X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=plugins%2Fdfsound%2Fspu.c;h=637e8521629f69968c20055a5fee89b3377e41fa;hp=79e534cd5da246976935acfa9e7d6642a51dbfc8;hb=HEAD;hpb=1ab7621a76d1ef82fde77322c12d4076889bed01 diff --git a/plugins/dfsound/spu.c b/plugins/dfsound/spu.c index 79e534cd..ed5b4b5c 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 +18,36 @@ * * ***************************************************************************/ +#include #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 "spu_config.h" +#include "spu.h" + +#ifdef __arm__ +#include "arm_features.h" +#endif + +#ifdef HAVE_ARMV7 + #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"); @@ -55,68 +68,14 @@ static char * libraryInfo = N_("P.E.Op.S. Sound Driver V1.7\nCoded by Pete B // globals -// psx buffer / addresses - -unsigned short regArea[10000]; -unsigned short spuMem[256*1024]; -unsigned char * spuMemC; -unsigned char * pSpuIrq=0; -unsigned char * pSpuBuffer; -unsigned char * pMixIrq=0; - -// user settings - -int iVolume=3; -int iXAPitch=1; -int iUseTimer=2; -int iSPUIRQWait=1; -int iDebugMode=0; -int iRecordMode=0; -int iUseReverb=2; -int iUseInterpolation=2; - -// 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; +SPUInfo spu; +SPUConfig spu_config; -unsigned short spuCtrl=0; // some vars to store psx reg infos -unsigned short spuStat=0; -unsigned short spuIrq=0; -unsigned long spuAddr=0xffffffff; // address into spu mem -int bEndThread=0; // thread handlers -int bThreadEnded=0; -int bSpuInit=0; -int bSPUIsOpen=0; - -static pthread_t thread = (pthread_t)-1; // thread id (linux) - -unsigned long dwNewChannel=0; // flags for faster testing, if new channel starts -unsigned long dwChannelOn=0; -unsigned long dwPendingChanOff=0; - -void (CALLBACK *irqCallback)(void)=0; // func of main emu, called on spu irq -void (CALLBACK *cddavCallback)(unsigned short,unsigned short)=0; - -// certain globals (were local before, but with the new timeproc I need em global) - -static const int f[8][2] = { { 0, 0 }, - { 60, 0 }, - { 115, -52 }, - { 98, -55 }, - { 122, -60 } }; +static int iFMod[NSSIZE]; +static int RVB[NSSIZE * 2]; int ChanBuf[NSSIZE]; -int SSumLR[NSSIZE*2]; -int iFMod[NSSIZE]; -int iCycle = 0; -short * pS; -int lastch=-1; // last channel processed on spu irq in timer mode -static int lastns=0; // last ns pos -static int iSecureStart=0; // secure start counter +#define CDDA_BUFFER_SIZE (16384 * sizeof(uint32_t)) // must be power of 2 //////////////////////////////////////////////////////////////////////// // CODE AREA @@ -169,668 +128,1238 @@ static int iSecureStart=0; // secure start counter // / // - -INLINE void InterpolateUp(int ch) +static void InterpolateUp(sample_buf *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 + int *SB = sb->SB; + if (sb->sinc_old != sinc) + { + sb->sinc_old = sinc; + SB[32] = 1; + } + 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(sample_buf *sb, int sinc) { - if(s_chan[ch].sinc>=0x20000L) // we would skip at least one val? + int *SB = sb->SB; + 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]) #include "gauss_i.h" +#include "xa.c" -//////////////////////////////////////////////////////////////////////// +static void do_irq(int cycles_after) +{ + if (spu.spuStat & STAT_IRQ) + log_unhandled("spu: missed irq?\n"); + else + { + spu.spuStat |= STAT_IRQ; // asserted status? + if (spu.irqCallback) + spu.irqCallback(cycles_after); + } +} -#include "xa.c" +static int check_irq(int ch, unsigned char *pos) +{ + if((spu.spuCtrl & (CTRL_ON|CTRL_IRQ)) == (CTRL_ON|CTRL_IRQ) && pos == spu.pSpuIrq) + { + //printf("ch%d irq %04zx\n", ch, pos - spu.spuMemC); + do_irq(0); + return 1; + } + return 0; +} + +void check_irq_io(unsigned int addr) +{ + unsigned int irq_addr = regAreaGet(H_SPUirqAddr) << 3; + //addr &= ~7; // ? + if((spu.spuCtrl & (CTRL_ON|CTRL_IRQ)) == (CTRL_ON|CTRL_IRQ) && addr == irq_addr) + { + //printf("io irq %04x\n", irq_addr); + do_irq(0); + } +} + +void do_irq_io(int cycles_after) +{ + if ((spu.spuCtrl & (CTRL_ON|CTRL_IRQ)) == (CTRL_ON|CTRL_IRQ)) + { + do_irq(cycles_after); + } +} //////////////////////////////////////////////////////////////////////// // START SOUND... called by main thread to setup a new sound on a channel //////////////////////////////////////////////////////////////////////// -INLINE void StartSound(int ch) +static void ResetInterpolation(sample_buf *sb) +{ + memset(&sb->interp, 0, sizeof(sb->interp)); + sb->sinc_old = -1; +} + +static void StartSoundSB(sample_buf *sb) { + sb->SB[26] = 0; // init mixing vars + sb->SB[27] = 0; + ResetInterpolation(sb); +} + +static void StartSoundMain(int ch) +{ + SPUCHAN *s_chan = &spu.s_chan[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->prevflags = 2; + s_chan->iSBPos = 27; + s_chan->spos = 0; + s_chan->bStarting = 1; - s_chan[ch].s_1=0; // init mixing vars - s_chan[ch].s_2=0; - s_chan[ch].iSBPos=28; + s_chan->pCurr = spu.spuMemC + ((regAreaGetCh(ch, 6) & ~1) << 3); - s_chan[ch].SB[29]=0; // init our interpolation helpers - s_chan[ch].SB[30]=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 + spu.