X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=plugins%2Fdfsound%2Fspu.c;h=b091efcc96fd0b9d8e6937520d9de861b966a84f;hp=780b09173736f50440824e0679fe8944b23dc319;hb=63a4f6b6a3b0315590cd3009df2c92480ed2d98b;hpb=78c60846942cad7d41d6d23a7c3d2b905ee4b822 diff --git a/plugins/dfsound/spu.c b/plugins/dfsound/spu.c index 780b0917..b091efcc 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, 2010-2012,2014,2015 + ***************************************************************************/ /*************************************************************************** * * @@ -15,26 +18,35 @@ * * ***************************************************************************/ +#ifndef _WIN32 +#include // gettimeofday in xa.c +#define THREAD_ENABLED 1 +#endif #include "stdafx.h" #define _IN_SPU #include "externals.h" #include "registers.h" -#include "cfg.h" -#include "dsoundoss.h" -#include "regs.h" - -#ifdef ENABLE_NLS -#include -#include -#define _(x) gettext(x) -#define N_(x) (x) +#include "out.h" +#include "arm_features.h" +#include "spu_config.h" + +#ifdef __ARM_ARCH_7A__ + #define ssat32_to_16(v) \ + asm("ssat %0,#16,%1" : "=r" (v) : "r" (v)) #else -#define _(x) (x) -#define N_(x) (x) + #define ssat32_to_16(v) do { \ + if (v < -32768) v = -32768; \ + else if (v > 32767) v = 32767; \ + } 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"); @@ -55,51 +67,51 @@ 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=3; -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 long dwNoiseVal=1; // global noise generator -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) +#ifdef THREAD_ENABLED + +#include +#include +#include + +// worker thread state +static struct { + unsigned int pending:1; + unsigned int exit_thread:1; + int ns_to; + int ctrl; + int decode_pos; + int silentch; + int *sRVBStart; + unsigned char *ram; + unsigned int chmask; + unsigned int r_chan_end; + unsigned int r_decode_dirty; + pthread_t thread; + sem_t sem_avail; + sem_t sem_done; + struct { + int spos; + int sbpos; + int sinc; + int start; + int loop; + int ns_to; + ADSRInfoEx adsr; + // might want to add vol and fmod flags.. + } ch[24]; +} *worker; -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; +#else +static const void * const worker = NULL; +#endif // certain globals (were local before, but with the new timeproc I need em global) @@ -108,14 +120,11 @@ static const int f[8][2] = { { 0, 0 }, { 115, -52 }, { 98, -55 }, { 122, -60 } }; +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 @@ -168,70 +177,70 @@ static int iSecureStart=0; // secure start counter // / // - -INLINE void InterpolateUp(int ch) +static void InterpolateUp(int *SB, int sinc) { - if(s_chan[ch].SB[32]==1) // flag == 1? calc step and set flag... and don't change the value in this pass + if(SB[32]==1) // flag == 1? calc step and set flag... and don't change the value in this pass { - const int id1=s_chan[ch].SB[30]-s_chan[ch].SB[29]; // curr delta to next val - const int id2=s_chan[ch].SB[31]-s_chan[ch].SB[30]; // and next delta to next-next val :) + const int id1=SB[30]-SB[29]; // curr delta to next val + const int id2=SB[31]-SB[30]; // and next delta to next-next val :) - s_chan[ch].SB[32]=0; + SB[32]=0; if(id1>0) // curr delta positive { if(id2>16; else - s_chan[ch].SB[28]=(id1*s_chan[ch].sinc)/0x20000L; + SB[28]=(id1*sinc)>>17; } else // curr delta negative { if(id2>id1) - {s_chan[ch].SB[28]=id1;s_chan[ch].SB[32]=2;} + {SB[28]=id1;SB[32]=2;} else if(id2>(id1<<1)) - s_chan[ch].SB[28]=(id1*s_chan[ch].sinc)/0x10000L; + SB[28]=(id1*sinc)>>16; else - s_chan[ch].SB[28]=(id1*s_chan[ch].sinc)/0x20000L; + SB[28]=(id1*sinc)>>17; } } else - if(s_chan[ch].SB[32]==2) // flag 1: calc step and set flag... and don't change the value in this pass + if(SB[32]==2) // flag 1: calc step and set flag... and don't change the value in this pass { - s_chan[ch].SB[32]=0; + 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]; + SB[28]=(SB[28]*sinc)>>17; + //if(sinc<=0x8000) + // SB[29]=SB[30]-(SB[28]*((0x10000/sinc)-1)); + //else + SB[29]+=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]; + SB[29]+=SB[28]; } // // even easier interpolation on downsampling, also no special filter, again just "Pete's common sense" tm // -INLINE