X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=plugins%2Fdfsound%2Fspu.c;h=2c13ca16fef83f6879eb08b1e8d1254e294a7265;hp=06cc4761c0356fa6199b806fcf6c85c2ec8cc98c;hb=c4c66b22ae607aa8dc5e6983f7ab030ead01faae;hpb=be1cb678c445cb8d9a4d707c698f0dc600a490ea diff --git a/plugins/dfsound/spu.c b/plugins/dfsound/spu.c index 06cc4761..2c13ca16 100644 --- a/plugins/dfsound/spu.c +++ b/plugins/dfsound/spu.c @@ -5,7 +5,7 @@ copyright : (C) 2002 by Pete Bernert email : BlackDove@addcom.de - Portions (C) Gražvydas "notaz" Ignotas, 2010-2011 + Portions (C) Gražvydas "notaz" Ignotas, 2010-2012 ***************************************************************************/ /*************************************************************************** @@ -24,8 +24,9 @@ #include "externals.h" #include "registers.h" -#include "cfg.h" -#include "dsoundoss.h" +#include "out.h" +#include "arm_features.h" +#include "spu_config.h" #ifdef ENABLE_NLS #include @@ -37,7 +38,7 @@ #define N_(x) (x) #endif -#ifdef __arm__ +#ifdef __ARM_ARCH_7A__ #define ssat32_to_16(v) \ asm("ssat %0,#16,%1" : "=r" (v) : "r" (v)) #else @@ -47,6 +48,11 @@ } while (0) #endif +#define PSXCLK 33868800 /* 33.8688 MHz */ + +// intended to be ~1 frame +#define IRQ_NEAR_BLOCKS 32 + /* #if defined (USEMACOSX) static char * libraryName = N_("Mac OS X Sound"); @@ -67,54 +73,14 @@ static char * libraryInfo = N_("P.E.Op.S. Sound Driver V1.7\nCoded by Pete B // globals -// psx buffer / addresses - -unsigned short regArea[10000]; -unsigned short spuMem[256*1024]; -unsigned char * spuMemC; -unsigned char * pSpuIrq=0; -unsigned char * pSpuBuffer; -unsigned char * pMixIrq=0; - -// user settings - -int iVolume=768; // 1024 is 1.0 -int iXAPitch=1; -int iUseTimer=2; -int iSPUIRQWait=1; -int iDebugMode=0; -int iRecordMode=0; -int iUseReverb=2; -int iUseInterpolation=2; +SPUInfo spu; +SPUConfig spu_config; // MAIN infos struct for each channel SPUCHAN s_chan[MAXCHAN+1]; // channel + 1 infos (1 is security for fmod handling) REVERBInfo rvb; -unsigned int dwNoiseVal; // global noise generator -unsigned int dwNoiseCount; -int iSpuAsyncWait=0; - -unsigned short spuCtrl=0; // some vars to store psx reg infos -unsigned short spuStat=0; -unsigned short spuIrq=0; -unsigned long spuAddr=0xffffffff; // address into spu mem -int bEndThread=0; // thread handlers -int bThreadEnded=0; -int bSpuInit=0; -int bSPUIsOpen=0; - -static pthread_t thread = (pthread_t)-1; // thread id (linux) - -unsigned int dwNewChannel=0; // flags for faster testing, if new channel starts -unsigned int dwChannelOn=0; // not silent channels -unsigned int dwPendingChanOff=0; -unsigned int dwChannelDead=0; // silent+not useful channels - -void (CALLBACK *irqCallback)(void)=0; // func of main emu, called on spu irq -void (CALLBACK *cddavCallback)(unsigned short,unsigned short)=0; - // certain globals (were local before, but with the new timeproc I need em global) static const int f[8][2] = { { 0, 0 }, @@ -122,15 +88,11 @@ static const int f[8][2] = { { 0, 0 }, { 115, -52 }, { 98, -55 }, { 122, -60 } }; -int ChanBuf[NSSIZE+3]; -int SSumLR[(NSSIZE+3)*2]; +int ChanBuf[NSSIZE]; +int SSumLR[NSSIZE*2]; int iFMod[NSSIZE]; -int iCycle = 0; -short * pS; -int lastch=-1; // last channel processed on spu irq in timer mode -static int lastns=0; // last ns pos -static int iSecureStart=0; // secure start counter +#define CDDA_BUFFER_SIZE (16384 * sizeof(uint32_t)) // must be power of 2 //////////////////////////////////////////////////////////////////////// // CODE AREA @@ -220,9 +182,10 @@ INLINE void