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 ***************************************************************************/
25 #include "externals.h"
26 #include "registers.h"
28 #include "spu_config.h"
31 #include "arm_features.h"
35 #define ssat32_to_16(v) \
36 asm("ssat %0,#16,%1" : "=r" (v) : "r" (v))
38 #define ssat32_to_16(v) do { \
39 if (v < -32768) v = -32768; \
40 else if (v > 32767) v = 32767; \
44 #define PSXCLK 33868800 /* 33.8688 MHz */
46 // intended to be ~1 frame
47 #define IRQ_NEAR_BLOCKS 32
50 #if defined (USEMACOSX)
51 static char * libraryName = N_("Mac OS X Sound");
52 #elif defined (USEALSA)
53 static char * libraryName = N_("ALSA Sound");
54 #elif defined (USEOSS)
55 static char * libraryName = N_("OSS Sound");
56 #elif defined (USESDL)
57 static char * libraryName = N_("SDL Sound");
58 #elif defined (USEPULSEAUDIO)
59 static char * libraryName = N_("PulseAudio Sound");
61 static char * libraryName = N_("NULL Sound");
64 static char * libraryInfo = N_("P.E.Op.S. Sound Driver V1.7\nCoded by Pete Bernert and the P.E.Op.S. team\n");
72 static int iFMod[NSSIZE];
73 static int RVB[NSSIZE * 2];
76 #define CDDA_BUFFER_SIZE (16384 * sizeof(uint32_t)) // must be power of 2
78 ////////////////////////////////////////////////////////////////////////
80 ////////////////////////////////////////////////////////////////////////
82 // dirty inline func includes
87 ////////////////////////////////////////////////////////////////////////
88 // helpers for simple interpolation
91 // easy interpolation on upsampling, no special filter, just "Pete's common sense" tm
93 // instead of having n equal sample values in a row like:
97 // we compare the current delta change with the next delta change.
99 // if curr_delta is positive,
101 // - and next delta is smaller (or changing direction):
105 // - and next delta significant (at least twice) bigger:
109 // - and next delta is nearly same:
114 // if curr_delta is negative,
116 // - and next delta is smaller (or changing direction):
120 // - and next delta significant (at least twice) bigger:
124 // - and next delta is nearly same:
129 static void InterpolateUp(int *SB, int sinc)
131 if(SB[32]==1) // flag == 1? calc step and set flag... and don't change the value in this pass
133 const int id1=SB[30]-SB[29]; // curr delta to next val
134 const int id2=SB[31]-SB[30]; // and next delta to next-next val :)
138 if(id1>0) // curr delta positive
141 {SB[28]=id1;SB[32]=2;}
144 SB[28]=(id1*sinc)>>16;
146 SB[28]=(id1*sinc)>>17;
148 else // curr delta negative
151 {SB[28]=id1;SB[32]=2;}
154 SB[28]=(id1*sinc)>>16;
156 SB[28]=(id1*sinc)>>17;
160 if(SB[32]==2) // flag 1: calc step and set flag... and don't change the value in this pass
164 SB[28]=(SB[28]*sinc)>>17;
166 // SB[29]=SB[30]-(SB[28]*((0x10000/sinc)-1));
170 else // no flags? add bigger val (if possible), calc smaller step, set flag1
175 // even easier interpolation on downsampling, also no special filter, again just "Pete's common sense" tm
178 static void InterpolateDown(int *SB, int sinc)
180 if(sinc>=0x20000L) // we would skip at least one val?
182 SB[29]+=(SB[30]-SB[29])/2; // add easy weight
183 if(sinc>=0x30000L) // we would skip even more vals?
184 SB[29]+=(SB[31]-SB[30])/2; // add additional next weight
188 ////////////////////////////////////////////////////////////////////////
189 // helpers for gauss interpolation
191 #define gval0 (((short*)(&SB[29]))[gpos&3])
192 #define gval(x) ((int)((short*)(&SB[29]))[(gpos+x)&3])
196 ////////////////////////////////////////////////////////////////////////
200 static void do_irq(void)
202 //if(!(spu.spuStat & STAT_IRQ))
204 spu.spuStat |= STAT_IRQ; // asserted status?
