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];
257 s_chan->pCurr = spu.spuMemC + ((regAreaGetCh(ch, 6) & ~1) << 3);
259 spu.dwNewChannel&=~(1<<ch); // clear new channel bit
260 spu.dwChannelDead&=~(1<<ch);
261 spu.dwChannelsAudible|=1<<ch;
264 static void StartSound(int ch)
267 StartSoundSB(spu.SB + ch * SB_SIZE);
270 ////////////////////////////////////////////////////////////////////////
271 // ALL KIND OF HELPERS
272 ////////////////////////////////////////////////////////////////////////
274 INLINE int FModChangeFrequency(int *SB, int pitch, int ns)
276 unsigned int NP=pitch;
279 NP=((32768L+iFMod[ns])*NP)>>15;
281 if(NP>0x3fff) NP=0x3fff;
284 sinc=NP<<4; // calc frequency
286 SB[32]=1; // reset interpolation
291 ////////////////////////////////////////////////////////////////////////
293 INLINE void StoreInterpolationVal(int *SB, int sinc, int fa, int fmod_freq)
295 if(fmod_freq) // fmod freq channel
301 if(spu_config.iUseInterpolation>=2) // gauss/cubic interpolation
309 if(spu_config.iUseInterpolation==1) // simple interpolation
312 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'
315 SB[32] = 1; // -> flag: calc new interolation
317 else SB[29]=fa; // no interpolation
321 ////////////////////////////////////////////////////////////////////////
323 INLINE int iGetInterpolationVal(int *SB, int sinc, int spos, int fmod_freq)
327 if(fmod_freq) return SB[29];
329 switch(spu_config.iUseInterpolation)
331 //--------------------------------------------------//
332 case 3: // cubic interpolation
338 fa = gval(3) - 3*gval(2) + 3*gval(1) - gval0;
339 fa *= (xd - (2<<15)) / 6;
341 fa += gval(2) - gval(1) - gval(1) + gval0;
342 fa *= (xd - (1<<15)) >> 1;
344 fa += gval(1) - gval0;
350 //--------------------------------------------------//
351 case 2: // gauss interpolation
354 vl = (spos >> 6) & ~3;
356 vr=(gauss[vl]*(int)gval0) >> 15;
357 vr+=(gauss[vl+1]*gval(1)) >> 15;
358 vr+=(gauss[vl+2]*gval(2)) >> 15;
359 vr+=(gauss[vl+3]*gval(3)) >> 15;
362 //--------------------------------------------------//
363 case 1: // simple interpolation
365 if(sinc<0x10000L) // -> upsampling?
366 InterpolateUp(SB, sinc); // --> interpolate up
367 else InterpolateDown(SB, sinc); // --> else down
370 //--------------------------------------------------//
371 default: // no interpolation
375 //--------------------------------------------------//
381 static void decode_block_data(int *dest, const unsigned char *src, int predict_nr, int shift_factor)
383 static const int f[16][2] = {
391 int fa, s_1, s_2, d, s;
396 for (nSample = 0; nSample < 28; src++)
399 s = (int)(signed short)((d & 0x0f) << 12);
401 fa = s >> shift_factor;
402 fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6);
405 dest[nSample++] = fa;
407 s = (int)(signed short)((d & 0xf0) << 8);
408 fa = s >> shift_factor;
409 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) // ?
