i = yb_r;
+  yt_i += vi_r;
+  x[1].real = yt_i;
+  yb_r -= vi_r;
+  x[3].real = yb_r;
+
+  yt_i = x[0].imag;
+  yb_i = yt_i - x[2].imag;
+  yt_i += x[2].imag;
+
+  yt_r = yt_i;
+  yt_r += u_i;
+  x[0].imag = yt_r;
+  yt_i -= u_i;
+  x[2].imag = yt_i;
+  yt_r = yb_i;
+  yt_r += vi_i;
+  x[1].imag = yt_r;
+  yb_i -= vi_i;
+  x[3].imag = yb_i;
+}
+
+
+void fft_8 (complex_t *x)
+{
+  /* delta_p = diag{1, sqrt(i)} here */
+  /* x[k] = sum_{i=0..7} x[i] * w^{i*k}, w=e^{-2*pi/8} 
+   */
+  register float wT1_r, wT1_i, wB1_r, wB1_i, wT2_r, wT2_i, wB2_r, wB2_i;
+  
+  wT1_r = x[1].real;
+  wT1_i = x[1].imag;
+  wB1_r = x[3].real;
+  wB1_i = x[3].imag;
+
+  x[1] = x[2];
+  x[2] = x[4];
+  x[3] = x[6];
+  fft_4(&x[0]);
+
+  
+  /* x[0] x[4] */
+  wT2_r = x[5].real;
+  wT2_r += x[7].real;
+  wT2_r += wT1_r;
+  wT2_r += wB1_r;
+  wT2_i = wT2_r;
+  wT2_r += x[0].real;
+  wT2_i = x[0].real - wT2_i;
+  x[0].real = wT2_r;
+  x[4].real = wT2_i;
+
+  wT2_i = x[5].imag;
+  wT2_i += x[7].imag;
+  wT2_i += wT1_i;
+  wT2_i += wB1_i;
+  wT2_r = wT2_i;
+  wT2_r += x[0].imag;
+  wT2_i = x[0].imag - wT2_i;
+  x[0].imag = wT2_r;
+  x[4].imag = wT2_i;
+  
+  /* x[2] x[6] */
+  wT2_r = x[5].imag;
+  wT2_r -= x[7].imag;
+  wT2_r += wT1_i;
+  wT2_r -= wB1_i;
+  wT2_i = wT2_r;
+  wT2_r += x[2].real;
+  wT2_i = x[2].real - wT2_i;
+  x[2].real = wT2_r;
+  x[6].real = wT2_i;
+
+  wT2_i = x[5].real;
+  wT2_i -= x[7].real;
+  wT2_i += wT1_r;
+  wT2_i -= wB1_r;
+  wT2_r = wT2_i;
+  wT2_r += x[2].imag;
+  wT2_i = x[2].imag - wT2_i;
+  x[2].imag = wT2_i;
+  x[6].imag = wT2_r;
+  
+
+  /* x[1] x[5] */
+  wT2_r = wT1_r;
+  wT2_r += wB1_i;
+  wT2_r -= x[5].real;
+  wT2_r -= x[7].imag;
+  wT2_i = wT1_i;
+  wT2_i -= wB1_r;
+  wT2_i -= x[5].imag;
+  wT2_i += x[7].real;
+
+  wB2_r = wT2_r;
+  wB2_r += wT2_i;
+  wT2_i -= wT2_r;
+  wB2_r *= HSQRT2;
+  wT2_i *= HSQRT2;
+  wT2_r = wB2_r;
+  wB2_r += x[1].real;
+  wT2_r =  x[1].real - wT2_r;
+
+  wB2_i = x[5].real;
+  x[1].real = wB2_r;
+  x[5].real = wT2_r;
+
+  wT2_r = wT2_i;
+  wT2_r += x[1].imag;
+  wT2_i = x[1].imag - wT2_i;
+  wB2_r = x[5].imag;
+  x[1].imag = wT2_r;
+  x[5].imag = wT2_i;
+
+  /* x[3] x[7] */
+  wT1_r -= wB1_i;
+  wT1_i += wB1_r;
+  wB1_r = wB2_i - x[7].imag;
+  wB1_i = wB2_r + x[7].real;
+  wT1_r -= wB1_r;
+  wT1_i -= wB1_i;
+  wB1_r = wT1_r + wT1_i;
+  wB1_r *= HSQRT2;
+  wT1_i -= wT1_r;
+  wT1_i *= HSQRT2;
+  wB2_r = x[3].real;
+  wB2_i = wB2_r + wT1_i;
+  wB2_r -= wT1_i;
+  x[3].real = wB2_i;
+  x[7].real = wB2_r;
+  wB2_i = x[3].imag;
+  wB2_r = wB2_i + wB1_r;
+  wB2_i -= wB1_r;
+  x[3].imag = wB2_i;
+  x[7].imag = wB2_r;
+}
+
+
+void fft_asmb(int k, complex_t *x, complex_t *wTB,