void InterpolateDown(int ch) +static void InterpolateDown(int *SB, int sinc) { - if(s_chan[ch].sinc>=0x20000L) // we would skip at least one val? + if(sinc>=0x20000L) // we would skip at least one val? { - s_chan[ch].SB[29]+=(s_chan[ch].SB[30]-s_chan[ch].SB[29])/2; // add easy weight - if(s_chan[ch].sinc>=0x30000L) // we would skip even more vals? - s_chan[ch].SB[29]+=(s_chan[ch].SB[31]-s_chan[ch].SB[30])/2;// add additional next weight + SB[29]+=(SB[30]-SB[29])/2; // add easy weight + if(sinc>=0x30000L) // we would skip even more vals? + SB[29]+=(SB[31]-SB[30])/2; // add additional next weight } } //////////////////////////////////////////////////////////////////////// // 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*)(&SB[29]))[gpos&3]) +#define gval(x) ((int)((short*)(&SB[29]))[(gpos+x)&3]) #include "gauss_i.h" @@ -239,6 +248,26 @@ 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 //////////////////////////////////////////////////////////////////////// @@ -248,139 +277,91 @@ INLINE void StartSound(int ch) StartADSR(ch); StartREVERB(ch); - // 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].prevflags=2; - s_chan[ch].s_1=0; // init mixing vars - s_chan[ch].s_2=0; - s_chan[ch].iSBPos=28; + s_chan[ch].SB[26]=0; // init mixing vars + s_chan[ch].SB[27]=0; + s_chan[ch].iSBPos=27; + s_chan[ch].SB[28]=0; s_chan[ch].SB[29]=0; // init our interpolation helpers s_chan[ch].SB[30]=0; + s_chan[ch].SB[31]=0; + s_chan[ch].spos=0; - if(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< take it and calc steps - s_chan[ch].sinc=s_chan[ch].iRawPitch<<4; - if(!s_chan[ch].sinc) s_chan[ch].sinc=1; - if(iUseInterpolation==1) s_chan[ch].SB[32]=1; // -> freq change in simle imterpolation mode: set flag -} - -//////////////////////////////////////////////////////////////////////// - -INLINE void FModChangeFrequency(int ch,int ns) +INLINE int FModChangeFrequency(int *SB, int pitch, int ns) { - int NP=s_chan[ch].iRawPitch; + unsigned int NP=pitch; + int sinc; - NP=((32768L+iFMod[ns])*NP)/32768L; + NP=((32768L+iFMod[ns])*NP)>>15; if(NP>0x3fff) NP=0x3fff; if(NP<0x1) NP=0x1; - NP=(44100L*NP)/(4096L); // calc frequency - - 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; - if(iUseInterpolation==1) // freq change in simple interpolation mode - s_chan[ch].SB[32]=1; + sinc=NP<<4; // calc frequency + if(spu_config.iUseInterpolation==1) // freq change in simple interpolation mode + SB[32]=1; iFMod[ns]=0; -} - -//////////////////////////////////////////////////////////////////////// - -// noise handler... just produces some noise data -// surely wrong... and no noise frequency (spuCtrl&0x3f00) will be used... -// and sometimes the noise will be used as fmod modulation... pfff - -INLINE int iGetNoiseVal(int ch) -{ - int fa; - if((dwNoiseVal<<=1)&0x80000000L) - { - dwNoiseVal^=0x0040001L; - fa=((dwNoiseVal>>2)&0x7fff); - fa=-fa; - } - else fa=(dwNoiseVal>>2)&0x7fff; - - // mmm... depending on the noise freq we allow bigger/smaller changes to the previous val - fa=s_chan[ch].iOldNoise+((fa-s_chan[ch].iOldNoise)/((0x001f-((spuCtrl&0x3f00)>>9))+1)); - if(fa>32767L) fa=32767L; - if(fa<-32767L) fa=-32767L; - s_chan[ch].iOldNoise=fa; - - if(iUseInterpolation<2) // no gauss/cubic interpolation? - s_chan[ch].SB[29] = fa; // -> store noise val in "current sample" slot - return fa; -} + return sinc; +} //////////////////////////////////////////////////////////////////////// -INLINE void StoreInterpolationVal(int ch,int fa) +INLINE void StoreInterpolationVal(int *SB, int sinc, int fa, int fmod_freq) { - if(s_chan[ch].bFMod==2) // fmod freq channel - s_chan[ch].SB[29]=fa; + if(fmod_freq) // fmod freq channel + SB[29]=fa; else { - if((spuCtrl&0x4000)==0) fa=0; // muted? - else // else adjust - { - if(fa>32767L) fa=32767L; - if(fa<-32767L) fa=-32767L; - } + ssat32_to_16(fa); - if(iUseInterpolation>=2) // gauss/cubic interpolation - { - int gpos = s_chan[ch].SB[28]; - gval0 = fa; + if(spu_config.iUseInterpolation>=2) // gauss/cubic interpolation + { + int gpos = SB[28]; + gval0 = fa; gpos = (gpos+1) & 3; - s_chan[ch].SB[28] = gpos; + 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' - s_chan[ch].SB[30] = s_chan[ch].SB[31]; - s_chan[ch].SB[31] = fa; - s_chan[ch].SB[32] = 1; // -> flag: calc new interolation + SB[28] = 0; + 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' + SB[30] = SB[31]; + SB[31] = fa; + SB[32] = 1; // -> flag: calc new interolation } - else s_chan[ch].SB[29]=fa; // no interpolation + else SB[29]=fa; // no interpolation } } //////////////////////////////////////////////////////////////////////// -INLINE int iGetInterpolationVal(int ch) +INLINE int iGetInterpolationVal(int *SB, int sinc, int spos, int fmod_freq) { int fa; - if(s_chan[ch].bFMod==2) return s_chan[ch].SB[29]; + if(fmod_freq) return SB[29]; - switch(iUseInterpolation) - { + switch(spu_config.iUseInterpolation) + { //--------------------------------------------------// case 3: // cubic interpolation { long xd;int gpos; - xd = ((s_chan[ch].spos) >> 1)+1; - gpos = s_chan[ch].SB[28]; + xd = (spos >> 1)+1; + gpos = SB[28]; fa = gval(3) - 3*gval(2) + 3*gval(1) - gval0; fa *= (xd - (2<<15)) / 6; @@ -398,9 +379,9 @@ INLINE int iGetInterpolationVal(int ch) case 2: // gauss interpolation { int vl, vr;int gpos; - vl = (s_chan[ch].spos >> 6) & ~3; - gpos = s_chan[ch].SB[28]; - vr=(gauss[vl]*gval0)&~2047; + vl = (spos >> 6) & ~3; + gpos = SB[28]; + 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; @@ -409,15 +390,15 @@ INLINE int iGetInterpolationVal(int ch) //--------------------------------------------------// case 1: // simple interpolation { - if(s_chan[ch].sinc<0x10000L) // -> upsampling? - InterpolateUp(ch); // --> interpolate up - else InterpolateDown(ch); // --> else down - fa=s_chan[ch].SB[29]; + if(sinc<0x10000L) // -> upsampling? + InterpolateUp(SB, sinc); // --> interpolate up + else InterpolateDown(SB, sinc); // --> else down + fa=SB[29]; } break; //--------------------------------------------------// default: // no interpolation { - fa=s_chan[ch].SB[29]; + fa=SB[29]; } break; //--------------------------------------------------// } @@ -425,408 +406,789 @@ INLINE int iGetInterpolationVal(int ch) return fa; } -static void do_irq(void) +static void decode_block_data(int *dest, const unsigned char *src, int predict_nr, int shift_factor) { - if(!(spuStat & STAT_IRQ)) + int nSample; + int fa, s_1, s_2, d, s; + + s_1 = dest[27]; + s_2 = dest[26]; + + for (nSample = 0; nSample < 28; src++) { - spuStat |= STAT_IRQ; - if(irqCallback) irqCallback(); + d = (int)*src; + s = (int)(signed short)((d & 0x0f) << 12); + + fa = s >> shift_factor; + fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6); + s_2=s_1;s_1=fa; + + dest[nSample++] = fa; + + s = (int)(signed short)((d & 0xf0) << 8); + fa = s >> shift_factor; + fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6); + s_2=s_1;s_1=fa; + + dest[nSample++] = fa; } } -static int decode_block(int ch) +static int decode_block(int ch, int *SB) { unsigned char *start; - unsigned int nSample; - int predict_nr,shift_factor,flags,d,s; - int fa,s_1,s_2; + int predict_nr, shift_factor, flags; int ret = 0; - s_chan[ch].iSBPos=0; + start = s_chan[ch].pCurr; // set up the current pos + if (start == spu.spuMemC) // ? + ret = 1; - 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 + if (!(s_chan[ch].prevflags & 2)) + ret = 1; + + start = s_chan[ch].pLoop; } + else + check_irq(ch, start); // hack, see check_irq below.. - //////////////////////////////////////////// irq check + predict_nr = (int)start[0]; + shift_factor = predict_nr & 0xf; + predict_nr >>= 4; - if(spuCtrl&CTRL_IRQ) - { - if(pSpuIrq == start) // irq address reached? - { - do_irq(); // -> call main emu + decode_block_data(SB, start + 2, predict_nr, shift_factor); + + flags = start[1]; + if (flags & 4) + s_chan[ch].pLoop = start; // loop adress + + start += 16; + + if (flags & 1) { // 1: stop/loop + start = s_chan[ch].pLoop; + check_irq(ch, start); // hack.. :( + } + + if (start - spu.spuMemC >= 0x80000) + start = spu.spuMemC; + + s_chan[ch].pCurr = start; // store values for next cycle + s_chan[ch].prevflags = flags; + + return ret; +} + +// do block, but ignore sample data +static int skip_block(int ch) +{ + unsigned char *start = s_chan[ch].pCurr; + int flags; + int ret = 0; + + if (s_chan[ch].prevflags & 1) { + if (!