CMS 3D CMS Logo

 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Properties Friends Macros Pages
List of all members | Public Member Functions | Static Public Member Functions | Private Attributes
SiPixelTemplate Class Reference

#include <SiPixelTemplate.h>

Public Member Functions

float chi2xavg (int i)
 averaage x chi^2 in 4 charge bins More...
 
float chi2xavgc2m (int i)
 1st pass chi2 min search: average x-chisq for merged clusters More...
 
float chi2xavgone ()
 //!< average x chi^2 for 1 pixel clusters More...
 
float chi2xmin (int i)
 minimum y chi^2 in 4 charge bins More...
 
float chi2xminc2m (int i)
 1st pass chi2 min search: minimum x-chisq for merged clusters More...
 
float chi2xminone ()
 //!< minimum of x chi^2 for 1 pixel clusters More...
 
float chi2yavg (int i)
 average y chi^2 in 4 charge bins More...
 
float chi2yavgc2m (int i)
 1st pass chi2 min search: average y-chisq for merged clusters More...
 
float chi2yavgone ()
 //!< average y chi^2 for 1 pixel clusters More...
 
float chi2ymin (int i)
 minimum y chi^2 in 4 charge bins More...
 
float chi2yminc2m (int i)
 1st pass chi2 min search: minimum y-chisq for merged clusters More...
 
float chi2yminone ()
 //!< minimum of y chi^2 for 1 pixel clusters More...
 
float clslenx ()
 x-size of smaller interpolated template in pixels More...
 
float clsleny ()
 y-size of smaller interpolated template in pixels More...
 
int cxtemp ()
 Return central pixel of x-template pixels above readout threshold. More...
 
int cytemp ()
 Return central pixel of y template pixels above readout threshold. More...
 
float dxone ()
 mean offset/correction for one pixel x-clusters More...
 
float dxtwo ()
 mean offset/correction for one double-pixel x-clusters More...
 
float dyone ()
 mean offset/correction for one pixel y-clusters More...
 
float dytwo ()
 mean offset/correction for one double-pixel y-clusters More...
 
float fbin (int i)
 Return lower bound of Qbin definition. More...
 
bool interpolate (int id, float cotalpha, float cotbeta, float locBz)
 
bool interpolate (int id, float cotalpha, float cotbeta)
 
float kappavav ()
 kappa parameter for Vavilov distribution More...
 
float kappavav2 ()
 kappa parameter for 2-cluster Vavilov distribution More...
 
float lorxbias ()
 signed lorentz x-width (microns) More...
 
float lorxwidth ()
 signed lorentz x-width (microns) More...
 
float lorybias ()
 signed lorentz y-width (microns) More...
 
float lorywidth ()
 signed lorentz y-width (microns) More...
 
float mpvvav ()
 most probable charge in Vavilov distribution (not actually for larger kappa) More...
 
float mpvvav2 ()
 most probable charge in 2-cluster Vavilov distribution (not actually for larger kappa) More...
 
float pixmax ()
 maximum pixel charge More...
 
float qavg ()
 average cluster charge for this set of track angles More...
 
int qbin (int id, float cotalpha, float cotbeta, float locBz, float qclus, float &pixmx, float &sigmay, float &deltay, float &sigmax, float &deltax, float &sy1, float &dy1, float &sy2, float &dy2, float &sx1, float &dx1, float &sx2, float &dx2)
 
int qbin (int id, float cotalpha, float cotbeta, float qclus)
 
int qbin (int id, float cotbeta, float qclus)
 
void qbin_dist (int id, float cotalpha, float cotbeta, float qbin_frac[4], float &ny1_frac, float &ny2_frac, float &nx1_frac, float &nx2_frac)
 
float qmin ()
 minimum cluster charge for valid hit (keeps 99.9% of simulated hits) More...
 
float qmin (int i)
 minimum cluster charge for valid hit (keeps 99.9% or 99.8% of simulated hits) More...
 
float qscale ()
 charge scaling factor More...
 
float r_qMeas_qTrue ()
 ratio of measured to true cluster charge More...
 
float s50 ()
 1/2 of the pixel threshold signal in electrons More...
 
float sigmavav ()
 "sigma" scale fctor for Vavilov distribution More...
 
float sigmavav2 ()
 "sigma" scale fctor for 2-cluster Vavilov distribution More...
 
bool simpletemplate2D (float xhitp, float yhitp, std::vector< bool > &ydouble, std::vector< bool > &xdouble, float template2d[13+2][21+2])
 Make simple 2-D templates from track angles set in interpolate and hit position. More...
 
 SiPixelTemplate (const std::vector< SiPixelTemplateStore > &thePixelTemp)
 Constructor for cases in which template store already exists. More...
 
float ss50 ()
 1/2 of the pixel threshold signal in electrons More...
 
float sxmax ()
 average pixel signal for x-projection of cluster More...
 
float sxone ()
 rms for one pixel x-clusters More...
 
float sxtwo ()
 rms for one double-pixel x-clusters More...
 
float symax ()
 average pixel signal for y-projection of cluster More...
 
float syone ()
 rms for one pixel y-clusters More...
 
float sytwo ()
 rms for one double-pixel y-clusters More...
 
void temperrors (int id, float cotalpha, float cotbeta, int qBin, float &sigmay, float &sigmax, float &sy1, float &sy2, float &sx1, float &sx2)
 
void vavilov2_pars (double &mpv, double &sigma, double &kappa)
 
void vavilov_pars (double &mpv, double &sigma, double &kappa)
 
float xavg (int i)
 average x-bias of reconstruction binned in 4 charge bins More...
 
float xavgc2m (int i)
 1st pass chi2 min search: average x-bias of reconstruction binned in 4 charge bins More...
 
float xflcorr (int binq, float qflx)
 
float xgsig (int i)
 average sigma_x from Gaussian fit binned in 4 charge bins More...
 
float xgx0 (int i)
 average x0 from Gaussian fit binned in 4 charge bins More...
 
float xrms (int i)
 average x-rms of reconstruction binned in 4 charge bins More...
 
float xrmsc2m (int i)
 1st pass chi2 min search: average x-rms of reconstruction binned in 4 charge bins More...
 
void xsigma2 (int fxpix, int lxpix, float sxthr, float xsum[13+4], float xsig2[13+4])
 
float xsize ()
 pixel x-size (microns) More...
 
void xtemp (int fxbin, int lxbin, float xtemplate[41][13+4])
 
void xtemp3d (int j, int k, std::vector< float > &xtemplate)
 
void xtemp3d_int (int nxpix, int &nxbins)
 
float xxratio ()
 fractional distance in x between cotalpha templates More...
 
float yavg (int i)
 average y-bias of reconstruction binned in 4 charge bins More...
 
float yavgc2m (int i)
 1st pass chi2 min search: average y-bias of reconstruction binned in 4 charge bins More...
 
float yflcorr (int binq, float qfly)
 
float ygsig (int i)
 average sigma_y from Gaussian fit binned in 4 charge bins More...
 
float ygx0 (int i)
 average y0 from Gaussian fit binned in 4 charge bins More...
 
float yratio ()
 fractional distance in y between cotbeta templates More...
 
float yrms (int i)
 average y-rms of reconstruction binned in 4 charge bins More...
 
float yrmsc2m (int i)
 1st pass chi2 min search: average y-rms of reconstruction binned in 4 charge bins More...
 
void ysigma2 (int fypix, int lypix, float sythr, float ysum[21+4], float ysig2[21+4])
 
void ysigma2 (float qpixel, int index, float &ysig2)
 
float ysize ()
 pixel y-size (microns) More...
 
void ytemp (int fybin, int lybin, float ytemplate[41][21+4])
 
void ytemp3d (int j, int k, std::vector< float > &ytemplate)
 
void ytemp3d_int (int nypix, int &nybins)
 
float yxratio ()
 fractional distance in y between cotalpha templates slices More...
 
float zsize ()
 pixel z-size or thickness (microns) More...
 

Static Public Member Functions

static void postInit (std::vector< SiPixelTemplateStore > &thePixelTemp_)
 
static bool pushfile (int filenum, std::vector< SiPixelTemplateStore > &thePixelTemp_)
 
static bool pushfile (const SiPixelTemplateDBObject &dbobject, std::vector< SiPixelTemplateStore > &thePixelTemp_)
 

Private Attributes

float abs_cotb_
 absolute value of cot beta More...
 
float chi2xavg_ [4]
 average x chi^2 in 4 charge bins More...
 
float chi2xavgc2m_ [4]
 1st pass chi2 min search: average x-chisq for merged clusters More...
 
float chi2xavgone_
 average x chi^2 for 1 pixel clusters More...
 
float chi2xmin_ [4]
 minimum of x chi^2 in 4 charge bins More...
 
float chi2xminc2m_ [4]
 1st pass chi2 min search: minimum x-chisq for merged clusters More...
 
float chi2xminone_
 minimum of x chi^2 for 1 pixel clusters More...
 
float chi2yavg_ [4]
 average y chi^2 in 4 charge bins More...
 
float chi2yavgc2m_ [4]
 1st pass chi2 min search: average y-chisq for merged clusters More...
 
float chi2yavgone_
 average y chi^2 for 1 pixel clusters More...
 
float chi2ymin_ [4]
 minimum of y chi^2 in 4 charge bins More...
 
float chi2yminc2m_ [4]
 1st pass chi2 min search: minimum y-chisq for merged clusters More...
 
float chi2yminone_
 minimum of y chi^2 for 1 pixel clusters More...
 
float clslenx_
 x-cluster length of smaller interpolated template in pixels More...
 
float clsleny_
 y-cluster length of smaller interpolated template in pixels More...
 
float cota_current_
 current cot alpha More...
 
float cotb_current_
 current cot beta More...
 
float dxone_
 mean offset/correction for one pixel x-clusters More...
 
float dxtwo_
 mean offset/correction for one double-pixel x-clusters More...
 
float dyone_
 mean offset/correction for one pixel y-clusters More...
 
float dytwo_
 mean offset/correction for one double-pixel y-clusters More...
 
float fbin_ [3]
 The QBin definitions in Q_clus/Q_avg. More...
 
int id_current_
 current id More...
 
int index_id_
 current index More...
 
float kappavav2_
 kappa parameter for 2-cluster Vavilov distribution More...
 
float kappavav_
 kappa parameter for Vavilov distribution More...
 