205 if(spu.irqCallback) spu.irqCallback();
209 static int check_irq(int ch, unsigned char *pos)
211 if((spu.spuCtrl & (CTRL_ON|CTRL_IRQ)) == (CTRL_ON|CTRL_IRQ) && pos == spu.pSpuIrq)
213 //printf("ch%d irq %04x\n", ch, pos - spu.spuMemC);
220 void check_irq_io(unsigned int addr)
222 unsigned int irq_addr = regAreaGet(H_SPUirqAddr) << 3;
224 if((spu.spuCtrl & (CTRL_ON|CTRL_IRQ)) == (CTRL_ON|CTRL_IRQ) && addr == irq_addr)
226 //printf("io irq %04x\n", irq_addr);
231 ////////////////////////////////////////////////////////////////////////
232 // START SOUND... called by main thread to setup a new sound on a channel
233 ////////////////////////////////////////////////////////////////////////
235 static void StartSoundSB(int *SB)
237 SB[26]=0; // init mixing vars
241 SB[29]=0; // init our interpolation helpers
246 static void StartSoundMain(int ch)
248 SPUCHAN *s_chan = &spu.s_chan[ch];
253 s_chan->prevflags = 2;
256 s_chan->bStarting = 1;
258 s_chan->pCurr = spu.spuMemC + ((regAreaGetCh(ch, 6) & ~1) << 3);
260 spu.dwNewChannel&=~(1<<ch); // clear new channel bit
261 spu.dwChannelDead&=~(1<<ch);
262 spu.dwChannelsAudible|=1<<ch;
265 static void StartSound(int ch)
268 StartSoundSB(spu.SB + ch * SB_SIZE);
271 ////////////////////////////////////////////////////////////////////////
272 // ALL KIND OF HELPERS
273 ////////////////////////////////////////////////////////////////////////
275 INLINE int FModChangeFrequency(int *SB, int pitch, int ns)
277 pitch = (signed short)pitch;
278 pitch = ((32768 + iFMod[ns]) * pitch) >> 15;
284 SB[32] = 1; // reset interpolation
289 ////////////////////////////////////////////////////////////////////////
291 INLINE void StoreInterpolationVal(int *SB, int sinc, int fa, int fmod_freq)
293 if(fmod_freq) // fmod freq channel
299 if(spu_config.iUseInterpolation>=2) // gauss/cubic interpolation
307 if(spu_config.iUseInterpolation==1) // simple interpolation
310 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'
313 SB[32] = 1; // -> flag: calc new interolation
315 else SB[29]=fa; // no interpolation
319 ////////////////////////////////////////////////////////////////////////
321 INLINE int iGetInterpolationVal(int *SB, int sinc, int spos, int fmod_freq)
325 if(fmod_freq) return SB[29];
327 switch(spu_config.iUseInterpolation)
329 //--------------------------------------------------//
330 case 3: // cubic interpolation
336 fa = gval(3) - 3*gval(2) + 3*gval(1) - gval0;
337 fa *= (xd - (2<<15)) / 6;
339 fa += gval(2) - gval(1) - gval(1) + gval0;
340 fa *= (xd - (1<<15)) >> 1;
342 fa += gval(1) - gval0;
348 //--------------------------------------------------//
349 case 2: // gauss interpolation
352 vl = (spos >> 6) & ~3;
354 vr=(gauss[vl]*(int)gval0) >> 15;
355 vr+=(gauss[vl+1]*gval(1)) >> 15;
356 vr+=(gauss[vl+2]*gval(2)) >> 15;
357 vr+=(gauss[vl+3]*gval(3)) >> 15;
360 //--------------------------------------------------//
361 case 1: // simple interpolation
363 if(sinc<0x10000L) // -> upsampling?
364 InterpolateUp(SB, sinc); // --> interpolate up
365 else InterpolateDown(SB, sinc); // --> else down
368 //--------------------------------------------------//
369 default: // no interpolation
373 //--------------------------------------------------//
379 static void decode_block_data(int *dest, const unsigned char *src, int predict_nr, int shift_factor)
381 static const int f[16][2] = {
389 int fa, s_1, s_2, d, s;
394 for (nSample = 0; nSample < 28; src++)
397 s = (int)(signed short)((d & 0x0f) << 12);
399 fa = s >> shift_factor;
400 fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6);
404 dest[nSample++] = fa;
406 s = (int)(signed short)((d & 0xf0) << 8);
407 fa = s >> shift_factor;
408 fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6);
412 dest[nSample++] = fa;
416 static int decode_block(void *unused, int ch, int *SB)
418 SPUCHAN *s_chan = &spu.s_chan[ch];
419 unsigned char *start;
420 int predict_nr, shift_factor, flags;
423 start = s_chan->pCurr; // set up the current pos
424 if (start - spu.spuMemC < 0x1000) { // ?
425 //log_unhandled("ch%02d plays decode bufs @%05lx\n",
426 // ch, (long)(start - spu.spuMemC));
430 if (s_chan->prevflags & 1) // 1: stop/loop
432 if (!(s_chan->prevflags & 2))
435 start = s_chan->pLoop;
438 check_irq(ch, start);
440 predict_nr = start[0];
441 shift_factor = predict_nr & 0xf;
444 decode_block_data(SB, start + 2, predict_nr, shift_factor);
447 if (flags & 4 && !s_chan->bIgnoreLoop)
448 s_chan->pLoop = start; // loop adress
452 s_chan->pCurr = start; // store values for next cycle
453 s_chan->prevflags = flags;