+	     const complex_t *d, const complex_t *d_3)
+{
+  register complex_t  *x2k, *x3k, *x4k, *wB;
+  register float a_r, a_i, a1_r, a1_i, u_r, u_i, v_r, v_i;
+
+  x2k = x + 2 * k;
+  x3k = x2k + 2 * k;
+  x4k = x3k + 2 * k;
+  wB = wTB + 2 * k;
+  
+  TRANSZERO(x[0],x2k[0],x3k[0],x4k[0]);
+  TRANS(x[1],x2k[1],x3k[1],x4k[1],wTB[1],wB[1],d[1],d_3[1]);
+  
+  --k;
+  for(;;) {
+     TRANS(x[2],x2k[2],x3k[2],x4k[2],wTB[2],wB[2],d[2],d_3[2]);
+     TRANS(x[3],x2k[3],x3k[3],x4k[3],wTB[3],wB[3],d[3],d_3[3]);
+     if (!--k) break;
+     x += 2;
+     x2k += 2;
+     x3k += 2;
+     x4k += 2;
+     d += 2;
+     d_3 += 2;
+     wTB += 2;
+     wB += 2;
+  }
+ 
+}
+
+void fft_asmb16(complex_t *x, complex_t *wTB)
+{
+  register float a_r, a_i, a1_r, a1_i, u_r, u_i, v_r, v_i;
+  int k = 2;
+
+  /* transform x[0], x[8], x[4], x[12] */
+  TRANSZERO(x[0],x[4],x[8],x[12]);
+
+  /* transform x[1], x[9], x[5], x[13] */
+  TRANS(x[1],x[5],x[9],x[13],wTB[1],wTB[5],delta16[1],delta16_3[1]);
+
+  /* transform x[2], x[10], x[6], x[14] */
+  TRANSHALF_16(x[2],x[6],x[10],x[14]);
+
+  /* transform x[3], x[11], x[7], x[15] */
+  TRANS(x[3],x[7],x[11],x[15],wTB[3],wTB[7],delta16[3],delta16_3[3]);
+
+} 
+
+
+void fft_64p_c (complex_t *a)
+{
+  fft_8(&a[0]); fft_4(&a[8]); fft_4(&a[12]);
+  fft_asmb16(&a[0], &a[8]);
+  
+  fft_8(&a[16]), fft_8(&a[24]);
+  fft_asmb(4, &a[0], &a[16],&delta32[0], &delta32_3[0]);
+
+  fft_8(&a[32]); fft_4(&a[40]); fft_4(&a[44]);
+  fft_asmb16(&a[32], &a[40]);
+
+  fft_8(&a[48]); fft_4(&a[56]); fft_4(&a[60]);
+  fft_asmb16(&a[48], &a[56]);
+
+  fft_asmb(8, &a[0], &a[32],&delta64[0], &delta64_3[0]);
+}
+
+
+void fft_128p_c (complex_t *a)
+{
+  fft_8(&a[0]); fft_4(&a[8]); fft_4(&a[12]);
+  fft_asmb16(&a[0], &a[8]);
+  
+  fft_8(&a[16]), fft_8(&a[24]);
+  fft_asmb(4, &a[0], &a[16],&delta32[0], &delta32_3[0]);
+
+  fft_8(&a[32]); fft_4(&a[40]); fft_4(&a[44]);
+  fft_asmb16(&a[32], &a[40]);
+
+  fft_8(&a[48]); fft_4(&a[56]); fft_4(&a[60]);
+  fft_asmb16(&a[48], &a[56]);
+
+  fft_asmb(8, &a[0], &a[32],&delta64[0], &delta64_3[0]);
+
+  fft_8(&a[64]); fft_4(&a[72]); fft_4(&a[76]);
+  /* fft_16(&a[64]); */
+  fft_asmb16(&a[64], &a[72]);
+
+  fft_8(&a[80]); fft_8(&a[88]);
+  
+  /* fft_32(&a[64]); */
+  fft_asmb(4, &a[64], &a[80],&delta32[0], &delta32_3[0]);
+
+  fft_8(&a[96]); fft_4(&a[104]), fft_4(&a[108]);
+  /* fft_16(&a[96]); */
+  fft_asmb16(&a[96], &a[104]);
+
+  fft_8(&a[112]), fft_8(&a[120]);
+  /* fft_32(&a[96]); */
+  fft_asmb(4, &a[96], &a[112], &delta32[0], &delta32_3[0]);
+  
+  /* fft_128(&a[0]); */
+  fft_asmb(16, &a[0], &a[64], &delta128[0], &delta128_3[0]);
+}

src/ac3_decoder/ac3_srfft.h

+static complex_t delta16[4] = 