float lorxbias_
 Lorentz x-bias. More...
 
float lorxwidth_
 Lorentz x-width. More...
 
float lorybias_
 Lorentz y-bias. More...
 
float lorywidth_
 Lorentz y-width (sign corrected for fpix frame) More...
 
float mpvvav2_
 most probable charge in 2-cluster Vavilov distribution (not actually for larger kappa) More...
 
float mpvvav_
 most probable charge in Vavilov distribution (not actually for larger kappa) More...
 
float nxbins_
 number of bins in each dimension of the x-splitting template More...
 
float nybins_
 number of bins in each dimension of the y-splitting template More...
 
float pixmax_
 maximum pixel charge More...
 
float qavg_
 average cluster charge for this set of track angles More...
 
float qavg_avg_
 average of cluster charge less than qavg More...
 
float qmin2_
 tighter minimum cluster charge for valid hit (keeps 99.8% of simulated hits) More...
 
float qmin_
 minimum cluster charge for valid hit (keeps 99.9% of simulated hits) More...
 
float qscale_
 charge scaling factor More...
 
float r_qMeas_qTrue_
 ratio of measured to true cluster charges More...
 
float s50_
 1/2 of the pixel threshold signal in adc units More...
 
float sigmavav2_
 "sigma" scale fctor for 2-cluster Vavilov distribution More...
 
float sigmavav_
 "sigma" scale fctor for Vavilov distribution More...
 
float ss50_
 1/2 of the pixel threshold signal in adc units More...
 
bool success_
 true if cotalpha, cotbeta are inside of the acceptance (dynamically loaded) More...
 
float sxmax_
 average pixel signal for x-projection of cluster More...
 
float sxone_
 rms for one pixel x-clusters More...
 
float sxparmax_
 maximum pixel signal for parameterization of x uncertainties More...
 
float sxtwo_
 rms for one double-pixel x-clusters More...
 
float symax_
 average pixel signal for y-projection of cluster More...
 
float syone_
 rms for one pixel y-clusters More...
 
float syparmax_
 maximum pixel signal for parameterization of y uncertainties More...
 
float sytwo_
 rms for one double-pixel y-clusters More...
 
boost::multi_array< float, 2 > temp2dx_
 2d-primitive for spltting 3-d template More...
 
boost::multi_array< float, 2 > temp2dy_
 2d-primitive for spltting 3-d template More...
 
const std::vector
< SiPixelTemplateStore > & 
thePixelTemp_
 
float xavg_ [4]
 average x-bias of reconstruction binned in 4 charge bins More...
 
float xavgc2m_ [4]
 1st pass chi2 min search: average x-bias of reconstruction binned in 4 charge bins More...
 
float xflparhh_ [4][6]
 Aqfl-parameterized x-correction in 4 charge bins for larger cotbeta, cotalpha. More...
 
float xflparhl_ [4][6]
 Aqfl-parameterized x-correction in 4 charge bins for larger cotbeta, smaller cotalpha. More...
 
float xflparlh_ [4][6]
 Aqfl-parameterized x-correction in 4 charge bins for smaller cotbeta, larger cotalpha. More...
 
float xflparll_ [4][6]
 Aqfl-parameterized x-correction in 4 charge bins for smaller cotbeta, cotalpha. More...
 
float xgsig_ [4]
 sigma from Gaussian fit binned in 4 charge bins More...
 
float xgx0_ [4]
 average x0 from Gaussian fit binned in 4 charge bins More...
 
float xpar0_ [2][5]
 projected x-pixel uncertainty parameterization for central cotalpha More...
 
float xparh_ [2][5]
 projected x-pixel uncertainty parameterization for larger cotalpha More...
 
float xparhy0_ [2][5]
 projected x-pixel uncertainty parameterization for larger cotbeta (central alpha) More...
 
float xparl_ [2][5]
 projected x-pixel uncertainty parameterization for smaller cotalpha More...
 
float xparly0_ [2][5]
 projected x-pixel uncertainty parameterization for smaller cotbeta (central alpha) More...
 
float xrms_ [4]
 average x-rms of reconstruction binned in 4 charge bins More...
 
float xrmsc2m_ [4]
 1st pass chi2 min search: average x-rms of reconstruction binned in 4 charge bins More...
 
float xsize_
 Pixel x-size. More...
 
float xtemp_ [9][13+4]
 templates for x-reconstruction (binned over 5 central pixels) More...
 
float xxratio_
 fractional distance in x between cotalpha templates More...
 
float yavg_ [4]
 average y-bias of reconstruction binned in 4 charge bins More...
 
float yavgc2m_ [4]
 1st pass chi2 min search: average y-bias of reconstruction binned in 4 charge bins More...
 
float yflparh_ [4][6]
 Aqfl-parameterized y-correction in 4 charge bins for larger cotbeta. More...
 
float yflparl_ [4][6]
 Aqfl-parameterized y-correction in 4 charge bins for smaller cotbeta. More...
 
float ygsig_ [4]
 average sigma_y from Gaussian fit binned in 4 charge bins More...
 
float ygx0_ [4]
 average y0 from Gaussian fit binned in 4 charge bins More...
 
float yparh_ [2][5]
 projected y-pixel uncertainty parameterization for larger cotbeta More...
 
float yparl_ [2][5]
 projected y-pixel uncertainty parameterization for smaller cotbeta More...
 
float yratio_
 fractional distance in y between cotbeta templates More...
 
float yrms_ [4]
 average y-rms of reconstruction binned in 4 charge bins More...
 
float yrmsc2m_ [4]
 1st pass chi2 min search: average y-rms of reconstruction binned in 4 charge bins More...
 
float ysize_
 Pixel y-size. More...
 
float ytemp_ [9][21+4]
 templates for y-reconstruction (binned over 5 central pixels) More...
 
float yxratio_
 fractional distance in y between x-slices of cotalpha templates More...
 
float zsize_
 Pixel z-size (thickness) More...
 

Detailed Description

A template management class. SiPixelTemplate contains thePixelTemp (a std::vector of SiPixelTemplateStore, each of which is a collection of many SiPixelTemplateEntries). Each SiPixelTemplateStore corresponds to a given detector condition, and is valid for a range of runs. We allow more than one Store since the may change over time.

This class reads templates from files via pushfile() method.

The main functionality of SiPixelTemplate is interpolate(), which produces a template on the fly, given a specific track's alpha and beta. The results are kept in data members and accessed via inline getters.

The resulting template is then used by PixelTempReco2D() (a global function) which get the reference for SiPixelTemplate & templ and uses the current template to reconstruct the SiPixelRecHit.

Definition at line 235 of file SiPixelTemplate.h.

Constructor & Destructor Documentation

SiPixelTemplate::SiPixelTemplate ( const std::vector< SiPixelTemplateStore > &  thePixelTemp)
inline

Constructor for cases in which template store already exists.

Definition at line 237 of file SiPixelTemplate.h.

References cota_current_, cotb_current_, id_current_, and index_id_.

Member Function Documentation

float SiPixelTemplate::chi2xavg ( int  i)
inline

averaage x chi^2 in 4 charge bins

Definition at line 417 of file SiPixelTemplate.h.

References assert(), chi2xavg_, edm::hlt::Exception, and i.

Referenced by SiPixelTemplateReco::PixelTempReco2D().

float SiPixelTemplate::chi2xavgc2m ( int  i)
inline

1st pass chi2 min search: average x-chisq for merged clusters

Definition at line 473 of file SiPixelTemplate.h.

References assert(), chi2xavgc2m_, edm::hlt::Exception, and i.

Referenced by SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::chi2xavgone ( )
inline

//!< average x chi^2 for 1 pixel clusters

Definition at line 497 of file SiPixelTemplate.h.

References chi2xavgone_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::chi2xmin ( int  i)
inline

minimum y chi^2 in 4 charge bins

Definition at line 424 of file SiPixelTemplate.h.

References assert(), chi2xmin_, edm::hlt::Exception, and i.

Referenced by SiPixelTemplateReco::PixelTempReco2D().

float SiPixelTemplate::chi2xminc2m ( int  i)
inline

1st pass chi2 min search: minimum x-chisq for merged clusters

Definition at line 480 of file SiPixelTemplate.h.

References assert(), chi2xminc2m_, edm::hlt::Exception, and i.

float SiPixelTemplate::chi2xminone ( )
inline

//!< minimum of x chi^2 for 1 pixel clusters

Definition at line 498 of file SiPixelTemplate.h.

References chi2xminone_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::chi2yavg ( int  i)
inline

average y chi^2 in 4 charge bins

Definition at line 403 of file SiPixelTemplate.h.

References assert(), chi2yavg_, edm::hlt::Exception, and i.

Referenced by SiPixelTemplateReco::PixelTempReco2D().

float SiPixelTemplate::chi2yavgc2m ( int  i)
inline

1st pass chi2 min search: average y-chisq for merged clusters

Definition at line 445 of file SiPixelTemplate.h.

References assert(), chi2yavgc2m_, edm::hlt::Exception, and i.

Referenced by SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::chi2yavgone ( )
inline

//!< average y chi^2 for 1 pixel clusters

Definition at line 495 of file SiPixelTemplate.h.

References chi2yavgone_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::chi2ymin ( int  i)
inline

minimum y chi^2 in 4 charge bins

Definition at line 410 of file SiPixelTemplate.h.

References assert(), chi2ymin_, edm::hlt::Exception, and i.

Referenced by SiPixelTemplateReco::PixelTempReco2D().

float SiPixelTemplate::chi2yminc2m ( int  i)
inline

1st pass chi2 min search: minimum y-chisq for merged clusters

Definition at line 452 of file SiPixelTemplate.h.

References assert(), chi2yminc2m_, edm::hlt::Exception, and i.

float SiPixelTemplate::chi2yminone ( )
inline

//!< minimum of y chi^2 for 1 pixel clusters

Definition at line 496 of file SiPixelTemplate.h.

References chi2yminone_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::clslenx ( )
inline

x-size of smaller interpolated template in pixels

Definition at line 343 of file SiPixelTemplate.h.

References clslenx_.

float SiPixelTemplate::clsleny ( )
inline

y-size of smaller interpolated template in pixels

Definition at line 342 of file SiPixelTemplate.h.

References clsleny_.

int SiPixelTemplate::cxtemp ( )

Return central pixel of x-template pixels above readout threshold.

