Major optimization of atrac3 codec for Coldfire targets. Moving several number cruncher arrays to IRAM. Decoder is sped up by +50% on h300, decoder now finally is realtime on Coldfire targets as well.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@24677 a1c6a512-1295-4272-9138-f99709370657

2 files changed, 20 insertions, 8 deletions

diff --git a/apps/codecs/libatrac/atrac3.c b/apps/codecs/libatrac/atrac3.c
index b2f675ef18..3c6ecc9197 100644
--- a/apps/codecs/libatrac/atrac3.c
+++ b/apps/codecs/libatrac/atrac3.c
@@ -55,9 +55,12 @@
 #define FFMIN(a,b) ((a) > (b) ? (b) : (a))
 #define FFSWAP(type,a,b) do{type SWAP_tmp= b; b= a; a= SWAP_tmp;}while(0)
 
-static int32_t          qmf_window[48] IBSS_ATTR;
-static VLC              spectral_coeff_tab[7];
-static channel_unit     channel_units[2] IBSS_ATTR_LARGE_IRAM;
+static VLC          spectral_coeff_tab[7];
+static int32_t      qmf_window[48] IBSS_ATTR;
+static int32_t      atrac3_spectrum [2][1024] IBSS_ATTR __attribute__((aligned(16)));
+static int32_t      atrac3_IMDCT_buf[2][ 512] IBSS_ATTR __attribute__((aligned(16)));
+static int32_t      atrac3_prevFrame[2][1024] IBSS_ATTR;
+static channel_unit channel_units[2] IBSS_ATTR_LARGE_IRAM;
 
 
 /**
@@ -425,7 +428,7 @@ static void inverseQuantizeSpectrum(int *mantissas, int32_t *pOut,
  * @return outSubbands  subband counter, fix for broken specification/files
  */
 
-int decodeSpectrum (GetBitContext *gb, int32_t *pOut) ICODE_ATTR;
+int decodeSpectrum (GetBitContext *gb, int32_t *pOut) ICODE_ATTR_LARGE_IRAM;
 int decodeSpectrum (GetBitContext *gb, int32_t *pOut)
 {
     int   numSubbands, codingMode, cnt, first, last, subbWidth;
@@ -1228,7 +1231,14 @@ int atrac3_decode_init(ATRAC3Context *q, RMContext *rmctx)
         q->matrix_coeff_index_next[i] = 3;
     }
    
+    /* Link the iram'ed arrays to the decoder's data structure */
     q->pUnits = channel_units;
+    q->pUnits[0].spectrum  = &atrac3_spectrum [0][0];
+    q->pUnits[1].spectrum  = &atrac3_spectrum [1][0];
+    q->pUnits[0].IMDCT_buf = &atrac3_IMDCT_buf[0][0];
+    q->pUnits[1].IMDCT_buf = &atrac3_IMDCT_buf[1][0];
+    q->pUnits[0].prevFrame = &atrac3_prevFrame[0][0];
+    q->pUnits[1].prevFrame = &atrac3_prevFrame[1][0];
     
     return 0;
 }
diff --git a/apps/codecs/libatrac/atrac3.h b/apps/codecs/libatrac/atrac3.h
index a817db2b55..5bed9c60bf 100644
--- a/apps/codecs/libatrac/atrac3.h
+++ b/apps/codecs/libatrac/atrac3.h
@@ -3,10 +3,12 @@
 
 #if (CONFIG_CPU == PP5022) || (CONFIG_CPU == PP5024) || (CONFIG_CPU == MCF5250)
 /* PP5022/24 and MCF5250 have larger IRAM */
-#define IBSS_ATTR_LARGE_IRAM IBSS_ATTR
+#define IBSS_ATTR_LARGE_IRAM  IBSS_ATTR
+#define ICODE_ATTR_LARGE_IRAM ICODE_ATTR
 #else
 /* other CPUs IRAM is not large enough */
 #define IBSS_ATTR_LARGE_IRAM
+#define ICODE_ATTR_LARGE_IRAM
 #endif
 
 /* These structures are needed to store the parsed gain control data. */
@@ -30,12 +32,12 @@ typedef struct {
     int               bandsCoded;
     int               numComponents;
     tonal_component   components[64];
-    int32_t           prevFrame[1024];
+    int32_t           *prevFrame;
     int               gcBlkSwitch;
     gain_block        gainBlock[2];
 
-    int32_t           spectrum[1024] __attribute__((aligned(16)));
-    int32_t           IMDCT_buf[1024] __attribute__((aligned(16)));
+    int32_t           *spectrum;
+    int32_t           *IMDCT_buf;
 
     int32_t           delayBuf1[46]; ///<qmf delay buffers
     int32_t           delayBuf2[46];