InterpolateUp(int ch) s_chan[ch].SB[32]=0; s_chan[ch].SB[28]=(s_chan[ch].SB[28]*s_chan[ch].sinc)/0x20000L; - if(s_chan[ch].sinc<=0x8000) - s_chan[ch].SB[29]=s_chan[ch].SB[30]-(s_chan[ch].SB[28]*((0x10000/s_chan[ch].sinc)-1)); - else s_chan[ch].SB[29]+=s_chan[ch].SB[28]; + //if(s_chan[ch].sinc<=0x8000) + // s_chan[ch].SB[29]=s_chan[ch].SB[30]-(s_chan[ch].SB[28]*((0x10000/s_chan[ch].sinc)-1)); + //else + s_chan[ch].SB[29]+=s_chan[ch].SB[28]; } else // no flags? add bigger val (if possible), calc smaller step, set flag1 s_chan[ch].SB[29]+=s_chan[ch].SB[28]; @@ -245,8 +208,8 @@ INLINE void InterpolateDown(int ch) //////////////////////////////////////////////////////////////////////// // helpers for gauss interpolation -#define gval0 (((short*)(&s_chan[ch].SB[29]))[gpos]) -#define gval(x) (((short*)(&s_chan[ch].SB[29]))[(gpos+x)&3]) +#define gval0 (((short*)(&s_chan[ch].SB[29]))[gpos&3]) +#define gval(x) ((int)((short*)(&s_chan[ch].SB[29]))[(gpos+x)&3]) #include "gauss_i.h" @@ -254,6 +217,26 @@ INLINE void InterpolateDown(int ch) #include "xa.c" +static void do_irq(void) +{ + //if(!(spu.spuStat & STAT_IRQ)) + { + spu.spuStat |= STAT_IRQ; // asserted status? + if(spu.irqCallback) spu.irqCallback(); + } +} + +static int check_irq(int ch, unsigned char *pos) +{ + if((spu.spuCtrl & CTRL_IRQ) && pos == spu.pSpuIrq) + { + //printf("ch%d irq %04x\n", ch, pos - spu.spuMemC); + do_irq(); + return 1; + } + return 0; +} + //////////////////////////////////////////////////////////////////////// // START SOUND... called by main thread to setup a new sound on a channel //////////////////////////////////////////////////////////////////////// @@ -275,11 +258,11 @@ INLINE void StartSound(int ch) s_chan[ch].SB[29]=0; // init our interpolation helpers s_chan[ch].SB[30]=0; - if(iUseInterpolation>=2) // gauss interpolation? + if(spu_config.iUseInterpolation>=2) // gauss interpolation? {s_chan[ch].spos=0x30000L;s_chan[ch].SB[28]=0;} // -> start with more decoding else {s_chan[ch].spos=0x10000L;s_chan[ch].SB[31]=0;} // -> no/simple interpolation starts with one 44100 decoding - dwNewChannel&=~(1<=2) // gauss/cubic interpolation + if(spu_config.iUseInterpolation>=2) // gauss/cubic interpolation { int gpos = s_chan[ch].SB[28]; gval0 = fa; @@ -322,7 +305,7 @@ INLINE void StoreInterpolationVal(int ch,int fa) s_chan[ch].SB[28] = gpos; } else - if(iUseInterpolation==1) // simple interpolation + if(spu_config.iUseInterpolation==1) // simple interpolation { s_chan[ch].SB[28] = 0; s_chan[ch].SB[29] = s_chan[ch].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' @@ -336,19 +319,19 @@ INLINE void StoreInterpolationVal(int ch,int fa) //////////////////////////////////////////////////////////////////////// -INLINE int iGetInterpolationVal(int ch) +INLINE int iGetInterpolationVal(int ch, int spos) { int fa; if(s_chan[ch].bFMod==2) return s_chan[ch].SB[29]; - switch(iUseInterpolation) + switch(spu_config.iUseInterpolation) { //--------------------------------------------------// case 3: // cubic interpolation { long xd;int gpos; - xd = ((s_chan[ch].spos) >> 1)+1; + xd = (spos >> 1)+1; gpos = s_chan[ch].SB[28]; fa = gval(3) - 3*gval(2) + 3*gval(1) - gval0; @@ -367,9 +350,9 @@ INLINE int iGetInterpolationVal(int ch) case 2: // gauss interpolation { int vl, vr;int gpos; - vl = (s_chan[ch].spos >> 6) & ~3; + vl = (spos >> 6) & ~3; gpos = s_chan[ch].SB[28]; - vr=(gauss[vl]*gval0)&~2047; + vr=(gauss[vl]*(int)gval0)&~2047; vr+=(gauss[vl+1]*gval(1))&~2047; vr+=(gauss[vl+2]*gval(2))&~2047; vr+=(gauss[vl+3]*gval(3))&~2047; @@ -394,15 +377,6 @@ INLINE int iGetInterpolationVal(int ch) return fa; } -static void do_irq(void) -{ - if(!