Definition at line 1992 of file SiPixelTemplate.cc.

References BXM1, BXM2, BXSIZE, and j.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

1994 {
1995  // Retrieve already interpolated quantities
1996 
1997  // Local variables
1998  int j;
1999 
2000  // Analyze only pixels along the central entry
2001  // First, find the maximum signal and then work out to the edges
2002 
2003  float sigmax = 0.f;
2004  float qedge = 2.*s50_;
2005  int jmax = -1;
2006 
2007  for(j=0; j<BXSIZE; ++j) {
2008  if(xtemp_[4][j] > sigmax) {
2009  sigmax = xtemp_[4][j];
2010  jmax = j;
2011  }
2012  }
2013  if(sigmax < qedge) {qedge = s50_;}
2014  if(sigmax < qedge || jmax<1 || jmax>BXM2) {return -1;}
2015 
2016  // Now search forward and backward
2017 
2018  int jend = jmax;
2019 
2020  for(j=jmax+1; j<BXM1; ++j) {
2021  if(xtemp_[4][j] < qedge) break;
2022  jend = j;
2023  }
2024 
2025  int jbeg = jmax;
2026 
2027  for(j=jmax-1; j>0; --j) {
2028  if(xtemp_[4][j] < qedge) break;
2029  jbeg = j;
2030  }
2031 
2032  return (jbeg+jend)/2;
2033 
2034 } // End cxtemp
#define BXSIZE
#define BXM1
float s50_
1/2 of the pixel threshold signal in adc units
#define BXM2
int j
Definition: DBlmapReader.cc:9
float xtemp_[9][13+4]
templates for x-reconstruction (binned over 5 central pixels)
int SiPixelTemplate::cytemp ( )

Return central pixel of y template pixels above readout threshold.

Definition at line 1943 of file SiPixelTemplate.cc.

References BYM1, BYM2, BYSIZE, and j.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

1945 {
1946  // Retrieve already interpolated quantities
1947 
1948  // Local variables
1949  int j;
1950 
1951 // Analyze only pixels along the central entry
1952 // First, find the maximum signal and then work out to the edges
1953 
1954  float sigmax = 0.f;
1955  float qedge = 2.*s50_;
1956  int jmax = -1;
1957 
1958  for(j=0; j<BYSIZE; ++j) {
1959  if(ytemp_[4][j] > sigmax) {
1960  sigmax = ytemp_[4][j];
1961  jmax = j;
1962  }
1963  }
1964  if(sigmax < qedge) {qedge = s50_;}
1965  if(sigmax < qedge || jmax<1 || jmax>BYM2) {return -1;}
1966 
1967 // Now search forward and backward
1968 
1969  int jend = jmax;
1970 
1971  for(j=jmax+1; j<BYM1; ++j) {
1972  if(ytemp_[4][j] < qedge) break;
1973  jend = j;
1974  }
1975 
1976  int jbeg = jmax;
1977 
1978 for(j=jmax-1; j>0; --j) {
1979  if(ytemp_[4][j] < qedge) break;
1980  jbeg = j;
1981  }
1982 
1983  return (jbeg+jend)/2;
1984 
1985 } // End cytemp
#define BYSIZE
float ytemp_[9][21+4]
templates for y-reconstruction (binned over 5 central pixels)
float s50_
1/2 of the pixel threshold signal in adc units
#define BYM1
int j
Definition: DBlmapReader.cc:9
#define BYM2
float SiPixelTemplate::dxone ( )
inline

mean offset/correction for one pixel x-clusters

Definition at line 330 of file SiPixelTemplate.h.

References dxone_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::dxtwo ( )
inline

mean offset/correction for one double-pixel x-clusters

Definition at line 332 of file SiPixelTemplate.h.

References dxtwo_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::dyone ( )
inline

mean offset/correction for one pixel y-clusters

Definition at line 325 of file SiPixelTemplate.h.

References dyone_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::dytwo ( )
inline

mean offset/correction for one double-pixel y-clusters

Definition at line 327 of file SiPixelTemplate.h.

References dytwo_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::fbin ( int  i)
inline

Return lower bound of Qbin definition.

Definition at line 487 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, fbin_, and i.

bool SiPixelTemplate::interpolate ( int  id,
float  cotalpha,
float  cotbeta,
float  locBz 
)

Interpolate input alpha and beta angles to produce a working template for each individual hit.

Parameters
id- (input) index of the template to use
cotalpha- (input) the cotangent of the alpha track angle (see CMS IN 2004/014)
cotbeta- (input) the cotangent of the beta track angle (see CMS IN 2004/014)
locBz- (input) the sign of this quantity is used to determine whether to flip cot(beta)<0 quantities from cot(beta)>0 (FPix only) for FPix IP-related tracks, locBz < 0 for cot(beta) > 0 and locBz > 0 for cot(beta) < 0

Definition at line 981 of file SiPixelTemplate.cc.

References funct::abs(), assert(), BXM1, BXM2, BYM1, BYM2, BYM3, edm::hlt::Exception, f, i, j, mathSSE::sqrt(), TXSIZE, and TYSIZE.

Referenced by interpolate(), SiPixelTemplateReco::PixelTempReco2D(), SiPixelTemplateSplit::PixelTempSplit(), and SiPixelGaussianSmearingRecHitConverterAlgorithm::smearHit().

