copyright : (C) 2002 by Pete Bernert
email : BlackDove@addcom.de
- Portions (C) Gražvydas "notaz" Ignotas, 2010-2012,2014
+ Portions (C) Gražvydas "notaz" Ignotas, 2010-2012,2014,2015
***************************************************************************/
/***************************************************************************
* *
***************************************************************************/
-#ifndef _WIN32
+#if !defined(_WIN32) && !defined(NO_OS)
#include <sys/time.h> // gettimeofday in xa.c
+#define THREAD_ENABLED 1
#endif
#include "stdafx.h"
#include "externals.h"
#include "registers.h"
#include "out.h"
-#include "arm_features.h"
#include "spu_config.h"
-#ifdef __ARM_ARCH_7A__
+#ifdef __arm__
+#include "arm_features.h"
+#endif
+
+#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
// START SOUND... called by main thread to setup a new sound on a channel
////////////////////////////////////////////////////////////////////////
-INLINE void StartSound(int ch)
+static void StartSoundSB(int *SB)
+{
+ SB[26]=0; // init mixing vars
+ SB[27]=0;
+
+ SB[28]=0;
+ SB[29]=0; // init our interpolation helpers
+ SB[30]=0;
+ SB[31]=0;
+}
+
+static void StartSoundMain(int ch)
{
+ SPUCHAN *s_chan = &spu.s_chan[ch];
+
StartADSR(ch);
StartREVERB(ch);
- s_chan[ch].bStop=0;
- s_chan[ch].prevflags=2;
+ s_chan->prevflags=2;
+ s_chan->iSBPos=27;
+ s_chan->spos=0;
- s_chan[ch].SB[26]=0; // init mixing vars
- s_chan[ch].SB[27]=0;
- s_chan[ch].iSBPos=27;
-
- s_chan[ch].SB[28]=0;
- s_chan[ch].SB[29]=0; // init our interpolation helpers
- s_chan[ch].SB[30]=0;
- s_chan[ch].SB[31]=0;
- s_chan[ch].spos=0;
+ s_chan->pCurr = spu.spuMemC+((regAreaGet(ch,6)&~1)<<3);
spu.dwNewChannel&=~(1<<ch); // clear new channel bit
spu.dwChannelOn|=1<<ch;
spu.dwChannelDead&=~(1<<ch);
}
+static void StartSound(int ch)
+{
+ StartSoundMain(ch);
+ StartSoundSB(spu.SB + ch * SB_SIZE);
+}
+
////////////////////////////////////////////////////////////////////////
// ALL KIND OF HELPERS
////////////////////////////////////////////////////////////////////////
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;
}
}
-static int decode_block(int ch, int *SB)
+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 ret = 0;
- start = s_chan[ch].pCurr; // set up the current pos
+ start = s_chan->pCurr; // set up the current pos
if (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))
+ if (!(s_chan->prevflags & 2))
ret = 1;
- start = s_chan[ch].pLoop;
+ start = s_chan->pLoop;
}
else
check_irq(ch, start); // hack, see check_irq below..
- predict_nr = (int)start[0];
+ predict_nr = start[0];
shift_factor = predict_nr & 0xf;
predict_nr >>= 4;
decode_block_data(SB, start + 2, predict_nr, shift_factor);
flags = start[1];
- if (flags & 4)
- s_chan[ch].pLoop = start; // loop adress
+ if (flags & 4 && (!s_chan->bIgnoreLoop))
+ s_chan->pLoop = start; // loop adress
start += 16;
if (flags & 1) { // 1: stop/loop
- start = s_chan[ch].pLoop;
+ start = s_chan->pLoop;
check_irq(ch, start); // hack.. :(
}
if (start - spu.spuMemC >= 0x80000)
start = spu.spuMemC;
- 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;
return ret;
}
// do block, but ignore sample data
static int skip_block(int ch)
{
- unsigned char *start = s_chan[ch].pCurr;
+ SPUCHAN *s_chan = &spu.s_chan[ch];
+ unsigned char *start = s_chan->pCurr;
int flags;
int ret = 0;
- if (s_chan[ch].prevflags & 1) {
- if (!(s_chan[ch].prevflags & 2))
+ if (s_chan->prevflags & 1) {
+ if (!(s_chan->prevflags & 2))
ret = 1;
- start = s_chan[ch].pLoop;
+ start = s_chan->pLoop;
}
else
check_irq(ch, start);
flags = start[1];
if (flags & 4)
- s_chan[ch].pLoop = start;
+ s_chan->pLoop = start;
start += 16;
if (flags & 1) {
- start = s_chan[ch].pLoop;
+ start = s_chan->pLoop;
check_irq(ch, start);
}
