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=d7dc1b6441fb60587fcd55a0df11701ce36b1ed6;hb=63a4f6b6a3b0315590cd3009df2c92480ed2d98b;hpb=76d1d09c1224af8d663ce63dc5b32425bd62cb29 diff --git a/plugins/dfsound/spu.c b/plugins/dfsound/spu.c index d7dc1b64..b091efcc 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, 2010-2012,2014,2015 ***************************************************************************/ /*************************************************************************** @@ -18,24 +18,19 @@ * * ***************************************************************************/ +#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" - -#ifdef ENABLE_NLS -#include -#include -#define _(x) gettext(x) -#define N_(x) (x) -#else -#define _(x) (x) -#define N_(x) (x) -#endif +#include "out.h" +#include "arm_features.h" +#include "spu_config.h" #ifdef __ARM_ARCH_7A__ #define ssat32_to_16(v) \ @@ -47,6 +42,11 @@ } 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"); @@ -67,53 +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=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; +#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; -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; +#else +static const void * const worker = NULL; +#endif // certain globals (were local before, but with the new timeproc I need em global) @@ -122,15 +120,9 @@ 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 @@ -185,71 +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)); + SB[28]=(SB[28]*sinc)>>17; + //if(sinc<=0x8000) + // SB[29]=SB[30]-(SB[28]*((0x10000/sinc)-1)); //else - s_chan[ch].SB[29]+=s_chan[ch].SB[28]; + 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" @@ -257,6 +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 //////////////////////////////////////////////////////////////////////// @@ -266,42 +277,40 @@ 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].SB[26]=0; // init mixing vars s_chan[ch].SB[27]=0; - s_chan[ch].iSBPos=28; + 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<>15; if(NP>0x3fff) NP=0x3fff; if(NP<0x1) NP=0x1; sinc=NP<<4; // calc frequency - if(iUseInterpolation==1) // freq change in simple interpolation mode - s_chan[ch].SB[32]=1; + if(spu_config.iUseInterpolation==1) // freq change in simple interpolation mode + SB[32]=1; iFMod[ns]=0; return sinc; @@ -309,50 +318,50 @@ INLINE int FModChangeFrequency(int ch,int ns) //////////////////////////////////////////////////////////////////////// -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 { 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; @@ -370,9 +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; @@ -381,15 +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; //--------------------------------------------------// } @@ -397,15 +406,6 @@ 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; @@ -434,61 +434,48 @@ static void decode_block_data(int *dest, const unsigned char *src, int predict_n } } -static int decode_block(int ch) +static int decode_block(int ch, int *SB) { unsigned char *start; - int predict_nr,shift_factor,flags; + int predict_nr, shift_factor, flags; int ret = 0; - start=s_chan[ch].pCurr; // set up the current pos - if(dwPendingChanOff&(1< turn everything off - dwPendingChanOff&=~(1< call main emu + if (!(s_chan[ch].prevflags & 2)) ret = 1; - } + + start = s_chan[ch].pLoop; } + else + check_irq(ch, start); // hack, see check_irq below.. - predict_nr=(int)start[0]; - shift_factor=predict_nr&0xf; + predict_nr = (int)start[0]; + shift_factor = predict_nr & 0xf; predict_nr >>= 4; - decode_block_data(s_chan[ch].SB, start + 2, predict_nr, shift_factor); - - //////////////////////////////////////////// flag handler + decode_block_data(SB, start + 2, predict_nr, shift_factor); - flags=(int)start[1]; - if(flags&4) - s_chan[ch].pLoop=start; // loop adress + flags = 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; } @@ -497,68 +484,138 @@ static int decode_block(int ch) static int skip_block(int ch) { unsigned char *start = s_chan[ch].pCurr; - int flags = start[1]; + int flags; int ret = 0; - if(start == pSpuIrq) - { - do_irq(); - ret = 1; + if (s_chan[ch].prevflags & 1) { + if (!