X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=plugins%2Fdfsound%2Fspu.c;h=c6a06fd446072b7dc87dbb6de0b97e1169978f93;hp=2c13ca16fef83f6879eb08b1e8d1254e294a7265;hb=de4a0279efefdd2e4595c8fc27f1564f4bff9341;hpb=c4c66b22ae607aa8dc5e6983f7ab030ead01faae diff --git a/plugins/dfsound/spu.c b/plugins/dfsound/spu.c index 2c13ca16..c6a06fd4 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 + Portions (C) Gražvydas "notaz" Ignotas, 2010-2012,2014,2015 ***************************************************************************/ /*************************************************************************** @@ -18,6 +18,10 @@ * * ***************************************************************************/ +#if !defined(_WIN32) && !defined(NO_OS) +#include // gettimeofday in xa.c +#define THREAD_ENABLED 1 +#endif #include "stdafx.h" #define _IN_SPU @@ -25,17 +29,10 @@ #include "externals.h" #include "registers.h" #include "out.h" -#include "arm_features.h" #include "spu_config.h" -#ifdef ENABLE_NLS -#include -#include -#define _(x) gettext(x) -#define N_(x) (x) -#else -#define _(x) (x) -#define N_(x) (x) +#ifdef __arm__ +#include "arm_features.h" #endif #ifdef __ARM_ARCH_7A__ @@ -78,19 +75,49 @@ 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; +#if defined(THREAD_ENABLED) || defined(WANT_THREAD_CODE) + +// worker thread state +static struct spu_worker { + unsigned int pending:1; + unsigned int exit_thread:1; + unsigned int stale_cache:1; + int ns_to; + int ctrl; + int decode_pos; + int silentch; + unsigned int chmask; + 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]; + struct { + struct { + int adsrState; + int adsrEnvelopeVol; + } ch[24]; + unsigned int chan_end; + unsigned int decode_dirty; + } r; +} *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 }, - { 60, 0 }, - { 115, -52 }, - { 98, -55 }, - { 122, -60 } }; +static int iFMod[NSSIZE]; int ChanBuf[NSSIZE]; -int SSumLR[NSSIZE*2]; -int iFMod[NSSIZE]; +int *SSumLR; #define CDDA_BUFFER_SIZE (16384 * sizeof(uint32_t)) // must be power of 2 @@ -145,71 +172,70 @@ 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" @@ -243,45 +269,46 @@ static int check_irq(int ch, unsigned char *pos) INLINE void StartSound(int ch) { + SPUCHAN *s_chan = &spu.s_chan[ch]; + int *SB = spu.SB + ch * SB_SIZE; + 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[ch].SB[26]=0; // init mixing vars - s_chan[ch].SB[27]=0; - s_chan[ch].iSBPos=28; + s_chan->iSBPos=27; + SB[26]=0; // init mixing vars + SB[27]=0; - s_chan[ch].SB[29]=0; // init our interpolation helpers - s_chan[ch].SB[30]=0; - - if(spu_config.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 + SB[28]=0; + SB[29]=0; // init our interpolation helpers + SB[30]=0; + SB[31]=0; + s_chan->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; @@ -289,50 +316,50 @@ 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; @@ -351,7 +378,7 @@ 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; @@ -361,15 +388,15 @@ 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; //--------------------------------------------------// } @@ -379,6 +406,13 @@ INLINE int iGetInterpolationVal(int ch, int spos) 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; @@ -405,56 +439,49 @@ 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) { + SPUCHAN *s_chan = &spu.s_chan[ch]; unsigned char *start; - int predict_nr,shift_factor,flags; - int stop = 0; + int predict_nr, shift_factor, flags; int ret = 0; - start = s_chan[ch].pCurr; // set up the current pos - if(start == spu.spuMemC) // ? - stop = 1; + start = s_chan->pCurr; // set up the current pos + if (start == spu.spuMemC) // ? + ret = 1; - if(s_chan[ch].prevflags&1) // 1: stop/loop + if (s_chan->prevflags & 1) // 1: stop/loop { - if(!