- s_chan[ch].pCurr = start;
- s_chan[ch].prevflags = flags;
+ s_chan->pCurr = start;
+ s_chan->prevflags = flags;
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;
- 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;
+ block = s_chan->pLoop;
if (block == spu.pSpuIrq) // hack.. (see decode_block)
break;
}
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_to) \
+static noinline int do_samples_##name( \
+ int (*decode_f)(void *context, int ch, int *SB), void *ctx, \
+ int ch, int ns_to, int *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 ns, d, fa; \
int ret = ns_to; \
interp_start; \
{ \
fmod_code; \
\
- spos += sinc; \
- while (spos >= 0x10000) \
+ *spos += sinc; \
+ while (*spos >= 0x10000) \
{ \
- fa = SB[sbpos++]; \
- if (sbpos >= 28) \
+ fa = SB[(*sbpos)++]; \
+ if (*sbpos >= 28) \
{ \
- sbpos = 0; \
- d = decode_block(ch, SB); \
+ *sbpos = 0; \
+ d = decode_f(ctx, ch, SB); \
if (d && ns < ret) \
ret = ns; \
} \
\
interp1_code; \
- spos -= 0x10000; \
+ *spos -= 0x10000; \
} \
\
interp2_code; \
} \
\
- 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(SB, s_chan[ch].iRawPitch, ns)
+ if(spu.s_chan[ch].bFMod==1 && iFMod[ns]) \
+ sinc = FModChangeFrequency(SB, spu.s_chan[ch].iRawPitch, ns)
make_do_samples(default, fmod_recv_check, ,
- StoreInterpolationVal(SB, sinc, fa, s_chan[ch].bFMod==2),
- ChanBuf[ns] = iGetInterpolationVal(SB, sinc, spos, s_chan[ch].bFMod==2), )
+ StoreInterpolationVal(SB, sinc, fa, spu.s_chan[ch].bFMod==2),
+ ChanBuf[ns] = iGetInterpolationVal(SB, sinc, *spos, spu.s_chan[ch].bFMod==2), )
make_do_samples(noint, , fa = SB[29], , ChanBuf[ns] = fa, SB[29] = fa)
#define simple_interp_store \
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;
- s_chan[ch].spos += s_chan[ch].iSBPos << 16;
+ spos += s_chan->iSBPos << 16;
for (ns = 0; ns < ns_to; ns++)
{
- s_chan[ch].spos += s_chan[ch].sinc;
- while (s_chan[ch].spos >= 28*0x10000)
+ spos += sinc;
+ while (spos >= 28*0x10000)
{
d = skip_block(ch);
if (d && ns < ret)
ret = ns;
- s_chan[ch].spos -= 28*0x10000;
+ spos -= 28*0x10000;
}
}
- s_chan[ch].iSBPos = s_chan[ch].spos >> 16;
- s_chan[ch].spos &= 0xffff;
+ s_chan->iSBPos = spos >> 16;
+ s_chan->spos = spos & 0xffff;
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--)
static void do_silent_chans(int ns_to, int silentch)
{
+ unsigned int mask;
+ SPUCHAN *s_chan;
int ch;
- for (ch = 0; ch < MAXCHAN; ch++)
+ mask = silentch & 0xffffff;
+ for (ch = 0; mask != 0; ch++, mask >>= 1)
{
- if (!(silentch & (1<<ch))) continue; // already handled
+ if (!(mask & 1)) continue;
if (spu.dwChannelDead & (1<<ch)) continue;
- if (s_chan[ch].pCurr > spu.pSpuIrq && s_chan[ch].pLoop > spu.pSpuIrq)
+
+ s_chan = &spu.s_chan[ch];
+ if (s_chan->pCurr > spu.pSpuIrq && s_chan->pLoop > spu.pSpuIrq)
continue;
- s_chan[ch].spos += s_chan[ch].iSBPos << 16;
- s_chan[ch].iSBPos = 0;
+ s_chan->spos += s_chan->iSBPos << 16;
+ s_chan->iSBPos = 0;
- s_chan[ch].spos += s_chan[ch].sinc * ns_to;
- while (s_chan[ch].spos >= 28 * 0x10000)
+ s_chan->spos += s_chan->sinc * ns_to;
+ while (s_chan->spos >= 28 * 0x10000)
{
- unsigned char *start = s_chan[ch].pCurr;
+ unsigned char *start = s_chan->pCurr;
skip_block(ch);
- if (start == s_chan[ch].pCurr || start - spu.spuMemC < 0x1000)
+ if (start == s_chan->pCurr || start - spu.spuMemC < 0x1000)
{
// looping on self or stopped(?)
spu.dwChannelDead |= 1<<ch;
- s_chan[ch].spos = 0;
+ s_chan->spos = 0;
break;
}
- s_chan[ch].spos -= 28 * 0x10000;
+ s_chan->spos -= 28 * 0x10000;
}
}
}
static void do_channels(int ns_to)
{
unsigned int mask;
- int ch, d;
-
- InitREVERB(ns_to);
+ int do_rvb, ch, d;
+ SPUCHAN *s_chan;
+ int *SB, sinc;
+
+ 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.dwChannelOn & 0xffffff;
for (ch = 0; mask != 0; ch++, mask >>= 1) // loop em all...