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 + pitch = (signed short)pitch; + pitch = ((32768 + fmod_buf[ns]) * pitch) >> 15; + pitch &= 0xffff; + if (pitch > 0x3fff) + pitch = 0x3fff; + + fmod_buf[ns] = 0; + + return pitch << 4; +} + +INLINE void StoreInterpolationGaussCubic(sample_buf *sb, int fa) +{ + int gpos = sb->interp.gauss.pos & 3; + sb->interp.gauss.val[gpos++] = fa; + sb->interp.gauss.pos = gpos & 3; } -//////////////////////////////////////////////////////////////////////// +#define gval(x) (int)sb->interp.gauss.val[(gpos + x) & 3] -INLINE void FModChangeFrequency(int ch,int ns) +INLINE int GetInterpolationCubic(const sample_buf *sb, int spos) { - int NP=s_chan[ch].iRawPitch; + int gpos = sb->interp.gauss.pos; + int xd = (spos >> 1) + 1; + int fa; - NP=((32768L+iFMod[ns])*NP)/32768L; + fa = gval(3) - 3*gval(2) + 3*gval(1) - gval(0); + fa *= (xd - (2<<15)) / 6; + fa >>= 15; + fa += gval(2) - gval(1) - gval(1) + gval(0); + fa *= (xd - (1<<15)) >> 1; + fa >>= 15; + fa += gval(1) - gval(0); + fa *= xd; + fa >>= 15; + fa = fa + gval(0); + return fa; +} - if(NP>0x3fff) NP=0x3fff; - if(NP<0x1) NP=0x1; +INLINE int GetInterpolationGauss(const sample_buf *sb, int spos) +{ + int gpos = sb->interp.gauss.pos; + int vl = (spos >> 6) & ~3; + int vr; + vr = (gauss[vl+0] * gval(0)) >> 15; + vr += (gauss[vl+1] * gval(1)) >> 15; + vr += (gauss[vl+2] * gval(2)) >> 15; + vr += (gauss[vl+3] * gval(3)) >> 15; + return vr; +} - NP=(44100L*NP)/(4096L); // calc frequency +static void decode_block_data(int *dest, const unsigned char *src, int predict_nr, int shift_factor) +{ + static const int f[16][2] = { + { 0, 0 }, + { 60, 0 }, + { 115, -52 }, + { 98, -55 }, + { 122, -60 } + }; + int nSample; + int fa, s_1, s_2, d, s; + + s_1 = dest[27]; + s_2 = dest[26]; + + for (nSample = 0; nSample < 28; src++) + { + d = (int)*src; + s = (int)(signed short)((d & 0x0f) << 12); - 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; - iFMod[ns]=0; -} + fa = s >> shift_factor; + fa += ((s_1 * f[predict_nr][0])>>6) + ((s_2 * f[predict_nr][1])>>6); + ssat32_to_16(fa); + 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); + ssat32_to_16(fa); + s_2 = s_1; s_1 = fa; -// 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 + dest[nSample++] = fa; + } +} -INLINE int iGetNoiseVal(int ch) +static int decode_block(void *unused, int ch, int *SB) { - int fa; + SPUCHAN *s_chan = &spu.s_chan[ch]; + unsigned char *start; + int predict_nr, shift_factor, flags; + int ret = 0; - if((dwNoiseVal<<=1)&0x80000000L) - { - dwNoiseVal^=0x0040001L; - fa=((dwNoiseVal>>2)&0x7fff); - fa=-fa; - } - else fa=(dwNoiseVal>>2)&0x7fff; + start = s_chan->pCurr; // set up the current pos + if (start - spu.spuMemC < 0x1000) { // ? + //log_unhandled("ch%02d plays decode bufs @%05lx\n", + // ch, (long)(start - spu.spuMemC)); + ret = 1; + } - // 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 (s_chan->prevflags & 1) // 1: stop/loop + { + if (!(s_chan->prevflags & 2)) + ret = 1; - if(iUseInterpolation<2) // no gauss/cubic interpolation? - s_chan[ch].SB[29] = fa; // -> store noise val in "current sample" slot - return fa; -} + start = s_chan->pLoop; + } -//////////////////////////////////////////////////////////////////////// + check_irq(ch, start); -INLINE void StoreInterpolationVal(int ch,int fa) -{ - if(s_chan[ch].bFMod==2) // fmod freq channel - s_chan[ch].SB[29]=fa; - else - { - if((spuCtrl&0x4000)==0) fa=0; // muted? - else // else adjust - { - if(fa>32767L) fa=32767L; - if(fa<-32767L) fa=-32767L; - } + predict_nr = start[0]; + shift_factor = predict_nr & 0xf; + predict_nr >>= 4; - if(iUseInterpolation>=2) // gauss/cubic interpolation - { - int gpos = s_chan[ch].SB[28]; - gval0 = fa; - gpos = (gpos+1) & 3; - s_chan[ch].SB[28] = gpos; - } - else - if(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 - } - else s_chan[ch].SB[29]=fa; // no interpolation - } -} + decode_block_data(SB, start + 2, predict_nr, shift_factor); -//////////////////////////////////////////////////////////////////////// + flags = start[1]; + if (flags & 4 && !s_chan->bIgnoreLoop) + s_chan->pLoop = start; // loop adress + + start += 16; + + s_chan->pCurr = start; // store values for next cycle + s_chan->prevflags = flags; + s_chan->bStarting = 0; -INLINE int iGetInterpolationVal(int ch) + return ret; +} + +// do block, but ignore sample data +static int skip_block(int ch) { - int fa; + SPUCHAN *s_chan = &spu.s_chan[ch]; + unsigned char *start = s_chan->pCurr; + int flags; + int ret = 0; - if(s_chan[ch].bFMod==2) return s_chan[ch].SB[29]; + if (s_chan->prevflags & 1) { + if (!(s_chan->prevflags & 2)) + ret = 1; - switch(iUseInterpolation) - { - //--------------------------------------------------// - case 3: // cubic interpolation - { - long xd;int gpos; - xd = ((s_chan[ch].spos) >> 1)+1; - gpos = s_chan[ch].SB[28]; - - fa = gval(3) - 3*gval(2) + 3*gval(1) - gval0; - fa *= (xd - (2<<15)) / 6; - fa >>= 15; - fa += gval(2) - gval(1) - gval(1) + gval0; - fa *= (xd - (1<<15)) >> 1; - fa >>= 15; - fa += gval(1) - gval0; - fa *= xd; - fa >>= 15; - fa = fa + gval0; - - } break; - //--------------------------------------------------// - 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; - vr+=(gauss[vl+1]*gval(1))&~2047; - vr+=(gauss[vl+2]*gval(2))&~2047; - vr+=(gauss[vl+3]*gval(3))&~2047; - fa = vr>>11; - } break; - //--------------------------------------------------// - 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]; - } break; - //--------------------------------------------------// - default: // no interpolation - { - fa=s_chan[ch].SB[29]; - } break; - //--------------------------------------------------// - } + start = s_chan->pLoop; + } - return fa; + check_irq(ch, start); + + flags = start[1]; + if (flags & 4 && !s_chan->bIgnoreLoop) + s_chan->pLoop = start; + + start += 16; + + s_chan->pCurr = start; + s_chan->prevflags = flags; + s_chan->bStarting = 0; + + return ret; } -static void do_irq(void) +// 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) { - if(!