spu: put reverb on the thread too

and more stuff to work struct

7 files changed, 247 insertions, 204 deletions

diff --git a/plugins/dfsound/externals.h b/plugins/dfsound/externals.h
index 4832fac..f6fc440 100644
--- a/plugins/dfsound/externals.h
+++ b/plugins/dfsound/externals.h
@@ -28,6 +28,11 @@
 #define noinline

 #define unlikely(x) x

 #endif

+#if defined(__GNUC__) && !defined(_TMS320C6X)

+#define preload __builtin_prefetch

+#else

+#define preload(...)

+#endif

 

 #define PSE_LT_SPU                  4

 #define PSE_SPU_ERR_SUCCESS         0

@@ -121,8 +126,6 @@ typedef struct
 

  int VolLeft;

  int VolRight;

- int iRVBLeft;

- int iRVBRight;

 

  int FB_SRC_A;       // (offset)

  int FB_SRC_B;       // (offset)

@@ -224,9 +227,10 @@ typedef struct
  int             iRightXAVol;

 

  SPUCHAN       * s_chan;

- int           * SB;

+ REVERBInfo    * rvb;

 

- int           * RVB;

+ // buffers

+ int           * SB;

  int           * SSumLR;

 

  int             pad[29];

@@ -240,7 +244,6 @@ typedef struct
 #ifndef _IN_SPU

 

 extern SPUInfo spu;

-extern REVERBInfo rvb;

 

 void do_samples(unsigned int cycles_to, int do_sync);

 void schedule_next_irq(void);

diff --git a/plugins/dfsound/registers.c b/plugins/dfsound/registers.c
index 7a49b1c..4588fa7 100644
--- a/plugins/dfsound/registers.c
+++ b/plugins/dfsound/registers.c
@@ -146,14 +146,14 @@ void CALLBACK SPUwriteRegister(unsigned long reg, unsigned short val,
     //-------------------------------------------------//

     case H_SPUReverbAddr:

       if(val==0xFFFF || val<=0x200)

-       {rvb.StartAddr=rvb.CurrAddr=0;}

+       {spu.rvb->StartAddr=spu.rvb->CurrAddr=0;}

       else

        {

         const long iv=(unsigned long)val<<2;

-        if(rvb.StartAddr!=iv)

+        if(spu.rvb->StartAddr!=iv)

          {

-          rvb.StartAddr=(unsigned long)val<<2;

-          rvb.CurrAddr=rvb.StartAddr;

+          spu.rvb->StartAddr=(unsigned long)val<<2;

+          spu.rvb->CurrAddr=spu.rvb->StartAddr;

          }

        }

       goto rvbd;

@@ -163,11 +163,11 @@ void CALLBACK SPUwriteRegister(unsigned long reg, unsigned short val,
       goto upd_irq;

     //-------------------------------------------------//

     case H_SPUrvolL:

-      rvb.VolLeft=val;

+      spu.rvb->VolLeft=val;

       break;

     //-------------------------------------------------//

     case H_SPUrvolR:

-      rvb.VolRight=val;

+      spu.rvb->VolRight=val;

       break;

     //-------------------------------------------------//

 

@@ -246,38 +246,38 @@ void CALLBACK SPUwriteRegister(unsigned long reg, unsigned short val,
       ReverbOn(16,24,val);

       break;

     //-------------------------------------------------//

-    case H_Reverb+0   : rvb.FB_SRC_A=val*4;         goto rvbd;

-    case H_Reverb+2   : rvb.FB_SRC_B=val*4;         goto rvbd;

-    case H_Reverb+4   : rvb.IIR_ALPHA=(short)val;   goto rvbd;

-    case H_Reverb+6   : rvb.ACC_COEF_A=(short)val;  goto rvbd;

-    case H_Reverb+8   : rvb.ACC_COEF_B=(short)val;  goto rvbd;

-    case H_Reverb+10  : rvb.ACC_COEF_C=(short)val;  goto rvbd;

-    case H_Reverb+12  : rvb.ACC_COEF_D=(short)val;  goto rvbd;

-    case H_Reverb+14  : rvb.IIR_COEF=(short)val;    goto rvbd;

-    case H_Reverb+16  : rvb.FB_ALPHA=(short)val;    goto rvbd;

-    case H_Reverb+18  : rvb.FB_X=(short)val;        goto rvbd;

-    case H_Reverb+20  : rvb.IIR_DEST_A0=val*4;      goto rvbd;

-    case H_Reverb+22  : rvb.IIR_DEST_A1=val*4;      goto rvbd;

-    case H_Reverb+24  : rvb.ACC_SRC_A0=val*4;       goto rvbd;

-    case H_Reverb+26  : rvb.ACC_SRC_A1=val*4;       goto rvbd;

-    case H_Reverb+28  : rvb.ACC_SRC_B0=val*4;       goto rvbd;

-    case H_Reverb+30  : rvb.ACC_SRC_B1=val*4;       goto rvbd;

-    case H_Reverb+32  : rvb.IIR_SRC_A0=val*4;       goto rvbd;

-    case H_Reverb+34  : rvb.IIR_SRC_A1=val*4;       goto rvbd;

-    case H_Reverb+36  : rvb.IIR_DEST_B0=val*4;      goto rvbd;

-    case H_Reverb+38  : rvb.IIR_DEST_B1=val*4;      goto rvbd;

-    case H_Reverb+40  : rvb.ACC_SRC_C0=val*4;       goto rvbd;

-    case H_Reverb+42  : rvb.ACC_SRC_C1=val*4;       goto rvbd;

-    case H_Reverb+44  : rvb.ACC_SRC_D0=val*4;       goto rvbd;

-    case H_Reverb+46  : rvb.ACC_SRC_D1=val*4;       goto rvbd;

-    case H_Reverb+48  : rvb.IIR_SRC_B1=val*4;       goto rvbd;

-    case H_Reverb+50  : rvb.IIR_SRC_B0=val*4;       goto rvbd;