454 s_chan->bStarting = 0;
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;
484 s_chan->bStarting = 0;
489 // if irq is going to trigger sooner than in upd_samples, set upd_samples
490 static void scan_for_irq(int ch, unsigned int *upd_samples)
492 SPUCHAN *s_chan = &spu.s_chan[ch];
493 int pos, sinc, sinc_inv, end;
494 unsigned char *block;
497 block = s_chan->pCurr;
500 end = pos + *upd_samples * sinc;
501 if (s_chan->prevflags & 1) // 1: stop/loop
502 block = s_chan->pLoop;
504 pos += (28 - s_chan->iSBPos) << 16;
507 if (block == spu.pSpuIrq)
511 if (flags & 1) { // 1: stop/loop
512 block = s_chan->pLoop;
519 sinc_inv = s_chan->sinc_inv;
521 sinc_inv = s_chan->sinc_inv = (0x80000000u / (uint32_t)sinc) << 1;
524 *upd_samples = (((uint64_t)pos * sinc_inv) >> 32) + 1;
525 //xprintf("ch%02d: irq sched: %3d %03d\n",
526 // ch, *upd_samples, *upd_samples * 60 * 263 / 44100);
530 #define make_do_samples(name, fmod_code, interp_start, interp1_code, interp2_code, interp_end) \
531 static noinline int do_samples_##name( \
532 int (*decode_f)(void *context, int ch, int *SB), void *ctx, \
533 int ch, int ns_to, int *SB, int sinc, int *spos, int *sbpos) \
539 for (ns = 0; ns < ns_to; ns++) \
544 while (*spos >= 0x10000) \
546 fa = SB[(*sbpos)++]; \
550 d = decode_f(ctx, ch, SB); \
567 #define fmod_recv_check \
568 if(spu.s_chan[ch].bFMod==1 && iFMod[ns]) \
569 sinc = FModChangeFrequency(SB, spu.s_chan[ch].iRawPitch, ns)
571 make_do_samples(default, fmod_recv_check, ,
572 StoreInterpolationVal(SB, sinc, fa, spu.s_chan[ch].bFMod==2),
573 ChanBuf[ns] = iGetInterpolationVal(SB, sinc, *spos, spu.s_chan[ch].bFMod==2), )
574 make_do_samples(noint, , fa = SB[29], , ChanBuf[ns] = fa, SB[29] = fa)
576 #define simple_interp_store \
583 #define simple_interp_get \
584 if(sinc<0x10000) /* -> upsampling? */ \
585 InterpolateUp(SB, sinc); /* --> interpolate up */ \
586 else InterpolateDown(SB, sinc); /* --> else down */ \
589 make_do_samples(simple, , ,
590 simple_interp_store, simple_interp_get, )
592 static int do_samples_skip(int ch, int ns_to)
594 SPUCHAN *s_chan = &spu.s_chan[ch];
595 int spos = s_chan->spos;
596 int sinc = s_chan->sinc;
597 int ret = ns_to, ns, d;
599 spos += s_chan->iSBPos << 16;
601 for (ns = 0; ns < ns_to; ns++)
604 while (spos >= 28*0x10000)
613 s_chan->iSBPos = spos >> 16;
614 s_chan->spos = spos & 0xffff;
619 static void do_lsfr_samples(int ns_to, int ctrl,
620 unsigned int *dwNoiseCount, unsigned int *dwNoiseVal)
622 unsigned int counter = *dwNoiseCount;
623 unsigned int val = *dwNoiseVal;
624 unsigned int level, shift, bit;
627 // modified from DrHell/shalma, no fraction
628 level = (ctrl >> 10) & 0x0f;
629 level = 0x8000 >> level;
631 for (ns = 0; ns < ns_to; ns++)
634 if (counter >= level)
637 shift = (val >> 10) & 0x1f;
638 bit = (0x69696969 >> shift) & 1;
639 bit ^= (val >> 15) & 1;
640 val = (val << 1) | bit;
643 ChanBuf[ns] = (signed short)val;
646 *dwNoiseCount = counter;
650 static int do_samples_noise(int ch, int ns_to)
654 ret = do_samples_skip(ch, ns_to);
656 do_lsfr_samples(ns_to, spu.spuCtrl, &spu.dwNoiseCount, &spu.dwNoiseVal);
662 // asm code; lv and rv must be 0-3fff
663 extern void mix_chan(int *SSumLR, int count, int lv, int rv);
664 extern void mix_chan_rvb(int *SSumLR, int count, int lv, int rv, int *rvb);
666 static void mix_chan(int *SSumLR, int count, int lv, int rv)
668 const int *src = ChanBuf;
675 l = (sval * lv) >> 14;
676 r = (sval * rv) >> 14;
682 static void mix_chan_rvb(int *SSumLR, int count, int lv, int rv, int *rvb)
684 const int *src = ChanBuf;
693 l = (sval * lv) >> 14;
694 r = (sval * rv) >> 14;
703 // 0x0800-0x0bff Voice 1
704 // 0x0c00-0x0fff Voice 3
705 static noinline void do_decode_bufs(unsigned short *mem, int which,
706 int count, int decode_pos)
708 unsigned short *dst = &mem[0x800/2 + which*0x400/2];
709 const int *src = ChanBuf;
710 int cursor = decode_pos;
715 dst[cursor] = *src++;
719 // decode_pos is updated and irqs are checked later, after voice loop
722 static void do_silent_chans(int ns_to, int silentch)
728 mask = silentch & 0xffffff;
729 for (ch = 0; mask != 0; ch++, mask >>= 1)
731 if (!(mask & 1)) continue;
732 if (spu.dwChannelDead & (1<<ch)) continue;
734 s_chan = &spu.s_chan[ch];
735 if (s_chan->pCurr > spu.pSpuIrq && s_chan->pLoop > spu.pSpuIrq)
738 s_chan->spos += s_chan->iSBPos << 16;
741 s_chan->spos += s_chan->sinc * ns_to;
742 while (s_chan->spos >= 28 * 0x10000)
744 unsigned char *start = s_chan->pCurr;
747 if (start == s_chan->pCurr || start - spu.spuMemC < 0x1000)
749 // looping on self or stopped(?)
