Changeset 4752
- Timestamp:
- 08/27/04 11:36:24 (20 years ago)
- Location:
- trunk/MagicSoft/Mars
- Files:
-
- 2 added
- 12 edited
Legend:
- Unmodified
- Added
- Removed
-
trunk/MagicSoft/Mars/Changelog
r4748 r4752 19 19 20 20 -*-*- END OF LINE -*-*- 21 2004/08/27: Thomas Bretz 22 23 * Makefile: 24 - added comments how to link statically 25 26 * callisto.cc: 27 - fixed some output 28 29 * mbadpixels/Makefile: 30 - added a comment 31 32 * mbase/BaseLinkDef.h, mbase/Makefile: 33 - added MArrayI 34 35 * mbase/MArrayI.[h,cc]: 36 - added 37 38 * mbase/MArrayD.cc: 39 - fixed some comments 40 41 * mcalib/MCalibrateData.[h,cc]: 42 - unified CalibratePedestal and CalibrateData. Calling GetConvFactor twice 43 took a lot of time. 44 45 * mjobs/MJCalibrateSignal.cc, mjobs/MJPedestal.cc: 46 - added two empty lines to output if finished 47 48 * mpedestal/MPedPhotCam.cc: 49 - use faster MArrays in ReCalc 50 - accelerated GetPixelContent 51 52 * msignal/MExtractTimeFastSpline.cc: 53 - accelerated a bit by defining 54 Float_t higainklo = fHiGainSecondDeriv[klo]; 55 Float_t higainkhi = fHiGainSecondDeriv[khi]; 56 instead of accesing the arrays many times inside the loops. 57 Somebody should do the same for logain. 58 59 60 21 61 2004/08/26: Wolfgang Wittek 22 62 -
trunk/MagicSoft/Mars/Makefile
r4722 r4752 106 106 @echo " Linking $@ ..." 107 107 $(CXX) $(CXXFLAGS) $(ROOTGLIBS) $(SOLIB) $@.o $(MARS_LIB) -o $@ 108 109 # Use this to link the programs statically - for gprof 110 #$(PROGRAMS): $(OBJS) $(HEADERS) $(PROGRAMS:=.o) 111 # @echo " Linking $@ ..." 112 # $(CXX) $(CXXFLAGS) $(ROOTGLIBS) $(OBJS) $(SUBDIRS:=/*.o) $@.o $(MARS_LIB) -o $@ 108 113 else 109 114 $(DYLIB): $(LIBRARIES) $(OBJS) $(HEADERS) -
trunk/MagicSoft/Mars/callisto.cc
r4748 r4752 325 325 return 1; 326 326 } 327 328 gLog << flush << all << endl << endl;329 327 } 330 328 … … 380 378 return 1; 381 379 } 382 383 gLog << flush << all << endl << endl;384 380 } 385 381 -
trunk/MagicSoft/Mars/mbase/BaseLinkDef.h
r4747 r4752 60 60 #pragma link C++ class MArrayS; 61 61 #pragma link C++ class MArrayD; 62 #pragma link C++ class MArrayI; 62 63 63 64 #pragma link C++ class MTime+; -
trunk/MagicSoft/Mars/mbase/MArrayD.cc
r4747 r4752 26 26 ////////////////////////////////////////////////////////////////////////////// 27 27 // 28 // MArray S28 // MArrayD 29 29 // 30 // Array of UShort_t. It is almost the same than TArrayD but derives from30 // Array of Double_t. It is almost the same than TArrayD but derives from 31 31 // TObject 32 32 // -
trunk/MagicSoft/Mars/mbase/Makefile
r4747 r4752 52 52 MArrayS.cc \ 53 53 MArrayD.cc \ 54 MArrayI.cc \ 54 55 MTime.cc \ 55 56 MClone.cc \ -
trunk/MagicSoft/Mars/mcalib/MCalibrateData.cc