981  {
982  // Interpolate for a new set of track angles
983 
984  // Local variables
985  int i, j;
986  int ilow, ihigh, iylow, iyhigh, Ny, Nxx, Nyx, imidy, imaxx;
987  float yratio, yxratio, xxratio, sxmax, qcorrect, qxtempcor, symax, chi2xavgone, chi2xminone, cotb, cotalpha0, cotbeta0;
988  bool flip_y;
989  // std::vector <float> xrms(4), xgsig(4), xrmsc2m(4);
990  float chi2xavg[4], chi2xmin[4], chi2xavgc2m[4], chi2xminc2m[4];
991 
992 
993  // Check to see if interpolation is valid
994  if(id != id_current_ || cotalpha != cota_current_ || cotbeta != cotb_current_) {
995 
996  cota_current_ = cotalpha; cotb_current_ = cotbeta; success_ = true;
997 
998  if(id != id_current_) {
999 
1000  // Find the index corresponding to id
1001 
1002  index_id_ = -1;
1003  for(i=0; i<(int)thePixelTemp_.size(); ++i) {
1004 
1005  //std::cout<<i<<" "<<id<<" "<<thePixelTemp_[i].head.ID<<std::endl;
1006 
1007  if(id == thePixelTemp_[i].head.ID) {
1008 
1009  index_id_ = i;
1010  id_current_ = id;
1011 
1012  // Copy the charge scaling factor to the private variable
1013 
1014  qscale_ = thePixelTemp_[index_id_].head.qscale;
1015 
1016  // Copy the pseudopixel signal size to the private variable
1017 
1018  s50_ = thePixelTemp_[index_id_].head.s50;
1019 
1020  // Copy Qbinning info to private variables
1021 
1022  for(j=0; j<3; ++j) {fbin_[j] = thePixelTemp_[index_id_].head.fbin[j];}
1023  //std::cout<<" set fbin "<< fbin_[0]<<" "<<fbin_[1]<<" "<<fbin_[2]<<std::endl;
1024 
1025  // Pixel sizes to the private variables
1026 
1027  xsize_ = thePixelTemp_[index_id_].head.xsize;
1028  ysize_ = thePixelTemp_[index_id_].head.ysize;
1029  zsize_ = thePixelTemp_[index_id_].head.zsize;
1030 
1031  break;
1032  }
1033  }
1034  }
1035 
1036 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1037  if(index_id_ < 0 || index_id_ >= (int)thePixelTemp_.size()) {
1038  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::interpolate can't find needed template ID = " << id << std::endl;
1039  }
1040 #else
1041  assert(index_id_ >= 0 && index_id_ < (int)thePixelTemp_.size());
1042 #endif
1043 
1044 // Interpolate the absolute value of cot(beta)
1045 
1046  abs_cotb_ = std::abs(cotbeta);
1047 
1048 // qcorrect corrects the cot(alpha)=0 cluster charge for non-zero cot(alpha)
1049 
1050  cotalpha0 = thePixelTemp_[index_id_].enty[0].cotalpha;
1051  qcorrect=std::sqrt((1.f+cotbeta*cotbeta+cotalpha*cotalpha)/(1.f+cotbeta*cotbeta+cotalpha0*cotalpha0));
1052 
1053 // for some cosmics, the ususal gymnastics are incorrect
1054  if(thePixelTemp_[index_id_].head.Dtype == 0) {
1055  cotb = abs_cotb_;
1056  flip_y = false;
1057  if(cotbeta < 0.f) {flip_y = true;}
1058  } else {
1059  if(locBz < 0.f) {
1060  cotb = cotbeta;
1061  flip_y = false;
1062  } else {
1063  cotb = -cotbeta;
1064  flip_y = true;
1065  }
1066  }
1067 
1068  Ny = thePixelTemp_[index_id_].head.NTy;
1069  Nyx = thePixelTemp_[index_id_].head.NTyx;
1070  Nxx = thePixelTemp_[index_id_].head.NTxx;
1071 
1072 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1073  if(Ny < 2 || Nyx < 1 || Nxx < 2) {
1074  throw cms::Exception("DataCorrupt") << "template ID = " << id_current_ << "has too few entries: Ny/Nyx/Nxx = " << Ny << "/" << Nyx << "/" << Nxx << std::endl;
1075  }
1076 #else
1077  assert(Ny > 1 && Nyx > 0 && Nxx > 1);
1078 #endif
1079  imaxx = Nyx - 1;
1080  imidy = Nxx/2;
1081 
1082 // next, loop over all y-angle entries
1083 
1084  ilow = 0;
1085  yratio = 0.f;
1086 
1087  if(cotb >= thePixelTemp_[index_id_].enty[Ny-1].cotbeta) {
1088 
1089  ilow = Ny-2;
1090  yratio = 1.;
1091  success_ = false;
1092 
1093  } else {
1094 
1095  if(cotb >= thePixelTemp_[index_id_].enty[0].cotbeta) {
1096 
1097  for (i=0; i<Ny-1; ++i) {
1098 
1099  if( thePixelTemp_[index_id_].enty[i].cotbeta <= cotb && cotb < thePixelTemp_[index_id_].enty[i+1].cotbeta) {
1100 
1101  ilow = i;
1102  yratio = (cotb - thePixelTemp_[index_id_].enty[i].cotbeta)/(thePixelTemp_[index_id_].enty[i+1].cotbeta - thePixelTemp_[index_id_].enty[i].cotbeta);
1103  break;
1104  }
1105  }
1106  } else { success_ = false; }
1107  }
1108 
1109  ihigh=ilow + 1;
1110 
1111 // Interpolate/store all y-related quantities (flip displacements when flip_y)
1112 
1113  yratio_ = yratio;
1114  qavg_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].qavg + yratio*thePixelTemp_[index_id_].enty[ihigh].qavg;
1115  qavg_ *= qcorrect;
1116  symax = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].symax + yratio*thePixelTemp_[index_id_].enty[ihigh].symax;
1117  syparmax_ = symax;
1118  sxmax = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].sxmax + yratio*thePixelTemp_[index_id_].enty[ihigh].sxmax;
1119  dyone_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].dyone + yratio*thePixelTemp_[index_id_].enty[ihigh].dyone;
1120  if(flip_y) {dyone_ = -dyone_;}
1121  syone_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].syone + yratio*thePixelTemp_[index_id_].enty[ihigh].syone;
1122  dytwo_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].dytwo + yratio*thePixelTemp_[index_id_].enty[ihigh].dytwo;
1123  if(flip_y) {dytwo_ = -dytwo_;}
1124  sytwo_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].sytwo + yratio*thePixelTemp_[index_id_].enty[ihigh].sytwo;
1125  qmin_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].qmin + yratio*thePixelTemp_[index_id_].enty[ihigh].qmin;
1126  qmin_ *= qcorrect;
1127  qmin2_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].qmin2 + yratio*thePixelTemp_[index_id_].enty[ihigh].qmin2;
1128  qmin2_ *= qcorrect;
1129  mpvvav_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].mpvvav + yratio*thePixelTemp_[index_id_].enty[ihigh].mpvvav;
1130  mpvvav_ *= qcorrect;
1131  sigmavav_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].sigmavav + yratio*thePixelTemp_[index_id_].enty[ihigh].sigmavav;
1132  kappavav_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].kappavav + yratio*thePixelTemp_[index_id_].enty[ihigh].kappavav;
1133  mpvvav2_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].mpvvav2 + yratio*thePixelTemp_[index_id_].enty[ihigh].mpvvav2;
1134  mpvvav2_ *= qcorrect;
1135  sigmavav2_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].sigmavav2 + yratio*thePixelTemp_[index_id_].enty[ihigh].sigmavav2;
1136  kappavav2_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].kappavav2 + yratio*thePixelTemp_[index_id_].enty[ihigh].kappavav2;
1137  clsleny_ = fminf(thePixelTemp_[index_id_].enty[ilow].clsleny, thePixelTemp_[index_id_].enty[ihigh].clsleny);
1138  qavg_avg_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].qavg_avg + yratio*thePixelTemp_[index_id_].enty[ihigh].qavg_avg;
1139  qavg_avg_ *= qcorrect;
1140  for(i=0; i<2 ; ++i) {
1141  for(j=0; j<5 ; ++j) {
1142 // Charge loss switches sides when cot(beta) changes sign
1143  if(flip_y) {
1144  yparl_[1-i][j] = thePixelTemp_[index_id_].enty[ilow].ypar[i][j];
1145  yparh_[1-i][j] = thePixelTemp_[index_id_].enty[ihigh].ypar[i][j];
1146  } else {
1147  yparl_[i][j] = thePixelTemp_[index_id_].enty[ilow].ypar[i][j];
1148  yparh_[i][j] = thePixelTemp_[index_id_].enty[ihigh].ypar[i][j];
1149  }
1150  xparly0_[i][j] = thePixelTemp_[index_id_].enty[ilow].xpar[i][j];
1151  xparhy0_[i][j] = thePixelTemp_[index_id_].enty[ihigh].xpar[i][j];
1152  }
1153  }
1154  for(i=0; i<4; ++i) {
1155  yavg_[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].yavg[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].yavg[i];
1156  if(flip_y) {yavg_[i] = -yavg_[i];}
1157  yrms_[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].yrms[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].yrms[i];
1158 // ygx0_[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].ygx0[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].ygx0[i];
1159 // if(flip_y) {ygx0_[i] = -ygx0_[i];}
1160 // ygsig_[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].ygsig[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].ygsig[i];
1161 // xrms[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].xrms[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].xrms[i];
1162 // xgsig[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].xgsig[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].xgsig[i];
1163  chi2yavg_[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2yavg[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2yavg[i];
1164  chi2ymin_[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2ymin[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2ymin[i];
1165  chi2xavg[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2xavg[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2xavg[i];
1166  chi2xmin[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2xmin[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2xmin[i];
1167  yavgc2m_[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].yavgc2m[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].yavgc2m[i];
1168  if(flip_y) {yavgc2m_[i] = -yavgc2m_[i];}
1169  yrmsc2m_[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].yrmsc2m[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].yrmsc2m[i];
1170  chi2yavgc2m_[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2yavgc2m[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2yavgc2m[i];
1171 // if(flip_y) {chi2yavgc2m_[i] = -chi2yavgc2m_[i];}
1172  chi2yminc2m_[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2yminc2m[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2yminc2m[i];
1173 // xrmsc2m[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].xrmsc2m[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].xrmsc2m[i];
1174  chi2xavgc2m[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2xavgc2m[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2xavgc2m[i];
1175  chi2xminc2m[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2xminc2m[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2xminc2m[i];
1176  for(j=0; j<6 ; ++j) {
1177  yflparl_[i][j] = thePixelTemp_[index_id_].enty[ilow].yflpar[i][j];
1178  yflparh_[i][j] = thePixelTemp_[index_id_].enty[ihigh].yflpar[i][j];
1179 
1180 // Since Q_fl is odd under cotbeta, it flips qutomatically, change only even terms
1181 
1182  if(flip_y && (j == 0 || j == 2 || j == 4)) {
1183  yflparl_[i][j] = - yflparl_[i][j];
1184  yflparh_[i][j] = - yflparh_[i][j];
1185  }
1186  }
1187  }
1188 
1190 
1191  chi2yavgone_=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2yavgone + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2yavgone;
1192  chi2yminone_=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2yminone + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2yminone;
1193  chi2xavgone=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2xavgone + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2xavgone;
1194  chi2xminone=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].chi2xminone + yratio*thePixelTemp_[index_id_].enty[ihigh].chi2xminone;
1195  // for(i=0; i<10; ++i) {
1196 // pyspare[i]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].yspare[i] + yratio*thePixelTemp_[index_id_].enty[ihigh].yspare[i];
1197 // }
1198 
1199 // Interpolate and build the y-template
1200 
1201  for(i=0; i<9; ++i) {
1202  ytemp_[i][0] = 0.f;
1203  ytemp_[i][1] = 0.f;
1204  ytemp_[i][BYM2] = 0.f;
1205  ytemp_[i][BYM1] = 0.f;
1206  for(j=0; j<TYSIZE; ++j) {
1207 
1208 // Flip the basic y-template when the cotbeta is negative
1209 
1210  if(flip_y) {
1211  ytemp_[8-i][BYM3-j]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].ytemp[i][j] + yratio*thePixelTemp_[index_id_].enty[ihigh].ytemp[i][j];
1212  } else {
1213  ytemp_[i][j+2]=(1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].ytemp[i][j] + yratio*thePixelTemp_[index_id_].enty[ihigh].ytemp[i][j];
1214  }
1215  }
1216  }
1217 
1218 // next, loop over all x-angle entries, first, find relevant y-slices
1219 
1220  iylow = 0;
1221  yxratio = 0.f;
1222 
1223  if(abs_cotb_ >= thePixelTemp_[index_id_].entx[Nyx-1][0].cotbeta) {
1224 
1225  iylow = Nyx-2;
1226  yxratio = 1.f;
1227 
1228  } else if(abs_cotb_ >= thePixelTemp_[index_id_].entx[0][0].cotbeta) {
1229 
1230  for (i=0; i<Nyx-1; ++i) {
1231 
1232  if( thePixelTemp_[index_id_].entx[i][0].cotbeta <= abs_cotb_ && abs_cotb_ < thePixelTemp_[index_id_].entx[i+1][0].cotbeta) {
1233 
1234  iylow = i;
1235  yxratio = (abs_cotb_ - thePixelTemp_[index_id_].entx[i][0].cotbeta)/(thePixelTemp_[index_id_].entx[i+1][0].cotbeta - thePixelTemp_[index_id_].entx[i][0].cotbeta);
1236  break;
1237  }
1238  }
1239  }
1240 
1241  iyhigh=iylow + 1;
1242 
1243  ilow = 0;
1244  xxratio = 0.f;
1245 
1246  if(cotalpha >= thePixelTemp_[index_id_].entx[0][Nxx-1].cotalpha) {
1247 
1248  ilow = Nxx-2;
1249  xxratio = 1.f;
1250  success_ = false;
1251 
1252  } else {
1253 
1254  if(cotalpha >= thePixelTemp_[index_id_].entx[0][0].cotalpha) {
1255 
1256  for (i=0; i<Nxx-1; ++i) {
1257 
1258  if( thePixelTemp_[index_id_].entx[0][i].cotalpha <= cotalpha && cotalpha < thePixelTemp_[index_id_].entx[0][i+1].cotalpha) {
1259 
1260  ilow = i;
1261  xxratio = (cotalpha - thePixelTemp_[index_id_].entx[0][i].cotalpha)/(thePixelTemp_[index_id_].entx[0][i+1].cotalpha - thePixelTemp_[index_id_].entx[0][i].cotalpha);
1262  break;
1263  }
1264  }
1265  } else { success_ = false; }
1266  }
1267 
1268  ihigh=ilow + 1;
1269 
1270 // Interpolate/store all x-related quantities
1271 
1272  yxratio_ = yxratio;
1273  xxratio_ = xxratio;
1274 
1275 // sxparmax defines the maximum charge for which the parameters xpar are defined (not rescaled by cotbeta)
1276 
1277  sxparmax_ = (1.f - xxratio)*thePixelTemp_[index_id_].entx[imaxx][ilow].sxmax + xxratio*thePixelTemp_[index_id_].entx[imaxx][ihigh].sxmax;
1278  sxmax_ = sxparmax_;
1279  if(thePixelTemp_[index_id_].entx[imaxx][imidy].sxmax != 0.f) {sxmax_=sxmax_/thePixelTemp_[index_id_].entx[imaxx][imidy].sxmax*sxmax;}
1280  symax_ = (1.f - xxratio)*thePixelTemp_[index_id_].entx[imaxx][ilow].symax + xxratio*thePixelTemp_[index_id_].entx[imaxx][ihigh].symax;
1281  if(thePixelTemp_[index_id_].entx[imaxx][imidy].symax != 0.f) {symax_=symax_/thePixelTemp_[index_id_].entx[imaxx][imidy].symax*symax;}
1282  dxone_ = (1.f - xxratio)*thePixelTemp_[index_id_].entx[0][ilow].dxone + xxratio*thePixelTemp_[index_id_].entx[0][ihigh].dxone;
1283  sxone_ = (1.f - xxratio)*thePixelTemp_[index_id_].entx[0][ilow].sxone + xxratio*thePixelTemp_[index_id_].entx[0][ihigh].sxone;
1284  dxtwo_ = (1.f - xxratio)*thePixelTemp_[index_id_].entx[0][ilow].dxtwo + xxratio*thePixelTemp_[index_id_].entx[0][ihigh].dxtwo;
1285  sxtwo_ = (1.f - xxratio)*thePixelTemp_[index_id_].entx[0][ilow].sxtwo + xxratio*thePixelTemp_[index_id_].entx[0][ihigh].sxtwo;
1286  clslenx_ = fminf(thePixelTemp_[index_id_].entx[0][ilow].clslenx, thePixelTemp_[index_id_].entx[0][ihigh].clslenx);
1287  for(i=0; i<2 ; ++i) {
1288  for(j=0; j<5 ; ++j) {
1289  xpar0_[i][j] = thePixelTemp_[index_id_].entx[imaxx][imidy].xpar[i][j];
1290  xparl_[i][j] = thePixelTemp_[index_id_].entx[imaxx][ilow].xpar[i][j];
1291  xparh_[i][j] = thePixelTemp_[index_id_].entx[imaxx][ihigh].xpar[i][j];
1292  }
1293  }
1294 
1295 // pixmax is the maximum allowed pixel charge (used for truncation)
1296 
1297  pixmax_=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iylow][ilow].pixmax + xxratio*thePixelTemp_[index_id_].entx[iylow][ihigh].pixmax)
1298  +yxratio*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].pixmax + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].pixmax);
1299 
1300 
1301  r_qMeas_qTrue_=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iylow][ilow].r_qMeas_qTrue + xxratio*thePixelTemp_[index_id_].entx[iylow][ihigh].r_qMeas_qTrue)
1302  +yxratio*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].r_qMeas_qTrue + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].r_qMeas_qTrue);
1303 
1304 
1305  for(i=0; i<4; ++i) {
1306  xavg_[i]=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iylow][ilow].xavg[i] + xxratio*thePixelTemp_[index_id_].entx[iylow][ihigh].xavg[i])
1307  +yxratio*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].xavg[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].xavg[i]);
1308 
1309  xrms_[i]=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iylow][ilow].xrms[i] + xxratio*thePixelTemp_[index_id_].entx[iylow][ihigh].xrms[i])
1310  +yxratio*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].xrms[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].xrms[i]);
1311 
1312 // xgx0_[i]=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iylow][ilow].xgx0[i] + xxratio*thePixelTemp_[index_id_].entx[iylow][ihigh].xgx0[i])
1313 // +yxratio*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].xgx0[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].xgx0[i]);
1314 
1315 // xgsig_[i]=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iylow][ilow].xgsig[i] + xxratio*thePixelTemp_[index_id_].entx[iylow][ihigh].xgsig[i])
1316 // +yxratio*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].xgsig[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].xgsig[i]);
1317 
1318  xavgc2m_[i]=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iylow][ilow].xavgc2m[i] + xxratio*thePixelTemp_[index_id_].entx[iylow][ihigh].xavgc2m[i])
1319  +yxratio*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].xavgc2m[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].xavgc2m[i]);
1320 
1321  xrmsc2m_[i]=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iylow][ilow].xrmsc2m[i] + xxratio*thePixelTemp_[index_id_].entx[iylow][ihigh].xrmsc2m[i])
1322  +yxratio*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].xrmsc2m[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].xrmsc2m[i]);
1323 
1324 // chi2xavgc2m_[i]=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iylow][ilow].chi2xavgc2m[i] + xxratio*thePixelTemp_[index_id_].entx[iylow][ihigh].chi2xavgc2m[i])
1325 // +yxratio*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].chi2xavgc2m[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].chi2xavgc2m[i]);
1326 
1327 // chi2xminc2m_[i]=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iylow][ilow].chi2xminc2m[i] + xxratio*thePixelTemp_[index_id_].entx[iylow][ihigh].chi2xminc2m[i])
1328 // +yxratio*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].chi2xminc2m[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].chi2xminc2m[i]);
1329 //
1330 // Try new interpolation scheme
1331 //
1332 // chi2xavg_[i]=((1.f - xxratio)*thePixelTemp_[index_id_].entx[imaxx][ilow].chi2xavg[i] + xxratio*thePixelTemp_[index_id_].entx[imaxx][ihigh].chi2xavg[i]);
1333 // if(thePixelTemp_[index_id_].entx[imaxx][imidy].chi2xavg[i] != 0.f) {chi2xavg_[i]=chi2xavg_[i]/thePixelTemp_[index_id_].entx[imaxx][imidy].chi2xavg[i]*chi2xavg[i];}
1334 //
1335 // chi2xmin_[i]=((1.f - xxratio)*thePixelTemp_[index_id_].entx[imaxx][ilow].chi2xmin[i] + xxratio*thePixelTemp_[index_id_].entx[imaxx][ihigh].chi2xmin[i]);
1336 // if(thePixelTemp_[index_id_].entx[imaxx][imidy].chi2xmin[i] != 0.f) {chi2xmin_[i]=chi2xmin_[i]/thePixelTemp_[index_id_].entx[imaxx][imidy].chi2xmin[i]*chi2xmin[i];}
1337 //
1338  chi2xavg_[i]=((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].chi2xavg[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].chi2xavg[i]);
1339  if(thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xavg[i] != 0.f) {chi2xavg_[i]=chi2xavg_[i]/thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xavg[i]*chi2xavg[i];}
1340 
1341  chi2xmin_[i]=((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].chi2xmin[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].chi2xmin[i]);
1342  if(thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xmin[i] != 0.f) {chi2xmin_[i]=chi2xmin_[i]/thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xmin[i]*chi2xmin[i];}
1343 
1344  chi2xavgc2m_[i]=((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].chi2xavgc2m[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].chi2xavgc2m[i]);
1345  if(thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xavgc2m[i] != 0.f) {chi2xavgc2m_[i]=chi2xavgc2m_[i]/thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xavgc2m[i]*chi2xavgc2m[i];}
1346 
1347  chi2xminc2m_[i]=((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].chi2xminc2m[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].chi2xminc2m[i]);
1348  if(thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xminc2m[i] != 0.f) {chi2xminc2m_[i]=chi2xminc2m_[i]/thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xminc2m[i]*chi2xminc2m[i];}
1349 
1350  for(j=0; j<6 ; ++j) {
1351  xflparll_[i][j] = thePixelTemp_[index_id_].entx[iylow][ilow].xflpar[i][j];
1352  xflparlh_[i][j] = thePixelTemp_[index_id_].entx[iylow][ihigh].xflpar[i][j];
1353  xflparhl_[i][j] = thePixelTemp_[index_id_].entx[iyhigh][ilow].xflpar[i][j];
1354  xflparhh_[i][j] = thePixelTemp_[index_id_].entx[iyhigh][ihigh].xflpar[i][j];
1355  }
1356  }
1357 
1358 // Do the spares next
1359 
1360  chi2xavgone_=((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].chi2xavgone + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].chi2xavgone);
1361  if(thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xavgone != 0.f) {chi2xavgone_=chi2xavgone_/thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xavgone*chi2xavgone;}
1362 
1363  chi2xminone_=((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].chi2xminone + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].chi2xminone);
1364  if(thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xminone != 0.f) {chi2xminone_=chi2xminone_/thePixelTemp_[index_id_].entx[iyhigh][imidy].chi2xminone*chi2xminone;}
