X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=plugins%2Fdfsound%2Fspu.c;h=d6cd952bb5d95447a1fa698320ad714be8ff73a6;hp=7e6cfb2ba1072313be348e1aaf099f3180dd3b36;hb=1775933abd26d2e282c4e0b4093778d204a0038d;hpb=ee849648a1bd5d12bd48c98f3b51bb96d7984d43 diff --git a/plugins/dfsound/spu.c b/plugins/dfsound/spu.c index 7e6cfb2b..d6cd952b 100644 --- a/plugins/dfsound/spu.c +++ b/plugins/dfsound/spu.c @@ -4,6 +4,9 @@ begin : Wed May 15 2002 copyright : (C) 2002 by Pete Bernert email : BlackDove@addcom.de + + Portions (C) Gražvydas "notaz" Ignotas, 2010-2011 + ***************************************************************************/ /*************************************************************************** * * @@ -20,6 +23,7 @@ #define _IN_SPU #include "externals.h" +#include "registers.h" #include "cfg.h" #include "dsoundoss.h" #include "regs.h" @@ -34,6 +38,17 @@ #define N_(x) (x) #endif +#ifdef __arm__ + #define ssat32_to_16(v) \ + asm("ssat %0,#16,%1" : "=r" (v) : "r" (v)) +#else + #define ssat32_to_16(v) do { \ + if (v < -32768) v = -32768; \ + else if (v > 32767) v = 32767; \ + } while (0) +#endif + +/* #if defined (USEMACOSX) static char * libraryName = N_("Mac OS X Sound"); #elif defined (USEALSA) @@ -49,6 +64,7 @@ static char * libraryName = N_("NULL Sound"); #endif static char * libraryInfo = N_("P.E.Op.S. Sound Driver V1.7\nCoded by Pete Bernert and the P.E.Op.S. team\n"); +*/ // globals @@ -71,7 +87,6 @@ int iDebugMode=0; int iRecordMode=0; int iUseReverb=2; int iUseInterpolation=2; -int iDisStereo=0; // MAIN infos struct for each channel @@ -93,19 +108,21 @@ int bSPUIsOpen=0; static pthread_t thread = (pthread_t)-1; // thread id (linux) unsigned long dwNewChannel=0; // flags for faster testing, if new channel starts +unsigned long dwChannelOn=0; +unsigned long dwPendingChanOff=0; 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[5][2] = { { 0, 0 }, +static const int f[8][2] = { { 0, 0 }, { 60, 0 }, { 115, -52 }, { 98, -55 }, { 122, -60 } }; -int SSumR[NSSIZE]; -int SSumL[NSSIZE]; +int ChanBuf[NSSIZE]; +int SSumLR[NSSIZE*2]; int iFMod[NSSIZE]; int iCycle = 0; short * pS; @@ -245,16 +262,15 @@ INLINE void StartSound(int ch) StartADSR(ch); StartREVERB(ch); - s_chan[ch].pCurr=s_chan[ch].pStart; // set sample start + // fussy timing issues - do in VoiceOn + //s_chan[ch].pCurr=s_chan[ch].pStart; // set sample start + //s_chan[ch].bStop=0; + //s_chan[ch].bOn=1; s_chan[ch].s_1=0; // init mixing vars s_chan[ch].s_2=0; s_chan[ch].iSBPos=28; - s_chan[ch].bNew=0; // init channel flags - s_chan[ch].bStop=0; - s_chan[ch].bOn=1; - s_chan[ch].SB[29]=0; // init our interpolation helpers s_chan[ch].SB[30]=0; @@ -279,9 +295,10 @@ INLINE void VoiceChangeFrequency(int ch) //////////////////////////////////////////////////////////////////////// -INLINE void FModChangeFrequency(int ch,int ns) +INLINE int FModChangeFrequency(int ch,int ns) { int NP=s_chan[ch].iRawPitch; + int sinc; NP=((32768L+iFMod[ns])*NP)/32768L; @@ -292,11 +309,13 @@ INLINE void FModChangeFrequency(int ch,int ns) s_chan[ch].iActFreq=NP; s_chan[ch].iUsedFreq=NP; - s_chan[ch].sinc=(((NP/10)<<16)/4410); - if(!s_chan[ch].sinc) s_chan[ch].sinc=1; + sinc=(((NP/10)<<16)/4410); + if(!sinc) sinc=1; if(iUseInterpolation==1) // freq change in simple interpolation mode - s_chan[ch].SB[32]=1; + s_chan[ch].SB[32]=1; iFMod[ns]=0; + + return sinc; } //////////////////////////////////////////////////////////////////////// @@ -423,6 +442,210 @@ INLINE int iGetInterpolationVal(int ch) return fa; } +static void do_irq(void) +{ + if(!