427 if (s_chan->prevflags & 1) // 1: stop/loop
429 if (!(s_chan->prevflags & 2))
432 start = s_chan->pLoop;
435 check_irq(ch, start);
437 predict_nr = start[0];
438 shift_factor = predict_nr & 0xf;
441 decode_block_data(SB, start + 2, predict_nr, shift_factor);
444 if (flags & 4 && !s_chan->bIgnoreLoop)
445 s_chan->pLoop = start; // loop adress
449 s_chan->pCurr = start; // store values for next cycle
450 s_chan->prevflags = flags;
455 // do block, but ignore sample data
456 static int skip_block(int ch)
458 SPUCHAN *s_chan = &spu.s_chan[ch];
459 unsigned char *start = s_chan->pCurr;
463 if (s_chan->prevflags & 1) {
464 if (!(s_chan->prevflags & 2))
467 start = s_chan->pLoop;
470 check_irq(ch, start);
473 if (flags & 4 && !s_chan->bIgnoreLoop)
474 s_chan->pLoop = start;
478 s_chan->pCurr = start;
479 s_chan->prevflags = flags;
484 // if irq is going to trigger sooner than in upd_samples, set upd_samples
485 static void scan_for_irq(int ch, unsigned int *upd_samples)
487 SPUCHAN *s_chan = &spu.s_chan[ch];
488 int pos, sinc, sinc_inv, end;
489 unsigned char *block;
492 block = s_chan->pCurr;
495 end = pos + *upd_samples * sinc;
497 pos += (28 - s_chan->iSBPos) << 16;
500 if (block == spu.pSpuIrq)
504 if (flags & 1) { // 1: stop/loop
505 block = s_chan->pLoop;
512 sinc_inv = s_chan->sinc_inv;
514 sinc_inv = s_chan->sinc_inv = (0x80000000u / (uint32_t)sinc) << 1;
517 *upd_samples = (((uint64_t)pos * sinc_inv) >> 32) + 1;
518 //xprintf("ch%02d: irq sched: %3d %03d\n",
519 // ch, *upd_samples, *upd_samples * 60 * 263 / 44100);
523 #define make_do_samples(name, fmod_code, interp_start, interp1_code, interp2_code, interp_end) \
524 static noinline int do_samples_##name( \
525 int (*decode_f)(void *context, int ch, int *SB), void *ctx, \
526 int ch, int ns_to, int *SB, int sinc, int *spos, int *sbpos) \
532 for (ns = 0; ns < ns_to; ns++) \
537 while (*spos >= 0x10000) \
539 fa = SB[(*sbpos)++]; \
543 d = decode_f(ctx, ch, SB); \
560 #define fmod_recv_check \
561 if(spu.s_chan[ch].bFMod==1 && iFMod[ns]) \
562 sinc = FModChangeFrequency(SB, spu.s_chan[ch].iRawPitch, ns)
564 make_do_samples(default, fmod_recv_check, ,
565 StoreInterpolationVal(SB, sinc, fa, spu.s_chan[ch].bFMod==2),
566 ChanBuf[ns] = iGetInterpolationVal(SB, sinc, *spos, spu.s_chan[ch].bFMod==2), )
567 make_do_samples(noint, , fa = SB[29], , ChanBuf[ns] = fa, SB[29] = fa)
569 #define simple_interp_store \
576 #define simple_interp_get \
577 if(sinc<0x10000) /* -> upsampling? */ \
578 InterpolateUp(SB, sinc); /* --> interpolate up */ \
579 else InterpolateDown(SB, sinc); /* --> else down */ \
582 make_do_samples(simple, , ,
583 simple_interp_store, simple_interp_get, )
585 static int do_samples_skip(int ch, int ns_to)
587 SPUCHAN *s_chan = &spu.s_chan[ch];
588 int spos = s_chan->spos;
589 int sinc = s_chan->sinc;
590 int ret = ns_to, ns, d;
592 spos += s_chan->iSBPos << 16;
594 for (ns = 0; ns < ns_to; ns++)
597 while (spos >= 28*0x10000)
606 s_chan->iSBPos = spos >> 16;
607 s_chan->spos = spos & 0xffff;
612 static void do_lsfr_samples(int ns_to, int ctrl,
613 unsigned int *dwNoiseCount, unsigned int *dwNoiseVal)
615 unsigned int counter = *dwNoiseCount;
616 unsigned int val = *dwNoiseVal;
617 unsigned int level, shift, bit;
620 // modified from DrHell/shalma, no fraction
621 level = (ctrl >> 10) & 0x0f;
622 level = 0x8000 >> level;
624 for (ns = 0; ns < ns_to; ns++)
627 if (counter >= level)
630 shift = (val >> 10) & 0x1f;
631 bit = (0x69696969 >> shift) & 1;
632 bit ^= (val >> 15) & 1;
633 val = (val << 1) | bit;
636 ChanBuf[ns] = (signed short)val;
639 *dwNoiseCount = counter;
643 static int do_samples_noise(int ch, int ns_to)
647 ret = do_samples_skip(ch, ns_to);
649 do_lsfr_samples(ns_to, spu.spuCtrl, &spu.dwNoiseCount, &spu.dwNoiseVal);
655 // asm code; lv and rv must be 0-3fff
656 extern void mix_chan(int *SSumLR, int count, int lv, int rv);
657 extern void mix_chan_rvb(int *SSumLR, int count, int lv, int rv, int *rvb);
659 static void mix_chan(int *SSumLR, int count, int lv, int rv)
661 const int *src = ChanBuf;
668 l = (sval * lv) >> 14;
669 r = (sval * rv) >> 14;
675 static void mix_chan_rvb(int *SSumLR, int count, int lv, int rv, int *rvb)
677 const int *src = ChanBuf;
686 l = (sval * lv) >> 14;
687 r = (sval * rv) >> 14;
696 // 0x0800-0x0bff Voice 1
697 // 0x0c00-0x0fff Voice 3
698 static noinline void do_decode_bufs(unsigned short *mem, int which,
699 int count, int decode_pos)
701 unsigned short *dst = &mem[0x800/2 + which*0x400/2];
702 const int *src = ChanBuf;
703 int cursor = decode_pos;
708 dst[cursor] = *src++;
712 // decode_pos is updated and irqs are checked later, after voice loop
715 static void do_silent_chans(int ns_to, int silentch)
721 mask = silentch & 0xffffff;
722 for (ch = 0; mask != 0; ch++, mask >>= 1)
724 if (!(mask & 1)) continue;
725 if (spu.dwChannelDead & (1<<ch)) continue;
727 s_chan = &spu.s_chan[ch];
728 if (s_chan->pCurr > spu.pSpuIrq && s_chan->pLoop > spu.pSpuIrq)
731 s_chan->spos += s_chan->iSBPos << 16;
734 s_chan->spos += s_chan->sinc * ns_to;
735 while (s_chan->spos >= 28 * 0x10000)
737 unsigned char *start = s_chan->pCurr;
740 if (start == s_chan->pCurr || start - spu.spuMemC < 0x1000)
742 // looping on self or stopped(?)
