Move MDCT reconstruction window code over to new trig function. Improves accuracy significantly and slightly reduces code size. Codec SNR now appears to be limited by truncation to 16 bit. Comparison to MS decoder gives > 91 dB of agreement, and a lower RMS error verses the source wav then MS. Additionally, move one commonly accessed table into IRAM.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@13813 a1c6a512-1295-4272-9138-f99709370657

3 files changed, 16 insertions, 45 deletions

diff --git a/apps/codecs/libwma/wmadec.h b/apps/codecs/libwma/wmadec.h
index 0de50795c4..e3e5d389ae 100644
--- a/apps/codecs/libwma/wmadec.h
+++ b/apps/codecs/libwma/wmadec.h
@@ -131,7 +131,7 @@ typedef struct WMADecodeContext
     fixed32 coefs[MAX_CHANNELS][BLOCK_MAX_SIZE];
     MDCTContext mdct_ctx[BLOCK_NB_SIZES];
     fixed32 *windows[BLOCK_NB_SIZES];
-    FFTComplex mdct_tmp[BLOCK_MAX_SIZE]; /* temporary storage for imdct */
+    FFTComplex *mdct_tmp; /* temporary storage for imdct */
     /* output buffer for one frame and the last for IMDCT windowing */
     fixed32 frame_out[MAX_CHANNELS][BLOCK_MAX_SIZE * 2];
     /* last frame info */
diff --git a/apps/codecs/libwma/wmadeci.c b/apps/codecs/libwma/wmadeci.c
index 1857e6a70d..bb9b96abe0 100644
--- a/apps/codecs/libwma/wmadeci.c
+++ b/apps/codecs/libwma/wmadeci.c
@@ -114,7 +114,9 @@ fixed32 tcos0[1024], tcos1[512], tcos2[256], tcos3[128], tcos4[64];        //the
 fixed32 tsin0[1024], tsin1[512], tsin2[256], tsin3[128], tsin4[64];
 
 FFTComplex *exparray[5];                                    //these are the fft lookup tables
+
 uint16_t *revarray[5];
+
 FFTComplex  exptab0[512] IBSS_ATTR;//, exptab1[256], exptab2[128], exptab3[64], exptab4[32];    //folded these in!
 uint16_t revtab0[1024], revtab1[512], revtab2[256], revtab3[128], revtab4[64];
 
@@ -122,6 +124,7 @@ uint16_t *runtabarray[2], *levtabarray[2];
 
 uint16_t runtab0[1336], runtab1[1336], levtab0[1336], levtab1[1336];                //these could be made smaller since only one can be 1336
 
+FFTComplex mdct_tmp[BLOCK_MAX_SIZE] IBSS_ATTR;                  /* temporary storage for imdct */
 
 //may also be too large by ~ 1KB each?
 static VLC_TYPE vlcbuf1[6144][2];
@@ -1080,6 +1083,7 @@ int wma_decode_init(WMADecodeContext* s, asf_waveformatex_t *wfx)
     exparray[0] = exptab0; //exparray[1] = exptab1; exparray[2] = exptab2; exparray[3] = exptab3; exparray[4] = exptab4;
     revarray[0]=revtab0; revarray[1]=revtab1; revarray[2]=revtab2; revarray[3]=revtab3; revarray[4]=revtab4;
 
+        s->mdct_tmp = mdct_tmp; /* temporary storage for imdct */
     for(i = 0; i < s->nb_block_sizes; ++i)
     {
         ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1);
@@ -1108,12 +1112,13 @@ int wma_decode_init(WMADecodeContext* s, asf_waveformatex_t *wfx)
 