(s_chan[ch].prevflags & 2)) ret = 1; - } + + start = s_chan[ch].pLoop; + } + else + check_irq(ch, start); + + flags = start[1]; + if (flags & 4) + s_chan[ch].pLoop = start; + + start += 16; + + if (flags & 1) { + start = s_chan[ch].pLoop; + check_irq(ch, start); } - s_1=s_chan[ch].s_1; - s_2=s_chan[ch].s_2; + s_chan[ch].pCurr = start; + s_chan[ch].prevflags = flags; + + return ret; +} + +#ifdef THREAD_ENABLED + +static int decode_block_work(int ch, int *SB) +{ + int predict_nr, shift_factor, flags; + const unsigned char *ram = worker->ram; + int start = worker->ch[ch].start; + int loop = worker->ch[ch].loop; - predict_nr=(int)*start;start++; - shift_factor=predict_nr&0xf; + predict_nr = ram[start]; + shift_factor = predict_nr & 0xf; predict_nr >>= 4; - flags=(int)*start;start++; - // -------------------------------------- // + decode_block_data(SB, ram + start + 2, predict_nr, shift_factor); - for (nSample=0;nSample<28;start++) - { - d=(int)*start; - s=((d&0xf)<<12); - if(s&0x8000) s|=0xffff0000; + flags = ram[start + 1]; + if (flags & 4) + loop = start; // loop adress - 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); + start += 16; - s_chan[ch].SB[nSample++]=fa; + if (flags & 1) // 1: stop/loop + start = loop; - 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; + worker->ch[ch].start = start & 0x7ffff; + worker->ch[ch].loop = loop; - s_chan[ch].SB[nSample++]=fa; - } + return 0; +} - //////////////////////////////////////////// flag handler +#endif - if((flags&4) && (!s_chan[ch].bIgnoreLoop)) - s_chan[ch].pLoop=start-16; // loop adress +// 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; - if(flags&1) // 1: stop/loop + 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(!(flags&2)) - dwPendingChanOff|=1<> 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 noinline int do_samples_##name(int (*decode_f)(int ch, int *SB), int ch, \ + int ns_to, int *SB, int sinc, int *spos, int *sbpos) \ +{ \ + int ns, d, fa; \ + int ret = ns_to; \ + interp_start; \ + \ + for (ns = 0; ns < ns_to; ns++) \ + { \ + fmod_code; \ + \ + *spos += sinc; \ + while (*spos >= 0x10000) \ + { \ + fa = SB[(*sbpos)++]; \ + if (*sbpos >= 28) \ + { \ + *sbpos = 0; \ + d = decode_f(ch, SB); \ + if (d && ns < ret) \ + ret = ns; \ + } \ + \ + interp1_code; \ + *spos -= 0x10000; \ + } \ + \ + interp2_code; \ + } \ + \ + interp_end; \ + \ + return ret; \ +} + +#define fmod_recv_check \ + if(s_chan[ch].bFMod==1 && iFMod[ns]) \ + sinc = FModChangeFrequency(SB, s_chan[ch].iRawPitch, ns) + +make_do_samples(default, fmod_recv_check, , + StoreInterpolationVal(SB, sinc, fa, s_chan[ch].bFMod==2), + ChanBuf[ns] = iGetInterpolationVal(SB, sinc, *spos, s_chan[ch].bFMod==2), ) +make_do_samples(noint, , fa = SB[29], , ChanBuf[ns] = fa, SB[29] = fa) + +#define simple_interp_store \ + SB[28] = 0; \ + SB[29] = SB[30]; \ + SB[30] = SB[31]; \ + SB[31] = fa; \ + SB[32] = 1 + +#define simple_interp_get \ + if(sinc<0x10000) /* -> upsampling? */ \ + InterpolateUp(SB, sinc); /* --> interpolate up */ \ + else InterpolateDown(SB, sinc); /* --> else down */ \ + ChanBuf[ns] = SB[29] - if (start - spuMemC >= 0x80000) - start = (unsigned char*)-1; +make_do_samples(simple, , , + simple_interp_store, simple_interp_get, ) + +static int do_samples_skip(int ch, int ns_to) +{ + int ret = ns_to, ns, d; + + s_chan[ch].spos += s_chan[ch].iSBPos << 16; + + for (ns = 0; ns < ns_to; ns++) + { + s_chan[ch].spos += s_chan[ch].sinc; + while (s_chan[ch].spos >= 28*0x10000) + { + d = skip_block(ch); + if (d && ns < ret) + ret = ns; + s_chan[ch].spos -= 28*0x10000; + } + } - s_chan[ch].pCurr=start; // store values for next cycle - s_chan[ch].s_1=s_1; - s_chan[ch].s_2=s_2; + s_chan[ch].iSBPos = s_chan[ch].spos >> 16; + s_chan[ch].spos &= 0xffff; return ret; } -//////////////////////////////////////////////////////////////////////// -// 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! -//////////////////////////////////////////////////////////////////////// +static void do_lsfr_samples(int ns_to, int ctrl, + unsigned int *dwNoiseCount, unsigned int *dwNoiseVal) +{ + unsigned int counter = *dwNoiseCount; + unsigned int val = *dwNoiseVal; + unsigned int level, shift, bit; + int ns; + + // modified from DrHell/shalma, no fraction + level = (ctrl >> 10) & 0x0f; + level = 0x8000 >> level; -// 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 + for (ns = 