1 /***************************************************************************
4 begin : Wed May 15 2002
5 copyright : (C) 2002 by Pete Bernert
6 email : BlackDove@addcom.de
8 Portions (C) GraÅžvydas "notaz" Ignotas, 2010-2012,2014,2015
10 ***************************************************************************/
11 /***************************************************************************
13 * This program is free software; you can redistribute it and/or modify *
14 * it under the terms of the GNU General Public License as published by *
15 * the Free Software Foundation; either version 2 of the License, or *
16 * (at your option) any later version. See also the license.txt file for *
17 * additional informations. *
19 ***************************************************************************/
21 #if !defined(_WIN32) && !defined(NO_OS)
22 #include <sys/time.h> // gettimeofday in xa.c
23 #define THREAD_ENABLED 1
29 #include "externals.h"
30 #include "registers.h"
32 #include "spu_config.h"
35 #include "arm_features.h"
39 #define ssat32_to_16(v) \
40 asm("ssat %0,#16,%1" : "=r" (v) : "r" (v))
42 #define ssat32_to_16(v) do { \
43 if (v < -32768) v = -32768; \
44 else if (v > 32767) v = 32767; \
48 #define PSXCLK 33868800 /* 33.8688 MHz */
50 // intended to be ~1 frame
51 #define IRQ_NEAR_BLOCKS 32
54 #if defined (USEMACOSX)
55 static char * libraryName = N_("Mac OS X Sound");
56 #elif defined (USEALSA)
57 static char * libraryName = N_("ALSA Sound");
58 #elif defined (USEOSS)
59 static char * libraryName = N_("OSS Sound");
60 #elif defined (USESDL)
61 static char * libraryName = N_("SDL Sound");
62 #elif defined (USEPULSEAUDIO)
63 static char * libraryName = N_("PulseAudio Sound");
65 static char * libraryName = N_("NULL Sound");
68 static char * libraryInfo = N_("P.E.Op.S. Sound Driver V1.7\nCoded by Pete Bernert and the P.E.Op.S. team\n");
76 static int iFMod[NSSIZE];
77 static int RVB[NSSIZE * 2];
80 #define CDDA_BUFFER_SIZE (16384 * sizeof(uint32_t)) // must be power of 2
82 ////////////////////////////////////////////////////////////////////////
84 ////////////////////////////////////////////////////////////////////////
86 // dirty inline func includes
91 ////////////////////////////////////////////////////////////////////////
92 // helpers for simple interpolation
95 // easy interpolation on upsampling, no special filter, just "Pete's common sense" tm
97 // instead of having n equal sample values in a row like:
101 // we compare the current delta change with the next delta change.
103 // if curr_delta is positive,
105 // - and next delta is smaller (or changing direction):
109 // - and next delta significant (at least twice) bigger:
113 // - and next delta is nearly same:
118 // if curr_delta is negative,
120 // - and next delta is smaller (or changing direction):
124 // - and next delta significant (at least twice) bigger:
128 // - and next delta is nearly same:
133 static void InterpolateUp(int *SB, int sinc)
135 if(SB[32]==1) // flag == 1? calc step and set flag... and don't change the value in this pass
137 const int id1=SB[30]-SB[29]; // curr delta to next val
138 const int id2=SB[31]-SB[30]; // and next delta to next-next val :)
142 if(id1>0) // curr delta positive
145 {SB[28]=id1;SB[32]=2;}
148 SB[28]=(id1*sinc)>>16;
150 SB[28]=(id1*sinc)>>17;
152 else // curr delta negative
155 {SB[28]=id1;SB[32]=2;}
158 SB[28]=(id1*sinc)>>16;
160 SB[28]=(id1*sinc)>>17;
164 if(SB[32]==2) // flag 1: calc step and set flag... and don't change the value in this pass
168 SB[28]=(SB[28]*sinc)>>17;
170 // SB[29]=SB[30]-(SB[28]*((0x10000/sinc)-1));
174 else // no flags? add bigger val (if possible), calc smaller step, set flag1
179 // even easier interpolation on downsampling, also no special filter, again just "Pete's common sense" tm
182 static void InterpolateDown(int *SB, int sinc)
184 if(sinc>=0x20000L) // we would skip at least one val?
186 SB[29]+=(SB[30]-SB[29])/2; // add easy weight
187 if(sinc>=0x30000L) // we would skip even more vals?
188 SB[29]+=(SB[31]-SB[30])/2; // add additional next weight
192 ////////////////////////////////////////////////////////////////////////
193 // helpers for gauss interpolation
195 #define gval0 (((short*)(&SB[29]))[gpos&3])
196 #define gval(x) ((int)((short*)(&SB[29]))[(gpos+x)&3])
200 ////////////////////////////////////////////////////////////////////////
204 static void do_irq(void)
206 //if(!(spu.spuStat & STAT_IRQ))
208 spu.spuStat |= STAT_IRQ; // asserted status?