1365  // for(i=0; i<10; ++i) {
1366 // pxspare[i]=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iylow][ilow].xspare[i] + xxratio*thePixelTemp_[index_id_].entx[iylow][ihigh].xspare[i])
1367 // +yxratio*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].xspare[i] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].xspare[i]);
1368 // }
1369 
1370 // Interpolate and build the x-template
1371 
1372 // qxtempcor corrects the total charge to the actual track angles (not actually needed for the template fits, but useful for Guofan)
1373 
1374  cotbeta0 = thePixelTemp_[index_id_].entx[iyhigh][0].cotbeta;
1375  qxtempcor=std::sqrt((1.f+cotbeta*cotbeta+cotalpha*cotalpha)/(1.f+cotbeta0*cotbeta0+cotalpha*cotalpha));
1376 
1377  for(i=0; i<9; ++i) {
1378  xtemp_[i][0] = 0.f;
1379  xtemp_[i][1] = 0.f;
1380  xtemp_[i][BXM2] = 0.f;
1381  xtemp_[i][BXM1] = 0.f;
1382  for(j=0; j<TXSIZE; ++j) {
1383 // Take next largest x-slice for the x-template (it reduces bias in the forward direction after irradiation)
1384 // xtemp_[i][j+2]=(1.f - xxratio)*thePixelTemp_[index_id_].entx[imaxx][ilow].xtemp[i][j] + xxratio*thePixelTemp_[index_id_].entx[imaxx][ihigh].xtemp[i][j];
1385 // xtemp_[i][j+2]=(1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].xtemp[i][j] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].xtemp[i][j];
1386  xtemp_[i][j+2]=qxtempcor*((1.f - xxratio)*thePixelTemp_[index_id_].entx[iyhigh][ilow].xtemp[i][j] + xxratio*thePixelTemp_[index_id_].entx[iyhigh][ihigh].xtemp[i][j]);
1387  }
1388  }
1389 
1390  lorywidth_ = thePixelTemp_[index_id_].head.lorywidth;
1391  lorxwidth_ = thePixelTemp_[index_id_].head.lorxwidth;
1392  lorybias_ = thePixelTemp_[index_id_].head.lorybias;
1393  lorxbias_ = thePixelTemp_[index_id_].head.lorxbias;
1394  if(locBz > 0.f) {lorywidth_ = -lorywidth_; lorybias_ = -lorybias_;}
1395 
1396 
1397  }
1398 
1399  return success_;
1400 } // interpolate
float yrmsc2m(int i)
1st pass chi2 min search: average y-rms of reconstruction binned in 4 charge bins ...
float cota_current_
current cot alpha
float mpvvav_
most probable charge in Vavilov distribution (not actually for larger kappa)
int i
Definition: DBlmapReader.cc:9
float xavg_[4]
average x-bias of reconstruction binned in 4 charge bins
float fbin_[3]
The QBin definitions in Q_clus/Q_avg.
float chi2xminone()
//!&lt; minimum of x chi^2 for 1 pixel clusters
float clsleny_
y-cluster length of smaller interpolated template in pixels
float pixmax_
maximum pixel charge
float chi2yavgc2m(int i)
1st pass chi2 min search: average y-chisq for merged clusters
float symax()
average pixel signal for y-projection of cluster
float kappavav2()
kappa parameter for 2-cluster Vavilov distribution
float lorybias_
Lorentz y-bias.
float yavg(int i)
average y-bias of reconstruction binned in 4 charge bins
float abs_cotb_
absolute value of cot beta
float xparhy0_[2][5]
projected x-pixel uncertainty parameterization for larger cotbeta (central alpha) ...
float chi2yminc2m_[4]
1st pass chi2 min search: minimum y-chisq for merged clusters
float chi2xavgc2m_[4]
1st pass chi2 min search: average x-chisq for merged clusters
#define TXSIZE
float xrmsc2m_[4]
1st pass chi2 min search: average x-rms of reconstruction binned in 4 charge bins ...
float chi2xmin(int i)
minimum y chi^2 in 4 charge bins
#define BXM1
float mpvvav()
most probable charge in Vavilov distribution (not actually for larger kappa)
float clslenx()
x-size of smaller interpolated template in pixels
assert(m_qm.get())
float sxmax_
average pixel signal for x-projection of cluster
float chi2yminone_
minimum of y chi^2 for 1 pixel clusters
float chi2ymin(int i)
minimum y chi^2 in 4 charge bins
float yratio()
fractional distance in y between cotbeta templates
float chi2xavgone_
average x chi^2 for 1 pixel clusters
float qmin()
minimum cluster charge for valid hit (keeps 99.9% of simulated hits)
float chi2xavg_[4]
average x chi^2 in 4 charge bins
float ytemp_[9][21+4]
templates for y-reconstruction (binned over 5 central pixels)
float mpvvav2()
most probable charge in 2-cluster Vavilov distribution (not actually for larger kappa) ...
float sytwo()
rms for one double-pixel y-clusters
float cotb_current_
current cot beta
float s50_
1/2 of the pixel threshold signal in adc units
float yavgc2m_[4]
1st pass chi2 min search: average y-bias of reconstruction binned in 4 charge bins ...
float chi2xminc2m_[4]
1st pass chi2 min search: minimum x-chisq for merged clusters
float chi2yminone()
//!&lt; minimum of y chi^2 for 1 pixel clusters
float sxone()
rms for one pixel x-clusters
float qavg_
average cluster charge for this set of track angles
#define BYM1
float r_qMeas_qTrue_
ratio of measured to true cluster charges
#define BXM2
float lorxbias_
Lorentz x-bias.
float xflparlh_[4][6]
Aqfl-parameterized x-correction in 4 charge bins for smaller cotbeta, larger cotalpha.
void ytemp(int fybin, int lybin, float ytemplate[41][21+4])
float chi2xavgc2m(int i)
1st pass chi2 min search: average x-chisq for merged clusters
float chi2ymin_[4]
minimum of y chi^2 in 4 charge bins
float dxone()
mean offset/correction for one pixel x-clusters
float sxparmax_
maximum pixel signal for parameterization of x uncertainties
float kappavav2_
kappa parameter for 2-cluster Vavilov distribution
float sigmavav_
&quot;sigma&quot; scale fctor for Vavilov distribution
float kappavav()
kappa parameter for Vavilov distribution
float chi2yavg(int i)
average y chi^2 in 4 charge bins
float yrms(int i)
average y-rms of reconstruction binned in 4 charge bins
T sqrt(T t)
Definition: SSEVec.h:48
float lorywidth_
Lorentz y-width (sign corrected for fpix frame)
float clsleny()
y-size of smaller interpolated template in pixels
Abs< T >::type abs(const T &t)
Definition: Abs.h:22
int j
Definition: DBlmapReader.cc:9
float sigmavav2_
&quot;sigma&quot; scale fctor for 2-cluster Vavilov distribution
#define BYM2
double f[11][100]
float sigmavav2()
&quot;sigma&quot; scale fctor for 2-cluster Vavilov distribution
float yparl_[2][5]
projected y-pixel uncertainty parameterization for smaller cotbeta
float xpar0_[2][5]
projected x-pixel uncertainty parameterization for central cotalpha
float sxtwo()
rms for one double-pixel x-clusters
float ysize_
Pixel y-size.
#define BYM3
float xsize_
Pixel x-size.
float dytwo()
mean offset/correction for one double-pixel y-clusters
float kappavav_
kappa parameter for Vavilov distribution
bool success_
true if cotalpha, cotbeta are inside of the acceptance (dynamically loaded)
float syone_
rms for one pixel y-clusters
float xflparhh_[4][6]
Aqfl-parameterized x-correction in 4 charge bins for larger cotbeta, cotalpha.
float chi2yminc2m(int i)
1st pass chi2 min search: minimum y-chisq for merged clusters
float yflparl_[4][6]
Aqfl-parameterized y-correction in 4 charge bins for smaller cotbeta.
float chi2yavgone_
average y chi^2 for 1 pixel clusters
float dyone_
mean offset/correction for one pixel y-clusters
float yavgc2m(int i)
1st pass chi2 min search: average y-bias of reconstruction binned in 4 charge bins ...
float sxone_
rms for one pixel x-clusters
float dxtwo_
mean offset/correction for one double-pixel x-clusters
float chi2xminc2m(int i)
1st pass chi2 min search: minimum x-chisq for merged clusters
const std::vector< SiPixelTemplateStore > & thePixelTemp_
#define TYSIZE
float qscale_
charge scaling factor
float xparh_[2][5]
projected x-pixel uncertainty parameterization for larger cotalpha
void xtemp(int fxbin, int lxbin, float xtemplate[41][13+4])
float qmin2_
tighter minimum cluster charge for valid hit (keeps 99.8% of simulated hits)
float yrms_[4]
average y-rms of reconstruction binned in 4 charge bins
float xavgc2m_[4]
1st pass chi2 min search: average x-bias of reconstruction binned in 4 charge bins ...
float xflparll_[4][6]
Aqfl-parameterized x-correction in 4 charge bins for smaller cotbeta, cotalpha.
float syone()
rms for one pixel y-clusters
float sytwo_
rms for one double-pixel y-clusters
float qavg_avg_
average of cluster charge less than qavg
float chi2yavgone()
//!&lt; average y chi^2 for 1 pixel clusters
float yxratio()
fractional distance in y between cotalpha templates slices
float chi2xminone_
minimum of x chi^2 for 1 pixel clusters
float qavg()
average cluster charge for this set of track angles
float xparl_[2][5]
projected x-pixel uncertainty parameterization for smaller cotalpha
float sxtwo_
rms for one double-pixel x-clusters
float sxmax()
average pixel signal for x-projection of cluster
float clslenx_
x-cluster length of smaller interpolated template in pixels
float yflparh_[4][6]
Aqfl-parameterized y-correction in 4 charge bins for larger cotbeta.
float xparly0_[2][5]
projected x-pixel uncertainty parameterization for smaller cotbeta (central alpha) ...
float xrms_[4]
average x-rms of reconstruction binned in 4 charge bins
float symax_
average pixel signal for y-projection of cluster
float yparh_[2][5]
projected y-pixel uncertainty parameterization for larger cotbeta
float chi2xavgone()
//!&lt; average x chi^2 for 1 pixel clusters
float xxratio_
fractional distance in x between cotalpha templates
float chi2yavg_[4]
average y chi^2 in 4 charge bins
float sigmavav()
&quot;sigma&quot; scale fctor for Vavilov distribution
float xflparhl_[4][6]
Aqfl-parameterized x-correction in 4 charge bins for larger cotbeta, smaller cotalpha.
float yxratio_
fractional distance in y between x-slices of cotalpha templates
float qmin_
minimum cluster charge for valid hit (keeps 99.9% of simulated hits)
float dxone_
mean offset/correction for one pixel x-clusters
float lorxwidth_
Lorentz x-width.
float chi2xavg(int i)
averaage x chi^2 in 4 charge bins
float yavg_[4]
average y-bias of reconstruction binned in 4 charge bins
int index_id_
current index
float mpvvav2_
most probable charge in 2-cluster Vavilov distribution (not actually for larger kappa) ...
float syparmax_
maximum pixel signal for parameterization of y uncertainties
float yrmsc2m_[4]
1st pass chi2 min search: average y-rms of reconstruction binned in 4 charge bins ...
float xtemp_[9][13+4]
templates for x-reconstruction (binned over 5 central pixels)
float dyone()
mean offset/correction for one pixel y-clusters
float dytwo_
mean offset/correction for one double-pixel y-clusters
float dxtwo()
mean offset/correction for one double-pixel x-clusters
int id_current_
current id
float zsize_
Pixel z-size (thickness)
float yratio_
fractional distance in y between cotbeta templates
float chi2yavgc2m_[4]
1st pass chi2 min search: average y-chisq for merged clusters
float xxratio()
fractional distance in x between cotalpha templates
float chi2xmin_[4]
minimum of x chi^2 in 4 charge bins
bool SiPixelTemplate::interpolate ( int  id,
float  cotalpha,
float  cotbeta 
)