743 spu.dwChannelDead |= 1<<ch;
748 s_chan->spos -= 28 * 0x10000;
753 static void do_channels(int ns_to)
760 do_rvb = spu.rvb->StartAddr && spu_config.iUseReverb;
762 memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2);
764 mask = spu.dwNewChannel & 0xffffff;
765 for (ch = 0; mask != 0; ch++, mask >>= 1) {
770 mask = spu.dwChannelsAudible & 0xffffff;
771 for (ch = 0; mask != 0; ch++, mask >>= 1) // loop em all...
773 if (!(mask & 1)) continue; // channel not playing? next
775 s_chan = &spu.s_chan[ch];
776 SB = spu.SB + ch * SB_SIZE;
778 if (spu.s_chan[ch].bNewPitch)
779 SB[32] = 1; // reset interpolation
780 spu.s_chan[ch].bNewPitch = 0;
783 d = do_samples_noise(ch, ns_to);
784 else if (s_chan->bFMod == 2
785 || (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 0))
786 d = do_samples_noint(decode_block, NULL, ch, ns_to,
787 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
788 else if (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 1)
789 d = do_samples_simple(decode_block, NULL, ch, ns_to,
790 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
792 d = do_samples_default(decode_block, NULL, ch, ns_to,
793 SB, sinc, &s_chan->spos, &s_chan->iSBPos);
795 d = MixADSR(&s_chan->ADSRX, d);
797 spu.dwChannelsAudible &= ~(1 << ch);
798 s_chan->ADSRX.State = ADSR_RELEASE;
799 s_chan->ADSRX.EnvelopeVol = 0;
800 memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0]));
803 if (ch == 1 || ch == 3)
805 do_decode_bufs(spu.spuMem, ch/2, ns_to, spu.decode_pos);
806 spu.decode_dirty_ch |= 1 << ch;
809 if (s_chan->bFMod == 2) // fmod freq channel
810 memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
811 if (s_chan->bRVBActive && do_rvb)
812 mix_chan_rvb(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume, RVB);
814 mix_chan(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume);
817 if (spu.rvb->StartAddr) {
819 REVERBDo(spu.SSumLR, RVB, ns_to, spu.rvb->CurrAddr);
821 spu.rvb->CurrAddr += ns_to / 2;
822 while (spu.rvb->CurrAddr >= 0x40000)
823 spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr;
827 static void do_samples_finish(int *SSumLR, int ns_to,
828 int silentch, int decode_pos);
830 // optional worker thread handling
832 #if HAVE_PTHREAD || defined(WANT_THREAD_CODE)
834 // worker thread state
835 static struct spu_worker {
838 unsigned int exit_thread;
839 unsigned int i_ready;
840 unsigned int i_reaped;
841 unsigned int last_boot_cnt; // dsp
842 unsigned int ram_dirty;
844 // aligning for C64X_DSP
845 unsigned int _pad0[128/4];
850 unsigned int active; // dsp
851 unsigned int boot_cnt;
853 unsigned int _pad1[128/4];
860 unsigned int channels_new;
861 unsigned int channels_on;
862 unsigned int channels_silent;
871 unsigned short ns_to;
872 unsigned short bNoise:1;
873 unsigned short bFMod:2;
874 unsigned short bRVBActive:1;
875 unsigned short bNewPitch:1;
878 int SSumLR[NSSIZE * 2];
882 #define WORK_MAXCNT (sizeof(worker->i) / sizeof(worker->i[0]))
883 #define WORK_I_MASK (WORK_MAXCNT - 1)
885 static void thread_work_start(void);
886 static void thread_work_wait_sync(struct work_item *work, int force);
887 static void thread_sync_caches(void);
888 static int thread_get_i_done(void);
890 static int decode_block_work(void *context, int ch, int *SB)
892 const unsigned char *ram = spu.spuMemC;
893 int predict_nr, shift_factor, flags;
894 struct work_item *work = context;
895 int start = work->ch[ch].start;
896 int loop = work->ch[ch].loop;
898 predict_nr = ram[start];
899 shift_factor = predict_nr & 0xf;
902 decode_block_data(SB, ram + start + 2, predict_nr, shift_factor);
904 flags = ram[start + 1];
906 loop = start; // loop adress
910 if (flags & 1) // 1: stop/loop
913 work->ch[ch].start = start & 0x7ffff;
914 work->ch[ch].loop = loop;
919 static void queue_channel_work(int ns_to, unsigned int silentch)
921 struct work_item *work;
926 work = &worker->i[worker->i_ready & WORK_I_MASK];
928 work->ctrl = spu.spuCtrl;
929 work->decode_pos = spu.decode_pos;
930 work->channels_silent = silentch;
932 mask = work->channels_new = spu.dwNewChannel & 0xffffff;
933 for (ch = 0; mask != 0; ch++, mask >>= 1) {
938 mask = work->channels_on = spu.dwChannelsAudible & 0xffffff;
939 spu.decode_dirty_ch |= mask & 0x0a;
941 for (ch = 0; mask != 0; ch++, mask >>= 1)
943 if (!(mask & 1)) continue;
945 s_chan = &spu.s_chan[ch];
946 work->ch[ch].spos = s_chan->spos;
947 work->ch[ch].sbpos = s_chan->iSBPos;
948 work->ch[ch].sinc = s_chan->sinc;
949 work->ch[ch].adsr = s_chan->ADSRX;
950 work->ch[ch].vol_l = s_chan->iLeftVolume;
951 work->ch[ch].vol_r = s_chan->iRightVolume;
952 work->ch[ch].start = s_chan->pCurr - spu.spuMemC;
953 work->ch[ch].loop = s_chan->pLoop - spu.spuMemC;
954 work->ch[ch].bNoise = s_chan->bNoise;
955 work->ch[ch].bFMod = s_chan->bFMod;
956 work->ch[ch].bRVBActive = s_chan->bRVBActive;
957 work->ch[ch].bNewPitch = s_chan->bNewPitch;
958 if (s_chan->prevflags & 1)
959 work->ch[ch].start = work->ch[ch].loop;
961 d = do_samples_skip(ch, ns_to);
962 work->ch[ch].ns_to = d;
964 // note: d is not accurate on skip
965 d = SkipADSR(&s_chan->ADSRX, d);
967 spu.dwChannelsAudible &= ~(1 << ch);
968 s_chan->ADSRX.State = ADSR_RELEASE;
969 s_chan->ADSRX.EnvelopeVol = 0;
971 s_chan->bNewPitch = 0;
975 if (spu.rvb->StartAddr) {
976 if (spu_config.iUseReverb)
977 work->rvb_addr = spu.rvb->CurrAddr;
979 spu.rvb->CurrAddr += ns_to / 2;
980 while (spu.rvb->CurrAddr >= 0x40000)
981 spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr;
988 static void do_channel_work(struct work_item *work)
991 int *SB, sinc, spos, sbpos;
997 memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2);
999 mask = work->channels_new;
1000 for (ch = 0; mask != 0; ch++, mask >>= 1) {
1002 StartSoundSB(spu.SB + ch * SB_SIZE);
1005 mask = work->channels_on;
1006 for (ch = 0; mask != 0; ch++, mask >>= 1)
1008 if (!(mask & 1)) continue;
1010 d = work->ch[ch].ns_to;
1011 spos = work->ch[ch].spos;
1012 sbpos = work->ch[ch].sbpos;
1013 sinc = work->ch[ch].sinc;
1015 SB = spu.SB + ch * SB_SIZE;
1016 if (work->ch[ch].bNewPitch)
1017 SB[32] = 1; // reset interpolation
1019 if (work->ch[ch].bNoise)
1020 do_lsfr_samples(d, work->ctrl, &spu.dwNoiseCount, &spu.dwNoiseVal);
1021 else if (work->ch[ch].bFMod == 2
1022 || (work->ch[ch].bFMod == 0 && spu_config.iUseInterpolation == 0))
1023 do_samples_noint(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1024 else if (work->ch[ch].bFMod == 0 && spu_config.iUseInterpolation == 1)