209 if(spu.irqCallback) spu.irqCallback();
213 static int check_irq(int ch, unsigned char *pos)
215 if((spu.spuCtrl & CTRL_IRQ) && pos == spu.pSpuIrq)
217 //printf("ch%d irq %04x\n", ch, pos - spu.spuMemC);
224 ////////////////////////////////////////////////////////////////////////
225 // START SOUND... called by main thread to setup a new sound on a channel
226 ////////////////////////////////////////////////////////////////////////
228 static void StartSoundSB(int *SB)
230 SB[26]=0; // init mixing vars
234 SB[29]=0; // init our interpolation helpers
239 static void StartSoundMain(int ch)
241 SPUCHAN *s_chan = &spu.s_chan[ch];
250 s_chan->pCurr = spu.spuMemC + ((regAreaGetCh(ch, 6) & ~1) << 3);
252 spu.dwNewChannel&=~(1<<ch); // clear new channel bit
253 spu.dwChannelDead&=~(1<<ch);
254 if (s_chan->iRawPitch)
255 spu.dwChannelsAudible|=1<<ch;
258 static void StartSound(int ch)
261 StartSoundSB(spu.SB + ch * SB_SIZE);
264 ////////////////////////////////////////////////////////////////////////
265 // ALL KIND OF HELPERS
266 ////////////////////////////////////////////////////////////////////////
268 INLINE int FModChangeFrequency(int *SB, int pitch, int ns)
270 unsigned int NP=pitch;
273 NP=((32768L+iFMod[ns])*NP)>>15;
275 if(NP>0x3fff) NP=0x3fff;
278 sinc=NP<<4; // calc frequency
279 if(spu_config.iUseInterpolation==1) // freq change in simple interpolation mode
286 ////////////////////////////////////////////////////////////////////////
288 INLINE void StoreInterpolationVal(int *SB, int sinc, int fa, int fmod_freq)
290 if(fmod_freq) // fmod freq channel
296 if(spu_config.iUseInterpolation>=2) // gauss/cubic interpolation
304 if(spu_config.iUseInterpolation==1) // simple interpolation
307 SB[29] = 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'
310 SB[32] = 1; // -> flag: calc new interolation
312 else SB[29]=fa; // no interpolation
316 ////////////////////////////////////////////////////////////////////////
318 INLINE int iGetInterpolationVal(int *SB, int sinc, int spos, int fmod_freq)
322 if(fmod_freq) return SB[29];
324 switch(spu_config.iUseInterpolation)
326 //--------------------------------------------------//
327 case 3: // cubic interpolation
333 fa = gval(3) - 3*gval(2) + 3*gval(1) - gval0;
334 fa *= (xd - (2<<15)) / 6;
336 fa += gval(2) - gval(1) - gval(1) + gval0;
337 fa *= (xd - (1<<15)) >> 1;
339 fa += gval(1) - gval0;
345 //--------------------------------------------------//
346 case 2: // gauss interpolation
349 vl = (spos >> 6) & ~3;
351 vr=(gauss[vl]*(int)gval0) >> 15;
352 vr+=(gauss[vl+1]*gval(1)) >> 15;
353 vr+=(gauss[vl+2]*gval(2)) >> 15;
354 vr+=(gauss[vl+3]*gval(3)) >> 15;
357 //--------------------------------------------------//
358 case 1: // simple interpolation
360 if(sinc<0x10000L) // -> upsampling?
361 InterpolateUp(SB, sinc); // --> interpolate up
362 else InterpolateDown(SB, sinc); // --> else down
365 //--------------------------------------------------//
366 default: // no interpolation
370 //--------------------------------------------------//
376 static void decode_block_data(int *dest, const unsigned char *src, int predict_nr, int shift_factor)
378 static const int f[16][2] = {
386 int fa, s_1, s_2, d, s;
391 for (nSample = 0; nSample < 28; src++)
394 s = (int)(signed short)((d & 0x0f) << 12);
396 fa = s >> shift_factor;
397 fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6);
400 dest[nSample++] = fa;
402 s = (int)(signed short)((d & 0xf0) << 8);
403 fa = s >> shift_factor;
404 fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6);
407 dest[nSample++] = fa;
411 static int decode_block(void *unused, int ch, int *SB)
413 SPUCHAN *s_chan = &spu.s_chan[ch];
414 unsigned char *start;
415 int predict_nr, shift_factor, flags;
418 start = s_chan->pCurr; // set up the current pos
419 if (start == spu.spuMemC) // ?
