copyright : (C) 2002 by Pete Bernert
email : BlackDove@addcom.de
- Portions (C) Gražvydas "notaz" Ignotas, 2010-2012
+ Portions (C) Gražvydas "notaz" Ignotas, 2010-2012,2014,2015
***************************************************************************/
/***************************************************************************
* *
***************************************************************************/
+#include <assert.h>
#include "stdafx.h"
#define _IN_SPU
#include "externals.h"
#include "registers.h"
#include "out.h"
-#include "arm_features.h"
#include "spu_config.h"
+#include "spu.h"
-#ifdef ENABLE_NLS
-#include <libintl.h>
-#include <locale.h>
-#define _(x) gettext(x)
-#define N_(x) (x)
-#else
-#define _(x) (x)
-#define N_(x) (x)
+#ifdef __arm__
+#include "arm_features.h"
#endif
-#ifdef __ARM_ARCH_7A__
+#ifdef HAVE_ARMV7
#define ssat32_to_16(v) \
asm("ssat %0,#16,%1" : "=r" (v) : "r" (v))
#else
SPUInfo spu;
SPUConfig spu_config;
-// MAIN infos struct for each channel
-
-SPUCHAN s_chan[MAXCHAN+1]; // channel + 1 infos (1 is security for fmod handling)
-REVERBInfo rvb;
-
-// certain globals (were local before, but with the new timeproc I need em global)
-
-static const int f[8][2] = { { 0, 0 },
- { 60, 0 },
- { 115, -52 },
- { 98, -55 },
- { 122, -60 } };
+static int iFMod[NSSIZE];
+static int RVB[NSSIZE * 2];
int ChanBuf[NSSIZE];
-int SSumLR[NSSIZE*2];
-int iFMod[NSSIZE];
#define CDDA_BUFFER_SIZE (16384 * sizeof(uint32_t)) // must be power of 2
// /
//
-
-INLINE void InterpolateUp(int ch)
+static void InterpolateUp(sample_buf *sb, int sinc)
{
- if(s_chan[ch].SB[32]==1) // flag == 1? calc step and set flag... and don't change the value in this pass
+ int *SB = sb->SB;
+ if (sb->sinc_old != sinc)
+ {
+ sb->sinc_old = sinc;
+ SB[32] = 1;
+ }
+ if(SB[32]==1) // flag == 1? calc step and set flag... and don't change the value in this pass
{
- const int id1=s_chan[ch].SB[30]-s_chan[ch].SB[29]; // curr delta to next val
- const int id2=s_chan[ch].SB[31]-s_chan[ch].SB[30]; // and next delta to next-next val :)
+ const int id1=SB[30]-SB[29]; // curr delta to next val
+ const int id2=SB[31]-SB[30]; // and next delta to next-next val :)
- s_chan[ch].SB[32]=0;
+ SB[32]=0;
if(id1>0) // curr delta positive
{
if(id2<id1)
- {s_chan[ch].SB[28]=id1;s_chan[ch].SB[32]=2;}
+ {SB[28]=id1;SB[32]=2;}
else
if(id2<(id1<<1))
- s_chan[ch].SB[28]=(id1*s_chan[ch].sinc)/0x10000L;
+ SB[28]=(id1*sinc)>>16;
else
- s_chan[ch].SB[28]=(id1*s_chan[ch].sinc)/0x20000L;
+ SB[28]=(id1*sinc)>>17;
}
else // curr delta negative
{
if(id2>id1)
- {s_chan[ch].SB[28]=id1;s_chan[ch].SB[32]=2;}
+ {SB[28]=id1;SB[32]=2;}
else
if(id2>(id1<<1))
- s_chan[ch].SB[28]=(id1*s_chan[ch].sinc)/0x10000L;
+ SB[28]=(id1*sinc)>>16;
else
- s_chan[ch].SB[28]=(id1*s_chan[ch].sinc)/0x20000L;
+ SB[28]=(id1*sinc)>>17;
}
}
else
- if(s_chan[ch].SB[32]==2) // flag 1: calc step and set flag... and don't change the value in this pass
+ if(SB[32]==2) // flag 1: calc step and set flag... and don't change the value in this pass
{
- s_chan[ch].SB[32]=0;
+ SB[32]=0;
- s_chan[ch].SB[28]=(s_chan[ch].SB[28]*s_chan[ch].sinc)/0x20000L;
- //if(s_chan[ch].sinc<=0x8000)
- // s_chan[ch].SB[29]=s_chan[ch].SB[30]-(s_chan[ch].SB[28]*((0x10000/s_chan[ch].sinc)-1));
+ SB[28]=(SB[28]*sinc)>>17;
+ //if(sinc<=0x8000)
+ // SB[29]=SB[30]-(SB[28]*((0x10000/sinc)-1));
//else
- s_chan[ch].SB[29]+=s_chan[ch].SB[28];
+ SB[29]+=SB[28];
}
else // no flags? add bigger val (if possible), calc smaller step, set flag1
- s_chan[ch].SB[29]+=s_chan[ch].SB[28];
+ SB[29]+=SB[28];
}
//
// even easier interpolation on downsampling, also no special filter, again just "Pete's common sense" tm
//
-INLINE void InterpolateDown(int ch)
+static void InterpolateDown(sample_buf *sb, int sinc)
{
- if(s_chan[ch].sinc>=0x20000L) // we would skip at least one val?
+ int *SB = sb->SB;
+ if(sinc>=0x20000L) // we would skip at least one val?
{
- s_chan[ch].SB[29]+=(s_chan[ch].SB[30]-s_chan[ch].SB[29])/2; // add easy weight
- if(s_chan[ch].sinc>=0x30000L) // we would skip even more vals?
- s_chan[ch].SB[29]+=(s_chan[ch].SB[31]-s_chan[ch].SB[30])/2;// add additional next weight
+ SB[29]+=(SB[30]-SB[29])/2; // add easy weight
+ if(sinc>=0x30000L) // we would skip even more vals?
+ SB[29]+=(SB[31]-SB[30])/2; // add additional next weight
}
}
////////////////////////////////////////////////////////////////////////
-// helpers for gauss interpolation
-
-#define gval0 (((short*)(&s_chan[ch].SB[29]))[gpos&3])
-#define gval(x) ((int)((short*)(&s_chan[ch].SB[29]))[(gpos+x)&3])
#include "gauss_i.h"
-
-////////////////////////////////////////////////////////////////////////
-
#include "xa.c"
-static void do_irq(void)
+static void do_irq(int cycles_after)
{
- //if(!(spu.spuStat & STAT_IRQ))
+ if (spu.spuStat & STAT_IRQ)
+ log_unhandled("spu: missed irq?\n");
+ else
{
spu.spuStat |= STAT_IRQ; // asserted status?
- if(spu.irqCallback) spu.irqCallback();
+ if (spu.irqCallback)
+ spu.irqCallback(cycles_after);
}
}
static int check_irq(int ch, unsigned char *pos)
{
- if((spu.spuCtrl & CTRL_IRQ) && pos == spu.pSpuIrq)
+ if((spu.spuCtrl & (CTRL_ON|CTRL_IRQ)) == (CTRL_ON|CTRL_IRQ) && pos == spu.pSpuIrq)
{
- //printf("ch%d irq %04x\n", ch, pos - spu.spuMemC);
- do_irq();
+ //printf("ch%d irq %04zx\n", ch, pos - spu.spuMemC);
+ do_irq(0);
return 1;
}
return 0;
}
+void check_irq_io(unsigned int addr)
+{
+ unsigned int irq_addr = regAreaGet(H_SPUirqAddr) << 3;
+ //addr &= ~7; // ?
