Increase wmapro decoder output sample depth to 24 bits, makes the mean error between our output and the original floating point decoder ~5e-8. (FS#11498 by Buschel)

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@27703 a1c6a512-1295-4272-9138-f99709370657

4 files changed, 19 insertions, 16 deletions

diff --git a/apps/codecs/libwmapro/wma.h b/apps/codecs/libwmapro/wma.h
index 84abeb349b..6fa7b94724 100644
--- a/apps/codecs/libwmapro/wma.h
+++ b/apps/codecs/libwmapro/wma.h
@@ -26,6 +26,7 @@
 #include "ffmpeg_put_bits.h"
 
 #define WMAPRO_FRACT (17)
+#define WMAPRO_DSP_SAMPLE_DEPTH (WMAPRO_FRACT + 8)
 
 /* size of blocks */
 #define BLOCK_MIN_BITS 7
diff --git a/apps/codecs/libwmapro/wmapro_math.h b/apps/codecs/libwmapro/wmapro_math.h
index f5cfa5f1d9..17b4829362 100644
--- a/apps/codecs/libwmapro/wmapro_math.h
+++ b/apps/codecs/libwmapro/wmapro_math.h
@@ -223,21 +223,22 @@ static inline void vector_fixmul_window(int32_t *dst, const int32_t *src0,
         asm volatile ( \
             "ldmia %[src]!, {r1-r4}   \n\t" \
             "smull r0, r5, r1, %[mul] \n\t" \
-            "mov   r0, r0, lsr #24    \n\t" \
-            "orr   r0, r0, r5, lsl #8 \n\t" \
+            "mov   r0, r0, lsr #16    \n\t" \
+            "orr   r0, r0, r5, lsl #16\n\t" \
             "smull r1, r5, r2, %[mul] \n\t" \
-            "mov   r1, r1, lsr #24    \n\t" \
-            "orr   r1, r1, r5, lsl #8 \n\t" \
+            "mov   r1, r1, lsr #16    \n\t" \
+            "orr   r1, r1, r5, lsl #16\n\t" \
             "smull r2, r5, r3, %[mul] \n\t" \
-            "mov   r2, r2, lsr #24    \n\t" \
-            "orr   r2, r2, r5, lsl #8 \n\t" \
+            "mov   r2, r2, lsr #16    \n\t" \
+            "orr   r2, r2, r5, lsl #16\n\t" \
             "smull r3, r5, r4, %[mul] \n\t" \
-            "mov   r3, r3, lsr #24    \n\t" \
-            "orr   r3, r3, r5, lsl #8 \n\t" \
+            "mov   r3, r3, lsr #16    \n\t" \
+            "orr   r3, r3, r5, lsl #16\n\t" \
             "stmia %[dst]!, {r0-r3}   \n"   \
             : [dst]"+r"(dst), [src]"+r"(src) \
             : [mul]"r"(mul) \
             : "r0", "r1", "r2", "r3", "r4", "r5", "memory");
+/* Disable ColdFire version until a correct version is written 
 #elif defined (CPU_COLDFIRE)
     #define VECT_MUL_SCALAR_KERNEL(dst, src, mul) \
         int32_t tmp; \
@@ -270,12 +271,13 @@ static inline void vector_fixmul_window(int32_t *dst, const int32_t *src0,
               [tmp] "=d" (tmp)                   \
             : [mul] "r" (mul)                    \
             : "d0", "d1", "d2", "d3", "memory", "cc");
+*/
 #else
     #define VECT_MUL_SCALAR_KERNEL(dst, src, mul) \
-        dst[i  ] = fixmul24(src[i  ], mul); \
-        dst[i+1] = fixmul24(src[i+1], mul); \
-        dst[i+2] = fixmul24(src[i+2], mul); \
-        dst[i+3] = fixmul24(src[i+3], mul);
+        dst[i  ] = fixmul16(src[i  ], mul); \
+        dst[i+1] = fixmul16(src[i+1], mul); \
+        dst[i+2] = fixmul16(src[i+2], mul); \
+        dst[i+3] = fixmul16(src[i+3], mul);
 #endif /* CPU_ARM, CPU_COLDFIRE */
 
 static inline void vector_fixmul_scalar(int32_t *dst, const int32_t *src, 
diff --git a/apps/codecs/libwmapro/wmaprodec.c b/apps/codecs/libwmapro/wmaprodec.c
index e5a5fd94eb..018ea7c304 100644
--- a/apps/codecs/libwmapro/wmaprodec.c
+++ b/apps/codecs/libwmapro/wmaprodec.c
@@ -125,7 +125,7 @@
 #define FFMAX(a,b) ((a) > (b) ? (a) : (b))
 
 /* Define some multiple used constants */
-#define SQRT2_FRACT24   0x016A09E6 /* 0x016A09E6 = (sqrt(2)*(1<<24)) */
+#define SQRT2_FRACT16   0x00016A0A /* 0x00016A0A = (sqrt(2)*(1<<16)) */
 #define COS_PI4_FRACT16 0x0000B505 /* 0x0000B505 = (cos(pi/4)<<16) */
 #define ONE_FRACT16     0x00010000 /* 0x00010000 = (1<<16) */
 
@@ -1058,10 +1058,10 @@ static void inverse_channel_transform(WMAProDecodeCtx *s)
                     int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
                     vector_fixmul_scalar(ch_data[0] + sfb[0],
                                          ch_data[0] + sfb[0],
-                                         SQRT2_FRACT24, len);
+                                         SQRT2_FRACT16, len);
                     vector_fixmul_scalar(ch_data[1] + sfb[0],
                                          ch_data[1] + sfb[0],
-                                         SQRT2_FRACT24, len);
+                                         SQRT2_FRACT16, len);
 
                 }
             }
diff --git a/apps/codecs/wmapro.c b/apps/codecs/wmapro.c
index 30e60e5716..75bbd24cda 100644
--- a/apps/codecs/wmapro.c
+++ b/apps/codecs/wmapro.c
@@ -44,7 +44,7 @@ enum codec_status codec_main(void)
     int size;                   /* Size of the input frame to the decoder */
 
     /* Generic codec initialisation */
-    ci->configure(DSP_SET_SAMPLE_DEPTH, WMAPRO_FRACT);
+    ci->configure(DSP_SET_SAMPLE_DEPTH, WMAPRO_DSP_SAMPLE_DEPTH);
     
 
 next_track: