ShuriZma
/
desmume
mirror of https://github.com/TASEmulators/desmume.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

spu: fix the ancient bug in looping adpcm sounds which slew performance. this is what made the adpcm caching option necessary as a workaround. so--get rid of adpcm caching.

This commit is contained in:
zeromus 2009-10-12 07:41:16 +00:00
parent 715b2a0c4c
commit 451366ab66
5 changed files with 38 additions and 226 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
desmume/src/NDSSystem.h
View File

@ -429,7 +429,6 @@ extern struct TCommonSettings {
, DebugConsole(false) , DebugConsole(false)
, single_core(true) , single_core(true)
, spuInterpolationMode(SPUInterpolation_Linear) , spuInterpolationMode(SPUInterpolation_Linear)
, spuAdpcmCache(false)
, gfx3d_flushMode(0) , gfx3d_flushMode(0)
, manualBackupType(0) , manualBackupType(0)
{ {
@ -464,7 +463,6 @@ extern struct TCommonSettings {
} wifi; } wifi;
SPUInterpolationMode spuInterpolationMode; SPUInterpolationMode spuInterpolationMode;
bool spuAdpcmCache;
//this is a temporary hack until we straighten out the flushing logic and/or gxfifo //this is a temporary hack until we straighten out the flushing logic and/or gxfifo
int gfx3d_flushMode; int gfx3d_flushMode;

242
desmume/src/SPU.cpp
View File

@ -32,6 +32,8 @@
#define M_PI 3.1415926535897932386 #define M_PI 3.1415926535897932386
#endif #endif
#define K_ADPCM_LOOPING_RECOVERY_INDEX 99999
#include "debug.h" #include "debug.h"
#include "MMU.h" #include "MMU.h"
#include "SPU.h" #include "SPU.h"
@ -106,157 +108,6 @@ static FORCEINLINE T MinMax(T val, T min, T max)
////////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////////
class ADPCMCacheItem
{
public:
ADPCMCacheItem()
: raw_copy(NULL)
, decoded(NULL)
, next(NULL)
, prev(NULL)
, lockCount(0)
{}
~ADPCMCacheItem() {
delete[] raw_copy;
delete[] decoded;
}
void unlock() {
lockCount--;
}
void lock() {
lockCount++;
}
u32 addr;
s8* raw_copy; //for memcmp
u32 raw_len;
u32 decode_len;
s16* decoded; //s16 decoded samples
ADPCMCacheItem *next, *prev; //double linked list
int lockCount;
};
//notes on the cache:
//I am really unhappy with the ref counting. this needs to be automatic.
//We could do something better than a linear search through cache items, but it may not be worth it.
//Also we may need to rescan more often (every time a sample loops)
class ADPCMCache
{
public:
ADPCMCache()
: list_front(NULL)
, list_back(NULL)
, cache_size(0)
{}
ADPCMCacheItem *list_front, *list_back;
//this ought to be enough for anyone
static const u32 kMaxCacheSize = 8*1024*1024;
//this is not really precise, it is off by a constant factor
u32 cache_size;
void evict(const u32 target = kMaxCacheSize) {
//evicts old cache items until it is less than the max cache size
//this means we actually can exceed the cache by the size of the next item.
//if we really wanted to hold ourselves to it, we could evict to kMaxCacheSize-nextItemSize
while(cache_size > target)
{
ADPCMCacheItem *oldest = list_back;
while(oldest && oldest->lockCount>0) oldest = oldest->prev; //find an unlocked one
if(!oldest)
{
//nothing we can do, everything in the cache is locked. maybe we're leaking.
//just quit trying to evict
return;
}
list_remove(oldest);
cache_size -= oldest->raw_len*2;
//printf("evicting! totalsize:%d\n",cache_size);
delete oldest;
}
}
//DO NOT USE THIS METHOD WITHOUT MAKING SURE YOU HAVE
//FREED THE CURRENT ADPCMCacheItem FIRST!
//we should do this with some kind of smart pointers, but i am too lazy
ADPCMCacheItem* scan(channel_struct *chan)
{
u32 addr = chan->addr;
s8* raw = chan->buf8;
u32 raw_len = chan->totlength * 4;
for(ADPCMCacheItem* curr = list_front;curr;curr=curr->next)
{
if(curr->addr != addr) continue;
if(curr->raw_len != raw_len) continue;
if(memcmp(curr->raw_copy,raw,raw_len))
{
//we found a cached item for the current address, but the data is stale.
//for a variety of complicated reasons, we need to throw it out right this instant.
list_remove(curr);
delete curr;
break;
}
curr->lock();
list_remove(curr);
list_push_front(curr);
return curr;
}
//item was not found. recruit an existing one (the oldest), or create a new one
evict(); //reduce the size of the cache if necessary
ADPCMCacheItem* newitem = new ADPCMCacheItem();
newitem->lock();
newitem->addr = addr;
newitem->raw_len = raw_len;
newitem->raw_copy = new s8[raw_len];
memcpy(newitem->raw_copy,chan->buf8,raw_len);
u32 decode_len = newitem->decode_len = raw_len*2;
cache_size += newitem->decode_len;
newitem->decoded = new s16[decode_len];
int index = chan->buf8[2] & 0x7F;
s16 pcm16b = (s16)((chan->buf8[1] << 8) | chan->buf8[0]);
s16 pcm16b_last = pcm16b;
for(u32 i = 8; i < decode_len; i++)
{
const u32 shift = (i&1)<<2;
const u32 data4bit = (((u32)chan->buf8[i >> 1]) >> shift);
const s32 diff = precalcdifftbl[index][data4bit & 0xF];
index = precalcindextbl[index][data4bit & 0x7];
pcm16b_last = pcm16b;
pcm16b = MinMax(pcm16b+diff, -0x8000, 0x7FFF);
newitem->decoded[i] = pcm16b;
}
//printf("new cacheitem! totalsize:%d\n",cache_size);
list_push_front(newitem);
return newitem;
}
void list_remove(ADPCMCacheItem* item) {
if(item->next) item->next->prev = item->prev;
if(item->prev) item->prev->next = item->next;
if(item == list_front) list_front = item->next;
if(item == list_back) list_back = item->prev;
}
void list_push_front(ADPCMCacheItem* item)
{
item->next = list_front;
if(list_front) list_front->prev = item;
else list_back = item;
item->prev = NULL;
list_front = item;
}
} adpcmCache;
//////////////////////////////////////////////////////////////////////////////
int SPU_ChangeSoundCore(int coreid, int buffersize) int SPU_ChangeSoundCore(int coreid, int buffersize)
{ {
int i; int i;
@ -380,9 +231,6 @@ void SPU_Reset(void)
//todo - check success? //todo - check success?
} }
//keep the cache tidy
adpcmCache.evict(0);
// Reset Registers // Reset Registers
for (i = 0x400; i < 0x51D; i++) for (i = 0x400; i < 0x51D; i++)
T1WriteByte(MMU.ARM7_REG, i, 0); T1WriteByte(MMU.ARM7_REG, i, 0);
@ -395,12 +243,6 @@ void SPU_struct::reset()
memset(sndbuf,0,bufsize*2*4); memset(sndbuf,0,bufsize*2*4);
memset(outbuf,0,bufsize*2*2); memset(outbuf,0,bufsize*2*2);
for(int i = 0; i < 16; i++)
{
if(channels[i].cacheItem)
channels[i].cacheItem->unlock();
}
memset((void *)channels, 0, sizeof(channel_struct) * 16); memset((void *)channels, 0, sizeof(channel_struct) * 16);
for(int i = 0; i < 16; i++) for(int i = 0; i < 16; i++)
@ -425,11 +267,6 @@ SPU_struct::~SPU_struct()
{ {
if(sndbuf) delete[] sndbuf; if(sndbuf) delete[] sndbuf;
if(outbuf) delete[] outbuf; if(outbuf) delete[] outbuf;
for(int i = 0; i < 16; i++)
{
if(channels[i].cacheItem)
channels[i].cacheItem->unlock();
}
} }
void SPU_DeInit(void) void SPU_DeInit(void)
@ -484,13 +321,9 @@ void SPU_struct::KeyOn(int channel)
thischan.index = thischan.buf8[2] & 0x7F; thischan.index = thischan.buf8[2] & 0x7F;
thischan.lastsampcnt = 7; thischan.lastsampcnt = 7;
thischan.sampcnt = 8; thischan.sampcnt = 8;
thischan.loop_index = K_ADPCM_LOOPING_RECOVERY_INDEX;
// thischan.loopstart = thischan.loopstart << 3; // thischan.loopstart = thischan.loopstart << 3;
// thischan.length = (thischan.length << 3) + thischan.loopstart; // thischan.length = (thischan.length << 3) + thischan.loopstart;
if(thischan.cacheItem) thischan.cacheItem->unlock();
thischan.cacheItem = NULL;
if(CommonSettings.spuAdpcmCache)
if(this != SPU_core)
thischan.cacheItem = adpcmCache.scan(&thischan);
break; break;
} }
case 3: // PSG case 3: // PSG
@ -701,10 +534,10 @@ template<SPUInterpolationMode INTERPOLATE_MODE> static FORCEINLINE void Fetch8Bi
*data = (s32)chan->buf8[loc] << 8; *data = (s32)chan->buf8[loc] << 8;
} }
template<SPUInterpolationMode INTERPOLATE_MODE, bool ADPCM_CACHED> static FORCEINLINE void Fetch16BitData(const channel_struct * const chan, s32 *data) template<SPUInterpolationMode INTERPOLATE_MODE> static FORCEINLINE void Fetch16BitData(const channel_struct * const chan, s32 *data)
{ {
const s16* const buf16 = ADPCM_CACHED ? chan->cacheItem->decoded : chan->buf16; const s16* const buf16 = chan->buf16;
const int shift = ADPCM_CACHED ? 3 : 1; const int shift = 1;
if(INTERPOLATE_MODE != SPUInterpolation_None) if(INTERPOLATE_MODE != SPUInterpolation_None)
{ {
u32 loc = sputrunc(chan->sampcnt); u32 loc = sputrunc(chan->sampcnt);
@ -720,13 +553,8 @@ template<SPUInterpolationMode INTERPOLATE_MODE, bool ADPCM_CACHED> static FORCEI
*data = (s32)buf16[sputrunc(chan->sampcnt)]; *data = (s32)buf16[sputrunc(chan->sampcnt)];
} }
template<SPUInterpolationMode INTERPOLATE_MODE, bool ADPCM_CACHED> static FORCEINLINE void FetchADPCMData(channel_struct * const chan, s32 * const data) template<SPUInterpolationMode INTERPOLATE_MODE> static FORCEINLINE void FetchADPCMData(channel_struct * const chan, s32 * const data)
{ {
if(ADPCM_CACHED)
{
return Fetch16BitData<INTERPOLATE_MODE,true>(chan, data);
}
// No sense decoding, just return the last sample // No sense decoding, just return the last sample
if (chan->lastsampcnt != sputrunc(chan->sampcnt)){ if (chan->lastsampcnt != sputrunc(chan->sampcnt)){
@ -741,6 +569,12 @@ template<SPUInterpolationMode INTERPOLATE_MODE, bool ADPCM_CACHED> static FORCEI
chan->pcm16b_last = chan->pcm16b; chan->pcm16b_last = chan->pcm16b;
chan->pcm16b = MinMax(chan->pcm16b+diff, -0x8000, 0x7FFF); chan->pcm16b = MinMax(chan->pcm16b+diff, -0x8000, 0x7FFF);
if(i == (chan->loopstart<<3)) {
if(chan->loop_index != K_ADPCM_LOOPING_RECOVERY_INDEX) printf("over-snagging\n");
chan->loop_pcm16b = chan->pcm16b;
chan->loop_index = chan->index;
}
} }
chan->lastsampcnt = sputrunc(chan->sampcnt); chan->lastsampcnt = sputrunc(chan->sampcnt);
@ -851,12 +685,21 @@ static FORCEINLINE void TestForLoop2(SPU_struct *SPU, channel_struct *chan)
{ {
while (chan->sampcnt > chan->double_totlength_shifted) while (chan->sampcnt > chan->double_totlength_shifted)
chan->sampcnt -= chan->double_totlength_shifted - (double)(chan->loopstart << 3); chan->sampcnt -= chan->double_totlength_shifted - (double)(chan->loopstart << 3);
//chan->sampcnt = (double)(chan->loopstart << 3);
if(chan->loop_index == K_ADPCM_LOOPING_RECOVERY_INDEX)
{
chan->pcm16b = (s16)((chan->buf8[1] << 8) | chan->buf8[0]); chan->pcm16b = (s16)((chan->buf8[1] << 8) | chan->buf8[0]);
chan->index = chan->buf8[2] & 0x7F; chan->index = chan->buf8[2] & 0x7F;
chan->lastsampcnt = 7; chan->lastsampcnt = 7;
} }
else else
{
chan->pcm16b = chan->loop_pcm16b;
chan->index = chan->loop_index;
chan->lastsampcnt = (chan->loopstart << 3);
}
}
else
{ {
chan->status = CHANSTAT_STOPPED; chan->status = CHANSTAT_STOPPED;
if(SPU == SPU_core) if(SPU == SPU_core)
@ -876,8 +719,8 @@ template<int CHANNELS> FORCEINLINE static void SPU_Mix(SPU_struct* SPU, channel_
} }
} }
template<int FORMAT, SPUInterpolationMode INTERPOLATE_MODE, int CHANNELS, bool CACHED> template<int FORMAT, SPUInterpolationMode INTERPOLATE_MODE, int CHANNELS>
FORCEINLINE static void _____SPU_ChanUpdate(SPU_struct* const SPU, channel_struct* const chan) FORCEINLINE static void ____SPU_ChanUpdate(SPU_struct* const SPU, channel_struct* const chan)
{ {
for (; SPU->bufpos < SPU->buflength; SPU->bufpos++) for (; SPU->bufpos < SPU->buflength; SPU->bufpos++)
{ {
@ -887,8 +730,8 @@ template<int FORMAT, SPUInterpolationMode INTERPOLATE_MODE, int CHANNELS, bool C
switch(FORMAT) switch(FORMAT)
{ {
case 0: Fetch8BitData<INTERPOLATE_MODE>(chan, &data); break; case 0: Fetch8BitData<INTERPOLATE_MODE>(chan, &data); break;
case 1: Fetch16BitData<INTERPOLATE_MODE,false>(chan, &data); break; case 1: Fetch16BitData<INTERPOLATE_MODE>(chan, &data); break;
case 2: FetchADPCMData<INTERPOLATE_MODE,CACHED>(chan, &data); break; case 2: FetchADPCMData<INTERPOLATE_MODE>(chan, &data); break;
case 3: FetchPSGData(chan, &data); break; case 3: FetchPSGData(chan, &data); break;
} }
SPU_Mix<CHANNELS>(SPU, chan, data); SPU_Mix<CHANNELS>(SPU, chan, data);
@ -902,14 +745,6 @@ template<int FORMAT, SPUInterpolationMode INTERPOLATE_MODE, int CHANNELS, bool C
} }
} }
template<int FORMAT, SPUInterpolationMode INTERPOLATE_MODE, int CHANNELS>
FORCEINLINE static void ____SPU_ChanUpdate(SPU_struct* const SPU, channel_struct* const chan)
{
if(FORMAT == 2 && chan->cacheItem)
_____SPU_ChanUpdate<FORMAT,INTERPOLATE_MODE,CHANNELS,true>(SPU,chan);
else _____SPU_ChanUpdate<FORMAT,INTERPOLATE_MODE,CHANNELS,false>(SPU,chan);
}
template<int FORMAT, SPUInterpolationMode INTERPOLATE_MODE> template<int FORMAT, SPUInterpolationMode INTERPOLATE_MODE>
FORCEINLINE static void ___SPU_ChanUpdate(const bool actuallyMix, SPU_struct* const SPU, channel_struct* const chan) FORCEINLINE static void ___SPU_ChanUpdate(const bool actuallyMix, SPU_struct* const SPU, channel_struct* const chan)
{ {
@ -1238,7 +1073,7 @@ void WAV_WavSoundUpdate(void* soundData, int numSamples)
void spu_savestate(EMUFILE* os) void spu_savestate(EMUFILE* os)
{ {
//version //version
write32le(2,os); write32le(3,os);
SPU_struct *spu = SPU_core; SPU_struct *spu = SPU_core;
@ -1297,7 +1132,7 @@ bool spu_loadstate(EMUFILE* is, int size)
read32le(&chan.length,is); read32le(&chan.length,is);
chan.totlength = chan.length + chan.loopstart; chan.totlength = chan.length + chan.loopstart;
chan.double_totlength_shifted = (double)(chan.totlength << format_shift[chan.format]); chan.double_totlength_shifted = (double)(chan.totlength << format_shift[chan.format]);
if(version != 1) if(version >= 2)
{ {
read64le(&temp64,is); chan.sampcnt = u64_to_double(temp64); read64le(&temp64,is); chan.sampcnt = u64_to_double(temp64);
read64le(&temp64,is); chan.sampinc = u64_to_double(temp64); read64le(&temp64,is); chan.sampinc = u64_to_double(temp64);
@ -1314,31 +1149,20 @@ bool spu_loadstate(EMUFILE* is, int size)
read16le(&chan.x,is); read16le(&chan.x,is);
read16le(&chan.psgnoise_last,is); read16le(&chan.psgnoise_last,is);
//hopefully trigger a recovery of the adpcm looping system
chan.loop_index = K_ADPCM_LOOPING_RECOVERY_INDEX;
//fixup the pointers which we had are supposed to keep cached //fixup the pointers which we had are supposed to keep cached
chan.buf8 = (s8*)&MMU.MMU_MEM[1][(chan.addr>>20)&0xFF][(chan.addr & MMU.MMU_MASK[1][(chan.addr >> 20) & 0xFF])]; chan.buf8 = (s8*)&MMU.MMU_MEM[1][(chan.addr>>20)&0xFF][(chan.addr & MMU.MMU_MASK[1][(chan.addr >> 20) & 0xFF])];
} }
if(version==2) { if(version>=2) {
read64le(&temp64,is); samples = u64_to_double(temp64); read64le(&temp64,is); samples = u64_to_double(temp64);
} }
//copy the core spu (the more accurate) to the user spu //copy the core spu (the more accurate) to the user spu
if(SPU_user) { if(SPU_user) {
for(int i=0;i<16;i++)
{
channel_struct &chan = SPU_user->channels[i];
if(chan.cacheItem) chan.cacheItem->unlock();
}
memcpy(SPU_user->channels,SPU_core->channels,sizeof(SPU_core->channels)); memcpy(SPU_user->channels,SPU_core->channels,sizeof(SPU_core->channels));
if(CommonSettings.spuAdpcmCache)
for(int i=0;i<16;i++)
{
channel_struct &chan = SPU_user->channels[i];
if(chan.format == 2)
chan.cacheItem = adpcmCache.scan(&chan);
}
} }
return true; return true;

6
desmume/src/SPU.h
View File

@ -68,12 +68,9 @@ extern SoundInterface_struct SNDDummy;
extern SoundInterface_struct SNDFile; extern SoundInterface_struct SNDFile;
extern int SPU_currentCoreNum; extern int SPU_currentCoreNum;
class ADPCMCacheItem;
struct channel_struct struct channel_struct
{ {
channel_struct() channel_struct()
: cacheItem(NULL)
{} {}
u32 num; u32 num;
u8 vol; u8 vol;
@ -99,10 +96,11 @@ struct channel_struct
// ADPCM specific // ADPCM specific
u32 lastsampcnt; u32 lastsampcnt;
s16 pcm16b, pcm16b_last; s16 pcm16b, pcm16b_last;
s16 loop_pcm16b;
int index; int index;
int loop_index;
u16 x; u16 x;
s16 psgnoise_last; s16 psgnoise_last;
ADPCMCacheItem *cacheItem;
} ; } ;
class SPU_struct class SPU_struct

8
desmume/src/windows/main.cpp
View File

@ -2151,7 +2151,6 @@ int _main()
sndcoretype = GetPrivateProfileInt("Sound","SoundCore2", SNDCORE_DIRECTX, IniName); sndcoretype = GetPrivateProfileInt("Sound","SoundCore2", SNDCORE_DIRECTX, IniName);
sndbuffersize = GetPrivateProfileInt("Sound","SoundBufferSize", 735 * 4, IniName); sndbuffersize = GetPrivateProfileInt("Sound","SoundBufferSize", 735 * 4, IniName);
CommonSettings.spuInterpolationMode = (SPUInterpolationMode)GetPrivateProfileInt("Sound","SPUInterpolation", 1, IniName); CommonSettings.spuInterpolationMode = (SPUInterpolationMode)GetPrivateProfileInt("Sound","SPUInterpolation", 1, IniName);
CommonSettings.spuAdpcmCache = GetPrivateProfileInt("Sound","SPUAdpcmCache",0,IniName)!=0;
EnterCriticalSection(&win_execute_sync); EnterCriticalSection(&win_execute_sync);
int spu_ret = SPU_ChangeSoundCore(sndcoretype, sndbuffersize); int spu_ret = SPU_ChangeSoundCore(sndcoretype, sndbuffersize);
@ -4816,9 +4815,6 @@ LRESULT CALLBACK SoundSettingsDlgProc(HWND hDlg, UINT uMsg, WPARAM wParam, LPARA
SendDlgItemMessage(hDlg, IDC_SPU_INTERPOLATION_CB, CB_ADDSTRING, 0, (LPARAM)"Cosine (slowest, sounds best)"); SendDlgItemMessage(hDlg, IDC_SPU_INTERPOLATION_CB, CB_ADDSTRING, 0, (LPARAM)"Cosine (slowest, sounds best)");
SendDlgItemMessage(hDlg, IDC_SPU_INTERPOLATION_CB, CB_SETCURSEL, (int)CommonSettings.spuInterpolationMode, 0); SendDlgItemMessage(hDlg, IDC_SPU_INTERPOLATION_CB, CB_SETCURSEL, (int)CommonSettings.spuInterpolationMode, 0);
//setup cache setting
CheckDlgButton(hDlg, IDC_SPU_CACHE, CommonSettings.spuAdpcmCache?BST_CHECKED:BST_UNCHECKED );
// Setup Sound Buffer Size Edit Text // Setup Sound Buffer Size Edit Text
sprintf(tempstr, "%d", sndbuffersize); sprintf(tempstr, "%d", sndbuffersize);
SetDlgItemText(hDlg, IDC_SOUNDBUFFERET, tempstr); SetDlgItemText(hDlg, IDC_SOUNDBUFFERET, tempstr);
@ -4880,10 +4876,6 @@ LRESULT CALLBACK SoundSettingsDlgProc(HWND hDlg, UINT uMsg, WPARAM wParam, LPARA
CommonSettings.spuInterpolationMode = (SPUInterpolationMode)SendDlgItemMessage(hDlg, IDC_SPU_INTERPOLATION_CB, CB_GETCURSEL, 0, 0); CommonSettings.spuInterpolationMode = (SPUInterpolationMode)SendDlgItemMessage(hDlg, IDC_SPU_INTERPOLATION_CB, CB_GETCURSEL, 0, 0);
WritePrivateProfileInt("Sound","SPUInterpolation",(int)CommonSettings.spuInterpolationMode, IniName); WritePrivateProfileInt("Sound","SPUInterpolation",(int)CommonSettings.spuInterpolationMode, IniName);
//write cache setting
CommonSettings.spuAdpcmCache = IsDlgButtonChecked(hDlg, IDC_SPU_CACHE) != 0;
WritePrivateProfileInt("Sound","SPUAdpcmCache",CommonSettings.spuAdpcmCache?1:0, IniName);
return TRUE; return TRUE;
} }
case IDCANCEL: case IDCANCEL:

BIN
desmume/src/windows/resources.rc
View File

Binary file not shown.