+ if((spu.spuCtrl & (CTRL_ON|CTRL_IRQ)) == (CTRL_ON|CTRL_IRQ) && addr == irq_addr)
+ {
+ //printf("io irq %04x\n", irq_addr);
+ do_irq(0);
+ }
+}
+
+void do_irq_io(int cycles_after)
+{
+ if ((spu.spuCtrl & (CTRL_ON|CTRL_IRQ)) == (CTRL_ON|CTRL_IRQ))
+ {
+ do_irq(cycles_after);
+ }
+}
+
////////////////////////////////////////////////////////////////////////
// START SOUND... called by main thread to setup a new sound on a channel
////////////////////////////////////////////////////////////////////////
-INLINE void StartSound(int ch)
+static void ResetInterpolation(sample_buf *sb)
{
- StartADSR(ch);
- StartREVERB(ch);
+ memset(&sb->interp, 0, sizeof(sb->interp));
+ sb->sinc_old = -1;
+}
- // fussy timing issues - do in VoiceOn
- //s_chan[ch].pCurr=s_chan[ch].pStart; // set sample start
- //s_chan[ch].bStop=0;
- //s_chan[ch].bOn=1;
+static void StartSoundSB(sample_buf *sb)
+{
+ sb->SB[26] = 0; // init mixing vars
+ sb->SB[27] = 0;
+ ResetInterpolation(sb);
+}
- s_chan[ch].SB[26]=0; // init mixing vars
- s_chan[ch].SB[27]=0;
- s_chan[ch].iSBPos=28;
+static void StartSoundMain(int ch)
+{
+ SPUCHAN *s_chan = &spu.s_chan[ch];
- s_chan[ch].SB[29]=0; // init our interpolation helpers
- s_chan[ch].SB[30]=0;
+ StartADSR(ch);
+ StartREVERB(ch);
- if(spu_config.iUseInterpolation>=2) // gauss interpolation?
- {s_chan[ch].spos=0x30000L;s_chan[ch].SB[28]=0;} // -> start with more decoding
- else {s_chan[ch].spos=0x10000L;s_chan[ch].SB[31]=0;} // -> no/simple interpolation starts with one 44100 decoding
+ s_chan->prevflags = 2;
+ s_chan->iSBPos = 27;
+ s_chan->spos = 0;
+ s_chan->bStarting = 1;
+
+ s_chan->pCurr = spu.spuMemC + ((regAreaGetCh(ch, 6) & ~1) << 3);
spu.dwNewChannel&=~(1<<ch); // clear new channel bit
+ spu.dwChannelDead&=~(1<<ch);
+ spu.dwChannelsAudible|=1<<ch;
+}
+
+static void StartSound(int ch)
+{
+ StartSoundMain(ch);
+ StartSoundSB(&spu.sb[ch]);
}
////////////////////////////////////////////////////////////////////////
// ALL KIND OF HELPERS
////////////////////////////////////////////////////////////////////////
-INLINE int FModChangeFrequency(int ch,int ns)
+INLINE int FModChangeFrequency(int pitch, int ns, int *fmod_buf)
{
- unsigned int NP=s_chan[ch].iRawPitch;
- int sinc;
-
- NP=((32768L+iFMod[ns])*NP)/32768L;
+ pitch = (signed short)pitch;
+ pitch = ((32768 + fmod_buf[ns]) * pitch) >> 15;
+ pitch &= 0xffff;
+ if (pitch > 0x3fff)
+ pitch = 0x3fff;
- if(NP>0x3fff) NP=0x3fff;
- if(NP<0x1) NP=0x1;
+ fmod_buf[ns] = 0;
- sinc=NP<<4; // calc frequency
- if(spu_config.iUseInterpolation==1) // freq change in simple interpolation mode
- s_chan[ch].SB[32]=1;
- iFMod[ns]=0;
-
- return sinc;
+ return pitch << 4;
}
-////////////////////////////////////////////////////////////////////////
-
-INLINE void StoreInterpolationVal(int ch,int fa)
+INLINE void StoreInterpolationGaussCubic(sample_buf *sb, int fa)
{
- if(s_chan[ch].bFMod==2) // fmod freq channel
- s_chan[ch].SB[29]=fa;
- else
- {
- ssat32_to_16(fa);
-
- if(spu_config.iUseInterpolation>=2) // gauss/cubic interpolation
- {
- int gpos = s_chan[ch].SB[28];
- gval0 = fa;
- gpos = (gpos+1) & 3;
- s_chan[ch].SB[28] = gpos;
- }
- else
- if(spu_config.iUseInterpolation==1) // simple interpolation
- {
- s_chan[ch].SB[28] = 0;
- s_chan[ch].SB[29] = s_chan[ch].SB[30]; // -> helpers for simple linear interpolation: delay real val for two slots, and calc the two deltas, for a 'look at the future behaviour'
- s_chan[ch].SB[30] = s_chan[ch].SB[31];
- s_chan[ch].SB[31] = fa;
- s_chan[ch].SB[32] = 1; // -> flag: calc new interolation
- }
- else s_chan[ch].SB[29]=fa; // no interpolation
- }
+ int gpos = sb->interp.gauss.pos & 3;
+ sb->interp.gauss.val[gpos++] = fa;
+ sb->interp.gauss.pos = gpos & 3;
}
-////////////////////////////////////////////////////////////////////////
+#define gval(x) (int)sb->interp.gauss.val[(gpos + x) & 3]
-INLINE int iGetInterpolationVal(int ch, int spos)
+INLINE int GetInterpolationCubic(const sample_buf *sb, int spos)
{
+ int gpos = sb->interp.gauss.pos;
+ int xd = (spos >> 1) + 1;
int fa;
- if(s_chan[ch].bFMod==2) return s_chan[ch].SB[29];
-
- switch(spu_config.iUseInterpolation)
- {
- //--------------------------------------------------//
- case 3: // cubic interpolation
- {
- long xd;int gpos;
- xd = (spos >> 1)+1;
- gpos = s_chan[ch].SB[28];
-
- fa = gval(3) - 3*gval(2) + 3*gval(1) - gval0;
- fa *= (xd - (2<<15)) / 6;
- fa >>= 15;
- fa += gval(2) - gval(1) - gval(1) + gval0;
- fa *= (xd - (1<<15)) >> 1;
- fa >>= 15;
- fa += gval(1) - gval0;
- fa *= xd;
- fa >>= 15;
- fa = fa + gval0;
-
- } break;
- //--------------------------------------------------//
- case 2: // gauss interpolation
- {
- int vl, vr;int gpos;
- vl = (spos >> 6) & ~3;
- gpos = s_chan[ch].SB[28];
- vr=(gauss[vl]*(int)gval0)&~2047;
- vr+=(gauss[vl+1]*gval(1))&~2047;
- vr+=(gauss[vl+2]*gval(2))&~2047;
- vr+=(gauss[vl+3]*gval(3))&~2047;
- fa = vr>>11;
- } break;
- //--------------------------------------------------//
- case 1: // simple interpolation
- {
- if(s_chan[ch].sinc<0x10000L) // -> upsampling?
- InterpolateUp(ch); // --> interpolate up
- else InterpolateDown(ch); // --> else down
- fa=s_chan[ch].SB[29];
- } break;
- //--------------------------------------------------//
- default: // no interpolation
- {
- fa=s_chan[ch].SB[29];
- } break;
- //--------------------------------------------------//
- }
-
+ fa = gval(3) - 3*gval(2) + 3*gval(1) - gval(0);
+ fa *= (xd - (2<<15)) / 6;
+ fa >>= 15;
+ fa += gval(2) - gval(1) - gval(1) + gval(0);
+ fa *= (xd - (1<<15)) >> 1;
+ fa >>= 15;
+ fa += gval(1) - gval(0);
+ fa *= xd;
+ fa >>= 15;
+ fa = fa + gval(0);
return fa;
}
+INLINE int GetInterpolationGauss(const sample_buf *sb, int spos)
+{
+ int gpos = sb->interp.gauss.pos;
+ int vl = (spos >> 6) & ~3;
+ int vr;
+ vr = (gauss[vl+0] * gval(0)) >> 15;
+ vr += (gauss[vl+1] * gval(1)) >> 15;
+ vr += (gauss[vl+2] * gval(2)) >> 15;
+ vr += (gauss[vl+3] * gval(3)) >> 15;
+ return vr;
+}
+
static void decode_block_data(int *dest, const unsigned char *src, int predict_nr, int shift_factor)
{
+ static const int f[16][2] = {
+ { 0, 0 },
+ { 60, 0 },
+ { 115, -52 },
+ { 98, -55 },
+ { 122, -60 }
+ };
int nSample;
int fa, s_1, s_2, d, s;
d = (int)*src;
s = (int)(signed short)((d & 0x0f) << 12);
- fa = s >> shift_factor;
+ fa = s >> shift_factor;
fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6);
- s_2=s_1;s_1=fa;
+ ssat32_to_16(fa);
+ s_2 = s_1; s_1 = fa;
dest[nSample++] = fa;
s = (int)(signed short)((d & 0xf0) << 8);
- fa = s >> shift_factor;
+ fa = s >> shift_factor;
fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6);
- s_2=s_1;s_1=fa;
+ ssat32_to_16(fa);
+ s_2 = s_1; s_1 = fa;
dest[nSample++] = fa;
}
}
-static int decode_block(int ch)
+static int decode_block(void *unused, int ch, int *SB)
{
+ SPUCHAN *s_chan = &spu.s_chan[ch];
unsigned char *start;
- int predict_nr,shift_factor,flags;
- int stop = 0;
+ int predict_nr, shift_factor, flags;
int ret = 0;
- start = s_chan[ch].pCurr; // set up the current pos
- if(start == spu.spuMemC) // ?