(spuStat & STAT_IRQ)) - { - spuStat |= STAT_IRQ; - if(irqCallback) irqCallback(); - } -} - static void decode_block_data(int *dest, const unsigned char *src, int predict_nr, int shift_factor) { int nSample; @@ -435,26 +409,28 @@ static int decode_block(int ch) { unsigned char *start; int predict_nr,shift_factor,flags; + int stop = 0; int ret = 0; - start=s_chan[ch].pCurr; // set up the current pos - if(dwPendingChanOff&(1< turn everything off - dwPendingChanOff&=~(1< call main emu - ret = 1; - } + spu.dwChannelOn &= ~(1< turn everything off + s_chan[ch].bStop = 1; + s_chan[ch].ADSRX.EnvelopeVol = 0; } predict_nr=(int)start[0]; @@ -463,29 +439,22 @@ static int decode_block(int ch) decode_block_data(s_chan[ch].SB, start + 2, predict_nr, shift_factor); - //////////////////////////////////////////// flag handler - flags=(int)start[1]; if(flags&4) s_chan[ch].pLoop=start; // loop adress start+=16; - if(flags&1) // 1: stop/loop - { - if(!(flags&2)) - dwPendingChanOff|=1<= 0x80000) { - // most likely wrong - start = spuMemC; - printf("ch%d oflow\n", ch); - } + if (start - spu.spuMemC >= 0x80000) + start = spu.spuMemC; s_chan[ch].pCurr = start; // store values for next cycle - s_chan[ch].bJump = flags & 1; + s_chan[ch].prevflags = flags; return ret; } @@ -495,28 +464,67 @@ static int skip_block(int ch) { unsigned char *start = s_chan[ch].pCurr; int flags = start[1]; - int ret = 0; + int ret = check_irq(ch, start); - if(start == pSpuIrq) - { - do_irq(); - ret = 1; - } + if(s_chan[ch].prevflags & 1) + start = s_chan[ch].pLoop; if(flags & 4) s_chan[ch].pLoop = start; - s_chan[ch].pCurr += 16; + start += 16; if(flags & 1) - s_chan[ch].pCurr = s_chan[ch].pLoop; + start = s_chan[ch].pLoop; + + s_chan[ch].pCurr = start; + s_chan[ch].prevflags = flags; - s_chan[ch].bJump = flags & 1; return ret; } +// if irq is going to trigger sooner than in upd_samples, set upd_samples +static void scan_for_irq(int ch, unsigned int *upd_samples) +{ + int pos, sinc, sinc_inv, end; + unsigned char *block; + int flags; + + block = s_chan[ch].pCurr; + pos = s_chan[ch].spos; + sinc = s_chan[ch].sinc; + end = pos + *upd_samples * sinc; + + pos += (28 - s_chan[ch].iSBPos) << 16; + while (pos < end) + { + if (block == spu.pSpuIrq) + break; + flags = block[1]; + block += 16; + if (flags & 1) { // 1: stop/loop + block = s_chan[ch].pLoop; + if (block == spu.pSpuIrq) // hack.. (see decode_block) + break; + } + pos += 28 << 16; + } + + if (pos < end) + { + sinc_inv = s_chan[ch].sinc_inv; + if (sinc_inv == 0) + sinc_inv = s_chan[ch].sinc_inv = (0x80000000u / (uint32_t)sinc) << 1; + + pos -= s_chan[ch].spos; + *upd_samples = (((uint64_t)pos * sinc_inv) >> 32) + 1; + //xprintf("ch%02d: irq sched: %3d %03d\n", + // ch, *upd_samples, *upd_samples * 60 * 263 / 44100); + } +} + #define make_do_samples(name, fmod_code, interp_start, interp1_code, interp2_code, interp_end) \ -static int do_samples_##name(int ch, int ns, int ns_to) \ +static noinline int do_samples_##name(int ch, int ns, int ns_to) \ { \ int sinc = s_chan[ch].sinc; \ int spos = s_chan[ch].spos; \ @@ -530,29 +538,25 @@ static int do_samples_##name(int ch, int ns, int ns_to) \ { \ fmod_code; \ \ + spos += sinc; \ while (spos >= 0x10000) \ { \ - if(sbpos == 28) \ + fa = SB[sbpos++]; \ + if(sbpos >= 28) \ { \ sbpos = 0; \ d = decode_block(ch); \ - if(d && iSPUIRQWait) \ - { \ - ret = ns; \ - goto out; \ - } \ + if(d) \ + ret = /*ns_to =*/ ns + 1; \ } \ \ - fa = SB[sbpos++]; \ interp1_code; \ spos -= 0x10000; \ } \ \ interp2_code; \ - spos += sinc; \ } \ \ -out: \ s_chan[ch].sinc = sinc; \ s_chan[ch].spos = spos; \ s_chan[ch].iSBPos = sbpos; \ @@ -567,7 +571,7 @@ out: \ make_do_samples(default, fmod_recv_check, , StoreInterpolationVal(ch, fa), - ChanBuf[ns] = iGetInterpolationVal(ch), ) + ChanBuf[ns] = iGetInterpolationVal(ch, spos), ) make_do_samples(noint, , fa = s_chan[ch].SB[29], , ChanBuf[ns] = fa, s_chan[ch].SB[29] = fa) #define simple_interp_store \ @@ -586,44 +590,47 @@ make_do_samples(noint, , fa = s_chan[ch].SB[29], , ChanBuf[ns] = fa, s_chan[ch]. make_do_samples(simple, , , simple_interp_store, simple_interp_get, ) -static int do_samples_noise(int ch, int ns, int ns_to) +static noinline int do_samples_noise(int ch, int ns, int ns_to) { int level, shift, bit; + int ret = -1, d; s_chan[ch].spos += s_chan[ch].sinc * (ns_to - ns); while (s_chan[ch].spos >= 28*0x10000) { - skip_block(ch); + d = skip_block(ch); + if (d) + ret = ns_to; s_chan[ch].spos -= 28*0x10000; } // modified from DrHell/shalma, no fraction - level = (spuCtrl >> 10) & 0x0f; + level = (spu.spuCtrl >> 10) & 0x0f; level = 0x8000 >> level; for (; ns < ns_to; ns++) { - dwNoiseCount += 2; - if (dwNoiseCount >= level) + spu.dwNoiseCount += 2; + if (spu.dwNoiseCount >= level) { - dwNoiseCount -= level; - shift = (dwNoiseVal >> 10) & 0x1f; + spu.dwNoiseCount -= level; + shift = (spu.dwNoiseVal >> 10) & 0x1f; bit = (0x69696969 >> shift) & 1; - if (dwNoiseVal & 0x8000) + if (spu.dwNoiseVal & 0x8000) bit ^= 1; - dwNoiseVal = (dwNoiseVal << 1) | bit; + spu.dwNoiseVal = (spu.dwNoiseVal << 1) | bit; } - ChanBuf[ns] = (signed short)dwNoiseVal; + ChanBuf[ns] = (signed short)spu.dwNoiseVal; } - return -1; + return ret; } -#ifdef __arm__ -// asm code +#ifdef HAVE_ARMV5 +// asm code; lv and rv must be 0-3fff extern void mix_chan(int start, int count, int lv, int rv); -extern void mix_chan_rvb(int start, int count, int lv, int rv); +extern void mix_chan_rvb(int start, int count, int lv, int rv, int *rvb); #else static void mix_chan(int start, int count, int lv, int rv) { @@ -642,10 +649,10 @@ static void mix_chan(int start, int count, int lv, int rv) } } -static void mix_chan_rvb(int start, int count, int lv, int rv) +static void mix_chan_rvb(int start, int count, int lv, int rv, int *rvb) { int *dst = SSumLR + start * 2; - int *drvb = sRVBStart + start * 2; + int *drvb = rvb + start * 2; const int *src = ChanBuf + start; int l, r; @@ -663,125 +670,147 @@ static void mix_chan_rvb(int start, int count, int lv, int rv) } #endif -//////////////////////////////////////////////////////////////////////// -// MAIN SPU FUNCTION -// here is the main job handler... thread, timer or direct func call -// basically the whole sound processing is done in this fat func! -//////////////////////////////////////////////////////////////////////// +// 0x0800-0x0bff Voice 1 +// 0x0c00-0x0fff Voice 3 +static noinline void do_decode_bufs(int which, int start, int count) +{ + const int *src = ChanBuf + start; + unsigned short *dst = &spu.spuMem[0x800/2 + which*0x400/2]; + int cursor = spu.decode_pos + start; -// 5 ms waiting phase, if buffer is full and no new sound has to get started -// .. can be made smaller (smallest val: 1 ms), but bigger waits give -// better performance + while (count-- > 0) + { + cursor &= 0x1ff; + dst[cursor] = *src++; + cursor++; + } -#define PAUSE_W 5 -#define PAUSE_L 5000 + // decode_pos is updated and irqs are checked later, after voice loop +} +//////////////////////////////////////////////////////////////////////// +// MAIN SPU FUNCTION +// here is the main job handler... +// basically the whole sound processing is done in this fat func! //////////////////////////////////////////////////////////////////////// -static void *MAINThread(void *arg) +void do_samples(unsigned int cycles_to) { - int volmult = iVolume; - int ns,ns_from,ns_to; + const int ns_from = 0; + int ns,ns_to,ns_len; + int volmult = spu_config.iVolume; int ch,d,silentch; - int bIRQReturn=0; + int cycle_diff; - while(!bEndThread) // until we are shutting down + cycle_diff = cycles_to - spu.cycles_played; + if (cycle_diff < -2*1048576 || cycle_diff > 2*1048576) { - // ok, at the beginning we are looking if there is - // enuff free place in the dsound/oss buffer to - // fill in new data, or if there is a new channel to start. - // if not, we wait (thread) or return (timer/spuasync) - // until enuff free place is available/a new channel gets - // started - - if(dwNewChannel) // new channel should start immedately? - { // (at least one bit 0 ... MAXCHANNEL is set?) - iSecureStart++; // -> set iSecure - if(iSecureStart>5) iSecureStart=0; // (if it is set 5 times - that means on 5 tries a new samples has been started - in a row, we will reset it, to give the sound update a chance) - } - else iSecureStart=0; // 0: no new channel should start + //xprintf("desync %u %d\n", cycles_to, cycle_diff); + spu.cycles_played = cycles_to; + return; + } - while(!iSecureStart && !bEndThread && // no new start? no thread end? - (SoundGetBytesBuffered()>TESTSIZE)) // and still enuff data in sound buffer? - { - iSecureStart=0; // reset secure + if (cycle_diff < 2 * 768) + return; - if(iUseTimer) return 0; // linux no-thread mode? bye - usleep(PAUSE_L); // else sleep for x ms (linux) + ns_to = (cycle_diff / 768 + 1) & ~1; + if (ns_to > NSSIZE) { + // should never happen + //xprintf("ns_to oflow %d %d\n", ns_to, NSSIZE); + ns_to = NSSIZE; + } - if(dwNewChannel) iSecureStart=1; // if a new channel kicks in (or, of course, sound buffer runs low), we will leave the loop - } + ////////////////////////////////////////////////////// + // special irq handling in the decode buffers (0x0000-0x1000) + // we know: + // the decode buffers are located in spu memory in the following way: + // 0x0000-0x03ff CD audio left + // 0x0400-0x07ff CD audio right + // 0x0800-0x0bff Voice 1 + // 0x0c00-0x0fff Voice 3 + // and decoded data is 16 bit for one sample + // we assume: + // even if voices 1/3 are off or no cd audio is playing, the internal + // play positions will move on and wrap after 0x400 bytes. + // Therefore: we just need a pointer from spumem+0 to spumem+3ff, and + // increase this pointer on each sample by 2 bytes. If this pointer + // (or 0x400 offsets of this pointer) hits the spuirq address, we generate + // an IRQ. - //--------------------------------------------------// continue from irq handling in timer mode? + if (unlikely((spu.spuCtrl & CTRL_IRQ) + && spu.pSpuIrq < spu.spuMemC+0x1000)) + { + int irq_pos = (spu.pSpuIrq - spu.spuMemC) / 2 & 0x1ff; + int left = (irq_pos - spu.decode_pos) & 0x1ff; + if (0 < left && left <= ns_to) + { + //xprintf("decoder irq %x\n", spu.decode_pos); + do_irq(); + } + } - ns_from=0; - ns_to=NSSIZE; - ch=0; - if(lastch>=0) // will be -1 if no continue is pending - { - ch=lastch; ns_from=lastns; lastch=-1; // -> setup all kind of vars to continue - } + InitREVERB(ns_to); - silentch=~(dwChannelOn|dwNewChannel); + { + silentch=~(spu.dwChannelOn|spu.dwNewChannel); //--------------------------------------------------// //- main