Interpolate input alpha and beta angles to produce a working template for each individual hit.

Parameters
id- (input) index of the template to use
cotalpha- (input) the cotangent of the alpha track angle (see CMS IN 2004/014)
cotbeta- (input) the cotangent of the beta track angle (see CMS IN 2004/014)

Definition at line 1412 of file SiPixelTemplate.cc.

References f, and interpolate().

1413 {
1414  // Interpolate for a new set of track angles
1415 
1416  // Local variables
1417  float locBz;
1418  locBz = -1.f;
1419  if(cotbeta < 0.f) {locBz = -locBz;}
1420  return SiPixelTemplate::interpolate(id, cotalpha, cotbeta, locBz);
1421 }
bool interpolate(int id, float cotalpha, float cotbeta, float locBz)
double f[11][100]
float SiPixelTemplate::kappavav ( )
inline

kappa parameter for Vavilov distribution

Definition at line 507 of file SiPixelTemplate.h.

References kappavav_.

float SiPixelTemplate::kappavav2 ( )
inline

kappa parameter for 2-cluster Vavilov distribution

Definition at line 510 of file SiPixelTemplate.h.

References kappavav2_.

float SiPixelTemplate::lorxbias ( )
inline

signed lorentz x-width (microns)

Definition at line 504 of file SiPixelTemplate.h.

References lorxbias_.

Referenced by PixelCPETemplateReco::localPosition().

float SiPixelTemplate::lorxwidth ( )
inline

signed lorentz x-width (microns)

Definition at line 500 of file SiPixelTemplate.h.

References lorxwidth_.

float SiPixelTemplate::lorybias ( )
inline

signed lorentz y-width (microns)

Definition at line 503 of file SiPixelTemplate.h.

References lorybias_.

Referenced by PixelCPETemplateReco::localPosition().

float SiPixelTemplate::lorywidth ( )
inline

signed lorentz y-width (microns)

Definition at line 499 of file SiPixelTemplate.h.

References lorywidth_.

float SiPixelTemplate::mpvvav ( )
inline

most probable charge in Vavilov distribution (not actually for larger kappa)

Definition at line 505 of file SiPixelTemplate.h.

References mpvvav_.

float SiPixelTemplate::mpvvav2 ( )
inline

most probable charge in 2-cluster Vavilov distribution (not actually for larger kappa)

Definition at line 508 of file SiPixelTemplate.h.

References mpvvav2_.

float SiPixelTemplate::pixmax ( )
inline

maximum pixel charge

Definition at line 320 of file SiPixelTemplate.h.

References pixmax_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

void SiPixelTemplate::postInit ( std::vector< SiPixelTemplateStore > &  thePixelTemp_)
static

Definition at line 944 of file SiPixelTemplate.cc.

References citk::for().

944  {
945  /*
946  std::cout << "SiPixelTemplate size " << thePixelTemp_.size() << std::endl;
947 #ifndef SI_PIXEL_TEMPLATE_USE_BOOST
948  std::cout <<"uses C arrays" << std::endl;
949 #endif
950 
951  int i=0;
952  for (auto & templ : thePixelTemp_) {
953  std::cout << i <<':' << templ.head.ID << ' ' << templ.head.NTy <<','<< templ.head.NTyx <<','<< templ.head.NTxx << std::endl;
954  for ( auto iy=1; iy<templ.head.NTy; ++iy ) { auto & ent = templ.enty[iy]; std::cout << ent.cotbeta <<',' << ent.cotbeta-templ.enty[iy-1].cotbeta << ' '; }
955  std::cout << std::endl;
956  for ( auto ix=1; ix<templ.head.NTxx; ++ix ){ auto & ent = templ.entx[0][ix]; std::cout << ent.cotalpha <<','<< ent.cotalpha-templ.entx[0][ix-1].cotalpha << ' ';}
957  std::cout << std::endl;
958  ++i;
959  }
960  */
961 
962  for (auto & templ : thePixelTemp_) {
963  for ( auto iy=0; iy<templ.head.NTy; ++iy ) templ.cotbetaY[iy]=templ.enty[iy].cotbeta;
964  for ( auto iy=0; iy<templ.head.NTyx; ++iy ) templ.cotbetaX[iy]= templ.entx[iy][0].cotbeta;
965  for ( auto ix=0; ix<templ.head.NTxx; ++ix ) templ.cotalphaX[ix]=templ.entx[0][ix].cotalpha;
966  }
967 
968 }
const std::vector< SiPixelTemplateStore > & thePixelTemp_
for(const auto &isodef:isoDefs)
bool SiPixelTemplate::pushfile ( int  filenum,
std::vector< SiPixelTemplateStore > &  thePixelTemp_ 
)
static

This routine initializes the global template structures from an external file template_summary_zpNNNN where NNNN are four digits of filenum.

Parameters
filenum- an integer NNNN used in the filename template_summary_zpNNNN

Definition at line 124 of file SiPixelTemplate.cc.

References SiPixelTemplateEntry::alpha, SiPixelTemplateEntry::beta, SiPixelTemplateHeader::Bfield, EnergyCorrector::c, SiPixelTemplateEntry::chi2xavg, SiPixelTemplateEntry::chi2xavgc2m, SiPixelTemplateEntry::chi2xavgone, SiPixelTemplateEntry::chi2xmin, SiPixelTemplateEntry::chi2xminc2m, SiPixelTemplateEntry::chi2xminone, SiPixelTemplateEntry::chi2yavg, SiPixelTemplateEntry::chi2yavgc2m, SiPixelTemplateEntry::chi2yavgone, SiPixelTemplateEntry::chi2ymin, SiPixelTemplateEntry::chi2yminc2m, SiPixelTemplateEntry::chi2yminone, SiPixelTemplateEntry::clslenx, SiPixelTemplateEntry::clsleny, SiPixelTemplateEntry::costrk, SiPixelTemplateEntry::cotalpha, SiPixelTemplateEntry::cotbeta, SiPixelTemplateHeader::Dtype, SiPixelTemplateEntry::dxone, SiPixelTemplateEntry::dxtwo, SiPixelTemplateEntry::dyone, SiPixelTemplateEntry::dytwo, ENDL, SiPixelTemplateStore::entx, SiPixelTemplateStore::enty, SiPixelTemplateHeader::fbin, mergeVDriftHistosByStation::file, SiPixelTemplateHeader::fluence, SiPixelTemplateEntry::fracxone, SiPixelTemplateEntry::fracxtwo, SiPixelTemplateEntry::fracyone, SiPixelTemplateEntry::fracytwo, SiPixelTemplateStore::head, i, SiPixelTemplateHeader::ID, recoMuon::in, create_public_lumi_plots::in_file, j, relval_steps::k, SiPixelTemplateEntry::kappavav, SiPixelTemplateEntry::kappavav2, prof2calltree::l, LOGERROR, LOGINFO, SiPixelTemplateHeader::lorxbias, SiPixelTemplateHeader::lorxwidth, SiPixelTemplateHeader::lorybias, SiPixelTemplateHeader::lorywidth, SiPixelTemplateEntry::mpvvav, SiPixelTemplateEntry::mpvvav2, SiPixelTemplateHeader::NTxx, SiPixelTemplateHeader::NTy, SiPixelTemplateHeader::NTyx, SiPixelTemplateEntry::pixmax, SiPixelTemplateEntry::qavg, SiPixelTemplateEntry::qavg_avg, SiPixelTemplateEntry::qbfrac, SiPixelTemplateEntry::qmin, SiPixelTemplateEntry::qmin2, SiPixelTemplateHeader::qscale, SiPixelTemplateEntry::r_qMeas_qTrue, SiPixelTemplateEntry::runnum, SiPixelTemplateHeader::s50, SiPixelTemplateEntry::sigmavav, SiPixelTemplateEntry::sigmavav2, SiPixelTemplateEntry::spare, SiPixelTemplateHeader::ss50, AlCaHLTBitMon_QueryRunRegistry::string, SiPixelTemplateEntry::sxmax, SiPixelTemplateEntry::sxone, SiPixelTemplateEntry::sxtwo, SiPixelTemplateEntry::symax, SiPixelTemplateEntry::syone, SiPixelTemplateEntry::sytwo, SiPixelTemplateHeader::temperature, SiPixelTemplateHeader::templ_version, SiPixelTemplateHeader::title, TXSIZE, TYSIZE, SiPixelTemplateHeader::Vbias, SiPixelTemplateEntry::xavg, SiPixelTemplateEntry::xavgc2m, SiPixelTemplateEntry::xavggen, SiPixelTemplateEntry::xflpar, SiPixelTemplateEntry::xgsig, SiPixelTemplateEntry::xgsiggen, SiPixelTemplateEntry::xgx0, SiPixelTemplateEntry::xgx0gen, SiPixelTemplateEntry::xpar, SiPixelTemplateEntry::xrms, SiPixelTemplateEntry::xrmsc2m, SiPixelTemplateEntry::xrmsgen, SiPixelTemplateHeader::xsize, SiPixelTemplateEntry::xtemp, SiPixelTemplateEntry::yavg, SiPixelTemplateEntry::yavgc2m, SiPixelTemplateEntry::yavggen, SiPixelTemplateEntry::yflpar, SiPixelTemplateEntry::ygsig, SiPixelTemplateEntry::ygsiggen, SiPixelTemplateEntry::ygx0, SiPixelTemplateEntry::ygx0gen, SiPixelTemplateEntry::ypar, SiPixelTemplateEntry::yrms, SiPixelTemplateEntry::yrmsc2m, SiPixelTemplateEntry::yrmsgen, SiPixelTemplateHeader::ysize, SiPixelTemplateEntry::ytemp, and SiPixelTemplateHeader::zsize.

Referenced by PixelCPEGeneric::PixelCPEGeneric(), PixelCPETemplateReco::PixelCPETemplateReco(), SiPixelGaussianSmearingRecHitConverterAlgorithm::SiPixelGaussianSmearingRecHitConverterAlgorithm(), and TrackClusterSplitter::TrackClusterSplitter().

