/* PUBLIC */
Fpa_chunk flist_delete(Fpa_chunk f, Term x)
{
if (f == NULL)
fprintf(stderr, "WARNING: flist_delete, item %p not found (1)!\n", x);
else if (FLT(x,FLAST(f)))
f->next = flist_delete(f->next, x);
else {
int n = f->n;
int i = FMAX - n;
while (i < FMAX && FLT(x,f->d[i]))
i++;
if (x != f->d[i])
fprintf(stderr, "WARNING: flist_delete, item %p not found (2)!\n", x);
else {
/* delete and close the hole */
int j;
for (j = i; j > FMAX-n; j--)
f->d[j] = f->d[j-1];
f->d[j] = NULL;
f->n = n-1;
if (f->n == 0) {
/* delete this chunk */
Fpa_chunk next = f->next;
free_fpa_chunk(f);
f = next;
}
else {
/* try to join this chunk with the next */
f = consolidate(f);
}
}
}
return f;
} /* flist_delete */
// feed adaptation data from a batch file containing entries (rawFile alignmentFile)
void FMLLREstimator::feedAdaptationData(const char *strBatchFile, const char *strAlignmentFormat,
double *dLikelihood) {
BatchFile batchFile(strBatchFile,"features|alignment");
batchFile.load();
for(unsigned int i=0 ; i < batchFile.size() ; ++i) {
//for(int i=0 ; i < 5 ; ++i) {
// load the alignment
Alignment *alignment = NULL;
if (strcmp(strAlignmentFormat,"text") == 0) {
AlignmentFile alignmentFile(m_phoneSet);
VPhoneAlignment *vPhoneAlignment = alignmentFile.load(batchFile.getField(i,"alignment"));
assert(vPhoneAlignment);
alignment = AlignmentFile::toAlignment(m_phoneSet,m_hmmManager,vPhoneAlignment);
AlignmentFile::destroyPhoneAlignment(vPhoneAlignment);
} else {
alignment = Alignment::load(batchFile.getField(i,"alignment"),NULL);
assert(alignment);
}
// load the feature vectors
FeatureFile featureFile(batchFile.getField(i,"features"),MODE_READ);
featureFile.load();
Matrix<float> *mFeatures = featureFile.getFeatureVectors();
// load and apply the transform
/*
Transform *transform = new Transform();
transform->load("/data/daniel/tasks/wsj/experiments/may16th_2013_CMNUtterance/5/fmllr1/transforms/440m.fmllr.bin");
Matrix<float> *mFeaturesX = transform->apply(*mFeatures);
mFeatures = mFeaturesX;
delete transform;
*/
// check consistency
if (mFeatures->getRows() != alignment->getFrames()) {
BVC_ERROR << "inconsistent number of feature vectors / alignment file";
}
// accumulate adaptation data
double dLikelihoodAlignment = 0.0;
feedAdaptationData(*mFeatures,alignment,&dLikelihoodAlignment);
BVC_VERB << "loaded file: " << batchFile.getField(i,"alignment") << " likelihood: " << FLT(10,2)
<< dLikelihoodAlignment << " (" << mFeatures->getRows() << "frames)";
*dLikelihood += dLikelihoodAlignment;
// clean-up
delete alignment;
delete mFeatures;
}
double dLikelihoodFrame = (*dLikelihood)/m_fOccupancyTotal;
BVC_VERB << "total likelihood: " << FLT(20,6) << *dLikelihood << " (likelihood per frame: "
<< FLT(8,4) << dLikelihoodFrame << ")";
}
static void
dump_immediate_verbose(
struct tgsi_full_immediate *imm,
unsigned ignored )
{
unsigned i;
TXT( "\nDataType : " );
ENM( imm->Immediate.DataType, TGSI_IMMS );
if( ignored ) {
TXT( "\nPadding : " );
UIX( imm->Immediate.Padding );
}
for( i = 0; i < imm->Immediate.Size - 1; i++ ) {
EOL();
switch( imm->Immediate.DataType ) {
case TGSI_IMM_FLOAT32:
TXT( "\nFloat: " );
FLT( imm->u.ImmediateFloat32[i].Float );
break;
default:
assert( 0 );
}
}
}
static void
dump_imm_data(struct tgsi_iterate_context *iter,
union tgsi_immediate_data *data,
unsigned num_tokens,
unsigned data_type)
{
struct dump_ctx *ctx = (struct dump_ctx *)iter;
unsigned i ;
TXT( " {" );
assert( num_tokens <= 4 );
for (i = 0; i < num_tokens; i++) {
switch (data_type) {
case TGSI_IMM_FLOAT32:
FLT( data[i].Float );
break;
case TGSI_IMM_UINT32:
UID(data[i].Uint);
break;
case TGSI_IMM_INT32:
SID(data[i].Int);
break;
default:
assert( 0 );
}
if (i < num_tokens - 1)
TXT( ", " );
}
TXT( "}" );
}
// print the features (debugging)
void FeatureFile::print(MatrixBase<float> &mFeatures, unsigned int iDelta) {
assert((mFeatures.getCols()%iDelta) == 0);
cout << endl;
for(unsigned int i=0 ; i < mFeatures.getRows() ; ++i) {
for(unsigned int j=0 ; j < iDelta ; ++j) {
for(unsigned int h=0 ; h < mFeatures.getCols()/iDelta ; ++h) {
cout << " " << FLT(9,6) << mFeatures(i,h+j*(mFeatures.getCols()/iDelta));
}
cout << endl;
}
cout << endl;
}
}
void ENTRYPOINT three_d_box( three_d_handle handle,
float x1, float x2, float y1, float y2, float z1, float z2,
unsigned red, unsigned green, unsigned blue )
{
if( hThreedDLL == NULL ) return;
// Use a trick to pass floats rather than as doubles
// (_Call16 is a varargs function, which makes all float
// parameters passed as doubles)
#define FLT(x) (*(long *) &(x))
_Call16( three_d_box_Proc, "wddddddwww", handle,
FLT(x1), FLT(x2), FLT(y1), FLT(y2), FLT(z1), FLT(z2),
red, green, blue );
}
static void
dump_imm_data(struct tgsi_iterate_context *iter,
union tgsi_immediate_data *data,
unsigned num_tokens,
unsigned data_type)
{
struct dump_ctx *ctx = (struct dump_ctx *)iter;
unsigned i ;
TXT( " {" );
assert( num_tokens <= 4 );
for (i = 0; i < num_tokens; i++) {
switch (data_type) {
case TGSI_IMM_FLOAT64: {
union di d;
d.ui = data[i].Uint | (uint64_t)data[i+1].Uint << 32;
DBL( d.d );
i++;
break;
}
case TGSI_IMM_FLOAT32:
if (ctx->dump_float_as_hex)
HFLT( data[i].Float );
else
FLT( data[i].Float );
break;
case TGSI_IMM_UINT32:
UID(data[i].Uint);
break;
case TGSI_IMM_INT32:
SID(data[i].Int);
break;
default:
assert( 0 );
}
if (i < num_tokens - 1)
TXT( ", " );
}
TXT( "}" );
}
/*! This function represents the core of the SPH density computation. The
* target particle may either be local, or reside in the communication
* buffer.
*/
int cs_update_weight_evaluate(int target, int mode, int *nexport, int *nsend_local)
{
int j, n;
int startnode, numngb, listindex = 0;
double h, h2, hinv, hinv3;
double wk;
double r, r2, u;
MyLongDouble weighted_numngb;
MyDouble *pos;
weighted_numngb = 0;
if(mode == 0)
{
pos = P[target].Pos;
h = PPP[target].Hsml;
}
else
{
pos = UpdateweightGet[target].Pos;
h = UpdateweightGet[target].Hsml;
}
h2 = h * h;
hinv = 1.0 / h;
#ifndef TWODIMS
hinv3 = hinv * hinv * hinv;
#else
hinv3 = hinv * hinv / boxSize_Z;
#endif
if(mode == 0)
{
startnode = All.MaxPart; /* root node */
}
else
{
startnode = UpdateweightGet[target].NodeList[0];
startnode = Nodes[startnode].u.d.nextnode; /* open it */
}
numngb = 0;
while(startnode >= 0)
{
while(startnode >= 0)
{
numngb = cs_ngb_treefind_variable_phases(pos, h, target, &startnode, mode, nexport, nsend_local);
if(numngb < 0)
return -1;
for(n = 0; n < numngb; n++)
{
j = Ngblist[n];
r2 = R2ngblist[n];
r = sqrt(r2);
u = r * hinv;
if(u < 0.5)
{
wk = hinv3 * (KERNEL_COEFF_1 + KERNEL_COEFF_2 * (u - 1) * u * u);
}
else
{
wk = hinv3 * KERNEL_COEFF_5 * (1.0 - u) * (1.0 - u) * (1.0 - u);
}
weighted_numngb += FLT(NORM_COEFF * wk / hinv3); /* 4.0/3 * PI = 4.188790204786 */
}
}
if(mode == 1)
{
listindex++;
if(listindex < NODELISTLENGTH)
{
startnode = UpdateweightGet[target].NodeList[listindex];
if(startnode >= 0)
startnode = Nodes[startnode].u.d.nextnode; /* open it */
}
}
}
if(mode == 0)
{
PPP[target].n.dNumNgb = weighted_numngb;
}
else
{
UpdateweightResult[target].Ngb = weighted_numngb;
}
return 0;
}
// Read in the config file for the whole application
int ConfigAppLoad()
{
TCHAR szConfig[MAX_PATH];
TCHAR szLine[1024];
FILE* h;
#ifdef _UNICODE
setlocale(LC_ALL, "");
#endif
CreateConfigName(szConfig);
if ((h = _tfopen(szConfig, _T("rt"))) == NULL) {
return 1;
}
// Go through each line of the config file
while (_fgetts(szLine, sizeof(szLine), h)) {
int nLen = _tcslen(szLine);
// Get rid of the linefeed at the end
if (szLine[nLen - 1] == 10) {
szLine[nLen - 1] = 0;
nLen--;
}
#define VAR(x) { TCHAR* szValue = LabelCheck(szLine,_T(#x)); \
if (szValue) x = _tcstol(szValue, NULL, 0); }
#define VAR64(x) { TCHAR* szValue = LabelCheck(szLine,_T(#x)); \
if (szValue) x = (long long)_tcstod(szValue, NULL); }
#define FLT(x) { TCHAR* szValue = LabelCheck(szLine,_T(#x)); \
if (szValue) x = _tcstod(szValue, NULL); }
#define STR(x) { TCHAR* szValue = LabelCheck(szLine,_T(#x) _T(" ")); \
if (szValue) _tcscpy(x,szValue); }
VAR(nIniVersion);
// Emulation
VAR(bBurnUseASMCPUEmulation);
// Video
VAR(nVidDepth); VAR(nVidRefresh);
VAR(nVidRotationAdjust);
// horizontal oriented
VAR(nVidHorWidth); VAR(nVidHorHeight);
VAR(bVidArcaderesHor);
VAR(VidPreset[0].nWidth); VAR(VidPreset[0].nHeight);
VAR(VidPreset[1].nWidth); VAR(VidPreset[1].nHeight);
VAR(VidPreset[2].nWidth); VAR(VidPreset[2].nHeight);
VAR(VidPreset[3].nWidth); VAR(VidPreset[3].nHeight);
VAR(nScreenSizeHor);
// vertical oriented
VAR(nVidVerWidth); VAR(nVidVerHeight);
VAR(bVidArcaderesVer);
VAR(VidPresetVer[0].nWidth); VAR(VidPresetVer[0].nHeight);
VAR(VidPresetVer[1].nWidth); VAR(VidPresetVer[1].nHeight);
VAR(VidPresetVer[2].nWidth); VAR(VidPresetVer[2].nHeight);
VAR(VidPresetVer[3].nWidth); VAR(VidPresetVer[3].nHeight);
VAR(nScreenSizeVer);
VAR(nWindowSize);
VAR(nWindowPosX); VAR(nWindowPosY);
VAR(bDoGamma);
VAR(bVidUseHardwareGamma);
VAR(bHardwareGammaOnly);
FLT(nGamma);
VAR(bVidFullStretch);
VAR(bVidCorrectAspect);
VAR(bVidAutoSwitchFull);
VAR(bVidTripleBuffer);
VAR(bVidVSync);
VAR(bVidScanlines);
VAR(nVidScanIntensity);
VAR(bMonitorAutoCheck);
VAR(nVidScrnAspectX);
VAR(nVidScrnAspectY);
VAR(bForce60Hz);
VAR(bAlwaysDrawFrames);
VAR(bVidUsePlaceholder);
STR(szPlaceHolder);
VAR(nVidSelect);
VAR(nVidBlitterOpt[0]);
VAR64(nVidBlitterOpt[1]);
VAR(nVidBlitterOpt[2]);
VAR(nVidBlitterOpt[3]);
// DirectDraw blitter
VAR(bVidScanHalf);
// Direct3D blitter
VAR(bVidBilinear);
VAR(bVidScanDelay);
VAR(bVidScanRotate);
VAR(bVidScanBilinear);
VAR(nVidFeedbackIntensity);
VAR(nVidFeedbackOverSaturation);
FLT(fVidScreenAngle);
FLT(fVidScreenCurvature);
VAR(bVidForce16bit);
VAR(nVidTransferMethod);
// DirectX Graphics blitter
FLT(dVidCubicB);
FLT(dVidCubicC);
// Sound
VAR(nAudSelect);
VAR(nAudSegCount);
VAR(nInterpolation);
VAR(nFMInterpolation);
VAR(nAudSampleRate[0]);
VAR(nAudDSPModule[0]);
VAR(nAudSampleRate[1]);
VAR(nAudDSPModule[1]);
// Other
STR(szLocalisationTemplate);
STR(szGamelistLocalisationTemplate);
VAR(nVidSDisplayStatus);
VAR(nMinChatFontSize);
VAR(nMaxChatFontSize);
VAR(bModelessMenu);
VAR(nSplashTime);
VAR(bDrvSaveAll);
VAR(nAppThreadPriority);
VAR(bAlwaysProcessKeyboardInput);
VAR(bAutoPause);
VAR(bSaveInputs);
VAR(nLoadMenuShowX);
VAR(nLoadMenuBoardTypeFilter);
VAR(nLoadMenuGenreFilter);
VAR(nLoadMenuFamilyFilter);
STR(szAppRomPaths[0]);
STR(szAppRomPaths[1]);
STR(szAppRomPaths[2]);
STR(szAppRomPaths[3]);
STR(szAppRomPaths[4]);
STR(szAppRomPaths[5]);
STR(szAppRomPaths[6]);
STR(szAppRomPaths[7]);
STR(szAppPreviewsPath);
STR(szAppTitlesPath);
STR(szAppSelectPath);
STR(szAppVersusPath);
STR(szAppHowtoPath);
STR(szAppScoresPath);
STR(szAppBossesPath);
STR(szAppGameoverPath);
STR(szAppFlyersPath);
STR(szAppMarqueesPath);
STR(szAppControlsPath);
STR(szAppCabinetsPath);
STR(szAppPCBsPath);
STR(szAppCheatsPath);
STR(szAppHistoryPath);
STR(szAppListsPath);
STR(szAppDatListsPath);
STR(szAppIpsPath);
STR(szAppIconsPath);
STR(szAppArchivesPath);
STR(szAppHiscorePath);
VAR(nMenuUITheme);
VAR(bNoChangeNumLock);
VAR(bSaveCRoms);
VAR(EnableHiscores);
VAR(nSelectedLanguage);
VAR(bEnableIcons);
VAR(nIconsSize);
STR(szPrevGames[0]);
STR(szPrevGames[1]);
STR(szPrevGames[2]);
STR(szPrevGames[3]);
STR(szPrevGames[4]);
STR(szPrevGames[5]);
STR(szPrevGames[6]);
STR(szPrevGames[7]);
STR(szPrevGames[8]);
STR(szPrevGames[9]);
// Default Controls
VAR(nPlayerDefaultControls[0]);
STR(szPlayerDefaultIni[0]);
VAR(nPlayerDefaultControls[1]);
STR(szPlayerDefaultIni[1]);
VAR(nPlayerDefaultControls[2]);
STR(szPlayerDefaultIni[2]);
VAR(nPlayerDefaultControls[3]);
STR(szPlayerDefaultIni[3]);
#undef STR
#undef FLT
#undef VAR
#undef VAR64
}
fclose(h);
return 0;
}
// Write out the config file for the whole application
int ConfigAppSave()
{
TCHAR szConfig[MAX_PATH];
FILE *h;
if (bCmdOptUsed) {
return 1;
}
#ifdef _UNICODE
setlocale(LC_ALL, "");
#endif
CreateConfigName(szConfig);
if ((h = _tfopen(szConfig, _T("wt"))) == NULL) {
return 1;
}
// Write title
_ftprintf(h, _T("// ") _T(APP_TITLE) _T(" v%s --- Main Config File\n\n"), szAppBurnVer);
_ftprintf(h, _T("// Don't edit this file manually unless you know what you're doing\n"));
_ftprintf(h, _T("// ") _T(APP_TITLE) _T(" will restore default settings when this file is deleted\n"));
#define VAR(x) _ftprintf(h, _T(#x) _T(" %d\n"), x)
#define VAR64(x) _ftprintf(h, _T(#x) _T(" %lf\n"), (float)x)
#define FLT(x) _ftprintf(h, _T(#x) _T(" %lf\n"), x)
#define STR(x) _ftprintf(h, _T(#x) _T(" %s\n"), x)
_ftprintf(h, _T("\n// The application version this file was saved from\n"));
// We can't use the macros for this!
