Adapted codecs and DSP system to having MACSR saved in thread context.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@7485 a1c6a512-1295-4272-9138-f99709370657

11 files changed, 44 insertions, 30 deletions

diff --git a/apps/codecs/Tremor/vorbisfile.c b/apps/codecs/Tremor/vorbisfile.c
index 92f42a313b..70c0cdb24e 100644
--- a/apps/codecs/Tremor/vorbisfile.c
+++ b/apps/codecs/Tremor/vorbisfile.c
@@ -762,6 +762,12 @@ int ov_clear(OggVorbis_File *vf){
 
 int ov_open_callbacks(void *f,OggVorbis_File *vf,char *initial,long ibytes,
     ov_callbacks callbacks){
+  #if defined(CPU_COLDFIRE) && !defined(SIMULATOR)
+  /* this seems to be the closest we get to an init function, let's init emac
+     here. rounding is disabled because of MULT31_SHIFT15, which will be
+     inaccurate with rounding in its current incarnation */
+  coldfire_set_macsr(EMAC_FRACTIONAL | EMAC_SATURATE);
+  #endif
   int ret=_ov_open1(f,vf,initial,ibytes,callbacks);
   if(ret)return ret;
   return _ov_open2(vf);
@@ -1287,15 +1293,12 @@ int ov_pcm_seek_page(OggVorbis_File *vf,ogg_int64_t pos){
 int ov_pcm_seek(OggVorbis_File *vf,ogg_int64_t pos){
   ogg_packet op={0,0,0,0,0,0};
   ogg_page og={0,0,0,0};
+
   int thisblock,lastblock=0;
   int ret=ov_pcm_seek_page(vf,pos);
   if(ret<0)return(ret);
   _make_decode_ready(vf);
 
-#if defined(CPU_COLDFIRE) && !defined(SIMULATOR)
-  mcf5249_init_mac();
-#endif
-
   /* discard leading packets we don't need for the lapping of the
      position we want; don't decode them */
 
@@ -1553,10 +1556,6 @@ long ov_read(OggVorbis_File *vf,char *buffer,int bytes_req,int *bitstream){
 
   if(vf->ready_state<OPENED)return(OV_EINVAL);
 
-#if defined(CPU_COLDFIRE) && !defined(SIMULATOR)
-  mcf5249_init_mac();
-#endif
-
   while(1){
     if(vf->ready_state==INITSET){
       samples=vorbis_synthesis_pcmout(&vf->vd,&pcm);
@@ -1622,10 +1621,6 @@ long ov_read_fixed(OggVorbis_File *vf,ogg_int32_t ***pcm_channels,int length,
                   int *bitstream){
   if(vf->ready_state<OPENED)return(OV_EINVAL);
 
-#if CONFIG_CPU == MCF5249
-  mcf5249_init_mac();
-#endif
-
   while(1){
     if(vf->ready_state==INITSET){
       ogg_int32_t **pcm;
diff --git a/apps/codecs/codec.h b/apps/codecs/codec.h
index c00d2bda95..0da653cf04 100644
--- a/apps/codecs/codec.h
+++ b/apps/codecs/codec.h
@@ -20,6 +20,7 @@
 /* Global declarations to be used in rockbox software codecs */
 
 #include "config.h"
+#include "system.h"
 
 #include <sys/types.h>
 
diff --git a/apps/codecs/liba52/parse.c b/apps/codecs/liba52/parse.c
index d5b323c701..9e7ed47d56 100644
--- a/apps/codecs/liba52/parse.c
+++ b/apps/codecs/liba52/parse.c
@@ -59,7 +59,7 @@ a52_state_t * a52_init (uint32_t mm_accel)
     int i;
 