1025 do_samples_simple(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1027 do_samples_default(decode_block_work, work, ch, d, SB, sinc, &spos, &sbpos);
1029 d = MixADSR(&work->ch[ch].adsr, d);
1031 work->ch[ch].adsr.EnvelopeVol = 0;
1032 memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0]));
1035 if (ch == 1 || ch == 3)
1036 do_decode_bufs(spu.spuMem, ch/2, ns_to, work->decode_pos);
1038 if (work->ch[ch].bFMod == 2) // fmod freq channel
1039 memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0]));
1040 if (work->ch[ch].bRVBActive && work->rvb_addr)
1041 mix_chan_rvb(work->SSumLR, ns_to,
1042 work->ch[ch].vol_l, work->ch[ch].vol_r, RVB);
1044 mix_chan(work->SSumLR, ns_to, work->ch[ch].vol_l, work->ch[ch].vol_r);
1048 REVERBDo(work->SSumLR, RVB, ns_to, work->rvb_addr);
1051 static void sync_worker_thread(int force)
1053 struct work_item *work;
1054 int done, used_space;
1056 // rvb offsets will change, thread may be using them
1057 force |= spu.rvb->dirty && spu.rvb->StartAddr;
1059 done = thread_get_i_done() - worker->i_reaped;
1060 used_space = worker->i_ready - worker->i_reaped;
1062 //printf("done: %d use: %d dsp: %u/%u\n", done, used_space,
1063 // worker->boot_cnt, worker->i_done);
1065 while ((force && used_space > 0) || used_space >= WORK_MAXCNT || done > 0) {
1066 work = &worker->i[worker->i_reaped & WORK_I_MASK];
1067 thread_work_wait_sync(work, force);
1069 do_samples_finish(work->SSumLR, work->ns_to,
1070 work->channels_silent, work->decode_pos);
1073 done = thread_get_i_done() - worker->i_reaped;
1074 used_space = worker->i_ready - worker->i_reaped;
1077 thread_sync_caches();
1082 static void queue_channel_work(int ns_to, int silentch) {}
1083 static void sync_worker_thread(int force) {}
1085 static const void * const worker = NULL;
1087 #endif // HAVE_PTHREAD || defined(WANT_THREAD_CODE)
1089 ////////////////////////////////////////////////////////////////////////
1090 // MAIN SPU FUNCTION
1091 // here is the main job handler...
1092 ////////////////////////////////////////////////////////////////////////
1094 void do_samples(unsigned int cycles_to, int do_direct)
1096 unsigned int silentch;
1100 cycle_diff = cycles_to - spu.cycles_played;
1101 if (cycle_diff < -2*1048576 || cycle_diff > 2*1048576)
1103 //xprintf("desync %u %d\n", cycles_to, cycle_diff);
1104 spu.cycles_played = cycles_to;
1108 silentch = ~(spu.dwChannelsAudible | spu.dwNewChannel) & 0xffffff;
1110 do_direct |= (silentch == 0xffffff);
1112 sync_worker_thread(do_direct);
1114 if (cycle_diff < 2 * 768)
1117 ns_to = (cycle_diff / 768 + 1) & ~1;
1118 if (ns_to > NSSIZE) {
1119 // should never happen
1120 //xprintf("ns_to oflow %d %d\n", ns_to, NSSIZE);
1124 //////////////////////////////////////////////////////
1125 // special irq handling in the decode buffers (0x0000-0x1000)
1127 // the decode buffers are located in spu memory in the following way:
1128 // 0x0000-0x03ff CD audio left
1129 // 0x0400-0x07ff CD audio right
1130 // 0x0800-0x0bff Voice 1
1131 // 0x0c00-0x0fff Voice 3
1132 // and decoded data is 16 bit for one sample
1134 // even if voices 1/3 are off or no cd audio is playing, the internal
1135 // play positions will move on and wrap after 0x400 bytes.