{
if (!(mask & 1)) continue; // channel not playing? next
- if (s_chan[ch].bNoise)
+ s_chan = &spu.s_chan[ch];
+ SB = spu.SB + ch * SB_SIZE;
+ sinc = s_chan->sinc;
+
+ if (s_chan->bNoise)
d = do_samples_noise(ch, ns_to);
- else if (s_chan[ch].bFMod == 2
- || (s_chan[ch].bFMod == 0 && spu_config.iUseInterpolation == 0))
- d = do_samples_noint(ch, ns_to);
- else if (s_chan[ch].bFMod == 0 && spu_config.iUseInterpolation == 1)
- d = do_samples_simple(ch, ns_to);
+ else if (s_chan->bFMod == 2
+ || (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 0))
+ d = do_samples_noint(decode_block, NULL, ch, ns_to,
+ SB, sinc, &s_chan->spos, &s_chan->iSBPos);
+ else if (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 1)
+ d = do_samples_simple(decode_block, NULL, ch, ns_to,
+ SB, sinc, &s_chan->spos, &s_chan->iSBPos);
else
- d = do_samples_default(ch, ns_to);
+ d = do_samples_default(decode_block, NULL, ch, ns_to,
+ SB, sinc, &s_chan->spos, &s_chan->iSBPos);
- d = MixADSR(ch, d);
+ d = MixADSR(&s_chan->ADSRX, d);
if (d < ns_to) {
spu.dwChannelOn &= ~(1 << ch);
- s_chan[ch].bStop = 1;
- s_chan[ch].ADSRX.EnvelopeVol = 0;
+ s_chan->ADSRX.EnvelopeVol = 0;
memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0]));
}
spu.decode_dirty_ch |= 1 << ch;
}
- if (s_chan[ch].bFMod == 2) // fmod freq channel
+ if (s_chan->bFMod == 2) // fmod freq channel
+ memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
+ 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)
+ 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 defined(THREAD_ENABLED) || defined(WANT_THREAD_CODE)
+
+// worker thread state
+static struct spu_worker {
+ union {
+ struct {
+ unsigned int exit_thread;
+ unsigned int i_ready;
+ unsigned int i_reaped;
+ unsigned int last_boot_cnt; // dsp
+ unsigned int ram_dirty;
+ };
+ // 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;
+ int ns_to;
+ short vol_l;
+ short vol_r;
+ ADSRInfoEx adsr;
+ // might also want to add fmod flags..
+ } 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)
+{
+ 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;
+
+ mask = work->channels_new = spu.dwNewChannel & 0xffffff;
+ for (ch = 0; mask != 0; ch++, mask >>= 1) {
+ if (mask & 1)
+ StartSoundMain(ch);
+ }
+
+ mask = work->channels_on = spu.dwChannelOn & 0xffffff;
+ 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;
+ if (s_chan->prevflags & 1)
+ work->ch[ch].start = work->ch[ch].loop;
+
+ d = do_samples_skip(ch, ns_to);
+ work->ch[ch].ns_to = d;
+
+ // note: d is not accurate on skip
+ d = SkipADSR(&s_chan->ADSRX, d);
+ if (d < ns_to) {
+ spu.dwChannelOn &= ~(1 << ch);
+ s_chan->ADSRX.EnvelopeVol = 0;
+ }
+ }
+
+ 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;
+ unsigned int decode_dirty_ch = 0;
+ const SPUCHAN *s_chan;
+ int *SB, sinc, spos, sbpos;
+ int d, ch, ns_to;
+
+ ns_to = work->ns_to;
+
+ 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 + ch * SB_SIZE);
+ }
+
+ 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;
+ sinc = work->ch[ch].sinc;
+
+ s_chan = &spu.s_chan[ch];
+ SB = spu.SB + ch * SB_SIZE;
+
+ if (s_chan->bNoise)
+ do_lsfr_samples(d, work->ctrl, &spu.dwNoiseCount, &spu.dwNoiseVal);
+ else if (s_chan->bFMod == 2
+ || (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 0))
+ do_samples_noint(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
+ else if (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 1)
+ do_samples_simple(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
+ else
+ do_samples_default(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
+
+ d = MixADSR(&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);
+ decode_dirty_ch |= 1 << ch;
+ }
+
+ if (s_chan->bFMod == 2) // fmod freq channel
memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
- if (s_chan[ch].bRVBActive)
- mix_chan_rvb(0, ns_to, s_chan[ch].iLeftVolume, s_chan[ch].iRightVolume, spu.sRVBStart);
+ if (s_chan->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(0, ns_to, s_chan[ch].iLeftVolume, s_chan[ch].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)
+{
+ 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);
+
+ 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
+
+static void queue_channel_work(int ns_to, int silentch) {}
+static void sync_worker_thread(int force) {}
+
+static const void * const worker = NULL;
+
+#endif // THREAD_ENABLED
+
////////////////////////////////////////////////////////////////////////
// MAIN SPU FUNCTION
// here is the main job handler...