0; ns < ns_to; ns++) + { + counter += 2; + if (counter >= level) + { + counter -= level; + shift = (val >> 10) & 0x1f; + bit = (0x69696969 >> shift) & 1; + bit ^= (val >> 15) & 1; + val = (val << 1) | bit; + } -#define PAUSE_W 5 -#define PAUSE_L 5000 + ChanBuf[ns] = (signed short)val; + } -//////////////////////////////////////////////////////////////////////// + *dwNoiseCount = counter; + *dwNoiseVal = val; +} -static void *MAINThread(void *arg) +static int do_samples_noise(int ch, int ns_to) { - int fa,ns,ns_from,ns_to; -#if !defined(_MACOSX) && !defined(__arm__) - int voldiv = iVolume; + int ret; + + ret = do_samples_skip(ch, ns_to); + + do_lsfr_samples(ns_to, spu.spuCtrl, &spu.dwNoiseCount, &spu.dwNoiseVal); + + return ret; +} + +#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, int *rvb); #else - const int voldiv = 2; +static void mix_chan(int start, int count, int lv, int rv) +{ + int *dst = SSumLR + start * 2; + const int *src = ChanBuf + start; + int l, r; + + while (count--) + { + int sval = *src++; + + l = (sval * lv) >> 14; + r = (sval * rv) >> 14; + *dst++ += l; + *dst++ += r; + } +} + +static void mix_chan_rvb(int start, int count, int lv, int rv, int *rvb) +{ + int *dst = SSumLR + start * 2; + int *drvb = rvb + start * 2; + const int *src = ChanBuf + start; + int l, r; + + while (count--) + { + int sval = *src++; + + l = (sval * lv) >> 14; + r = (sval * rv) >> 14; + *dst++ += l; + *dst++ += r; + *drvb++ += l; + *drvb++ += r; + } +} #endif - int ch,d; - int bIRQReturn=0; - while(!bEndThread) // until we are shutting down +// 0x0800-0x0bff Voice 1 +// 0x0c00-0x0fff Voice 3 +static noinline void do_decode_bufs(unsigned short *mem, int which, + int count, int decode_pos) +{ + unsigned short *dst = &mem[0x800/2 + which*0x400/2]; + const int *src = ChanBuf; + int cursor = decode_pos; + + while (count-- > 0) { - // 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 + cursor &= 0x1ff; + dst[cursor] = *src++; + cursor++; + } + + // decode_pos is updated and irqs are checked later, after voice loop +} + +static void do_silent_chans(int ns_to, int silentch) +{ + int ch; - while(!iSecureStart && !bEndThread && // no new start? no thread end? - (SoundGetBytesBuffered()>TESTSIZE)) // and still enuff data in sound buffer? + for (ch = 0; ch < MAXCHAN; ch++) + { + if (!(silentch & (1< spu.pSpuIrq && s_chan[ch].pLoop > spu.pSpuIrq) + continue; + + s_chan[ch].spos += s_chan[ch].iSBPos << 16; + s_chan[ch].iSBPos = 0; + + s_chan[ch].spos += s_chan[ch].sinc * ns_to; + while (s_chan[ch].spos >= 28 * 0x10000) { - iSecureStart=0; // reset secure + unsigned char *start = s_chan[ch].pCurr; - if(iUseTimer) return 0; // linux no-thread mode? bye - usleep(PAUSE_L); // else sleep for x ms (linux) + skip_block(ch); + if (start == s_chan[ch].pCurr || start - spu.spuMemC < 0x1000) + { + // looping on self or stopped(?) + spu.dwChannelDead |= 1<=0) // will be -1 if no continue is pending + mask = spu.dwChannelOn & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) // loop em all... + { + if (!(mask & 1)) continue; // channel not playing? next + + SB = s_chan[ch].SB; + sinc = s_chan[ch].sinc; + + if (s_chan[ch].bNoise) + d = do_samples_noise(ch, ns_to); + else if (s_chan[ch].bFMod == 2 + || (s_chan[ch].bFMod == 0 && spu_config.iUseInterpolation == 0)) + d = do_samples_noint(decode_block, ch, ns_to, + SB, sinc, &s_chan[ch].spos, &s_chan[ch].iSBPos); + else if (s_chan[ch].bFMod == 0 && spu_config.iUseInterpolation == 1) + d = do_samples_simple(decode_block, ch, ns_to, + SB, sinc, &s_chan[ch].spos, &s_chan[ch].iSBPos); + else + d = do_samples_default(decode_block, ch, ns_to, + SB, sinc, &s_chan[ch].spos, &s_chan[ch].iSBPos); + + d = MixADSR(&s_chan[ch].ADSRX, d); + if (d < ns_to) { + spu.dwChannelOn &= ~(1 << ch); + s_chan[ch].ADSRX.EnvelopeVol = 0; + memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0])); + } + + if (ch == 1 || ch == 3) { - ch=lastch; ns_from=lastns; lastch=-1; // -> setup all kind of vars to continue + do_decode_bufs(spu.spuMem, ch/2, ns_to, spu.decode_pos); + spu.decode_dirty_ch |= 1 << ch; } - //--------------------------------------------------// - //- main channel loop -// - //--------------------------------------------------// - { - for(;chns_to = ns_to; + worker->ctrl = spu.spuCtrl; + worker->decode_pos = spu.decode_pos; + worker->silentch = silentch; + worker->sRVBStart = spu.sRVBStart; + worker->ram = spu.spuMemC; + + mask = worker->chmask = spu.dwChannelOn & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) + { + if (!