(spuStat & STAT_IRQ)) + SPUCHAN *s_chan = &spu.s_chan[ch]; + int pos, sinc, sinc_inv, end; + unsigned char *block; + int flags; + + block = s_chan->pCurr; + pos = s_chan->spos; + sinc = s_chan->sinc; + end = pos + *upd_samples * sinc; + if (s_chan->prevflags & 1) // 1: stop/loop + block = s_chan->pLoop; + + pos += (28 - s_chan->iSBPos) << 16; + while (pos < end) + { + if (block == spu.pSpuIrq) + break; + flags = block[1]; + block += 16; + if (flags & 1) { // 1: stop/loop + block = s_chan->pLoop; + } + pos += 28 << 16; + } + + if (pos < end) { - spuStat |= STAT_IRQ; - if(irqCallback) irqCallback(); + sinc_inv = s_chan->sinc_inv; + if (sinc_inv == 0) + sinc_inv = s_chan->sinc_inv = (0x80000000u / (uint32_t)sinc) << 1; + + pos -= s_chan->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); } } -static int decode_block(int ch) +#define make_do_samples(name, fmod_code, interp_start, interp_store, interp_get, interp_end) \ +static noinline int name(int *dst, \ + int (*decode_f)(void *context, int ch, int *SB), void *ctx, \ + int ch, int ns_to, sample_buf *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->SB[(*sbpos)++]; \ + if (*sbpos >= 28) \ + { \ + *sbpos = 0; \ + d = decode_f(ctx, ch, sb->SB); \ + if (d && ns < ret) \ + ret = ns; \ + } \ + \ + interp_store; \ + *spos -= 0x10000; \ + } \ + \ + interp_get; \ + } \ + \ + interp_end; \ + \ + return ret; \ +} + +// helpers for simple linear interpolation: delay real val for two slots, +// and calc the two deltas, for a 'look at the future behaviour' +#define simple_interp_store \ + sb->SB[28] = 0; \ + sb->SB[29] = sb->SB[30]; \ + sb->SB[30] = sb->SB[31]; \ + sb->SB[31] = fa; \ + sb->SB[32] = 1 + +#define simple_interp_get \ + if(sinc<0x10000) /* -> upsampling? */ \ + InterpolateUp(sb, sinc); /* --> interpolate up */ \ + else InterpolateDown(sb, sinc); /* --> else down */ \ + dst[ns] = sb->SB[29] + +make_do_samples(do_samples_nointerp, , fa = sb->SB[29], + , dst[ns] = fa, sb->SB[29] = fa) +make_do_samples(do_samples_simple, , , + simple_interp_store, simple_interp_get, ) +make_do_samples(do_samples_gauss, , , + StoreInterpolationGaussCubic(sb, fa), + dst[ns] = GetInterpolationGauss(sb, *spos), ) +make_do_samples(do_samples_cubic, , , + StoreInterpolationGaussCubic(sb, fa), + dst[ns] = GetInterpolationCubic(sb, *spos), ) +make_do_samples(do_samples_fmod, + sinc = FModChangeFrequency(spu.s_chan[ch].iRawPitch, ns, iFMod), , + StoreInterpolationGaussCubic(sb, fa), + dst[ns] = GetInterpolationGauss(sb, *spos), ) + +INLINE int do_samples_adpcm(int *dst, + int (*decode_f)(void *context, int ch, int *SB), void *ctx, + int ch, int ns_to, int fmod, sample_buf *sb, int sinc, int *spos, int *sbpos) { - unsigned char *start; - unsigned int nSample; - int predict_nr,shift_factor,flags,d,s; - int fa,s_1,s_2; - int ret = 0; + int interp = spu.interpolation; + if (fmod == 1) + return do_samples_fmod(dst, decode_f, ctx, ch, ns_to, sb, sinc, spos, sbpos); + if (fmod) + interp = 2; + switch (interp) { + case 0: + return do_samples_nointerp(dst, decode_f, ctx, ch, ns_to, sb, sinc, spos, sbpos); + case 1: + return do_samples_simple (dst, decode_f, ctx, ch, ns_to, sb, sinc, spos, sbpos); + default: + return do_samples_gauss (dst, decode_f, ctx, ch, ns_to, sb, sinc, spos, sbpos); + case 3: + return do_samples_cubic (dst, decode_f, ctx, ch, ns_to, sb, sinc, spos, sbpos); + } +} - s_chan[ch].iSBPos=0; +static int do_samples_skip(int ch, int ns_to) +{ + SPUCHAN *s_chan = &spu.s_chan[ch]; + int spos = s_chan->spos; + int sinc = s_chan->sinc; + int ret = ns_to, ns, d; + + spos += s_chan->iSBPos << 16; - 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 + spos += sinc; + while (spos >= 28*0x10000) + { + d = skip_block(ch); + if (d && ns < ret) + ret = ns; + spos -= 28*0x10000; + } } - //////////////////////////////////////////// irq check + s_chan->iSBPos = spos >> 16; + s_chan->spos = spos & 0xffff; + + return ret; +} + +static int do_samples_skip_fmod(int ch, int ns_to, int *fmod_buf) +{ + SPUCHAN *s_chan = &spu.s_chan[ch]; + int spos = s_chan->spos; + int ret = ns_to, ns, d; + + spos += s_chan->iSBPos << 16; - if(spuCtrl&CTRL_IRQ) + for (ns = 0; ns < ns_to; ns++) { - if(pSpuIrq == start) // irq address reached? + spos += FModChangeFrequency(s_chan->iRawPitch, ns, fmod_buf); + while (spos >= 28*0x10000) { - do_irq(); // -> call main emu - ret = 1; + d = skip_block(ch); + if (d && ns < ret) + ret = ns; + spos -= 28*0x10000; } } - s_1=s_chan[ch].s_1; - s_2=s_chan[ch].s_2; + s_chan->iSBPos = spos >> 16; + s_chan->spos = spos & 0xffff; - predict_nr=(int)*start;start++; - shift_factor=predict_nr&0xf; - predict_nr >>= 4; - flags=(int)*start;start++; + return ret; +} + +static void do_lsfr_samples(int *dst, 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; - for (nSample=0;nSample<28;start++) + for (ns = 0; ns < ns_to; ns++) { - d=(int)*start; - s=((d&0xf)<<12); - if(s&0x8000) s|=0xffff0000; + counter += 2; + if (counter >= level) + { + counter -= level; + shift = (val >> 10) & 0x1f; + bit = (0x69696969 >> shift) & 1; + bit ^= (val >> 15) & 1; + val = (val << 1) | bit; + } - 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); + dst[ns] = (signed short)val; + } - s_chan[ch].SB[nSample++]=fa; + *dwNoiseCount = counter; + *dwNoiseVal = val; +} - 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; +static int do_samples_noise(int *dst, int ch, int ns_to) +{ + int ret; - s_chan[ch].SB[nSample++]=fa; - } + ret = do_samples_skip(ch, ns_to); - //////////////////////////////////////////// flag handler + do_lsfr_samples(dst, ns_to, spu.spuCtrl, &spu.dwNoiseCount, &spu.dwNoiseVal); - if((flags&4) && (!s_chan[ch].bIgnoreLoop)) - s_chan[ch].pLoop=start-16; // loop adress + return ret; +} - if(flags&1) // 1: stop/loop - { - if(!(flags&2)) - dwPendingChanOff|=1<= 0x80000) - start = (unsigned char*)-1; + l = (sval * lv) >> 14; + r = (sval * rv) >> 14; + *SSumLR++ += l; + *SSumLR++ += r; + } +} - s_chan[ch].pCurr=start; // store values for next cycle - s_chan[ch].s_1=s_1; - s_chan[ch].s_2=s_2; +static void mix_chan_rvb(int *SSumLR, int count, int lv, int rv, int *rvb) +{ + const int *src = ChanBuf; + int *dst = SSumLR; + int *drvb = rvb; + int l, r; - return ret; + while (count--) + { + int sval = *src++; + + l = (sval * lv) >> 14; + r = (sval * rv) >> 14; + *dst++ += l; + *dst++ += r; + *drvb++ += l; + *drvb++ += r; + } } +#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 fa,ns,ns_from,ns_to; -#if !defined(_MACOSX) && !defined(__arm__) - int voldiv = iVolume; -#else - const int voldiv = 2; -#endif - int ch,d; - int bIRQReturn=0; + unsigned int mask; + SPUCHAN *s_chan; + int ch; - while(!bEndThread) // until we are shutting down + mask = silentch & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) { - // 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 (!(mask & 1)) continue; + if (spu.dwChannelDead & (1<pCurr > spu.pSpuIrq && s_chan->pLoop > spu.pSpuIrq) + continue; - while(!iSecureStart && !bEndThread && // no new start? no thread end? - (SoundGetBytesBuffered()>TESTSIZE)) // and still enuff data in sound buffer? + s_chan->spos += s_chan->iSBPos << 16; + s_chan->iSBPos = 0; + + s_chan->spos += s_chan->sinc * ns_to; + while (s_chan->spos >= 28 * 0x10000) { - iSecureStart=0; // reset secure + unsigned char *start = s_chan->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->pCurr || start - spu.spuMemC < 0x1000) + { + // looping on self or stopped(?) + spu.dwChannelDead |= 1<spos = 0; + break; + } - if(dwNewChannel) iSecureStart=1; // if a new channel kicks in (or, of course, sound buffer runs low), we will leave the loop + s_chan->spos -= 28 * 0x10000; } + } +} - //--------------------------------------------------// continue from irq handling in timer mode? +static void do_channels(int ns_to) +{ + unsigned int mask; + int do_rvb, ch, d; + SPUCHAN *s_chan; - ns_from=0; - ns_to=NSSIZE; - ch=0; - if(lastch>=0) // will be -1 if no continue is pending - { - ch=lastch; ns_from=lastns; lastch=-1; // -> setup all kind of vars to continue + if (unlikely(spu.interpolation != spu_config.iUseInterpolation)) + { + spu.interpolation = spu_config.iUseInterpolation; + mask = spu.dwChannelsAudible & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) + if (mask & 1) + ResetInterpolation(&spu.sb[ch]); + } + + do_rvb = spu.rvb->StartAddr && spu_config.iUseReverb; + if (do_rvb) + memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2); + + mask = spu.dwNewChannel & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) { + if (mask & 1) + StartSound(ch); + } + + mask = spu.dwChannelsAudible & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) // loop em all... + { + if (!(mask & 1)) continue; // channel not playing? next + + s_chan = &spu.s_chan[ch]; + if (s_chan->bNoise) + d = do_samples_noise(ChanBuf, ch, ns_to); + else + d = do_samples_adpcm(ChanBuf, decode_block, NULL, ch, ns_to, s_chan->bFMod, + &spu.sb[ch], s_chan->sinc, &s_chan->spos, &s_chan->iSBPos); + + if (!s_chan->bStarting) { + d = MixADSR(ChanBuf, &s_chan->ADSRX, d); + if (d < ns_to) { + spu.dwChannelsAudible &= ~(1 << ch); + s_chan->ADSRX.State = ADSR_RELEASE; + s_chan->ADSRX.EnvelopeVol = 0; + memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0])); } + } - //--------------------------------------------------// - //- main channel loop -// - //--------------------------------------------------// + if (ch == 1 || ch == 3) { - for(;chbFMod == 2) // fmod freq channel + memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0])); + if (!