- stop = 1;
+ start = s_chan->pCurr; // set up the current pos
+ if (start - spu.spuMemC < 0x1000) { // ?
+ //log_unhandled("ch%02d plays decode bufs @%05lx\n",
+ // ch, (long)(start - spu.spuMemC));
+ ret = 1;
+ }
- if(s_chan[ch].prevflags&1) // 1: stop/loop
+ if (s_chan->prevflags & 1) // 1: stop/loop
{
- if(!(s_chan[ch].prevflags&2))
- stop = 1;
+ if (!(s_chan->prevflags & 2))
+ ret = 1;
- start = s_chan[ch].pLoop;
+ start = s_chan->pLoop;
}
- else
- ret = check_irq(ch, start); // hack, see check_irq below..
- if(stop)
- {
- spu.dwChannelOn &= ~(1<<ch); // -> turn everything off
- s_chan[ch].bStop = 1;
- s_chan[ch].ADSRX.EnvelopeVol = 0;
- }
+ check_irq(ch, start);
- predict_nr=(int)start[0];
- shift_factor=predict_nr&0xf;
+ predict_nr = start[0];
+ shift_factor = predict_nr & 0xf;
predict_nr >>= 4;
- decode_block_data(s_chan[ch].SB, start + 2, predict_nr, shift_factor);
-
- flags=(int)start[1];
- if(flags&4)
- s_chan[ch].pLoop=start; // loop adress
+ decode_block_data(SB, start + 2, predict_nr, shift_factor);
- start+=16;
+ flags = start[1];
+ if (flags & 4 && !s_chan->bIgnoreLoop)
+ s_chan->pLoop = start; // loop adress
- if(flags&1) { // 1: stop/loop
- start = s_chan[ch].pLoop;
- ret |= check_irq(ch, start); // hack.. :(
- }
-
- if (start - spu.spuMemC >= 0x80000)
- start = spu.spuMemC;
+ start += 16;
- s_chan[ch].pCurr = start; // store values for next cycle
- s_chan[ch].prevflags = flags;
+ s_chan->pCurr = start; // store values for next cycle
+ s_chan->prevflags = flags;
+ s_chan->bStarting = 0;
return ret;
}
// do block, but ignore sample data
static int skip_block(int ch)
{
- unsigned char *start = s_chan[ch].pCurr;
- int flags = start[1];
- int ret = check_irq(ch, start);
+ SPUCHAN *s_chan = &spu.s_chan[ch];
+ unsigned char *start = s_chan->pCurr;
+ int flags;
+ int ret = 0;
- if(s_chan[ch].prevflags & 1)
- start = s_chan[ch].pLoop;
+ if (s_chan->prevflags & 1) {
+ if (!(s_chan->prevflags & 2))
+ ret = 1;
- if(flags & 4)
- s_chan[ch].pLoop = start;
+ start = s_chan->pLoop;
+ }
- start += 16;
+ check_irq(ch, start);
+
+ flags = start[1];
+ if (flags & 4 && !s_chan->bIgnoreLoop)
+ s_chan->pLoop = start;
- if(flags & 1)
- start = s_chan[ch].pLoop;
+ start += 16;
- s_chan[ch].pCurr = start;
- s_chan[ch].prevflags = flags;
+ s_chan->pCurr = start;
+ s_chan->prevflags = flags;
+ s_chan->bStarting = 0;
return ret;
}
// if irq is going to trigger sooner than in upd_samples, set upd_samples
static void scan_for_irq(int ch, unsigned int *upd_samples)
{
+ SPUCHAN *s_chan = &spu.s_chan[ch];
int pos, sinc, sinc_inv, end;
unsigned char *block;
int flags;
- block = s_chan[ch].pCurr;
- pos = s_chan[ch].spos;
- sinc = s_chan[ch].sinc;
+ block = s_chan->pCurr;
+ pos = s_chan->spos;
+ sinc = s_chan->sinc;
end = pos + *upd_samples * sinc;
+ if (s_chan->prevflags & 1) // 1: stop/loop
+ block = s_chan->pLoop;
- pos += (28 - s_chan[ch].iSBPos) << 16;
+ pos += (28 - s_chan->iSBPos) << 16;
while (pos < end)
{
if (block == spu.pSpuIrq)
flags = block[1];
block += 16;
if (flags & 1) { // 1: stop/loop
- block = s_chan[ch].pLoop;
- if (block == spu.pSpuIrq) // hack.. (see decode_block)
- break;
+ block = s_chan->pLoop;
}
pos += 28 << 16;
}
if (pos < end)
{
- sinc_inv = s_chan[ch].sinc_inv;
+ sinc_inv = s_chan->sinc_inv;
if (sinc_inv == 0)
- sinc_inv = s_chan[ch].sinc_inv = (0x80000000u / (uint32_t)sinc) << 1;
+ sinc_inv = s_chan->sinc_inv = (0x80000000u / (uint32_t)sinc) << 1;
- pos -= s_chan[ch].spos;
+ pos -= s_chan->spos;
*upd_samples = (((uint64_t)pos * sinc_inv) >> 32) + 1;
//xprintf("ch%02d: irq sched: %3d %03d\n",
// ch, *upd_samples, *upd_samples * 60 * 263 / 44100);
}
}
-#define make_do_samples(name, fmod_code, interp_start, interp1_code, interp2_code, interp_end) \
-static noinline int do_samples_##name(int ch, int ns, int ns_to) \
+#define make_do_samples(name, fmod_code, interp_start, interp_store, interp_get, interp_end) \
+static noinline int name(int *dst, \
+ int (*decode_f)(void *context, int ch, int *SB), void *ctx, \
+ int ch, int ns_to, sample_buf *sb, int sinc, int *spos, int *sbpos) \
{ \
- int sinc = s_chan[ch].sinc; \
- int spos = s_chan[ch].spos; \
- int sbpos = s_chan[ch].iSBPos; \
- int *SB = s_chan[ch].SB; \
- int ret = -1; \
- int d, fa; \
+ int ns, d, fa; \
+ int ret = ns_to; \
interp_start; \
\
- for (; ns < ns_to; ns++) \
+ for (ns = 0; ns < ns_to; ns++) \
{ \
fmod_code; \
\
- spos += sinc; \
- while (spos >= 0x10000) \
+ *spos += sinc; \
+ while (*spos >= 0x10000) \
{ \
- fa = SB[sbpos++]; \
- if(sbpos >= 28) \
+ fa = sb->SB[(*sbpos)++]; \
+ if (*sbpos >= 28) \
{ \
- sbpos = 0; \
- d = decode_block(ch); \
- if(d) \
- ret = /*ns_to =*/ ns + 1; \
+ *sbpos = 0; \
+ d = decode_f(ctx, ch, sb->SB); \
+ if (d && ns < ret) \
+ ret = ns; \
} \
\
- interp1_code; \
- spos -= 0x10000; \
+ interp_store; \
+ *spos -= 0x10000; \
} \
\
- interp2_code; \
+ interp_get; \
} \
\
- s_chan[ch].sinc = sinc; \
- s_chan[ch].spos = spos; \
- s_chan[ch].iSBPos = sbpos; \
interp_end; \
\
return ret; \
}
-#define fmod_recv_check \
- if(s_chan[ch].bFMod==1 && iFMod[ns]) \
- sinc = FModChangeFrequency(ch,ns)
-
-make_do_samples(default, fmod_recv_check, ,
- StoreInterpolationVal(ch, fa),
- ChanBuf[ns] = iGetInterpolationVal(ch, spos), )
-make_do_samples(noint, , fa = s_chan[ch].SB[29], , ChanBuf[ns] = fa, s_chan[ch].SB[29] = fa)
-
+// helpers for simple linear interpolation: delay real val for two slots,
+// and calc the two deltas, for a 'look at the future behaviour'
#define simple_interp_store \
- s_chan[ch].SB[28] = 0; \
- s_chan[ch].SB[29] = s_chan[ch].SB[30]; \
- s_chan[ch].SB[30] = s_chan[ch].SB[31]; \
- s_chan[ch].SB[31] = fa; \
- s_chan[ch].SB[32] = 1
+ sb->SB[28] = 0; \
+ sb->SB[29] = sb->SB[30]; \
+ sb->SB[30] = sb->SB[31]; \
+ sb->SB[31] = fa; \
+ sb->SB[32] = 1
#define simple_interp_get \
- if(sinc<0x10000) /* -> upsampling? */ \
- InterpolateUp(ch); /* --> interpolate up */ \
- else InterpolateDown(ch); /* --> else down */ \
- ChanBuf[ns] = s_chan[ch].SB[29]
-
-make_do_samples(simple, , ,
+ if(sinc<0x10000) /* -> upsampling? */ \
+ InterpolateUp(sb, sinc); /* --> interpolate up */ \