+ { {1.00000000000000,  0.00000000000000},
+   {0.92387953251129, -0.38268343236509},
+   {0.70710678118655, -0.70710678118655},
+   {0.38268343236509, -0.92387953251129}};
+
+static complex_t delta16_3[4] = 
+ { {1.00000000000000,  0.00000000000000},
+   {0.38268343236509, -0.92387953251129},
+   {-0.70710678118655, -0.70710678118655},
+   {-0.92387953251129, 0.38268343236509}};
+
+static complex_t delta32[8] = 
+ { {1.00000000000000,  0.00000000000000},
+   {0.98078528040323, -0.19509032201613},
+   {0.92387953251129, -0.38268343236509},
+   {0.83146961230255, -0.55557023301960},
+   {0.70710678118655, -0.70710678118655},
+   {0.55557023301960, -0.83146961230255},
+   {0.38268343236509, -0.92387953251129},
+   {0.19509032201613, -0.98078528040323}};
+
+static complex_t delta32_3[8] = 
+ { {1.00000000000000,  0.00000000000000},
+   {0.83146961230255, -0.55557023301960},
+   {0.38268343236509, -0.92387953251129},
+   {-0.19509032201613, -0.98078528040323},
+   {-0.70710678118655, -0.70710678118655},
+   {-0.98078528040323, -0.19509032201613},
+   {-0.92387953251129, 0.38268343236509},
+   {-0.55557023301960, 0.83146961230255}};
+
+static complex_t delta64[16] = 
+ { {1.00000000000000,  0.00000000000000},
+   {0.99518472667220, -0.09801714032956},
+   {0.98078528040323, -0.19509032201613},
+   {0.95694033573221, -0.29028467725446},
+   {0.92387953251129, -0.38268343236509},
+   {0.88192126434836, -0.47139673682600},
+   {0.83146961230255, -0.55557023301960},
+   {0.77301045336274, -0.63439328416365},
+   {0.70710678118655, -0.70710678118655},
+   {0.63439328416365, -0.77301045336274},
+   {0.55557023301960, -0.83146961230255},
+   {0.47139673682600, -0.88192126434835},
+   {0.38268343236509, -0.92387953251129},
+   {0.29028467725446, -0.95694033573221},
+   {0.19509032201613, -0.98078528040323},
+   {0.09801714032956, -0.99518472667220}};
+
+static complex_t delta64_3[16] = 
+ { {1.00000000000000,  0.00000000000000},
+   {0.95694033573221, -0.29028467725446},
+   {0.83146961230255, -0.55557023301960},
+   {0.63439328416365, -0.77301045336274},
+   {0.38268343236509, -0.92387953251129},
+   {0.09801714032956, -0.99518472667220},
+   {-0.19509032201613, -0.98078528040323},
+   {-0.47139673682600, -0.88192126434836},
+   {-0.70710678118655, -0.70710678118655},
+   {-0.88192126434835, -0.47139673682600},
+   {-0.98078528040323, -0.19509032201613},
+   {-0.99518472667220, 0.09801714032956},
+   {-0.92387953251129, 0.38268343236509},
+   {-0.77301045336274, 0.63439328416365},
+   {-0.55557023301960, 0.83146961230255},
+   {-0.29028467725446, 0.95694033573221}};
+
+static complex_t delta128[32] = 
+ { {1.00000000000000,  0.00000000000000},