1136 // Therefore: we just need a pointer from spumem+0 to spumem+3ff, and
1137 // increase this pointer on each sample by 2 bytes. If this pointer
1138 // (or 0x400 offsets of this pointer) hits the spuirq address, we generate
1141 if (unlikely((spu.spuCtrl & CTRL_IRQ)
1142 && spu.pSpuIrq < spu.spuMemC+0x1000))
1144 int irq_pos = (spu.pSpuIrq - spu.spuMemC) / 2 & 0x1ff;
1145 int left = (irq_pos - spu.decode_pos) & 0x1ff;
1146 if (0 < left && left <= ns_to)
1148 //xprintf("decoder irq %x\n", spu.decode_pos);
1152 check_irq_io(spu.spuAddr);
1154 if (unlikely(spu.rvb->dirty))
1157 if (do_direct || worker == NULL || !spu_config.iUseThread) {
1159 do_samples_finish(spu.SSumLR, ns_to, silentch, spu.decode_pos);
1162 queue_channel_work(ns_to, silentch);
1163 //sync_worker_thread(1); // uncomment for debug
1166 // advance "stopped" channels that can cause irqs
1167 // (all chans are always playing on the real thing..)
1168 if (spu.spuCtrl & CTRL_IRQ)
1169 do_silent_chans(ns_to, silentch);
1171 spu.cycles_played += ns_to * 768;
1172 spu.decode_pos = (spu.decode_pos + ns_to) & 0x1ff;
1175 static void do_samples_finish(int *SSumLR, int ns_to,
1176 int silentch, int decode_pos)
1178 int vol_l = ((int)regAreaGet(H_SPUmvolL) << 17) >> 17;
1179 int vol_r = ((int)regAreaGet(H_SPUmvolR) << 17) >> 17;
1183 // must clear silent channel decode buffers
1184 if(unlikely(silentch & spu.decode_dirty_ch & (1<<1)))
1186 memset(&spu.spuMem[0x800/2], 0, 0x400);
1187 spu.decode_dirty_ch &= ~(1<<1);
1189 if(unlikely(silentch & spu.decode_dirty_ch & (1<<3)))
1191 memset(&spu.spuMem[0xc00/2], 0, 0x400);
1192 spu.decode_dirty_ch &= ~(1<<3);
1195 MixXA(SSumLR, ns_to, decode_pos);
1197 vol_l = vol_l * spu_config.iVolume >> 10;
1198 vol_r = vol_r * spu_config.iVolume >> 10;
1200 if (!(spu.spuCtrl & 0x4000) || !(vol_l | vol_r))
1203 memset(spu.pS, 0, ns_to * 2 * sizeof(spu.pS[0]));
1204 memset(SSumLR, 0, ns_to * 2 * sizeof(SSumLR[0]));
1205 spu.pS += ns_to * 2;
1208 for (ns = 0; ns < ns_to * 2; )
1210 d = SSumLR[ns]; SSumLR[ns] = 0;
1211 d = d * vol_l >> 14;
1216 d = SSumLR[ns]; SSumLR[ns] = 0;
1217 d = d * vol_r >> 14;
1224 void schedule_next_irq(void)
1226 unsigned int upd_samples;
1229 if (spu.scheduleCallback == NULL)
1232 upd_samples = 44100 / 50;
1234 for (ch = 0; ch < MAXCHAN; ch++)
1236 if (spu.dwChannelDead & (1 << ch))
1238 if ((unsigned long)(spu.pSpuIrq - spu.s_chan[ch].pCurr) > IRQ_NEAR_BLOCKS * 16
1239 && (unsigned long)(spu.pSpuIrq - spu.s_chan[ch].pLoop) > IRQ_NEAR_BLOCKS * 16)
1241 if (spu.s_chan[ch].sinc == 0)
1244 scan_for_irq(ch, &upd_samples);
1247 if (unlikely(spu.pSpuIrq < spu.spuMemC + 0x1000))
1249 int irq_pos = (spu.pSpuIrq - spu.spuMemC) / 2 & 0x1ff;
1250 int left = (irq_pos - spu.decode_pos) & 0x1ff;
1251 if (0 < left && left < upd_samples) {
1252 //xprintf("decode: %3d (%3d/%3d)\n", left, spu.decode_pos, irq_pos);
1257 if (upd_samples < 44100 / 50)
1258 spu.scheduleCallback(upd_samples * 768);
1261 // SPU ASYNC... even newer epsxe func
1262 // 1 time every 'cycle' cycles... harhar
1264 // rearmed: called dynamically now
1266 void CALLBACK SPUasync(unsigned int cycle, unsigned int flags)
1268 do_samples(cycle, spu_config.iUseFixedUpdates);
1270 if (spu.spuCtrl & CTRL_IRQ)
1271 schedule_next_irq();
1274 out_current->feed(spu.pSpuBuffer, (unsigned char *)spu.pS - spu.pSpuBuffer);
1275 spu.pS = (short *)spu.pSpuBuffer;
1277 if (spu_config.iTempo) {
