Use more IRAM on S5L870x to speed up wma by ~3%.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@29905 a1c6a512-1295-4272-9138-f99709370657

3 files changed, 36 insertions, 19 deletions

diff --git a/apps/codecs/libwma/wmadata.h b/apps/codecs/libwma/wmadata.h
index 343b6e9e92..07a55df19a 100644
--- a/apps/codecs/libwma/wmadata.h
+++ b/apps/codecs/libwma/wmadata.h
@@ -1426,7 +1426,7 @@ static const CoefVLCTable coef_vlcs[6] = {
 };
 
 /*table of the values of 10^(index*.05)*/
-const fixed64 pow_table[] =
+const fixed64 pow_table[] ICONST_ATTR_WMA_XL_IRAM =
     {   0x199a, 0x1cb9, 0x203a, 0x2429, 0x2893, 0x2d86, 0x3314, 0x3950, 0x404e,
         0x4827, 0x50f4, 0x5ad5, 0x65ea, 0x725a, 0x804e, 0x8ff6, 0xa186, 0xb53c,
         0xcb59, 0xe429, 0x10000LL,0x11f3dLL,0x14249LL,0x1699cLL,0x195bcLL,
@@ -1499,7 +1499,7 @@ const fixed32 pow_10_to_yover16[] ICONST_ATTR=
     0x27100000, 0x2d1bd1e1, 0x341736de, 0x3c2743e8, 0x4576cb4a
 };
 
-const fixed32 pow_a_table[] =
+const fixed32 pow_a_table[] ICONST_ATTR_WMA_XL_IRAM =
 {
     0x1004,0x1008,0x100c,0x1010,0x1014,0x1018,0x101c,0x1021,0x1025,0x1029,0x102d,
     0x1031,0x1036,0x103a,0x103e,0x1043,0x1047,0x104b,0x1050,0x1054,0x1059,0x105d,
@@ -1522,7 +1522,7 @@ const fixed32 pow_a_table[] =
  */
 
 
-const fixed64 lsp_pow_e_table[] =
+const fixed64 lsp_pow_e_table[] ICONST_ATTR_WMA_XL_IRAM =
 {
     0xb504f30000000000LL, 0x9837f00000000000LL, 0x8000000000000000LL, 0x6ba27e8000000000LL, 0x5a82798000000000LL,
     0x4c1bf80000000000LL, 0x4000000000000000LL, 0x35d13f4000000000LL, 0x2d413cc000000000LL, 0x260dfc0000000000LL,
diff --git a/apps/codecs/libwma/wmadec.h b/apps/codecs/libwma/wmadec.h
index d7fa79d268..76429dede1 100644
--- a/apps/codecs/libwma/wmadec.h
+++ b/apps/codecs/libwma/wmadec.h
@@ -51,16 +51,27 @@
 
 #define LSP_POW_BITS 7
 
-/*define IRAM for targets with 48k/80k IRAM split*/
-#ifndef IBSS_ATTR_WMA_LARGE_IRAM
-#if (CONFIG_CPU == PP5022) || (CONFIG_CPU == PP5024) || defined(CPU_S5L870X) || (CONFIG_CPU == MCF5250)
-/* PP5022/24, MCF5250 have 128KB of IRAM. 80KB are allocated for codecs */
+
+#if (CONFIG_CPU == PP5022) || (CONFIG_CPU == PP5024) || (CONFIG_CPU == MCF5250)
+/* PP5022/24 and MCF5250 have 128KB of IRAM. 80KB are allocated for codecs */
+#define IBSS_ATTR_WMA_LARGE_IRAM IBSS_ATTR
+#define IBSS_ATTR_WMA_XL_IRAM
+#define ICONST_ATTR_WMA_XL_IRAM
+
+#elif defined(CPU_S5L870X)
+/* S5L870x has even more IRAM. Use it. */
 #define IBSS_ATTR_WMA_LARGE_IRAM IBSS_ATTR
+#define IBSS_ATTR_WMA_XL_IRAM    IBSS_ATTR
+#define ICONST_ATTR_WMA_XL_IRAM  ICONST_ATTR
+
 #else
 /* other PP's and MCF5249 have 96KB of IRAM */
 #define IBSS_ATTR_WMA_LARGE_IRAM
+#define IBSS_ATTR_WMA_XL_IRAM
+#define ICONST_ATTR_WMA_XL_IRAM
+
 #endif
-#endif
+
 