(s_chan[ch].prevflags & 2)) + ret = 1; + + start = s_chan[ch].pLoop; } + else + check_irq(ch, start); - if(flags & 4) + flags = start[1]; + if (flags & 4) s_chan[ch].pLoop = start; - s_chan[ch].pCurr += 16; + start += 16; + + if (flags & 1) { + start = s_chan[ch].pLoop; + check_irq(ch, start); + } - if(flags & 1) - s_chan[ch].pCurr = s_chan[ch].pLoop; + s_chan[ch].pCurr = start; + s_chan[ch].prevflags = flags; - s_chan[ch].bJump = flags & 1; 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 = ram[start]; + shift_factor = predict_nr & 0xf; + predict_nr >>= 4; + + decode_block_data(SB, ram + start + 2, predict_nr, shift_factor); + + flags = ram[start + 1]; + if (flags & 4) + loop = start; // loop adress + + start += 16; + + if (flags & 1) // 1: stop/loop + start = loop; + + worker->ch[ch].start = start & 0x7ffff; + worker->ch[ch].loop = loop; + + return 0; +} + +#endif + +// 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 (*decode_f)(int ch, int *SB), int ch, \ + int ns_to, int *SB, int sinc, int *spos, int *sbpos) \ { \ - int sinc = s_chan[ch].sinc; \ - int spos = s_chan[ch].spos; \ - int sbpos = s_chan[ch].iSBPos; \ - int *SB = s_chan[ch].SB; \ - int ret = -1; \ - int d, fa; \ + int ns, d, fa; \ + int ret = ns_to; \ interp_start; \ \ - for (; ns < ns_to; ns++) \ + for (ns = 0; ns < ns_to; ns++) \ { \ fmod_code; \ \ - while (spos >= 0x10000) \ + *spos += sinc; \ + while (*spos >= 0x10000) \ { \ - if(sbpos == 28) \ + fa = SB[(*sbpos)++]; \ + if (*sbpos >= 28) \ { \ - sbpos = 0; \ - d = decode_block(ch); \ - if(d && iSPUIRQWait) \ - { \ + *sbpos = 0; \ + d = decode_f(ch, SB); \ + if (d && ns < ret) \ ret = ns; \ - goto out; \ - } \ } \ \ - fa = SB[sbpos++]; \ interp1_code; \ - spos -= 0x10000; \ + *spos -= 0x10000; \ } \ \ interp2_code; \ - spos += sinc; \ } \ \ -out: \ - s_chan[ch].sinc = sinc; \ - s_chan[ch].spos = spos; \ - s_chan[ch].iSBPos = sbpos; \ interp_end; \ \ return ret; \ @@ -566,67 +623,99 @@ out: \ #define fmod_recv_check \ if(s_chan[ch].bFMod==1 && iFMod[ns]) \ - sinc = FModChangeFrequency(ch,ns) + sinc = FModChangeFrequency(SB, s_chan[ch].iRawPitch, 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) + 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 \ - 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 + 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(ch); /* --> interpolate up */ \ - else InterpolateDown(ch); /* --> else down */ \ - ChanBuf[ns] = s_chan[ch].SB[29] + if(sinc<0x10000) /* -> upsampling? */ \ + InterpolateUp(SB, sinc); /* --> interpolate up */ \ + else InterpolateDown(SB, sinc); /* --> else down */ \ + ChanBuf[ns] = SB[29] make_do_samples(simple, , , simple_interp_store, simple_interp_get, ) -static int do_samples_noise(int ch, int ns, int ns_to) +static int do_samples_skip(int ch, int ns_to) { - int level, shift, bit; + int ret = ns_to, ns, d; + + s_chan[ch].spos += s_chan[ch].iSBPos << 16; - s_chan[ch].spos += s_chan[ch].sinc * (ns_to - ns); - while (s_chan[ch].spos >= 28*0x10000) + for (ns = 0; ns < ns_to; ns++) { - skip_block(ch); - s_chan[ch].spos -= 28*0x10000; + 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].iSBPos = s_chan[ch].spos >> 16; + s_chan[ch].spos &= 0xffff; + + return ret; +} + +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 = (spuCtrl >> 10) & 0x0f; + level = (ctrl >> 10) & 0x0f; level = 0x8000 >> level; - for (; ns < ns_to; ns++) + for (ns = 0; ns < ns_to; ns++) { - dwNoiseCount += 2; - if (dwNoiseCount >= level) + counter += 2; + if (counter >= level) { - dwNoiseCount -= level; - shift = (dwNoiseVal >> 10) & 0x1f; + counter -= level; + shift = (val >> 10) & 0x1f; bit = (0x69696969 >> shift) & 1; - if (dwNoiseVal & 0x8000) - bit ^= 1; - dwNoiseVal = (dwNoiseVal << 1) | bit; + bit ^= (val >> 