_ftprintf(h, _T("nIniVersion 0x%06X"), nBurnVer);
_ftprintf(h, _T("\n\n\n"));
_ftprintf(h, _T("// --- emulation --------------------------------------------------------------\n"));
_ftprintf(h, _T("\n// If non-zero, use A68K for MC68000 emulation\n"));
VAR(bBurnUseASMCPUEmulation);
_ftprintf(h, _T("\n\n\n"));
_ftprintf(h, _T("// --- Video ------------------------------------------------------------------\n"));
// Horizontal oriented
_ftprintf(h, _T("\n// (Horizontal Oriented) The display mode to use for fullscreen\n"));
VAR(nVidHorWidth); VAR(nVidHorHeight);
_ftprintf(h, _T("\n// (Horizontal Oriented) If non-zero, use the same fullscreen resolution as the original arcade game\n"));
VAR(bVidArcaderesHor);
_ftprintf(h, _T("\n// (Horizontal Oriented) The preset resolutions appearing in the menu\n"));
VAR(VidPreset[0].nWidth); VAR(VidPreset[0].nHeight);
VAR(VidPreset[1].nWidth); VAR(VidPreset[1].nHeight);
VAR(VidPreset[2].nWidth); VAR(VidPreset[2].nHeight);
VAR(VidPreset[3].nWidth); VAR(VidPreset[3].nHeight);
_ftprintf(h, _T("\n// (Horizontal Oriented) Full-screen size (0 = use display mode variables)\n"));
VAR(nScreenSizeHor);
// Vertical oriented
_ftprintf(h, _T("\n// (Vertical Oriented) The display mode to use for fullscreen\n"));
VAR(nVidVerWidth); VAR(nVidVerHeight);
_ftprintf(h, _T("\n// (Vertical Oriented) If non-zero, use the same fullscreen resolution as the original arcade game\n"));
VAR(bVidArcaderesVer);
_ftprintf(h, _T("\n// (Vertical Oriented) The preset resolutions appearing in the menu\n"));
VAR(VidPresetVer[0].nWidth); VAR(VidPresetVer[0].nHeight);
VAR(VidPresetVer[1].nWidth); VAR(VidPresetVer[1].nHeight);
VAR(VidPresetVer[2].nWidth); VAR(VidPresetVer[2].nHeight);
VAR(VidPresetVer[3].nWidth); VAR(VidPresetVer[3].nHeight);
_ftprintf(h, _T("\n// (Vertical Oriented) Full-screen size (0 = use display mode variables)\n"));
VAR(nScreenSizeVer);
_ftprintf(h, _T("\n// Full-screen bit depth\n"));
VAR(nVidDepth);
_ftprintf(h, _T("\n// Specify the refresh rate, 0 = default (changing this will not work with many video cards)\n"));
VAR(nVidRefresh);
_ftprintf(h, _T("\n// If non-zero, do not rotate the graphics for vertical games\n"));
VAR(nVidRotationAdjust);
_ftprintf(h, _T("\n// Initial window size (0 = autosize)\n"));
VAR(nWindowSize);
_ftprintf(h, _T("\n// Window position\n"));
VAR(nWindowPosX); VAR(nWindowPosY);
_ftprintf(h, _T("\n// If non-zero, perform gamma correction\n"));
VAR(bDoGamma);
_ftprintf(h, _T("\n// If non-zero, use the video hardware to correct gamma\n"));
VAR(bVidUseHardwareGamma);
_ftprintf(h, _T("\n// If non-zero, don't fall back on software gamma correction\n"));
VAR(bHardwareGammaOnly);
_ftprintf(h, _T("\n// Gamma to correct with\n"));
FLT(nGamma);
_ftprintf(h, _T("\n// If non-zero, auto-switch to fullscreen after loading game\n"));
VAR(bVidAutoSwitchFull);
_ftprintf(h, _T("\n// If non-zero, allow stretching of the image to any size\n"));
VAR(bVidFullStretch);
_ftprintf(h, _T("\n// If non-zero, stretch the image to the largest size preserving aspect ratio\n"));
VAR(bVidCorrectAspect);
_ftprintf(h, _T("\n// If non-zero, try to use a triple buffer in fullscreen\n"));
VAR(bVidTripleBuffer);
_ftprintf(h, _T("\n// If non-zero, try to synchronise blits with the display\n"));
VAR(bVidVSync);
_ftprintf(h, _T("\n// Transfer method: 0 = blit from system memory / use driver/DirectX texture management;\n"));
_ftprintf(h, _T("// 1 = copy to a video memory surface, then use bltfast();\n"));
_ftprintf(h, _T("// -1 = autodetect for DirectDraw, equals 1 for Direct3D\n"));
VAR(nVidTransferMethod);
_ftprintf(h, _T("\n// If non-zero, draw scanlines to simulate a low-res monitor\n"));
VAR(bVidScanlines);
_ftprintf(h, _T("\n// Maximum scanline intensity\n"));
VAR(nVidScanIntensity);
_ftprintf(h, _T("\n// If non-zero, rotate scanlines and RGB effects for rotated games\n"));
VAR(bVidScanRotate);
_ftprintf(h, _T("\n// The selected blitter module\n"));
VAR(nVidSelect);
_ftprintf(h, _T("\n// Options for the blitter modules\n"));
VAR(nVidBlitterOpt[0]);
VAR64(nVidBlitterOpt[1]);
VAR(nVidBlitterOpt[2]);
VAR(nVidBlitterOpt[3]);
_ftprintf(h, _T("\n// If non-zero, attempt to auto-detect the monitor aspect ratio\n"));
VAR(bMonitorAutoCheck);
_ftprintf(h, _T("\n// The aspect ratio of the monitor\n"));
VAR(nVidScrnAspectX);
VAR(nVidScrnAspectY);
_ftprintf(h, _T("\n// If non-zero, force all games to use a 60Hz refresh rate\n"));
VAR(bForce60Hz);
_ftprintf(h, _T("\n// If non-zero, skip frames when needed to keep the emulation running at full speed\n"));
VAR(bAlwaysDrawFrames);
_ftprintf(h, _T("\n// If non-zero, use a placeholder image when no game is loaded\n"));
VAR(bVidUsePlaceholder);
_ftprintf(h, _T("\n// The filename of the placeholder image to use (empty filename = use built-in)\n"));
STR(szPlaceHolder);
_ftprintf(h, _T("\n"));
_ftprintf(h, _T("// --- DirectDraw blitter module settings -------------------------------------\n"));
_ftprintf(h, _T("\n// If non-zero, draw scanlines at 50%% intensity\n"));
VAR(bVidScanHalf);
_ftprintf(h, _T("\n"));
_ftprintf(h, _T("// --- Direct3D 7 blitter module settings -------------------------------------\n"));
_ftprintf(h, _T("\n// If non-zero, use bi-linear filtering to display the image\n"));
VAR(bVidBilinear);
_ftprintf(h, _T("\n// If non-zero, simulate slow phosphors (feedback)\n"));
VAR(bVidScanDelay);
_ftprintf(h, _T("\n// If non-zero, use bi-linear filtering for the scanlines\n"));
VAR(bVidScanBilinear);
_ftprintf(h, _T("\n// Feedback amount for slow phosphor simulation\n"));
VAR(nVidFeedbackIntensity);
_ftprintf(h, _T("\n// Oversaturation amount for slow phosphor simulation\n"));
VAR(nVidFeedbackOverSaturation);
_ftprintf(h, _T("\n// Angle at wich the emulated screen is tilted (in radians)\n"));
FLT(fVidScreenAngle);
_ftprintf(h, _T("\n// Angle of the sphere segment used for the 3D screen (in radians)\n"));
FLT(fVidScreenCurvature);
_ftprintf(h, _T("\n// If non-zero, force 16 bit emulation even in 32-bit screenmodes\n"));
VAR(bVidForce16bit);
_ftprintf(h, _T("\n"));
_ftprintf(h, _T("// --- DirectX Graphics 9 blitter module settings -----------------------------\n"));
_ftprintf(h, _T("\n// The filter parameters for the cubic filter\n"));
FLT(dVidCubicB);
FLT(dVidCubicC);
_ftprintf(h, _T("\n\n\n"));
_ftprintf(h, _T("// --- Sound ------------------------------------------------------------------\n"));
_ftprintf(h, _T("\n// The selected audio plugin\n"));
VAR(nAudSelect);
_ftprintf(h, _T("\n// Number of frames in sound buffer (= sound lag)\n"));
VAR(nAudSegCount);
_ftprintf(h, _T("\n// The order of PCM/ADPCM interpolation\n"));
VAR(nInterpolation);
_ftprintf(h, _T("\n// The order of FM interpolation\n"));
VAR(nFMInterpolation);
_ftprintf(h, _T("\n"));
_ftprintf(h, _T("// --- DirectSound plugin settings --------------------------------------------\n"));
_ftprintf(h, _T("\n// Sample rate\n"));
VAR(nAudSampleRate[0]);
_ftprintf(h, _T("\n// DSP module to use for sound enhancement: 0 = none, 1 = low-pass filter\n"));
VAR(nAudDSPModule[0]);
_ftprintf(h, _T("\n"));
_ftprintf(h, _T("// --- XAudio2 plugin settings ------------------------------------------------\n"));
_ftprintf(h, _T("\n// Sample rate\n"));
VAR(nAudSampleRate[1]);
_ftprintf(h, _T("\n// DSP module to use for sound enhancement: 0 = none, 1 = low-pass filter, 2 = reverb\n"));
VAR(nAudDSPModule[1]);
_ftprintf(h, _T("\n\n\n"));
_ftprintf(h, _T("// --- UI ---------------------------------------------------------------------\n"));
_ftprintf(h, _T("\n// Filename of the active UI translation template\n"));
STR(szLocalisationTemplate);
_ftprintf(h, _T("\n// Filename of the active game list translation template\n"));
STR(szGamelistLocalisationTemplate);
_ftprintf(h, _T("\n// 1 = display pause/record/replay/kaillera icons in the upper right corner of the display\n"));
VAR(nVidSDisplayStatus);
_ftprintf(h, _T("\n// Minimum height (in pixels) of the font used for the Kaillera chat function (used for arcade resolution)\n"));
VAR(nMinChatFontSize);
_ftprintf(h, _T("\n// Maximum height (in pixels) of the font used for the Kaillera chat function (used for 1280x960 or higher).\n"));
VAR(nMaxChatFontSize);
_ftprintf(h, _T("\n// Make the menu modeless\n"));
VAR(bModelessMenu);
_ftprintf(h, _T("\n// Minimum length of time to display the splash screen (in milliseconds)\n"));
VAR(nSplashTime);
_ftprintf(h, _T("\n// If non-zero, load and save all ram (the state)\n"));
VAR(bDrvSaveAll);
_ftprintf(h, _T("\n// The thread priority for the application. Do *NOT* edit this manually\n"));
VAR(nAppThreadPriority);
_ftprintf(h, _T("\n// If non-zero, process keyboard input even when the application loses focus\n"));
VAR(bAlwaysProcessKeyboardInput);
_ftprintf(h, _T("\n// If non-zero, pause when the application loses focus\n"));
VAR(bAutoPause);
_ftprintf(h, _T("\n// If non-zero, save the inputs for each game\n"));
VAR(bSaveInputs);
_ftprintf(h, _T("\n"));
_ftprintf(h, _T("// --- Load Game Dialog -------------------------------------------------------\n"));
_ftprintf(h, _T("\n// Load game dialog options\n"));
VAR(nLoadMenuShowX);
_ftprintf(h, _T("\n// Load game dialog board type filter options\n"));
VAR(nLoadMenuBoardTypeFilter);
_ftprintf(h, _T("\n// Load game dialog genre filter options\n"));
VAR(nLoadMenuGenreFilter);
_ftprintf(h, _T("\n// Load game dialog family filter options\n"));
VAR(nLoadMenuFamilyFilter);
_ftprintf(h, _T("\n// The paths to search for rom zips (include trailing backslash)\n"));
STR(szAppRomPaths[0]);
STR(szAppRomPaths[1]);
STR(szAppRomPaths[2]);
STR(szAppRomPaths[3]);
STR(szAppRomPaths[4]);
STR(szAppRomPaths[5]);
STR(szAppRomPaths[6]);
STR(szAppRomPaths[7]);
_ftprintf(h, _T("\n// The paths to search for support files (include trailing backslash)\n"));
STR(szAppPreviewsPath);
STR(szAppTitlesPath);
STR(szAppSelectPath);
STR(szAppVersusPath);
STR(szAppHowtoPath);
STR(szAppScoresPath);
STR(szAppBossesPath);
STR(szAppGameoverPath);
STR(szAppFlyersPath);
STR(szAppMarqueesPath);
STR(szAppControlsPath);
STR(szAppCabinetsPath);
STR(szAppPCBsPath);
STR(szAppCheatsPath);
STR(szAppHistoryPath);
STR(szAppListsPath);
STR(szAppDatListsPath);
STR(szAppIpsPath);
STR(szAppIconsPath);
STR(szAppArchivesPath);
STR(szAppHiscorePath);
_ftprintf(h, _T("\n\n\n"));
_ftprintf(h, _T("// --- miscellaneous ---------------------------------------------------------\n"));
_ftprintf(h, _T("\n// If non-zero, use an image menu theme.\n"));
VAR(nMenuUITheme);
_ftprintf(h, _T("\n// If non-zero, don't change the status of the Num Lock key.\n"));
VAR(bNoChangeNumLock);
_ftprintf(h, _T("\n// If non-zero, write the decrypted Neo-Geo 'C' roms to disc.\n"));
VAR(bSaveCRoms);
_ftprintf(h, _T("\n// If non-zero, enable high score saving support.\n"));
VAR(EnableHiscores);
_ftprintf(h, _T("\n// The language index to use for the IPS Patch Manager dialog.\n"));
VAR(nSelectedLanguage);
_ftprintf(h, _T("\n// If non-zero, display drivers icons.\n"));
VAR(bEnableIcons);
_ftprintf(h, _T("\n// Specify icons display size, 0 = 16x16 , 1 = 24x24, 2 = 32x32.\n"));
VAR(nIconsSize);
_ftprintf(h, _T("\n// Previous games list.\n"));
STR(szPrevGames[0]);
STR(szPrevGames[1]);
STR(szPrevGames[2]);
STR(szPrevGames[3]);
STR(szPrevGames[4]);
STR(szPrevGames[5]);
STR(szPrevGames[6]);
STR(szPrevGames[7]);
STR(szPrevGames[8]);
STR(szPrevGames[9]);
_ftprintf(h, _T("\n// Player default controls, number is the index of the configuration in the input dialog\n"));
VAR(nPlayerDefaultControls[0]);
STR(szPlayerDefaultIni[0]);
VAR(nPlayerDefaultControls[1]);
STR(szPlayerDefaultIni[1]);
VAR(nPlayerDefaultControls[2]);
STR(szPlayerDefaultIni[2]);
VAR(nPlayerDefaultControls[3]);
STR(szPlayerDefaultIni[3]);
_ftprintf(h, _T("\n\n\n"));
#undef STR
#undef FLT
#undef VAR
fclose(h);
return 0;
}
/*! This function computes the gravitational potential for ALL the particles.