+   {0.99879545620517, -0.04906767432742},
+   {0.99518472667220, -0.09801714032956},
+   {0.98917650996478, -0.14673047445536},
+   {0.98078528040323, -0.19509032201613},
+   {0.97003125319454, -0.24298017990326},
+   {0.95694033573221, -0.29028467725446},
+   {0.94154406518302, -0.33688985339222},
+   {0.92387953251129, -0.38268343236509},
+   {0.90398929312344, -0.42755509343028},
+   {0.88192126434836, -0.47139673682600},
+   {0.85772861000027, -0.51410274419322},
+   {0.83146961230255, -0.55557023301960},
+   {0.80320753148064, -0.59569930449243},
+   {0.77301045336274, -0.63439328416365},
+   {0.74095112535496, -0.67155895484702},
+   {0.70710678118655, -0.70710678118655},
+   {0.67155895484702, -0.74095112535496},
+   {0.63439328416365, -0.77301045336274},
+   {0.59569930449243, -0.80320753148064},
+   {0.55557023301960, -0.83146961230255},
+   {0.51410274419322, -0.85772861000027},
+   {0.47139673682600, -0.88192126434835},
+   {0.42755509343028, -0.90398929312344},
+   {0.38268343236509, -0.92387953251129},
+   {0.33688985339222, -0.94154406518302},
+   {0.29028467725446, -0.95694033573221},
+   {0.24298017990326, -0.97003125319454},
+   {0.19509032201613, -0.98078528040323},
+   {0.14673047445536, -0.98917650996478},
+   {0.09801714032956, -0.99518472667220},
+   {0.04906767432742, -0.99879545620517}};
+
+static complex_t delta128_3[32] = 
+ { {1.00000000000000,  0.00000000000000},
+   {0.98917650996478, -0.14673047445536},
+   {0.95694033573221, -0.29028467725446},
+   {0.90398929312344, -0.42755509343028},
+   {0.83146961230255, -0.55557023301960},
+   {0.74095112535496, -0.67155895484702},
+   {0.63439328416365, -0.77301045336274},
+   {0.51410274419322, -0.85772861000027},
+   {0.38268343236509, -0.92387953251129},
+   {0.24298017990326, -0.97003125319454},
+   {0.09801714032956, -0.99518472667220},
+   {-0.04906767432742, -0.99879545620517},
+   {-0.19509032201613, -0.98078528040323},
+   {-0.33688985339222, -0.94154406518302},
+   {-0.47139673682600, -0.88192126434836},
+   {-0.59569930449243, -0.80320753148065},
+   {-0.70710678118655, -0.70710678118655},
+   {-0.80320753148065, -0.59569930449243},
+   {-0.88192126434835, -0.47139673682600},
+   {-0.94154406518302, -0.33688985339222},
+   {-0.98078528040323, -0.19509032201613},
+   {-0.99879545620517, -0.04906767432742},
+   {-0.99518472667220, 0.09801714032956},
+   {-0.97003125319454, 0.24298017990326},
+   {-0.92387953251129, 0.38268343236509},
+   {-0.85772861000027, 0.51410274419322},
+   {-0.77301045336274, 0.63439328416365},
+   {-0.67155895484702, 0.74095112535496},
+   {-0.55557023301960, 0.83146961230255},