750 spu.dwChannelDead |= 1<<ch;
755 s_chan->spos -= 28 * 0x10000;
760 static void do_channels(int ns_to)
767 do_rvb = spu.rvb->StartAddr && spu_config.iUseReverb;
769 memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2);
771 mask = spu.dwNewChannel & 0xffffff;
772 for (ch = 0; mask != 0; ch++, mask >>= 1) {
777 mask = spu.dwChannelsAudible & 0xffffff;
778 for (ch = 0; mask != 0; ch++, mask >>= 1) // loop em all...
780 if (!(mask & 1)) continue; // channel not playing? next
782 s_chan = &spu.s_chan[ch];
783 SB = spu.SB + ch * SB_SIZE;
785 if (spu.s_chan[ch].bNewPitch)
786 SB[32] = 1; // reset interpolation
787 spu.s_chan[ch].bNewPitch = 0;
790 d = do_samples_noise(ch, ns_to);
791 else if (s_chan->bFMod == 2
792 || (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 0))
793 d = do_samples_noint(decode_block, NULL, ch, ns_to,
794 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
795 else if (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 1)
796 d = do_samples_simple(decode_block, NULL, ch, ns_to,
797 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
799 d = do_samples_default(decode_block, NULL, ch, ns_to,
800 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
802 if (!s_chan->bStarting) {
803 d = MixADSR(&s_chan->ADSRX, d);
805 spu.dwChannelsAudible &= ~(1 << ch);
806 s_chan->ADSRX.State = ADSR_RELEASE;
807 s_chan->ADSRX.EnvelopeVol = 0;
808 memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0]));
812 if (ch == 1 || ch == 3)
814 do_decode_bufs(spu.spuMem, ch/2, ns_to, spu.decode_pos);
815 spu.decode_dirty_ch |= 1 << ch;
818 if (s_chan->bFMod == 2) // fmod freq channel
819 memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
820 if (s_chan->bRVBActive && do_rvb)
821 mix_chan_rvb(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume, RVB);
823 mix_chan(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume);
826 MixXA(spu.SSumLR, RVB, ns_to, spu.decode_pos);
828 if (spu.rvb->StartAddr) {
830 REVERBDo(spu.SSumLR, RVB, ns_to, spu.rvb->CurrAddr);
832 spu.rvb->CurrAddr += ns_to / 2;
833 while (spu.rvb->CurrAddr >= 0x40000)
834 spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr;
838 static void do_samples_finish(int *SSumLR, int ns_to,
839 int silentch, int decode_pos);
841 // optional worker thread handling
843 #if P_HAVE_PTHREAD || defined(WANT_THREAD_CODE)
845 // worker thread state
846 static struct spu_worker {
849 unsigned int exit_thread;
850 unsigned int i_ready;
851 unsigned int i_reaped;
852 unsigned int last_boot_cnt; // dsp
853 unsigned int ram_dirty;
855 // aligning for C64X_DSP
856 unsigned int _pad0[128/4];
861 unsigned int active; // dsp
862 unsigned int boot_cnt;
864 unsigned int _pad1[128/4];
871 unsigned int channels_new;
872 unsigned int channels_on;
873 unsigned int channels_silent;
882 unsigned short ns_to;
883 unsigned short bNoise:1;
884 unsigned short bFMod:2;
885 unsigned short bRVBActive:1;
886 unsigned short bNewPitch:1;
889 int SSumLR[NSSIZE * 2];
893 #define WORK_MAXCNT (sizeof(worker->i) / sizeof(worker->i[0]))
894 #define WORK_I_MASK (WORK_MAXCNT - 1)
896 static void thread_work_start(void);
897 static void thread_work_wait_sync(struct work_item *work, int force);
898 static void thread_sync_caches(void);
899 static int thread_get_i_done(void);
901 static int decode_block_work(void *context, int ch, int *SB)
903 const unsigned char *ram = spu.spuMemC;
904 int predict_nr, shift_factor, flags;
905 struct work_item *work = context;
906 int start = work->ch[ch].start;
907 int loop = work->ch[ch].loop;
909 predict_nr = ram[start];
910 shift_factor = predict_nr & 0xf;
913 decode_block_data(SB, ram + start + 2, predict_nr, shift_factor);
915 flags = ram[start + 1];
917 loop = start; // loop adress
921 if (flags & 1) // 1: stop/loop
924 work->ch[ch].start = start & 0x7ffff;
925 work->ch[ch].loop = loop;
930 static void queue_channel_work(int ns_to, unsigned int silentch)
932 struct work_item *work;
937 work = &worker->i[worker->i_ready & WORK_I_MASK];
939 work->ctrl = spu.spuCtrl;
940 work->decode_pos = spu.decode_pos;
941 work->channels_silent = silentch;
943 mask = work->channels_new = spu.dwNewChannel & 0xffffff;
944 for (ch = 0; mask != 0; ch++, mask >>= 1) {
949 mask = work->channels_on = spu.dwChannelsAudible & 0xffffff;
950 spu.decode_dirty_ch |= mask & 0x0a;