(s_chan[ch].prevflags&2)) - stop = 1; + if (!(s_chan->prevflags & 2)) + ret = 1; - start = s_chan[ch].pLoop; + start = s_chan->pLoop; } else - ret = check_irq(ch, start); // hack, see check_irq below.. + check_irq(ch, start); // hack, see check_irq below.. - if(stop) - { - spu.dwChannelOn &= ~(1< turn everything off - s_chan[ch].bStop = 1; - s_chan[ch].ADSRX.EnvelopeVol = 0; - } - - predict_nr=(int)start[0]; - shift_factor=predict_nr&0xf; + predict_nr = start[0]; + 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=(int)start[1]; - if(flags&4) - s_chan[ch].pLoop=start; // loop adress + flags = start[1]; + if (flags & 4) + s_chan->pLoop = start; // loop adress - start+=16; + start += 16; - if(flags&1) { // 1: stop/loop - start = s_chan[ch].pLoop; - ret |= check_irq(ch, start); // hack.. :( + if (flags & 1) { // 1: stop/loop + start = s_chan->pLoop; + check_irq(ch, start); // hack.. :( } if (start - spu.spuMemC >= 0x80000) start = spu.spuMemC; - s_chan[ch].pCurr = start; // store values for next cycle - s_chan[ch].prevflags = flags; + s_chan->pCurr = start; // store values for next cycle + s_chan->prevflags = flags; return ret; } @@ -462,40 +489,83 @@ static int decode_block(int ch) // do block, but ignore sample data static int skip_block(int ch) { - unsigned char *start = s_chan[ch].pCurr; - int flags = start[1]; - int ret = check_irq(ch, start); + SPUCHAN *s_chan = &spu.s_chan[ch]; + unsigned char *start = s_chan->pCurr; + int flags; + int ret = 0; - if(s_chan[ch].prevflags & 1) - start = s_chan[ch].pLoop; + if (s_chan->prevflags & 1) { + if (!(s_chan->prevflags & 2)) + ret = 1; - if(flags & 4) - s_chan[ch].pLoop = start; + start = s_chan->pLoop; + } + else + check_irq(ch, start); + + flags = start[1]; + if (flags & 4) + s_chan->pLoop = start; start += 16; - if(flags & 1) - start = s_chan[ch].pLoop; + if (flags & 1) { + start = s_chan->pLoop; + check_irq(ch, start); + } - s_chan[ch].pCurr = start; - s_chan[ch].prevflags = flags; + s_chan->pCurr = start; + s_chan->prevflags = flags; return ret; } +#if defined(THREAD_ENABLED) || defined(WANT_THREAD_CODE) + +static int decode_block_work(int ch, int *SB) +{ + const unsigned char *ram = spu.spuMemC; + int predict_nr, shift_factor, flags; + 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) { + SPUCHAN *s_chan = &spu.s_chan[ch]; 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; + block = s_chan->pCurr; + pos = s_chan->spos; + sinc = s_chan->sinc; end = pos + *upd_samples * sinc; - pos += (28 - s_chan[ch].iSBPos) << 16; + pos += (28 - s_chan->iSBPos) << 16; while (pos < end) { if (block == spu.pSpuIrq) @@ -503,7 +573,7 @@ static void scan_for_irq(int ch, unsigned int *upd_samples) flags = block[1]; block += 16; if (flags & 1) { // 1: stop/loop - block = s_chan[ch].pLoop; + block = s_chan->pLoop; if (block == spu.pSpuIrq) // hack.. (see decode_block) break; } @@ -512,11 +582,11 @@ static void scan_for_irq(int ch, unsigned int *upd_samples) if (pos < end) { - sinc_inv = s_chan[ch].sinc_inv; + sinc_inv = s_chan->sinc_inv; if (sinc_inv == 0) - sinc_inv = s_chan[ch].sinc_inv = (0x80000000u / (uint32_t)sinc) << 1; + sinc_inv = s_chan->sinc_inv = (0x80000000u / (uint32_t)sinc) << 1; - pos -= s_chan[ch].spos; + 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); @@ -524,106 +594,130 @@ 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, 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; \ \ - 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); \ - if(d) \ - ret = /*ns_to =*/ ns + 1; \ + *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; \ } #define fmod_recv_check \ - if(s_chan[ch].bFMod==1 && iFMod[ns]) \ - sinc = FModChangeFrequency(ch,ns) + if(spu.s_chan[ch].bFMod==1 && iFMod[ns]) \ + sinc = FModChangeFrequency(SB, spu.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, spu.s_chan[ch].bFMod==2), + ChanBuf[ns] = iGetInterpolationVal(SB, sinc, *spos, spu.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 noinline 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 = -1, d; + SPUCHAN *s_chan = &spu.s_chan[ch]; + int ret = ns_to, ns, d; + + s_chan->spos += s_chan->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++) { - d = skip_block(ch); - if (d) - ret = ns_to; - s_chan[ch].spos -= 28*0x10000; + s_chan->spos += s_chan->sinc; + while (s_chan->spos >= 28*0x10000) + { + d = skip_block(ch); + if (d && ns < ret) + ret = ns; + s_chan->spos -= 28*0x10000; + } } + s_chan->iSBPos = s_chan->spos >> 16; + s_chan->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 = (spu.spuCtrl >> 10) & 0x0f; + level = (ctrl >> 10) & 0x0f; level = 0x8000 >> level; - for (; ns < ns_to; ns++) + for (ns = 0; ns < ns_to; ns++) { - spu.dwNoiseCount += 2; - if (spu.dwNoiseCount >= level) + counter += 2; + if (counter >= level) { - spu.dwNoiseCount -= level; - shift = (spu.dwNoiseVal >> 10) & 0x1f; + counter -= level; + shift = (val >> 10) & 0x1f; bit = (0x69696969 >> shift) & 1; - if (spu.dwNoiseVal & 0x8000) - bit ^= 1; - spu.dwNoiseVal = (spu.dwNoiseVal << 1) | bit; + bit ^= (val >> 15) & 1; + val = (val << 1) | bit; } - ChanBuf[ns] = (signed short)spu.dwNoiseVal; + ChanBuf[ns] = (signed short)val; } + *dwNoiseCount = counter; + *dwNoiseVal = val; +} + +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; } @@ -672,11 +766,12 @@ static void mix_chan_rvb(int start, int count, int lv, int rv, int *rvb) // 0x0800-0x0bff Voice 1 // 0x0c00-0x0fff Voice 3 -static noinline void do_decode_bufs(int which, int start, int count) +static noinline void do_decode_bufs(unsigned short *mem, int which, + int count, int decode_pos) { - const int *src = ChanBuf + start; - unsigned short *dst = &spu.spuMem[0x800/2 + which*0x400/2]; - int cursor = spu.decode_pos + start; + unsigned short *dst = &mem[0x800/2 + which*0x400/2]; + const int *src = ChanBuf; + int cursor = decode_pos; while (count-- > 0) { @@ -688,18 +783,248 @@ static noinline void do_decode_bufs(int which, int start, int count) // decode_pos is updated and irqs are checked later, after voice loop } +static void do_silent_chans(int ns_to, int silentch) +{ + unsigned int mask; + SPUCHAN *s_chan; + int ch; + + mask = silentch & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) + { + if (!(mask & 1)) continue; + if (spu.dwChannelDead & (1<pCurr > spu.pSpuIrq && s_chan->pLoop > spu.pSpuIrq) + continue; + + 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) + { + unsigned char *start = s_chan->pCurr; + + skip_block(ch); + if (start == s_chan->pCurr || start - spu.spuMemC < 0x1000) + { + // looping on self or stopped(?) + spu.dwChannelDead |= 1<spos = 0; + break; + } + + s_chan->spos -= 28 * 0x10000; + } + } +} + +static void do_channels(int ns_to) +{ + unsigned int mask; + SPUCHAN *s_chan; + int *SB, sinc; + int ch, d; + + InitREVERB(ns_to); + + mask = spu.dwChannelOn & 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]; + SB = spu.SB + ch * SB_SIZE; + sinc = s_chan->sinc; + + if (s_chan->bNoise) + d = do_samples_noise(ch, ns_to); + else if (s_chan->bFMod == 2 + || (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 0)) + d = do_samples_noint(decode_block, ch, ns_to, + SB, sinc, &s_chan->spos, &s_chan->iSBPos); + else if (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 1) + d = do_samples_simple(decode_block, ch, ns_to, + SB, sinc, &s_chan->spos, &s_chan->iSBPos); + else + d = do_samples_default(decode_block, ch, ns_to, + SB, sinc, &s_chan->spos, &s_chan->iSBPos); + + d = MixADSR(&s_chan->ADSRX, d); + if (d < ns_to) { + spu.dwChannelOn &= ~(1 << ch); + s_chan->ADSRX.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, spu.decode_pos); + spu.decode_dirty_ch |= 1 << ch; + } + + if (s_chan->bFMod == 2) // fmod freq channel + memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0])); + if (s_chan->bRVBActive) + mix_chan_rvb(0, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume, spu.sRVBStart); + else + mix_chan(0, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume); + } +} + +static void do_samples_finish(int ns_to, int silentch, int decode_pos); + +// optional worker thread handling + +#if defined(THREAD_ENABLED) || defined(WANT_THREAD_CODE) + +static void thread_work_start(void); +static void thread_work_wait_sync(void); +static void thread_sync_caches(void); + +static void queue_channel_work(int ns_to, int silentch) +{ + const SPUCHAN *s_chan; + unsigned int mask; + int ch; + + worker->ns_to = ns_to; + worker->ctrl = spu.spuCtrl; + worker->decode_pos = spu.decode_pos; + worker->silentch = silentch; + + mask = worker->chmask = spu.dwChannelOn & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) + { + if (!(mask & 1)) continue; + + s_chan = &spu.s_chan[ch]; + worker->ch[ch].spos = s_chan->spos; + worker->ch[ch].sbpos = s_chan->iSBPos; + worker->ch[ch].sinc = s_chan->sinc; + worker->ch[ch].adsr = s_chan->ADSRX; + worker->ch[ch].start = s_chan->pCurr - spu.spuMemC; + worker->ch[ch].loop = s_chan->pLoop - spu.spuMemC; + if (s_chan->prevflags & 1) + worker->ch[ch].start = worker->ch[ch].loop; + + worker->ch[ch].ns_to = do_samples_skip(ch, ns_to); + } + + worker->pending = 1; + thread_work_start(); +} + +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; + SPUCHAN *s_chan; + + ns_to = worker->ns_to; + memset(spu.sRVBStart, 0, ns_to * sizeof(spu.