+   {-0.42755509343028, 0.90398929312344},
+   {-0.29028467725446, 0.95694033573221},
+   {-0.14673047445536, 0.98917650996478}};
+
+#define HSQRT2 0.707106781188;
+
+#define TRANSZERO(A0,A4,A8,A12) { \
+  u_r = wTB[0].real; \
+  v_i = u_r - wTB[k*2].real; \
+  u_r += wTB[k*2].real; \
+  u_i = wTB[0].imag; \
+  v_r = wTB[k*2].imag - u_i; \
+  u_i += wTB[k*2].imag; \
+  a_r = A0.real; \
+  a_i = A0.imag; \
+  a1_r = a_r; \
+  a1_r += u_r; \
+  A0.real = a1_r; \
+  a_r -= u_r; \
+  A8.real = a_r; \
+  a1_i = a_i; \
+  a1_i += u_i; \
+  A0.imag = a1_i; \
+  a_i -= u_i; \
+  A8.imag = a_i; \
+  a1_r = A4.real; \
+  a1_i = A4.imag; \
+  a_r = a1_r; \
+  a_r -= v_r; \
+  A4.real = a_r; \
+  a1_r += v_r; \
+  A12.real = a1_r; \
+  a_i = a1_i; \
+  a_i -= v_i; \
+  A4.imag = a_i; \
+  a1_i += v_i; \
+  A12.imag = a1_i; \
+  }
+
+#define TRANSHALF_16(A2,A6,A10,A14) {\
+  u_r = wTB[2].real; \
+  a_r = u_r; \
+  u_i = wTB[2].imag; \
+  u_r += u_i; \
+  u_i -= a_r; \
+  a_r = wTB[6].real; \
+  a1_r = a_r; \
+  a_i = wTB[6].imag; \
+  a_r = a_i - a_r; \
+  a_i += a1_r; \
+  v_i = u_r - a_r; \
+  u_r += a_r; \
+  v_r = u_i + a_i; \
+  u_i -= a_i; \
+  v_i *= HSQRT2; \
+  v_r *= HSQRT2; \
+  u_r *= HSQRT2; \
+  u_i *= HSQRT2; \
+  a_r = A2.real; \
+  a_i = A2.imag; \
+  a1_r = a_r; \
+  a1_r += u_r; \
+  A2.real = a1_r; \
+  a_r -= u_r; \
+  A10.real = a_r; \
+  a1_i = a_i; \
+  a1_i += u_i; \
+  A2.imag = a1_i; \
+  a_i -= u_i; \
+  A10.imag = a_i; \
+  a1_r = A6.real; \
+  a1_i = A6.imag;  \
+  a_r = a1_r; \
+  a1_r += v_r; \
+  A6.real = a1_r; \
+  a_r -= v_r; \
+  A14.real = a_r; \
+  a_i = a1_i; \
+  a1_i -= v_i; \
+  A6.imag = a1_i; \
+  a_i += v_i; \
+  A14.imag = a_i; \
+  }
+
+#define TRANS(A1,A5,A9,A13,WT,WB,D,D3) { \
+  u_r = WT.real; \
+  a_r = u_r; \
+  a_r *= D.imag; \
+  u_r *= D.real; \
+  a_i = WT.imag; \
+  a1_i = a_i; \
+  a1_i *= D.real; \
+  a_i *= D.imag; \
+  u_r -= a_i; \
+  u_i = a_r; \
+  u_i += a1_i; \
+  a_r = WB.real; \
+  a1_r = a_r; \
+  a1_r *= D3.real; \
+  a_r *= D3.imag; \
+  a_i = WB.imag; \
+  a1_i = a_i; \
+  a_i *= D3.real; \
+  a1_i *= D3.imag; \
+  a1_r -= a1_i; \
+  a_r += a_i; \
+  v_i = u_r - a1_r; \
+  u_r += a1_r; \
+  v_r = a_r - u_i; \
+  u_i += a_r; \
+  a_r = A1.real; \
+  a_i = A1.imag; \
+  a1_r = a_r; \
+  a1_r += u_r; \
+  A1.real = a1_r; \
+  a_r -= u_r; \
+  A9.real = a_r; \
+  a1_i = a_i; \
+  a1_i += u_i; \
+  A1.imag = a1_i; \
+  a_i -= u_i; \
+  A9.imag = a_i; \
+  a1_r = A5.real; \
+  a1_i = A5.imag;  \
+  a_r = a1_r; \
+  a1_r -= v_r; \
+  A5.real = a1_r; \
+  a_r += v_r; \
+  A13.real = a_r; \
+  a_i = a1_i; \
+  a1_i -= v_i; \
+  A5.imag = a1_i; \
+  a_i += v_i; \
+  A13.imag = a_i; \
+  }