952 for (ch = 0; mask != 0; ch++, mask >>= 1)
954 if (!(mask & 1)) continue;
956 s_chan = &spu.s_chan[ch];
957 work->ch[ch].spos = s_chan->spos;
958 work->ch[ch].sbpos = s_chan->iSBPos;
959 work->ch[ch].sinc = s_chan->sinc;
960 work->ch[ch].adsr = s_chan->ADSRX;
961 work->ch[ch].vol_l = s_chan->iLeftVolume;
962 work->ch[ch].vol_r = s_chan->iRightVolume;
963 work->ch[ch].start = s_chan->pCurr - spu.spuMemC;
964 work->ch[ch].loop = s_chan->pLoop - spu.spuMemC;
965 work->ch[ch].bNoise = s_chan->bNoise;
966 work->ch[ch].bFMod = s_chan->bFMod;
967 work->ch[ch].bRVBActive = s_chan->bRVBActive;
968 work->ch[ch].bNewPitch = s_chan->bNewPitch;
969 if (s_chan->prevflags & 1)
970 work->ch[ch].start = work->ch[ch].loop;
972 d = do_samples_skip(ch, ns_to);
973 work->ch[ch].ns_to = d;
975 if (!s_chan->bStarting) {
976 // note: d is not accurate on skip
977 d = SkipADSR(&s_chan->ADSRX, d);
979 spu.dwChannelsAudible &= ~(1 << ch);
980 s_chan->ADSRX.State = ADSR_RELEASE;
981 s_chan->ADSRX.EnvelopeVol = 0;
984 s_chan->bNewPitch = 0;
988 if (spu.rvb->StartAddr) {
989 if (spu_config.iUseReverb)
990 work->rvb_addr = spu.rvb->CurrAddr;
992 spu.rvb->CurrAddr += ns_to / 2;
993 while (spu.rvb->CurrAddr >= 0x40000)
994 spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr;
1001 static void do_channel_work(struct work_item *work)
1004 int *SB, sinc, spos, sbpos;
1007 ns_to = work->ns_to;
1010 memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2);
1012 mask = work->channels_new;
1013 for (ch = 0; mask != 0; ch++, mask >>= 1) {
1015 StartSoundSB(spu.SB + ch * SB_SIZE);
1018 mask = work->channels_on;
1019 for (ch = 0; mask != 0; ch++, mask >>= 1)
1021 if (!(mask & 1)) continue;
1023 d = work->ch[ch].ns_to;
1024 spos = work->ch[ch].spos;
1025 sbpos = work->ch[ch].sbpos;
1026 sinc = work->ch[ch].sinc;
1028 SB = spu.SB + ch * SB_SIZE;
1029 if (work->ch[ch].bNewPitch)
1030 SB[32] = 1; // reset interpolation
1032 if (work->ch[ch].bNoise)
1033 do_lsfr_samples(d, work->ctrl, &spu.dwNoiseCount, &spu.dwNoiseVal);
1034 else if (work->ch[ch].bFMod == 2
1035 || (work->ch[ch].bFMod == 0 && spu_config.iUseInterpolation == 0))
1036 do_samples_noint(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1037 else if (work->ch[ch].bFMod == 0 && spu_config.iUseInterpolation == 1)
1038 do_samples_simple(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1040 do_samples_default(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1042 d = MixADSR(&work->ch[ch].adsr, d);
1044 work->ch[ch].adsr.EnvelopeVol = 0;
1045 memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0]));
1048 if (ch == 1 || ch == 3)
1049 do_decode_bufs(spu.spuMem, ch/2, ns_to, work->decode_pos);
1051 if (work->ch[ch].bFMod == 2) // fmod freq channel
1052 memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
1053 if (work->ch[ch].bRVBActive && work->rvb_addr)
1054 mix_chan_rvb(work->SSumLR, ns_to,
1055 work->ch[ch].vol_l, work->ch[ch].vol_r, RVB);
1057 mix_chan(work->SSumLR, ns_to, work->ch[ch].vol_l, work->ch[ch].vol_r);
1061 REVERBDo(work->SSumLR, RVB, ns_to, work->rvb_addr);
1064 static void sync_worker_thread(int force)
1066 struct work_item *work;
1067 int done, used_space;
1069 // rvb offsets will change, thread may be using them
1070 force |= spu.rvb->dirty && spu.rvb->StartAddr;
1072 done = thread_get_i_done() - worker->i_reaped;
1073 used_space = worker->i_ready - worker->i_reaped;
1075 //printf("done: %d use: %d dsp: %u/%u\n", done, used_space,
1076 // worker->boot_cnt, worker->i_done);
1078 while ((force && used_space > 0) || used_space >= WORK_MAXCNT || done > 0) {
1079 work = &worker->i[worker->i_reaped & WORK_I_MASK];
1080 thread_work_wait_sync(work, force);
1082 MixXA(work->SSumLR, RVB, work->ns_to, work->decode_pos);
1083 do_samples_finish(work->SSumLR, work->ns_to,
1084 work->channels_silent, work->decode_pos);
1087 done = thread_get_i_done() - worker->i_reaped;
1088 used_space = worker->i_ready - worker->i_reaped;
1091 thread_sync_caches();
1096 static void queue_channel_work(int ns_to, int silentch) {}
1097 static void sync_worker_thread(int force) {}
1099 static const void * const worker = NULL;
1101 #endif // P_HAVE_PTHREAD || defined(WANT_THREAD_CODE)
1103 ////////////////////////////////////////////////////////////////////////
1104 // MAIN SPU FUNCTION
1105 // here is the main job handler...