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; + + s_chan = &spu.s_chan[ch]; + SB = spu.SB + ch * SB_SIZE; + + if (s_chan->bNoise) + do_lsfr_samples(d, worker->ctrl, &spu.dwNoiseCount, &spu.dwNoiseVal); + else if (s_chan->bFMod == 2 + || (s_chan->bFMod == 0 && spu_config.iUseInterpolation == 0)) + do_samples_noint(decode_block_work, ch, d, SB, sinc, &spos, &sbpos); + else if (s_chan->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])); + } + worker->r.ch[ch].adsrState = worker->ch[ch].adsr.State; + worker->r.ch[ch].adsrEnvelopeVol = worker->ch[ch].adsr.EnvelopeVol; + + if (ch == 1 || ch == 3) + { + do_decode_bufs(spu.spuMem, ch/2, ns_to, worker->decode_pos); + decode_dirty_ch |= 1 << ch; + } + + if (s_chan->bFMod == 2) // fmod freq channel + memcpy(iFMod, &ChanBuf, ns_to * sizeof(iFMod[0])); + if (s_chan->bRVBActive) + mix_chan_rvb(0, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume, spu.sRVBStart); + else + mix_chan(0, ns_to, s_chan->iLeftVolume, s_chan->iRightVolume); + } + + worker->r.chan_end = endmask; + worker->r.decode_dirty = decode_dirty_ch; +} + +static void sync_worker_thread(int do_direct) +{ + unsigned int mask; + int ch; + + if (do_direct) + thread_sync_caches(); + if (!worker->pending) + return; + + thread_work_wait_sync(); + 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 (spu.s_chan[ch].ADSRX.State != ADSR_RELEASE) + spu.s_chan[ch].ADSRX.State = worker->r.ch[ch].adsrState; + spu.s_chan[ch].ADSRX.EnvelopeVol = worker->r.ch[ch].adsrEnvelopeVol; + } + + 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(int do_direct) {} + +#endif // THREAD_ENABLED + //////////////////////////////////////////////////////////////////////// // MAIN SPU FUNCTION // here is the main job handler... -// basically the whole sound processing is done in this fat func! //////////////////////////////////////////////////////////////////////// -void do_samples(unsigned int cycles_to) +void do_samples(unsigned int cycles_to, int do_direct) { - const int ns_from = 0; - int ns,ns_to,ns_len; - int volmult = spu_config.iVolume; - int ch,d,silentch; + unsigned int mask; + int ch, ns_to; + int silentch; int cycle_diff; cycle_diff = cycles_to - spu.cycles_played; @@ -710,6 +1035,10 @@ void do_samples(unsigned int cycles_to) return; } + do_direct |= (cycle_diff < 64 * 768); + if (worker != NULL) + sync_worker_thread(do_direct); + if (cycle_diff < 2 * 768) return; @@ -749,75 +1078,42 @@ void do_samples(unsigned int cycles_to) } } - InitREVERB(ns_to); + mask = spu.dwNewChannel & 0xffffff; + for (ch = 0; mask != 0; ch++, mask >>= 1) { + if (mask & 1) + StartSound(ch); + } - { - silentch=~(spu.dwChannelOn|spu.dwNewChannel); + silentch = ~spu.dwChannelOn & 0xffffff; - //--------------------------------------------------// - //- main channel loop -// - //--------------------------------------------------// - { - for(ch=0;ch spu.pSpuIrq && s_chan[ch].pLoop > spu.pSpuIrq) - continue; - - s_chan[ch].spos += s_chan[ch].sinc * (ns_to - ns_from); - while(s_chan[ch].spos >= 28 * 0x10000) - { - unsigned char *start = s_chan[ch].pCurr; - - skip_block(ch); - if(start == s_chan[ch].pCurr || start - spu.spuMemC < 0x1000) - { - // looping on self or stopped(?) - spu.dwChannelDead |= 1< IRQ_NEAR_BLOCKS * 16 - && (unsigned long)(spu.pSpuIrq - s_chan[ch].pLoop) > IRQ_NEAR_BLOCKS * 16) + 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; scan_for_irq(ch, &upd_samples); @@ -910,7 +1201,7 @@ 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(); @@ -963,8 +1254,8 @@ int CALLBACK