-    case H_Reverb+52  : rvb.MIX_DEST_A0=val*4;      goto rvbd;

-    case H_Reverb+54  : rvb.MIX_DEST_A1=val*4;      goto rvbd;

-    case H_Reverb+56  : rvb.MIX_DEST_B0=val*4;      goto rvbd;

-    case H_Reverb+58  : rvb.MIX_DEST_B1=val*4;      goto rvbd;

-    case H_Reverb+60  : rvb.IN_COEF_L=(short)val;   goto rvbd;

-    case H_Reverb+62  : rvb.IN_COEF_R=(short)val;   goto rvbd;

+    case H_Reverb+0   : spu.rvb->FB_SRC_A=val*4;         goto rvbd;

+    case H_Reverb+2   : spu.rvb->FB_SRC_B=val*4;         goto rvbd;

+    case H_Reverb+4   : spu.rvb->IIR_ALPHA=(short)val;   goto rvbd;

+    case H_Reverb+6   : spu.rvb->ACC_COEF_A=(short)val;  goto rvbd;

+    case H_Reverb+8   : spu.rvb->ACC_COEF_B=(short)val;  goto rvbd;

+    case H_Reverb+10  : spu.rvb->ACC_COEF_C=(short)val;  goto rvbd;

+    case H_Reverb+12  : spu.rvb->ACC_COEF_D=(short)val;  goto rvbd;

+    case H_Reverb+14  : spu.rvb->IIR_COEF=(short)val;    goto rvbd;

+    case H_Reverb+16  : spu.rvb->FB_ALPHA=(short)val;    goto rvbd;

+    case H_Reverb+18  : spu.rvb->FB_X=(short)val;        goto rvbd;

+    case H_Reverb+20  : spu.rvb->IIR_DEST_A0=val*4;      goto rvbd;

+    case H_Reverb+22  : spu.rvb->IIR_DEST_A1=val*4;      goto rvbd;

+    case H_Reverb+24  : spu.rvb->ACC_SRC_A0=val*4;       goto rvbd;

+    case H_Reverb+26  : spu.rvb->ACC_SRC_A1=val*4;       goto rvbd;

+    case H_Reverb+28  : spu.rvb->ACC_SRC_B0=val*4;       goto rvbd;

+    case H_Reverb+30  : spu.rvb->ACC_SRC_B1=val*4;       goto rvbd;

+    case H_Reverb+32  : spu.rvb->IIR_SRC_A0=val*4;       goto rvbd;

+    case H_Reverb+34  : spu.rvb->IIR_SRC_A1=val*4;       goto rvbd;

+    case H_Reverb+36  : spu.rvb->IIR_DEST_B0=val*4;      goto rvbd;

+    case H_Reverb+38  : spu.rvb->IIR_DEST_B1=val*4;      goto rvbd;

+    case H_Reverb+40  : spu.rvb->ACC_SRC_C0=val*4;       goto rvbd;

+    case H_Reverb+42  : spu.rvb->ACC_SRC_C1=val*4;       goto rvbd;

+    case H_Reverb+44  : spu.rvb->ACC_SRC_D0=val*4;       goto rvbd;

+    case H_Reverb+46  : spu.rvb->ACC_SRC_D1=val*4;       goto rvbd;

+    case H_Reverb+48  : spu.rvb->IIR_SRC_B1=val*4;       goto rvbd;

+    case H_Reverb+50  : spu.rvb->IIR_SRC_B0=val*4;       goto rvbd;

+    case H_Reverb+52  : spu.rvb->MIX_DEST_A0=val*4;      goto rvbd;

+    case H_Reverb+54  : spu.rvb->MIX_DEST_A1=val*4;      goto rvbd;

+    case H_Reverb+56  : spu.rvb->MIX_DEST_B0=val*4;      goto rvbd;

+    case H_Reverb+58  : spu.rvb->MIX_DEST_B1=val*4;      goto rvbd;

+    case H_Reverb+60  : spu.rvb->IN_COEF_L=(short)val;   goto rvbd;

+    case H_Reverb+62  : spu.rvb->IN_COEF_R=(short)val;   goto rvbd;

    }

  return;

 

@@ -287,7 +287,7 @@ upd_irq:
  return;

 

 rvbd:

- rvb.dirty = 1; // recalculate on next update

+ spu.rvb->dirty = 1; // recalculate on next update

 }

 

 ////////////////////////////////////////////////////////////////////////

diff --git a/plugins/dfsound/reverb.c b/plugins/dfsound/reverb.c
index 7e32b8e..2ff6edc 100644
--- a/plugins/dfsound/reverb.c
+++ b/plugins/dfsound/reverb.c
@@ -50,41 +50,40 @@ INLINE int rvb2ram_offs(int curr, int space, int iOff)
 

 // get_buffer content helper: takes care about wraps

 #define g_buffer(var) \

- ((int)(signed short)spu.spuMem[rvb2ram_offs(curr_addr, space, rvb.n##var)])

+ ((int)(signed short)spu.spuMem[rvb2ram_offs(curr_addr, space, rvb->n##var)])

 

 // saturate iVal and store it as var

 #define s_buffer(var, iVal) \

  ssat32_to_16(iVal); \

- spu.spuMem[rvb2ram_offs(curr_addr, space, rvb.n##var)] = iVal

+ spu.spuMem[rvb2ram_offs(curr_addr, space, rvb->n##var)] = iVal

 

 #define s_buffer1(var, iVal) \

  ssat32_to_16(iVal); \

- spu.spuMem[rvb2ram_offs(curr_addr, space, rvb.n##var + 1)] = iVal

+ spu.spuMem[rvb2ram_offs(curr_addr, space, rvb->n##var + 1)] = iVal

 

 ////////////////////////////////////////////////////////////////////////

 

 // portions based on spu2-x from PCSX2

-static void MixREVERB(int *SSumLR, int *RVB, int ns_to)

+static void MixREVERB(int *SSumLR, int *RVB, int ns_to, int curr_addr)