(spuStat & STAT_IRQ)) + { + spuStat |= STAT_IRQ; + if(irqCallback) irqCallback(); + } +} + +static int decode_block(int ch) +{ + unsigned char *start; + unsigned int nSample; + int predict_nr,shift_factor,flags,d,s; + int fa,s_1,s_2; + int ret = 0; + + s_chan[ch].iSBPos=0; + + start=s_chan[ch].pCurr; // set up the current pos + if(start == (unsigned char*)-1 || // special "stop" sign + (dwPendingChanOff&(1< turn everything off + dwPendingChanOff&=~(1< and done for this channel + } + + //////////////////////////////////////////// irq check + + if(spuCtrl&CTRL_IRQ) + { + if(pSpuIrq == start) // irq address reached? + { + do_irq(); // -> call main emu + ret = 1; + } + } + + s_1=s_chan[ch].s_1; + s_2=s_chan[ch].s_2; + + predict_nr=(int)*start;start++; + shift_factor=predict_nr&0xf; + predict_nr >>= 4; + flags=(int)*start;start++; + + // -------------------------------------- // + + for (nSample=0;nSample<28;start++) + { + d=(int)*start; + s=((d&0xf)<<12); + if(s&0x8000) s|=0xffff0000; + + fa=(s >> shift_factor); + fa=fa + ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6); + s_2=s_1;s_1=fa; + s=((d & 0xf0) << 8); + + s_chan[ch].SB[nSample++]=fa; + + if(s&0x8000) s|=0xffff0000; + fa=(s>>shift_factor); + fa=fa + ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6); + s_2=s_1;s_1=fa; + + s_chan[ch].SB[nSample++]=fa; + } + + //////////////////////////////////////////// flag handler + + if((flags&4) && (!s_chan[ch].bIgnoreLoop)) + s_chan[ch].pLoop=start-16; // loop adress + + if(flags&1) // 1: stop/loop + { + if(!(flags&2)) + dwPendingChanOff|=1<= 0x80000) + start = (unsigned char*)-1; + + s_chan[ch].pCurr=start; // store values for next cycle + s_chan[ch].s_1=s_1; + s_chan[ch].s_2=s_2; + + return ret; +} + +// do block, but ignore sample data +static int skip_block(int ch) +{ + unsigned char *start = s_chan[ch].pCurr; + int flags = start[1]; + int ret = 0; + + // Tron Bonne hack, probably wrong (could be wrong memory contents..) + if(flags & ~7) flags = 0; + + if(start == pSpuIrq) + { + do_irq(); + ret = 1; + } + + if((flags & 4) && !s_chan[ch].bIgnoreLoop) + s_chan[ch].pLoop=start; + + s_chan[ch].pCurr += 16; + + if(flags & 1) + s_chan[ch].pCurr = s_chan[ch].pLoop; + + return ret; +} + +#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) \ +{ \ + int sinc = s_chan[ch].sinc; \ + int spos = s_chan[ch].spos; \ + int ret = -1; \ + int d, fa; \ + interp_start; \ + \ + for (; ns < ns_to; ns++) \ + { \ + fmod_code; \ + \ + while (spos >= 0x10000) \ + { \ + if(s_chan[ch].iSBPos == 28) \ + { \ + d = decode_block(ch); \ + if(d && iSPUIRQWait) \ + { \ + ret = ns; \ + goto out; \ + } \ + } \ + \ + fa = s_chan[ch].SB[s_chan[ch].iSBPos++]; \ + interp1_code; \ + spos -= 0x10000; \ + } \ + \ + interp2_code; \ + spos += sinc; \ + } \ + \ +out: \ + s_chan[ch].sinc = sinc; \ + s_chan[ch].spos = spos; \ + interp_end; \ + \ + return ret; \ +} + +#define fmod_recv_check \ + if(s_chan[ch].bFMod==1 && iFMod[ns]) \ + sinc = FModChangeFrequency(ch,ns) + +make_do_samples(default, fmod_recv_check, , + StoreInterpolationVal(ch, fa), + ChanBuf[ns] = iGetInterpolationVal(ch), ) +make_do_samples(noint, , fa = s_chan[ch].SB[29], , ChanBuf[ns] = fa, s_chan[ch].SB[29] = fa) + +#define simple_interp_store \ + s_chan[ch].SB[28] = 0; \ + s_chan[ch].SB[29] = s_chan[ch].SB[30]; \ + s_chan[ch].SB[30] = s_chan[ch].SB[31]; \ + s_chan[ch].SB[31] = fa; \ + s_chan[ch].SB[32] = 1 + +#define simple_interp_get \ + if(sinc<0x10000) /* -> upsampling? */ \ + InterpolateUp(ch); /* --> interpolate up */ \ + else InterpolateDown(ch); /* --> else down */ \ + ChanBuf[ns] = s_chan[ch].SB[29] + +make_do_samples(simple, , , + simple_interp_store, simple_interp_get, ) + +static int do_samples_noise(int ch, int ns, int ns_to) +{ + s_chan[ch].spos += s_chan[ch].sinc * (ns_to - ns); + while (s_chan[ch].spos >= 28*0x10000) + { + skip_block(ch); + s_chan[ch].spos -= 28*0x10000; + } + + for (; ns < ns_to; ns++) + ChanBuf[ns] = iGetNoiseVal(ch); + + return -1; +} + //////////////////////////////////////////////////////////////////////// // MAIN SPU FUNCTION // here is the main job handler... thread, timer or direct func call @@ -440,14 +663,13 @@ INLINE int iGetInterpolationVal(int ch) static void *MAINThread(void *arg) { - int s_1,s_2,fa,ns; + int ns,ns_from,ns_to; #if !defined(_MACOSX) && !defined(__arm__) int voldiv = iVolume; #else const int voldiv = 2; #endif - unsigned char * start;unsigned int nSample; - int ch,predict_nr,shift_factor,flags,d,s; + int ch,d; int bIRQReturn=0; while(!bEndThread) // until we are shutting down @@ -479,190 +701,122 @@ static void *MAINThread(void *arg) //--------------------------------------------------// continue from irq handling in timer mode? + ns_from=0; + ns_to=NSSIZE; + ch=0; if(lastch>=0) // will be -1 if no continue is pending { - ch=lastch; ns=lastns; lastch=-1; // -> setup all kind of vars to continue - goto GOON; // -> directly jump to the continue point + ch=lastch; ns_from=lastns; lastch=-1; // -> setup all kind of vars to continue } //--------------------------------------------------// //- main channel loop -// //--------------------------------------------------// { - for(ch=0;ch=0) { - if(s_chan[ch].bFMod==1 && iFMod[ns]) // fmod freq channel - FModChangeFrequency(ch,ns); - - while(s_chan[ch].spos>=0x10000L) - { - if(s_chan[ch].iSBPos==28) // 28 reached? - { - start=s_chan[ch].pCurr; // set up the current pos - - if (start == (unsigned char*)-1) // special "stop" sign - { - s_chan[ch].bOn=0; // -> turn everything off - s_chan[ch].ADSRX.lVolume=0; - s_chan[ch].ADSRX.EnvelopeVol=0; - goto ENDX; // -> and done for this channel - } - - s_chan[ch].iSBPos=0; - - //////////////////////////////////////////// spu irq handler here? mmm... do it later - - s_1=s_chan[ch].s_1; - s_2=s_chan[ch].s_2; - - predict_nr=(int)*start;start++; - shift_factor=predict_nr&0xf; - predict_nr >>= 4; - flags=(int)*start;start++; - - // -------------------------------------- // - - for (nSample=0;nSample<28;start++) - { - d=(int)*start; - s=((d&0xf)<<12); - if(s&0x8000) s|=0xffff0000; - - fa=(s >> shift_factor); - fa=fa + ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6); - s_2=s_1;s_1=fa; - s=((d & 0xf0) << 8); - - s_chan[ch].SB[nSample++]=fa; - - if(s&0x8000) s|=0xffff0000; - fa=(s>>shift_factor); - fa=fa + ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6); - s_2=s_1;s_1=fa; - - s_chan[ch].SB[nSample++]=fa; - } - - //////////////////////////////////////////// irq check - - if(irqCallback && (spuCtrl&0x40)) // some callback and irq active? - { - if((pSpuIrq > start-16 && // irq address reached? - pSpuIrq <= start) || - ((flags&1) && // special: irq on looping addr, when stop/loop flag is set - (pSpuIrq > s_chan[ch].pLoop-16 && - pSpuIrq <= s_chan[ch].pLoop))) - { - s_chan[ch].iIrqDone=1; // -> debug flag - irqCallback(); // -> call main emu - - if(iSPUIRQWait) // -> option: wait after irq for main emu - { - iSpuAsyncWait=1; - bIRQReturn=1; - } - } - } - - //////////////////////////////////////////// flag handler - - if((flags&4) && (!s_chan[ch].bIgnoreLoop)) - s_chan[ch].pLoop=start-16; // loop adress - - if(flags&1) // 1: stop/loop - { - // We play this block out first... - //if(!(flags&2)) // 1+2: do loop... otherwise: stop - if(flags!=3 || s_chan[ch].pLoop==NULL) // PETE: if we don't check exactly for 3, loop hang ups will happen (DQ4, for example) - { // and checking if pLoop is set avoids crashes, yeah - start = (unsigned char*)-1; - } - else - { - start = s_chan[ch].pLoop; - } - } - - s_chan[ch].pCurr=start; // store values for next cycle - s_chan[ch].s_1=s_1; - s_chan[ch].s_2=s_2; - - if(bIRQReturn) // special return for "spu irq - wait for cpu action" - { - bIRQReturn=0; - if(iUseTimer!=2) - { - DWORD dwWatchTime=timeGetTime_spu()+2500; - - while(iSpuAsyncWait && !bEndThread && - timeGetTime_spu() store 1T sample data, use that to do fmod on next channel - else // no fmod freq channel + for(ns=ns_from;ns>14; + r=(sval*rv)>>14; + SSumLR[ns*2] +=l; + SSumLR[ns*2+1]+=r; ////////////////////////////////////////////// // now let us store sound data for reverb - if(s_chan[ch].bRVBActive) StoreREVERB(ch,ns); + if(s_chan[ch].bRVBActive) StoreREVERB(ch,ns,l,r); } + } + } + } + + // advance "stopped" channels that can cause irqs + // (all chans are always playing on the real thing..) + if(!bIRQReturn && (spuCtrl&CTRL_IRQ)) + for(ch=0;ch pSpuIrq && s_chan[ch].pLoop > pSpuIrq) + continue; - //////////////////////////////////////////////// - // ok, go on until 1 ms data of this channel is collected + if(s_chan[ch].iActFreq!=s_chan[ch].iUsedFreq) // new psx frequency? + VoiceChangeFrequency(ch); - ns++; - s_chan[ch].spos += s_chan[ch].sinc; + s_chan[ch].spos += s_chan[ch].sinc * NSSIZE; + while(s_chan[ch].spos >= 28 * 0x10000) + { + unsigned char *start=s_chan[ch].pCurr; + bIRQReturn |= skip_block(ch); + if(start == s_chan[ch].pCurr) + { + // looping on self + s_chan[ch].pCurr=(unsigned char *)-1; + break; + } + + s_chan[ch].spos -= 28 * 0x10000; } -ENDX: ; } - } + + if(bIRQReturn && iSPUIRQWait) // special return for "spu irq - wait for cpu action" + { + iSpuAsyncWait=1; + bIRQReturn=0; + if(iUseTimer!