+ else InterpolateDown(sb, sinc); /* --> else down */ \
+ dst[ns] = sb->SB[29]
+
+make_do_samples(do_samples_nointerp, , fa = sb->SB[29],
+ , dst[ns] = fa, sb->SB[29] = fa)
+make_do_samples(do_samples_simple, , ,
simple_interp_store, simple_interp_get, )
+make_do_samples(do_samples_gauss, , ,
+ StoreInterpolationGaussCubic(sb, fa),
+ dst[ns] = GetInterpolationGauss(sb, *spos), )
+make_do_samples(do_samples_cubic, , ,
+ StoreInterpolationGaussCubic(sb, fa),
+ dst[ns] = GetInterpolationCubic(sb, *spos), )
+make_do_samples(do_samples_fmod,
+ sinc = FModChangeFrequency(spu.s_chan[ch].iRawPitch, ns, iFMod), ,
+ StoreInterpolationGaussCubic(sb, fa),
+ dst[ns] = GetInterpolationGauss(sb, *spos), )
+
+INLINE int do_samples_adpcm(int *dst,
+ int (*decode_f)(void *context, int ch, int *SB), void *ctx,
+ int ch, int ns_to, int fmod, sample_buf *sb, int sinc, int *spos, int *sbpos)
+{
+ int interp = spu.interpolation;
+ if (fmod == 1)
+ return do_samples_fmod(dst, decode_f, ctx, ch, ns_to, sb, sinc, spos, sbpos);
+ if (fmod)
+ interp = 2;
+ switch (interp) {
+ case 0:
+ return do_samples_nointerp(dst, decode_f, ctx, ch, ns_to, sb, sinc, spos, sbpos);
+ case 1:
+ return do_samples_simple (dst, decode_f, ctx, ch, ns_to, sb, sinc, spos, sbpos);
+ default:
+ return do_samples_gauss (dst, decode_f, ctx, ch, ns_to, sb, sinc, spos, sbpos);
+ case 3:
+ return do_samples_cubic (dst, decode_f, ctx, ch, ns_to, sb, sinc, spos, sbpos);
+ }
+}
+
+static int do_samples_skip(int ch, int ns_to)
+{
+ SPUCHAN *s_chan = &spu.s_chan[ch];
+ int spos = s_chan->spos;
+ int sinc = s_chan->sinc;
+ int ret = ns_to, ns, d;
+
+ spos += s_chan->iSBPos << 16;
+
+ for (ns = 0; ns < ns_to; ns++)
+ {
+ spos += sinc;
+ while (spos >= 28*0x10000)
+ {
+ d = skip_block(ch);
+ if (d && ns < ret)
+ ret = ns;
+ spos -= 28*0x10000;
+ }
+ }
+
+ s_chan->iSBPos = spos >> 16;
+ s_chan->spos = spos & 0xffff;
+
+ return ret;
+}
-static noinline int do_samples_noise(int ch, int ns, int ns_to)
+static int do_samples_skip_fmod(int ch, int ns_to, int *fmod_buf)
{
- int level, shift, bit;
- int ret = -1, d;
+ SPUCHAN *s_chan = &spu.s_chan[ch];
+ int spos = s_chan->spos;
+ int ret = ns_to, ns, d;
- s_chan[ch].spos += s_chan[ch].sinc * (ns_to - ns);
- while (s_chan[ch].spos >= 28*0x10000)
+ spos += s_chan->iSBPos << 16;
+
+ for (ns = 0; ns < ns_to; ns++)
{
- d = skip_block(ch);
- if (d)
- ret = ns_to;
- s_chan[ch].spos -= 28*0x10000;
+ spos += FModChangeFrequency(s_chan->iRawPitch, ns, fmod_buf);
+ while (spos >= 28*0x10000)
+ {
+ d = skip_block(ch);
+ if (d && ns < ret)
+ ret = ns;
+ spos -= 28*0x10000;
+ }
}
+ s_chan->iSBPos = spos >> 16;
+ s_chan->spos = spos & 0xffff;
+
+ return ret;
+}
+
+static void do_lsfr_samples(int *dst, int ns_to, int ctrl,
+ unsigned int *dwNoiseCount, unsigned int *dwNoiseVal)
+{
+ unsigned int counter = *dwNoiseCount;
+ unsigned int val = *dwNoiseVal;
+ unsigned int level, shift, bit;
+ int ns;
+
// modified from DrHell/shalma, no fraction
- level = (spu.spuCtrl >> 10) & 0x0f;
+ level = (ctrl >> 10) & 0x0f;
level = 0x8000 >> level;
- for (; ns < ns_to; ns++)
+ for (ns = 0; ns < ns_to; ns++)
{
- spu.dwNoiseCount += 2;
- if (spu.dwNoiseCount >= level)
+ counter += 2;
+ if (counter >= level)
{
- spu.dwNoiseCount -= level;
- shift = (spu.dwNoiseVal >> 10) & 0x1f;
+ counter -= level;
+ shift = (val >> 10) & 0x1f;
bit = (0x69696969 >> shift) & 1;
- if (spu.dwNoiseVal & 0x8000)
- bit ^= 1;
- spu.dwNoiseVal = (spu.dwNoiseVal << 1) | bit;
+ bit ^= (val >> 15) & 1;
+ val = (val << 1) | bit;
}
- ChanBuf[ns] = (signed short)spu.dwNoiseVal;
+ dst[ns] = (signed short)val;
}
+ *dwNoiseCount = counter;
+ *dwNoiseVal = val;
+}
+
+static int do_samples_noise(int *dst, int ch, int ns_to)
+{
+ int ret;
+
+ ret = do_samples_skip(ch, ns_to);
+
+ do_lsfr_samples(dst, ns_to, spu.spuCtrl, &spu.dwNoiseCount, &spu.dwNoiseVal);
+
return ret;
}
#ifdef HAVE_ARMV5
// asm code; lv and rv must be 0-3fff
-extern void mix_chan(int start, int count, int lv, int rv);
-extern void mix_chan_rvb(int start, int count, int lv, int rv, int *rvb);
+extern void mix_chan(int *SSumLR, int count, int lv, int rv);
+extern void mix_chan_rvb(int *SSumLR, int count, int lv, int rv, int *rvb);
#else
-static void mix_chan(int start, int count, int lv, int rv)
+static void mix_chan(int *SSumLR, int count, int lv, int rv)
{
- int *dst = SSumLR + start * 2;
- const int *src = ChanBuf + start;
+ const int *src = ChanBuf;
int l, r;
while (count--)
l = (sval * lv) >> 14;
r = (sval * rv) >> 14;
- *dst++ += l;
- *dst++ += r;
+ *SSumLR++ += l;
+ *SSumLR++ += r;
}
}
-static void mix_chan_rvb(int start, int count, int lv, int rv, int *rvb)
+static void mix_chan_rvb(int *SSumLR, int count, int lv, int rv, int *rvb)
{
- int *dst = SSumLR + start * 2;
- int *drvb = rvb + start * 2;
- const int *src = ChanBuf + start;
+ const int *src = ChanBuf;
+ int *dst = SSumLR;
+ int *drvb = rvb;
int l, r;
while (count--)
// 0x0800-0x0bff Voice 1
// 0x0c00-0x0fff Voice 3
-static noinline void do_decode_bufs(int which, int start, int count)
+static noinline void do_decode_bufs(unsigned short *mem, int which,
+ int count, int decode_pos)
{
- const int *src = ChanBuf + start;
- unsigned short *dst = &spu.spuMem[0x800/2 + which*0x400/2];
- int cursor = spu.decode_pos + start;
+ unsigned short *dst = &mem[0x800/2 + which*0x400/2];
+ const int *src = ChanBuf;
+ int cursor = decode_pos;
while (count-- > 0)
{
// decode_pos is updated and irqs are checked later, after voice loop
}
+static void do_silent_chans(int ns_to, int silentch)
+{
+ unsigned int mask;
+ SPUCHAN *s_chan;
+ int ch;
+
+ mask = silentch & 0xffffff;
+ for (ch = 0; mask != 0; ch++, mask >>= 1)
+ {
+ if (!(mask & 1)) continue;
+ if (spu.dwChannelDead & (1<<ch)) continue;
+
+ s_chan = &spu.s_chan[ch];
+ if (s_chan->pCurr > spu.pSpuIrq && s_chan->pLoop > spu.pSpuIrq)
+ continue;
+
+ s_chan->spos += s_chan->iSBPos << 16;
+ s_chan->iSBPos = 0;
+
+ s_chan->spos += s_chan->sinc * ns_to;
+ while (s_chan->spos >= 28 * 0x10000)
+ {
+ unsigned char *start = s_chan->pCurr;
+
+ skip_block(ch);
+ if (start == s_chan->pCurr || start - spu.spuMemC < 0x1000)
+ {
+ // looping on self or stopped(?)