1106 ////////////////////////////////////////////////////////////////////////
1108 void do_samples(unsigned int cycles_to, int do_direct)
1110 unsigned int silentch;
1114 cycle_diff = cycles_to - spu.cycles_played;
1115 if (cycle_diff < -2*1048576 || cycle_diff > 2*1048576)
1117 //xprintf("desync %u %d\n", cycles_to, cycle_diff);
1118 spu.cycles_played = cycles_to;
1122 silentch = ~(spu.dwChannelsAudible | spu.dwNewChannel) & 0xffffff;
1124 do_direct |= (silentch == 0xffffff);
1126 sync_worker_thread(do_direct);
1128 if (cycle_diff < 2 * 768)
1131 ns_to = (cycle_diff / 768 + 1) & ~1;
1132 if (ns_to > NSSIZE) {
1133 // should never happen
1134 //xprintf("ns_to oflow %d %d\n", ns_to, NSSIZE);
1138 //////////////////////////////////////////////////////
1139 // special irq handling in the decode buffers (0x0000-0x1000)
1141 // the decode buffers are located in spu memory in the following way:
1142 // 0x0000-0x03ff CD audio left
1143 // 0x0400-0x07ff CD audio right
1144 // 0x0800-0x0bff Voice 1
1145 // 0x0c00-0x0fff Voice 3
1146 // and decoded data is 16 bit for one sample
1148 // even if voices 1/3 are off or no cd audio is playing, the internal
1149 // play positions will move on and wrap after 0x400 bytes.
1150 // Therefore: we just need a pointer from spumem+0 to spumem+3ff, and
1151 // increase this pointer on each sample by 2 bytes. If this pointer
1152 // (or 0x400 offsets of this pointer) hits the spuirq address, we generate
1155 if (unlikely((spu.spuCtrl & CTRL_IRQ)
1156 && spu.pSpuIrq < spu.spuMemC+0x1000))
1158 int irq_pos = (spu.pSpuIrq - spu.spuMemC) / 2 & 0x1ff;
1159 int left = (irq_pos - spu.decode_pos) & 0x1ff;
1160 if (0 < left && left <= ns_to)
1162 //xprintf("decoder irq %x\n", spu.decode_pos);
1166 if (!spu.cycles_dma_end || (int)(spu.cycles_dma_end - cycles_to) < 0) {
1167 spu.cycles_dma_end = 0;
1168 check_irq_io(spu.spuAddr);
1171 if (unlikely(spu.rvb->dirty))
1174 if (do_direct || worker == NULL || !spu_config.iUseThread) {
1176 do_samples_finish(spu.SSumLR, ns_to, silentch, spu.decode_pos);
1179 queue_channel_work(ns_to, silentch);
1180 //sync_worker_thread(1); // uncomment for debug
1183 // advance "stopped" channels that can cause irqs
1184 // (all chans are always playing on the real thing..)
1185 if (spu.spuCtrl & CTRL_IRQ)
1186 do_silent_chans(ns_to, silentch);
1188 spu.cycles_played += ns_to * 768;
1189 spu.decode_pos = (spu.decode_pos + ns_to) & 0x1ff;
1191 static int ccount; static time_t ctime; ccount++;
1192 if (time(NULL) != ctime)
1193 { printf("%d\n", ccount); ccount = 0; ctime = time(NULL); }
1197 static void do_samples_finish(int *SSumLR, int ns_to,
1198 int silentch, int decode_pos)
1200 int vol_l = ((int)regAreaGet(H_SPUmvolL) << 17) >> 17;
1201 int vol_r = ((int)regAreaGet(H_SPUmvolR) << 17) >> 17;
1205 // must clear silent channel decode buffers
1206 if(unlikely(silentch & spu.decode_dirty_ch & (1<<1)))
1208 memset(&spu.spuMem[0x800/2], 0, 0x400);
1209 spu.decode_dirty_ch &= ~(1<<1);
1211 if(unlikely(silentch & spu.decode_dirty_ch & (1<<3)))
1213 memset(&spu.spuMem[0xc00/2], 0, 0x400);
1214 spu.decode_dirty_ch &= ~(1<<3);
1217 vol_l = vol_l * spu_config.iVolume >> 10;
1218 vol_r = vol_r * spu_config.iVolume >> 10;
1220 if (!(spu.spuCtrl & CTRL_MUTE) || !(vol_l | vol_r))
1223 memset(spu.pS, 0, ns_to * 2 * sizeof(spu.pS[0]));
1224 memset(SSumLR, 0, ns_to * 2 * sizeof(SSumLR[0]));
1225 spu.pS += ns_to * 2;
1228 for (ns = 0; ns < ns_to * 2; )
1230 d = SSumLR[ns]; SSumLR[ns] = 0;
1231 d = d * vol_l >> 14;
1236 d = SSumLR[ns]; SSumLR[ns] = 0;
1237 d = d * vol_r >> 14;
1244 void schedule_next_irq(void)
1246 unsigned int upd_samples;
1249 if (spu.scheduleCallback == NULL)
1252 upd_samples = 44100 / 50;
1254 for (ch = 0; ch < MAXCHAN; ch++)
1256 if (spu.dwChannelDead & (1 << ch))
1258 if ((unsigned long)(spu.pSpuIrq - spu.s_chan[ch].pCurr) > IRQ_NEAR_BLOCKS * 16
