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=07bc4b3789090b7617fc9a0db7c35b325aa7f9cc;hb=63a4f6b6a3b0315590cd3009df2c92480ed2d98b;hpb=215ff9e69c0b845f24e7a3aa9faeef06d9276145 diff --git a/plugins/dfsound/spu.c b/plugins/dfsound/spu.c index 07bc4b37..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, 2010-2012,2014 + Portions (C) Gražvydas "notaz" Ignotas, 2010-2012,2014,2015 ***************************************************************************/ /*************************************************************************** @@ -20,6 +20,7 @@ #ifndef _WIN32 #include // gettimeofday in xa.c +#define THREAD_ENABLED 1 #endif #include "stdafx.h" @@ -74,6 +75,44 @@ SPUConfig spu_config; SPUCHAN s_chan[MAXCHAN+1]; // channel + 1 infos (1 is security for fmod handling) REVERBInfo rvb; +#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; + +#else +static const void * const worker = NULL; +#endif + // certain globals (were local before, but with the new timeproc I need em global) static const int f[8][2] = { { 0, 0 }, @@ -138,71 +177,70 @@ int iFMod[NSSIZE]; // / // - -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&3]) -#define gval(x) ((int)((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" @@ -239,10 +277,7 @@ 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; @@ -255,25 +290,27 @@ INLINE void StartSound(int ch) s_chan[ch].spos=0; spu.dwNewChannel&=~(1<>15; if(NP>0x3fff) NP=0x3fff; if(NP<0x1) NP=0x1; sinc=NP<<4; // calc frequency if(spu_config.iUseInterpolation==1) // freq change in simple interpolation mode - s_chan[ch].SB[32]=1; + SB[32]=1; iFMod[ns]=0; return sinc; @@ -281,50 +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(spu_config.iUseInterpolation>=2) // gauss/cubic interpolation - { - int gpos = s_chan[ch].SB[28]; - gval0 = fa; + { + int gpos = SB[28]; + gval0 = fa; gpos = (gpos+1) & 3; - s_chan[ch].SB[28] = gpos; + SB[28] = gpos; } else 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, int spos) +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(spu_config.iUseInterpolation) - { + { //--------------------------------------------------// case 3: // cubic interpolation { long xd;int gpos; xd = (spos >> 1)+1; - gpos = s_chan[ch].SB[28]; + gpos = SB[28]; fa = gval(3) - 3*gval(2) + 3*gval(1) - gval0; fa *= (xd - (2<<15)) / 6; @@ -343,7 +380,7 @@ INLINE int iGetInterpolationVal(int ch, int spos) { int vl, vr;int gpos; vl = (spos >> 6) & ~3; - gpos = s_chan[ch].SB[28]; + gpos = SB[28]; vr=(gauss[vl]*(int)gval0)&~2047; vr+=(gauss[vl+1]*gval(1))&~2047; vr+=(gauss[vl+2]*gval(2))&~2047; @@ -353,15 +390,15 @@ INLINE int iGetInterpolationVal(int ch, int spos) //--------------------------------------------------// 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; //--------------------------------------------------// } @@ -397,7 +434,7 @@ 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; @@ -421,7 +458,7 @@ static int decode_block(int ch) shift_factor = predict_nr & 0xf; predict_nr >>= 4; - decode_block_data(s_chan[ch].SB, start + 2, predict_nr, shift_factor); + decode_block_data(SB, start + 2, predict_nr, shift_factor); flags = start[1]; if (flags & 4) @@ -476,6 +513,38 @@ static int skip_block(int ch) 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) { @@ -517,12 +586,9 @@ static void scan_for_irq(int ch, unsigned int *upd_samples) } #define make_do_samples(name, fmod_code, interp_start, interp1_code, interp2_code, interp_end) \ -static noinline int do_samples_##name(int ch, 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 ns, d, fa; \ int ret = ns_to; \ interp_start; \ @@ -531,28 +597,25 @@ static noinline int do_samples_##name(int ch, int ns_to) \ { \ fmod_code; \ \ - spos += sinc; \ - while (spos >= 0x10000) \ + *spos += sinc; \ + while (*spos >= 0x10000) \ { \ - fa = SB[sbpos++]; \ - if (sbpos >= 28) \ + fa = SB[(*sbpos)++]; \ + if (*sbpos >= 28) \ { \ - sbpos = 0; \ - d = decode_block(ch); \ + *sbpos = 0; \ + d = decode_f(ch, SB); \ if (d && ns < ret) \ ret = ns; \ } \ \ interp1_code; \ - spos -= 0x10000; \ + *spos -= 0x10000; \ } \ \ interp2_code; \ } \ \ - s_chan[ch].sinc = sinc; \ - s_chan[ch].spos = spos; \ - s_chan[ch].iSBPos = sbpos; \ interp_end; \ \ return ret; \ @@ -560,12 +623,12 @@ static noinline int do_samples_##name(int ch, int ns_to) \ #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, spos), ) -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 \ SB[28] = 0; \ @@ -575,10 +638,10 @@ make_do_samples(noint, , fa = s_chan[ch].SB[29], , ChanBuf[ns] = 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] + 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, ) @@ -747,6 +810,7 @@ static void do_silent_chans(int ns_to, int silentch) static void do_channels(int ns_to) { unsigned int mask; + int *SB, sinc; int ch, d; InitREVERB(ns_to); @@ -756,20 +820,25 @@ static void do_channels(int ns_to) { 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(ch, ns_to); + 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(ch, ns_to); + d = do_samples_simple(decode_block, ch, ns_to, + SB, sinc, &s_chan[ch].spos, &s_chan[ch].iSBPos); else - d = do_samples_default(ch, ns_to); + d = do_samples_default(decode_block, ch, ns_to, + SB, sinc, &s_chan[ch].spos, &s_chan[ch].iSBPos); - d = MixADSR(ch, d); + d = MixADSR(&s_chan[ch].ADSRX, d); if (d < ns_to) { spu.dwChannelOn &= ~(1 << ch); - s_chan[ch].bStop = 1; s_chan[ch].ADSRX.EnvelopeVol = 0; memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0])); } @@ -789,14 +858,142 @@ static void do_channels(int ns_to) } } +static void do_samples_finish(int ns_to, int silentch, int decode_pos); + +// 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) + { + 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 + 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); + } + + 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_finish(int ns_to, int silentch); - -void do_samples(unsigned int cycles_to) +void do_samples(unsigned int cycles_to, int do_sync) { unsigned int mask; int ch, ns_to; @@ -850,7 +1047,8 @@ void do_samples(unsigned int cycles_to) } } - silentch = ~(spu.dwChannelOn|spu.dwNewChannel); + if (worker != NULL) + sync_worker_thread(); mask = spu.dwNewChannel & 0xffffff; for (ch = 0; mask != 0; ch++, mask >>= 1) { @@ -858,14 +1056,22 @@ void do_samples(unsigned int cycles_to) StartSound(ch); } - if (spu.dwChannelOn == 0) + silentch = ~spu.dwChannelOn & 0xffffff; + + if (spu.dwChannelOn == 0) { InitREVERB(ns_to); + do_samples_finish(ns_to, silentch, spu.decode_pos); + } else { - do_channels(ns_to); + 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); + } } - do_samples_finish(ns_to, silentch); - // advance "stopped" channels that can cause irqs // (all chans are always playing on the real thing..) if (spu.spuCtrl & CTRL_IRQ) @@ -875,7 +1081,7 @@ void do_samples(unsigned int cycles_to) spu.decode_pos = (spu.decode_pos + ns_to) & 0x1ff; } -void do_samples_finish(int ns_to, int silentch) +static void do_samples_finish(int ns_to, int silentch, int decode_pos) { int volmult = spu_config.iVolume; int ns; @@ -895,7 +1101,7 @@ void do_samples_finish(int ns_to, int silentch) //---------------------------------------------------// // mix XA infos (if any) - MixXA(ns_to, spu.decode_pos); + MixXA(ns_to, decode_pos); /////////////////////////////////////////////////////// // mix all channels (including reverb) into one buffer @@ -967,7 +1173,7 @@ void schedule_next_irq(void) void CALLBACK SPUasync(unsigned int cycle, unsigned int flags) { - do_samples(cycle); + do_samples(cycle, 0); if (spu.spuCtrl & CTRL_IRQ) schedule_next_irq(); @@ -1072,7 +1278,79 @@ void RemoveStreams(void) spu.CDDAStart = NULL; } -// INIT/EXIT STUFF +#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) +{ +} + +#endif // SPUINIT: this func will be called first by the main emu long CALLBACK SPUinit(void) @@ -1091,6 +1369,8 @@ long CALLBACK SPUinit(void) if (spu_config.iVolume == 0) spu_config.iVolume = 768; // 1024 is 1.0 + init_spu_thread(); + return 0; } @@ -1125,6 +1405,8 @@ long CALLBACK SPUshutdown(void) RemoveStreams(); // no more streaming spu.bSpuInit=0; + exit_spu_thread(); + return 0; }