         //fixed32 n2 = itofix32(n<<1);        //2x the window length
         //alpha = fixdiv32(M_PI_F, n2);        //PI / (2x Window length) == PI<<(s->frame_len_bits - i+1)
-        //printf("two values of alpha %16.10lf %16.10lf\n", fixtof64(alpha), fixtof64(M_PI_F>>(s->frame_len_bits - i+1)));
-        alpha = M_PI_F>>(s->frame_len_bits - i+1);
+
+        //alpha = M_PI_F>>(s->frame_len_bits - i+1);
+        alpha = (1<<15)>>(s->frame_len_bits - i+1);     /* this calculates 0.5/(2*n) */
         for(j=0;j<n;++j)
         {
             fixed32 j2 = itofix32(j) + 0x8000;
-            window[j] = fixsin32(fixmul32(j2,alpha));        //alpha between 0 and pi/2
+             window[j] = fsincos(fixmul32(j2,alpha)<<16, 0);        //alpha between 0 and pi/2
 
         }
         //printf("created window\n");
@@ -1192,43 +1197,7 @@ int wma_decode_init(WMADecodeContext* s, asf_waveformatex_t *wfx)
     return 0;
 }
 
-#if 0
-/* interpolate values for a bigger or smaller block. The block must
-   have multiple sizes */
-static void interpolate_array(fixed32 *scale, int old_size, int new_size)
-{
-    int i, j, jincr, k;
-    fixed32 v;
-
 
-
-    if (new_size > old_size)
-    {
-        jincr = new_size / old_size;
-        j = new_size;
-        for(i = old_size - 1; i >=0; --i)
-        {
-            v = scale[i];
-            k = jincr;
-            do
-            {
-                scale[--j] = v;
-            }
-            while (--k);
-        }
-    }
-    else if (new_size < old_size)
-    {
-        j = 0;
-        jincr = old_size / new_size;
-        for(i = 0; i < new_size; ++i)
-        {
-            scale[i] = scale[j];
-            j += jincr;
-        }
-    }
-}
-#endif
 /* compute x^-0.25 with an exponent and mantissa table. We use linear
    interpolation to reduce the mantissa table size at a small speed
    expense (linear interpolation approximately doubles the number of
@@ -1958,9 +1927,9 @@ static int wma_decode_frame(WMADecodeContext *s, int16_t *samples)
 }
 
 int wma_decode_superframe(WMADecodeContext* s,
-                                 void *data,
+                                 void *data,    /*output*/
                                  int *data_size,
-                                 uint8_t *buf,
+                                 uint8_t *buf,  /*input*/
                                  int buf_size)
 {
     //WMADecodeContext *s = avctx->priv_data;
diff --git a/apps/codecs/libwma/wmafixed.c b/apps/codecs/libwma/wmafixed.c
index 7c38009754..3a902dd9ef 100644
--- a/apps/codecs/libwma/wmafixed.c
+++ b/apps/codecs/libwma/wmafixed.c
@@ -68,7 +68,7 @@ fixed32 fixmul32(fixed32 x, fixed32 y)
     return (fixed32)temp;
 }
 
-
+#endif
 /*
         Special fixmul32 that does a 16.16 x 1.31 multiply that returns a 16.16 value.
         this is needed because the fft constants are all normalized to be less then 1
@@ -76,7 +76,7 @@ fixed32 fixmul32(fixed32 x, fixed32 y)
 
 
 */
-
+#ifndef CPU_ARM
 fixed32 fixmul32b(fixed32 x, fixed32 y)
 {
     fixed64 temp;
@@ -88,10 +88,10 @@ fixed32 fixmul32b(fixed32 x, fixed32 y)
 
     return (fixed32)temp;
 }
-
 #endif
 
 
+
 /*
         Not performance senstitive code here
 
@@ -275,6 +275,7 @@ long fsincos(unsigned long phase, fixed32 *cos)
 
 */
 
+#if 0
 fixed32 fixsin32(fixed32 x)
 {
 
@@ -325,3 +326,4 @@ fixed32 fixcos32(fixed32 x)
 {
     return fixsin32(x - (M_PI_F>>1))*-1;
 }
+#endif