1278 if (!out_current->busy())
1279 // cause more samples to be generated
1280 // (and break some games because of bad sync)
1281 spu.cycles_played -= 44100 / 60 / 2 * 768;
1286 // SPU UPDATE... new epsxe func
1287 // 1 time every 32 hsync lines
1288 // (312/32)x50 in pal
1289 // (262/32)x60 in ntsc
1291 // since epsxe 1.5.2 (linux) uses SPUupdate, not SPUasync, I will
1292 // leave that func in the linux port, until epsxe linux is using
1293 // the async function as well
1295 void CALLBACK SPUupdate(void)
1301 void CALLBACK SPUplayADPCMchannel(xa_decode_t *xap, unsigned int cycle, int is_start)
1304 if(!xap->freq) return; // no xa freq ? bye
1307 do_samples(cycle, 1); // catch up to prevent source underflows later
1309 FeedXA(xap); // call main XA feeder
1313 int CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes, unsigned int cycle, int is_start)
1315 if (!pcm) return -1;
1316 if (nbytes<=0) return -1;
1319 do_samples(cycle, 1); // catch up to prevent source underflows later
1321 return FeedCDDA((unsigned char *)pcm, nbytes);
1324 // to be called after state load
1325 void ClearWorkingState(void)
1327 memset(iFMod, 0, sizeof(iFMod));
1328 spu.pS=(short *)spu.pSpuBuffer; // setup soundbuffer pointer
1331 // SETUPSTREAMS: init most of the spu buffers
1332 static void SetupStreams(void)
1334 spu.pSpuBuffer = (unsigned char *)malloc(32768); // alloc mixing buffer
1335 spu.SSumLR = calloc(NSSIZE * 2, sizeof(spu.SSumLR[0]));
1337 spu.XAStart = malloc(44100 * sizeof(uint32_t)); // alloc xa buffer
1338 spu.XAEnd = spu.XAStart + 44100;
1339 spu.XAPlay = spu.XAStart;
1340 spu.XAFeed = spu.XAStart;
1342 spu.CDDAStart = malloc(CDDA_BUFFER_SIZE); // alloc cdda buffer
1343 spu.CDDAEnd = spu.CDDAStart + 16384;
1344 spu.CDDAPlay = spu.CDDAStart;
1345 spu.CDDAFeed = spu.CDDAStart;
1347 ClearWorkingState();
1350 // REMOVESTREAMS: free most buffer
1351 static void RemoveStreams(void)
1353 free(spu.pSpuBuffer); // free mixing buffer
1354 spu.pSpuBuffer = NULL;
1357 free(spu.XAStart); // free XA buffer
1359 free(spu.CDDAStart); // free CDDA buffer
1360 spu.CDDAStart = NULL;
1363 #if defined(C64X_DSP)
1365 /* special code for TI C64x DSP */
1366 #include "spu_c64x.c"
1370 #include <pthread.h>
1371 #include <semaphore.h>
1380 /* generic pthread implementation */
1382 static void thread_work_start(void)
1384 sem_post(&t.sem_avail);
1387 static void thread_work_wait_sync(struct work_item *work, int force)
1389 sem_wait(&t.sem_done);
1392 static int thread_get_i_done(void)
1394 return worker->i_done;
1397 static void thread_sync_caches(void)
1401 static void *spu_worker_thread(void *unused)
1403 struct work_item *work;
1406 sem_wait(&t.sem_avail);
1407 if (worker->exit_thread)
1410 work = &worker->i[worker->i_done & WORK_I_MASK];
1411 do_channel_work(work);
1414 sem_post(&t.sem_done);
1420 static void init_spu_thread(void)
1424 if (sysconf(_SC_NPROCESSORS_ONLN) <= 1)
1427 worker = calloc(1, sizeof(*worker));
1430 ret = sem_init(&t.sem_avail, 0, 0);
1432 goto fail_sem_avail;
1433 ret = sem_init(&t.sem_done, 0, 0);
1437 ret = pthread_create(&t.thread, NULL, spu_worker_thread, NULL);
1441 spu_config.iThreadAvail = 1;
1445 sem_destroy(&t.sem_done);
1447 sem_destroy(&t.sem_avail);
1451 spu_config.iThreadAvail = 0;
1454 static void exit_spu_thread(void)
1458 worker->exit_thread = 1;
1459 sem_post(&t.sem_avail);
1460 pthread_join(t.thread, NULL);
1461 sem_destroy(&t.sem_done);