* First, the (short-range) tree potential is computed, and then, if needed,
* the long range PM potential is added.
*/
void compute_potential(void)
{
int i;
#ifndef NOGRAVITY
int j, k, ret, sendTask, recvTask;
int ndone, ndone_flag, dummy;
int ngrp, place, nexport, nimport;
double fac;
MPI_Status status;
double r2;
if(All.ComovingIntegrationOn)
set_softenings();
if(ThisTask == 0)
{
printf("Start computation of potential for all particles...\n");
fflush(stdout);
}
CPU_Step[CPU_MISC] += measure_time();
if(TreeReconstructFlag)
{
if(ThisTask == 0)
printf("Tree construction.\n");
CPU_Step[CPU_MISC] += measure_time();
#if defined(SFR) || defined(BLACK_HOLES)
rearrange_particle_sequence();
#endif
force_treebuild(NumPart, NULL);
CPU_Step[CPU_TREEBUILD] += measure_time();
TreeReconstructFlag = 0;
if(ThisTask == 0)
printf("Tree construction done.\n");
}
/* allocate buffers to arrange communication */
All.BunchSize =
(int) ((All.BufferSize * 1024 * 1024) / (sizeof(struct data_index) + sizeof(struct data_nodelist) +
sizeof(struct gravdata_in) + sizeof(struct potdata_out) +
sizemax(sizeof(struct gravdata_in),
sizeof(struct potdata_out))));
DataIndexTable = (struct data_index *) mymalloc(All.BunchSize * sizeof(struct data_index));
DataNodeList = (struct data_nodelist *) mymalloc(All.BunchSize * sizeof(struct data_nodelist));
for(i = 0; i < NumPart; i++)
if(P[i].Ti_current != All.Ti_Current)
drift_particle(i, All.Ti_Current);
i = 0; /* beginn with this index */
do
{
for(j = 0; j < NTask; j++)
{
Send_count[j] = 0;
Exportflag[j] = -1;
}
/* do local particles and prepare export list */
for(nexport = 0; i < NumPart; i++)
{
#ifndef PMGRID
ret = force_treeevaluate_potential(i, 0, &nexport, Send_count);
#else
ret = force_treeevaluate_potential_shortrange(i, 0, &nexport, Send_count);
#endif
if(ret < 0)
break; /* export buffer has filled up */
}
#ifdef MYSORT
mysort_dataindex(DataIndexTable, nexport, sizeof(struct data_index), data_index_compare);
#else
qsort(DataIndexTable, nexport, sizeof(struct data_index), data_index_compare);
#endif
MPI_Allgather(Send_count, NTask, MPI_INT, Sendcount_matrix, NTask, MPI_INT, MPI_COMM_WORLD);
for(j = 0, nimport = 0, Recv_offset[0] = 0, Send_offset[0] = 0; j < NTask; j++)
{
Recv_count[j] = Sendcount_matrix[j * NTask + ThisTask];
nimport += Recv_count[j];
if(j > 0)
{
Send_offset[j] = Send_offset[j - 1] + Send_count[j - 1];
Recv_offset[j] = Recv_offset[j - 1] + Recv_count[j - 1];
}
}
GravDataGet = (struct gravdata_in *) mymalloc(nimport * sizeof(struct gravdata_in));
GravDataIn = (struct gravdata_in *) mymalloc(nexport * sizeof(struct gravdata_in));
/* prepare particle data for export */
for(j = 0; j < nexport; j++)
{
place = DataIndexTable[j].Index;
for(k = 0; k < 3; k++)
GravDataIn[j].Pos[k] = P[place].Pos[k];
#ifdef UNEQUALSOFTENINGS
GravDataIn[j].Type = P[place].Type;
#ifdef ADAPTIVE_GRAVSOFT_FORGAS
if(P[place].Type == 0)
GravDataIn[j].Soft = SphP[place].Hsml;
#endif
#endif
GravDataIn[j].OldAcc = P[place].OldAcc;
for(k = 0; k < NODELISTLENGTH; k++)
GravDataIn[j].NodeList[k] = DataNodeList[DataIndexTable[j].IndexGet].NodeList[k];
}
/* exchange particle data */
for(ngrp = 1; ngrp < (1 << PTask); ngrp++)
{
sendTask = ThisTask;
recvTask = ThisTask ^ ngrp;
if(recvTask < NTask)
{
if(Send_count[recvTask] > 0 || Recv_count[recvTask] > 0)
{
/* get the particles */
MPI_Sendrecv(&GravDataIn[Send_offset[recvTask]],
Send_count[recvTask] * sizeof(struct gravdata_in), MPI_BYTE,
recvTask, TAG_POTENTIAL_A,
&GravDataGet[Recv_offset[recvTask]],
Recv_count[recvTask] * sizeof(struct gravdata_in), MPI_BYTE,
recvTask, TAG_POTENTIAL_A, MPI_COMM_WORLD, &status);
}
}
}
myfree(GravDataIn);
PotDataResult = (struct potdata_out *) mymalloc(nimport * sizeof(struct potdata_out));
PotDataOut = (struct potdata_out *) mymalloc(nexport * sizeof(struct potdata_out));
/* now do the particles that were sent to us */
for(j = 0; j < nimport; j++)
{
#ifndef PMGRID
force_treeevaluate_potential(j, 1, &dummy, &dummy);
#else
force_treeevaluate_potential_shortrange(j, 1, &dummy, &dummy);
#endif
}
if(i >= NumPart)
ndone_flag = 1;
else
ndone_flag = 0;
MPI_Allreduce(&ndone_flag, &ndone, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
/* get the result */
for(ngrp = 1; ngrp < (1 << PTask); ngrp++)
{
sendTask = ThisTask;
recvTask = ThisTask ^ ngrp;
if(recvTask < NTask)
{
if(Send_count[recvTask] > 0 || Recv_count[recvTask] > 0)
{
/* send the results */
MPI_Sendrecv(&PotDataResult[Recv_offset[recvTask]],
Recv_count[recvTask] * sizeof(struct potdata_out),
MPI_BYTE, recvTask, TAG_POTENTIAL_B,
&PotDataOut[Send_offset[recvTask]],
Send_count[recvTask] * sizeof(struct potdata_out),
MPI_BYTE, recvTask, TAG_POTENTIAL_B, MPI_COMM_WORLD, &status);
}
}
}
/* add the results to the local particles */
for(j = 0; j < nexport; j++)
{
place = DataIndexTable[j].Index;
P[place].p.dPotential += PotDataOut[j].Potential;
}
myfree(PotDataOut);
myfree(PotDataResult);
myfree(GravDataGet);
}
while(ndone < NTask);
myfree(DataNodeList);
myfree(DataIndexTable);
/* add correction to exclude self-potential */
for(i = 0; i < NumPart; i++)
{
#ifdef FLTROUNDOFFREDUCTION
P[i].p.Potential = FLT(P[i].p.dPotential);
#endif
/* remove self-potential */
P[i].p.Potential += P[i].Mass / All.SofteningTable[P[i].Type];
if(All.ComovingIntegrationOn)
if(All.PeriodicBoundariesOn)
P[i].p.Potential -= 2.8372975 * pow(P[i].Mass, 2.0 / 3) *
pow(All.Omega0 * 3 * All.Hubble * All.Hubble / (8 * M_PI * All.G), 1.0 / 3);
}
/* multiply with the gravitational constant */
for(i = 0; i < NumPart; i++)
P[i].p.Potential *= All.G;
#ifdef PMGRID
#ifdef PERIODIC
pmpotential_periodic();
#ifdef PLACEHIGHRESREGION
i = pmpotential_nonperiodic(1);
if(i == 1) /* this is returned if a particle lied outside allowed range */
{
pm_init_regionsize();
pm_setup_nonperiodic_kernel();
i = pmpotential_nonperiodic(1); /* try again */
}
if(i == 1)
endrun(88686);
#endif
#else
i = pmpotential_nonperiodic(0);
if(i == 1) /* this is returned if a particle lied outside allowed range */
{
pm_init_regionsize();
pm_setup_nonperiodic_kernel();
i = pmpotential_nonperiodic(0); /* try again */
}
if(i == 1)
endrun(88687);
#ifdef PLACEHIGHRESREGION
i = pmpotential_nonperiodic(1);
if(i == 1) /* this is returned if a particle lied outside allowed range */
{
pm_init_regionsize();
i = pmpotential_nonperiodic(1);
}
if(i != 0)
endrun(88688);
#endif
#endif
#endif
if(All.ComovingIntegrationOn)
{
#ifndef PERIODIC
fac = -0.5 * All.Omega0 * All.Hubble * All.Hubble;
for(i = 0; i < NumPart; i++)
{
for(k = 0, r2 = 0; k < 3; k++)
r2 += P[i].Pos[k] * P[i].Pos[k];
P[i].p.Potential += fac * r2;
}
#endif
}
else
{
fac = -0.5 * All.OmegaLambda * All.Hubble * All.Hubble;
if(fac != 0)
{
for(i = 0; i < NumPart; i++)
{
for(k = 0, r2 = 0; k < 3; k++)
r2 += P[i].Pos[k] * P[i].Pos[k];
P[i].p.Potential += fac * r2;
}
}
}
if(ThisTask == 0)
{
printf("potential done.\n");
fflush(stdout);
}
#else
for(i = 0; i < NumPart; i++)
P[i].Potential = 0;
#endif
CPU_Step[CPU_POTENTIAL] += measure_time();
}
static void
sobel (TileDrawable *drawable,
gint do_horizontal,
gint do_vertical,
gint keep_sign)
{
GPixelRgn srcPR, destPR;
gint width, height;
gint num_channels;
guchar *dest;
guchar *prev_row;
guchar *cur_row;
guchar *next_row;
guchar *pr;
guchar *cr;
guchar *nr;
guchar *tmp;
gint row, col;
gint x1, y1, x2, y2;
gint alpha;
gint counter;
GPrecisionType precision = gimp_drawable_precision (drawable->id);
/* Get the input area. This is the bounding box of the selection in
* the image (or the entire image if there is no selection). Only
* operating on the input area is simply an optimization. It doesn't
* need to be done for correct operation. (It simply makes it go
* faster, since fewer pixels need to be operated on).
*/
gimp_drawable_mask_bounds (drawable->id, &x1, &y1, &x2, &y2);
gimp_progress_init (_("Sobel Edge Detect"));
/* Get the size of the input image. (This will/must be the same
* as the size of the output image.
*/
width = drawable->width;
height = drawable->height;
num_channels = drawable->num_channels;
alpha = gimp_drawable_has_alpha (drawable -> id);
/* allocate generic data row buffers */
prev_row = (guchar *) malloc ((x2 - x1 + 2) * drawable->bpp);
cur_row = (guchar *) malloc ((x2 - x1 + 2) * drawable->bpp);
next_row = (guchar *) malloc ((x2 - x1 + 2) * drawable->bpp);
dest = (guchar *) malloc ((x2 - x1) * drawable->bpp);
/* initialize the pixel regions */
gimp_pixel_rgn_init (&srcPR, drawable, 0, 0, width, height, FALSE, FALSE);
gimp_pixel_rgn_init (&destPR, drawable, 0, 0, width, height, TRUE, TRUE);
/* generic data pointers to the three rows */
pr = prev_row + drawable->bpp;
cr = cur_row + drawable->bpp;
nr = next_row + drawable->bpp;
sobel_prepare_row (&srcPR, pr, x1, y1 - 1, (x2 - x1));
sobel_prepare_row (&srcPR, cr, x1, y1, (x2 - x1));
counter = 0;
/* loop through the rows, applying the sobel convolution */
for (row = y1; row < y2; row++)
{
/* prepare the next row */
sobel_prepare_row (&srcPR, nr, x1, row + 1, (x2 - x1));
switch(precision)
{
case PRECISION_U8:
{
guint8* d = (guint8*)dest;
guint8* p = (guint8*)pr;
guint8* c = (guint8*)cr;
guint8* n = (guint8*)nr;
gint gradient, hor_gradient, ver_gradient;
for (col = 0; col < (x2 - x1) * num_channels; col++) {
if (do_horizontal)
hor_gradient = (p[col - num_channels] + 2 * p[col] + p[col + num_channels]) -
(n[col - num_channels] + 2 * n[col] + n[col + num_channels]);
else
hor_gradient = 0;
if (do_vertical)
ver_gradient = (p[col - num_channels] + 2 * c[col - num_channels] + n[col - num_channels]) -
(p[col + num_channels] + 2 * c[col + num_channels] + n[col + num_channels]);
else
ver_gradient = 0;
if (do_vertical && do_horizontal)
gradient = ROUND_TO_INT (RMS (hor_gradient, ver_gradient)) / 5.66;
else
{
if (keep_sign)
gradient = 127 + (ROUND_TO_INT ((hor_gradient + ver_gradient) / 8.0));
else
gradient = ROUND_TO_INT (abs (hor_gradient + ver_gradient) / 4.0);
}
if (alpha && (((col + 1) % num_channels) == 0))
{
*d++ = (counter == 0) ? 0 : 255;
counter = 0;
}
else
{
*d++ = gradient;
if (gradient > 10) counter ++;
}
}
}
break;
case PRECISION_U16:
{
guint16* d = (guint16*)dest;
guint16* p = (guint16*)pr;
guint16* c = (guint16*)cr;
guint16* n = (guint16*)nr;
gint gradient, hor_gradient, ver_gradient;
for (col = 0; col < (x2 - x1) * num_channels; col++) {
if (do_horizontal)
hor_gradient = (p[col - num_channels] + 2 * p[col] + p[col + num_channels]) -
(n[col - num_channels] + 2 * n[col] + n[col + num_channels]);
else
hor_gradient = 0;
if (do_vertical)
ver_gradient = (p[col - num_channels] + 2 * c[col - num_channels] + n[col - num_channels]) -
(p[col + num_channels] + 2 * c[col + num_channels] + n[col + num_channels]);
else
ver_gradient = 0;
if (do_vertical && do_horizontal)
gradient = ROUND_TO_INT (RMS (hor_gradient, ver_gradient)) / 5.66;
else
{
if (keep_sign)
gradient = 32767 + (ROUND_TO_INT ((hor_gradient + ver_gradient) / 8.0));
else
gradient = ROUND_TO_INT (abs (hor_gradient + ver_gradient) / 4.0);
}
if (alpha && (((col + 1) % num_channels) == 0))
{
*d++ = (counter == 0) ? 0 : 65535;
counter = 0;
}
else
{
*d++ = gradient;
if (gradient > (10/255.0) * 65535) counter ++;
}
}
}
break;
case PRECISION_FLOAT:
{
gfloat* d = (gfloat*)dest;
gfloat* p = (gfloat*)pr;
gfloat* c = (gfloat*)cr;
gfloat* n = (gfloat*)nr;
gfloat gradient, hor_gradient, ver_gradient;
for (col = 0; col < (x2 - x1) * num_channels; col++) {
if (do_horizontal)
hor_gradient = (p[col - num_channels] + 2 * p[col] + p[col + num_channels]) -
(n[col - num_channels] + 2 * n[col] + n[col + num_channels]);
else
hor_gradient = 0;
if (do_vertical)
ver_gradient = (p[col - num_channels] + 2 * c[col - num_channels] + n[col - num_channels]) -
(p[col + num_channels] + 2 * c[col + num_channels] + n[col + num_channels]);
else
ver_gradient = 0;
if (do_vertical && do_horizontal)
gradient = (RMS (hor_gradient, ver_gradient)) / 5.66;
else
{
if (keep_sign)
gradient = .5 + (hor_gradient + ver_gradient) / 8.0;
else
gradient = abs (hor_gradient + ver_gradient) / 4.0;
}
if (alpha && (((col + 1) % num_channels) == 0))
{
*d++ = (counter == 0) ? 0.0 : 1.0;
counter = 0;
}
else
{
*d++ = gradient;
if (gradient > (10/255.0)) counter ++;
}
}
}
break;
case PRECISION_FLOAT16:
{
guint16* d = (guint16*)dest;
guint16* p = (guint16*)pr;
guint16* c = (guint16*)cr;
guint16* n = (guint16*)nr;
gfloat gradient, hor_gradient, ver_gradient;
ShortsFloat u,v,s, u1, v1, s1;
for (col = 0; col < (x2 - x1) * num_channels; col++) {
if (do_horizontal)
hor_gradient = (
FLT(p[col - num_channels],u) +
2 * FLT(p[col], v) +
FLT(p[col + num_channels], s)
) -
(
FLT(n[col - num_channels],u1) +
2 * FLT(n[col], v1)
+ FLT(n[col + num_channels], s1)
);
else
hor_gradient = 0;
if (do_vertical)
ver_gradient = (
FLT(p[col - num_channels], u) +
2 * FLT (c[col - num_channels],v) +
FLT(n[col - num_channels],s)
) -
(
FLT(p[col + num_channels], u1) +
2 * FLT (c[col + num_channels],v1) +
FLT(n[col + num_channels], s1)
);
else
ver_gradient = 0;
if (do_vertical && do_horizontal)
gradient = (RMS (hor_gradient, ver_gradient)) / 5.66;
else
{
if (keep_sign)
gradient = .5 + (hor_gradient + ver_gradient) / 8.0;
else
gradient = abs (hor_gradient + ver_gradient) / 4.0;
}
if (alpha && (((col + 1) % num_channels) == 0))
{
*d++ = (counter == 0) ? 0.0 : 1.0;
counter = 0;
}
else
{
*d++ = FLT16(gradient, u);
if (gradient > (10/255.0)) counter ++;
}
}
}
break;
}
/* store the dest */
gimp_pixel_rgn_set_row (&destPR, dest, x1, row, (x2 - x1));
/* shuffle the row pointers */
tmp = pr;
pr = cr;
cr = nr;
nr = tmp;
if ((row % 5) == 0)
gimp_progress_update ((double) row / (double) (y2 - y1));
}
/* update the sobeled region */
gimp_drawable_flush (drawable);
gimp_drawable_merge_shadow (drawable->id, TRUE);
gimp_drawable_update (drawable->id, x1, y1, (x2 - x1), (y2 - y1));
free (prev_row);
free (cur_row);
free (next_row);
free (dest);
}
/*
* dooper() will execute a constant operation in a node and return
* the node to be the result of the operation.
*/
static EXPR *dooper P1 (EXPR *, ep)
{
EXPRTYPE type = ep->nodetype;
ep->nodetype = ep->v.p[0]->nodetype;
switch (ep->v.p[0]->nodetype) {
#ifdef FLOAT_SUPPORT
#ifndef FLOAT_BOOTSTRAP
RVAL f;
#endif /* FLOAT_BOOTSTRAP */
case en_fcon:
#ifndef FLOAT_BOOTSTRAP
FASSIGN (f, ep->v.p[0]->v.f);
switch (type) {
case en_uminus:
FASSIGN (ep->v.f, f);
FNEG (ep->v.f);
break;
case en_test:
ep->v.i = FTST (f) ? (IVAL) 1 : (IVAL) 0;
ep->nodetype = en_icon;
break;
case en_not:
ep->v.i = FTST (f) ? (IVAL) 0 : (IVAL) 1;
ep->nodetype = en_icon;
break;
case en_cast:
if (is_real_floating_type (ep->etp)) {
ep->v.f = f;
} else if (is_bool (ep->etp)) {
ep->v.u = (UVAL) (f ? 1 : 0);
ep->nodetype = en_icon;
} else {
FTOL (ep->v.i, f);
ep->v.i = strip_icon (ep->v.i, ep->etp);
ep->nodetype = en_icon;
}
break;
case en_add:
FADD3 (ep->v.f, f, ep->v.p[1]->v.f);
break;
case en_sub:
FSUB3 (ep->v.f, f, ep->v.p[1]->v.f);
break;
case en_mul:
FMUL3 (ep->v.f, f, ep->v.p[1]->v.f);
break;
case en_div:
if (FTST (ep->v.p[1]->v.f)) {
FDIV3 (ep->v.f, f, ep->v.p[1]->v.f);
} else {
ep->nodetype = en_div;
}
break;
case en_eq:
ep->v.i = (IVAL) FEQ (f, ep->v.p[1]->v.f);
ep->nodetype = en_icon;
break;
case en_ne:
ep->v.i = (IVAL) FNE (f, ep->v.p[1]->v.f);
ep->nodetype = en_icon;
break;
case en_land:
ep->v.i = (IVAL) (FTST (f) && FTST (ep->v.p[1]->v.f));
ep->nodetype = en_icon;
break;
case en_lor:
ep->v.i = (IVAL) (FTST (f) || FTST (ep->v.p[1]->v.f));
ep->nodetype = en_icon;
break;
case en_lt:
ep->v.i = (IVAL) FLT (f, ep->v.p[1]->v.f);
ep->nodetype = en_icon;
break;
case en_le:
ep->v.i = (IVAL) FLE (f, ep->v.p[1]->v.f);
ep->nodetype = en_icon;
break;
case en_gt:
ep->v.i = (IVAL) FGT (f, ep->v.p[1]->v.f);
ep->nodetype = en_icon;
break;
case en_ge:
ep->v.i = (IVAL) FGE (f, ep->v.p[1]->v.f);
ep->nodetype = en_icon;
break;
default:
CANNOT_REACH_HERE ();
break;
}
break;
#endif /* FLOAT_BOOTSTRAP */
#endif /* FLOAT_SUPPORT */
case en_icon:
if (is_unsigned_type (ep->v.p[0]->etp)) {
UVAL u = ep->v.p[0]->v.u;
switch (type) {
case en_uminus:
/*
* unary minus on an unsigned is normally a mistake so we must
* fool the compiler into not giving a warning.
*/
ep->v.u = (UVAL) (-(IVAL) u);
break;
case en_test:
ep->v.u = (u ? (UVAL) 1 : (UVAL) 0);
break;
case en_not:
ep->v.u = (u ? (UVAL) 0 : (UVAL) 1);
break;
case en_compl:
ep->v.u = (UVAL) strip_icon ((IVAL) ~u, ep->etp);
break;
case en_cast:
if (is_bool (ep->etp)) {
ep->v.u = (UVAL) (u ? 1 : 0);
#ifdef FLOAT_SUPPORT
} else if (is_real_floating_type (ep->etp)) {
ep->nodetype = en_fcon;
UTOF (ep->v.f, u);
break;
#endif /* FLOAT_SUPPORT */
} else {
ep->v.u = (UVAL) strip_icon ((IVAL) u, ep->etp);
}
break;
case en_add:
ep->v.u = u + ep->v.p[1]->v.u;
break;
case en_sub:
ep->v.u = u - ep->v.p[1]->v.u;
break;
case en_mul:
ep->v.u = u * ep->v.p[1]->v.u;
break;
case en_div:
if (ep->v.p[1]->v.u == (UVAL) 0) {
ep->nodetype = en_div;
} else {
ep->v.u = u / ep->v.p[1]->v.u;
}
break;
case en_mod:
if (ep->v.p[1]->v.u == (UVAL) 0) {
ep->nodetype = en_mod;
} else {
ep->v.u = u % ep->v.p[1]->v.u;
}
break;
case en_and:
ep->v.u = u & ep->v.p[1]->v.u;
break;
case en_or:
ep->v.u = u | ep->v.p[1]->v.u;
break;
case en_xor:
ep->v.u = u ^ ep->v.p[1]->v.u;
break;
case en_eq:
ep->v.u = (UVAL) (u == ep->v.p[1]->v.u);
break;
case en_ne:
ep->v.u = (UVAL) (u != ep->v.p[1]->v.u);
break;
case en_land:
ep->v.u = (UVAL) (u && ep->v.p[1]->v.u);
break;
case en_lor:
ep->v.u = (UVAL) (u || ep->v.p[1]->v.u);
break;
case en_lt:
ep->v.u = (UVAL) (u < ep->v.p[1]->v.u);
break;
case en_le:
ep->v.u = (UVAL) (u <= ep->v.p[1]->v.u);
break;
case en_gt:
ep->v.u = (UVAL) (u > ep->v.p[1]->v.u);
break;
case en_ge:
ep->v.u = (UVAL) (u >= ep->v.p[1]->v.u);
break;
case en_lsh:
ep->v.u = u << ep->v.p[1]->v.u;
break;
case en_rsh:
ep->v.u = u >> ep->v.p[1]->v.u;
break;
default:
CANNOT_REACH_HERE ();
break;
}
} else {
/* PUBLIC */
Fpa_chunk flist_insert(Fpa_chunk f, Term x)
{
if (f == NULL) {
f = get_fpa_chunk();
FLAST(f) = x;
f->n = 1;
}
else if (f->n == FMAX) {
if (FLT(x,FLAST(f)))
f->next = flist_insert(f->next, x);
else if (x == FLAST(f))
fprintf(stderr, "WARNING: flist_insert, item %p already here (1)!\n", x);
else if (FGT(x,FFIRST(f))) {
/*
This special case isn't necessary. It is to improve performance.
The application for which I'm writing this inserts items in
increasing order (most of the time), and this prevents a lot of
half-empty chunks in that case.
*/
Fpa_chunk f2 = flist_insert(NULL, x);
f2->next = f;
f = f2;
}
else {
/* split this chunk in half */
Fpa_chunk f2 = get_fpa_chunk();
int move = FMAX / 2;
int i, j;
for (i = 0, j = FMAX-move; i < move; i++, j++) {
f2->d[j] = f->d[i];
f->d[i] = NULL;
}
f2->n = move;
f->n = FMAX - move;
f2->next = f;
f = flist_insert(f2, x);
}
}
else {
if (f->next && FLE(x,FFIRST(f->next)))
f->next = flist_insert(f->next, x);
else {
/* insert into this pa_chunk */
int n = f->n;
int i = FMAX - n;
while (i < FMAX && FLT(x,f->d[i]))
i++;
if (i < FMAX && x == f->d[i])
fprintf(stderr, "WARNING: flist_insert, item %p already here (2)!\n", x);
else if (i == FMAX - n) {
f->d[i-1] = x;
f->n = n+1;
}
else {
/* insert at i-1, shifting the rest */
int j;
for (j = FMAX-n; j < i; j++)
f->d[j-1] = f->d[j];
f->d[i-1] = x;
f->n = n+1;
}
}
}
return f;
} /* flist_insert */
int main() {
int i,j,n,pt;
#ifdef TEST
class quadtree<2> testtree,testtree2;
TinyVector<FLT,2> x1(0.0,0.0),x2(1.0,1.0),x3(1.1,1.1);
FLT vtest[17][2] = {{0.1,0.1},{0.2,0.4},{.9,0.2},{.125,0.7},{0.51,0.53},{0.85,0.15},{0.25,0.85},{0.99,0.99},{0.001,.3},{0.01,0.3},{0.2,0.1},
{2.0,2.0},{0.9,0.9},{0.9,0.9},{0.9,0.9},{0.9,0.9},{0.9,0.9}};
blitz::Array<blitz::TinyVector<FLT,2>,1> vtest2(17);
FLT dist;
for(i=0;i<17;++i) {
vtest2(i)(0) = vtest[i][0];
vtest2(i)(1) = vtest[i][1];
}
testtree.init(vtest2, 2000, x1, x2);
printf("%f %f %f %f\n",testtree.xmin(0),testtree.xmax(0),testtree.xmin(1),testtree.xmax(1));
for(i=0;i<11;++i) {
printf("%d\n",i);
testtree.addpt(i);
}
// /* Test what happens for point outside domain */
// printf("point outside test\n");
// testtree.addpt(11);
//
// /* Test what happens for same point 5 times */
// printf("5 times test\n");
// for(i=12;i<17;++i)
// testtree.addpt(i);
quadtree<2> copytree(testtree);
dist = testtree.nearpt(9, i);
printf("%d %f\n",i,dist);
dist = copytree.nearpt(9, i);
printf("%d %f\n",i,dist);
vtest2(0)[0] = 0.9;
vtest2(0)[1] = 0.9;
testtree.update(0,10);
printf("update ok\n");
testtree.reinit();
dist = testtree.nearpt(vtest2(0), i);
printf("%d %f\n",i,dist);
dist = testtree.nearpt(0, i);
printf("%d %f\n",i,dist);
dist = testtree.nearpt(x3.data(), i);
printf("%d %f\n",i,dist);
testtree.output("out");
testtree.output("out",quadtree<2>::text);
/* What happens if I insert point twice? */
testtree.addpt(0);
// class quadtree<3> octtree;
// FLT y1[3] = {0.0,0.0,0.0},y2[3] ={1.0,1.0,1.0};
// FLT wtest[9][3] = {{0.1,0.1,0.1},{0.1,0.1,0.9},{0.1,0.9,0.1},{0.1,0.9,0.9},{0.9,0.1,0.1},{0.9,0.1,0.9},{0.9,0.9,0.1},{0.9,0.9,0.9}, {0.9,0.85,0.9}};
// octtree.init(wtest, 20, y1, y2);
// for(i=0;i<9;++i) {
// printf("%d\n",i);
// octtree.addpt(i);
// }
//
// quadtree<3> copytree3(octtree);
//
// dist = octtree.nearpt(8, i);
// printf("%d %f\n",i,dist);
// dist = copytree3.nearpt(8, i);
// printf("%d %f\n",i,dist);
//
//
// wtest[0][0] = 0.9;
// wtest[0][1] = 0.9;
// wtest[0][2] = 0.87;
// octtree.update(0,9);
//
// printf("update ok\n");
//
// octtree.reinit();
//
// dist = octtree.nearpt(0, i);
// printf("%d %f\n",i,dist);
//
// octtree.output("out3");
// octtree.output("out3",quadtree<3>::text);
//
// /* What happens if I insert 4 points outside of domain? */
// quadtree<2> bigtree;
// bigtree.init(vtest,100000, x1, x2);
// for(i=0;i<16;++i) {
// printf("%d\n",i);
// bigtree.addpt(i);
// }
// bigtree.output("huh",quadtree<2>::text);
#endif
#ifdef FINDUNIQUE
int nvrts = 556732;
class quadtree<3> dups;
FLT (*vrts)[3];
FLT xmax[3],xmin[3];
FLT dist1;
int tvrtx[3];
int *ipoint;
ipoint = new int[nvrts];
if (ipoint == 0) printf("shit\n");
vrts = (FLT (*)[3]) malloc(nvrts*3*sizeof(double));
if (vrts == NULL) printf("shit\n");
dups.allocate(vrts,nvrts);
scanf("%lf %lf %lf\n",&vrts[0][0],&vrts[0][1],&vrts[0][2]);
for(n=0;n<3;++n) {
xmax[n] = vrts[0][n];
xmin[n] = vrts[0][n];
}
for(i=1;i<nvrts;++i) {
scanf("%lf %lf %lf",&vrts[i][0],&vrts[i][1],&vrts[i][2]);
for(n=0;n<3;++n) {
xmax[n] = MAX(vrts[i][n],xmax[n]);
xmin[n] = MIN(vrts[i][n],xmin[n]);
}
}
for(i=0;i<nvrts;++i)
for(n=0;n<3;++n)
vrts[i][n] -= xmax[n];
for(n=0;n<3;++n) {
xmin[n] -= xmax[n];
xmax[n] = 0.0;
}
dups.init(vrts,nvrts,xmin,xmax);
dups.addpt(0);
printf("%17.10e %17.10e %17.10e\n",vrts[0][0],vrts[0][1],vrts[0][2]);
ipoint[0] = 0;
int count = 1;
for(i=0;i<nvrts;i+=4) {
for(j=0;j<3;++j) {
dist1 = dups.nearpt(vrts[i+j], pt);
if (fabs((vrts[i+j][0]-vrts[pt][0])/xmin[0]) > 1.0e-4
|| fabs((vrts[i+j][1]-vrts[pt][1])/xmin[1]) > 1.0e-4
|| fabs((vrts[i+j][2]-vrts[pt][2])/xmin[2]) > 1.0e-4) {
printf("%17.10e %17.10e %17.10e\n",vrts[i+j][0],vrts[i+j][1],vrts[i+j][2]);
dups.addpt(i+j);
ipoint[i+j] = count;
pt = count++;
}
else {
pt = ipoint[pt];
}
tvrtx[j] = pt;
}
printf("# %d %d %d\n",tvrtx[0]+1,tvrtx[1]+1,tvrtx[2]+1);
}
#endif
return(0);
}
/* This function updates the weights for SN before exploding. This is
* necessary due to the fact that gas particles neighbours of a given star
* could have been transformed into stars and they need to be taken off the
* neighbour list for the exploding star.