15) & 1; + val = (val << 1) | bit; } - ChanBuf[ns] = (signed short)dwNoiseVal; + ChanBuf[ns] = (signed short)val; } - return -1; + *dwNoiseCount = counter; + *dwNoiseVal = val; } -#ifdef __ARM_ARCH_7A__ -// asm code +static int do_samples_noise(int ch, int ns_to) +{ + 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); +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) { @@ -645,10 +734,10 @@ 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; @@ -666,264 +755,440 @@ 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! -//////////////////////////////////////////////////////////////////////// - -// 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 +// 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; -#define PAUSE_W 5 -#define PAUSE_L 5000 + while (count-- > 0) + { + cursor &= 0x1ff; + dst[cursor] = *src++; + cursor++; + } -//////////////////////////////////////////////////////////////////////// + // decode_pos is updated and irqs are checked later, after voice loop +} -static void *MAINThread(void *arg) +static void do_silent_chans(int ns_to, int silentch) { - int volmult = iVolume; - int ns,ns_from,ns_to; - int ch,d,silentch; - int bIRQReturn=0; + int ch; - while(!bEndThread) // until we are shutting down + for (ch = 0; ch < MAXCHAN; ch++) { - // 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 + 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; - while(!iSecureStart && !bEndThread && // no new start? no thread end? - (SoundGetBytesBuffered()>TESTSIZE)) // and still enuff data in sound buffer? + 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; } - silentch=~(dwChannelOn|dwNewChannel); + 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, spu.sRVBStart); + else + mix_chan(0, ns_to, s_chan[ch].iLeftVolume, s_chan[ch].iRightVolume); + } +} + +static void do_samples_finish(int ns_to, int silentch, int decode_pos); - //--------------------------------------------------// - //- main channel loop -// - //--------------------------------------------------// +// optional worker thread handling + +#ifdef THREAD_ENABLED + +static void queue_channel_work(int ns_to, int silentch) +{ + unsigned int mask; + int ch; + + worker->ns_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); + } + + worker->pending = 1; + sem_post(&worker->sem_avail); +} + +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; + + ns_to = worker->ns_to; + memset(worker->sRVBStart, 0, ns_to * sizeof(worker->sRVBStart[0]) * 2); + + 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); + + 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])); + } + + if (ch == 1 || ch == 3) { - for(;ch=0) - { - bIRQReturn=1; - lastch=ch; - lastns=ns_to=d; - if(d==0) - break; - } - - MixADSR(ch, ns_from, ns_to); - - if(s_chan[ch].bFMod==2) // fmod freq channel - memcpy(iFMod, ChanBuf, sizeof(iFMod)); - else if(s_chan[ch].bRVBActive) - mix_chan_rvb(ns_from,ns_to-ns_from,s_chan[ch].iLeftVolume,s_chan[ch].iRightVolume); - else - mix_chan(ns_from,ns_to-ns_from,s_chan[ch].iLeftVolume,s_chan[ch].iRightVolume); - } + do_decode_bufs((void *)worker->ram, ch/2, ns_to, worker->decode_pos); + decode_dirty_ch |= 1 << ch; } - // 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; - - s_chan[ch].spos += s_chan[ch].sinc * NSSIZE; - 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) - { - // looping on self - dwChannelDead |= 1<sRVBStart); + else + mix_chan(0, ns_to, s_chan[ch].iLeftVolume, s_chan[ch].iRightVolume); + } + + worker->r_chan_end = endmask; + worker->r_decode_dirty = decode_dirty_ch; +} + +static void sync_worker_thread(void) +{ + unsigned int mask; + int ch; + + if (!worker->pending) + return; + + sem_wait(&worker->sem_done); + worker->pending = 0; + + mask = worker->chmask; + for (ch = 0; mask != 0; ch++, mask >>= 1) { + if (!