+ spu.dwChannelDead |= 1<<ch;
+ s_chan->spos = 0;
+ break;
+ }
+
+ s_chan->spos -= 28 * 0x10000;
+ }
+ }
+}
+
+static void do_channels(int ns_to)
+{
+ unsigned int mask;
+ int do_rvb, ch, d;
+ SPUCHAN *s_chan;
+
+ if (unlikely(spu.interpolation != spu_config.iUseInterpolation))
+ {
+ spu.interpolation = spu_config.iUseInterpolation;
+ mask = spu.dwChannelsAudible & 0xffffff;
+ for (ch = 0; mask != 0; ch++, mask >>= 1)
+ if (mask & 1)
+ ResetInterpolation(&spu.sb[ch]);
+ }
+
+ 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) {
+ if (mask & 1)
+ StartSound(ch);
+ }
+
+ mask = spu.dwChannelsAudible & 0xffffff;
+ for (ch = 0; mask != 0; ch++, mask >>= 1) // loop em all...
+ {
+ if (!(mask & 1)) continue; // channel not playing? next
+
+ s_chan = &spu.s_chan[ch];
+ if (s_chan->bNoise)
+ d = do_samples_noise(ChanBuf, ch, ns_to);
+ else
+ d = do_samples_adpcm(ChanBuf, decode_block, NULL, ch, ns_to, s_chan->bFMod,
+ &spu.sb[ch], s_chan->sinc, &s_chan->spos, &s_chan->iSBPos);
+
+ if (!s_chan->bStarting) {
+ d = MixADSR(ChanBuf, &s_chan->ADSRX, d);
+ if (d < ns_to) {
+ spu.dwChannelsAudible &= ~(1 << ch);
+ s_chan->ADSRX.State = ADSR_RELEASE;
+ s_chan->ADSRX.EnvelopeVol = 0;
+ memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0]));
+ }
+ }
+
+ if (ch == 1 || ch == 3)
+ {
+ do_decode_bufs(spu.spuMem, ch/2, ns_to, spu.decode_pos);
+ spu.decode_dirty_ch |= 1 << ch;
+ }
+
+ if (s_chan->bFMod == 2) // fmod freq channel
+ memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
+ if (!(spu.spuCtrl & CTRL_MUTE))
+ ;
+ else 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);
+ }
+
+ MixCD(spu.SSumLR, RVB, ns_to, spu.decode_pos);
+
+ if (spu.rvb->StartAddr) {
+ if (do_rvb)
+ 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 ns_to,
+ int silentch, int decode_pos);
+
+// optional worker thread handling
+
+#if P_HAVE_PTHREAD || defined(WANT_THREAD_CODE)
+
+// worker thread state
+static struct spu_worker {
+ union {
+ struct {
+ unsigned char exit_thread;
+ unsigned char prev_work_in_thread;
+ unsigned char pad[2];
+ unsigned int i_ready;
+ unsigned int i_reaped;
+ unsigned int last_boot_cnt; // dsp
+ unsigned int ram_dirty;
+ unsigned int channels_last;
+ };
+ // aligning for C64X_DSP
+ unsigned int _pad0[128/4];
+ };
+ union {
+ struct {
+ unsigned int i_done;
+ unsigned int active; // dsp
+ unsigned int boot_cnt;
+ };
+ unsigned int _pad1[128/4];
+ };
+ struct work_item {
+ int ns_to;
+ int ctrl;
+ int decode_pos;
+ int rvb_addr;
+ unsigned int channels_new;
+ unsigned int channels_on;
+ unsigned int channels_silent;
+ struct {
+ int spos;
+ int sbpos;
+ int sinc;
+ int start;
+ int loop;
+ short vol_l;
+ short vol_r;
+ unsigned short ns_to;
+ unsigned short bNoise:1;
+ unsigned short bFMod:2;
+ unsigned short bRVBActive:1;
+ unsigned short bStarting:1;
+ ADSRInfoEx adsr;
+ } ch[24];
+ int SSumLR[NSSIZE * 2];
+ } i[4];
+} *worker;
+
+#define WORK_MAXCNT (sizeof(worker->i) / sizeof(worker->i[0]))
+#define WORK_I_MASK (WORK_MAXCNT - 1)
+
+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)
+{
+ const unsigned char *ram = spu.spuMemC;
+ int predict_nr, shift_factor, flags;
+ struct work_item *work = context;
+ int start = work->ch[ch].start;
+ int loop = work->ch[ch].loop;
+
+ predict_nr = ram[start];
+ shift_factor = predict_nr & 0xf;
+ predict_nr >>= 4;
+
+ decode_block_data(SB, ram + start + 2, predict_nr, shift_factor);
+
+ flags = ram[start + 1];
+ if (flags & 4)
+ loop = start; // loop adress
+
+ start += 16;
+
+ if (flags & 1) // 1: stop/loop
+ start = loop;
+
+ work->ch[ch].start = start & 0x7ffff;
+ work->ch[ch].loop = loop;
+
+ return 0;
+}
+
+static void queue_channel_work(int ns_to, unsigned int silentch)
+{
+ int tmpFMod[NSSIZE];
+ struct work_item *work;
+ SPUCHAN *s_chan;
+ unsigned int mask;
+ int ch, d;
+
+ work = &worker->i[worker->i_ready & WORK_I_MASK];
+ work->ns_to = ns_to;
+ work->ctrl = spu.spuCtrl;
+ work->decode_pos = spu.decode_pos;
+ work->channels_silent = silentch;
+
+ if (!worker->prev_work_in_thread) {
+ // copy adpcm and interpolation state to sb_thread
+ worker->prev_work_in_thread = 1;
+ mask = spu.dwChannelsAudible & ~spu.dwNewChannel & 0xffffff;
+ for (ch = 0; mask != 0; ch++, mask >>= 1) {
+ if (mask & 1)
+ memcpy(spu.sb_thread[ch].SB, spu.sb[ch].SB, sizeof(spu.sb_thread[ch].SB));
+ }
+ }
+
+ mask = work->channels_new = spu.dwNewChannel & 0xffffff;
+ for (ch = 0; mask != 0; ch++, mask >>= 1) {
+ if (mask & 1)
+ StartSound(ch);
+ }
+
+ mask = work->channels_on = spu.dwChannelsAudible & 0xffffff;
+ worker->channels_last = mask;
+ spu.decode_dirty_ch |= mask & 0x0a;
+
+ for (ch = 0; mask != 0; ch++, mask >>= 1)
+ {
+ if (!(mask & 1)) continue;
+
+ s_chan = &spu.s_chan[ch];
+ work->ch[ch].spos = s_chan->spos;
+ 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;
+ work->ch[ch].bNoise = s_chan->bNoise;
+ work->ch[ch].bFMod = s_chan->bFMod;
+ work->ch[ch].bRVBActive = s_chan->bRVBActive;
+ work->ch[ch].bStarting = s_chan->bStarting;
+ if (s_chan->prevflags & 1)
+ work->ch[ch].start = work->ch[ch].loop;
+
+ if (unlikely(s_chan->bFMod == 2))
+ {
+ // sucks, have to do double work
+ if (s_chan->bNoise)
+ d = do_samples_noise(tmpFMod, ch, ns_to);
+ else
+ d = do_samples_gauss(tmpFMod, decode_block, NULL, ch, ns_to,
+ &spu.sb[ch], s_chan->sinc, &s_chan->spos, &s_chan->iSBPos);
+ if (!s_chan->bStarting) {
+ d = MixADSR(tmpFMod, &s_chan->ADSRX, d);
+ if (d < ns_to) {
+ spu.dwChannelsAudible &= ~(1 << ch);
+ s_chan->ADSRX.State = ADSR_RELEASE;