422 if (s_chan->prevflags & 1) // 1: stop/loop
424 if (!(s_chan->prevflags & 2))
427 start = s_chan->pLoop;
430 check_irq(ch, start);
432 predict_nr = start[0];
433 shift_factor = predict_nr & 0xf;
436 decode_block_data(SB, start + 2, predict_nr, shift_factor);
439 if (flags & 4 && !s_chan->bIgnoreLoop)
440 s_chan->pLoop = start; // loop adress
444 s_chan->pCurr = start; // store values for next cycle
445 s_chan->prevflags = flags;
450 // do block, but ignore sample data
451 static int skip_block(int ch)
453 SPUCHAN *s_chan = &spu.s_chan[ch];
454 unsigned char *start = s_chan->pCurr;
458 if (s_chan->prevflags & 1) {
459 if (!(s_chan->prevflags & 2))
462 start = s_chan->pLoop;
465 check_irq(ch, start);
468 if (flags & 4 && !s_chan->bIgnoreLoop)
469 s_chan->pLoop = start;
473 s_chan->pCurr = start;
474 s_chan->prevflags = flags;
479 // if irq is going to trigger sooner than in upd_samples, set upd_samples
480 static void scan_for_irq(int ch, unsigned int *upd_samples)
482 SPUCHAN *s_chan = &spu.s_chan[ch];
483 int pos, sinc, sinc_inv, end;
484 unsigned char *block;
487 block = s_chan->pCurr;
490 end = pos + *upd_samples * sinc;
492 pos += (28 - s_chan->iSBPos) << 16;
495 if (block == spu.pSpuIrq)
499 if (flags & 1) { // 1: stop/loop
500 block = s_chan->pLoop;
507 sinc_inv = s_chan->sinc_inv;
509 sinc_inv = s_chan->sinc_inv = (0x80000000u / (uint32_t)sinc) << 1;
512 *upd_samples = (((uint64_t)pos * sinc_inv) >> 32) + 1;
513 //xprintf("ch%02d: irq sched: %3d %03d\n",
514 // ch, *upd_samples, *upd_samples * 60 * 263 / 44100);
518 #define make_do_samples(name, fmod_code, interp_start, interp1_code, interp2_code, interp_end) \
519 static noinline int do_samples_##name( \
520 int (*decode_f)(void *context, int ch, int *SB), void *ctx, \
521 int ch, int ns_to, int *SB, int sinc, int *spos, int *sbpos) \
527 for (ns = 0; ns < ns_to; ns++) \
532 while (*spos >= 0x10000) \
534 fa = SB[(*sbpos)++]; \
538 d = decode_f(ctx, ch, SB); \
555 #define fmod_recv_check \
556 if(spu.s_chan[ch].bFMod==1 && iFMod[ns]) \
557 sinc = FModChangeFrequency(SB, spu.s_chan[ch].iRawPitch, ns)
559 make_do_samples(default, fmod_recv_check, ,
560 StoreInterpolationVal(SB, sinc, fa, spu.s_chan[ch].bFMod==2),
561 ChanBuf[ns] = iGetInterpolationVal(SB, sinc, *spos, spu.s_chan[ch].bFMod==2), )
562 make_do_samples(noint, , fa = SB[29], , ChanBuf[ns] = fa, SB[29] = fa)
564 #define simple_interp_store \
571 #define simple_interp_get \
572 if(sinc<0x10000) /* -> upsampling? */ \
573 InterpolateUp(SB, sinc); /* --> interpolate up */ \
574 else InterpolateDown(SB, sinc); /* --> else down */ \
577 make_do_samples(simple, , ,
578 simple_interp_store, simple_interp_get, )
580 static int do_samples_skip(int ch, int ns_to)
582 SPUCHAN *s_chan = &spu.s_chan[ch];
583 int spos = s_chan->spos;
584 int sinc = s_chan->sinc;
585 int ret = ns_to, ns, d;
587 spos += s_chan->iSBPos << 16;
589 for (ns = 0; ns < ns_to; ns++)
592 while (spos >= 28*0x10000)
601 s_chan->iSBPos = spos >> 16;
602 s_chan->spos = spos & 0xffff;
607 static void do_lsfr_samples(int ns_to, int ctrl,
608 unsigned int *dwNoiseCount, unsigned int *dwNoiseVal)
610 unsigned int counter = *dwNoiseCount;
611 unsigned int val = *dwNoiseVal;
612 unsigned int level, shift, bit;
615 // modified from DrHell/shalma, no fraction
616 level = (ctrl >> 10) & 0x0f;
617 level = 0x8000 >> level;
619 for (ns = 0; ns < ns_to; ns++)
622 if (counter >= level)
625 shift = (val >> 10) & 0x1f;
626 bit = (0x69696969 >> shift) & 1;
627 bit ^= (val >> 15) & 1;
628 val = (val << 1) | bit;
631 ChanBuf[ns] = (signed short)val;
634 *dwNoiseCount = counter;
638 static int do_samples_noise(int ch, int ns_to)
642 ret = do_samples_skip(ch, ns_to);
644 do_lsfr_samples(ns_to, spu.spuCtrl, &spu.dwNoiseCount, &spu.dwNoiseVal);
650 // asm code; lv and rv must be 0-3fff
651 extern void mix_chan(int *SSumLR, int count, int lv, int rv);
652 extern void mix_chan_rvb(int *SSumLR, int count, int lv, int rv, int *rvb);
654 static void mix_chan(int *SSumLR, int count, int lv, int rv)
656 const int *src = ChanBuf;
663 l = (sval * lv) >> 14;
664 r = (sval * rv) >> 14;
670 static void mix_chan_rvb(int *SSumLR, int count, int lv, int rv, int *rvb)
672 const int *src = ChanBuf;
681 l = (sval * lv) >> 14;
682 r = (sval * rv) >> 14;
691 // 0x0800-0x0bff Voice 1
692 // 0x0c00-0x0fff Voice 3
693 static noinline void do_decode_bufs(unsigned short *mem, int which,
694 int count, int decode_pos)
696 unsigned short *dst = &mem[0x800/2 + which*0x400/2];
697 const int *src = ChanBuf;
698 int cursor = decode_pos;
703 dst[cursor] = *src++;
707 // decode_pos is updated and irqs are checked later, after voice loop
710 static void do_silent_chans(int ns_to, int silentch)
716 mask = silentch & 0xffffff;
717 for (ch = 0; mask != 0; ch++, mask >>= 1)
719 if (!(mask & 1)) continue;
720 if (spu.dwChannelDead & (1<<ch)) continue;
722 s_chan = &spu.s_chan[ch];
723 if (s_chan->pCurr > spu.pSpuIrq && s_chan->pLoop > spu.pSpuIrq)
726 s_chan->spos += s_chan->iSBPos << 16;
729 s_chan->spos += s_chan->sinc * ns_to;
730 while (s_chan->spos >= 28 * 0x10000)
732 unsigned char *start = s_chan->pCurr;
735 if (start == s_chan->pCurr || start - spu.spuMemC < 0x1000)
737 // looping on self or stopped(?)
738 spu.dwChannelDead |= 1<<ch;
743 s_chan->spos -= 28 * 0x10000;
748 static void do_channels(int ns_to)
755 do_rvb = spu.rvb->StartAddr && spu_config.iUseReverb;
757 memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2);
759 mask = spu.dwNewChannel & 0xffffff;
760 for (ch = 0; mask != 0; ch++, mask >>= 1) {
765 mask = spu.dwChannelsAudible & 0xffffff;
766 for (ch = 0; mask != 0; ch++, mask >>= 1) // loop em all...
768 if (!(mask & 1)) continue; // channel not playing? next
770 s_chan = &spu.s_chan[ch];
771 SB = spu.SB + ch * SB_SIZE;
775 d = do_samples_noise(ch, ns_to);
776 else if (s_chan->bFMod == 2
777 || (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 0))
778 d = do_samples_noint(decode_block, NULL, ch, ns_to,
779 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
780 else if (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 1)
781 d = do_samples_simple(decode_block, NULL, ch, ns_to,
782 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
784 d = do_samples_default(decode_block, NULL, ch, ns_to,
785 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
787 d = MixADSR(&s_chan->ADSRX, d);
789 spu.dwChannelsAudible &= ~(1 << ch);
790 s_chan->ADSRX.State = ADSR_RELEASE;
791 s_chan->ADSRX.EnvelopeVol = 0;
792 memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0]));
795 if (ch == 1 || ch == 3)
797 do_decode_bufs(spu.spuMem, ch/2, ns_to, spu.decode_pos);
798 spu.decode_dirty_ch |= 1 << ch;
801 if (s_chan->bFMod == 2) // fmod freq channel
802 memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
803 if (s_chan->bRVBActive && do_rvb)
804 mix_chan_rvb(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume, RVB);
806 mix_chan(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume);
809 if (spu.rvb->StartAddr) {
811 REVERBDo(spu.SSumLR, RVB, ns_to, spu.rvb->CurrAddr);
813 spu.rvb->CurrAddr += ns_to / 2;
814 while (spu.rvb->CurrAddr >= 0x40000)
815 spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr;
819 static void do_samples_finish(int *SSumLR, int ns_to,
820 int silentch, int decode_pos);
822 // optional worker thread handling
824 #if defined(THREAD_ENABLED) || defined(WANT_THREAD_CODE)
826 // worker thread state
827 static struct spu_worker {
830 unsigned int exit_thread;
831 unsigned int i_ready;
832 unsigned int i_reaped;
833 unsigned int last_boot_cnt; // dsp
834 unsigned int ram_dirty;
836 // aligning for C64X_DSP
837 unsigned int _pad0[128/4];
842 unsigned int active; // dsp
843 unsigned int boot_cnt;
845 unsigned int _pad1[128/4];
852 unsigned int channels_new;
853 unsigned int channels_on;
854 unsigned int channels_silent;
865 // might also want to add fmod flags..
867 int SSumLR[NSSIZE * 2];
871 #define WORK_MAXCNT (sizeof(worker->i) / sizeof(worker->i[0]))
872 #define WORK_I_MASK (WORK_MAXCNT - 1)
874 static void thread_work_start(void);
875 static void thread_work_wait_sync(struct work_item *work, int force);
876 static void thread_sync_caches(void);
877 static int thread_get_i_done(void);
879 static int decode_block_work(void *context, int ch, int *SB)
881 const unsigned char *ram = spu.spuMemC;
882 int predict_nr, shift_factor, flags;
883 struct work_item *work = context;
884 int start = work->ch[ch].start;
885 int loop = work->ch[ch].loop;
887 predict_nr = ram[start];
888 shift_factor = predict_nr & 0xf;
891 decode_block_data(SB, ram + start + 2, predict_nr, shift_factor);
893 flags = ram[start + 1];
895 loop = start; // loop adress
899 if (flags & 1) // 1: stop/loop
902 work->ch[ch].start = start & 0x7ffff;
903 work->ch[ch].loop = loop;
908 static void queue_channel_work(int ns_to, unsigned int silentch)
910 struct work_item *work;
915 work = &worker->i[worker->i_ready & WORK_I_MASK];
917 work->ctrl = spu.spuCtrl;
918 work->decode_pos = spu.decode_pos;
919 work->channels_silent = silentch;
921 mask = work->channels_new = spu.dwNewChannel & 0xffffff;
922 for (ch = 0; mask != 0; ch++, mask >>= 1) {
927 mask = work->channels_on = spu.dwChannelsAudible & 0xffffff;
928 spu.decode_dirty_ch |= mask & 0x0a;
930 for (ch = 0; mask != 0; ch++, mask >>= 1)
932 if (!(mask & 1)) continue;
934 s_chan = &spu.s_chan[ch];
935 work->ch[ch].spos = s_chan->spos;
936 work->ch[ch].sbpos = s_chan->iSBPos;
937 work->ch[ch].sinc = s_chan->sinc;
938 work->ch[ch].adsr = s_chan->ADSRX;
939 work->ch[ch].vol_l = s_chan->iLeftVolume;
940 work->ch[ch].vol_r = s_chan->iRightVolume;
941 work->ch[ch].start = s_chan->pCurr - spu.spuMemC;
942 work->ch[ch].loop = s_chan->pLoop - spu.spuMemC;
943 if (s_chan->prevflags & 1)
944 work->ch[ch].start = work->ch[ch].loop;
946 d = do_samples_skip(ch, ns_to);
947 work->ch[ch].ns_to = d;
949 // note: d is not accurate on skip
950 d = SkipADSR(&s_chan->ADSRX, d);
952 spu.dwChannelsAudible &= ~(1 << ch);
953 s_chan->ADSRX.EnvelopeVol = 0;
958 if (spu.rvb->StartAddr) {
959 if (spu_config.iUseReverb)
960 work->rvb_addr = spu.rvb->CurrAddr;
962 spu.rvb->CurrAddr += ns_to / 2;
963 while (spu.rvb->CurrAddr >= 0x40000)
964 spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr;
971 static void do_channel_work(struct work_item *work)
974 unsigned int decode_dirty_ch = 0;
975 const SPUCHAN *s_chan;
976 int *SB, sinc, spos, sbpos;
982 memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2);
984 mask = work->channels_new;
985 for (ch = 0; mask != 0; ch++, mask >>= 1) {
987 StartSoundSB(spu.SB + ch * SB_SIZE);
990 mask = work->channels_on;
991 for (ch = 0; mask != 0; ch++, mask >>= 1)
993 if (!(mask & 1)) continue;
995 d = work->ch[ch].ns_to;
996 spos = work->ch[ch].spos;
997 sbpos = work->ch[ch].sbpos;
998 sinc = work->ch[ch].sinc;
1000 s_chan = &spu.s_chan[ch];
1001 SB = spu.SB + ch * SB_SIZE;
1004 do_lsfr_samples(d, work->ctrl, &spu.dwNoiseCount, &spu.dwNoiseVal);
1005 else if (s_chan->bFMod == 2
1006 || (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 0))
1007 do_samples_noint(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1008 else if (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 1)
1009 do_samples_simple(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1011 do_samples_default(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1013 d = MixADSR(&work->ch[ch].adsr, d);
1015 work->ch[ch].adsr.EnvelopeVol = 0;
1016 memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0]));
1019 if (ch == 1 || ch == 3)
1021 do_decode_bufs(spu.spuMem, ch/2, ns_to, work->decode_pos);
1022 decode_dirty_ch |= 1 << ch;
1025 if (s_chan->bFMod == 2) // fmod freq channel
1026 memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
1027 if (s_chan->bRVBActive && work->rvb_addr)
1028 mix_chan_rvb(work->SSumLR, ns_to,
1029 work->ch[ch].vol_l, work->ch[ch].vol_r, RVB);
1031 mix_chan(work->SSumLR, ns_to, work->ch[ch].vol_l, work->ch[ch].vol_r);
1035 REVERBDo(work->SSumLR, RVB, ns_to, work->rvb_addr);
1038 static void sync_worker_thread(int force)
1040 struct work_item *work;
1041 int done, used_space;
1043 // rvb offsets will change, thread may be using them
1044 force |= spu.rvb->dirty && spu.rvb->StartAddr;
1046 done = thread_get_i_done() - worker->i_reaped;
1047 used_space = worker->i_ready - worker->i_reaped;
1049 //printf("done: %d use: %d dsp: %u/%u\n", done, used_space,
1050 // worker->boot_cnt, worker->i_done);
1052 while ((force && used_space > 0) || used_space >= WORK_MAXCNT || done > 0) {
1053 work = &worker->i[worker->i_reaped & WORK_I_MASK];
1054 thread_work_wait_sync(work, force);
1056 do_samples_finish(work->SSumLR, work->ns_to,
1057 work->channels_silent, work->decode_pos);
1060 done = thread_get_i_done() - worker->i_reaped;
1061 used_space = worker->i_ready - worker->i_reaped;
1064 thread_sync_caches();
1069 static void queue_channel_work(int ns_to, int silentch) {}
1070 static void sync_worker_thread(int force) {}
1072 static const void * const worker = NULL;
1074 #endif // THREAD_ENABLED
1076 ////////////////////////////////////////////////////////////////////////
1077 // MAIN SPU FUNCTION
1078 // here is the main job handler...
1079 ////////////////////////////////////////////////////////////////////////
1081 void do_samples(unsigned int cycles_to, int do_direct)
1083 unsigned int silentch;
1087 cycle_diff = cycles_to - spu.cycles_played;
1088 if (cycle_diff < -2*1048576 || cycle_diff > 2*1048576)
1090 //xprintf("desync %u %d\n", cycles_to, cycle_diff);
1091 spu.cycles_played = cycles_to;
1095 silentch = ~(spu.dwChannelsAudible | spu.dwNewChannel) & 0xffffff;
1097 do_direct |= (silentch == 0xffffff);
1099 sync_worker_thread(do_direct);
1101 if (cycle_diff < 2 * 768)
1104 ns_to = (cycle_diff / 768 + 1) & ~1;
1105 if (ns_to > NSSIZE) {
1106 // should never happen
1107 //xprintf("ns_to oflow %d %d\n", ns_to, NSSIZE);
1111 //////////////////////////////////////////////////////
1112 // special irq handling in the decode buffers (0x0000-0x1000)
1114 // the decode buffers are located in spu memory in the following way:
1115 // 0x0000-0x03ff CD audio left
1116 // 0x0400-0x07ff CD audio right
1117 // 0x0800-0x0bff Voice 1
1118 // 0x0c00-0x0fff Voice 3
1119 // and decoded data is 16 bit for one sample
1121 // even if voices 1/3 are off or no cd audio is playing, the internal
1122 // play positions will move on and wrap after 0x400 bytes.
1123 // Therefore: we just need a pointer from spumem+0 to spumem+3ff, and
1124 // increase this pointer on each sample by 2 bytes. If this pointer
1125 // (or 0x400 offsets of this pointer) hits the spuirq address, we generate
1128 if (unlikely((spu.spuCtrl & CTRL_IRQ)
1129 && spu.pSpuIrq < spu.spuMemC+0x1000))
1131 int irq_pos = (spu.pSpuIrq - spu.spuMemC) / 2 & 0x1ff;
1132 int left = (irq_pos - spu.decode_pos) & 0x1ff;
1133 if (0 < left && left <= ns_to)
1135 //xprintf("decoder irq %x\n", spu.decode_pos);
1140 if (unlikely(spu.rvb->dirty))
1143 if (do_direct || worker == NULL || !spu_config.iUseThread) {
1145 do_samples_finish(spu.SSumLR, ns_to, silentch, spu.decode_pos);
1148 queue_channel_work(ns_to, silentch);
1151 // advance "stopped" channels that can cause irqs
1152 // (all chans are always playing on the real thing..)
1153 if (spu.spuCtrl & CTRL_IRQ)
1154 do_silent_chans(ns_to, silentch);
1156 spu.cycles_played += ns_to * 768;
1157 spu.decode_pos = (spu.decode_pos + ns_to) & 0x1ff;
1160 static void do_samples_finish(int *SSumLR, int ns_to,
1161 int silentch, int decode_pos)
1163 int vol_l = ((int)regAreaGet(H_SPUmvolL) << 17) >> 17;
1164 int vol_r = ((int)regAreaGet(H_SPUmvolR) << 17) >> 17;
1168 // must clear silent channel decode buffers
1169 if(unlikely(silentch & spu.decode_dirty_ch & (1<<1)))
1171 memset(&spu.spuMem[0x800/2], 0, 0x400);
1172 spu.decode_dirty_ch &= ~(1<<1);
1174 if(unlikely(silentch & spu.decode_dirty_ch & (1<<3)))
1176 memset(&spu.spuMem[0xc00/2], 0, 0x400);
1177 spu.decode_dirty_ch &= ~(1<<3);
1180 MixXA(SSumLR, ns_to, decode_pos);
1182 vol_l = vol_l * spu_config.iVolume >> 10;
1183 vol_r = vol_r * spu_config.iVolume >> 10;
1185 if (!(spu.spuCtrl & 0x4000) || !(vol_l | vol_r))
1188 memset(spu.pS, 0, ns_to * 2 * sizeof(spu.pS[0]));
1189 memset(SSumLR, 0, ns_to * 2 * sizeof(SSumLR[0]));
1190 spu.pS += ns_to * 2;
1193 for (ns = 0; ns < ns_to * 2; )
1195 d = SSumLR[ns]; SSumLR[ns] = 0;
1196 d = d * vol_l >> 14;
1201 d = SSumLR[ns]; SSumLR[ns] = 0;
1202 d = d * vol_r >> 14;
1209 void schedule_next_irq(void)
1211 unsigned int upd_samples;
1214 if (spu.scheduleCallback == NULL)
1217 upd_samples = 44100 / 50;
1219 for (ch = 0; ch < MAXCHAN; ch++)
1221 if (spu.dwChannelDead & (1 << ch))
1223 if ((unsigned long)(spu.pSpuIrq - spu.s_chan[ch].pCurr) > IRQ_NEAR_BLOCKS * 16
1224 && (unsigned long)(spu.pSpuIrq - spu.s_chan[ch].pLoop) > IRQ_NEAR_BLOCKS * 16)
1226 if (spu.s_chan[ch].sinc == 0)
1229 scan_for_irq(ch, &upd_samples);
1232 if (unlikely(spu.pSpuIrq < spu.spuMemC + 0x1000))
1234 int irq_pos = (spu.pSpuIrq - spu.spuMemC) / 2 & 0x1ff;
1235 int left = (irq_pos - spu.decode_pos) & 0x1ff;
1236 if (0 < left && left < upd_samples) {
1237 //xprintf("decode: %3d (%3d/%3d)\n", left, spu.decode_pos, irq_pos);
1242 if (upd_samples < 44100 / 50)
1243 spu.scheduleCallback(upd_samples * 768);
1246 // SPU ASYNC... even newer epsxe func
1247 // 1 time every 'cycle' cycles... harhar
1249 // rearmed: called dynamically now
1251 void CALLBACK SPUasync(unsigned int cycle, unsigned int flags)
1253 do_samples(cycle, spu_config.iUseFixedUpdates);
1255 if (spu.spuCtrl & CTRL_IRQ)
1256 schedule_next_irq();
1259 out_current->feed(spu.pSpuBuffer, (unsigned char *)spu.pS - spu.pSpuBuffer);
1260 spu.pS = (short *)spu.pSpuBuffer;
1262 if (spu_config.iTempo) {
1263 if (!out_current->busy())
1264 // cause more samples to be generated
1265 // (and break some games because of bad sync)
1266 spu.cycles_played -= 44100 / 60 / 2 * 768;
1271 // SPU UPDATE... new epsxe func
1272 // 1 time every 32 hsync lines
1273 // (312/32)x50 in pal
1274 // (262/32)x60 in ntsc
1276 // since epsxe 1.5.2 (linux) uses SPUupdate, not SPUasync, I will
1277 // leave that func in the linux port, until epsxe linux is using
1278 // the async function as well
1280 void CALLBACK SPUupdate(void)
1286 void CALLBACK SPUplayADPCMchannel(xa_decode_t *xap, unsigned int cycle, int is_start)
1289 if(!xap->freq) return; // no xa freq ? bye
1292 do_samples(cycle, 1); // catch up to prevent source underflows later
1294 FeedXA(xap); // call main XA feeder
1298 int CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes, unsigned int cycle, int is_start)
1300 if (!pcm) return -1;
1301 if (nbytes<=0) return -1;
1304 do_samples(cycle, 1); // catch up to prevent source underflows later
1306 return FeedCDDA((unsigned char *)pcm, nbytes);
1309 // to be called after state load
1310 void ClearWorkingState(void)
1312 memset(iFMod, 0, sizeof(iFMod));
1313 spu.pS=(short *)spu.pSpuBuffer; // setup soundbuffer pointer
1316 // SETUPSTREAMS: init most of the spu buffers
1317 static void SetupStreams(void)
1319 spu.pSpuBuffer = (unsigned char *)malloc(32768); // alloc mixing buffer
1320 spu.SSumLR = calloc(NSSIZE * 2, sizeof(spu.SSumLR[0]));
1322 spu.XAStart = malloc(44100 * sizeof(uint32_t)); // alloc xa buffer
1323 spu.XAEnd = spu.XAStart + 44100;
1324 spu.XAPlay = spu.XAStart;
1325 spu.XAFeed = spu.XAStart;
1327 spu.CDDAStart = malloc(CDDA_BUFFER_SIZE); // alloc cdda buffer
1328 spu.CDDAEnd = spu.CDDAStart + 16384;
1329 spu.CDDAPlay = spu.CDDAStart;
1330 spu.CDDAFeed = spu.CDDAStart;
1332 ClearWorkingState();
1335 // REMOVESTREAMS: free most buffer
1336 static void RemoveStreams(void)
1338 free(spu.pSpuBuffer); // free mixing buffer
1339 spu.pSpuBuffer = NULL;
1342 free(spu.XAStart); // free XA buffer
1344 free(spu.CDDAStart); // free CDDA buffer
1345 spu.CDDAStart = NULL;
1348 #if defined(C64X_DSP)
1350 /* special code for TI C64x DSP */
1351 #include "spu_c64x.c"
1353 #elif defined(THREAD_ENABLED)
1355 #include <pthread.h>
1356 #include <semaphore.h>
1365 /* generic pthread implementation */
1367 static void thread_work_start(void)
1369 sem_post(&t.sem_avail);
1372 static void thread_work_wait_sync(struct work_item *work, int force)
1374 sem_wait(&t.sem_done);
1377 static int thread_get_i_done(void)
1379 return worker->i_done;
1382 static void thread_sync_caches(void)
1386 static void *spu_worker_thread(void *unused)
1388 struct work_item *work;
1391 sem_wait(&t.sem_avail);
1392 if (worker->exit_thread)
1395 work = &worker->i[worker->i_done & WORK_I_MASK];
1396 do_channel_work(work);
1399 sem_post(&t.sem_done);
1405 static void init_spu_thread(void)
1409 if (sysconf(_SC_NPROCESSORS_ONLN) <= 1)
1412 worker = calloc(1, sizeof(*worker));
1415 ret = sem_init(&t.sem_avail, 0, 0);
1417 goto fail_sem_avail;
1418 ret = sem_init(&t.sem_done, 0, 0);
1422 ret = pthread_create(&t.thread, NULL, spu_worker_thread, NULL);
1426 spu_config.iThreadAvail = 1;
1430 sem_destroy(&t.sem_done);
1432 sem_destroy(&t.sem_avail);
1436 spu_config.iThreadAvail = 0;
1439 static void exit_spu_thread(void)
1443 worker->exit_thread = 1;
1444 sem_post(&t.sem_avail);
1445 pthread_join(t.thread, NULL);
1446 sem_destroy(&t.sem_done);
1447 sem_destroy(&t.sem_avail);
1452 #else // if !THREAD_ENABLED
1454 static void init_spu_thread(void)
1458 static void exit_spu_thread(void)
1464 // SPUINIT: this func will be called first by the main emu
1465 long CALLBACK SPUinit(void)
1469 spu.spuMemC = calloc(1, 512 * 1024);
1472 spu.s_chan = calloc(MAXCHAN+1, sizeof(spu.s_chan[0])); // channel + 1 infos (1 is security for fmod handling)
1473 spu.rvb = calloc(1, sizeof(REVERBInfo));
1474 spu.SB = calloc(MAXCHAN, sizeof(spu.SB[0]) * SB_SIZE);
1478 spu.pSpuIrq = spu.spuMemC;
1480 SetupStreams(); // prepare streaming
1482 if (spu_config.iVolume == 0)
1483 spu_config.iVolume = 768; // 1024 is 1.0
1487 for (i = 0; i < MAXCHAN; i++) // loop sound channels
1489 spu.s_chan[i].ADSRX.SustainLevel = 0xf; // -> init sustain
1490 spu.s_chan[i].ADSRX.SustainIncrease = 1;
1491 spu.s_chan[i].pLoop = spu.spuMemC;
1492 spu.s_chan[i].pCurr = spu.spuMemC;
1493 spu.s_chan[i].bIgnoreLoop = 0;
1496 spu.bSpuInit=1; // flag: we are inited
1501 // SPUOPEN: called by main emu after init
1502 long CALLBACK SPUopen(void)
1504 if (spu.bSPUIsOpen) return 0; // security for some stupid main emus
1506 SetupSound(); // setup sound (before init!)
1510 return PSE_SPU_ERR_SUCCESS;
1513 // SPUCLOSE: called before shutdown
1514 long CALLBACK SPUclose(void)
1516 if (!spu.bSPUIsOpen) return 0; // some security
1518 spu.bSPUIsOpen = 0; // no more open
1520 out_current->finish(); // no more sound handling
1525 // SPUSHUTDOWN: called by main emu on final exit
1526 long CALLBACK SPUshutdown(void)
1541 RemoveStreams(); // no more streaming
1547 // SPUTEST: we don't test, we are always fine ;)
1548 long CALLBACK SPUtest(void)
1553 // SPUCONFIGURE: call config dialog
1554 long CALLBACK SPUconfigure(void)
1559 // StartCfgTool("CFG");
1564 // SPUABOUT: show about window
1565 void CALLBACK SPUabout(void)
1570 // StartCfgTool("ABOUT");
1575 // this functions will be called once,
1576 // passes a callback that should be called on SPU-IRQ/cdda volume change
1577 void CALLBACK SPUregisterCallback(void (CALLBACK *callback)(void))
1579 spu.irqCallback = callback;
1582 void CALLBACK SPUregisterCDDAVolume(void (CALLBACK *CDDAVcallback)(short, short))
1584 spu.cddavCallback = CDDAVcallback;
1587 void CALLBACK SPUregisterScheduleCb(void (CALLBACK *callback)(unsigned int))
1589 spu.scheduleCallback = callback;
1592 // COMMON PLUGIN INFO FUNCS
1594 char * CALLBACK PSEgetLibName(void)
1596 return _(libraryName);
1599 unsigned long CALLBACK PSEgetLibType(void)
1604 unsigned long CALLBACK PSEgetLibVersion(void)
1606 return (1 << 16) | (6 << 8);
1609 char * SPUgetLibInfos(void)
1611 return _(libraryInfo);
1616 void spu_get_debug_info(int *chans_out, int *run_chans, int *fmod_chans_out, int *noise_chans_out)
1618 int ch = 0, fmod_chans = 0, noise_chans = 0, irq_chans = 0;
1620 if (spu.s_chan == NULL)
1623 for(;ch<MAXCHAN;ch++)
1625 if (!(spu.dwChannelsAudible & (1<<ch)))
1627 if (spu.s_chan[ch].bFMod == 2)
1628 fmod_chans |= 1 << ch;
1629 if (spu.s_chan[ch].bNoise)
1630 noise_chans |= 1 << ch;
1631 if((spu.spuCtrl&CTRL_IRQ) && spu.s_chan[ch].pCurr <= spu.pSpuIrq && spu.s_chan[ch].pLoop <= spu.pSpuIrq)
1632 irq_chans |= 1 << ch;
1635 *chans_out = spu.dwChannelsAudible;
1636 *run_chans = ~spu.dwChannelsAudible & ~spu.dwChannelDead & irq_chans;
1637 *fmod_chans_out = fmod_chans;
1638 *noise_chans_out = noise_chans;
1641 // vim:shiftwidth=1:expandtab