1259 && (unsigned long)(spu.pSpuIrq - spu.s_chan[ch].pLoop) > IRQ_NEAR_BLOCKS * 16)
1261 if (spu.s_chan[ch].sinc == 0)
1264 scan_for_irq(ch, &upd_samples);
1267 if (unlikely(spu.pSpuIrq < spu.spuMemC + 0x1000))
1269 int irq_pos = (spu.pSpuIrq - spu.spuMemC) / 2 & 0x1ff;
1270 int left = (irq_pos - spu.decode_pos) & 0x1ff;
1271 if (0 < left && left < upd_samples) {
1272 //xprintf("decode: %3d (%3d/%3d)\n", left, spu.decode_pos, irq_pos);
1277 if (upd_samples < 44100 / 50)
1278 spu.scheduleCallback(upd_samples * 768);
1281 // SPU ASYNC... even newer epsxe func
1282 // 1 time every 'cycle' cycles... harhar
1284 // rearmed: called dynamically now
1286 void CALLBACK SPUasync(unsigned int cycle, unsigned int flags)
1288 do_samples(cycle, 0);
1290 if (spu.spuCtrl & CTRL_IRQ)
1291 schedule_next_irq();
1294 out_current->feed(spu.pSpuBuffer, (unsigned char *)spu.pS - spu.pSpuBuffer);
1295 spu.pS = (short *)spu.pSpuBuffer;
1297 if (spu_config.iTempo) {
1298 if (!out_current->busy())
1299 // cause more samples to be generated
1300 // (and break some games because of bad sync)
1301 spu.cycles_played -= 44100 / 60 / 2 * 768;
1306 // SPU UPDATE... new epsxe func
1307 // 1 time every 32 hsync lines
1308 // (312/32)x50 in pal
1309 // (262/32)x60 in ntsc
1311 // since epsxe 1.5.2 (linux) uses SPUupdate, not SPUasync, I will
1312 // leave that func in the linux port, until epsxe linux is using
1313 // the async function as well
1315 void CALLBACK SPUupdate(void)
1321 void CALLBACK SPUplayADPCMchannel(xa_decode_t *xap, unsigned int cycle, int is_start)
1324 if(!xap->freq) return; // no xa freq ? bye
1327 do_samples(cycle, 1); // catch up to prevent source underflows later
1329 FeedXA(xap); // call main XA feeder
1330 spu.xapGlobal = xap; // store info for save states
1334 int CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes, unsigned int cycle, int is_start)
1336 if (!pcm) return -1;
1337 if (nbytes<=0) return -1;
1340 do_samples(cycle, 1); // catch up to prevent source underflows later
1342 FeedCDDA((unsigned char *)pcm, nbytes);
1346 // to be called after state load
1347 void ClearWorkingState(void)
1349 memset(iFMod, 0, sizeof(iFMod));
1350 spu.pS=(short *)spu.pSpuBuffer; // setup soundbuffer pointer
1353 // SETUPSTREAMS: init most of the spu buffers
1354 static void SetupStreams(void)
1356 spu.pSpuBuffer = (unsigned char *)malloc(32768); // alloc mixing buffer
1357 spu.SSumLR = calloc(NSSIZE * 2, sizeof(spu.SSumLR[0]));
1359 spu.XAStart = malloc(44100 * sizeof(uint32_t)); // alloc xa buffer
1360 spu.XAEnd = spu.XAStart + 44100;
1361 spu.XAPlay = spu.XAStart;
1362 spu.XAFeed = spu.XAStart;
1364 spu.CDDAStart = malloc(CDDA_BUFFER_SIZE); // alloc cdda buffer
1365 spu.CDDAEnd = spu.CDDAStart + CDDA_BUFFER_SIZE / sizeof(uint32_t);
1366 spu.CDDAPlay = spu.CDDAStart;
1367 spu.CDDAFeed = spu.CDDAStart;
1369 ClearWorkingState();
1372 // REMOVESTREAMS: free most buffer
1373 static void RemoveStreams(void)
1375 free(spu.pSpuBuffer); // free mixing buffer
1376 spu.pSpuBuffer = NULL;
1379 free(spu.XAStart); // free XA buffer
1381 free(spu.CDDAStart); // free CDDA buffer
1382 spu.CDDAStart = NULL;
1385 #if defined(C64X_DSP)
1387 /* special code for TI C64x DSP */
1388 #include "spu_c64x.c"
1390 #elif P_HAVE_PTHREAD
1392 #include <pthread.h>
1393 #include <semaphore.h>
1402 /* generic pthread implementation */
1404 static void thread_work_start(void)
1406 sem_post(&t.sem_avail);
1409 static void thread_work_wait_sync(struct work_item *work, int force)
1411 sem_wait(&t.sem_done);
1414 static int thread_get_i_done(void)
1416 return worker->i_done;
1419 static void thread_sync_caches(void)
1423 static void *spu_worker_thread(void *unused)
1425 struct work_item *work;
1428 sem_wait(&t.sem_avail);
1429 if (worker->exit_thread)