 {

- int l_old = rvb.iRVBLeft;

- int r_old = rvb.iRVBRight;

- int curr_addr = rvb.CurrAddr;

- int space = 0x40000 - rvb.StartAddr;

- int l = 0, r = 0, ns;

+ const REVERBInfo *rvb = spu.rvb;

+ int IIR_ALPHA = rvb->IIR_ALPHA;

+ int IIR_COEF = rvb->IIR_COEF;

+ int space = 0x40000 - rvb->StartAddr;

+ int l, r, ns;

 

  for (ns = 0; ns < ns_to * 2; )

   {

-   int IIR_ALPHA = rvb.IIR_ALPHA;

    int ACC0, ACC1, FB_A0, FB_A1, FB_B0, FB_B1;

    int mix_dest_a0, mix_dest_a1, mix_dest_b0, mix_dest_b1;

 

-   int input_L = RVB[ns]   * rvb.IN_COEF_L;

-   int input_R = RVB[ns+1] * rvb.IN_COEF_R;

+   int input_L = RVB[ns]   * rvb->IN_COEF_L;

+   int input_R = RVB[ns+1] * rvb->IN_COEF_R;

 

-   int IIR_INPUT_A0 = ((g_buffer(IIR_SRC_A0) * rvb.IIR_COEF) + input_L) >> 15;

-   int IIR_INPUT_A1 = ((g_buffer(IIR_SRC_A1) * rvb.IIR_COEF) + input_R) >> 15;

-   int IIR_INPUT_B0 = ((g_buffer(IIR_SRC_B0) * rvb.IIR_COEF) + input_L) >> 15;

-   int IIR_INPUT_B1 = ((g_buffer(IIR_SRC_B1) * rvb.IIR_COEF) + input_R) >> 15;

+   int IIR_INPUT_A0 = ((g_buffer(IIR_SRC_A0) * IIR_COEF) + input_L) >> 15;

+   int IIR_INPUT_A1 = ((g_buffer(IIR_SRC_A1) * IIR_COEF) + input_R) >> 15;

+   int IIR_INPUT_B0 = ((g_buffer(IIR_SRC_B0) * IIR_COEF) + input_L) >> 15;

+   int IIR_INPUT_B1 = ((g_buffer(IIR_SRC_B1) * IIR_COEF) + input_R) >> 15;

 

    int iir_dest_a0 = g_buffer(IIR_DEST_A0);

    int iir_dest_a1 = g_buffer(IIR_DEST_A1);

@@ -96,30 +95,34 @@ static void MixREVERB(int *SSumLR, int *RVB, int ns_to)
    int IIR_B0 = iir_dest_b0 + ((IIR_INPUT_B0 - iir_dest_b0) * IIR_ALPHA >> 15);

    int IIR_B1 = iir_dest_b1 + ((IIR_INPUT_B1 - iir_dest_b1) * IIR_ALPHA >> 15);

 

+   preload(SSumLR + ns + 64*2/4 - 4);

+

    s_buffer1(IIR_DEST_A0, IIR_A0);

    s_buffer1(IIR_DEST_A1, IIR_A1);

    s_buffer1(IIR_DEST_B0, IIR_B0);

    s_buffer1(IIR_DEST_B1, IIR_B1);

 

-   ACC0 = (g_buffer(ACC_SRC_A0) * rvb.ACC_COEF_A +

-           g_buffer(ACC_SRC_B0) * rvb.ACC_COEF_B +

-           g_buffer(ACC_SRC_C0) * rvb.ACC_COEF_C +

-           g_buffer(ACC_SRC_D0) * rvb.ACC_COEF_D) >> 15;

-   ACC1 = (g_buffer(ACC_SRC_A1) * rvb.ACC_COEF_A +

-           g_buffer(ACC_SRC_B1) * rvb.ACC_COEF_B +

-           g_buffer(ACC_SRC_C1) * rvb.ACC_COEF_C +

-           g_buffer(ACC_SRC_D1) * rvb.ACC_COEF_D) >> 15;

+   preload(RVB + ns + 64*2/4 - 4);

+

+   ACC0 = (g_buffer(ACC_SRC_A0) * rvb->ACC_COEF_A +

+           g_buffer(ACC_SRC_B0) * rvb->ACC_COEF_B +

+           g_buffer(ACC_SRC_C0) * rvb->ACC_COEF_C +

+           g_buffer(ACC_SRC_D0) * rvb->ACC_COEF_D) >> 15;

+   ACC1 = (g_buffer(ACC_SRC_A1) * rvb->ACC_COEF_A +

+           g_buffer(ACC_SRC_B1) * rvb->ACC_COEF_B +

+           g_buffer(ACC_SRC_C1) * rvb->ACC_COEF_C +

+           g_buffer(ACC_SRC_D1) * rvb->ACC_COEF_D) >> 15;

 

    FB_A0 = g_buffer(FB_SRC_A0);

    FB_A1 = g_buffer(FB_SRC_A1);

    FB_B0 = g_buffer(FB_SRC_B0);

    FB_B1 = g_buffer(FB_SRC_B1);

 

-   mix_dest_a0 = ACC0 - ((FB_A0 * rvb.FB_ALPHA) >> 15);

-   mix_dest_a1 = ACC1 - ((FB_A1 * rvb.FB_ALPHA) >> 15);

+   mix_dest_a0 = ACC0 - ((FB_A0 * rvb->FB_ALPHA) >> 15);

+   mix_dest_a1 = ACC1 - ((FB_A1 * rvb->FB_ALPHA) >> 15);

 

-   mix_dest_b0 = FB_A0 + (((ACC0 - FB_A0) * rvb.FB_ALPHA - FB_B0 * rvb.FB_X) >> 15);

-   mix_dest_b1 = FB_A1 + (((ACC1 - FB_A1) * rvb.FB_ALPHA - FB_B1 * rvb.FB_X) >> 15);

+   mix_dest_b0 = FB_A0 + (((ACC0 - FB_A0) * rvb->FB_ALPHA - FB_B0 * rvb->FB_X) >> 15);

+   mix_dest_b1 = FB_A1 + (((ACC1 - FB_A1) * rvb->FB_ALPHA - FB_B1 * rvb->FB_X) >> 15);

 

    s_buffer(MIX_DEST_A0, mix_dest_a0);

    s_buffer(MIX_DEST_A1, mix_dest_a1);

@@ -129,73 +132,60 @@ static void MixREVERB(int *SSumLR, int *RVB, int ns_to)
    l = (mix_dest_a0 + mix_dest_b0) / 2;

    r = (mix_dest_a1 + mix_dest_b1) / 2;

 

-   l = (l * rvb.VolLeft)  >> 15; // 15?