+ s_chan->ADSRX.EnvelopeVol = 0;
+ }
+ }
+ memset(&tmpFMod[d], 0, (ns_to - d) * sizeof(tmpFMod[d]));
+ work->ch[ch].ns_to = d;
+ continue;
+ }
+ if (unlikely(s_chan->bFMod))
+ d = do_samples_skip_fmod(ch, ns_to, tmpFMod);
+ else
+ d = do_samples_skip(ch, ns_to);
+ work->ch[ch].ns_to = d;
+
+ if (!s_chan->bStarting) {
+ // note: d is not accurate on skip
+ d = SkipADSR(&s_chan->ADSRX, d);
+ if (d < ns_to) {
+ spu.dwChannelsAudible &= ~(1 << ch);
+ s_chan->ADSRX.State = ADSR_RELEASE;
+ s_chan->ADSRX.EnvelopeVol = 0;
+ }
+ }
+ } // for (ch;;)
+
+ work->rvb_addr = 0;
+ if (spu.rvb->StartAddr) {
+ if (spu_config.iUseReverb)
+ 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();
+}
+
+static void do_channel_work(struct work_item *work)
+{
+ unsigned int mask;
+ int spos, sbpos;
+ int d, ch, ns_to;
+
+ ns_to = work->ns_to;
+
+ if (unlikely(spu.interpolation != spu_config.iUseInterpolation))
+ {
+ spu.interpolation = spu_config.iUseInterpolation;
+ mask = work->channels_on;
+ for (ch = 0; mask != 0; ch++, mask >>= 1)
+ if (mask & 1)
+ ResetInterpolation(&spu.sb_thread[ch]);
+ }
+
+ 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) {
+ if (mask & 1)
+ StartSoundSB(&spu.sb_thread[ch]);
+ }
+
+ mask = work->channels_on;
+ for (ch = 0; mask != 0; ch++, mask >>= 1)
+ {
+ if (!(mask & 1)) continue;
+
+ d = work->ch[ch].ns_to;
+ spos = work->ch[ch].spos;
+ sbpos = work->ch[ch].sbpos;
+
+ if (work->ch[ch].bNoise)
+ do_lsfr_samples(ChanBuf, d, work->ctrl, &spu.dwNoiseCount, &spu.dwNoiseVal);
+ else
+ do_samples_adpcm(ChanBuf, decode_block_work, work, ch, d, work->ch[ch].bFMod,
+ &spu.sb_thread[ch], work->ch[ch].sinc, &spos, &sbpos);
+
+ d = MixADSR(ChanBuf, &work->ch[ch].adsr, d);
+ if (d < ns_to) {
+ work->ch[ch].adsr.EnvelopeVol = 0;
+ memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0]));
+ }
+
+ if (ch == 1 || ch == 3)
+ do_decode_bufs(spu.spuMem, ch/2, ns_to, work->decode_pos);
+
+ if (work->ch[ch].bFMod == 2) // fmod freq channel
+ memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
+ if (work->ch[ch].bRVBActive && work->rvb_addr)
+ mix_chan_rvb(work->SSumLR, ns_to,
+ work->ch[ch].vol_l, work->ch[ch].vol_r, RVB);
+ else
+ 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_no_thread)
+{
+ int force = force_no_thread;
+ struct work_item *work;
+ int done, used_space;
+
+ // rvb offsets will change, thread may be using them
+ force |= spu.rvb->dirty && spu.rvb->StartAddr;
+
+ done = thread_get_i_done() - worker->i_reaped;
+ used_space = worker->i_ready - worker->i_reaped;
+
+ //printf("done: %d use: %d dsp: %u/%u\n", done, used_space,
+ // worker->boot_cnt, worker->i_done);
+
+ while ((force && used_space > 0) || used_space >= WORK_MAXCNT || done > 0) {
+ work = &worker->i[worker->i_reaped & WORK_I_MASK];
+ thread_work_wait_sync(work, force);
+
+ MixCD(work->SSumLR, RVB, work->ns_to, work->decode_pos);
+ 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_no_thread && worker->prev_work_in_thread) {
+ unsigned int ch, mask = worker->channels_last;
+ worker->prev_work_in_thread = 0;
+ thread_sync_caches();
+ for (ch = 0; mask != 0; ch++, mask >>= 1) {
+ if (mask & 1)
+ memcpy(spu.sb[ch].SB, spu.sb_thread[ch].SB, sizeof(spu.sb_thread[ch].SB));
+ }
+ }
+}
+
+#else
+
+static void queue_channel_work(int ns_to, int silentch) {}
+static void sync_worker_thread(int force_no_thread) {}
+
+static const void * const worker = NULL;
+
+#endif // P_HAVE_PTHREAD || defined(WANT_THREAD_CODE)
+
////////////////////////////////////////////////////////////////////////
// MAIN SPU FUNCTION
// here is the main job handler...
-// basically the whole sound processing is done in this fat func!
////////////////////////////////////////////////////////////////////////
-void do_samples(unsigned int cycles_to)
+void do_samples(unsigned int cycles_to, int force_no_thread)
{
- const int ns_from = 0;
- int ns,ns_to,ns_len;
- int volmult = spu_config.iVolume;
- int ch,d,silentch;
+ unsigned int silentch;
int cycle_diff;
+ int ns_to;
cycle_diff = cycles_to - spu.cycles_played;
if (cycle_diff < -2*1048576 || cycle_diff > 2*1048576)
{
- //xprintf("desync %u %d\n", cycles_to, cycle_diff);
+ log_unhandled("desync %u %d\n", cycles_to, cycle_diff);
spu.cycles_played = cycles_to;
return;
}
+ silentch = ~(spu.dwChannelsAudible | spu.dwNewChannel) & 0xffffff;
+
+ force_no_thread |= (silentch == 0xffffff);
+ if (worker != NULL)
+ sync_worker_thread(force_no_thread);
+
if (cycle_diff < 2 * 768)
return;
ns_to = (cycle_diff / 768 + 1) & ~1;
if (ns_to > NSSIZE) {
// should never happen
- //xprintf("ns_to oflow %d %d\n", ns_to, NSSIZE);
+ log_unhandled("ns_to oflow %d %d\n", ns_to, NSSIZE);
ns_to = NSSIZE;
}
if (0 < left && left <= ns_to)
{
//xprintf("decoder irq %x\n", spu.decode_pos);
- do_irq();
+ do_irq(0);
}
}
+ if (!spu.cycles_dma_end || (int)(spu.cycles_dma_end - cycles_to) < 0) {
+ spu.cycles_dma_end = 0;
+ check_irq_io(spu.spuAddr);
+ }
- InitREVERB(ns_to);
+ if (unlikely(spu.rvb->dirty))
+ REVERBPrep();
- {
- silentch=~(spu.dwChannelOn|spu.dwNewChannel);
+ if (force_no_thread || worker == NULL || !spu_config.iUseThread) {
+ do_channels(ns_to);
+ do_samples_finish(spu.SSumLR, ns_to, silentch, spu.decode_pos);
+ }
+ else {
+ queue_channel_work(ns_to, silentch);
+ //sync_worker_thread(1); // uncomment for debug
+ }
- //--------------------------------------------------//
- //- main channel loop -//
- //--------------------------------------------------//
- {
- for(ch=0;ch<MAXCHAN;ch++) // loop em all...