(spu.spuCtrl & CTRL_MUTE)) + ; + else if (s_chan->bRVBActive && do_rvb) + mix_chan_rvb(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume, RVB); + else + mix_chan(spu.SSumLR, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume); + } - if(s_chan[ch].bFMod==1 && iFMod[ns]) // fmod freq channel - FModChangeFrequency(ch,ns); + MixCD(spu.SSumLR, RVB, ns_to, spu.decode_pos); - 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 - } - } + if (spu.rvb->StartAddr) { + if (do_rvb) + REVERBDo(spu.SSumLR, RVB, ns_to, spu.rvb->CurrAddr); - fa=s_chan[ch].SB[s_chan[ch].iSBPos++]; // get sample data + spu.rvb->CurrAddr += ns_to / 2; + while (spu.rvb->CurrAddr >= 0x40000) + spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr; + } +} - StoreInterpolationVal(ch,fa); // store val for later interpolation +static void do_samples_finish(int *SSumLR, int ns_to, + int silentch, int decode_pos); + +// optional worker thread handling + +#if P_HAVE_PTHREAD || defined(WANT_THREAD_CODE) + +// worker thread state +static struct spu_worker { + union { + struct { + unsigned char exit_thread; + unsigned char prev_work_in_thread; + unsigned char pad[2]; + unsigned int i_ready; + unsigned int i_reaped; + unsigned int last_boot_cnt; // dsp + unsigned int ram_dirty; + unsigned int channels_last; + }; + // aligning for C64X_DSP + unsigned int _pad0[128/4]; + }; + union { + struct { + unsigned int i_done; + unsigned int active; // dsp + unsigned int boot_cnt; + }; + unsigned int _pad1[128/4]; + }; + struct work_item { + int ns_to; + int ctrl; + int decode_pos; + int rvb_addr; + unsigned int channels_new; + unsigned int channels_on; + unsigned int channels_silent; + struct { + int spos; + int sbpos; + int sinc; + int start; + int loop; + short vol_l; + short vol_r; + unsigned short ns_to; + unsigned short bNoise:1; + unsigned short bFMod:2; + unsigned short bRVBActive:1; + unsigned short bStarting:1; + ADSRInfoEx adsr; + } ch[24]; + int SSumLR[NSSIZE * 2]; + } i[4]; +} *worker; + +#define WORK_MAXCNT (sizeof(worker->i) / sizeof(worker->i[0])) +#define WORK_I_MASK (WORK_MAXCNT - 1) + +static void thread_work_start(void); +static void thread_work_wait_sync(struct work_item *work, int force); +static void thread_sync_caches(void); +static int thread_get_i_done(void); + +static int decode_block_work(void *context, int ch, int *SB) +{ + const unsigned char *ram = spu.spuMemC; + int predict_nr, shift_factor, flags; + struct work_item *work = context; + int start = work->ch[ch].start; + int loop = work->ch[ch].loop; + + predict_nr = ram[start]; + shift_factor = predict_nr & 0xf; + predict_nr >>= 4; - s_chan[ch].spos -= 0x10000L; - } + decode_block_data(SB, ram + start + 2, predict_nr, shift_factor); - if(s_chan[ch].bNoise) - fa=iGetNoiseVal(ch); // get noise val - else fa=iGetInterpolationVal(ch); // get sample val - ChanBuf[ns]=fa; + flags = ram[start + 1]; + if (flags & 4) + loop = start; // loop adress - //////////////////////////////////////////////// - // ok, go on until 1 ms data of this channel is collected + start += 16; - s_chan[ch].spos += s_chan[ch].sinc; -ENDX: ; - } + if (flags & 1) // 1: stop/loop + start = loop; - MixADSR(ch, ns_from, ns_to); + work->ch[ch].start = start & 0x7ffff; + work->ch[ch].loop = loop; - if(s_chan[ch].bFMod==2) // fmod freq channel - memcpy(iFMod, ChanBuf, sizeof(iFMod)); - else for(ns=ns_from;nsi[worker->i_ready & WORK_I_MASK]; + work->ns_to = ns_to; + work->ctrl = spu.spuCtrl; + work->decode_pos = spu.decode_pos; + work->channels_silent = silentch; + + if (!worker->prev_work_in_thread) { + // copy adpcm and interpolation state to sb_thread + worker->prev_work_in_thread = 1; + mask = spu.dwChannelsAudible & ~spu.dwNewChannel & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) { + if (mask & 1) + memcpy(spu.sb_thread[ch].SB, spu.sb[ch].SB, sizeof(spu.sb_thread[ch].SB)); + } + } - SSumLR[ns*2] +=(sval*s_chan[ch].iLeftVolume)/0x4000L; - SSumLR[ns*2+1]+=(sval*s_chan[ch].iRightVolume)/0x4000L; + mask = work->channels_new = spu.dwNewChannel & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) { + if (mask & 1) + StartSound(ch); + } - ////////////////////////////////////////////// - // now let us store sound data for reverb + mask = work->channels_on = spu.dwChannelsAudible & 0xffffff; + worker->channels_last = mask; + spu.decode_dirty_ch |= mask & 0x0a; - if(s_chan[ch].bRVBActive) StoreREVERB(ch,ns,sval); - } - } - } + for (ch = 0; mask != 0; ch++, mask >>= 1) + { + if (!(mask & 1)) continue; + + s_chan = &spu.s_chan[ch]; + work->ch[ch].spos = s_chan->spos; + work->ch[ch].sbpos = s_chan->iSBPos; + work->ch[ch].sinc = s_chan->sinc; + work->ch[ch].adsr = s_chan->ADSRX; + work->ch[ch].vol_l = s_chan->iLeftVolume; + work->ch[ch].vol_r = s_chan->iRightVolume; + work->ch[ch].start = s_chan->pCurr - spu.spuMemC; + work->ch[ch].loop = s_chan->pLoop - spu.spuMemC; + work->ch[ch].bNoise = s_chan->bNoise; + work->ch[ch].bFMod = s_chan->bFMod; + work->ch[ch].bRVBActive = s_chan->bRVBActive; + work->ch[ch].bStarting = s_chan->bStarting; + if (s_chan->prevflags & 1) + work->ch[ch].start = work->ch[ch].loop; + + if (unlikely(s_chan->bFMod == 2)) + { + // sucks, have to do double work + if (s_chan->bNoise) + d = do_samples_noise(tmpFMod, ch, ns_to); + else + d = do_samples_gauss(tmpFMod, decode_block, NULL, ch, ns_to, + &spu.sb[ch], s_chan->sinc, &s_chan->spos, &s_chan->iSBPos); + if (!s_chan->bStarting) { + d = MixADSR(tmpFMod, &s_chan->ADSRX, d); + if (d < ns_to) { + spu.dwChannelsAudible &= ~(1 << ch); + s_chan->ADSRX.State = ADSR_RELEASE; + s_chan->ADSRX.EnvelopeVol = 0; + } + } + memset(&tmpFMod[d], 0, (ns_to - d) * sizeof(tmpFMod[d])); + work->ch[ch].ns_to = d; + continue; + } + if (unlikely(s_chan->bFMod)) + d = do_samples_skip_fmod(ch, ns_to, tmpFMod); + else + d = do_samples_skip(ch, ns_to); + work->ch[ch].ns_to = d; + + if (!s_chan->bStarting) { + // note: d is not accurate on skip + d = SkipADSR(&s_chan->ADSRX, d); + if (d < ns_to) { + spu.dwChannelsAudible &= ~(1 << ch); + s_chan->ADSRX.State = ADSR_RELEASE; + s_chan->ADSRX.EnvelopeVol = 0; } + } + } // for (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; - - 