-   r = (r * rvb.VolRight) >> 15;

+   l = (l * rvb->VolLeft)  >> 15; // 15?

+   r = (r * rvb->VolRight) >> 15;

 

-   SSumLR[ns++] += (l + l_old) / 2;

-   SSumLR[ns++] += (r + r_old) / 2;

    SSumLR[ns++] += l;

    SSumLR[ns++] += r;

-

-   l_old = l;

-   r_old = r;

+   SSumLR[ns++] += l;

+   SSumLR[ns++] += r;

 

    curr_addr++;

-   if (curr_addr >= 0x40000) curr_addr = rvb.StartAddr;

+   if (curr_addr >= 0x40000) curr_addr = rvb->StartAddr;

   }

-

- rvb.iRVBLeft = l;

- rvb.iRVBRight = r;

- rvb.CurrAddr = curr_addr;

 }

 

-static void MixREVERB_off(int *SSumLR, int ns_to)

+static void MixREVERB_off(int *SSumLR, int ns_to, int curr_addr)

 {

- int l_old = rvb.iRVBLeft;

- int r_old = rvb.iRVBRight;

- int curr_addr = rvb.CurrAddr;

- int space = 0x40000 - rvb.StartAddr;

- int l = 0, r = 0, ns;

+ const REVERBInfo *rvb = spu.rvb;

+ int space = 0x40000 - rvb->StartAddr;

+ int l, r, ns;

 

  for (ns = 0; ns < ns_to * 2; )

   {

+   preload(SSumLR + ns + 64*2/4 - 4);

+

    l = (g_buffer(MIX_DEST_A0) + g_buffer(MIX_DEST_B0)) / 2;

    r = (g_buffer(MIX_DEST_A1) + g_buffer(MIX_DEST_B1)) / 2;

 

-   l = (l * rvb.VolLeft)  >> 15;

-   r = (r * rvb.VolRight) >> 15;

+   l = (l * rvb->VolLeft)  >> 15;

+   r = (r * rvb->VolRight) >> 15;

 

-   SSumLR[ns++] += (l + l_old) / 2;

-   SSumLR[ns++] += (r + r_old) / 2;

    SSumLR[ns++] += l;

    SSumLR[ns++] += r;

-

-   l_old = l;

-   r_old = r;

+   SSumLR[ns++] += l;

+   SSumLR[ns++] += r;

 

    curr_addr++;

-   if (curr_addr >= 0x40000) curr_addr = rvb.StartAddr;

+   if (curr_addr >= 0x40000) curr_addr = rvb->StartAddr;

   }

-

- rvb.iRVBLeft = l;

- rvb.iRVBRight = r;

- rvb.CurrAddr = curr_addr;

 }

 

-static void prepare_offsets(void)

+static void REVERBPrep(void)

 {

- int space = 0x40000 - rvb.StartAddr;

+ REVERBInfo *rvb = spu.rvb;

+ int space = 0x40000 - rvb->StartAddr;

  int t;

  #define prep_offs(v) \

-   t = rvb.v; \

+   t = rvb->v; \

    while (t >= space) \

      t -= space; \

-   rvb.n##v = t

+   rvb->n##v = t

  #define prep_offs2(d, v1, v2) \

-   t = rvb.v1 - rvb.v2; \

+   t = rvb->v1 - rvb->v2; \

    while (t >= space) \

      t -= space; \

-   rvb.n##d = t

+   rvb->n##d = t

 

  prep_offs(IIR_SRC_A0);

  prep_offs(IIR_SRC_A1);

@@ -224,37 +214,18 @@ static void prepare_offsets(void)
 

 #undef prep_offs

 #undef prep_offs2

- rvb.dirty = 0;

+ rvb->dirty = 0;

 }

 

-INLINE void REVERBDo(int *SSumLR, int *RVB, int ns_to)

+INLINE void REVERBDo(int *SSumLR, int *RVB, int ns_to, int curr_addr)

 {

- if (!rvb.StartAddr)                                   // reverb is off

- {

-  rvb.iRVBLeft = rvb.iRVBRight = 0;

-  return;

- }

-

  if (spu.spuCtrl & 0x80)                               // -> reverb on? oki

  {

-  if (unlikely(rvb.dirty))

-   prepare_offsets();

-

-  MixREVERB(SSumLR, RVB, ns_to);

- }

- else if (rvb.VolLeft || rvb.VolRight)

- {

-  if (unlikely(rvb.dirty))

-   prepare_offsets();

-

-  MixREVERB_off(SSumLR, ns_to);

+  MixREVERB(SSumLR, RVB, ns_to, curr_addr);

  }

- else                                                  // -> reverb off

+ else if (spu.rvb->VolLeft || spu.rvb->VolRight)

  {

-  // reverb runs anyway

-  rvb.CurrAddr += ns_to / 2;

-  while (rvb.CurrAddr >= 0x40000)

-   rvb.CurrAddr -= 0x40000 - rvb.StartAddr;

+  MixREVERB_off(SSumLR, ns_to, curr_addr);

  }

 }

 

diff --git a/plugins/dfsound/spu.c b/plugins/dfsound/spu.c
index ec31b0c..8681d35 100644
--- a/plugins/dfsound/spu.c
+++ b/plugins/dfsound/spu.c
@@ -73,11 +73,8 @@ static char * libraryInfo     = N_("P.E.Op.S. Sound Driver V1.7\nCoded by Pete B
 SPUInfo         spu;
 SPUConfig       spu_config;
 