     #if defined(CPU_COLDFIRE) && !defined(SIMULATOR)
-    asm volatile ("move.l #0x30, %macsr"); /* frac. mode with rounding */
+    coldfire_set_macsr(EMAC_FRACTIONAL | EMAC_ROUND | EMAC_SATURATE);
     #endif
     /* 
       this needs to come back if we ever want two decoder instances
diff --git a/apps/codecs/libmad/imdct_mcf5249.S b/apps/codecs/libmad/imdct_mcf5249.S
index 62b33c3f07..1afd910fa8 100644
--- a/apps/codecs/libmad/imdct_mcf5249.S
+++ b/apps/codecs/libmad/imdct_mcf5249.S
@@ -32,7 +32,6 @@ III_imdct_s:
      * get more low bits out of the accext01 register _before_ doing the
      * movclrs.
      */
-    move.l #0xb0, %macsr                /* frac mode, saturation, rounding */
     sub.l %a0, %a0                      /* clear loop variable */
 .imdctloop:                             /* outer loop label */
     lea.l imdct_s, %a1                  /* load pointer to imdct coefs in a1 */
diff --git a/apps/codecs/libmad/layer3.c b/apps/codecs/libmad/layer3.c
index b5fa2be6f5..6d528ea52c 100644
--- a/apps/codecs/libmad/layer3.c
+++ b/apps/codecs/libmad/layer3.c
@@ -1775,8 +1775,6 @@ void imdct36(mad_fixed_t const X[18], mad_fixed_t x[36])
   mad_fixed_t t[16];
   /* assumes FRACBITS = 28 */
   asm volatile (
-    "move.l #0xb0, %%d0\n\t"        /* frac. mode, saturate, round */
-    "move.l %%d0, %%macsr\n\t"
     "move.l (4*4, %[X]), %%d0\n\t"
     "move.l #0x0ec835e8, %%d1\n\t"
     "mac.l %%d0, %%d1, (13*4, %[X]), %%d0, %%acc0\n\t"
diff --git a/apps/codecs/libmad/synth.c b/apps/codecs/libmad/synth.c
index bdec4258fb..4315e239e8 100644
--- a/apps/codecs/libmad/synth.c
+++ b/apps/codecs/libmad/synth.c
@@ -42,6 +42,11 @@ void mad_synth_init(struct mad_synth *synth)
   synth->pcm.samplerate = 0;
   synth->pcm.channels   = 0;
   synth->pcm.length     = 0;
+  #if defined(CPU_COLDFIRE) && !defined(SIMULATOR)
+  /* init the emac unit here, since this function should always be called
+     before using libmad */
+  coldfire_set_macsr(EMAC_FRACTIONAL | EMAC_SATURATE | EMAC_ROUND);
+  #endif
 }
 
 /*
@@ -582,8 +587,6 @@ void synth_full(struct mad_synth *synth, struct mad_frame const *frame,
   mad_fixed_t const (*Dptr)[32];
   mad_fixed64hi_t hi;
 
-  asm volatile("move.l #0x20, %macsr"); /* fractional mode */
-  
   for (ch = 0; ch < nch; ++ch) {
     sbsample = &frame->sbsample[ch];
     filter   = &synth->filter[ch];
diff --git a/apps/codecs/libmusepack/mpc_decoder.c b/apps/codecs/libmusepack/mpc_decoder.c
index 0c8b8b76a7..a37dc4583c 100644
--- a/apps/codecs/libmusepack/mpc_decoder.c
+++ b/apps/codecs/libmusepack/mpc_decoder.c
@@ -1173,6 +1173,10 @@ void mpc_decoder_setup(mpc_decoder *d, mpc_reader *r)
   mpc_decoder_initialisiere_quantisierungstabellen(d, 1.0f);
   mpc_decoder_init_huffman_sv6(d);
   mpc_decoder_init_huffman_sv7(d);
+
+  #if defined(CPU_COLDFIRE)&& !defined(SIMULATOR)
+  coldfire_set_macsr(EMAC_FRACTIONAL | EMAC_ROUND | EMAC_SATURATE);
+  #endif
 }
 
 static void mpc_decoder_set_streaminfo(mpc_decoder *d, mpc_streaminfo *si)
diff --git a/apps/codecs/libmusepack/synth_filter.c b/apps/codecs/libmusepack/synth_filter.c
index 7644551d16..51ffcce921 100644
--- a/apps/codecs/libmusepack/synth_filter.c
+++ b/apps/codecs/libmusepack/synth_filter.c
@@ -335,9 +335,6 @@ static void Synthese_Filter_float_internal(MPC_SAMPLE_FORMAT * OutData,MPC_SAMPL
 {
     mpc_uint32_t n;
     