r4639 r4752 265 265 } 266 266 267 // Sizes might have changed 268 if (fPedestalFlag && (Int_t)fPedestal->GetSize() != fSignals->GetSize()) 269 { 270 *fLog << err << "Size mismatch of MPedestalCam and MCalibrationCam... abort." << endl; 271 return kFALSE; 272 } 273 267 274 if(fCalibrationMode == kBlindPixel && !fQEs->IsBlindPixelMethodValid()) 268 275 { … … 315 322 316 323 if (TestPedestalFlag(kRun)) 317 if (!CalibratePedestal()) 318 return kFALSE; 324 Calibrate(kFALSE, kTRUE); 319 325 320 326 return kTRUE; … … 323 329 // -------------------------------------------------------------------------- 324 330 // 325 // Calibrate the Pedestal values326 //327 Bool_t MCalibrateData::CalibratePedestal()328 {329 //330 // Fill MPedPhot container using the informations from331 // MPedestalCam, MExtractedSignalCam and MCalibrationCam332 //333 const Float_t slices = fSignals->GetNumUsedHiGainFADCSlices();334 const Float_t sqrtslices = TMath::Sqrt(slices);335 336 // is pixid equal to pixidx ?337 if ((Int_t)fPedestal->GetSize() != fSignals->GetSize())338 {339 *fLog << err << "Sizes of MPedestalCam and MCalibrationCam are different" << endl;340 return kFALSE;341 }342 343 const Int_t n = fPedestal->GetSize();344 for (Int_t pixid=0; pixid<n; pixid++)345 {346 const MPedestalPix &ped = (*fPedestal)[pixid];347 348 // pedestals/(used FADC slices) in [ADC] counts349 const Float_t pedes = ped.GetPedestal() * slices;350 const Float_t pedrms = ped.GetPedestalRms() * sqrtslices;351 352 //353 // get phe/ADC conversion factor354 //355 Float_t hiloconv;356 Float_t hiloconverr;357 Float_t calibConv;358 Float_t calibConvVar;359 Float_t calibFFactor;360 if (!GetConversionFactor(pixid, hiloconv, hiloconverr,361 calibConv, calibConvVar, calibFFactor))362 continue;363 364 //365 // pedestals/(used FADC slices) in [number of photons]366 //367 const Float_t pedphot = pedes * calibConv;368 const Float_t pedphotrms = pedrms * calibConv;369 370 (*fPedPhot)[pixid].Set(pedphot, pedphotrms);371 }372 373 fPedPhot->SetReadyToSave();374 375 return kTRUE;376 }377 378 // --------------------------------------------------------------------------379 //380 331 // Get conversion factor and its error from MCalibrationCam 381 332 // 382 333 Bool_t MCalibrateData::GetConversionFactor(UInt_t pixidx, Float_t &hiloconv, Float_t &hiloconverr, 383 Float_t &calibConv, Float_t &calibConvVar, Float_t &calibFFactor) 334 Float_t &calibConv, Float_t &calibConvVar, 335 Float_t &calibFFactor) const 384 336 { 385 337 // … … 486 438 } 487 439 488 void MCalibrateData::CalibrateData() 489 { 490 const UInt_t npix = fSignals->GetSize(); 440 void MCalibrateData::Calibrate(Bool_t data, Bool_t pedestal) const 441 { 442 if (!data && !pedestal) 443 return; 444 445 const UInt_t npix = fSignals->GetSize(); 446 const Float_t slices = fSignals->GetNumUsedHiGainFADCSlices(); 447 const Float_t sqrtslices = TMath::Sqrt(slices); 491 448 492 449 Float_t hiloconv; 493 450 Float_t hiloconverr; 494 Float_t calib rationConversionFactor;495 Float_t calib rationConversionFactorErr;451 Float_t calibConv; 452 Float_t calibConvErr; 496 453 Float_t calibFFactor; 497 454 … … 499 456 { 500 457 if (!GetConversionFactor(pixidx, hiloconv, hiloconverr, 501 calib