-// MAIN infos struct for each channel
-
-REVERBInfo      rvb;
-
 static int iFMod[NSSIZE];
+static int RVB[NSSIZE * 2];
 int ChanBuf[NSSIZE];
 
 #define CDDA_BUFFER_SIZE (16384 * sizeof(uint32_t)) // must be power of 2
@@ -763,11 +760,13 @@ static void do_silent_chans(int ns_to, int silentch)
 static void do_channels(int ns_to)
 {
  unsigned int mask;
+ int do_rvb, ch, d;
  SPUCHAN *s_chan;
  int *SB, sinc;
- int ch, d;
 
- memset(spu.RVB, 0, ns_to * sizeof(spu.RVB[0]) * 2);
+ do_rvb = spu.rvb->StartAddr && spu_config.iUseReverb;
+ if (do_rvb)
+  memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2);
 
  mask = spu.dwNewChannel & 0xffffff;
  for (ch = 0; mask != 0; ch++, mask >>= 1) {
@@ -812,14 +811,27 @@ static void do_channels(int ns_to)
 
    if (s_chan->bFMod == 2)                         // fmod freq channel
     memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
-   if (s_chan->bRVBActive)
-    mix_chan_rvb(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume, spu.RVB);
+   if (s_chan->bRVBActive && do_rvb)
+    mix_chan_rvb(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume, RVB);
    else
     mix_chan(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume);
   }
+
+  if (spu.rvb->StartAddr) {
+   if (do_rvb) {
+    if (unlikely(spu.rvb->dirty))
+     REVERBPrep();
+
+    REVERBDo(spu.SSumLR, RVB, ns_to, spu.rvb->CurrAddr);
+   }
+
+   spu.rvb->CurrAddr += ns_to / 2;
+   while (spu.rvb->CurrAddr >= 0x40000)
+    spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr;
+  }
 }
 
-static void do_samples_finish(int *SSumLR, int *RVB, int ns_to,
+static void do_samples_finish(int *SSumLR, int ns_to,
  int silentch, int decode_pos);
 
 // optional worker thread handling
@@ -833,8 +845,7 @@ static struct spu_worker {
    unsigned int exit_thread;
    unsigned int i_ready;
    unsigned int i_reaped;
-   unsigned int req_sent; // dsp
-   unsigned int last_boot_cnt;
+   unsigned int last_boot_cnt; // dsp
   };
   // aligning for C64X_DSP
   unsigned int _pad0[128/4];
@@ -851,6 +862,7 @@ static struct spu_worker {
   int ns_to;
   int ctrl;
   int decode_pos;
+  int rvb_addr;
   unsigned int channels_new;
   unsigned int channels_on;
   unsigned int channels_silent;
@@ -861,10 +873,11 @@ static struct spu_worker {
    int start;
    int loop;
    int ns_to;
+   short vol_l;
+   short vol_r;
    ADSRInfoEx adsr;
-   // might want to add vol and fmod flags..
+   // might also want to add fmod flags..
   } ch[24];
-  int RVB[NSSIZE * 2];
   int SSumLR[NSSIZE * 2];
  } i[4];
 } *worker;
@@ -874,6 +887,7 @@ static struct spu_worker {
 
 static void thread_work_start(void);
 static void thread_work_wait_sync(struct work_item *work, int force);
+static void thread_sync_caches(void);
 static int  thread_get_i_done(void);
 
 static int decode_block_work(void *context, int ch, int *SB)
@@ -936,6 +950,8 @@ static void queue_channel_work(int ns_to, unsigned int silentch)
    work->ch[ch].sbpos = s_chan->iSBPos;
    work->ch[ch].sinc = s_chan->sinc;
    work->ch[ch].adsr = s_chan->ADSRX;
+   work->ch[ch].vol_l = s_chan->iLeftVolume;
+   work->ch[ch].vol_r = s_chan->iRightVolume;
    work->ch[ch].start = s_chan->pCurr - spu.spuMemC;
    work->ch[ch].loop = s_chan->pLoop - spu.spuMemC;
    if (s_chan->prevflags & 1)
@@ -952,6 +968,19 @@ static void queue_channel_work(int ns_to, unsigned int silentch)
    }
   }
 
+ work->rvb_addr = 0;
+ if (spu.rvb->StartAddr) {
+  if (spu_config.iUseReverb) {
+   if (unlikely(spu.rvb->dirty))
+    REVERBPrep();
+   work->rvb_addr = spu.rvb->CurrAddr;
+  }
+
+  spu.rvb->CurrAddr += ns_to / 2;
+  while (spu.rvb->CurrAddr >= 0x40000)
+   spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr;
+ }
+
  worker->i_ready++;
  thread_work_start();
 }
@@ -965,7 +994,9 @@ static void do_channel_work(struct work_item *work)
  SPUCHAN *s_chan;
 
  ns_to = work->ns_to;
- memset(work->RVB, 0, ns_to * sizeof(work->RVB[0]) * 2);
+
+ if (work->rvb_addr)
+  memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2);
 
  mask = work->channels_new;
  for (ch = 0; mask != 0; ch++, mask >>= 1) {
@@ -1010,12 +1041,15 @@ static void do_channel_work(struct work_item *work)
 
    if (s_chan->bFMod == 2)                         // fmod freq channel
     memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
-   if (s_chan->bRVBActive)
+   if (s_chan->bRVBActive && work->rvb_addr)
     mix_chan_rvb(work->SSumLR, ns_to,
-      s_chan->iLeftVolume, s_chan->iRightVolume, work->RVB);
+      work->ch[ch].vol_l, work->ch[ch].vol_r, RVB);
    else
-    mix_chan(work->SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume);
+    mix_chan(work->SSumLR, ns_to, work->ch[ch].vol_l, work->ch[ch].vol_r);
   }
+
+  if (work->rvb_addr)
+   REVERBDo(work->SSumLR, RVB, ns_to, work->rvb_addr);
 }
 
 static void sync_worker_thread(int force)
@@ -1032,13 +1066,15 @@ static void sync_worker_thread(int force)
   work = &worker->i[worker->i_reaped & WORK_I_MASK];
   thread_work_wait_sync(work, force);
 