*/
void cs_update_weights(void)
{
MyFloat *Left, *Right;
int i, j, ndone, ndone_flag, npleft, dummy, iter = 0;
int ngrp, sendTask, recvTask, place, nexport, nimport;
long long ntot;
double dmax1, dmax2;
double desnumngb;
if(ThisTask == 0)
{
printf("... start update weights phase = %d ...\n", Flag_phase);
fflush(stdout);
}
Left = (MyFloat *) mymalloc(NumPart * sizeof(MyFloat));
Right = (MyFloat *) mymalloc(NumPart * sizeof(MyFloat));
for(i = FirstActiveParticle; i >= 0; i = NextActiveParticle[i])
{
if(P[i].Type == 6 || P[i].Type == 7)
{
Left[i] = Right[i] = 0;
}
}
/* allocate buffers to arrange communication */
Ngblist = (int *) mymalloc(NumPart * sizeof(int));
R2ngblist = (double *) mymalloc(NumPart * sizeof(double));
All.BunchSize =
(int) ((All.BufferSize * 1024 * 1024) / (sizeof(struct data_index) + sizeof(struct data_nodelist) +
sizeof(struct updateweight_in) +
sizeof(struct updateweight_out) +
sizemax(sizeof(struct updateweight_in),
sizeof(struct updateweight_out))));
DataIndexTable = (struct data_index *) mymalloc(All.BunchSize * sizeof(struct data_index));
DataNodeList = (struct data_nodelist *) mymalloc(All.BunchSize * sizeof(struct data_nodelist));
desnumngb = All.DesNumNgb;
/* we will repeat the whole thing for those particles where we didn't find enough neighbours */
do
{
i = FirstActiveParticle; /* begin with this index */
do
{
for(j = 0; j < NTask; j++)
{
Send_count[j] = 0;
Exportflag[j] = -1;
}
/* do local particles and prepare export list */
for(nexport = 0; i >= 0; i = NextActiveParticle[i])
{
if((P[i].Type == 6 || P[i].Type == 7) && P[i].TimeBin >= 0)
{
if(cs_update_weight_evaluate(i, 0, &nexport, Send_count) < 0)
break;
}
}
#ifdef MYSORT
mysort_dataindex(DataIndexTable, nexport, sizeof(struct data_index), data_index_compare);
#else
qsort(DataIndexTable, nexport, sizeof(struct data_index), data_index_compare);
#endif
MPI_Allgather(Send_count, NTask, MPI_INT, Sendcount_matrix, NTask, MPI_INT, MPI_COMM_WORLD);
for(j = 0, nimport = 0, Recv_offset[0] = 0, Send_offset[0] = 0; j < NTask; j++)
{
Recv_count[j] = Sendcount_matrix[j * NTask + ThisTask];
nimport += Recv_count[j];
if(j > 0)
{
Send_offset[j] = Send_offset[j - 1] + Send_count[j - 1];
Recv_offset[j] = Recv_offset[j - 1] + Recv_count[j - 1];
}
}
UpdateweightGet = (struct updateweight_in *) mymalloc(nimport * sizeof(struct updateweight_in));
UpdateweightIn = (struct updateweight_in *) mymalloc(nexport * sizeof(struct updateweight_in));
/* prepare particle data for export */
for(j = 0; j < nexport; j++)
{
place = DataIndexTable[j].Index;
UpdateweightIn[j].Pos[0] = P[place].Pos[0];
UpdateweightIn[j].Pos[1] = P[place].Pos[1];
UpdateweightIn[j].Pos[2] = P[place].Pos[2];
UpdateweightIn[j].Hsml = PPP[place].Hsml;
memcpy(UpdateweightIn[j].NodeList,
DataNodeList[DataIndexTable[j].IndexGet].NodeList, NODELISTLENGTH * sizeof(int));
}
/* exchange particle data */
for(ngrp = 1; ngrp < (1 << PTask); ngrp++)
{
sendTask = ThisTask;
recvTask = ThisTask ^ ngrp;
if(recvTask < NTask)
{
if(Send_count[recvTask] > 0 || Recv_count[recvTask] > 0)
{
/* get the particles */
MPI_Sendrecv(&UpdateweightIn[Send_offset[recvTask]],
Send_count[recvTask] * sizeof(struct updateweight_in), MPI_BYTE,
recvTask, TAG_DENS_A,
&UpdateweightGet[Recv_offset[recvTask]],
Recv_count[recvTask] * sizeof(struct updateweight_in), MPI_BYTE,
recvTask, TAG_DENS_A, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
}
}
}
myfree(UpdateweightIn);
UpdateweightResult =
(struct updateweight_out *) mymalloc(nimport * sizeof(struct updateweight_out));
UpdateweightOut = (struct updateweight_out *) mymalloc(nexport * sizeof(struct updateweight_out));
/* now do the particles that were sent to us */
for(j = 0; j < nimport; j++)
cs_update_weight_evaluate(j, 1, &dummy, &dummy);
if(i < 0)
ndone_flag = 1;
else
ndone_flag = 0;
MPI_Allreduce(&ndone_flag, &ndone, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
/* get the result */
for(ngrp = 1; ngrp < (1 << PTask); ngrp++)
{
sendTask = ThisTask;
recvTask = ThisTask ^ ngrp;
if(recvTask < NTask)
{
if(Send_count[recvTask] > 0 || Recv_count[recvTask] > 0)
{
/* send the results */
MPI_Sendrecv(&UpdateweightResult[Recv_offset[recvTask]],
Recv_count[recvTask] * sizeof(struct updateweight_out),
MPI_BYTE, recvTask, TAG_DENS_B,
&UpdateweightOut[Send_offset[recvTask]],
Send_count[recvTask] * sizeof(struct updateweight_out),
MPI_BYTE, recvTask, TAG_DENS_B, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
}
}
}
/* add the result to the local particles */
for(j = 0; j < nexport; j++)
{
place = DataIndexTable[j].Index;
PPP[place].n.dNumNgb += UpdateweightOut[j].Ngb;
}
myfree(UpdateweightOut);
myfree(UpdateweightResult);
myfree(UpdateweightGet);
}
while(ndone < NTask);
/* do final operations on results */
for(i = FirstActiveParticle, npleft = 0; i >= 0; i = NextActiveParticle[i])
{
if(P[i].Type == 6 || P[i].Type == 7)
{
#ifdef FLTROUNDOFFREDUCTION
PPP[i].n.NumNgb = FLT(PPP[i].n.dNumNgb);
#endif
/* now check whether we had enough neighbours */
if(PPP[i].n.NumNgb < (desnumngb - All.MaxNumNgbDeviation) ||
(PPP[i].n.NumNgb > (desnumngb + All.MaxNumNgbDeviation)
&& PPP[i].Hsml > (1.01 * All.MinGasHsml)))
{
/* need to redo this particle */
npleft++;
if(Left[i] > 0 && Right[i] > 0)
if((Right[i] - Left[i]) < 1.0e-3 * Left[i])
{
/* this one should be ok */
npleft--;
P[i].TimeBin = -P[i].TimeBin - 1; /* Mark as inactive */
continue;
}
if(PPP[i].n.NumNgb < (desnumngb - All.MaxNumNgbDeviation))
Left[i] = DMAX(PPP[i].Hsml, Left[i]);
else
{
if(Right[i] != 0)
{
if(PPP[i].Hsml < Right[i])
Right[i] = PPP[i].Hsml;
}
else
Right[i] = PPP[i].Hsml;
}
if(iter >= MAXITER - 10)
{
printf
("i=%d task=%d ID=%d Hsml=%g Left=%g Right=%g Ngbs=%g Right-Left=%g\n pos=(%g|%g|%g)\n",
i, ThisTask, (int) P[i].ID, PPP[i].Hsml, Left[i], Right[i],
(float) PPP[i].n.NumNgb, Right[i] - Left[i], P[i].Pos[0], P[i].Pos[1], P[i].Pos[2]);
fflush(stdout);
}
if(Right[i] > 0 && Left[i] > 0)
PPP[i].Hsml = pow(0.5 * (pow(Left[i], 3) + pow(Right[i], 3)), 1.0 / 3);
else
{
if(Right[i] == 0 && Left[i] == 0)
endrun(8188); /* can't occur */
if(Right[i] == 0 && Left[i] > 0)
{
PPP[i].Hsml *= 1.26;
}
if(Right[i] > 0 && Left[i] == 0)
{
PPP[i].Hsml /= 1.26;
}
}
if(PPP[i].Hsml < All.MinGasHsml)
PPP[i].Hsml = All.MinGasHsml;
}
else
P[i].TimeBin = -P[i].TimeBin - 1; /* Mark as inactive */
/* CECILIA */
if(iter == MAXITER)
{
int old_type;
old_type = P[i].Type;
P[i].Type = 4; /* no SN mark any more */
PPP[i].n.NumNgb = 0;
printf("part=%d of type=%d was assigned NumNgb=%g and type=%d\n", i, old_type,
PPP[i].n.NumNgb, P[i].Type);
}
}
}
sumup_large_ints(1, &npleft, &ntot);
if(ntot > 0)
{
iter++;
if(iter > 0 && ThisTask == 0)
{
printf("ngb iteration %d: need to repeat for %d%09d particles.\n", iter,
(int) (ntot / 1000000000), (int) (ntot % 1000000000));
fflush(stdout);
}
if(iter > MAXITER)
{
#ifndef CS_FEEDBACK
printf("failed to converge in neighbour iteration in update_weights \n");
fflush(stdout);
endrun(1155);
#else
/* CECILIA */
if(Flag_phase == 2) /* HOT */
{
printf("Not enough hot neighbours for energy/metal distribution part=%d Type=%d\n", i,
P[i].Type);
fflush(stdout);
break;
/* endrun(1156); */
}
else
{
printf("Not enough cold neighbours for energy/metal distribution part=%d Type=%d\n", i,
P[i].Type);
fflush(stdout);
break;
}
#endif
}
}
}
while(ntot > 0);
myfree(DataNodeList);
myfree(DataIndexTable);
myfree(R2ngblist);
myfree(Ngblist);
myfree(Right);
myfree(Left);
/* mark as active again */
for(i = FirstActiveParticle; i >= 0; i = NextActiveParticle[i])
if(P[i].TimeBin < 0)
P[i].TimeBin = -P[i].TimeBin - 1;
/* collect some timing information */
if(ThisTask == 0)
{
printf("... update weights phase = %d done...\n", Flag_phase);
fflush(stdout);
}
}
void cs_find_hot_neighbours(void)
{
MyFloat *Left, *Right;
int nimport;
int i, j, n, ndone_flag, dummy;
int ndone, ntot, npleft;
int iter = 0;
int ngrp, sendTask, recvTask;
int place, nexport;
double dmax1, dmax2;
double xhyd, yhel, ne, mu, energy, temp;
double a3inv;
if(All.ComovingIntegrationOn)
a3inv = 1 / (All.Time * All.Time * All.Time);
else
a3inv = 1;
/* allocate buffers to arrange communication */
Left = (MyFloat *) mymalloc(NumPart * sizeof(MyFloat));
Right = (MyFloat *) mymalloc(NumPart * sizeof(MyFloat));
Ngblist = (int *) mymalloc(NumPart * sizeof(int));
All.BunchSize =
(int) ((All.BufferSize * 1024 * 1024) / (sizeof(struct data_index) + sizeof(struct data_nodelist) +
sizeof(struct hotngbs_in) + sizeof(struct hotngbs_out) +
sizemax(sizeof(struct hotngbs_in), sizeof(struct hotngbs_out))));
DataIndexTable = (struct data_index *) mymalloc(All.BunchSize * sizeof(struct data_index));
DataNodeList = (struct data_nodelist *) mymalloc(All.BunchSize * sizeof(struct data_nodelist));
CPU_Step[CPU_MISC] += measure_time();
for(n = FirstActiveParticle; n >= 0; n = NextActiveParticle[n])
{
if(P[n].Type == 0)
{
/* select reservoir and cold phase particles */
if(P[n].EnergySN > 0 && SphP[n].d.Density * a3inv > All.PhysDensThresh * All.DensFrac_Phase)
{
xhyd = P[n].Zm[6] / P[n].Mass;
yhel = (1 - xhyd) / (4. * xhyd);
ne = SphP[n].Ne;
mu = (1 + 4 * yhel) / (1 + yhel + ne);
energy = SphP[n].Entropy * P[n].Mass / GAMMA_MINUS1 * pow(SphP[n].d.Density * a3inv, GAMMA_MINUS1); /* Total Energys */
temp = GAMMA_MINUS1 / BOLTZMANN * energy / P[n].Mass * PROTONMASS * mu;
temp *= All.UnitEnergy_in_cgs / All.UnitMass_in_g; /* Temperature in Kelvin */
if(temp < All.Tcrit_Phase)
{
Left[n] = Right[n] = 0;
if(!(SphP[n].HotHsml > 0.))