(mask & 1)) continue; + + worker->ch[ch].spos = s_chan[ch].spos; + worker->ch[ch].sbpos = s_chan[ch].iSBPos; + worker->ch[ch].sinc = s_chan[ch].sinc; + worker->ch[ch].adsr = s_chan[ch].ADSRX; + worker->ch[ch].start = s_chan[ch].pCurr - spu.spuMemC; + worker->ch[ch].loop = s_chan[ch].pLoop - spu.spuMemC; + if (s_chan[ch].prevflags & 1) + worker->ch[ch].start = worker->ch[ch].loop; + + worker->ch[ch].ns_to = do_samples_skip(ch, ns_to); + } - while(s_chan[ch].spos>=0x10000L) - { - if(s_chan[ch].iSBPos==28) // 28 reached? - { - d = decode_block(ch); - if(d && iSPUIRQWait) // -> option: wait after irq for main emu - { - bIRQReturn=1; - lastch=ch; - lastns=ns_to=ns; - goto ENDX; // do remaining chans unil this ns - } - } + worker->pending = 1; + sem_post(&worker->sem_avail); +} - fa=s_chan[ch].SB[s_chan[ch].iSBPos++]; // get sample data +static void do_channel_work(void) +{ + unsigned int mask, endmask = 0; + unsigned int decode_dirty_ch = 0; + int *SB, sinc, spos, sbpos; + int d, ch, ns_to; - StoreInterpolationVal(ch,fa); // store val for later interpolation + ns_to = worker->ns_to; + memset(worker->sRVBStart, 0, ns_to * sizeof(worker->sRVBStart[0]) * 2); - s_chan[ch].spos -= 0x10000L; - } + mask = worker->chmask; + for (ch = 0; mask != 0; ch++, mask >>= 1) + { + if (!(mask & 1)) continue; + + d = worker->ch[ch].ns_to; + spos = worker->ch[ch].spos; + sbpos = worker->ch[ch].sbpos; + sinc = worker->ch[ch].sinc; + SB = s_chan[ch].SB; + + if (s_chan[ch].bNoise) + do_lsfr_samples(d, worker->ctrl, &spu.dwNoiseCount, &spu.dwNoiseVal); + else if (s_chan[ch].bFMod == 2 + || (s_chan[ch].bFMod == 0 && spu_config.iUseInterpolation == 0)) + do_samples_noint(decode_block_work, ch, d, SB, sinc, &spos, &sbpos); + else if (s_chan[ch].bFMod == 0 && spu_config.iUseInterpolation == 1) + do_samples_simple(decode_block_work, ch, d, SB, sinc, &spos, &sbpos); + else + do_samples_default(decode_block_work, ch, d, SB, sinc, &spos, &sbpos); - if(s_chan[ch].bNoise) - fa=iGetNoiseVal(ch); // get noise val - else fa=iGetInterpolationVal(ch); // get sample val + d = MixADSR(&worker->ch[ch].adsr, d); + if (d < ns_to) { + endmask |= 1 << ch; + worker->ch[ch].adsr.EnvelopeVol = 0; + memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0])); + } - sval = (MixADSR(ch) * fa) / 1024; // mix adsr + if (ch == 1 || ch == 3) + { + do_decode_bufs((void *)worker->ram, ch/2, ns_to, worker->decode_pos); + decode_dirty_ch |= 1 << ch; + } - if(s_chan[ch].bFMod==2) // fmod freq channel - iFMod[ns]=sval; // -> store 1T sample data, use that to do fmod on next channel - else // no fmod freq channel - { - ////////////////////////////////////////////// - // ok, left/right sound volume (psx volume goes from 0 ... 0x3fff) + if (s_chan[ch].bFMod == 2) // fmod freq channel + memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0])); + if (s_chan[ch].bRVBActive) + mix_chan_rvb(0, ns_to, s_chan[ch].iLeftVolume, s_chan[ch].iRightVolume, worker->sRVBStart); + else + mix_chan(0, ns_to, s_chan[ch].iLeftVolume, s_chan[ch].iRightVolume); + } - SSumLR[ns*2] +=(sval*s_chan[ch].iLeftVolume)/0x4000L; - SSumLR[ns*2+1]+=(sval*s_chan[ch].iRightVolume)/0x4000L; + worker->r_chan_end = endmask; + worker->r_decode_dirty = decode_dirty_ch; +} - ////////////////////////////////////////////// - // now let us store sound data for reverb +static void sync_worker_thread(void) +{ + unsigned int mask; + int ch; - if(s_chan[ch].bRVBActive) StoreREVERB(ch,ns,sval); - } + if (!worker->pending) + return; - //////////////////////////////////////////////// - // ok, go on until 1 ms data of this channel is collected + sem_wait(&worker->sem_done); + worker->pending = 0; - s_chan[ch].spos += s_chan[ch].sinc; - } -ENDX: ; - } - } + mask = worker->chmask; + for (ch = 0; mask != 0; ch++, mask >>= 1) { + if (!(mask & 1)) continue; - // 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; - - if(s_chan[ch].iActFreq!