-  do_samples_finish(work->SSumLR, work->RVB, work->ns_to,
+  do_samples_finish(work->SSumLR, work->ns_to,
    work->channels_silent, work->decode_pos);
 
   worker->i_reaped++;
   done = thread_get_i_done() - worker->i_reaped;
   used_space = worker->i_ready - worker->i_reaped;
  }
+ if (force)
+  thread_sync_caches();
 }
 
 #else
@@ -1116,7 +1152,7 @@ void do_samples(unsigned int cycles_to, int do_direct)
 
   if (do_direct || worker == NULL || !spu_config.iUseThread) {
    do_channels(ns_to);
-   do_samples_finish(spu.SSumLR, spu.RVB, ns_to, silentch, spu.decode_pos);
+   do_samples_finish(spu.SSumLR, ns_to, silentch, spu.decode_pos);
   }
   else {
    queue_channel_work(ns_to, silentch);
@@ -1131,7 +1167,7 @@ void do_samples(unsigned int cycles_to, int do_direct)
   spu.decode_pos = (spu.decode_pos + ns_to) & 0x1ff;
 }
 
-static void do_samples_finish(int *SSumLR, int *RVB, int ns_to,
+static void do_samples_finish(int *SSumLR, int ns_to,
  int silentch, int decode_pos)
 {
   int volmult = spu_config.iVolume;
@@ -1150,17 +1186,8 @@ static void do_samples_finish(int *SSumLR, int *RVB, int ns_to,
     spu.decode_dirty_ch &= ~(1<<3);
    }
 
-  //---------------------------------------------------//
-  // mix XA infos (if any)
-
   MixXA(SSumLR, ns_to, decode_pos);
   
-  ///////////////////////////////////////////////////////
-  // mix all channels (including reverb) into one buffer
-
-  if(spu_config.iUseReverb)
-   REVERBDo(SSumLR, RVB, ns_to);
-
   if((spu.spuCtrl&0x4000)==0) // muted? (rare, don't optimize for this)
    {
     memset(spu.pS, 0, ns_to * 2 * sizeof(spu.pS[0]));
@@ -1283,12 +1310,9 @@ void ClearWorkingState(void)
 }
 
 // SETUPSTREAMS: init most of the spu buffers
-void SetupStreams(void)
+static void SetupStreams(void)
 { 
- int i;
-
  spu.pSpuBuffer = (unsigned char *)malloc(32768);      // alloc mixing buffer
- spu.RVB = calloc(NSSIZE * 2, sizeof(spu.RVB[0]));
  spu.SSumLR = calloc(NSSIZE * 2, sizeof(spu.SSumLR[0]));
 
  spu.XAStart =                                         // alloc xa buffer
@@ -1303,26 +1327,14 @@ void SetupStreams(void)
  spu.CDDAPlay  = spu.CDDAStart;
  spu.CDDAFeed  = spu.CDDAStart;
 
- for(i=0;i<MAXCHAN;i++)                                // loop sound channels
-  {
-   spu.s_chan[i].ADSRX.SustainLevel = 0xf;             // -> init sustain
-   spu.s_chan[i].ADSRX.SustainIncrease = 1;
-   spu.s_chan[i].pLoop=spu.spuMemC;
-   spu.s_chan[i].pCurr=spu.spuMemC;
-  }
-
  ClearWorkingState();
-
- spu.bSpuInit=1;                                       // flag: we are inited
 }
 
 // REMOVESTREAMS: free most buffer
-void RemoveStreams(void)
+static void RemoveStreams(void)
 { 
  free(spu.pSpuBuffer);                                 // free mixing buffer
  spu.pSpuBuffer = NULL;
- free(spu.RVB);                                        // free reverb buffer
- spu.RVB = NULL;
  free(spu.SSumLR);
  spu.SSumLR = NULL;
  free(spu.XAStart);                                    // free XA buffer
@@ -1365,6 +1377,10 @@ static int thread_get_i_done(void)
  return worker->i_done;
 }
 
+static void thread_sync_caches(void)
+{
+}
+
 static void *spu_worker_thread(void *unused)
 {
  struct work_item *work;
@@ -1446,11 +1462,13 @@ static void exit_spu_thread(void)
 // SPUINIT: this func will be called first by the main emu
 long CALLBACK SPUinit(void)
 {
+ int i;
+
  spu.spuMemC = calloc(1, 512 * 1024);
- memset((void *)&rvb, 0, sizeof(REVERBInfo));
  InitADSR();
 
  spu.s_chan = calloc(MAXCHAN+1, sizeof(spu.s_chan[0])); // channel + 1 infos (1 is security for fmod handling)
+ spu.rvb = calloc(1, sizeof(REVERBInfo));
  spu.SB = calloc(MAXCHAN, sizeof(spu.SB[0]) * SB_SIZE);
 
  spu.spuAddr = 0;
@@ -1464,6 +1482,16 @@ long CALLBACK SPUinit(void)
 
  init_spu_thread();
 
+ for (i = 0; i < MAXCHAN; i++)                         // loop sound channels
+  {
+   spu.s_chan[i].ADSRX.SustainLevel = 0xf;             // -> init sustain
+   spu.s_chan[i].ADSRX.SustainIncrease = 1;
+   spu.s_chan[i].pLoop = spu.spuMemC;
+   spu.s_chan[i].pCurr = spu.spuMemC;
+  }
+
+ spu.bSpuInit=1;                                       // flag: we are inited
+
  return 0;
 }
 