////////////////////////////////////////////////////////////////////////
-void do_samples_finish(int ns_to, int silentch);
-
-void do_samples(unsigned int cycles_to)
+void do_samples(unsigned int cycles_to, int do_direct)
{
- unsigned int mask;
- int ch, ns_to;
- int 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)
return;
}
+ silentch = ~(spu.dwChannelOn | spu.dwNewChannel) & 0xffffff;
+
+ do_direct |= (silentch == 0xffffff);
+ if (worker != NULL)
+ sync_worker_thread(do_direct);
+
if (cycle_diff < 2 * 768)
return;
}
}
- silentch = ~(spu.dwChannelOn|spu.dwNewChannel);
+ if (unlikely(spu.rvb->dirty))
+ REVERBPrep();
- mask = spu.dwNewChannel & 0xffffff;
- for (ch = 0; mask != 0; ch++, mask >>= 1) {
- if (mask & 1)
- StartSound(ch);
+ if (do_direct || worker == NULL || !spu_config.iUseThread) {
+ do_channels(ns_to);
+ do_samples_finish(spu.SSumLR, ns_to, silentch, spu.decode_pos);
}
-
- if (spu.dwChannelOn == 0)
- InitREVERB(ns_to);
else {
- do_channels(ns_to);
+ queue_channel_work(ns_to, silentch);
}
- do_samples_finish(ns_to, silentch);
-
// advance "stopped" channels that can cause irqs
// (all chans are always playing on the real thing..)
if (spu.spuCtrl & CTRL_IRQ)
spu.decode_pos = (spu.decode_pos + ns_to) & 0x1ff;
}
-void do_samples_finish(int ns_to, int silentch)
+static void do_samples_finish(int *SSumLR, int ns_to,
+ int silentch, int decode_pos)
{
int volmult = spu_config.iVolume;
int ns;
int d;
- if(unlikely(silentch & spu.decode_dirty_ch & (1<<1))) // must clear silent channel decode buffers
+ // 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, spu.decode_pos);
+ MixXA(SSumLR, ns_to, decode_pos);
- ///////////////////////////////////////////////////////
- // mix all channels (including reverb) into one buffer
-
- if(spu_config.iUseReverb)
- REVERBDo(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]));
{
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;
scan_for_irq(ch, &upd_samples);
void CALLBACK SPUasync(unsigned int cycle, unsigned int flags)
{
- do_samples(cycle);
+ do_samples(cycle, spu_config.iUseFixedUpdates);
if (spu.spuCtrl & CTRL_IRQ)
schedule_next_irq();
// 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.CDDAStart = malloc(CDDA_BUFFER_SIZE); // alloc cdda buffer
spu.CDDAEnd = spu.CDDAStart + 16384;
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 defined(THREAD_ENABLED)
+
+#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;
+
+ 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 !THREAD_ENABLED
+
+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;
+
+ spu.spuMemC = calloc(1, 512 * 1024);
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.SB = calloc(MAXCHAN, sizeof(spu.SB[0]) * SB_SIZE);
+
+ 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();
+
+ exit_spu_thread();
+
+ free(spu.spuMemC);
+ spu.spuMemC = NULL;
+ free(spu.SB);
+ spu.SB = NULL;
+ free(spu.s_chan);
+ spu.s_chan = NULL;
+ free(spu.rvb);
+ spu.rvb = NULL;
+
RemoveStreams(); // no more streaming
spu.bSpuInit=0;
spu.irqCallback = callback;
}
-void CALLBACK SPUregisterCDDAVolume(void (CALLBACK *CDDAVcallback)(unsigned short,unsigned short))
+void CALLBACK SPUregisterCDDAVolume(void (CALLBACK *CDDAVcallback)(short, short))
{
spu.cddavCallback = CDDAVcallback;
}
{
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)))
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;
}