rationConversionFactor, calibrationConversionFactorErr, calibFFactor))458 calibConv, calibConvErr, calibFFactor)) 502 459 continue; 503 460 504 const MExtractedSignalPix &sig = (*fSignals)[pixidx]; 505 506 Float_t signal = 0; 507 Float_t signalErr = 0.; 508 509 if (sig.IsLoGainUsed()) 510 { 511 signal = sig.GetExtractedSignalLoGain()*hiloconv; 512 signalErr = signal*hiloconverr; 513 } 514 else 515 { 516 if (sig.GetExtractedSignalHiGain() <= 9999.) 517 signal = sig.GetExtractedSignalHiGain(); 518 } 519 520 const Float_t nphot = signal*calibrationConversionFactor; 521 const Float_t nphotErr = calibFFactor*TMath::Sqrt(TMath::Abs(nphot)); 522 523 // 524 // The following part is the commented first version of the error calculation 525 // Contact Markus Gaug for questions (or wait for the next documentation update...) 526 // 527 /* 528 nphotErr = signal > 0 ? signalErr*signalErr / (signal * signal) : 0. 529 + calibConv > 0 ? calibConvVar / (calibConv * calibConv ) : 0.; 530 nphotErr = TMath::Sqrt(nphotErr) * nphot; 531 */ 532 533 MCerPhotPix *cpix = fCerPhotEvt->AddPixel(pixidx, nphot, nphotErr); 534 535 if (sig.GetNumHiGainSaturated() > 0) 536 cpix->SetPixelHGSaturated(); 537 538 if (sig.GetNumLoGainSaturated() > 0) 539 cpix->SetPixelSaturated(); 540 } 541 542 fCerPhotEvt->FixSize(); 543 fCerPhotEvt->SetReadyToSave(); 461 if (data) 462 { 463 const MExtractedSignalPix &sig = (*fSignals)[pixidx]; 464 465 Float_t signal = 0; 466 Float_t signalErr = 0.; 467 468 if (sig.IsLoGainUsed()) 469 { 470 signal = sig.GetExtractedSignalLoGain()*hiloconv; 471 signalErr = signal*hiloconverr; 472 } 473 else 474 { 475 if (sig.GetExtractedSignalHiGain() <= 9999.) 476 signal = sig.GetExtractedSignalHiGain(); 477 } 478 479 const Float_t nphot = signal*calibConv; 480 const Float_t nphotErr = calibFFactor*TMath::Sqrt(TMath::Abs(nphot)); 481 482 // 483 // The following part is the commented first version of the error calculation 484 // Contact Markus Gaug for questions (or wait for the next documentation update...) 485 // 486 /* 487 nphotErr = signal > 0 ? signalErr*signalErr / (signal * signal) : 0. 488 + calibConv > 0 ? calibConvVar / (calibConv * calibConv ) : 0.; 489 nphotErr = TMath::Sqrt(nphotErr) * nphot; 490 */ 491 492 MCerPhotPix *cpix = fCerPhotEvt->AddPixel(pixidx, nphot, nphotErr); 493 494 if (sig.GetNumHiGainSaturated() > 0) 495 cpix->SetPixelHGSaturated(); 496 497 if (sig.GetNumLoGainSaturated() > 0) 498 cpix->SetPixelSaturated(); 499 } 500 if (pedestal) 501 { 502 const MPedestalPix &ped = (*fPedestal)[pixidx]; 503 504 // pedestals/(used FADC slices) in [ADC] counts 505 const Float_t pedes = ped.GetPedestal() * slices; 506 const Float_t pedrms = ped.GetPedestalRms() * sqrtslices; 507 508 // 509 // pedestals/(used FADC slices) in [number of photons] 510 // 511 const Float_t pedphot = pedes * calibConv; 512 const Float_t pedphotrms = pedrms * calibConv; 513 514 (*fPedPhot)[pixidx].Set(pedphot, pedphotrms); 515 } 516 } 517 518 if (pedestal) 519 fPedPhot->SetReadyToSave(); 520 521 if (data) 522 { 523 fCerPhotEvt->FixSize(); 524 fCerPhotEvt->SetReadyToSave(); 525 } 544 526 } 545 527 … … 563 545 */ 564 546 565 if (fCalibrationMode!