@@ -1504,6 +1532,8 @@ long CALLBACK SPUshutdown(void)
  spu.SB = NULL;
  free(spu.s_chan);
  spu.s_chan = NULL;
+ free(spu.rvb);
+ spu.rvb = NULL;
 
  RemoveStreams();                                      // no more streaming
  spu.bSpuInit=0;
diff --git a/plugins/dfsound/spu_c64x.c b/plugins/dfsound/spu_c64x.c
index 200ab38..be10a6b 100644
--- a/plugins/dfsound/spu_c64x.c
+++ b/plugins/dfsound/spu_c64x.c
@@ -43,6 +43,8 @@ static struct {
 
  dsp_mem_region_t region;
  dsp_component_id_t compid;
+ unsigned int stale_caches:1;
+ unsigned int req_sent:1;
 } f;
 
 static void thread_work_start(void)
@@ -63,7 +65,7 @@ static void thread_work_start(void)
 
  // to start the DSP, dsp_rpc_send() must be used,
  // but before that, previous request must be finished
- if (worker->req_sent) {
+ if (f.req_sent) {
   if (worker->boot_cnt == worker->last_boot_cnt) {
    // hopefully still booting
    //printf("booting?\n");
@@ -74,7 +76,7 @@ static void thread_work_start(void)
   if (ret != 0) {
    fprintf(stderr, "dsp_rpc_recv failed: %d\n", ret);
    f.dsp_logbuf_print();
-   worker->req_sent = 0;
+   f.req_sent = 0;
    spu_config.iUseThread = 0;
    return;
   }
@@ -94,7 +96,7 @@ static void thread_work_start(void)
   spu_config.iUseThread = 0;
   return;
  }
- worker->req_sent = 1;
+ f.req_sent = 1;
 }
 
 static int thread_get_i_done(void)
@@ -108,14 +110,13 @@ static void thread_work_wait_sync(struct work_item *work, int force)
  int limit = 1000;
  int ns_to;
 
- ns_to = work->ns_to;
- f.dsp_cache_inv_virt(work->RVB, sizeof(work->RVB[0]) * 2 * ns_to);
- f.dsp_cache_inv_virt(work->SSumLR, sizeof(work->SSumLR[0]) * 2 * ns_to);
- __builtin_prefetch(work->RVB);
- __builtin_prefetch(work->SSumLR);
-
  while (worker->i_done == worker->i_reaped && limit-- > 0) {
-  if (!worker->active) {
+  if (!f.req_sent) {
+   printf("dsp: req not sent?\n");
+   break;
+  }
+
+  if (worker->boot_cnt != worker->last_boot_cnt && !worker->active) {
    printf("dsp: broken sync\n");
    worker->last_boot_cnt = ~0;
    break;
@@ -125,6 +126,13 @@ static void thread_work_wait_sync(struct work_item *work, int force)
   f.dsp_cache_inv_virt(&worker->i_done, 64);
  }
 
+ ns_to = work->ns_to;
+ f.dsp_cache_inv_virt(work->SSumLR, sizeof(work->SSumLR[0]) * 2 * ns_to);
+ preload(work->SSumLR);
+ preload(work->SSumLR + 64/4);
+
+ f.stale_caches = 1; // SB, spuMem
+
  if (limit == 0)
   printf("dsp: wait timeout\n");
 
@@ -132,7 +140,7 @@ static void thread_work_wait_sync(struct work_item *work, int force)
  if (worker->i_reaped != worker->i_done - 1)
   return;
 
- if (worker->req_sent && (force || worker->i_done == worker->i_ready)) {
+ if (f.req_sent && (force || worker->i_done == worker->i_ready)) {
   dsp_msg_t msg;
   int ret;
 
@@ -142,12 +150,20 @@ static void thread_work_wait_sync(struct work_item *work, int force)
    f.dsp_logbuf_print();
    spu_config.iUseThread = 0;
   }
-  worker->req_sent = 0;
+  f.req_sent = 0;
  }
+}
 
- if (force) {
+static void thread_sync_caches(void)
+{
+ if (f.stale_caches) {
   f.dsp_cache_inv_virt(spu.SB, sizeof(spu.SB[0]) * SB_SIZE * 24);
   f.dsp_cache_inv_virt(spu.spuMemC + 0x800, 0x800);
+  if (spu.rvb->StartAddr) {
+   int left = 0x40000 - spu.rvb->StartAddr;
+   f.dsp_cache_inv_virt(spu.spuMem + spu.rvb->StartAddr, left * 2);
+  }
+  f.stale_caches = 0;
  }
 }
 
@@ -220,9 +236,9 @@ static void init_spu_thread(void)
     mem->sizeof_region_mem, sizeof(*mem));
   goto fail_init;
  }
- if (mem->offsetof_s_chan1 != offsetof(typeof(*mem), s_chan[1])) {
+ if (mem->offsetof_s_chan1 != offsetof(typeof(*mem), in.s_chan[1])) {
   fprintf(stderr, "error: size mismatch 2: %d vs %zd\n",
-    mem->offsetof_s_chan1, offsetof(typeof(*mem), s_chan[1]));
+    mem->offsetof_s_chan1, offsetof(typeof(*mem), in.s_chan[1]));
   goto fail_init;
  }
  if (mem->offsetof_spos_3_20 != offsetof(typeof(*mem), worker.i[3].ch[20])) {
@@ -237,7 +253,9 @@ static void init_spu_thread(void)
  free(spu.SB);
  spu.SB = mem->SB;
  free(spu.s_chan);
- spu.s_chan = mem->s_chan;
+ spu.s_chan = mem->in.s_chan;
+ free(spu.rvb);
+ spu.rvb = &mem->in.rvb;
  worker = &mem->worker;
 
  printf("spu: C64x DSP ready (id=%d).\n", (int)f.compid);
@@ -267,8 +285,10 @@ static void exit_spu_thread(void)
  if (worker == NULL)
   return;
 