SPUplayCDDAchannel(short *pcm, int nbytes) // to be called after state load void ClearWorkingState(void) { - memset(SSumLR,0,sizeof(SSumLR)); // init some mixing buffers - memset(iFMod,0,sizeof(iFMod)); + memset(SSumLR, 0, NSSIZE * 2 * 4); // init some mixing buffers + memset(iFMod, 0, sizeof(iFMod)); spu.pS=(short *)spu.pSpuBuffer; // setup soundbuffer pointer } @@ -974,8 +1265,8 @@ void SetupStreams(void) int i; 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); + spu.sRVBStart = calloc(NSSIZE * 2, sizeof(spu.sRVBStart[0])); + SSumLR = calloc(NSSIZE * 2, sizeof(SSumLR[0])); spu.XAStart = // alloc xa buffer (uint32_t *)malloc(44100 * sizeof(uint32_t)); @@ -991,10 +1282,10 @@ void SetupStreams(void) for(i=0;i init sustain - s_chan[i].ADSRX.SustainIncrease = 1; - s_chan[i].pLoop=spu.spuMemC; - s_chan[i].pCurr=spu.spuMemC; + 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; } ClearWorkingState(); @@ -1009,24 +1300,131 @@ void RemoveStreams(void) spu.pSpuBuffer = NULL; free(spu.sRVBStart); // free reverb buffer spu.sRVBStart = NULL; + free(SSumLR); + SSumLR = NULL; free(spu.XAStart); // free XA buffer spu.XAStart = NULL; free(spu.CDDAStart); // free CDDA buffer spu.CDDAStart = NULL; } -// INIT/EXIT STUFF +#if defined(C64X_DSP) + +/* special code for TI C64x DSP */ +#include "spu_c64x.c" + +#elif defined(THREAD_ENABLED) + +#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) +{ + sem_post(&t.sem_avail); +} + +static void thread_work_wait_sync(void) +{ + sem_wait(&t.sem_done); +} + +static void thread_sync_caches(void) +{ +} + +static void *spu_worker_thread(void *unused) +{ + while (1) { + sem_wait(&t.sem_avail); + if (worker->exit_thread) + break; + + do_channel_work(); + + sem_post(&t.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(&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; + + return; + +fail_thread: + sem_destroy(&t.sem_done); +fail_sem_done: + sem_destroy(&t.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(&t.sem_avail); + pthread_join(t.thread, NULL); + sem_destroy(&t.sem_done); + sem_destroy(&t.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) { - spu.spuMemC = (unsigned char *)spu.spuMem; // just small setup + spu.spuMemC = calloc(1, 512 * 1024); memset((void *)&rvb, 0, sizeof(REVERBInfo)); InitADSR(); - spu.spuAddr = 0xffffffff; + spu.s_chan = calloc(MAXCHAN+1, sizeof(spu.s_chan[0])); // channel + 1 infos (1 is security for fmod handling) + spu.SB = calloc(MAXCHAN, sizeof(spu.SB[0]) * SB_SIZE); + + spu.spuAddr = 0; spu.decode_pos = 0; - memset((void *)s_chan, 0, sizeof(s_chan)); spu.pSpuIrq = spu.spuMemC; SetupStreams(); // prepare streaming @@ -1034,6 +1432,8 @@ long CALLBACK SPUinit(void) if (spu_config.iVolume == 0) spu_config.iVolume = 768; // 1024 is 1.0 + init_spu_thread(); + return 0; } @@ -1065,6 +1465,16 @@ long CALLBACK SPUclose(void) long CALLBACK SPUshutdown(void) { SPUclose(); + + exit_spu_thread(); + + free(spu.spuMemC); + spu.spuMemC = NULL; + free(spu.SB); + spu.SB = NULL; + free(spu.s_chan); + spu.s_chan = NULL; + RemoveStreams(); // no more streaming spu.bSpuInit=0; @@ -1144,15 +1554,18 @@ void spu_get_debug_info(int *chans_out, int *run_chans, int *fmod_chans_out, int { int ch = 0, fmod_chans = 0, noise_chans = 0, irq_chans = 0; + if (spu.s_chan == NULL) + return; + for(;ch