=kSkip) 566 CalibrateData(); 567 568 if (TestPedestalFlag(kEvent)) 569 if (!CalibratePedestal()) 570 return kFALSE; 571 572 return kTRUE; 547 Calibrate(fCalibrationMode!=kSkip, TestPedestalFlag(kEvent)); 548 return kTRUE; 573 549 } 574 550 -
trunk/MagicSoft/Mars/mcalib/MCalibrateData.h
r4632 r4752 53 53 TString fNamePedPhotContainer; // name of fPedPhot 54 54 55 Bool_t CalibratePedestal(); 56 void CalibrateData(); 55 void Calibrate(Bool_t data, Bool_t pedestal) const; 57 56 58 Bool_t GetConversionFactor(UInt_t, Float_t &, Float_t &, Float_t &, Float_t &, Float_t &) ;57 Bool_t GetConversionFactor(UInt_t, Float_t &, Float_t &, Float_t &, Float_t &, Float_t &) const; 59 58 60 59 Int_t PreProcess(MParList *pList); -
trunk/MagicSoft/Mars/mjobs/MJCalibrateSignal.cc
r4733 r4752 355 355 356 356 *fLog << all << GetDescriptor() << ": Done." << endl; 357 *fLog << endl << endl; 357 358 358 359 return kTRUE; -
trunk/MagicSoft/Mars/mjobs/MJPedestal.cc
r4733 r4752 650 650 651 651 *fLog << all << GetDescriptor() << ": Done." << endl; 652 *fLog << endl << endl; 652 653 653 654 return kTRUE; -
trunk/MagicSoft/Mars/mpedestal/MPedPhotCam.cc
r4609 r4752 39 39 #include "MPedPhotCam.h" 40 40 41 #include <TArrayI.h>42 #include <TArrayD.h>43 41 #include <TClonesArray.h> 42 43 #include "MArrayI.h" 44 #include "MArrayD.h" 44 45 45 46 #include "MLog.h" … … 233 234 const Int_t na = GetNumAreas(); 234 235 235 TArrayI acnt(na); 236 TArrayI scnt(ns); 237 TArrayD asumx(na); 238 TArrayD ssumx(ns); 239 TArrayD asumr(na); 240 TArrayD ssumr(ns); 236 // Using MArray instead of TArray because they don't do range checks 237 MArrayI acnt(na); 238 MArrayI scnt(ns); 239 MArrayD asumx(na); 240 MArrayD ssumx(ns); 241 MArrayD asumr(na); 242 MArrayD ssumr(ns); 241 243 242 244 for (int i=0; i<np; i++) … … 246 248 247 249 // Create sums for areas and sectors 250 const MPedPhotPix &pix = (*this)[i]; 251 252 const UInt_t ne = pix.GetNumEvents(); 253 const Float_t mean = ne*pix.GetMean(); 254 const Float_t rms = ne*pix.GetRms(); 255 248 256 const UInt_t aidx = geom[i].GetAidx(); 257 asumx[aidx] += mean; 258 asumr[aidx] += rms; 259 acnt[aidx] += ne; 260 249 261 const UInt_t sect = geom[i].GetSector(); 250 251 const MPedPhotPix &pix = (*this)[i]; 252 253 const UInt_t ne = pix.GetNumEvents(); 254 const Float_t mean = pix.GetMean(); 255 const Float_t rms = pix.GetRms(); 256 257 asumx[aidx] += ne*mean; 258 asumr[aidx] += ne*rms; 259 acnt[aidx] += ne; 260 261 ssumx[sect] += ne*mean; 262 ssumr[sect] += ne*rms; 262 ssumx[sect] += mean; 263 ssumr[sect] += rms; 263 264 scnt[sect] += ne; 264 265 } … … 306 307 Bool_t MPedPhotCam::GetPixelContent(Double_t &val, Int_t idx, const MGeomCam &cam, Int_t type) const 307 308 { 308 309 const Float_t ped = (*this)[idx].GetMean();310 const Float_t rms = (*this)[idx].GetRms();311 const UInt_t nevt = (*this)[idx].GetNumEvents();312 313 314 Float_t pederr;315 Float_t rmserr;316 317 if (nevt>0)318 {319 pederr = rms/TMath::Sqrt((Float_t)nevt);320 rmserr = rms/TMath::Sqrt((Float_t)nevt)/2.;321 }322 else323 {324 pederr = -1;325 rmserr = -1;326 }327 328 309 switch (type) 329 310 { 330 311 case 0: 331 val = ped;//(*this)[idx].GetMean();312 val = (*this)[idx].GetMean(); 332 313 break; 333 314 case 1: 334 val = rms; //(*this)[idx].GetRms();315 val = (*this)[idx].GetRms(); 335 316 break; 336 317 case 2: 337 val = pederr; // new318 val = (*this)[idx].GetNumEvents()>0 ? (*this)[idx].GetRms()/TMath::Sqrt((Float_t)(*this)[idx].GetNumEvents()) : -1; 338 319 break; 339 320 case 3: 340 val = rmserr; // new321 val = (*this)[idx].GetNumEvents()>0 ? (*this)[idx].GetRms()/TMath::Sqrt((Float_t)(*this)[idx].GetNumEvents())/2. : -1; 341 322 break; 342 323 default: -
trunk/MagicSoft/Mars/msignal/MExtractTimeFastSpline.cc
r4732 r4752 200 200 pp = fHiGainSecondDeriv[i-1] + 4.; 201 201 fHiGainSecondDeriv[i] = -1.0/pp; 202 fHiGainFirstDeriv [i]= *(p+1) - 2.* *(p) + *(p-1);203 fHiGainFirstDeriv [i] = (6.0* fHiGainFirstDeriv[i]-fHiGainFirstDeriv[i-1])/pp;202 const Double_t deriv = *(p+1) - 2.* *(p) + *(p-1); 203 fHiGainFirstDeriv [i] = (6.0*deriv-fHiGainFirstDeriv[i-1])/pp; 204 204 } 205 205 … … 232 232 // interval maxpos+1. 233 233 // 234 235 Float_t higainklo = fHiGainSecondDeriv[klo]; 236 Float_t higainkhi = fHiGainSecondDeriv[khi]; 234 237 while (x<upper-0.3) 235 238 { … … 241 244 y = a*klocont 242 245 + b*khicont 243 + (a*a*a-a)* fHiGainSecondDeriv[klo]244 + (b*b*b-b)* fHiGainSecondDeriv[khi];246 + (a*a*a-a)*higainklo 247 + (b*b*b-b)*higainkhi; 245 248 246 249 if (y > abmax) … … 265 268 khicont = (Float_t)*(first+khi); 266 269 270 higainklo = fHiGainSecondDeriv[klo]; 271 higainkhi = fHiGainSecondDeriv[khi]; 267 272 while (x<upper-0.3) 268 273 { … … 274 279 y = a* klocont 275 280 + b* khicont 276 + (a*a*a-a)* fHiGainSecondDeriv[klo]277 + (b*b*b-b)* fHiGainSecondDeriv[khi];281 + (a*a*a-a)*higainklo 282 + (b*b*b-b)*higainkhi; 278 283 279 284 if (y > abmax) … … 295 300 step = 0.04; // step size of 83 ps 296 301 302 higainklo = fHiGainSecondDeriv[klo]; 303 higainkhi = fHiGainSecondDeriv[khi]; 297 304 while (x<up) 298 305 { … … 304 311 y = a* klocont 305 312 + b* khicont 306 + (a*a*a-a)* fHiGainSecondDeriv[klo]307 + (b*b*b-b)* fHiGainSecondDeriv[khi];313 + (a*a*a-a)*higainklo 314 + (b*b*b-b)*higainkhi; 308 315 309 316 if (y > abmax) … … 329 336 b = x - lower; 330 337 338 higainklo = fHiGainSecondDeriv[klo]; 339 higainkhi = fHiGainSecondDeriv[khi]; 331 340 while (x>lo) 332 341 { … … 338 347 y = a* klocont 339 348 + b* khicont 340 + (a*a*a-a)* fHiGainSecondDeriv[klo]341 + (b*b*b-b)* fHiGainSecondDeriv[khi];349 + (a*a*a-a)*higainklo 350 + (b*b*b-b)*higainkhi; 342 351 343 352 if (y > abmax)
Note:
See TracChangeset
for help on using the changeset viewer.