(mask & 1)) continue; + + // 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; + } + + spu.dwChannelOn &= ~worker->r_chan_end; + spu.decode_dirty_ch |= worker->r_decode_dirty; + + do_samples_finish(worker->ns_to, worker->silentch, + worker->decode_pos); +} + +#else + +static void queue_channel_work(int ns_to, int silentch) {} +static void sync_worker_thread(void) {} + +#endif // THREAD_ENABLED + +//////////////////////////////////////////////////////////////////////// +// MAIN SPU FUNCTION +// here is the main job handler... +//////////////////////////////////////////////////////////////////////// + +void do_samples(unsigned int cycles_to, int do_sync) +{ + unsigned int mask; + int ch, ns_to; + int silentch; + int cycle_diff; + + 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: + // 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. - if(bIRQReturn && iSPUIRQWait) // special return for "spu irq - wait for cpu action" + 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) { - iSpuAsyncWait=1; - bIRQReturn=0; - if(iUseTimer!=2) - { - DWORD dwWatchTime=timeGetTime_spu()+2500; - - while(iSpuAsyncWait && !bEndThread && - timeGetTime_spu()>= 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); + + 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 + { + 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); + } - //---------------------------------------------------// - //- here we have another 1 ms of sound data //---------------------------------------------------// // mix XA infos (if any) - MixXA(); + MixXA(ns_to, decode_pos); /////////////////////////////////////////////////////// // mix all channels (including reverb) into one buffer - if(iUseReverb) - REVERBDo(); + if(spu_config.iUseReverb) + REVERBDo(ns_to); - if((spuCtrl&0x4000)==0) // muted? (rare, don't optimize for this) + if((spu.spuCtrl&0x4000)==0) // muted? (rare, don't optimize for this) { - memset(pS, 0, NSSIZE * 2 * sizeof(pS[0])); - pS += NSSIZE*2; + memset(spu.pS, 0, ns_to * 2 * sizeof(spu.pS[0])); + spu.pS += ns_to * 2; } else - for (ns = 0; ns < NSSIZE*2; ) + for (ns = 0; ns < ns_to * 2; ) { d = SSumLR[ns]; SSumLR[ns] = 0; d = d * volmult >> 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). +void schedule_next_irq(void) +{ + unsigned int upd_samples; + int ch; - if(pMixIrq) - { - for(ns=0;ns=pMixIrq+(ch*0x400) && pSpuIrqspuMemC+0x3ff) pMixIrq=spuMemC; - } - } + if (spu.scheduleCallback == NULL) + return; - InitREVERB(); + upd_samples = 44100 / 50; - // 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; - } - } + 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; - // end of big main loop... + scan_for_irq(ch, &upd_samples); + } - bThreadEnded = 1; + 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; + } + } - return 0; + 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, 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 @@ -937,7 +1202,6 @@ void CALLBACK SPUasync(unsigned long cycle) void CALLBACK SPUupdate(void) { - SPUasync(0); } // XA AUDIO @@ -959,37 +1223,12 @@ int CALLBACK SPUplayCDDAchannel(short *pcm, int 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 @@ -997,89 +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; + 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 + spu.CDDAStart = // alloc cdda buffer (uint32_t *)malloc(CDDA_BUFFER_SIZE); - CDDAEnd = CDDAStart + 16384; - CDDAPlay = CDDAStart; - CDDAFeed = CDDAStart; + 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; +} + +#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; } -// INIT/EXIT STUFF +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) +{ +} + +#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(); - 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; } @@ -1087,12 +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; } @@ -1102,6 +1403,9 @@ long CALLBACK SPUshutdown(void) { SPUclose(); RemoveStreams(); // no more streaming + spu.bSpuInit=0; + + exit_spu_thread(); return 0; } @@ -1138,12 +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 @@ -1176,18 +1485,18 @@ void spu_get_debug_info(int *chans_out, int *run_chans, int *fmod_chans_out, int for(;ch