channel loop -// //--------------------------------------------------// { - for(;ch=0) - { - bIRQReturn=1; - lastch=ch; - lastns=ns_to=d; - if(d==0) - break; - } + do_samples_default(ch, ns_from, ns_to); + + ns_len = ns_to - ns_from; MixADSR(ch, ns_from, ns_to); + if(ch==1 || ch==3) + { + do_decode_bufs(ch/2, ns_from, ns_len); + spu.decode_dirty_ch |= 1< pSpuIrq && s_chan[ch].pLoop > pSpuIrq) + if(spu.dwChannelDead&(1< spu.pSpuIrq && s_chan[ch].pLoop > spu.pSpuIrq) continue; - s_chan[ch].spos += s_chan[ch].sinc * NSSIZE; + s_chan[ch].spos += s_chan[ch].sinc * (ns_to - ns_from); while(s_chan[ch].spos >= 28 * 0x10000) { unsigned char *start = s_chan[ch].pCurr; - // no need for bIRQReturn since the channel is silent - iSpuAsyncWait |= skip_block(ch); - if(start == s_chan[ch].pCurr) + skip_block(ch); + if(start == s_chan[ch].pCurr || start - spu.spuMemC < 0x1000) { - // looping on self - dwChannelDead |= 1<> 10; ssat32_to_16(d); - *pS++ = d; + *spu.pS++ = d; ns++; d = SSumLR[ns]; SSumLR[ns] = 0; d = d * volmult >> 10; ssat32_to_16(d); - *pS++ = d; + *spu.pS++ = d; ns++; } - ////////////////////////////////////////////////////// - // special irq handling in the decode buffers (0x0000-0x1000) - // we know: - // the decode buffers are located in spu memory in the following way: - // 0x0000-0x03ff CD audio left - // 0x0400-0x07ff CD audio right - // 0x0800-0x0bff Voice 1 - // 0x0c00-0x0fff Voice 3 - // and decoded data is 16 bit for one sample - // we assume: - // even if voices 1/3 are off or no cd audio is playing, the internal - // play positions will move on and wrap after 0x400 bytes. - // Therefore: we just need a pointer from spumem+0 to spumem+3ff, and - // increase this pointer on each sample by 2 bytes. If this pointer - // (or 0x400 offsets of this pointer) hits the spuirq address, we generate - // an IRQ. Only problem: the "wait for cpu" option is kinda hard to do here - // in some of Peops timer modes. So: we ignore this option here (for now). + spu.cycles_played += ns_to * 768; - if(pMixIrq) - { - for(ns=0;ns=pMixIrq+(ch*0x400) && pSpuIrqspuMemC+0x3ff) pMixIrq=spuMemC; - } - } + spu.decode_pos = (spu.decode_pos + ns_to) & 0x1ff; + } +} - InitREVERB(); +void schedule_next_irq(void) +{ + unsigned int upd_samples; + int ch; - // feed the sound - // wanna have around 1/60 sec (16.666 ms) updates - if (iCycle++ > 16/FRAG_MSECS) - { - SoundFeedStreamData((unsigned char *)pSpuBuffer, - ((unsigned char *)pS) - ((unsigned char *)pSpuBuffer)); - pS = (short *)pSpuBuffer; - iCycle = 0; - } - } + if (spu.scheduleCallback == NULL) + return; - // end of big main loop... + upd_samples = 44100 / 50; - bThreadEnded = 1; + for (ch = 0; ch < MAXCHAN; ch++) + { + if (spu.dwChannelDead & (1 << ch)) + continue; + if ((unsigned long)(spu.pSpuIrq - s_chan[ch].pCurr) > IRQ_NEAR_BLOCKS * 16 + && (unsigned long)(spu.pSpuIrq - s_chan[ch].pLoop) > IRQ_NEAR_BLOCKS * 16) + continue; - return 0; + scan_for_irq(ch, &upd_samples); + } + + if (unlikely(spu.pSpuIrq < spu.spuMemC + 0x1000)) + { + int irq_pos = (spu.pSpuIrq - spu.spuMemC) / 2 & 0x1ff; + int left = (irq_pos - spu.decode_pos) & 0x1ff; + if (0 < left && left < upd_samples) { + //xprintf("decode: %3d (%3d/%3d)\n", left, spu.decode_pos, irq_pos); + upd_samples = left; + } + } + + if (upd_samples < 44100 / 50) + spu.scheduleCallback(upd_samples * 768); } // SPU ASYNC... even newer epsxe func // 1 time every 'cycle' cycles... harhar -void CALLBACK SPUasync(unsigned long cycle) +// rearmed: called dynamically