 #define VLCBITS 7       /*7 is the lowest without glitching*/
 #define VLCMAX ((22+VLCBITS-1)/VLCBITS)
diff --git a/apps/codecs/libwma/wmadeci.c b/apps/codecs/libwma/wmadeci.c
index 50a1b47405..d63a7641c9 100644
--- a/apps/codecs/libwma/wmadeci.c
+++ b/apps/codecs/libwma/wmadeci.c
@@ -33,19 +33,25 @@ static void wma_lsp_to_curve_init(WMADecodeContext *s, int frame_len);
 
 /*declarations of statically allocated variables used to remove malloc calls*/
 
-fixed32 coefsarray[MAX_CHANNELS][BLOCK_MAX_SIZE] IBSS_ATTR MEM_ALIGN_ATTR;
+static fixed32 coefsarray[MAX_CHANNELS][BLOCK_MAX_SIZE] IBSS_ATTR MEM_ALIGN_ATTR;
 /*decode and window into IRAM on targets with at least 80KB of codec IRAM*/
-fixed32 frame_out_buf[MAX_CHANNELS][BLOCK_MAX_SIZE * 2] IBSS_ATTR_WMA_LARGE_IRAM MEM_ALIGN_ATTR;
+static fixed32 frame_out_buf[MAX_CHANNELS][BLOCK_MAX_SIZE * 2] IBSS_ATTR_WMA_LARGE_IRAM MEM_ALIGN_ATTR;
 
 /*MDCT reconstruction windows*/
-fixed32 stat0[2048] MEM_ALIGN_ATTR, stat1[1024] MEM_ALIGN_ATTR, 
-    stat2[512] MEM_ALIGN_ATTR, stat3[256] MEM_ALIGN_ATTR, stat4[128] MEM_ALIGN_ATTR;    
+static fixed32 stat0[2048] IBSS_ATTR_WMA_XL_IRAM MEM_ALIGN_ATTR;
+static fixed32 stat1[1024] IBSS_ATTR_WMA_XL_IRAM MEM_ALIGN_ATTR;
+static fixed32 stat2[ 512] IBSS_ATTR_WMA_XL_IRAM MEM_ALIGN_ATTR;
+static fixed32 stat3[ 256] IBSS_ATTR_WMA_XL_IRAM MEM_ALIGN_ATTR;
+static fixed32 stat4[ 128] IBSS_ATTR_WMA_XL_IRAM MEM_ALIGN_ATTR;
 
 /*VLC lookup tables*/
-uint16_t *runtabarray[2], *levtabarray[2];                                        
+static uint16_t *runtabarray[2];
+static uint16_t *levtabarray[2];         
 
-uint16_t runtab_big[1336] MEM_ALIGN_ATTR, runtab_small[1072] MEM_ALIGN_ATTR,
-    levtab_big[1336] MEM_ALIGN_ATTR, levtab_small[1072] MEM_ALIGN_ATTR;
+static uint16_t runtab_big[1336]   MEM_ALIGN_ATTR;
+static uint16_t runtab_small[1072] MEM_ALIGN_ATTR;
+static uint16_t levtab_big[1336]   MEM_ALIGN_ATTR;
+static uint16_t levtab_small[1072] MEM_ALIGN_ATTR;
 
 #define VLCBUF1SIZE 4598
 #define VLCBUF2SIZE 3574
@@ -54,11 +60,11 @@ uint16_t runtab_big[1336] MEM_ALIGN_ATTR, runtab_small[1072] MEM_ALIGN_ATTR,
 
 /*putting these in IRAM actually makes PP slower*/
 
-VLC_TYPE vlcbuf1[VLCBUF1SIZE][2] MEM_ALIGN_ATTR;
-VLC_TYPE vlcbuf2[VLCBUF2SIZE][2] MEM_ALIGN_ATTR;
+static VLC_TYPE vlcbuf1[VLCBUF1SIZE][2] IBSS_ATTR_WMA_XL_IRAM MEM_ALIGN_ATTR;
+static VLC_TYPE vlcbuf2[VLCBUF2SIZE][2] MEM_ALIGN_ATTR;
 /* This buffer gets reused for lsp tables */
-VLC_TYPE vlcbuf3[VLCBUF3SIZE][2] MEM_ALIGN_ATTR;
-VLC_TYPE vlcbuf4[VLCBUF4SIZE][2] MEM_ALIGN_ATTR;
+static VLC_TYPE vlcbuf3[VLCBUF3SIZE][2] MEM_ALIGN_ATTR;
+static VLC_TYPE vlcbuf4[VLCBUF4SIZE][2] MEM_ALIGN_ATTR;