SphP[n].HotHsml = All.InitialHotHsmlFactor * PPP[n].Hsml; /* Estimation of HotHsml : ONLY first step */
P[n].Type = 10; /* temporarily mark particles of interest with this number */
}
}
}
}
/* we will repeat the whole thing for those particles where we didn't find enough neighbours */
do
{
i = FirstActiveParticle; /* beginn with this index */
do
{
for(j = 0; j < NTask; j++)
{
Send_count[j] = 0;
Exportflag[j] = -1;
}
/* do local particles and prepare export list */
for(nexport = 0; i >= 0; i = NextActiveParticle[i])
if(P[i].Type == 10 && P[i].TimeBin >= 0)
{
if(cs_hotngbs_evaluate(i, 0, &nexport, Send_count) < 0)
break;
}
#ifdef MYSORT
mysort_dataindex(DataIndexTable, nexport, sizeof(struct data_index), data_index_compare);
#else
qsort(DataIndexTable, nexport, sizeof(struct data_index), data_index_compare);
#endif
MPI_Allgather(Send_count, NTask, MPI_INT, Sendcount_matrix, NTask, MPI_INT, MPI_COMM_WORLD);
for(j = 0, nimport = 0, Recv_offset[0] = 0, Send_offset[0] = 0; j < NTask; j++)
{
Recv_count[j] = Sendcount_matrix[j * NTask + ThisTask];
nimport += Recv_count[j];
if(j > 0)
{
Send_offset[j] = Send_offset[j - 1] + Send_count[j - 1];
Recv_offset[j] = Recv_offset[j - 1] + Recv_count[j - 1];
}
}
HotNgbsGet = (struct hotngbs_in *) mymalloc(nimport * sizeof(struct hotngbs_in));
HotNgbsIn = (struct hotngbs_in *) mymalloc(nexport * sizeof(struct hotngbs_in));
/* prepare particle data for export */
for(j = 0; j < nexport; j++)
{
place = DataIndexTable[j].Index;
HotNgbsIn[j].Pos[0] = P[place].Pos[0];
HotNgbsIn[j].Pos[1] = P[place].Pos[1];
HotNgbsIn[j].Pos[2] = P[place].Pos[2];
HotNgbsIn[j].HotHsml = SphP[place].HotHsml;
HotNgbsIn[j].Entropy = SphP[place].Entropy;
memcpy(HotNgbsIn[j].NodeList,
DataNodeList[DataIndexTable[j].IndexGet].NodeList, NODELISTLENGTH * sizeof(int));
}
for(ngrp = 1; ngrp < (1 << PTask); ngrp++)
{
sendTask = ThisTask;
recvTask = ThisTask ^ ngrp;
if(recvTask < NTask)
{
if(Send_count[recvTask] > 0 || Recv_count[recvTask] > 0)
{
/* get the particles */
MPI_Sendrecv(&HotNgbsIn[Send_offset[recvTask]],
Send_count[recvTask] * sizeof(struct hotngbs_in), MPI_BYTE,
recvTask, TAG_DENS_A,
&HotNgbsGet[Recv_offset[recvTask]],
Recv_count[recvTask] * sizeof(struct hotngbs_in), MPI_BYTE,
recvTask, TAG_DENS_A, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
}
}
}
myfree(HotNgbsIn);
HotNgbsResult = (struct hotngbs_out *) mymalloc(nimport * sizeof(struct hotngbs_out));
HotNgbsOut = (struct hotngbs_out *) mymalloc(nexport * sizeof(struct hotngbs_out));
/* now do the particles that need to be exported */
for(j = 0; j < nimport; j++)
cs_hotngbs_evaluate(j, 1, &dummy, &dummy);
if(i < 0)
ndone_flag = 1;
else
ndone_flag = 0;
MPI_Allreduce(&ndone_flag, &ndone, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
/* get the result */
for(ngrp = 1; ngrp < (1 << PTask); ngrp++)
{
sendTask = ThisTask;
recvTask = ThisTask ^ ngrp;
if(recvTask < NTask)
{
if(Send_count[recvTask] > 0 || Recv_count[recvTask] > 0)
{
/* send the results */
MPI_Sendrecv(&HotNgbsResult[Recv_offset[recvTask]],
Recv_count[recvTask] * sizeof(struct hotngbs_out),
MPI_BYTE, recvTask, TAG_DENS_B,
&HotNgbsOut[Send_offset[recvTask]],
Send_count[recvTask] * sizeof(struct hotngbs_out),
MPI_BYTE, recvTask, TAG_DENS_B, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
}
}
}
/* add the result to the local particles */
for(j = 0; j < nexport; j++)
{
place = DataIndexTable[j].Index;
SphP[place].da.dDensityAvg += HotNgbsOut[j].DensitySum;
SphP[place].ea.dEntropyAvg += HotNgbsOut[j].EntropySum;
SphP[place].HotNgbNum += HotNgbsOut[j].HotNgbNum;
}
myfree(HotNgbsOut);
myfree(HotNgbsResult);
myfree(HotNgbsGet);
}
while(ndone < NTask);
/* do final operations on results */
for(i = FirstActiveParticle, npleft = 0; i >= 0; i = NextActiveParticle[i])
{
if(P[i].Type == 10 && P[i].TimeBin >= 0)
{
#ifdef FLTROUNDOFFREDUCTION
SphP[i].da.DensityAvg = FLT(SphP[i].da.dDensityAvg);
SphP[i].ea.EntropyAvg = FLT(SphP[i].ea.dEntropyAvg);
#endif
if(SphP[i].HotNgbNum > 0)
{
SphP[i].da.DensityAvg /= SphP[i].HotNgbNum;
SphP[i].ea.EntropyAvg /= SphP[i].HotNgbNum;
}
else
{
SphP[i].da.DensityAvg = 0;
SphP[i].ea.EntropyAvg = 0;
}
/* now check whether we had enough neighbours */
if(SphP[i].HotNgbNum < (All.DesNumNgb - All.MaxNumHotNgbDeviation) ||
(SphP[i].HotNgbNum > (All.DesNumNgb + All.MaxNumHotNgbDeviation)))
{
/* need to redo this particle */
npleft++;
if(Left[i] > 0 && Right[i] > 0)
if((Right[i] - Left[i]) < 1.0e-3 * Left[i])
{
/* this one should be ok */
npleft--;
P[i].TimeBin = -P[i].TimeBin - 1; /* Mark as inactive */
continue;
}
if(SphP[i].HotNgbNum < (All.DesNumNgb - All.MaxNumHotNgbDeviation))
Left[i] = DMAX(SphP[i].HotHsml, Left[i]);
else
{
if(Right[i] != 0)
{
if(SphP[i].HotHsml < Right[i])
Right[i] = SphP[i].HotHsml;
}
else
Right[i] = SphP[i].HotHsml;
}
if(Left[i] > All.MaxHotHsmlParam * PPP[i].Hsml) /* prevent us from searching too far */
{
npleft--;
P[i].TimeBin = -P[i].TimeBin - 1; /* Mark as inactive */
/* Ad-hoc definition of SAvg and RhoAvg when there are no hot neighbours */
/* Note that a minimum nunmber of hot neighbours are required for promotion, see c_enrichment.c */
if(SphP[i].HotNgbNum == 0)
{
SphP[i].da.DensityAvg = SphP[i].d.Density / 100;
SphP[i].ea.EntropyAvg = SphP[i].Entropy * 1000;
printf("WARNING: Used ad-hoc values for SAvg and RhoAvg, No hot neighbours\n");
}
continue;
}
if(iter >= MAXITER_HOT - 10)
{
printf
("i=%d task=%d ID=%d Hsml=%g Left=%g Right=%g Ngbs=%g Right-Left=%g\n pos=(%g|%g|%g)\n",
i, ThisTask, P[i].ID, SphP[i].HotHsml, Left[i], Right[i],
(float) SphP[i].HotNgbNum, Right[i] - Left[i], P[i].Pos[0], P[i].Pos[1],
P[i].Pos[2]);
fflush(stdout);
}
if(Right[i] > 0 && Left[i] > 0)
SphP[i].HotHsml = pow(0.5 * (pow(Left[i], 3) + pow(Right[i], 3)), 1.0 / 3);
else
{
if(Right[i] == 0 && Left[i] == 0)
endrun(8188); /* can't occur */
if(Right[i] == 0 && Left[i] > 0)
SphP[i].HotHsml *= 1.26;
if(Right[i] > 0 && Left[i] == 0)
SphP[i].HotHsml /= 1.26;
}
}
else
P[i].TimeBin = -P[i].TimeBin - 1; /* Mark as inactive */
}
}
MPI_Allreduce(&npleft, &ntot, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
if(ntot > 0)
{
iter++;
if(iter > 0 && ThisTask == 0)
{
printf("hotngb iteration %d: need to repeat for %d particles.\n", iter, ntot);
fflush(stdout);
}
if(iter > MAXITER_HOT)
{
printf("failed to converge in hot-neighbour iteration\n");
fflush(stdout);
endrun(1155);
}
}
}
while(ntot > 0);
myfree(DataNodeList);
myfree(DataIndexTable);
myfree(Ngblist);
myfree(Right);
myfree(Left);
for(i = FirstActiveParticle; i >= 0; i = NextActiveParticle[i])
if(P[i].Type == 10)
{
P[i].Type = 0;
/* mark as active again */
if(P[i].TimeBin < 0)
P[i].TimeBin = -P[i].TimeBin - 1;
}
CPU_Step[CPU_HOTNGBS] += measure_time();
}
/* solves the recombination equation and updates the HI and HII abundances */
void radtransfer_update_chemistry(void)
{
int i, j, n;
double inter_dt, nH, temp, alpha, a3inv, gamma, molecular_weight;
#ifdef RT_PHOTOHEATING
double dtemp, e1, sigma1, de1, rate1;
double eV_to_erg = 1.60184e-12;
#endif
#ifdef RT_COOLING
double de2, de3, de4, de5, rate2, rate3, rate4, rate5;
#endif
double a, b, c, s = 1, q;
n = 1;
inter_dt = dt / FLT(n);
if(All.ComovingIntegrationOn)
a3inv = 1 / (All.Time * All.Time * All.Time);
else
a3inv = 1.0;
/* begin substepping */
for(j = 0; j < n; j++)
for(i = 0; i < N_gas; i++)
if(P[i].Type == 0)
#ifdef RT_VAR_DT
if(SphP[i].rt_flag == 1)
#endif
{
SphP[i].n_gamma /= P[i].Mass;
/* number of photons should be positive */
if(SphP[i].n_gamma < 0)
{
printf("NEGATIVE n_gamma: %g %d %d \n", SphP[i].n_gamma, i, ThisTask);
endrun(111);
}
nH = (SphP[i].d.Density * a3inv) / (PROTONMASS / All.UnitMass_in_g * All.HubbleParam); //physical
molecular_weight = 4 / (1 + 3 * HYDROGEN_MASSFRAC); //needed to obtain correct temperatures
/* molecular_weight should equal 1 in this case, but then HYDROGEN_MASSFRAC has to be set to 1 in allvars.h */
temp = SphP[i].Entropy * pow(SphP[i].d.Density * a3inv, GAMMA_MINUS1) * molecular_weight * (PROTONMASS / All.UnitMass_in_g) /
(BOLTZMANN / All.UnitEnergy_in_cgs);
//printf("temp %g %g %g\n", temp, molecular_weight, HYDROGEN_MASSFRAC);
/* collisional ionization rate */
#ifdef RT_COLLISIONAL_IONIZATION
gamma = 5.85e-11 * pow(temp,0.5) * exp(-157809.1/temp) / (1.0 + pow(temp/1e5,0.5));
gamma *= All.UnitTime_in_s / All.UnitLength_in_cm / All.UnitLength_in_cm / All.UnitLength_in_cm;
gamma *= All.HubbleParam * All.HubbleParam;
#else
gamma=0;
#endif
/* alpha_B recombination coefficient */
alpha = 2.59e-13 * pow(temp/1e4,-0.7);
alpha *= All.UnitTime_in_s / All.UnitLength_in_cm / All.UnitLength_in_cm / All.UnitLength_in_cm;
alpha *= All.HubbleParam * All.HubbleParam;
#ifndef SEMI_IMPLICIT
/* implicit scheme for ionization*/
a = inter_dt * alpha * nH;
b = -2.0 * inter_dt * alpha * nH - inter_dt * c_light * sigma * nH * SphP[i].n_gamma - 1.0;
c = SphP[i].nHI + inter_dt * alpha * nH;
if(b > 0)
s = 1.0;
if(b < 0)
s = -1.0;
if(b == 0)
s = 0.0;
q = -0.5 * (b + s * sqrt(b * b - 4.0 * a * c));
SphP[i].nHI = c / q;
#ifdef RT_NO_RECOMBINATIONS
SphP[i].nHII += inter_dt * c_light * sigma * nH * SphP[i].n_gamma * SphP[i].nHI;
SphP[i].nHI = 1.0 - SphP[i].nHII;
#endif
if(SphP[i].nHI < 0 && SphP[i].nHI > -EPSILON)
SphP[i].nHI = 0;
if(SphP[i].nHI > 1 && SphP[i].nHI < (1+EPSILON))
SphP[i].nHI = 1;
if(SphP[i].nHI < -EPSILON || SphP[i].nHI > (1+EPSILON))
{
printf("WRONG nHII: %g %d %d \n", SphP[i].nHII, i, ThisTask);
endrun(222);
}
SphP[i].nHII = 1 - SphP[i].nHI;
SphP[i].n_elec = SphP[i].nHII;
#else
/* semi-implicit scheme for ionization*/
SphP[i].nHII = (SphP[i].nHII + inter_dt * c_light * sigma * nH * SphP[i].n_gamma + inter_dt * gamma * nH * SphP[i].n_elec) /
(1.0 + inter_dt * c_light * sigma * nH * SphP[i].n_gamma +
inter_dt * alpha * nH * SphP[i].n_elec + inter_dt * gamma * nH * SphP[i].n_elec);
/* fraction should be between 0 and 1 */
if(SphP[i].nHII < 0 || SphP[i].nHII > 1)
{
printf("WRONG nHII: %g %d %d \n", SphP[i].nHII, i, ThisTask);
endrun(222);
}
SphP[i].nHI = 1 - SphP[i].nHII;
SphP[i].n_elec = SphP[i].nHII;
#endif
#ifdef RT_PHOTOHEATING
/*photoheating for 10^5K blackbody*/
sigma1 = 1.63e-18 / All.UnitLength_in_cm / All.UnitLength_in_cm; //code units
sigma1 *= All.HubbleParam * All.HubbleParam;
e1 = 29.65 * eV_to_erg; //real units
/* all rates in erg cm^3 s^-1 in code units and energy in ergs*/
/* photoheating rate */
rate1 = c_light * e1 * sigma1;
de1 = SphP[i].nHI * nH * SphP[i].n_gamma * nH * rate1;
dtemp = de1 * GAMMA_MINUS1 / nH / BOLTZMANN * inter_dt;
SphP[i].Entropy += dtemp * (BOLTZMANN / All.UnitEnergy_in_cgs) / pow(SphP[i].d.Density * a3inv, GAMMA_MINUS1) /
molecular_weight / (PROTONMASS / All.UnitMass_in_g);
#endif
#ifdef RT_COOLING
/* recombination cooling rate */
rate2 = 8.7e-27 * pow(temp,0.5) * pow(temp/1e3,-0.2) / (1.0 + pow(temp/1e6,0.7));
rate2 *= All.UnitTime_in_s / All.UnitLength_in_cm / All.UnitLength_in_cm / All.UnitLength_in_cm;
rate2 *= All.HubbleParam * All.HubbleParam;
/* collisional ionization cooling rate */
rate3 = 1.27e-21 * pow(temp,0.5) * exp(-157809.1/temp) / (1.0 + pow(temp/1e5,0.5));
rate3 *= All.UnitTime_in_s / All.UnitLength_in_cm / All.UnitLength_in_cm / All.UnitLength_in_cm;
rate3 *= All.HubbleParam * All.HubbleParam;
/* collisional excitation cooling rate */
rate4 = 7.5e-19 * (1.0 + pow(temp/1e5,0.5)) * exp(-118348/temp);
rate4 *= All.UnitTime_in_s / All.UnitLength_in_cm / All.UnitLength_in_cm / All.UnitLength_in_cm;
rate4 *= All.HubbleParam * All.HubbleParam;
/* Bremsstrahlung cooling rate */
rate5 = 1.42e-27 * 1.3 * pow(temp,0.5);
rate5 *= All.UnitTime_in_s / All.UnitLength_in_cm / All.UnitLength_in_cm / All.UnitLength_in_cm;
rate5 *= All.HubbleParam * All.HubbleParam;
de2 = SphP[i].nHII * nH * SphP[i].n_elec * nH * rate2;
de3 = SphP[i].nHI * nH * SphP[i].n_elec * nH * rate3;
de4 = SphP[i].nHI * nH * SphP[i].n_elec * nH * rate4;
de5 = SphP[i].nHII * nH * SphP[i].n_elec * nH * rate5;
dtemp = (de2 + de3 + de4 + de5) * GAMMA_MINUS1 / nH / BOLTZMANN * inter_dt;
SphP[i].Entropy -= dtemp * (BOLTZMANN / All.UnitEnergy_in_cgs) / pow(SphP[i].d.Density * a3inv, GAMMA_MINUS1) /
molecular_weight / (PROTONMASS / All.UnitMass_in_g);
#endif
SphP[i].n_gamma *= P[i].Mass;
}
}
int main()
{
fs_hash_init(FS_HASH_UMAC);
int passes = 0;
int fails = 0;
#if 1
#define DUMP(t) printf("%lld %s\n", fs_c.xsd_##t, #t);
DUMP(double);
DUMP(float);
DUMP(decimal);
DUMP(integer);
DUMP(boolean);
#endif
#define TEST1(f, a, x) printf("[%s] ", fs_value_equal(f(NULL, a), x) ? "PASS" : "FAIL"); printf("%s(", #f); fs_value_print(a); printf(") = "); fs_value_print(f(NULL, a)); printf("\n"); if (!fs_value_equal(f(NULL, a), x)) { fails++; printf(" should have been "); fs_value_print(x); printf("\n"); } else { passes++; }
#define TEST2(f, a, b, x) printf("[%s] ", fs_value_equal(f(NULL, a, b), x) ? "PASS" : "FAIL"); printf("%s(", #f); fs_value_print(a); printf(", "); fs_value_print(b); printf(") = "); fs_value_print(f(NULL, a, b)); printf("\n"); if (!fs_value_equal(f(NULL, a, b), x)) { fails++; printf(" should have been "); fs_value_print(x); printf("\n"); } else { passes++; }
#define TEST3(f, a, b, c, x) printf("[%s] ", fs_value_equal(f(NULL, a, b, c), x) ? "PASS" : "FAIL"); printf("%s(", #f); fs_value_print(a); printf(", "); fs_value_print(b); printf(", "); fs_value_print(c); printf(") = "); fs_value_print(f(NULL, a, b, c)); printf("\n"); if (!fs_value_equal(f(NULL, a, b, c), x)) { fails++; printf(" should have been "); fs_value_print(x); printf("\n"); } else { passes++; }
#define URI(x) fs_value_uri(x)
#define RID(x) fs_value_rid(x)
#define BND(x) fs_value_rid(0x8000000000000000LL | x)
#define STR(x) fs_value_string(x)
#define PLN(x) fs_value_plain(x)
#define PLN_L(x, l) fs_value_plain_with_lang(x, l)
#define DBL(x) fs_value_double(x)
#define FLT(x) fs_value_float(x)
#define DEC(x) fs_value_decimal(x)
#define INT(x) fs_value_integer(x)
#define BLN(x) fs_value_boolean(x)
#define DAT(x) fs_value_datetime(x)
#define DAT_S(x) fs_value_datetime_from_string(x)
#define ERR() fs_value_error(FS_ERROR_INVALID_TYPE, NULL)
#define BLK() fs_value_blank()
TEST2(fn_numeric_add, BLK(), BLK(), ERR());
TEST2(fn_numeric_add, INT(1), URI("test:"), ERR());
TEST2(fn_numeric_add, STR("2"), INT(3), ERR());
TEST2(fn_numeric_add, INT(2), INT(3), INT(5));
TEST2(fn_numeric_add, DEC(2.5), DEC(-1), DEC(1.5));
TEST2(fn_numeric_add, DEC(-2.5), DEC(-1.5), DEC(-4));
TEST2(fn_numeric_subtract, INT(2), INT(3), INT(-1));
TEST2(fn_numeric_add, DBL(1), INT(2), DBL(3));
TEST2(fn_numeric_add, FLT(17000), INT(2), FLT(17002));
TEST2(fn_numeric_subtract, FLT(17000), INT(2), FLT(16998));
TEST2(fn_numeric_subtract, DEC(17000.5), INT(2), DEC(16998.5));
TEST2(fn_numeric_multiply, URI("http://example.com/"), INT(2), ERR());
TEST2(fn_numeric_multiply, DEC(3.5), INT(2), DEC(7));
TEST2(fn_numeric_multiply, DEC(35), DBL(-0.1), DBL(-3.5));
TEST2(fn_numeric_multiply, INT(1), FLT(23), FLT(23));
TEST2(fn_numeric_multiply, INT(10), INT(23), INT(230));
TEST2(fn_numeric_divide, INT(9), URI("http://example.org/"), ERR());
TEST2(fn_numeric_divide, INT(9), INT(2), DEC(4.5));
TEST2(fn_numeric_divide, INT(-9), INT(2), DEC(-4.5));
TEST2(fn_numeric_divide, INT(-9), INT(-2), DEC(4.5));
TEST2(fn_numeric_divide, DBL(90), FLT(10), DBL(9));
TEST2(fn_numeric_divide, FLT(99), FLT(-10), FLT(-9.9));
TEST2(fn_equal, URI("http://example.com/"), DEC(23), BLN(0));
TEST2(fn_equal, URI("http://example.com/"), URI("http://example.com/"), BLN(1));
TEST2(fn_equal, INT(23), DEC(23), BLN(1));
TEST2(fn_equal, fs_value_decimal_from_string("-23.0"), DEC(-23), BLN(1));
TEST2(fn_equal, STR("foo"), PLN("foo"), BLN(0));
TEST2(fn_equal, PLN("foo"), PLN("foo"), BLN(1));
TEST2(fn_equal, BLN(0), BLN(0), BLN(1));
TEST2(fn_greater_than, STR("BBB"), STR("AAA"), BLN(1));
TEST2(fn_greater_than, PLN("BBB"), PLN("AAA"), BLN(1));
TEST2(fn_greater_than, PLN("AAA"), PLN("BBB"), BLN(0));
TEST2(fn_less_than, PLN("BBB"), PLN("AAA"), BLN(0));
TEST2(fn_less_than, PLN("AAA"), PLN("BBB"), BLN(1));
TEST2(fn_less_than, INT(20), INT(15), BLN(0));
TEST2(fn_numeric_equal, INT(23), INT(23), BLN(1));
TEST2(fn_numeric_equal, INT(23), DEC(23), BLN(1));
TEST2(fn_numeric_equal, INT(23), FLT(23), BLN(1));
TEST2(fn_numeric_equal, INT(23), DBL(23), BLN(1));
TEST2(fn_numeric_equal, fn_minus(NULL, INT(23)), DBL(-23), BLN(1));
TEST2(fn_datetime_equal, DAT(1000), DAT(1000), BLN(1));
TEST2(fn_datetime_equal, DAT(time(NULL)), DAT(1000), BLN(0));
TEST2(fn_numeric_less_than, INT(0), DBL(23), BLN(1));
TEST2(fn_numeric_less_than, INT(23), DBL(23), BLN(0));
TEST2(fn_numeric_less_than, DBL(22.99999), INT(23), BLN(1));
TEST2(fn_numeric_less_than, DEC(-18.51), DEC(-18.5), BLN(1));
TEST2(fn_numeric_less_than, DBL(-18.51), DBL(-18.5), BLN(1));
TEST2(fn_numeric_less_than, DEC(-18.5), DEC(-18.51), BLN(0));
TEST2(fn_numeric_less_than, DBL(-18.5), DBL(-18.51), BLN(0));
TEST2(fn_numeric_greater_than, DEC(-121.98882), DEC(-121.739856), BLN(0));
TEST2(fn_numeric_greater_than, DEC(37.67473), DEC(37.677954), BLN(0));
TEST2(fn_numeric_greater_than, INT(0), DBL(23), BLN(0));
TEST2(fn_numeric_greater_than, INT(23), DBL(23), BLN(0));
TEST2(fn_numeric_greater_than, DBL(22.99999), INT(23), BLN(0));
TEST2(fn_numeric_greater_than, DEC(-18.51), DEC(-18.5), BLN(0));
TEST2(fn_numeric_greater_than, DBL(-18.51), DBL(-18.5), BLN(0));
TEST2(fn_numeric_greater_than, DEC(-18.5), DEC(-18.51), BLN(1));
TEST2(fn_numeric_greater_than, DBL(-18.5), DBL(-18.51), BLN(1));
TEST2(fn_logical_and, BLN(1), BLN(0), BLN(0));
TEST2(fn_logical_and, BLN(1), BLN(1), BLN(1));
TEST2(fn_logical_and, INT(0), INT(1), BLN(0));
TEST2(fn_logical_and, INT(1), INT(1), BLN(1));
TEST2(fn_logical_and, STR("true"), INT(1), BLN(1));
TEST2(fn_logical_and, STR("false"), INT(1), BLN(1));
TEST2(fn_logical_and, INT(1), ERR(), ERR());
TEST2(fn_logical_and, ERR(), INT(1), ERR());
TEST2(fn_logical_and, INT(0), ERR(), BLN(0));
TEST2(fn_logical_and, ERR(), INT(0), BLN(0));
TEST2(fn_logical_and, ERR(), ERR(), ERR());
TEST2(fn_logical_or, BLN(1), BLN(0), BLN(1));
TEST2(fn_logical_or, BLN(1), BLN(1), BLN(1));
TEST2(fn_logical_or, INT(0), INT(1), BLN(1));
TEST2(fn_logical_or, INT(1), INT(1), BLN(1));
TEST2(fn_logical_or, STR("true"), INT(32), BLN(1));
TEST2(fn_logical_or, STR("false"), INT(1), BLN(1));
TEST2(fn_logical_or, INT(1), ERR(), BLN(1));
TEST2(fn_logical_or, ERR(), INT(1), BLN(1));
TEST2(fn_logical_or, INT(0), ERR(), ERR());
TEST2(fn_logical_or, ERR(), INT(0), ERR());
TEST2(fn_logical_or, ERR(), ERR(), ERR());
TEST2(fn_compare, STR("AAA"), STR("BBB"), INT(-1));
TEST2(fn_compare, STR("BBB"), STR("BBB"), INT(0));
TEST2(fn_compare, STR("BBB"), STR("AAA"), INT(1));
TEST2(fn_compare, PLN("BBB"), PLN("AAA"), INT(1));
TEST2(fn_compare, STR("BBB"), PLN("BBB"), ERR());
TEST2(fn_compare, STR("http://example.com/"), URI("http://example.com/"), ERR());
TEST2(fn_compare, URI("http://example.com/"), URI("http://example.com/"), ERR());
TEST2(fn_compare, INT(1), PLN("BBB"), ERR());
TEST3(fn_matches, PLN("foobar"), PLN("foo"), BLK(), BLN(1));
TEST3(fn_matches, PLN("foobar"), PLN("^foo"), BLK(), BLN(1));
TEST3(fn_matches, PLN("foobar"), PLN("bar$"), BLK(), BLN(1));
TEST3(fn_matches, PLN("foobar"), PLN("BAR"), STR("i"), BLN(1));
TEST3(fn_matches, PLN("foobar"), PLN("^FOOB[AO]R$"), STR("i"), BLN(1));
TEST3(fn_matches, PLN("foobar"), PLN("^bar"), BLK(), BLN(0));
TEST3(fn_matches, PLN("foobar"), PLN("^foo$"), BLK(), BLN(0));
TEST3(fn_matches, PLN("foobar"), PLN("foo bar"), PLN("x"), BLN(1));
TEST3(fn_matches, INT(23), PLN("foo bar"), PLN("x"), ERR());
TEST3(fn_matches, PLN("foobar"), DAT(1000), PLN("x"), ERR());
TEST1(fn_bound, URI("http://example.com/"), BLN(1));
TEST1(fn_bound, STR("http"), BLN(1));
TEST1(fn_bound, PLN(""), BLN(1));
TEST1(fn_bound, BND(100), BLN(1));
TEST1(fn_bound, RID(FS_RID_NULL), BLN(0));
TEST1(fn_is_blank, URI("http://example.com/"), BLN(0));
TEST1(fn_is_blank, STR("http"), BLN(0));
TEST1(fn_is_blank, PLN(""), BLN(0));
TEST1(fn_is_blank, BND(100), BLN(1));
TEST1(fn_is_blank, RID(FS_RID_NULL), BLN(0));
TEST1(fn_is_iri, URI("http://example.com/"), BLN(1));
TEST1(fn_is_iri, STR("http"), BLN(0));
TEST1(fn_is_iri, PLN(""), BLN(0));
TEST1(fn_is_iri, BND(100), BLN(0));
TEST1(fn_is_iri, RID(FS_RID_NULL), BLN(0));
TEST1(fn_is_literal, URI("http://example.com/"), BLN(0));
TEST1(fn_is_literal, STR("http"), BLN(1));
TEST1(fn_is_literal, PLN(""), BLN(1));
TEST1(fn_is_literal, BND(100), BLN(0));
TEST1(fn_is_literal, RID(FS_RID_NULL), BLN(0));
TEST1(fn_str, URI("http://example.com/"), PLN("http://example.com/"));
TEST1(fn_str, STR("http"), PLN("http"));
TEST1(fn_str, PLN(""), PLN(""));
TEST1(fn_str, BLN(1), PLN("true"));
TEST1(fn_str, INT(1), PLN("1"));
TEST1(fn_str, FLT(11.1), PLN("11.100000"));
TEST1(fn_str, DBL(11.1), PLN("11.100000"));
TEST1(fn_str, DEC(23), PLN("23.000000"));
TEST1(fn_str, DAT(1000), PLN("1970-01-01T00:16:40"));
TEST1(fn_str, BND(100), ERR());
TEST1(fn_str, RID(FS_RID_NULL), ERR());
TEST1(fn_lang, PLN("foo"), PLN(""));
TEST1(fn_lang, STR("foo"), PLN(""));
TEST1(fn_lang, PLN_L("foo", "en"), PLN("en"));
TEST1(fn_lang, PLN_L("foo", ""), PLN(""));
TEST1(fn_lang, BLN(1), PLN(""));
TEST1(fn_lang, INT(1), PLN(""));
TEST1(fn_lang, DEC(1.1), PLN(""));
TEST1(fn_lang, FLT(1.1), PLN(""));
TEST1(fn_lang, DBL(1.1), PLN(""));
TEST1(fn_lang, URI("http://example.org/"), ERR());
TEST1(fn_datatype, PLN("foo"), ERR());
TEST1(fn_datatype, STR("foo"), URI(XSD_STRING));
TEST1(fn_datatype, PLN_L("foo", "en"), ERR());
TEST1(fn_datatype, PLN_L("foo", ""), ERR());
TEST1(fn_datatype, BLN(1), URI(XSD_BOOLEAN));
TEST1(fn_datatype, INT(1), URI(XSD_INTEGER));
TEST1(fn_datatype, DEC(1.1), URI(XSD_DECIMAL));
TEST1(fn_datatype, FLT(1.1), URI(XSD_FLOAT));
TEST1(fn_datatype, DBL(1.1), URI(XSD_DOUBLE));
TEST1(fn_datatype, DAT(1000), URI(XSD_DATETIME));
TEST1(fn_datatype, URI("http://example.org/"), ERR());
TEST2(fn_equal, DAT_S("1975-01-07"), DAT(158284800), BLN(1));
TEST2(fn_equal, DAT_S("1975-01-07T00:00:01"), DAT(158284801), BLN(1));
TEST2(fn_equal, DAT_S("1975-01-07T01:00:01+0000"), DAT(158288401), BLN(1));
TEST2(fn_equal, DAT_S("1975-01-07T01:00:01-0900"), DAT(158320801), BLN(1));
TEST2(fn_cast, DBL(2.1), URI(XSD_INTEGER), INT(2));
TEST2(fn_cast, PLN("2.23"), URI(XSD_DOUBLE), DBL(2.23));
TEST2(fn_cast, PLN_L("2.23", "en"), URI(XSD_DOUBLE), DBL(2.23));
TEST2(fn_cast, STR("2.23"), URI(XSD_DOUBLE), DBL(2.23));
TEST2(fn_cast, URI("http://example.com/"), URI(XSD_STRING), ERR());
TEST2(fn_cast, DAT(1000), URI(XSD_STRING), STR("1970-01-01T00:16:40"));
TEST2(fn_cast, STR("1975-01-07T01:00:01-0900"), URI(XSD_DATETIME), DAT(158320801));
printf("\n=== pass %d, fail %d\n", passes, fails);
if (fails) {
return 1;
}
return 0;
}
// Viterbi search
void WFSADecoder::viterbi(float *fFeatureVectors, unsigned int iFeatureVectors) {
double dTimeBegin = TimeUtils::getTimeMilliseconds();
float fScore = 0.0;
HMMStateDecoding *hmmStateDecoding;
m_iFeatureVectors = iFeatureVectors;
m_hmmManager->resetHMMEmissionProbabilityComputation();
m_activeStateTable->beginUtterance();
unsigned int iLexUnitEndSentence = m_lexiconManager->m_lexUnitEndSentence->iLexUnit;
// create the initial set of active states (states coming from the initial state and non-epsilon arcs)
StateX *stateInitial = m_wfsAcceptor->getInitialState();
m_activeStateTable->activateStateInitial(stateInitial);
TransitionX *transition = NULL;
TransitionX *transitionEnd = NULL;
TransitionX *transitionAux = NULL;
ActiveState *activeStatesCurrent = NULL;
unsigned int iActiveStatesCurrent = 0;
//m_activeStateTable->printSubgraph(stateInitial);
//exit(-1);
// process the feature vectors
for(unsigned int t=0 ; t < iFeatureVectors ; ++t) {
m_fScoreBest = -FLT_MAX;
#ifdef SIMD
float *fFeatureVector = &(fFeatureVectors[t*FEATURE_VECTOR_LENGTH_ALIGNED_16]);
#else
float *fFeatureVector = &(fFeatureVectors[t*FEATURE_VECTOR_LENGTH]);
#endif
activeStatesCurrent = m_activeStateTable->getActiveStatesCurrent(&iActiveStatesCurrent);
m_activeStateTable->m_iStatesExpanded = 0;
m_activeStateTable->m_iStatesActivated = 0;
m_activeStateTable->m_iStatesPruned = 0;
// (1) process all the active states (these are non-epsilon states)
for(unsigned int i = 0 ; i < iActiveStatesCurrent ; ++i) {
// skip pruned states
if (activeStatesCurrent[i].state == NULL) {
continue;
}
bool bExpanded = false;
++m_activeStateTable->m_iStatesExpanded;
ActiveState &activeState = activeStatesCurrent[i];
// (1.1) self-loop (this is a simulated transition)
// compute emission probability
#ifdef SIMD
fScore = activeState.hmmStateDecoding->computeEmissionProbabilityNearestNeighborSIMD(fFeatureVector,t);
#else
fScore = activeState.hmmStateDecoding->computeEmissionProbabilityNearestNeighborPDE(fFeatureVector,t);
#endif
// preventive pruning goes here
if (activeState.fScore+fScore < (m_fScoreBest-m_fPruningLikelihood)) {
continue;
}
// lattice generation
int iWGToken = -1;
if (m_bLatticeGeneration) {
assert(activeState.iWGToken != -1);
iWGToken = m_activeStateTable->newWGToken(activeState.iWGToken);
WGToken *wgToken = iWGToken+m_activeStateTable->m_wgTokens;
// update the scores
for(int i=0 ; ((i < m_activeStateTable->m_iMaxWordSequencesState) && (wgToken[i].iWordSequence != -1)) ; ++i) {
wgToken[i].fScore += fScore;
}
}
// activate the state
m_activeStateTable->activateState(activeState.state,activeState.fScore+fScore,&m_fScoreBest,
activeState.hmmStateDecoding,activeState.iHistoryItem,iWGToken,0.0);
// (1.2) standard transitions
// get the first and last transitions from the state
transition = *activeState.state;
transitionEnd = *(activeState.state+1);
assert(transition <= transitionEnd);
unsigned int iTransitions = 0;
TransitionX *transitions = m_wfsAcceptor->getTransitions(&iTransitions);
assert(transition < (transitions+iTransitions));
assert(transitionEnd <= (transitions+iTransitions));
while(transition != transitionEnd) {
assert(transition != NULL);
// epsilon-transition
if (transition->iSymbol & EPSILON_TRANSITION) {
// lattice generation
int iWGToken = -1;
if (m_bLatticeGeneration) {
assert(activeState.iWGToken != -1);
iWGToken = m_activeStateTable->newWGToken(activeState.iWGToken);
WGToken *wgToken = iWGToken+m_activeStateTable->m_wgTokens;
// update the scores
for(int i=0 ; ((i < m_activeStateTable->m_iMaxWordSequencesState) && (wgToken[i].iWordSequence != -1)) ; ++i) {
wgToken[i].fScore += transition->fWeight;
}
}
// activate the state
m_activeStateTable->activateStateEpsilon(transition->state,activeState.fScore+transition->fWeight,
activeState.iHistoryItem,iWGToken,0.0);
}
// fake-transition
else if (transition->iSymbol & FAKE_TRANSITION) {
//bool bStop = true;
}
// lexical-unit transition
else if (transition->iSymbol & LEX_UNIT_TRANSITION) {
// create a new history item
int iHistoryItem = m_activeStateTable->newHistoryItem();
HistoryItem *historyItem = m_activeStateTable->m_historyItems+iHistoryItem;
historyItem->iPrev = activeState.iHistoryItem;
historyItem->iLexUnitPron = transition->iSymbol & LEX_UNIT_TRANSITION_COMPLEMENT;
historyItem->iEndFrame = t-1;
historyItem->fScore = activeState.fScore+transition->fWeight;
historyItem->iActive = -1;
// lattice generation
int iWGToken = -1;
int iWordSequence = -1;
if (m_bLatticeGeneration) {
// keep the N word sequences arriving to the state
assert(activeState.iWGToken != -1);
historyItem->iWGToken = activeState.iWGToken;
// checks
for(int i=0 ; i < m_iMaxWordSequencesState ; ++i) {
if ((historyItem->iWGToken+m_activeStateTable->m_wgTokens)[i].iWordSequence == -1) {
break;
}
bool bFlag = false;
if (m_activeStateTable->m_mWGTokenDeleted.find(historyItem->iWGToken+m_activeStateTable->m_wgTokens) != m_activeStateTable->m_mWGTokenDeleted.end()) {
//printf("%x !!\n",historyItem->iWGToken+m_activeStateTable->m_wgTokens);
bFlag = true;
}
assert(historyItem->iEndFrame > (m_activeStateTable->m_historyItems+ (historyItem->iWGToken+m_activeStateTable->m_wgTokens)[i].iHistoryItem)->iEndFrame);
}
// generate a new hash value for the new word sequence
iWordSequence = m_activeStateTable->hashWordSequence(historyItem);
}
if ((transition->iSymbol & LEX_UNIT_TRANSITION_COMPLEMENT) == iLexUnitEndSentence) {
//bool bStop = true;
}
for(transitionAux = *(transition->state) ; transitionAux != *(transition->state+1) ; ++transitionAux) {
// epsilon-transition
if (transitionAux->iSymbol & EPSILON_TRANSITION) {
// preventive pruning
if (activeState.fScore+transition->fWeight+transitionAux->fWeight < (m_fScoreBest-m_fPruningLikelihood)) {
continue;
}
// lattice generation
if (m_bLatticeGeneration) {
bExpanded = true;
iWGToken = m_activeStateTable->newWGToken(iWordSequence,
activeState.fScore+transition->fWeight+transitionAux->fWeight,iHistoryItem);
}
// activate the state
m_activeStateTable->activateStateEpsilon(transitionAux->state,
activeState.fScore+transition->fWeight+transitionAux->fWeight,iHistoryItem,iWGToken,0.0);
}
// fake transition
else if (transitionAux->iSymbol & FAKE_TRANSITION) {
//bool bStop = true;
}
// lex-unit transition (end-of-sentence)
else if (transitionAux->iSymbol & LEX_UNIT_TRANSITION) {
if ((transitionAux->iSymbol & LEX_UNIT_TRANSITION_COMPLEMENT) == iLexUnitEndSentence) {
//bool bStop = true;
}
}
// leaf-transition
else {
// compute emission probability
hmmStateDecoding = &m_hmmStatesDecoding[transitionAux->iSymbol];
#ifdef SIMD
fScore = hmmStateDecoding->computeEmissionProbabilityNearestNeighborSIMD(fFeatureVector,t);
#else
fScore = hmmStateDecoding->computeEmissionProbabilityNearestNeighborPDE(fFeatureVector,t);
#endif
// preventive pruning
if (activeState.fScore+transition->fWeight+transitionAux->fWeight+fScore < (m_fScoreBest-m_fPruningLikelihood)) {
continue;
}
// lattice generation
if (m_bLatticeGeneration) {
bExpanded = true;
iWGToken = m_activeStateTable->newWGToken(iWordSequence,
activeState.fScore+transition->fWeight+transitionAux->fWeight+fScore,iHistoryItem);
}
// activate the state
m_activeStateTable->activateState(transitionAux->state,
activeState.fScore+transition->fWeight+transitionAux->fWeight+fScore,
&m_fScoreBest,hmmStateDecoding,iHistoryItem,iWGToken,0.0);
//printf("new lexical unit: %f\n",activeState.fScore+transition->fWeight+transitionAux->fWeight+fScore);
}
}
}
// leaf-transition
else {
// compute emission probability
hmmStateDecoding = &m_hmmStatesDecoding[transition->iSymbol];
#ifdef SIMD
fScore = hmmStateDecoding->computeEmissionProbabilityNearestNeighborSIMD(fFeatureVector,t);
#else
fScore = hmmStateDecoding->computeEmissionProbabilityNearestNeighborPDE(fFeatureVector,t);
#endif
// preventive pruning goes here
if (activeState.fScore+transition->fWeight+fScore < (m_fScoreBest-m_fPruningLikelihood)) {
transition++;
continue;
}
// lattice generation
int iWGToken = -1;
if (m_bLatticeGeneration) {
assert(activeState.iWGToken != -1);
iWGToken = m_activeStateTable->newWGToken(activeState.iWGToken);
WGToken *wgToken = iWGToken+m_activeStateTable->m_wgTokens;
// update the scores
for(int i=0 ; ((i < m_activeStateTable->m_iMaxWordSequencesState) && (wgToken[i].iWordSequence != -1)) ; ++i) {
wgToken[i].fScore += transition->fWeight+fScore;
}
}
// activate the state
m_activeStateTable->activateState(transition->state,activeState.fScore+transition->fWeight+fScore,
&m_fScoreBest,hmmStateDecoding,activeState.iHistoryItem,iWGToken,0.0);
}
transition++;
}
}
//printf("# active states before processing epsilon-transitions: %d\n",);
// sanity check
if (m_bLatticeGeneration) {
// active nodes of next time frame
ActiveState *activeState = m_activeStateTable->m_activeStatesNext;
while(activeState != m_activeStateTable->m_activeStateAvailable) {
// make sure the WGToken structure keeps the historyItem
if (activeState->iWGToken != -1) {
assert(activeState->iHistoryItem != -1);
WGToken *wgToken = activeState->iWGToken+m_activeStateTable->m_wgTokens;
bool bFound = false;
for(int i=0 ; ((i < m_activeStateTable->m_iMaxWordSequencesState) && (wgToken[i].iWordSequence != -1)) ; ++i) {
if (wgToken[i].iHistoryItem == activeState->iHistoryItem) {
bFound = true;
}
}
assert(bFound);
}
++activeState;
}
}
// (3) process epsilon transitions in topological order
m_activeStateTable->processEpsilonTransitions(fFeatureVector,&m_fScoreBest);
if (t != iFeatureVectors-1) {
// (4) apply beam pruning
m_activeStateTable->beamPruning(&m_fScoreBest);
// move to the next time frame
m_activeStateTable->nextTimeFrame();
} else {
cout << "t=" << setw(5) << t << " bestScore: " << FLT(14,6) << m_fScoreBest << endl;
}
}
//m_activeStateTable->printInfo();
m_activeStateTable->endUtterance();
double dTimeEnd = TimeUtils::getTimeMilliseconds();
double dTimeSeconds = (dTimeEnd-dTimeBegin)/1000.0;
BVC_VERB << "decoding time: " << FLT(8,2) << dTimeSeconds << " seconds (RTF: " <<
FLT(5,2) << dTimeSeconds/(((float)iFeatureVectors)/100.0) << ")";
}