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; - } - } + work->rvb_addr = 0; + if (spu.rvb->StartAddr) { + if (spu_config.iUseReverb) + work->rvb_addr = spu.rvb->CurrAddr; - 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()CurrAddr += ns_to / 2; + while (spu.rvb->CurrAddr >= 0x40000) + spu.rvb->CurrAddr -= 0x40000 - spu.rvb->StartAddr; + } + worker->i_ready++; + thread_work_start(); +} - //---------------------------------------------------// - //- here we have another 1 ms of sound data - //---------------------------------------------------// - // mix XA infos (if any) +static void do_channel_work(struct work_item *work) +{ + unsigned int mask; + int spos, sbpos; + int d, ch, ns_to; - MixXA(); - - /////////////////////////////////////////////////////// - // mix all channels (including reverb) into one buffer + ns_to = work->ns_to; - for (ns = 0; ns < NSSIZE*2; ) - { - SSumLR[ns] += MixREVERBLeft(ns/2); + if (unlikely(spu.interpolation != spu_config.iUseInterpolation)) + { + spu.interpolation = spu_config.iUseInterpolation; + mask = work->channels_on; + for (ch = 0; mask != 0; ch++, mask >>= 1) + if (mask & 1) + ResetInterpolation(&spu.sb_thread[ch]); + } - d = SSumLR[ns] / voldiv; SSumLR[ns] = 0; - if (d < -32767) d = -32767; if (d > 32767) d = 32767; - *pS++ = d; - ns++; + if (work->rvb_addr) + memset(RVB, 0, ns_to * sizeof(RVB[0]) * 2); + + mask = work->channels_new; + for (ch = 0; mask != 0; ch++, mask >>= 1) { + if (mask & 1) + StartSoundSB(&spu.sb_thread[ch]); + } - SSumLR[ns] += MixREVERBRight(); + mask = work->channels_on; + for (ch = 0; mask != 0; ch++, mask >>= 1) + { + if (!(mask & 1)) continue; - d = SSumLR[ns] / voldiv; SSumLR[ns] = 0; - if(d < -32767) d = -32767; if(d > 32767) d = 32767; - *pS++ = d; - ns++; + d = work->ch[ch].ns_to; + spos = work->ch[ch].spos; + sbpos = work->ch[ch].sbpos; + + if (work->ch[ch].bNoise) + do_lsfr_samples(ChanBuf, d, work->ctrl, &spu.dwNoiseCount, &spu.dwNoiseVal); + else + do_samples_adpcm(ChanBuf, decode_block_work, work, ch, d, work->ch[ch].bFMod, + &spu.sb_thread[ch], work->ch[ch].sinc, &spos, &sbpos); + + d = MixADSR(ChanBuf, &work->ch[ch].adsr, d); + if (d < ns_to) { + work->ch[ch].adsr.EnvelopeVol = 0; + memset(&ChanBuf[d], 0, (ns_to - d) * sizeof(ChanBuf[0])); } - ////////////////////////////////////////////////////// + if (ch == 1 || ch == 3) + do_decode_bufs(spu.spuMem, ch/2, ns_to, work->decode_pos); + + if (work->ch[ch].bFMod == 2) // fmod freq channel + memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0])); + if (work->ch[ch].bRVBActive && work->rvb_addr) + mix_chan_rvb(work->SSumLR, ns_to, + work->ch[ch].vol_l, work->ch[ch].vol_r, RVB); + else + mix_chan(work->SSumLR, ns_to, work->ch[ch].vol_l, work->ch[ch].vol_r); + } + + if (work->rvb_addr) + REVERBDo(work->SSumLR, RVB, ns_to, work->rvb_addr); +} + +static void sync_worker_thread(int force_no_thread) +{ + int force = force_no_thread; + struct work_item *work; + int done, used_space; + + // rvb offsets will change, thread may be using them + force |= spu.rvb->dirty && spu.rvb->StartAddr; + + done = thread_get_i_done() - worker->i_reaped; + used_space = worker->i_ready - worker->i_reaped; + + //printf("done: %d use: %d dsp: %u/%u\n", done, used_space, + // worker->boot_cnt, worker->i_done); + + while ((force && used_space > 0) || used_space >= WORK_MAXCNT || done > 0) { + work = &worker->i[worker->i_reaped & WORK_I_MASK]; + thread_work_wait_sync(work, force); + + MixCD(work->SSumLR, RVB, work->ns_to, work->decode_pos); + do_samples_finish(work->SSumLR, work->ns_to, + work->channels_silent, work->decode_pos); + + worker->i_reaped++; + done = thread_get_i_done() - worker->i_reaped; + used_space = worker->i_ready - worker->i_reaped; + } + if (force_no_thread && worker->prev_work_in_thread) { + unsigned int ch, mask = worker->channels_last; + worker->prev_work_in_thread = 0; + thread_sync_caches(); + for (ch = 0; mask != 0; ch++, mask >>= 1) { + if (mask & 1) + memcpy(spu.sb[ch].SB, spu.sb_thread[ch].SB, sizeof(spu.sb_thread[ch].SB)); + } + } +} + +#else + +static void queue_channel_work(int ns_to, int silentch) {} +static void sync_worker_thread(int force_no_thread) {} + +static const void * const worker = NULL; + +#endif // P_HAVE_PTHREAD || defined(WANT_THREAD_CODE) + +//////////////////////////////////////////////////////////////////////// +// MAIN SPU FUNCTION +// here is the main job handler... +//////////////////////////////////////////////////////////////////////// + +void do_samples(unsigned int cycles_to, int force_no_thread) +{ + unsigned int silentch; + int cycle_diff; + int ns_to; + + cycle_diff = cycles_to - spu.cycles_played; + if (cycle_diff < -2*1048576 || cycle_diff > 2*1048576) + { + log_unhandled("desync %u %d\n", cycles_to, cycle_diff); + spu.cycles_played = cycles_to; + return; + } + + silentch = ~(spu.dwChannelsAudible | spu.dwNewChannel) & 0xffffff; + + force_no_thread |= (silentch == 0xffffff); + if (worker != NULL) + sync_worker_thread(force_no_thread); + + if (cycle_diff < 2 * 768) + return; + + ns_to = (cycle_diff / 768 + 1) & ~1; + if (ns_to > NSSIZE) { + // should never happen + log_unhandled("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(0); } } + if (!spu.cycles_dma_end || (int)(spu.cycles_dma_end - cycles_to) < 0) { + spu.cycles_dma_end = 0; + check_irq_io(spu.spuAddr); + } + + if (unlikely(spu.rvb->dirty)) + REVERBPrep(); + + if (force_no_thread || worker == NULL || !spu_config.iUseThread) { + do_channels(ns_to); + do_samples_finish(spu.SSumLR, ns_to, silentch, spu.decode_pos); + } + else { + queue_channel_work(ns_to, silentch); + //sync_worker_thread(1); // uncomment for debug + } + + // 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; + spu.spuStat = (spu.spuStat & ~0x800) | ((spu.decode_pos << 3) & 0x800); +#if 0 + static int ccount; static time_t ctime; ccount++; + if (time(NULL) != ctime) + { printf("%d\n", ccount); ccount = 0; ctime = time(NULL); } +#endif +} - InitREVERB(); +static void do_samples_finish(int *SSumLR, int ns_to, + int silentch, int decode_pos) +{ + int vol_l = ((int)regAreaGet(H_SPUmvolL) << 17) >> 17; + int vol_r = ((int)regAreaGet(H_SPUmvolR) << 17) >> 17; + int ns; + int d; - // feed the sound - // wanna have around 1/60 sec (16.666 ms) updates - if (iCycle++ > 16) + // must clear silent channel decode buffers + if(unlikely(silentch & spu.decode_dirty_ch & (1<<1))) { - 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... + vol_l = vol_l * spu_config.iVolume >> 10; + vol_r = vol_r * spu_config.iVolume >> 10; - bThreadEnded = 1; + if (!(vol_l | vol_r)) + { + // muted? (rare) + memset(spu.pS, 0, ns_to * 2 * sizeof(spu.pS[0])); + memset(SSumLR, 0, ns_to * 2 * sizeof(SSumLR[0])); + spu.pS += ns_to * 2; + } + else + for (ns = 0; ns < ns_to * 2; ) + { + d = SSumLR[ns]; SSumLR[ns] = 0; + d = d * vol_l >> 14; + ssat32_to_16(d); + *spu.pS++ = d; + ns++; - return 0; + d = SSumLR[ns]; SSumLR[ns] = 0; + d = d * vol_r >> 14; + 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 - spu.s_chan[ch].pCurr) > IRQ_NEAR_BLOCKS * 16 + && (unsigned long)(spu.pSpuIrq - spu.s_chan[ch].pLoop) > IRQ_NEAR_BLOCKS * 16) + continue; + if (spu.s_chan[ch].sinc == 0) + 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 @@ -844,152 +1373,255 @@ void CALLBACK SPUasync(unsigned long cycle) void CALLBACK SPUupdate(void) { - SPUasync(0); } // XA AUDIO -void CALLBACK SPUplayADPCMchannel(xa_decode_t *xap) +void CALLBACK SPUplayADPCMchannel(xa_decode_t *xap, unsigned int cycle, int is_start) { if(!xap) return; - if(!xap->freq) return; // no xa freq ? bye + if(!xap->freq) return; // no xa freq ? bye + + if (is_start) + spu.XAPlay = spu.XAFeed = spu.XAStart; + if (spu.XAPlay == spu.XAFeed) + do_samples(cycle, 1); // catch up to prevent source underflows later - FeedXA(xap); // call main XA feeder + FeedXA(xap); // call main XA feeder + spu.xapGlobal = xap; // store info for save states + spu.cdClearSamples = 512; } // CDDA AUDIO -void CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes) +int CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes, unsigned int cycle, int unused) { - if (!pcm) return; - if (nbytes<=0) return; + if (!pcm) return -1; + if (nbytes<=0) return -1; + + if (spu.CDDAPlay == spu.CDDAFeed) + do_samples(cycle, 1); // catch up to prevent source underflows later FeedCDDA((unsigned char *)pcm, nbytes); + spu.cdClearSamples = 512; + return 0; } -// SETUPTIMER: init of certain buffers and threads/timers -void SetupTimer(void) +void CALLBACK SPUsetCDvol(unsigned char ll, unsigned char lr, + unsigned char rl, unsigned char rr, unsigned int cycle) { - memset(SSumLR,0,sizeof(SSumLR)); // init some mixing buffers - memset(iFMod,0,NSSIZE*sizeof(int)); - pS=(short *)pSpuBuffer; // setup soundbuffer pointer + if (spu.XAPlay != spu.XAFeed || spu.CDDAPlay != spu.CDDAFeed) + do_samples(cycle, 1); + spu.cdv.ll = ll; + spu.cdv.lr = lr; + spu.cdv.rl = rl; + spu.cdv.rr = rr; +} - bEndThread=0; // init thread vars - bThreadEnded=0; - bSpuInit=1; // flag: we are inited +// to be called after state load +void ClearWorkingState(void) +{ + memset(iFMod, 0, sizeof(iFMod)); + spu.pS=(short *)spu.pSpuBuffer; // setup soundbuffer pointer +} - if(!iUseTimer) // linux: use thread - { - pthread_create(&thread, NULL, MAINThread, NULL); - } +// SETUPSTREAMS: init most of the spu buffers +static void SetupStreams(void) +{ + spu.pSpuBuffer = (unsigned char *)malloc(32768); // alloc mixing buffer + spu.SSumLR = calloc(NSSIZE * 2, sizeof(spu.SSumLR[0])); + + spu.XAStart = malloc(44100 * sizeof(uint32_t)); // alloc xa buffer + spu.XAEnd = spu.XAStart + 44100; + spu.XAPlay = spu.XAStart; + spu.XAFeed = spu.XAStart; + + spu.CDDAStart = malloc(CDDA_BUFFER_SIZE); // alloc cdda buffer + spu.CDDAEnd = spu.CDDAStart + CDDA_BUFFER_SIZE / sizeof(uint32_t); + spu.CDDAPlay = spu.CDDAStart; + spu.CDDAFeed = spu.CDDAStart; + + ClearWorkingState(); +} + +// REMOVESTREAMS: free most buffer +static void RemoveStreams(void) +{ + free(spu.pSpuBuffer); // free mixing buffer + spu.pSpuBuffer = NULL; + free(spu.SSumLR); + spu.SSumLR = NULL; + free(spu.XAStart); // free XA buffer + spu.XAStart = NULL; + free(spu.CDDAStart); // free CDDA buffer + spu.CDDAStart = NULL; } -// REMOVETIMER: kill threads/timers -void RemoveTimer(void) +#if defined(C64X_DSP) + +/* special code for TI C64x DSP */ +#include "spu_c64x.c" + +#elif P_HAVE_PTHREAD + +#include +#include +#include + +static struct { + pthread_t thread; + sem_t sem_avail; + sem_t sem_done; +} t; + +/* generic pthread implementation */ + +static void thread_work_start(void) { - bEndThread=1; // raise flag to end thread + sem_post(&t.sem_avail); +} - 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 - } +static void thread_work_wait_sync(struct work_item *work, int force) +{ + sem_wait(&t.sem_done); +} - bThreadEnded=0; // no more spu is running - bSpuInit=0; +static int thread_get_i_done(void) +{ + return worker->i_done; } -// SETUPSTREAMS: init most of the spu buffers -void SetupStreams(void) -{ - int i; +static void thread_sync_caches(void) +{ +} - pSpuBuffer=(unsigned char *)malloc(32768); // alloc mixing buffer +static void *spu_worker_thread(void *unused) +{ + struct work_item *work; - if(iUseReverb==1) i=88200*2; - else i=NSSIZE*2; + while (1) { + sem_wait(&t.sem_avail); + if (worker->exit_thread) + break; - sRVBStart = (int *)malloc(i*4); // alloc reverb buffer - memset(sRVBStart,0,i*4); - sRVBEnd = sRVBStart + i; - sRVBPlay = sRVBStart; + work = &worker->i[worker->i_done & WORK_I_MASK]; + do_channel_work(work); + worker->i_done++; - XAStart = // alloc xa buffer - (uint32_t *)malloc(44100 * sizeof(uint32_t)); - XAEnd = XAStart + 44100; - XAPlay = XAStart; - XAFeed = XAStart; + sem_post(&t.sem_done); + } - CDDAStart = // alloc cdda buffer - (uint32_t *)malloc(16384 * sizeof(uint32_t)); - CDDAEnd = CDDAStart + 16384; - CDDAPlay = CDDAStart; - CDDAFeed = CDDAStart; + return NULL; +} - for(i=0;i init sustain - s_chan[i].pLoop=spuMemC; - s_chan[i].pStart=spuMemC; - s_chan[i].pCurr=spuMemC; - } +static void init_spu_thread(void) +{ + int ret; + + spu.sb_thread = spu.sb_thread_; + + if (sysconf(_SC_NPROCESSORS_ONLN) <= 1) + return; + + worker = calloc(1, sizeof(*worker)); + if (worker == NULL) + return; + ret = sem_init(&t.sem_avail, 0, 0); + if (ret != 0) + goto fail_sem_avail; + ret = sem_init(&t.sem_done, 0, 0); + if (ret != 0) + goto fail_sem_done; + + ret = pthread_create(&t.thread, NULL, spu_worker_thread, NULL); + if (ret != 0) + goto fail_thread; + + spu_config.iThreadAvail = 1; + return; + +fail_thread: + sem_destroy(&t.sem_done); +fail_sem_done: + sem_destroy(&t.sem_avail); +fail_sem_avail: + free(worker); + worker = NULL; + spu_config.iThreadAvail = 0; +} - pMixIrq=spuMemC; // enable decoded buffer irqs by setting the address +static void exit_spu_thread(void) +{ + if (worker == NULL) + return; + worker->exit_thread = 1; + sem_post(&t.sem_avail); + pthread_join(t.thread, NULL); + sem_destroy(&t.sem_done); + sem_destroy(&t.sem_avail); + free(worker); + worker = NULL; } -// 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; +#else // if !P_HAVE_PTHREAD + +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 - memset((void *)&rvb, 0, sizeof(REVERBInfo)); + int i; + + memset(&spu, 0, sizeof(spu)); + spu.spuMemC = calloc(1, 512 * 1024 + 16); + // a guard for runaway channels - End+Mute + spu.spuMemC[512 * 1024 + 1] = 1; + 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.s_chan = calloc(MAXCHAN+1, sizeof(spu.s_chan[0])); // channel + 1 infos (1 is security for fmod handling) + spu.rvb = calloc(1, sizeof(REVERBInfo)); + + spu.spuAddr = 0; + spu.decode_pos = 0; + spu.pSpuIrq = spu.spuMemC; + SetupStreams(); // prepare streaming + if (spu_config.iVolume == 0) + spu_config.iVolume = 768; // 1024 is 1.0 + + init_spu_thread(); + + for (i = 0; i < MAXCHAN; i++) // loop sound channels + { + spu.s_chan[i].ADSRX.SustainLevel = 0xf; // -> init sustain + spu.s_chan[i].ADSRX.SustainIncrease = 1; + spu.s_chan[i].pLoop = spu.spuMemC; + spu.s_chan[i].pCurr = spu.spuMemC; + spu.s_chan[i].bIgnoreLoop = 0; + } + + spu.bSpuInit=1; // flag: we are inited + 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; } @@ -997,12 +1629,11 @@ 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; } @@ -1011,49 +1642,38 @@ long CALLBACK SPUclose(void) long CALLBACK SPUshutdown(void) { SPUclose(); - RemoveStreams(); // no more streaming - return 0; -} + exit_spu_thread(); -// SPUTEST: we don't test, we are always fine ;) -long CALLBACK SPUtest(void) -{ - return 0; -} + free(spu.spuMemC); + spu.spuMemC = NULL; + free(spu.s_chan); + spu.s_chan = NULL; + free(spu.rvb); + spu.rvb = NULL; -// SPUCONFIGURE: call config dialog -long CALLBACK SPUconfigure(void) -{ -#ifdef _MACOSX - DoConfiguration(); -#else -// StartCfgTool("CFG"); -#endif - return 0; -} + RemoveStreams(); // no more streaming + spu.bSpuInit=0; -// SPUABOUT: show about window -void CALLBACK SPUabout(void) -{ -#ifdef _MACOSX - DoAbout(); -#else -// StartCfgTool("ABOUT"); -#endif + return 0; } // SETUP CALLBACKS // this functions will be called once, // passes a callback that should be called on SPU-IRQ/cdda volume change -void CALLBACK SPUregisterCallback(void (CALLBACK *callback)(void)) +void CALLBACK SPUregisterCallback(void (CALLBACK *callback)(int)) +{ + spu.irqCallback = callback; +} + +void CALLBACK SPUregisterCDDAVolume(void (CALLBACK *CDDAVcallback)(short, short)) { - irqCallback = callback; + //spu.cddavCallback = CDDAVcallback; } -void CALLBACK SPUregisterCDDAVolume(void (CALLBACK *CDDAVcallback)(unsigned short,unsigned short)) +void CALLBACK SPUregisterScheduleCb(void (CALLBACK *callback)(unsigned int)) { - cddavCallback = CDDAVcallback; + spu.scheduleCallback = callback; } // COMMON PLUGIN INFO FUNCS @@ -1080,21 +1700,27 @@ char * SPUgetLibInfos(void) */ // debug -void spu_get_debug_info(int *chans_out, int *fmod_chans_out, int *noise_chans_out) +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; + int ch = 0, fmod_chans = 0, noise_chans = 0, irq_chans = 0; + + if (spu.s_chan == NULL) + return; for(;ch