1462 sem_destroy(&t.sem_avail);
1467 #else // if !HAVE_PTHREAD
1469 static void init_spu_thread(void)
1473 static void exit_spu_thread(void)
1479 // SPUINIT: this func will be called first by the main emu
1480 long CALLBACK SPUinit(void)
1484 spu.spuMemC = calloc(1, 512 * 1024);
1487 spu.s_chan = calloc(MAXCHAN+1, sizeof(spu.s_chan[0])); // channel + 1 infos (1 is security for fmod handling)
1488 spu.rvb = calloc(1, sizeof(REVERBInfo));
1489 spu.SB = calloc(MAXCHAN, sizeof(spu.SB[0]) * SB_SIZE);
1493 spu.pSpuIrq = spu.spuMemC;
1495 SetupStreams(); // prepare streaming
1497 if (spu_config.iVolume == 0)
1498 spu_config.iVolume = 768; // 1024 is 1.0
1502 for (i = 0; i < MAXCHAN; i++) // loop sound channels
1504 spu.s_chan[i].ADSRX.SustainLevel = 0xf; // -> init sustain
1505 spu.s_chan[i].ADSRX.SustainIncrease = 1;
1506 spu.s_chan[i].pLoop = spu.spuMemC;
1507 spu.s_chan[i].pCurr = spu.spuMemC;
1508 spu.s_chan[i].bIgnoreLoop = 0;
1511 spu.bSpuInit=1; // flag: we are inited
1516 // SPUOPEN: called by main emu after init
1517 long CALLBACK SPUopen(void)
1519 if (spu.bSPUIsOpen) return 0; // security for some stupid main emus
1521 SetupSound(); // setup sound (before init!)
1525 return PSE_SPU_ERR_SUCCESS;
1528 // SPUCLOSE: called before shutdown
1529 long CALLBACK SPUclose(void)
1531 if (!spu.bSPUIsOpen) return 0; // some security
1533 spu.bSPUIsOpen = 0; // no more open
1535 out_current->finish(); // no more sound handling
1540 // SPUSHUTDOWN: called by main emu on final exit
1541 long CALLBACK SPUshutdown(void)
1556 RemoveStreams(); // no more streaming
1562 // SPUTEST: we don't test, we are always fine ;)
1563 long CALLBACK SPUtest(void)
1568 // SPUCONFIGURE: call config dialog
1569 long CALLBACK SPUconfigure(void)
1574 // StartCfgTool("CFG");
1579 // SPUABOUT: show about window
1580 void CALLBACK SPUabout(void)
1585 // StartCfgTool("ABOUT");
1590 // this functions will be called once,
1591 // passes a callback that should be called on SPU-IRQ/cdda volume change
1592 void CALLBACK SPUregisterCallback(void (CALLBACK *callback)(void))
1594 spu.irqCallback = callback;
1597 void CALLBACK SPUregisterCDDAVolume(void (CALLBACK *CDDAVcallback)(short, short))
1599 spu.cddavCallback = CDDAVcallback;
1602 void CALLBACK SPUregisterScheduleCb(void (CALLBACK *callback)(unsigned int))
1604 spu.scheduleCallback = callback;
1607 // COMMON PLUGIN INFO FUNCS
1609 char * CALLBACK PSEgetLibName(void)
1611 return _(libraryName);
1614 unsigned long CALLBACK PSEgetLibType(void)
1619 unsigned long CALLBACK PSEgetLibVersion(void)
1621 return (1 << 16) | (6 << 8);
1624 char * SPUgetLibInfos(void)
1626 return _(libraryInfo);
1631 void spu_get_debug_info(int *chans_out, int *run_chans, int *fmod_chans_out, int *noise_chans_out)
1633 int ch = 0, fmod_chans = 0, noise_chans = 0, irq_chans = 0;
1635 if (spu.s_chan == NULL)
1638 for(;ch<MAXCHAN;ch++)
1640 if (!(spu.dwChannelsAudible & (1<<ch)))
1642 if (spu.s_chan[ch].bFMod == 2)
1643 fmod_chans |= 1 << ch;
1644 if (spu.s_chan[ch].bNoise)
1645 noise_chans |= 1 << ch;
1646 if((spu.spuCtrl&CTRL_IRQ) && spu.s_chan[ch].pCurr <= spu.pSpuIrq && spu.s_chan[ch].pLoop <= spu.pSpuIrq)
1647 irq_chans |= 1 << ch;
1650 *chans_out = spu.dwChannelsAudible;
1651 *run_chans = ~spu.dwChannelsAudible & ~spu.dwChannelDead & irq_chans;
1652 *fmod_chans_out = fmod_chans;
1653 *noise_chans_out = noise_chans;
1656 // vim:shiftwidth=1:expandtab