=s_chan[ch].iUsedFreq) // new psx frequency? - VoiceChangeFrequency(ch); - - s_chan[ch].spos += s_chan[ch].sinc * NSSIZE; - while(s_chan[ch].spos >= 28 * 0x10000) - { - unsigned char *start=s_chan[ch].pCurr; - int flags = start[1]; - - // Tron Bonne hack, probably wrong (could be wrong memory contents..) - if(flags & ~7) flags = 0; - - if(start == pSpuIrq) - { - do_irq(); - bIRQReturn = 1; - } - else if((flags & 1) && start == s_chan[ch].pLoop) - { - // looping on self - s_chan[ch].pCurr=(unsigned char *)-1; - break; - } - - 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; - - s_chan[ch].spos -= 28 * 0x10000; - } - } + // be sure there was no keyoff while thread was working + if (s_chan[ch].ADSRX.State != ADSR_RELEASE) + s_chan[ch].ADSRX.State = worker->ch[ch].adsr.State; + s_chan[ch].ADSRX.EnvelopeVol = worker->ch[ch].adsr.EnvelopeVol; + } - 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()r_chan_end; + spu.decode_dirty_ch |= worker->r_decode_dirty; + do_samples_finish(worker->ns_to, worker->silentch, + worker->decode_pos); +} - //---------------------------------------------------// - //- here we have another 1 ms of sound data - //---------------------------------------------------// - // mix XA infos (if any) +#else - MixXA(); - - /////////////////////////////////////////////////////// - // mix all channels (including reverb) into one buffer +static void queue_channel_work(int ns_to, int silentch) {} +static void sync_worker_thread(void) {} - for (ns = 0; ns < NSSIZE*2; ) - { - SSumLR[ns] += MixREVERBLeft(ns/2); +#endif // THREAD_ENABLED - d = SSumLR[ns] / voldiv; SSumLR[ns] = 0; - if (d < -32767) d = -32767; if (d > 32767) d = 32767; - *pS++ = d; - ns++; +//////////////////////////////////////////////////////////////////////// +// MAIN SPU FUNCTION +// here is the main job handler... +//////////////////////////////////////////////////////////////////////// - SSumLR[ns] += MixREVERBRight(); +void do_samples(unsigned int cycles_to, int do_sync) +{ + unsigned int mask; + int ch, ns_to; + int silentch; + int cycle_diff; - d = SSumLR[ns] / voldiv; SSumLR[ns] = 0; - if(d < -32767) d = -32767; if(d > 32767) d = 32767; - *pS++ = d; - ns++; - } + cycle_diff = cycles_to - spu.cycles_played; + if (cycle_diff < -2*1048576 || cycle_diff > 2*1048576) + { + //xprintf("desync %u %d\n", cycles_to, cycle_diff); + spu.cycles_played = cycles_to; + return; + } + + if (cycle_diff < 2 * 768) + return; + + 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; + } - ////////////////////////////////////////////////////// + ////////////////////////////////////////////////////// // special irq handling in the decode buffers (0x0000-0x1000) - // we know: + // 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: + // 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 + // 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). + // an IRQ. - if(pMixIrq) + if (unlikely((spu.spuCtrl & CTRL_IRQ) + && spu.pSpuIrq < spu.spuMemC+0x1000)) { - for(ns=0;ns=pMixIrq+(ch*0x400) && pSpuIrqspuMemC+0x3ff) pMixIrq=spuMemC; + //xprintf("decoder irq %x\n", spu.decode_pos); + do_irq(); } } - InitREVERB(); + if (worker != NULL) + sync_worker_thread(); + + mask = spu.dwNewChannel & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) { + if (mask & 1) + StartSound(ch); + } + + silentch = ~spu.dwChannelOn & 0xffffff; + + if (spu.dwChannelOn == 0) { + InitREVERB(ns_to); + do_samples_finish(ns_to, silentch, spu.decode_pos); + } + else { + if (do_sync || worker == NULL || !spu_config.iUseThread) { + do_channels(ns_to); + do_samples_finish(ns_to, silentch, spu.decode_pos); + } + else { + queue_channel_work(ns_to, silentch); + } + } + + // advance "stopped" channels that can cause irqs + // (all chans are always playing on the real thing..) + if (spu.spuCtrl & CTRL_IRQ) + do_silent_chans(ns_to, silentch); - // feed the sound - // wanna have around 1/60 sec (16.666 ms) updates - if (iCycle++ > 16) + spu.cycles_played += ns_to * 768; + spu.decode_pos = (spu.decode_pos + ns_to) & 0x1ff; +} + +static void do_samples_finish(int ns_to, int silentch, int decode_pos) +{ + int volmult = spu_config.iVolume; + int ns; + int d; + + if(unlikely(silentch & spu.decode_dirty_ch & (1<<1))) // must clear silent channel decode buffers { - SoundFeedStreamData((unsigned