- {
- if(spu.dwNewChannel&(1<<ch)) StartSound(ch); // start new sound
- if(!(spu.dwChannelOn&(1<<ch))) continue; // channel not playing? next
-
- if(s_chan[ch].bNoise)
- do_samples_noise(ch, ns_from, ns_to);
- else if(s_chan[ch].bFMod==2 || (s_chan[ch].bFMod==0 && spu_config.iUseInterpolation==0))
- do_samples_noint(ch, ns_from, ns_to);
- else if(s_chan[ch].bFMod==0 && spu_config.iUseInterpolation==1)
- do_samples_simple(ch, ns_from, ns_to);
- else
- do_samples_default(ch, ns_from, ns_to);
-
- ns_len = ns_to - ns_from;
-
- MixADSR(ch, ns_from, ns_to);
-
- if(ch==1 || ch==3)
- {
- do_decode_bufs(ch/2, ns_from, ns_len);
- spu.decode_dirty_ch |= 1<<ch;
- }
-
- if(s_chan[ch].bFMod==2) // fmod freq channel
- memcpy(&iFMod[ns_from], &ChanBuf[ns_from], ns_len * sizeof(iFMod[0]));
- if(s_chan[ch].bRVBActive)
- mix_chan_rvb(ns_from, ns_len, s_chan[ch].iLeftVolume, s_chan[ch].iRightVolume, spu.sRVBStart);
- else
- mix_chan(ns_from, ns_len, s_chan[ch].iLeftVolume, s_chan[ch].iRightVolume);
- }
- }
+ // advance "stopped" channels that can cause irqs
+ // (all chans are always playing on the real thing..)
+ if (spu.spuCtrl & CTRL_IRQ)
+ do_silent_chans(ns_to, silentch);
- // advance "stopped" channels that can cause irqs
- // (all chans are always playing on the real thing..)
- if(spu.spuCtrl&CTRL_IRQ)
- for(ch=0;ch<MAXCHAN;ch++)
- {
- if(!(silentch&(1<<ch))) continue; // already handled
- if(spu.dwChannelDead&(1<<ch)) continue;
- if(s_chan[ch].pCurr > spu.pSpuIrq && s_chan[ch].pLoop > spu.pSpuIrq)
- continue;
-
- s_chan[ch].spos += s_chan[ch].sinc * (ns_to - ns_from);
- while(s_chan[ch].spos >= 28 * 0x10000)
- {
- unsigned char *start = s_chan[ch].pCurr;
-
- skip_block(ch);
- if(start == s_chan[ch].pCurr || start - spu.spuMemC < 0x1000)
- {
- // looping on self or stopped(?)
- spu.dwChannelDead |= 1<<ch;
- s_chan[ch].spos = 0;
- break;
- }
-
- s_chan[ch].spos -= 28 * 0x10000;
- }
- }
+ spu.cycles_played += ns_to * 768;
+ spu.decode_pos = (spu.decode_pos + ns_to) & 0x1ff;
+#if 0
+ static int ccount; static time_t ctime; ccount++;
+ if (time(NULL) != ctime)
+ { printf("%d\n", ccount); ccount = 0; ctime = time(NULL); }
+#endif
+}
- if(unlikely(silentch & spu.decode_dirty_ch & (1<<1))) // must clear silent channel decode buffers
+static void do_samples_finish(int *SSumLR, int ns_to,
+ int silentch, int decode_pos)
+{
+ int vol_l = ((int)regAreaGet(H_SPUmvolL) << 17) >> 17;
+ int vol_r = ((int)regAreaGet(H_SPUmvolR) << 17) >> 17;
+ int ns;
+ int d;
+
+ // must clear silent channel decode buffers
+ if(unlikely(silentch & spu.decode_dirty_ch & (1<<1)))
{
memset(&spu.spuMem[0x800/2], 0, 0x400);
spu.decode_dirty_ch &= ~(1<<1);
spu.decode_dirty_ch &= ~(1<<3);
}
- //---------------------------------------------------//
- // mix XA infos (if any)
-
- MixXA(ns_to);
-
- ///////////////////////////////////////////////////////
- // mix all channels (including reverb) into one buffer
-
- if(spu_config.iUseReverb)
- REVERBDo(ns_to);
+ vol_l = vol_l * spu_config.iVolume >> 10;
+ vol_r = vol_r * spu_config.iVolume >> 10;
- if((spu.spuCtrl&0x4000)==0) // muted? (rare, don't optimize for this)
+ if (!(vol_l | vol_r))
{
+ // muted? (rare)
memset(spu.pS, 0, ns_to * 2 * sizeof(spu.pS[0]));
+ memset(SSumLR, 0, ns_to * 2 * sizeof(SSumLR[0]));
spu.pS += ns_to * 2;
}
else
for (ns = 0; ns < ns_to * 2; )
{
d = SSumLR[ns]; SSumLR[ns] = 0;
- d = d * volmult >> 10;
+ d = d * vol_l >> 14;
ssat32_to_16(d);
*spu.pS++ = d;
ns++;
d = SSumLR[ns]; SSumLR[ns] = 0;
- d = d * volmult >> 10;
+ d = d * vol_r >> 14;
ssat32_to_16(d);
*spu.pS++ = d;
ns++;
}
-
- spu.cycles_played += ns_to * 768;
-
- spu.decode_pos = (spu.decode_pos + ns_to) & 0x1ff;
- }
}
void schedule_next_irq(void)
{
if (spu.dwChannelDead & (1 << ch))
continue;
- if ((unsigned long)(spu.pSpuIrq - s_chan[ch].pCurr) > IRQ_NEAR_BLOCKS * 16
- && (unsigned long)(spu.pSpuIrq - s_chan[ch].pLoop) > IRQ_NEAR_BLOCKS * 16)
+ if ((unsigned long)(spu.pSpuIrq - spu.s_chan[ch].pCurr) > IRQ_NEAR_BLOCKS * 16
+ && (unsigned long)(spu.pSpuIrq - spu.s_chan[ch].pLoop) > IRQ_NEAR_BLOCKS * 16)
+ continue;
+ if (spu.s_chan[ch].sinc == 0)
continue;
scan_for_irq(ch, &upd_samples);
void CALLBACK SPUasync(unsigned int cycle, unsigned int flags)
{
- do_samples(cycle);
+ do_samples(cycle, 0);
if (spu.spuCtrl & CTRL_IRQ)
schedule_next_irq();
// XA AUDIO
-void CALLBACK SPUplayADPCMchannel(xa_decode_t *xap)
+void CALLBACK SPUplayADPCMchannel(xa_decode_t *xap, unsigned int cycle, int is_start)
{
if(!xap) return;
- if(!xap->freq) return; // no xa freq ? bye
+ if(!xap->freq) return; // no xa freq ? bye
+
+ if (is_start)
+ spu.XAPlay = spu.XAFeed = spu.XAStart;
+ if (spu.XAPlay == spu.XAFeed)
+ do_samples(cycle, 1); // catch up to prevent source underflows later
- FeedXA(xap); // call main XA feeder
+ FeedXA(xap); // call main XA feeder
+ spu.xapGlobal = xap; // store info for save states
}
// CDDA AUDIO
-int CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes)
+int CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes, unsigned int cycle, int unused)
{
if (!pcm) return -1;
if (nbytes<=0) return -1;
- return FeedCDDA((unsigned char *)pcm, nbytes);
+ if (spu.CDDAPlay == spu.CDDAFeed)
+ do_samples(cycle, 1); // catch up to prevent source underflows later
+
+ FeedCDDA((unsigned char *)pcm, nbytes);
+ return 0;
+}
+
+void CALLBACK SPUsetCDvol(unsigned char ll, unsigned char lr,
+ unsigned char rl, unsigned char rr, unsigned int cycle)
+{
+ if (spu.XAPlay != spu.XAFeed || spu.CDDAPlay != spu.CDDAFeed)
+ do_samples(cycle, 1);
+ spu.cdv.ll = ll;
+ spu.cdv.lr = lr;
+ spu.cdv.rl = rl;
+ spu.cdv.rr = rr;
}
// to be called after state load
void ClearWorkingState(void)
{
- memset(SSumLR,0,sizeof(SSumLR)); // init some mixing buffers
- memset(iFMod,0,sizeof(iFMod));
+ memset(iFMod, 0, sizeof(iFMod));
spu.pS=(short *)spu.pSpuBuffer; // setup soundbuffer pointer
}
// 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.sRVBStart = (int *)malloc(NSSIZE*2*4); // alloc reverb buffer
- memset(spu.sRVBStart,0,NSSIZE*2*4);