- if (worker->req_sent)
+ if (f.req_sent) {
   f.dsp_rpc_recv(&msg);
+  f.req_sent = 0;
+ }
 
  f.dsp_logbuf_print();
  f.dsp_shm_free(f.region);
@@ -277,6 +297,7 @@ static void exit_spu_thread(void)
  spu.spuMemC = NULL;
  spu.SB = NULL;
  spu.s_chan = NULL;
+ spu.rvb = NULL;
  worker = NULL;
 }
 
diff --git a/plugins/dfsound/spu_c64x.h b/plugins/dfsound/spu_c64x.h
index bb20cc3..8210e63 100644
--- a/plugins/dfsound/spu_c64x.h
+++ b/plugins/dfsound/spu_c64x.h
@@ -11,8 +11,12 @@ struct region_mem {
  // careful not to lose ARM writes by DSP overwriting
  // with old data when it's writing out neighbor cachelines
  int _pad1[128/4 - ((SB_SIZE * 24) & (128/4 - 1))];
- SPUCHAN s_chan[24 + 1];
- int _pad2[128/4 - ((sizeof(SPUCHAN) * 25 / 4) & (128/4 - 1))];
+ struct spu_in {
+  // these are not to be modified by DSP
+  SPUCHAN s_chan[24 + 1];
+  REVERBInfo rvb;
+ } in;
+ int _pad2[128/4 - ((sizeof(struct spu_in) / 4) & (128/4 - 1))];
  struct spu_worker worker;
  SPUConfig spu_config;
  // init/debug
diff --git a/plugins/dfsound/spu_c64x_dspcode.c b/plugins/dfsound/spu_c64x_dspcode.c
index 97d3028..b0352a9 100644
--- a/plugins/dfsound/spu_c64x_dspcode.c
+++ b/plugins/dfsound/spu_c64x_dspcode.c
@@ -31,6 +31,7 @@
 /* dummy deps, some bloat but avoids ifdef hell in SPU code.. */
 static void thread_work_start(void) {}
 static void thread_work_wait_sync(struct work_item *work, int force) {}
+static void thread_sync_caches(void) {}
 static int  thread_get_i_done(void) { return 0; }
 struct out_driver *out_current;
 void SetupSound(void) {}
@@ -38,7 +39,8 @@ void SetupSound(void) {}
 
 static void invalidate_cache(struct work_item *work)
 {
- syscalls.cache_inv(work, offsetof(typeof(*work), RVB), 1);
+ // see comment in writeout_cache()
+ //syscalls.cache_inv(work, offsetof(typeof(*work), SSumLR), 1);
  syscalls.cache_inv(spu.s_chan, sizeof(spu.s_chan[0]) * 24, 0);
  syscalls.cache_inv(work->SSumLR,
    sizeof(work->SSumLR[0]) * 2 * work->ns_to, 0);
@@ -48,14 +50,16 @@ static void writeout_cache(struct work_item *work)
 {
  int ns_to = work->ns_to;
 
- syscalls.cache_wb(work->RVB, sizeof(work->RVB[0]) * 2 * ns_to, 1);
  syscalls.cache_wb(work->SSumLR, sizeof(work->SSumLR[0]) * 2 * ns_to, 1);
+ // have to invalidate now, otherwise there is a race between
+ // DSP evicting dirty lines and ARM writing new data to this area
+ syscalls.cache_inv(work, offsetof(typeof(*work), SSumLR), 0);
 }
 
 static void do_processing(void)
 {
+ int left, dirty = 0, had_rvb = 0;
  struct work_item *work;
- int left, dirty = 0;
 
  while (worker->active)
  {
@@ -70,6 +74,8 @@ static void do_processing(void)
 
    work = &worker->i[worker->i_done & WORK_I_MASK];
    invalidate_cache(work);
+   had_rvb |= work->rvb_addr;
+   spu.spuCtrl = work->ctrl;
    do_channel_work(work);
    writeout_cache(work);
 
@@ -82,6 +88,11 @@ static void do_processing(void)
   if (dirty) {
    syscalls.cache_wb(spu.spuMemC + 0x800, 0x800, 1);
    syscalls.cache_wb(spu.SB, sizeof(spu.SB[0]) * SB_SIZE * 24, 1);
+   if (had_rvb) {
+    left = 0x40000 - spu.rvb->StartAddr;
+    syscalls.cache_wb(spu.spuMem + spu.rvb->StartAddr, left * 2, 1);
+    had_rvb = 0;
+   }
    dirty = 0;
    continue;
   }
@@ -105,12 +116,13 @@ static unsigned int exec(dsp_component_cmd_t cmd,
 
    spu.spuMemC = mem->spu_ram;
    spu.SB = mem->SB;
-   spu.s_chan = mem->s_chan;
+   spu.s_chan = mem->in.s_chan;
+   spu.rvb = &mem->in.rvb;
    worker = &mem->worker;
    memcpy(&spu_config, &mem->spu_config, sizeof(spu_config));
 
    mem->sizeof_region_mem = sizeof(*mem);
-   mem->offsetof_s_chan1 = offsetof(typeof(*mem), s_chan[1]);
+   mem->offsetof_s_chan1 = offsetof(typeof(*mem), in.s_chan[1]);
    mem->offsetof_spos_3_20 = offsetof(typeof(*mem), worker.i[3].ch[20]);
    // seems to be unneeded, no write-alloc? but just in case..
    syscalls.cache_wb(&mem->sizeof_region_mem, 3 * 4, 1);
@@ -126,7 +138,9 @@ static unsigned int exec(dsp_component_cmd_t cmd,
 
    // c64_tools lib does BCACHE_wbInvAll() when it receives mailbox irq,
    // but invalidate anyway in case c64_tools is ever fixed..
-   syscalls.cache_inv(mem, sizeof(mem->spu_ram) + sizeof(mem->SB), 0);
+   // XXX edit: don't bother as reverb is not handled, will fix if needed
+   //syscalls.cache_inv(mem, sizeof(mem->spu_ram) + sizeof(mem->SB), 0);
+   //syscalls.cache_inv(&mem->in, sizeof(mem->in), 0);
    break;
 
   default: