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_ON|CTRL_IRQ)) == (CTRL_ON|CTRL_IRQ) && pos == spu.pSpuIrq)
217 //printf("ch%d irq %04x\n", ch, pos - spu.spuMemC);
224 void check_irq_io(unsigned int addr)
226 unsigned int irq_addr = regAreaGet(H_SPUirqAddr) << 3;
228 if((spu.spuCtrl & (CTRL_ON|CTRL_IRQ)) == (CTRL_ON|CTRL_IRQ) && addr == irq_addr)
230 //printf("io irq %04x\n", irq_addr);
235 ////////////////////////////////////////////////////////////////////////
236 // START SOUND... called by main thread to setup a new sound on a channel
237 ////////////////////////////////////////////////////////////////////////
239 static void StartSoundSB(int *SB)
241 SB[26]=0; // init mixing vars
245 SB[29]=0; // init our interpolation helpers
250 static void StartSoundMain(int ch)
252 SPUCHAN *s_chan = &spu.s_chan[ch];
261 s_chan->pCurr = spu.spuMemC + ((regAreaGetCh(ch, 6) & ~1) << 3);
263 spu.dwNewChannel&=~(1<<ch); // clear new channel bit
264 spu.dwChannelDead&=~(1<<ch);
265 spu.dwChannelsAudible|=1<<ch;
268 static void StartSound(int ch)
271 StartSoundSB(spu.SB + ch * SB_SIZE);
274 ////////////////////////////////////////////////////////////////////////
275 // ALL KIND OF HELPERS
276 ////////////////////////////////////////////////////////////////////////
278 INLINE int FModChangeFrequency(int *SB, int pitch, int ns)
280 unsigned int NP=pitch;
283 NP=((32768L+iFMod[ns])*NP)>>15;
285 if(NP>0x3fff) NP=0x3fff;
288 sinc=NP<<4; // calc frequency
290 SB[32]=1; // reset interpolation
295 ////////////////////////////////////////////////////////////////////////
297 INLINE void StoreInterpolationVal(int *SB, int sinc, int fa, int fmod_freq)
299 if(fmod_freq) // fmod freq channel
305 if(spu_config.iUseInterpolation>=2) // gauss/cubic interpolation
313 if(spu_config.iUseInterpolation==1) // simple interpolation
316 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'
319 SB[32] = 1; // -> flag: calc new interolation
321 else SB[29]=fa; // no interpolation
325 ////////////////////////////////////////////////////////////////////////
327 INLINE int iGetInterpolationVal(int *SB, int sinc, int spos, int fmod_freq)
331 if(fmod_freq) return SB[29];
333 switch(spu_config.iUseInterpolation)
335 //--------------------------------------------------//
336 case 3: // cubic interpolation
342 fa = gval(3) - 3*gval(2) + 3*gval(1) - gval0;
343 fa *= (xd - (2<<15)) / 6;
345 fa += gval(2) - gval(1) - gval(1) + gval0;
346 fa *= (xd - (1<<15)) >> 1;
348 fa += gval(1) - gval0;
354 //--------------------------------------------------//
355 case 2: // gauss interpolation
358 vl = (spos >> 6) & ~3;
360 vr=(gauss[vl]*(int)gval0) >> 15;
361 vr+=(gauss[vl+1]*gval(1)) >> 15;
362 vr+=(gauss[vl+2]*gval(2)) >> 15;
363 vr+=(gauss[vl+3]*gval(3)) >> 15;
366 //--------------------------------------------------//
367 case 1: // simple interpolation
369 if(sinc<0x10000L) // -> upsampling?
370 InterpolateUp(SB, sinc); // --> interpolate up
371 else InterpolateDown(SB, sinc); // --> else down
374 //--------------------------------------------------//
375 default: // no interpolation
379 //--------------------------------------------------//
385 static void decode_block_data(int *dest, const unsigned char *src, int predict_nr, int shift_factor)
387 static const int f[16][2] = {
395 int fa, s_1, s_2, d, s;
400 for (nSample = 0; nSample < 28; src++)
403 s = (int)(signed short)((d & 0x0f) << 12);
405 fa = s >> shift_factor;
406 fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6);
409 dest[nSample++] = fa;
411 s = (int)(signed short)((d & 0xf0) << 8);
412 fa = s >> shift_factor;
413 fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6);
416 dest[nSample++] = fa;
420 static int decode_block(void *unused, int ch, int *SB)
422 SPUCHAN *s_chan = &spu.s_chan[ch];
423 unsigned char *start;
424 int predict_nr, shift_factor, flags;
427 start = s_chan->pCurr; // set up the current pos
428 if (start == spu.spuMemC) // ?
431 if (s_chan->prevflags & 1) // 1: stop/loop
433 if (!(s_chan->prevflags & 2))
436 start = s_chan->pLoop;
439 check_irq(ch, start);
441 predict_nr = start[0];
442 shift_factor = predict_nr & 0xf;
445 decode_block_data(SB, start + 2, predict_nr, shift_factor);
448 if (flags & 4 && !s_chan->bIgnoreLoop)
449 s_chan->pLoop = start; // loop adress
453 s_chan->pCurr = start; // store values for next cycle
454 s_chan->prevflags = flags;
459 // do block, but ignore sample data
460 static int skip_block(int ch)
462 SPUCHAN *s_chan = &spu.s_chan[ch];
463 unsigned char *start = s_chan->pCurr;
467 if (s_chan->prevflags & 1) {
468 if (!(s_chan->prevflags & 2))
471 start = s_chan->pLoop;
474 check_irq(ch, start);
477 if (flags & 4 && !s_chan->bIgnoreLoop)
478 s_chan->pLoop = start;
482 s_chan->pCurr = start;
483 s_chan->prevflags = flags;
488 // if irq is going to trigger sooner than in upd_samples, set upd_samples
489 static void scan_for_irq(int ch, unsigned int *upd_samples)
491 SPUCHAN *s_chan = &spu.s_chan[ch];
492 int pos, sinc, sinc_inv, end;
493 unsigned char *block;
496 block = s_chan->pCurr;
499 end = pos + *upd_samples * sinc;
501 pos += (28 - s_chan->iSBPos) << 16;
504 if (block == spu.pSpuIrq)
508 if (flags & 1) { // 1: stop/loop
509 block = s_chan->pLoop;
516 sinc_inv = s_chan->sinc_inv;
518 sinc_inv = s_chan->sinc_inv = (0x80000000u / (uint32_t)sinc) << 1;
521 *upd_samples = (((uint64_t)pos * sinc_inv) >> 32) + 1;
522 //xprintf("ch%02d: irq sched: %3d %03d\n",
523 // ch, *upd_samples, *upd_samples * 60 * 263 / 44100);
527 #define make_do_samples(name, fmod_code, interp_start, interp1_code, interp2_code, interp_end) \
528 static noinline int do_samples_##name( \
529 int (*decode_f)(void *context, int ch, int *SB), void *ctx, \
530 int ch, int ns_to, int *SB, int sinc, int *spos, int *sbpos) \
536 for (ns = 0; ns < ns_to; ns++) \
541 while (*spos >= 0x10000) \
543 fa = SB[(*sbpos)++]; \
547 d = decode_f(ctx, ch, SB); \
564 #define fmod_recv_check \
565 if(spu.s_chan[ch].bFMod==1 && iFMod[ns]) \
566 sinc = FModChangeFrequency(SB, spu.s_chan[ch].iRawPitch, ns)
568 make_do_samples(default, fmod_recv_check, ,
569 StoreInterpolationVal(SB, sinc, fa, spu.s_chan[ch].bFMod==2),
570 ChanBuf[ns] = iGetInterpolationVal(SB, sinc, *spos, spu.s_chan[ch].bFMod==2), )
571 make_do_samples(noint, , fa = SB[29], , ChanBuf[ns] = fa, SB[29] = fa)
573 #define simple_interp_store \
580 #define simple_interp_get \
581 if(sinc<0x10000) /* -> upsampling? */ \
582 InterpolateUp(SB, sinc); /* --> interpolate up */ \
583 else InterpolateDown(SB, sinc); /* --> else down */ \
586 make_do_samples(simple, , ,
587 simple_interp_store, simple_interp_get, )
589 static int do_samples_skip(int ch, int ns_to)
591 SPUCHAN *s_chan = &spu.s_chan[ch];
592 int spos = s_chan->spos;
593 int sinc = s_chan->sinc;
594 int ret = ns_to, ns, d;
596 spos += s_chan->iSBPos << 16;
598 for (ns = 0; ns < ns_to; ns++)
601 while (spos >= 28*0x10000)
610 s_chan->iSBPos = spos >> 16;
611 s_chan->spos = spos & 0xffff;
616 static void do_lsfr_samples(int ns_to, int ctrl,
617 unsigned int *dwNoiseCount, unsigned int *dwNoiseVal)
619 unsigned int counter = *dwNoiseCount;
620 unsigned int val = *dwNoiseVal;
621 unsigned int level, shift, bit;
624 // modified from DrHell/shalma, no fraction
625 level = (ctrl >> 10) & 0x0f;
626 level = 0x8000 >> level;
628 for (ns = 0; ns < ns_to; ns++)
631 if (counter >= level)
634 shift = (val >> 10) & 0x1f;
635 bit = (0x69696969 >> shift) & 1;
636 bit ^= (val >> 15) & 1;
637 val = (val << 1) | bit;
640 ChanBuf[ns] = (signed short)val;
643 *dwNoiseCount = counter;
647 static int do_samples_noise(int ch, int ns_to)
651 ret = do_samples_skip(ch, ns_to);
653 do_lsfr_samples(ns_to, spu.spuCtrl, &spu.dwNoiseCount, &spu.dwNoiseVal);
659 // asm code; lv and rv must be 0-3fff
660 extern void mix_chan(int *SSumLR, int count, int lv, int rv);
661 extern void mix_chan_rvb(int *SSumLR, int count, int lv, int rv, int *rvb);
663 static void mix_chan(int *SSumLR, int count, int lv, int rv)
665 const int *src = ChanBuf;
672 l = (sval * lv) >> 14;
673 r = (sval * rv) >> 14;
679 static void mix_chan_rvb(int *SSumLR, int count, int lv, int rv, int *rvb)
681 const int *src = ChanBuf;
690 l = (sval * lv) >> 14;
691 r = (sval * rv) >> 14;
700 // 0x0800-0x0bff Voice 1
701 // 0x0c00-0x0fff Voice 3
702 static noinline void do_decode_bufs(unsigned short *mem, int which,
703 int count, int decode_pos)
705 unsigned short *dst = &mem[0x800/2 + which*0x400/2];
706 const int *src = ChanBuf;
707 int cursor = decode_pos;
712 dst[cursor] = *src++;
716 // decode_pos is updated and irqs are checked later, after voice loop
719 static void do_silent_chans(int ns_to, int silentch)
725 mask = silentch & 0xffffff;
726 for (ch = 0; mask != 0; ch++, mask >>= 1)
728 if (!(mask & 1)) continue;
729 if (spu.dwChannelDead & (1<<ch)) continue;
731 s_chan = &spu.s_chan[ch];
732 if (s_chan->pCurr > spu.pSpuIrq && s_chan->pLoop > spu.pSpuIrq)
735 s_chan->spos += s_chan->iSBPos << 16;
738 s_chan->spos += s_chan->sinc * ns_to;
739 while (s_chan->spos >= 28 * 0x10000)
741 unsigned char *start = s_chan->pCurr;
744 if (start == s_chan->pCurr || start - spu.spuMemC < 0x1000)
746 // looping on self or stopped(?)
747 spu.dwChannelDead |= 1<<ch;
752 s_chan->spos -= 28 * 0x10000;
757 static void do_channels(int ns_to)
764 do_rvb = spu.rvb->StartAddr && spu_config.iUseReverb;
766 memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2);
768 mask = spu.dwNewChannel & 0xffffff;
769 for (ch = 0; mask != 0; ch++, mask >>= 1) {
774 mask = spu.dwChannelsAudible & 0xffffff;
775 for (ch = 0; mask != 0; ch++, mask >>= 1) // loop em all...
777 if (!(mask & 1)) continue; // channel not playing? next
779 s_chan = &spu.s_chan[ch];
780 SB = spu.SB + ch * SB_SIZE;
782 if (spu.s_chan[ch].bNewPitch)
783 SB[32] = 1; // reset interpolation
784 spu.s_chan[ch].bNewPitch = 0;
787 d = do_samples_noise(ch, ns_to);
788 else if (s_chan->bFMod == 2
789 || (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 0))
790 d = do_samples_noint(decode_block, NULL, ch, ns_to,
791 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
792 else if (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 1)
793 d = do_samples_simple(decode_block, NULL, ch, ns_to,
794 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
796 d = do_samples_default(decode_block, NULL, ch, ns_to,
797 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
799 d = MixADSR(&s_chan->ADSRX, d);
801 spu.dwChannelsAudible &= ~(1 << ch);
802 s_chan->ADSRX.State = ADSR_RELEASE;
803 s_chan->ADSRX.EnvelopeVol = 0;
804 memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0]));
807 if (ch == 1 || ch == 3)
809 do_decode_bufs(spu.spuMem, ch/2, ns_to, spu.decode_pos);
810 spu.decode_dirty_ch |= 1 << ch;
813 if (s_chan->bFMod == 2) // fmod freq channel
814 memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
815 if (s_chan->bRVBActive && do_rvb)
816 mix_chan_rvb(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume, RVB);
818 mix_chan(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume);
821 if (spu.rvb->StartAddr) {
823 REVERBDo(spu.SSumLR, RVB, ns_to, spu.rvb->CurrAddr);
825 spu.rvb->CurrAddr += ns_to / 2;
826 while (spu.rvb->CurrAddr >= 0x40000)
827 spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr;
831 static void do_samples_finish(int *SSumLR, int ns_to,
832 int silentch, int decode_pos);
834 // optional worker thread handling
836 #if defined(THREAD_ENABLED) || defined(WANT_THREAD_CODE)
838 // worker thread state
839 static struct spu_worker {
842 unsigned int exit_thread;
843 unsigned int i_ready;
844 unsigned int i_reaped;
845 unsigned int last_boot_cnt; // dsp
846 unsigned int ram_dirty;
848 // aligning for C64X_DSP
849 unsigned int _pad0[128/4];
854 unsigned int active; // dsp
855 unsigned int boot_cnt;
857 unsigned int _pad1[128/4];
864 unsigned int channels_new;
865 unsigned int channels_on;
866 unsigned int channels_silent;
875 unsigned short ns_to;
876 unsigned short bNoise:1;
877 unsigned short bFMod:2;
878 unsigned short bRVBActive:1;
879 unsigned short bNewPitch:1;
882 int SSumLR[NSSIZE * 2];
886 #define WORK_MAXCNT (sizeof(worker->i) / sizeof(worker->i[0]))
887 #define WORK_I_MASK (WORK_MAXCNT - 1)
889 static void thread_work_start(void);
890 static void thread_work_wait_sync(struct work_item *work, int force);
891 static void thread_sync_caches(void);
892 static int thread_get_i_done(void);
894 static int decode_block_work(void *context, int ch, int *SB)
896 const unsigned char *ram = spu.spuMemC;
897 int predict_nr, shift_factor, flags;
898 struct work_item *work = context;
899 int start = work->ch[ch].start;
900 int loop = work->ch[ch].loop;
902 predict_nr = ram[start];
903 shift_factor = predict_nr & 0xf;
906 decode_block_data(SB, ram + start + 2, predict_nr, shift_factor);
908 flags = ram[start + 1];
910 loop = start; // loop adress
914 if (flags & 1) // 1: stop/loop
917 work->ch[ch].start = start & 0x7ffff;
918 work->ch[ch].loop = loop;
923 static void queue_channel_work(int ns_to, unsigned int silentch)
925 struct work_item *work;
930 work = &worker->i[worker->i_ready & WORK_I_MASK];
932 work->ctrl = spu.spuCtrl;
933 work->decode_pos = spu.decode_pos;
934 work->channels_silent = silentch;
936 mask = work->channels_new = spu.dwNewChannel & 0xffffff;
937 for (ch = 0; mask != 0; ch++, mask >>= 1) {
942 mask = work->channels_on = spu.dwChannelsAudible & 0xffffff;
943 spu.decode_dirty_ch |= mask & 0x0a;
945 for (ch = 0; mask != 0; ch++, mask >>= 1)
947 if (!(mask & 1)) continue;
949 s_chan = &spu.s_chan[ch];
950 work->ch[ch].spos = s_chan->spos;
951 work->ch[ch].sbpos = s_chan->iSBPos;
952 work->ch[ch].sinc = s_chan->sinc;
953 work->ch[ch].adsr = s_chan->ADSRX;
954 work->ch[ch].vol_l = s_chan->iLeftVolume;
955 work->ch[ch].vol_r = s_chan->iRightVolume;
956 work->ch[ch].start = s_chan->pCurr - spu.spuMemC;
957 work->ch[ch].loop = s_chan->pLoop - spu.spuMemC;
958 work->ch[ch].bNoise = s_chan->bNoise;
959 work->ch[ch].bFMod = s_chan->bFMod;
960 work->ch[ch].bRVBActive = s_chan->bRVBActive;
961 work->ch[ch].bNewPitch = s_chan->bNewPitch;
962 if (s_chan->prevflags & 1)
963 work->ch[ch].start = work->ch[ch].loop;
965 d = do_samples_skip(ch, ns_to);
966 work->ch[ch].ns_to = d;
968 // note: d is not accurate on skip
969 d = SkipADSR(&s_chan->ADSRX, d);
971 spu.dwChannelsAudible &= ~(1 << ch);
972 s_chan->ADSRX.State = ADSR_RELEASE;
973 s_chan->ADSRX.EnvelopeVol = 0;
975 s_chan->bNewPitch = 0;
979 if (spu.rvb->StartAddr) {
980 if (spu_config.iUseReverb)
981 work->rvb_addr = spu.rvb->CurrAddr;
983 spu.rvb->CurrAddr += ns_to / 2;
984 while (spu.rvb->CurrAddr >= 0x40000)
985 spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr;
992 static void do_channel_work(struct work_item *work)
995 int *SB, sinc, spos, sbpos;
1001 memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2);
1003 mask = work->channels_new;
1004 for (ch = 0; mask != 0; ch++, mask >>= 1) {
1006 StartSoundSB(spu.SB + ch * SB_SIZE);
1009 mask = work->channels_on;
1010 for (ch = 0; mask != 0; ch++, mask >>= 1)
1012 if (!(mask & 1)) continue;
1014 d = work->ch[ch].ns_to;
1015 spos = work->ch[ch].spos;
1016 sbpos = work->ch[ch].sbpos;
1017 sinc = work->ch[ch].sinc;
1019 SB = spu.SB + ch * SB_SIZE;
1020 if (work->ch[ch].bNewPitch)
1021 SB[32] = 1; // reset interpolation
1023 if (work->ch[ch].bNoise)
1024 do_lsfr_samples(d, work->ctrl, &spu.dwNoiseCount, &spu.dwNoiseVal);
1025 else if (work->ch[ch].bFMod == 2
1026 || (work->ch[ch].bFMod == 0 && spu_config.iUseInterpolation == 0))
1027 do_samples_noint(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1028 else if (work->ch[ch].bFMod == 0 && spu_config.iUseInterpolation == 1)
1029 do_samples_simple(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1031 do_samples_default(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1033 d = MixADSR(&work->ch[ch].adsr, d);
1035 work->ch[ch].adsr.EnvelopeVol = 0;
1036 memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0]));
1039 if (ch == 1 || ch == 3)
1040 do_decode_bufs(spu.spuMem, ch/2, ns_to, work->decode_pos);
1042 if (work->ch[ch].bFMod == 2) // fmod freq channel
1043 memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
1044 if (work->ch[ch].bRVBActive && work->rvb_addr)
1045 mix_chan_rvb(work->SSumLR, ns_to,
1046 work->ch[ch].vol_l, work->ch[ch].vol_r, RVB);
1048 mix_chan(work->SSumLR, ns_to, work->ch[ch].vol_l, work->ch[ch].vol_r);
1052 REVERBDo(work->SSumLR, RVB, ns_to, work->rvb_addr);
1055 static void sync_worker_thread(int force)
1057 struct work_item *work;
1058 int done, used_space;
1060 // rvb offsets will change, thread may be using them
1061 force |= spu.rvb->dirty && spu.rvb->StartAddr;
1063 done = thread_get_i_done() - worker->i_reaped;
1064 used_space = worker->i_ready - worker->i_reaped;
1066 //printf("done: %d use: %d dsp: %u/%u\n", done, used_space,
1067 // worker->boot_cnt, worker->i_done);
1069 while ((force && used_space > 0) || used_space >= WORK_MAXCNT || done > 0) {
1070 work = &worker->i[worker->i_reaped & WORK_I_MASK];
1071 thread_work_wait_sync(work, force);
1073 do_samples_finish(work->SSumLR, work->ns_to,
1074 work->channels_silent, work->decode_pos);
1077 done = thread_get_i_done() - worker->i_reaped;
1078 used_space = worker->i_ready - worker->i_reaped;
1081 thread_sync_caches();
1086 static void queue_channel_work(int ns_to, int silentch) {}
1087 static void sync_worker_thread(int force) {}
1089 static const void * const worker = NULL;
1091 #endif // THREAD_ENABLED
1093 ////////////////////////////////////////////////////////////////////////
1094 // MAIN SPU FUNCTION
1095 // here is the main job handler...
1096 ////////////////////////////////////////////////////////////////////////
1098 void do_samples(unsigned int cycles_to, int do_direct)
1100 unsigned int silentch;
1104 cycle_diff = cycles_to - spu.cycles_played;
1105 if (cycle_diff < -2*1048576 || cycle_diff > 2*1048576)
1107 //xprintf("desync %u %d\n", cycles_to, cycle_diff);
1108 spu.cycles_played = cycles_to;
1112 silentch = ~(spu.dwChannelsAudible | spu.dwNewChannel) & 0xffffff;
1114 do_direct |= (silentch == 0xffffff);
1116 sync_worker_thread(do_direct);
1118 if (cycle_diff < 2 * 768)
1121 ns_to = (cycle_diff / 768 + 1) & ~1;
1122 if (ns_to > NSSIZE) {
1123 // should never happen
1124 //xprintf("ns_to oflow %d %d\n", ns_to, NSSIZE);
1128 //////////////////////////////////////////////////////
1129 // special irq handling in the decode buffers (0x0000-0x1000)
1131 // the decode buffers are located in spu memory in the following way:
1132 // 0x0000-0x03ff CD audio left
1133 // 0x0400-0x07ff CD audio right
1134 // 0x0800-0x0bff Voice 1
1135 // 0x0c00-0x0fff Voice 3
1136 // and decoded data is 16 bit for one sample
1138 // even if voices 1/3 are off or no cd audio is playing, the internal
1139 // play positions will move on and wrap after 0x400 bytes.
1140 // Therefore: we just need a pointer from spumem+0 to spumem+3ff, and
1141 // increase this pointer on each sample by 2 bytes. If this pointer
1142 // (or 0x400 offsets of this pointer) hits the spuirq address, we generate
1145 if (unlikely((spu.spuCtrl & CTRL_IRQ)
1146 && spu.pSpuIrq < spu.spuMemC+0x1000))
1148 int irq_pos = (spu.pSpuIrq - spu.spuMemC) / 2 & 0x1ff;
1149 int left = (irq_pos - spu.decode_pos) & 0x1ff;
1150 if (0 < left && left <= ns_to)
1152 //xprintf("decoder irq %x\n", spu.decode_pos);
1156 check_irq_io(spu.spuAddr);
1158 if (unlikely(spu.rvb->dirty))
1161 if (do_direct || worker == NULL || !spu_config.iUseThread) {
1163 do_samples_finish(spu.SSumLR, ns_to, silentch, spu.decode_pos);
1166 queue_channel_work(ns_to, silentch);
1167 //sync_worker_thread(1); // uncomment for debug
1170 // advance "stopped" channels that can cause irqs
1171 // (all chans are always playing on the real thing..)
1172 if (spu.spuCtrl & CTRL_IRQ)
1173 do_silent_chans(ns_to, silentch);
1175 spu.cycles_played += ns_to * 768;
1176 spu.decode_pos = (spu.decode_pos + ns_to) & 0x1ff;
1179 static void do_samples_finish(int *SSumLR, int ns_to,
1180 int silentch, int decode_pos)
1182 int vol_l = ((int)regAreaGet(H_SPUmvolL) << 17) >> 17;
1183 int vol_r = ((int)regAreaGet(H_SPUmvolR) << 17) >> 17;
1187 // must clear silent channel decode buffers
1188 if(unlikely(silentch & spu.decode_dirty_ch & (1<<1)))
1190 memset(&spu.spuMem[0x800/2], 0, 0x400);
1191 spu.decode_dirty_ch &= ~(1<<1);
1193 if(unlikely(silentch & spu.decode_dirty_ch & (1<<3)))
1195 memset(&spu.spuMem[0xc00/2], 0, 0x400);
1196 spu.decode_dirty_ch &= ~(1<<3);
1199 MixXA(SSumLR, ns_to, decode_pos);
1201 vol_l = vol_l * spu_config.iVolume >> 10;
1202 vol_r = vol_r * spu_config.iVolume >> 10;
1204 if (!(spu.spuCtrl & 0x4000) || !(vol_l | vol_r))
1207 memset(spu.pS, 0, ns_to * 2 * sizeof(spu.pS[0]));
1208 memset(SSumLR, 0, ns_to * 2 * sizeof(SSumLR[0]));
1209 spu.pS += ns_to * 2;
1212 for (ns = 0; ns < ns_to * 2; )
1214 d = SSumLR[ns]; SSumLR[ns] = 0;
1215 d = d * vol_l >> 14;
1220 d = SSumLR[ns]; SSumLR[ns] = 0;
1221 d = d * vol_r >> 14;
1228 void schedule_next_irq(void)
1230 unsigned int upd_samples;
1233 if (spu.scheduleCallback == NULL)
1236 upd_samples = 44100 / 50;
1238 for (ch = 0; ch < MAXCHAN; ch++)
1240 if (spu.dwChannelDead & (1 << ch))
1242 if ((unsigned long)(spu.pSpuIrq - spu.s_chan[ch].pCurr) > IRQ_NEAR_BLOCKS * 16
1243 && (unsigned long)(spu.pSpuIrq - spu.s_chan[ch].pLoop) > IRQ_NEAR_BLOCKS * 16)
1245 if (spu.s_chan[ch].sinc == 0)
1248 scan_for_irq(ch, &upd_samples);
1251 if (unlikely(spu.pSpuIrq < spu.spuMemC + 0x1000))
1253 int irq_pos = (spu.pSpuIrq - spu.spuMemC) / 2 & 0x1ff;
1254 int left = (irq_pos - spu.decode_pos) & 0x1ff;
1255 if (0 < left && left < upd_samples) {
1256 //xprintf("decode: %3d (%3d/%3d)\n", left, spu.decode_pos, irq_pos);
1261 if (upd_samples < 44100 / 50)
1262 spu.scheduleCallback(upd_samples * 768);
1265 // SPU ASYNC... even newer epsxe func
1266 // 1 time every 'cycle' cycles... harhar
1268 // rearmed: called dynamically now
1270 void CALLBACK SPUasync(unsigned int cycle, unsigned int flags)
1272 do_samples(cycle, spu_config.iUseFixedUpdates);
1274 if (spu.spuCtrl & CTRL_IRQ)
1275 schedule_next_irq();
1278 out_current->feed(spu.pSpuBuffer, (unsigned char *)spu.pS - spu.pSpuBuffer);
1279 spu.pS = (short *)spu.pSpuBuffer;
1281 if (spu_config.iTempo) {
1282 if (!out_current->busy())
1283 // cause more samples to be generated
1284 // (and break some games because of bad sync)
1285 spu.cycles_played -= 44100 / 60 / 2 * 768;
1290 // SPU UPDATE... new epsxe func
1291 // 1 time every 32 hsync lines
1292 // (312/32)x50 in pal
1293 // (262/32)x60 in ntsc
1295 // since epsxe 1.5.2 (linux) uses SPUupdate, not SPUasync, I will
1296 // leave that func in the linux port, until epsxe linux is using
1297 // the async function as well
1299 void CALLBACK SPUupdate(void)
1305 void CALLBACK SPUplayADPCMchannel(xa_decode_t *xap, unsigned int cycle, int is_start)
1308 if(!xap->freq) return; // no xa freq ? bye
1311 do_samples(cycle, 1); // catch up to prevent source underflows later
1313 FeedXA(xap); // call main XA feeder
1317 int CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes, unsigned int cycle, int is_start)
1319 if (!pcm) return -1;
1320 if (nbytes<=0) return -1;
1323 do_samples(cycle, 1); // catch up to prevent source underflows later
1325 return FeedCDDA((unsigned char *)pcm, nbytes);
1328 // to be called after state load
1329 void ClearWorkingState(void)
1331 memset(iFMod, 0, sizeof(iFMod));
1332 spu.pS=(short *)spu.pSpuBuffer; // setup soundbuffer pointer
1335 // SETUPSTREAMS: init most of the spu buffers
1336 static void SetupStreams(void)
1338 spu.pSpuBuffer = (unsigned char *)malloc(32768); // alloc mixing buffer
1339 spu.SSumLR = calloc(NSSIZE * 2, sizeof(spu.SSumLR[0]));
1341 spu.XAStart = malloc(44100 * sizeof(uint32_t)); // alloc xa buffer
1342 spu.XAEnd = spu.XAStart + 44100;
1343 spu.XAPlay = spu.XAStart;
1344 spu.XAFeed = spu.XAStart;
1346 spu.CDDAStart = malloc(CDDA_BUFFER_SIZE); // alloc cdda buffer
1347 spu.CDDAEnd = spu.CDDAStart + 16384;
1348 spu.CDDAPlay = spu.CDDAStart;
1349 spu.CDDAFeed = spu.CDDAStart;
1351 ClearWorkingState();
1354 // REMOVESTREAMS: free most buffer
1355 static void RemoveStreams(void)
1357 free(spu.pSpuBuffer); // free mixing buffer
1358 spu.pSpuBuffer = NULL;
1361 free(spu.XAStart); // free XA buffer
1363 free(spu.CDDAStart); // free CDDA buffer
1364 spu.CDDAStart = NULL;
1367 #if defined(C64X_DSP)
1369 /* special code for TI C64x DSP */
1370 #include "spu_c64x.c"
1372 #elif defined(THREAD_ENABLED)
1374 #include <pthread.h>
1375 #include <semaphore.h>
1384 /* generic pthread implementation */
1386 static void thread_work_start(void)
1388 sem_post(&t.sem_avail);
1391 static void thread_work_wait_sync(struct work_item *work, int force)
1393 sem_wait(&t.sem_done);
1396 static int thread_get_i_done(void)
1398 return worker->i_done;
1401 static void thread_sync_caches(void)
1405 static void *spu_worker_thread(void *unused)
1407 struct work_item *work;
1410 sem_wait(&t.sem_avail);
1411 if (worker->exit_thread)
1414 work = &worker->i[worker->i_done & WORK_I_MASK];
1415 do_channel_work(work);
1418 sem_post(&t.sem_done);
1424 static void init_spu_thread(void)
1428 if (sysconf(_SC_NPROCESSORS_ONLN) <= 1)
1431 worker = calloc(1, sizeof(*worker));
1434 ret = sem_init(&t.sem_avail, 0, 0);
1436 goto fail_sem_avail;
1437 ret = sem_init(&t.sem_done, 0, 0);
1441 ret = pthread_create(&t.thread, NULL, spu_worker_thread, NULL);
1445 spu_config.iThreadAvail = 1;
1449 sem_destroy(&t.sem_done);
1451 sem_destroy(&t.sem_avail);
1455 spu_config.iThreadAvail = 0;
1458 static void exit_spu_thread(void)
1462 worker->exit_thread = 1;
1463 sem_post(&t.sem_avail);
1464 pthread_join(t.thread, NULL);
1465 sem_destroy(&t.sem_done);
1466 sem_destroy(&t.sem_avail);
1471 #else // if !THREAD_ENABLED
1473 static void init_spu_thread(void)
1477 static void exit_spu_thread(void)
1483 // SPUINIT: this func will be called first by the main emu
1484 long CALLBACK SPUinit(void)
1488 spu.spuMemC = calloc(1, 512 * 1024);
1491 spu.s_chan = calloc(MAXCHAN+1, sizeof(spu.s_chan[0])); // channel + 1 infos (1 is security for fmod handling)
1492 spu.rvb = calloc(1, sizeof(REVERBInfo));
1493 spu.SB = calloc(MAXCHAN, sizeof(spu.SB[0]) * SB_SIZE);
1497 spu.pSpuIrq = spu.spuMemC;
1499 SetupStreams(); // prepare streaming
1501 if (spu_config.iVolume == 0)
1502 spu_config.iVolume = 768; // 1024 is 1.0
1506 for (i = 0; i < MAXCHAN; i++) // loop sound channels
1508 spu.s_chan[i].ADSRX.SustainLevel = 0xf; // -> init sustain
1509 spu.s_chan[i].ADSRX.SustainIncrease = 1;
1510 spu.s_chan[i].pLoop = spu.spuMemC;
1511 spu.s_chan[i].pCurr = spu.spuMemC;
1512 spu.s_chan[i].bIgnoreLoop = 0;
1515 spu.bSpuInit=1; // flag: we are inited
1520 // SPUOPEN: called by main emu after init
1521 long CALLBACK SPUopen(void)
1523 if (spu.bSPUIsOpen) return 0; // security for some stupid main emus
1525 SetupSound(); // setup sound (before init!)
1529 return PSE_SPU_ERR_SUCCESS;
1532 // SPUCLOSE: called before shutdown
1533 long CALLBACK SPUclose(void)
1535 if (!spu.bSPUIsOpen) return 0; // some security
1537 spu.bSPUIsOpen = 0; // no more open
1539 out_current->finish(); // no more sound handling
1544 // SPUSHUTDOWN: called by main emu on final exit
1545 long CALLBACK SPUshutdown(void)
1560 RemoveStreams(); // no more streaming
1566 // SPUTEST: we don't test, we are always fine ;)
1567 long CALLBACK SPUtest(void)
1572 // SPUCONFIGURE: call config dialog
1573 long CALLBACK SPUconfigure(void)
1578 // StartCfgTool("CFG");
1583 // SPUABOUT: show about window
1584 void CALLBACK SPUabout(void)
1589 // StartCfgTool("ABOUT");
1594 // this functions will be called once,
1595 // passes a callback that should be called on SPU-IRQ/cdda volume change
1596 void CALLBACK SPUregisterCallback(void (CALLBACK *callback)(void))
1598 spu.irqCallback = callback;
1601 void CALLBACK SPUregisterCDDAVolume(void (CALLBACK *CDDAVcallback)(short, short))
1603 spu.cddavCallback = CDDAVcallback;
1606 void CALLBACK SPUregisterScheduleCb(void (CALLBACK *callback)(unsigned int))
1608 spu.scheduleCallback = callback;
1611 // COMMON PLUGIN INFO FUNCS
1613 char * CALLBACK PSEgetLibName(void)
1615 return _(libraryName);
1618 unsigned long CALLBACK PSEgetLibType(void)
1623 unsigned long CALLBACK PSEgetLibVersion(void)
1625 return (1 << 16) | (6 << 8);
1628 char * SPUgetLibInfos(void)
1630 return _(libraryInfo);
1635 void spu_get_debug_info(int *chans_out, int *run_chans, int *fmod_chans_out, int *noise_chans_out)
1637 int ch = 0, fmod_chans = 0, noise_chans = 0, irq_chans = 0;
1639 if (spu.s_chan == NULL)
1642 for(;ch<MAXCHAN;ch++)
1644 if (!(spu.dwChannelsAudible & (1<<ch)))
1646 if (spu.s_chan[ch].bFMod == 2)
1647 fmod_chans |= 1 << ch;
1648 if (spu.s_chan[ch].bNoise)
1649 noise_chans |= 1 << ch;
1650 if((spu.spuCtrl&CTRL_IRQ) && spu.s_chan[ch].pCurr <= spu.pSpuIrq && spu.s_chan[ch].pLoop <= spu.pSpuIrq)
1651 irq_chans |= 1 << ch;
1654 *chans_out = spu.dwChannelsAudible;
1655 *run_chans = ~spu.dwChannelsAudible & ~spu.dwChannelDead & irq_chans;
1656 *fmod_chans_out = fmod_chans;
1657 *noise_chans_out = noise_chans;
1660 // vim:shiftwidth=1:expandtab