1432 work = &worker->i[worker->i_done & WORK_I_MASK];
1433 do_channel_work(work);
1436 sem_post(&t.sem_done);
1442 static void init_spu_thread(void)
1446 if (sysconf(_SC_NPROCESSORS_ONLN) <= 1)
1449 worker = calloc(1, sizeof(*worker));
1452 ret = sem_init(&t.sem_avail, 0, 0);
1454 goto fail_sem_avail;
1455 ret = sem_init(&t.sem_done, 0, 0);
1459 ret = pthread_create(&t.thread, NULL, spu_worker_thread, NULL);
1463 spu_config.iThreadAvail = 1;
1467 sem_destroy(&t.sem_done);
1469 sem_destroy(&t.sem_avail);
1473 spu_config.iThreadAvail = 0;
1476 static void exit_spu_thread(void)
1480 worker->exit_thread = 1;
1481 sem_post(&t.sem_avail);
1482 pthread_join(t.thread, NULL);
1483 sem_destroy(&t.sem_done);
1484 sem_destroy(&t.sem_avail);
1489 #else // if !P_HAVE_PTHREAD
1491 static void init_spu_thread(void)
1495 static void exit_spu_thread(void)
1501 // SPUINIT: this func will be called first by the main emu
1502 long CALLBACK SPUinit(void)
1506 memset(&spu, 0, sizeof(spu));
1507 spu.spuMemC = calloc(1, 512 * 1024);
1510 spu.s_chan = calloc(MAXCHAN+1, sizeof(spu.s_chan[0])); // channel + 1 infos (1 is security for fmod handling)
1511 spu.rvb = calloc(1, sizeof(REVERBInfo));
1512 spu.SB = calloc(MAXCHAN, sizeof(spu.SB[0]) * SB_SIZE);
1516 spu.pSpuIrq = spu.spuMemC;
1518 SetupStreams(); // prepare streaming
1520 if (spu_config.iVolume == 0)
1521 spu_config.iVolume = 768; // 1024 is 1.0
1525 for (i = 0; i < MAXCHAN; i++) // loop sound channels
1527 spu.s_chan[i].ADSRX.SustainLevel = 0xf; // -> init sustain
1528 spu.s_chan[i].ADSRX.SustainIncrease = 1;
1529 spu.s_chan[i].pLoop = spu.spuMemC;
1530 spu.s_chan[i].pCurr = spu.spuMemC;
1531 spu.s_chan[i].bIgnoreLoop = 0;
1534 spu.bSpuInit=1; // flag: we are inited
1539 // SPUOPEN: called by main emu after init
1540 long CALLBACK SPUopen(void)
1542 if (spu.bSPUIsOpen) return 0; // security for some stupid main emus
1544 SetupSound(); // setup sound (before init!)
1548 return PSE_SPU_ERR_SUCCESS;
1551 // SPUCLOSE: called before shutdown
1552 long CALLBACK SPUclose(void)
1554 if (!spu.bSPUIsOpen) return 0; // some security
1556 spu.bSPUIsOpen = 0; // no more open
1558 out_current->finish(); // no more sound handling
1563 // SPUSHUTDOWN: called by main emu on final exit
1564 long CALLBACK SPUshutdown(void)
1579 RemoveStreams(); // no more streaming
1586 // this functions will be called once,
1587 // passes a callback that should be called on SPU-IRQ/cdda volume change
1588 void CALLBACK SPUregisterCallback(void (CALLBACK *callback)(void))
1590 spu.irqCallback = callback;
1593 void CALLBACK SPUregisterCDDAVolume(void (CALLBACK *CDDAVcallback)(short, short))
1595 spu.cddavCallback = CDDAVcallback;
1598 void CALLBACK SPUregisterScheduleCb(void (CALLBACK *callback)(unsigned int))
1600 spu.scheduleCallback = callback;
1603 // COMMON PLUGIN INFO FUNCS
1605 char * CALLBACK PSEgetLibName(void)
1607 return _(libraryName);
1610 unsigned long CALLBACK PSEgetLibType(void)
1615 unsigned long CALLBACK PSEgetLibVersion(void)
1617 return (1 << 16) | (6 << 8);
1620 char * SPUgetLibInfos(void)
1622 return _(libraryInfo);
1627 void spu_get_debug_info(int *chans_out, int *run_chans, int *fmod_chans_out, int *noise_chans_out)
1629 int ch = 0, fmod_chans = 0, noise_chans = 0, irq_chans = 0;
1631 if (spu.s_chan == NULL)
1634 for(;ch<MAXCHAN;ch++)
1636 if (!(spu.dwChannelsAudible & (1<<ch)))
1638 if (spu.s_chan[ch].bFMod == 2)
1639 fmod_chans |= 1 << ch;
1640 if (spu.s_chan[ch].bNoise)
1641 noise_chans |= 1 << ch;
1642 if((spu.spuCtrl&CTRL_IRQ) && spu.s_chan[ch].pCurr <= spu.pSpuIrq && spu.s_chan[ch].pLoop <= spu.pSpuIrq)
1643 irq_chans |= 1 << ch;
1646 *chans_out = spu.dwChannelsAudible;
1647 *run_chans = ~spu.dwChannelsAudible & ~spu.dwChannelDead & irq_chans;
1648 *fmod_chans_out = fmod_chans;
1649 *noise_chans_out = noise_chans;
1652 // vim:shiftwidth=1:expandtab