char *)pSpuBuffer, - ((unsigned char *)pS) - ((unsigned char *)pSpuBuffer)); - pS = (short *)pSpuBuffer; - iCycle = 0; + memset(&spu.spuMem[0x800/2], 0, 0x400); + spu.decode_dirty_ch &= ~(1<<1); + } + if(unlikely(silentch & spu.decode_dirty_ch & (1<<3))) + { + memset(&spu.spuMem[0xc00/2], 0, 0x400); + spu.decode_dirty_ch &= ~(1<<3); } - } - // end of big main loop... + //---------------------------------------------------// + // mix XA infos (if any) - bThreadEnded = 1; + MixXA(ns_to, decode_pos); + + /////////////////////////////////////////////////////// + // mix all channels (including reverb) into one buffer - return 0; + if(spu_config.iUseReverb) + REVERBDo(ns_to); + + if((spu.spuCtrl&0x4000)==0) // muted? (rare, don't optimize for this) + { + memset(spu.pS, 0, ns_to * 2 * sizeof(spu.pS[0])); + spu.pS += ns_to * 2; + } + else + for (ns = 0; ns < ns_to * 2; ) + { + d = SSumLR[ns]; SSumLR[ns] = 0; + d = d * volmult >> 10; + ssat32_to_16(d); + *spu.pS++ = d; + ns++; + + d = SSumLR[ns]; SSumLR[ns] = 0; + d = d * volmult >> 10; + ssat32_to_16(d); + *spu.pS++ = d; + ns++; + } +} + +void schedule_next_irq(void) +{ + unsigned int upd_samples; + int ch; + + if (spu.scheduleCallback == NULL) + return; + + upd_samples = 44100 / 50; + + 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; + + 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) return; - iSpuAsyncWait=0; - } + do_samples(cycle, 0); - 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 @@ -840,7 +1202,6 @@ void CALLBACK SPUasync(unsigned long cycle) void CALLBACK SPUupdate(void) { - SPUasync(0); } // XA AUDIO @@ -854,45 +1215,20 @@ 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 @@ -900,92 +1236,152 @@ void SetupStreams(void) { int i; - pSpuBuffer=(unsigned char *)malloc(32768); // alloc mixing buffer - - if(iUseReverb==1) i=88200*2; - else i=NSSIZE*2; - - sRVBStart = (int *)malloc(i*4); // alloc reverb buffer - memset(sRVBStart,0,i*4); - sRVBEnd = sRVBStart + i; - sRVBPlay = sRVBStart; + 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); - 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].pStart=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; +} + +#ifdef THREAD_ENABLED + +static void *spu_worker_thread(void *unused) +{ + while (1) { + sem_wait(&worker->sem_avail); + if (worker->exit_thread) + break; + + do_channel_work(); + + sem_post(&worker->sem_done); + } + + return NULL; +} + +static void init_spu_thread(void) +{ + int ret; + + if (sysconf(_SC_NPROCESSORS_ONLN) <= 1) + return; + + worker = calloc(1, sizeof(*worker)); + if (worker == NULL) + return; + ret = sem_init(&worker->sem_avail, 0, 0); + if (ret != 0) + goto fail_sem_avail; + ret = sem_init(&worker->sem_done, 0, 0); + if (ret != 0) + goto fail_sem_done; + + ret = pthread_create(&worker->thread, NULL, spu_worker_thread, NULL); + if (ret != 0) + goto fail_thread; + + return; + +fail_thread: + sem_destroy(&worker->sem_done); +fail_sem_done: + sem_destroy(&worker->sem_avail); +fail_sem_avail: + free(worker); + worker = NULL; +} + +static void exit_spu_thread(void) +{ + if (worker == NULL) + return; + worker->exit_thread = 1; + sem_post(&worker->sem_avail); + pthread_join(worker->thread, NULL); + sem_destroy(&worker->sem_done); + sem_destroy(&worker->sem_avail); + free(worker); + worker = NULL; +} + +#else // if !THREAD_ENABLED + +static void init_spu_thread(void) +{ +} + +static void exit_spu_thread(void) +{ } -// INIT/EXIT STUFF +#endif // 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(); - iVolume = 3; - iReverbOff = -1; - 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 + + init_spu_thread(); + 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; } @@ -993,12 +1389,11 @@ 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; } @@ -1008,6 +1403,9 @@ long CALLBACK SPUshutdown(void) { SPUclose(); RemoveStreams(); // no more streaming + spu.bSpuInit=0; + + exit_spu_thread(); return 0; } @@ -1044,12 +1442,17 @@ 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 @@ -1075,4 +1478,27 @@ char * SPUgetLibInfos(void) } */ +// debug +void spu_get_debug_info(int *chans_out, int *run_chans, int *fmod_chans_out, int *noise_chans_out) +{ + int ch = 0, fmod_chans = 0, noise_chans = 0, irq_chans = 0; + + for(;ch