=2) + { + DWORD dwWatchTime=timeGetTime_spu()+2500; + + while(iSpuAsyncWait && !bEndThread && + timeGetTime_spu() 32767) dl = 32767; + if(iUseReverb) + REVERBDo(); - SSumR[ns] += MixREVERBRight(); - - dr = SSumR[ns] / voldiv; SSumR[ns] = 0; - if (dr < -32767) dr = -32767; if (dr > 32767) dr = 32767; - *pS++ = (dl + dr) / 2; - } + if((spuCtrl&0x4000)==0) // muted? (rare, don't optimize for this) + { + memset(pS, 0, NSSIZE * 2 * sizeof(pS[0])); + pS += NSSIZE*2; } - else // stereo: - for (ns = 0; ns < NSSIZE; ns++) + else + for (ns = 0; ns < NSSIZE*2; ) { - SSumL[ns] += MixREVERBLeft(ns); - - d = SSumL[ns] / voldiv; SSumL[ns] = 0; - if (d < -32767) d = -32767; if (d > 32767) d = 32767; + d = SSumLR[ns] / voldiv; SSumLR[ns] = 0; + ssat32_to_16(d); *pS++ = d; + ns++; - SSumR[ns] += MixREVERBRight(); - - d = SSumR[ns] / voldiv; SSumR[ns] = 0; - if(d < -32767) d = -32767; if(d > 32767) d = 32767; + d = SSumLR[ns] / voldiv; SSumLR[ns] = 0; + ssat32_to_16(d); *pS++ = d; + ns++; } ////////////////////////////////////////////////////// @@ -725,7 +868,7 @@ ENDX: ; // 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). - if(pMixIrq && irqCallback) + if(pMixIrq) { for(ns=0;ns=pMixIrq+(ch*0x400) && pSpuIrqspuMemC+0x3ff) pMixIrq=spuMemC; @@ -769,7 +912,7 @@ void CALLBACK SPUasync(unsigned long cycle) if(iSpuAsyncWait) { iSpuAsyncWait++; - if(iSpuAsyncWait<=64) return; + if(iSpuAsyncWait<=16) return; iSpuAsyncWait=0; } @@ -778,6 +921,11 @@ void CALLBACK SPUasync(unsigned long cycle) if(!bSpuInit) return; // -> no init, no call MAINThread(0); // -> linux high-compat mode + + // 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; } } @@ -817,8 +965,7 @@ void CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes) // SETUPTIMER: init of certain buffers and threads/timers void SetupTimer(void) { - memset(SSumR,0,NSSIZE*sizeof(int)); // init some mixing buffers - memset(SSumL,0,NSSIZE*sizeof(int)); + memset(SSumLR,0,sizeof(SSumLR)); // init some mixing buffers memset(iFMod,0,NSSIZE*sizeof(int)); pS=(short *)pSpuBuffer; // setup soundbuffer pointer @@ -873,16 +1020,14 @@ void SetupStreams(void) (uint32_t *)malloc(16384 * sizeof(uint32_t)); CDDAEnd = CDDAStart + 16384; CDDAPlay = CDDAStart; - CDDAFeed = CDDAStart + 1; + CDDAFeed = CDDAStart; for(i=0;i init sustain - s_chan[i].iMute=0; - s_chan[i].iIrqDone=0; + s_chan[i].ADSRX.SustainLevel = 0xf; // -> init sustain s_chan[i].pLoop=spuMemC; s_chan[i].pStart=spuMemC; s_chan[i].pCurr=spuMemC; @@ -914,7 +1059,6 @@ long CALLBACK SPUinit(void) InitADSR(); iVolume = 3; - iReverbOff = -1; spuIrq = 0; spuAddr = 0xffffffff; bEndThread = 0; @@ -923,7 +1067,7 @@ long CALLBACK SPUinit(void) pMixIrq = 0; memset((void *)s_chan, 0, (MAXCHAN + 1) * sizeof(SPUCHAN)); pSpuIrq = 0; - iSPUIRQWait = 0; + //iSPUIRQWait = 0; lastch = -1; //ReadConfigSPU(); // read user stuff @@ -1029,3 +1173,25 @@ char * SPUgetLibInfos(void) return _(libraryInfo); } */ + +// debug +void spu_get_debug_info(int *chans_out, int *fmod_chans_out, int *noise_chans_out) +{ + int ch = 0, fmod_chans = 0, noise_chans = 0; + + for(;ch