+ spu.SSumLR = calloc(NSSIZE * 2, sizeof(spu.SSumLR[0]));
- spu.XAStart = // alloc xa buffer
- (uint32_t *)malloc(44100 * sizeof(uint32_t));
+ spu.XAStart = malloc(44100 * sizeof(uint32_t)); // alloc xa buffer
spu.XAEnd = spu.XAStart + 44100;
spu.XAPlay = spu.XAStart;
spu.XAFeed = spu.XAStart;
- spu.CDDAStart = // alloc cdda buffer
- (uint32_t *)malloc(CDDA_BUFFER_SIZE);
- spu.CDDAEnd = spu.CDDAStart + 16384;
+ spu.CDDAStart = malloc(CDDA_BUFFER_SIZE); // alloc cdda buffer
+ spu.CDDAEnd = spu.CDDAStart + CDDA_BUFFER_SIZE / sizeof(uint32_t);
spu.CDDAPlay = spu.CDDAStart;
spu.CDDAFeed = spu.CDDAStart;
- for(i=0;i<MAXCHAN;i++) // loop sound channels
- {
- s_chan[i].ADSRX.SustainLevel = 0xf; // -> init sustain
- s_chan[i].ADSRX.SustainIncrease = 1;
- s_chan[i].pLoop=spu.spuMemC;
- 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.sRVBStart); // free reverb buffer
- spu.sRVBStart = NULL;
+ free(spu.SSumLR);
+ spu.SSumLR = NULL;
free(spu.XAStart); // free XA buffer
spu.XAStart = NULL;
free(spu.CDDAStart); // free CDDA buffer
spu.CDDAStart = NULL;
}
-// INIT/EXIT STUFF
+#if defined(C64X_DSP)
+
+/* special code for TI C64x DSP */
+#include "spu_c64x.c"
+
+#elif P_HAVE_PTHREAD
+
+#include <pthread.h>
+#include <semaphore.h>
+#include <unistd.h>
+
+static struct {
+ pthread_t thread;
+ sem_t sem_avail;
+ sem_t sem_done;
+} t;
+
+/* generic pthread implementation */
+
+static void thread_work_start(void)
+{
+ sem_post(&t.sem_avail);
+}
+
+static void thread_work_wait_sync(struct work_item *work, int force)
+{
+ sem_wait(&t.sem_done);
+}
+
+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;
+
+ while (1) {
+ sem_wait(&t.sem_avail);
+ if (worker->exit_thread)
+ break;
+
+ work = &worker->i[worker->i_done & WORK_I_MASK];
+ do_channel_work(work);
+ worker->i_done++;
+
+ sem_post(&t.sem_done);
+ }
+
+ return NULL;
+}
+
+static void init_spu_thread(void)
+{
+ int ret;
+
+ spu.sb_thread = spu.sb_thread_;
+
+ if (sysconf(_SC_NPROCESSORS_ONLN) <= 1)
+ return;
+
+ worker = calloc(1, sizeof(*worker));
+ if (worker == NULL)
+ return;
+ ret = sem_init(&t.sem_avail, 0, 0);
+ if (ret != 0)
+ goto fail_sem_avail;
+ ret = sem_init(&t.sem_done, 0, 0);
+ if (ret != 0)
+ goto fail_sem_done;
+
+ ret = pthread_create(&t.thread, NULL, spu_worker_thread, NULL);
+ if (ret != 0)
+ goto fail_thread;
+
+ spu_config.iThreadAvail = 1;
+ return;
+
+fail_thread:
+ sem_destroy(&t.sem_done);
+fail_sem_done:
+ sem_destroy(&t.sem_avail);
+fail_sem_avail:
+ free(worker);
+ worker = NULL;
+ spu_config.iThreadAvail = 0;
+}
+
+static void exit_spu_thread(void)
+{
+ if (worker == NULL)
+ return;
+ worker->exit_thread = 1;
+ sem_post(&t.sem_avail);
+ pthread_join(t.thread, NULL);
+ sem_destroy(&t.sem_done);
+ sem_destroy(&t.sem_avail);
+ free(worker);
+ worker = NULL;
+}
+
+#else // if !P_HAVE_PTHREAD
+
+static void init_spu_thread(void)
+{
+}
+
+static void exit_spu_thread(void)
+{
+}
+
+#endif
// SPUINIT: this func will be called first by the main emu
long CALLBACK SPUinit(void)
{
- spu.spuMemC = (unsigned char *)spu.spuMem; // just small setup
- memset((void *)&rvb, 0, sizeof(REVERBInfo));
+ int i;
+
+ memset(&spu, 0, sizeof(spu));
+ spu.spuMemC = calloc(1, 512 * 1024 + 16);
+ // a guard for runaway channels - End+Mute
+ spu.spuMemC[512 * 1024 + 1] = 1;
+
InitADSR();
- spu.spuAddr = 0xffffffff;
+ 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.spuAddr = 0;
spu.decode_pos = 0;
- memset((void *)s_chan, 0, sizeof(s_chan));
spu.pSpuIrq = spu.spuMemC;
SetupStreams(); // prepare streaming
if (spu_config.iVolume == 0)
spu_config.iVolume = 768; // 1024 is 1.0
+ 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.s_chan[i].bIgnoreLoop = 0;
+ }
+
+ spu.bSpuInit=1; // flag: we are inited
+
return 0;
}
long CALLBACK SPUshutdown(void)
{
SPUclose();
- RemoveStreams(); // no more streaming
- spu.bSpuInit=0;
- return 0;
-}
+ exit_spu_thread();
-// SPUTEST: we don't test, we are always fine ;)
-long CALLBACK SPUtest(void)
-{
- return 0;
-}
+ free(spu.spuMemC);
+ spu.spuMemC = NULL;
+ free(spu.s_chan);
+ spu.s_chan = NULL;
+ free(spu.rvb);
+ spu.rvb = NULL;
-// SPUCONFIGURE: call config dialog
-long CALLBACK SPUconfigure(void)
-{
-#ifdef _MACOSX
- DoConfiguration();
-#else
-// StartCfgTool("CFG");
-#endif
- return 0;
-}
+ RemoveStreams(); // no more streaming
+ spu.bSpuInit=0;
-// SPUABOUT: show about window
-void CALLBACK SPUabout(void)
-{
-#ifdef _MACOSX
- DoAbout();
-#else
-// StartCfgTool("ABOUT");
-#endif
+ return 0;
}
// SETUP CALLBACKS
// this functions will be called once,
// passes a callback that should be called on SPU-IRQ/cdda volume change
-void CALLBACK SPUregisterCallback(void (CALLBACK *callback)(void))
+void CALLBACK SPUregisterCallback(void (CALLBACK *callback)(int))
{
spu.irqCallback = callback;
}
-void CALLBACK SPUregisterCDDAVolume(void (CALLBACK *CDDAVcallback)(unsigned short,unsigned short))
+void CALLBACK SPUregisterCDDAVolume(void (CALLBACK *CDDAVcallback)(short, short))
{
- spu.cddavCallback = CDDAVcallback;
+ //spu.cddavCallback = CDDAVcallback;
}
void CALLBACK SPUregisterScheduleCb(void (CALLBACK *callback)(unsigned int))
{
int ch = 0, fmod_chans = 0, noise_chans = 0, irq_chans = 0;
+ if (spu.s_chan == NULL)
+ return;
+
for(;ch<MAXCHAN;ch++)
{
- if (!(spu.dwChannelOn & (1<<ch)))
+ if (!(spu.dwChannelsAudible & (1<<ch)))
continue;
- if (s_chan[ch].bFMod == 2)
+ if (spu.s_chan[ch].bFMod == 2)
fmod_chans |= 1 << ch;
- if (s_chan[ch].bNoise)
+ if (spu.s_chan[ch].bNoise)
noise_chans |= 1 << ch;
- if((spu.spuCtrl&CTRL_IRQ) && s_chan[ch].pCurr <= spu.pSpuIrq && s_chan[ch].pLoop <= spu.pSpuIrq)
+ if((spu.spuCtrl&CTRL_IRQ) && spu.s_chan[ch].pCurr <= spu.pSpuIrq && spu.s_chan[ch].pLoop <= spu.pSpuIrq)
irq_chans |= 1 << ch;
}
- *chans_out = spu.dwChannelOn;
- *run_chans = ~spu.dwChannelOn & ~spu.dwChannelDead & irq_chans;
+ *chans_out = spu.dwChannelsAudible;
+ *run_chans = ~spu.dwChannelsAudible & ~spu.dwChannelDead & irq_chans;
*fmod_chans_out = fmod_chans;
*noise_chans_out = noise_chans;
}