-    #if defined(CPU_COLDFIRE) && !defined(SIMULATOR)
-    asm volatile ("move.l #0x20, %macsr"); /* fractional emac mode */
-    #endif
     for ( n = 0; n < 36; n++, Y += 32 ) {
         V -= 64;
         Calculate_New_V ( Y, V );
diff --git a/apps/codecs/mpa.c b/apps/codecs/mpa.c
index 7beecd9edb..9970d57240 100644
--- a/apps/codecs/mpa.c
+++ b/apps/codecs/mpa.c
@@ -114,9 +114,6 @@ enum codec_status codec_start(struct codec_api* api)
     ci->memset(&Synth, 0, sizeof(struct mad_synth));
     ci->memset(&Timer, 0, sizeof(mad_timer_t));
     
-#if defined(CPU_COLDFIRE) && !defined(SIMULATOR)
-    mcf5249_init_mac();
-#endif
     mad_stream_init(&Stream);
     mad_frame_init(&Frame);
     mad_synth_init(&Synth);
diff --git a/apps/dsp.c b/apps/dsp.c
index 8064a881a3..23b7ea548d 100644
--- a/apps/dsp.c
+++ b/apps/dsp.c
@@ -43,7 +43,6 @@
 
 #if defined(CPU_COLDFIRE) && !defined(SIMULATOR)
 
-#define INIT() asm volatile ("move.l #0xb0, %macsr") /* frac, round, clip */
 /* Multiply two S.31 fractional integers and return the sign bit and the
  * 31 most significant bits of the result.
  */
@@ -89,7 +88,6 @@
 
 #else
 
-#define INIT()
 #define FRACMUL(x, y) (long) (((((long long) (x)) * ((long long) (y))) >> 31))
 #define FRACMUL_8(x, y) (long) (((((long long) (x)) * ((long long) (y))) >> 23))
 #define FRACMUL_8_LOOP(x, y, s) \
@@ -492,11 +490,17 @@ long dsp_process(char* dst, char* src[], long size)
     long factor;
     int samples;
 
+    #if defined(CPU_COLDFIRE) && !defined(SIMULATOR)
+    /* set emac unit for dsp processing, and save old macsr, we're running in
+       codec thread context at this point, so can't clobber it */
+    unsigned long old_macsr = coldfire_get_macsr();
+    coldfire_set_macsr(EMAC_FRACTIONAL | EMAC_ROUND | EMAC_SATURATE);
+    #endif
+    
     dsp = &dsp_conf[current_codec];
 
     factor = (dsp->stereo_mode != STEREO_MONO) ? 2 : 1;
     size /= dsp->sample_bytes * factor;
-    INIT();
     dsp_set_replaygain(false);
 
     while (size > 0)
@@ -510,7 +514,10 @@ long dsp_process(char* dst, char* src[], long size)
         dst += samples * sizeof(short) * 2;
         yield();
     }
-
+    #if defined(CPU_COLDFIRE) && !defined(SIMULATOR)
+    /* set old macsr again */
+    coldfire_set_macsr(old_macsr);
+    #endif
     return written * sizeof(short) * 2;
 }
 
diff --git a/firmware/export/system.h b/firmware/export/system.h
index 895b3f7529..07b7094615 100644
--- a/firmware/export/system.h
+++ b/firmware/export/system.h
@@ -125,8 +125,21 @@ enum {
      : /* %0 */ "d"(mask),   \
        /* %1 */ "a"(address))
 
-static inline void mcf5249_init_mac(void) {
-      asm volatile ("move.l #0x20, %macsr;");  /* frac, truncate, no saturation */
+#define EMAC_ROUND      0x10
+#define EMAC_FRACTIONAL 0x20
+#define EMAC_SATURATE   0x80
+
+static inline void coldfire_set_macsr(const unsigned long flags)
+{
+    asm volatile ("move.l %0, %%macsr" : : "r" (flags));
+}
+
+static inline unsigned long coldfire_get_macsr(void)
+{
+    unsigned long m;
+    
+    asm volatile ("move.l %%macsr, %0" : "=r" (m));
+    return m;
 }
 
 #endif