now + +void CALLBACK SPUasync(unsigned int cycle, unsigned int flags) { - if(iSpuAsyncWait) - { - iSpuAsyncWait++; - if(iSpuAsyncWait<=16/FRAG_MSECS) return; - iSpuAsyncWait=0; - } + do_samples(cycle); - if(iUseTimer==2) // special mode, only used in Linux by this spu (or if you enable the experimental Windows mode) - { - if(!bSpuInit) return; // -> no init, no call + if (spu.spuCtrl & CTRL_IRQ) + schedule_next_irq(); - MAINThread(0); // -> linux high-compat mode + if (flags & 1) { + out_current->feed(spu.pSpuBuffer, (unsigned char *)spu.pS - spu.pSpuBuffer); + spu.pS = (short *)spu.pSpuBuffer; - // abuse iSpuAsyncWait mechanism to reduce calls to above function - // to make it do larger chunks - // note: doing it less often than once per frame causes skips - iSpuAsyncWait=1; + if (spu_config.iTempo) { + if (!out_current->busy()) + // cause more samples to be generated + // (and break some games because of bad sync) + spu.cycles_played -= 44100 / 60 / 2 * 768; } + } } // SPU UPDATE... new epsxe func @@ -934,7 +939,6 @@ void CALLBACK SPUasync(unsigned long cycle) void CALLBACK SPUupdate(void) { - SPUasync(0); } // XA AUDIO @@ -948,45 +952,20 @@ void CALLBACK SPUplayADPCMchannel(xa_decode_t *xap) } // CDDA AUDIO -void CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes) +int CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes) { - if (!pcm) return; - if (nbytes<=0) return; + if (!pcm) return -1; + if (nbytes<=0) return -1; - FeedCDDA((unsigned char *)pcm, nbytes); + return FeedCDDA((unsigned char *)pcm, nbytes); } -// SETUPTIMER: init of certain buffers and threads/timers -void SetupTimer(void) +// to be called after state load +void ClearWorkingState(void) { memset(SSumLR,0,sizeof(SSumLR)); // init some mixing buffers - memset(iFMod,0,NSSIZE*sizeof(int)); - pS=(short *)pSpuBuffer; // setup soundbuffer pointer - - bEndThread=0; // init thread vars - bThreadEnded=0; - bSpuInit=1; // flag: we are inited - - if(!iUseTimer) // linux: use thread - { - pthread_create(&thread, NULL, MAINThread, NULL); - } -} - -// REMOVETIMER: kill threads/timers -void RemoveTimer(void) -{ - bEndThread=1; // raise flag to end thread - - if(!iUseTimer) // linux tread? - { - int i=0; - while(!bThreadEnded && i<2000) {usleep(1000L);i++;} // -> wait until thread has ended - if(thread!=(pthread_t)-1) {pthread_cancel(thread);thread=(pthread_t)-1;} // -> cancel thread anyway - } - - bThreadEnded=0; // no more spu is running - bSpuInit=0; + memset(iFMod,0,sizeof(iFMod)); + spu.pS=(short *)spu.pSpuBuffer; // setup soundbuffer pointer } // SETUPSTREAMS: init most of the spu buffers @@ -994,52 +973,46 @@ void SetupStreams(void) { int i; - pSpuBuffer=(unsigned char *)malloc(32768); // alloc mixing buffer - - if(iUseReverb==1) i=88200*2; - else i=NSSIZE*2; + spu.pSpuBuffer = (unsigned char *)malloc(32768); // alloc mixing buffer + spu.sRVBStart = (int *)malloc(NSSIZE*2*4); // alloc reverb buffer + memset(spu.sRVBStart,0,NSSIZE*2*4); - sRVBStart = (int *)malloc(i*4); // alloc reverb buffer - memset(sRVBStart,0,i*4); - sRVBEnd = sRVBStart + i; - sRVBPlay = sRVBStart; - - XAStart = // alloc xa buffer + spu.XAStart = // alloc xa buffer (uint32_t *)malloc(44100 * sizeof(uint32_t)); - XAEnd = XAStart + 44100; - XAPlay = XAStart; - XAFeed = XAStart; + spu.XAEnd = spu.XAStart + 44100; + spu.XAPlay = spu.XAStart; + spu.XAFeed = spu.XAStart; - CDDAStart = // alloc cdda buffer - (uint32_t *)malloc(16384 * sizeof(uint32_t)); - CDDAEnd = CDDAStart + 16384; - CDDAPlay = CDDAStart; - CDDAFeed = CDDAStart; + spu.CDDAStart = // alloc cdda buffer + (uint32_t *)malloc(CDDA_BUFFER_SIZE); + spu.CDDAEnd = spu.CDDAStart + 16384; + spu.CDDAPlay = spu.CDDAStart; + spu.CDDAFeed = spu.CDDAStart; for(i=0;i init sustain - s_chan[i].pLoop=spuMemC; - s_chan[i].pCurr=spuMemC; + s_chan[i].ADSRX.SustainIncrease = 1; + s_chan[i].pLoop=spu.spuMemC; + s_chan[i].pCurr=spu.spuMemC; } - pMixIrq=spuMemC; // enable decoded buffer irqs by setting the address + ClearWorkingState(); + + spu.bSpuInit=1; // flag: we are inited } // REMOVESTREAMS: free most buffer void RemoveStreams(void) { - free(pSpuBuffer); // free mixing buffer - pSpuBuffer = NULL; - free(sRVBStart); // free reverb buffer - sRVBStart = NULL; - free(XAStart); // free XA buffer - XAStart = NULL; - free(CDDAStart); // free CDDA buffer - CDDAStart = NULL; + free(spu.pSpuBuffer); // free mixing buffer + spu.pSpuBuffer = NULL; + free(spu.sRVBStart); // free reverb buffer + spu.sRVBStart = NULL; + free(spu.XAStart); // free XA buffer + spu.XAStart = NULL; + free(spu.CDDAStart); // free CDDA buffer + spu.CDDAStart = NULL; } // INIT/EXIT STUFF @@ -1047,36 +1020,31 @@ void RemoveStreams(void) // SPUINIT: this func will be called first by the main emu long CALLBACK SPUinit(void) { - spuMemC = (unsigned char *)spuMem; // just small setup + spu.spuMemC = (unsigned char *)spu.spuMem; // just small setup memset((void *)&rvb, 0, sizeof(REVERBInfo)); InitADSR(); - spuIrq = 0; - spuAddr = 0xffffffff; - bEndThread = 0; - bThreadEnded = 0; - spuMemC = (unsigned char *)spuMem; - pMixIrq = 0; - memset((void *)s_chan, 0, (MAXCHAN + 1) * sizeof(SPUCHAN)); - pSpuIrq = 0; - //iSPUIRQWait = 0; - lastch = -1; - - //ReadConfigSPU(); // read user stuff + spu.spuAddr = 0xffffffff; + spu.decode_pos = 0; + memset((void *)s_chan, 0, sizeof(s_chan)); + spu.pSpuIrq = spu.spuMemC; + SetupStreams(); // prepare streaming + if (spu_config.iVolume == 0) + spu_config.iVolume = 768; // 1024 is 1.0 + return 0; } // SPUOPEN: called by main emu after init long CALLBACK SPUopen(void) { - if (bSPUIsOpen) return 0; // security for some stupid main emus + if (spu.bSPUIsOpen) return 0; // security for some stupid main emus SetupSound(); // setup sound (before init!) - SetupTimer(); // timer for feeding data - bSPUIsOpen = 1; + spu.bSPUIsOpen = 1; return PSE_SPU_ERR_SUCCESS; } @@ -1084,12 +1052,11 @@ long CALLBACK SPUopen(void) // SPUCLOSE: called before shutdown long CALLBACK SPUclose(void) { - if (!bSPUIsOpen) return 0; // some security + if (!spu.bSPUIsOpen) return 0; // some security - bSPUIsOpen = 0; // no more open + spu.bSPUIsOpen = 0; // no more open - RemoveTimer(); // no more feeding - RemoveSound(); // no more sound handling + out_current->finish(); // no more sound handling return 0; } @@ -1099,6 +1066,7 @@ long CALLBACK SPUshutdown(void) { SPUclose(); RemoveStreams(); // no more streaming + spu.bSpuInit=0; return 0; } @@ -1135,12 +1103,17 @@ void CALLBACK SPUabout(void) // passes a callback that should be called on SPU-IRQ/cdda volume change void CALLBACK SPUregisterCallback(void (CALLBACK *callback)(void)) { - irqCallback = callback; + spu.irqCallback = callback; } void CALLBACK SPUregisterCDDAVolume(void (CALLBACK *CDDAVcallback)(unsigned short,unsigned short)) { - cddavCallback = CDDAVcallback; + spu.cddavCallback = CDDAVcallback; +} + +void CALLBACK SPUregisterScheduleCb(void (CALLBACK *callback)(unsigned int)) +{ + spu.scheduleCallback = callback; } // COMMON PLUGIN INFO FUNCS @@ -1173,18 +1146,18 @@ void spu_get_debug_info(int *chans_out, int *run_chans, int *fmod_chans_out, int for(;ch