125 {
126  // Add template stored in external file numbered filenum to theTemplateStore
127 
128  // Local variables
129  int i, j, k, l;
130  float qavg_avg;
131  const char *tempfile;
132  // char title[80]; remove this
133  char c;
134  const int code_version={17};
135 
136 
137 
138  // Create a filename for this run
139 
140  std::ostringstream tout;
141 
142  // Create different path in CMSSW than standalone
143 
144 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
145  tout << "CalibTracker/SiPixelESProducers/data/template_summary_zp"
146  << std::setw(4) << std::setfill('0') << std::right << filenum << ".out" << std::ends;
147  std::string tempf = tout.str();
148  edm::FileInPath file( tempf.c_str() );
149  tempfile = (file.fullPath()).c_str();
150 #else
151  tout << "template_summary_zp" << std::setw(4) << std::setfill('0') << std::right << filenum << ".out" << std::ends;
152  std::string tempf = tout.str();
153  tempfile = tempf.c_str();
154 #endif
155 
156  // open the template file
157 
158  std::ifstream in_file(tempfile, std::ios::in);
159 
160  if(in_file.is_open()) {
161 
162  // Create a local template storage entry
163 
164  SiPixelTemplateStore theCurrentTemp;
165 
166  // Read-in a header string first and print it
167 
168  for (i=0; (c=in_file.get()) != '\n'; ++i) {
169  if(i < 79) {theCurrentTemp.head.title[i] = c;}
170  }
171  if(i > 78) {i=78;}
172  theCurrentTemp.head.title[i+1] ='\0';
173  LOGINFO("SiPixelTemplate") << "Loading Pixel Template File - " << theCurrentTemp.head.title << ENDL;
174 
175  // next, the header information
176 
177  in_file >> theCurrentTemp.head.ID >> theCurrentTemp.head.templ_version >> theCurrentTemp.head.Bfield >> theCurrentTemp.head.NTy >> theCurrentTemp.head.NTyx >> theCurrentTemp.head.NTxx
178  >> theCurrentTemp.head.Dtype >> theCurrentTemp.head.Vbias >> theCurrentTemp.head.temperature >> theCurrentTemp.head.fluence >> theCurrentTemp.head.qscale
179  >> theCurrentTemp.head.s50 >> theCurrentTemp.head.lorywidth >> theCurrentTemp.head.lorxwidth >> theCurrentTemp.head.ysize >> theCurrentTemp.head.xsize >> theCurrentTemp.head.zsize;
180 
181  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 0A, no template load" << ENDL; return false;}
182 
183  if(theCurrentTemp.head.templ_version > 17) {
184  in_file >> theCurrentTemp.head.ss50 >> theCurrentTemp.head.lorybias
185  >> theCurrentTemp.head.lorxbias >> theCurrentTemp.head.fbin[0]
186  >> theCurrentTemp.head.fbin[1] >> theCurrentTemp.head.fbin[2];
187 
188  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 0B, no template load"
189  << ENDL; return false;}
190  } else {
191  theCurrentTemp.head.ss50 = theCurrentTemp.head.s50;
192  theCurrentTemp.head.lorybias = theCurrentTemp.head.lorywidth/2.f;
193  theCurrentTemp.head.lorxbias = theCurrentTemp.head.lorxwidth/2.f;
194  theCurrentTemp.head.fbin[0] = fbin0;
195  theCurrentTemp.head.fbin[1] = fbin1;
196  theCurrentTemp.head.fbin[2] = fbin2;
197  }
198 
199  LOGINFO("SiPixelTemplate") << "Template ID = " << theCurrentTemp.head.ID << ", Template Version " << theCurrentTemp.head.templ_version << ", Bfield = " << theCurrentTemp.head.Bfield
200  << ", NTy = " << theCurrentTemp.head.NTy << ", NTyx = " << theCurrentTemp.head.NTyx<< ", NTxx = " << theCurrentTemp.head.NTxx << ", Dtype = " << theCurrentTemp.head.Dtype
201  << ", Bias voltage " << theCurrentTemp.head.Vbias << ", temperature "
202  << theCurrentTemp.head.temperature << ", fluence " << theCurrentTemp.head.fluence << ", Q-scaling factor " << theCurrentTemp.head.qscale
203  << ", 1/2 multi dcol threshold " << theCurrentTemp.head.s50 << ", 1/2 single dcol threshold " << theCurrentTemp.head.ss50
204  << ", y Lorentz Width " << theCurrentTemp.head.lorywidth << ", y Lorentz Bias " << theCurrentTemp.head.lorybias
205  << ", x Lorentz width " << theCurrentTemp.head.lorxwidth << ", x Lorentz Bias " << theCurrentTemp.head.lorxbias
206  << ", Q/Q_avg fractions for Qbin defs " << theCurrentTemp.head.fbin[0] << ", " << theCurrentTemp.head.fbin[1]
207  << ", " << theCurrentTemp.head.fbin[2]
208  << ", pixel x-size " << theCurrentTemp.head.xsize << ", y-size " << theCurrentTemp.head.ysize << ", zsize " << theCurrentTemp.head.zsize << ENDL;
209 
210 
211  if(theCurrentTemp.head.templ_version < code_version) {LOGERROR("SiPixelTemplate") << "code expects version " << code_version << " finds "<< theCurrentTemp.head.templ_version <<", no template load" << ENDL; return false;}
212 
213 #ifdef SI_PIXEL_TEMPLATE_USE_BOOST
214 
215 // next, layout the 1-d/2-d structures needed to store template
216 
217  theCurrentTemp.enty.resize(boost::extents[theCurrentTemp.head.NTy]);
218 
219  theCurrentTemp.entx.resize(boost::extents[theCurrentTemp.head.NTyx][theCurrentTemp.head.NTxx]);
220 
221 #endif
222 
223 // next, loop over all y-angle entries
224 
225  for (i=0; i < theCurrentTemp.head.NTy; ++i) {
226 
227  in_file >> theCurrentTemp.enty[i].runnum >> theCurrentTemp.enty[i].costrk[0]
228  >> theCurrentTemp.enty[i].costrk[1] >> theCurrentTemp.enty[i].costrk[2];
229 
230  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 1, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
231 
232  // Calculate the alpha, beta, and cot(beta) for this entry
233 
234  theCurrentTemp.enty[i].alpha = static_cast<float>(atan2((double)theCurrentTemp.enty[i].costrk[2], (double)theCurrentTemp.enty[i].costrk[0]));
235 
236  theCurrentTemp.enty[i].cotalpha = theCurrentTemp.enty[i].costrk[0]/theCurrentTemp.enty[i].costrk[2];
237 
238  theCurrentTemp.enty[i].beta = static_cast<float>(atan2((double)theCurrentTemp.enty[i].costrk[2], (double)theCurrentTemp.enty[i].costrk[1]));
239 
240  theCurrentTemp.enty[i].cotbeta = theCurrentTemp.enty[i].costrk[1]/theCurrentTemp.enty[i].costrk[2];
241 
242  in_file >> theCurrentTemp.enty[i].qavg >> theCurrentTemp.enty[i].pixmax >> theCurrentTemp.enty[i].symax >> theCurrentTemp.enty[i].dyone
243  >> theCurrentTemp.enty[i].syone >> theCurrentTemp.enty[i].sxmax >> theCurrentTemp.enty[i].dxone >> theCurrentTemp.enty[i].sxone;
244 
245  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 2, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
246 
247  in_file >> theCurrentTemp.enty[i].dytwo >> theCurrentTemp.enty[i].sytwo >> theCurrentTemp.enty[i].dxtwo
248  >> theCurrentTemp.enty[i].sxtwo >> theCurrentTemp.enty[i].qmin >> theCurrentTemp.enty[i].clsleny >> theCurrentTemp.enty[i].clslenx;
249 
250  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 3, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
251 
252 
253  if(theCurrentTemp.enty[i].qmin <= 0.) {LOGERROR("SiPixelTemplate") << "Error in template ID " << theCurrentTemp.head.ID << " qmin = " << theCurrentTemp.enty[i].qmin << ", run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
254 
255 
256  for (j=0; j<2; ++j) {
257 
258  in_file >> theCurrentTemp.enty[i].ypar[j][0] >> theCurrentTemp.enty[i].ypar[j][1]
259  >> theCurrentTemp.enty[i].ypar[j][2] >> theCurrentTemp.enty[i].ypar[j][3] >> theCurrentTemp.enty[i].ypar[j][4];
260 
261  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 4, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
262 
263  }
264 
265  for (j=0; j<9; ++j) {
266 
267  for (k=0; k<TYSIZE; ++k) {in_file >> theCurrentTemp.enty[i].ytemp[j][k];}
268 
269  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 5, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
270  }
271 
272  for (j=0; j<2; ++j) {
273 
274  in_file >> theCurrentTemp.enty[i].xpar[j][0] >> theCurrentTemp.enty[i].xpar[j][1]
275  >> theCurrentTemp.enty[i].xpar[j][2] >> theCurrentTemp.enty[i].xpar[j][3] >> theCurrentTemp.enty[i].xpar[j][4];
276 
277  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 6, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
278 
279  }
280 
281  qavg_avg = 0.f;
282  for (j=0; j<9; ++j) {
283 
284  for (k=0; k<TXSIZE; ++k) {in_file >> theCurrentTemp.enty[i].xtemp[j][k]; qavg_avg += theCurrentTemp.enty[i].xtemp[j][k];}
285 
286  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 7, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
287  }
288  theCurrentTemp.enty[i].qavg_avg = qavg_avg/9.;
289 
290  for (j=0; j<4; ++j) {
291 
292  in_file >> theCurrentTemp.enty[i].yavg[j] >> theCurrentTemp.enty[i].yrms[j] >> theCurrentTemp.enty[i].ygx0[j] >> theCurrentTemp.enty[i].ygsig[j];
293 
294  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 8, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
295  }
296 
297  for (j=0; j<4; ++j) {
298 
299  in_file >> theCurrentTemp.enty[i].yflpar[j][0] >> theCurrentTemp.enty[i].yflpar[j][1] >> theCurrentTemp.enty[i].yflpar[j][2]
300  >> theCurrentTemp.enty[i].yflpar[j][3] >> theCurrentTemp.enty[i].yflpar[j][4] >> theCurrentTemp.enty[i].yflpar[j][5];
301 
302  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 9, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
303  }
304 
305  for (j=0; j<4; ++j) {
306 
307  in_file >> theCurrentTemp.enty[i].xavg[j] >> theCurrentTemp.enty[i].xrms[j] >> theCurrentTemp.enty[i].xgx0[j] >> theCurrentTemp.enty[i].xgsig[j];
308 
309  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 10, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
310  }
311 
312  for (j=0; j<4; ++j) {
313 
314  in_file >> theCurrentTemp.enty[i].xflpar[j][0] >> theCurrentTemp.enty[i].xflpar[j][1] >> theCurrentTemp.enty[i].xflpar[j][2]
315  >> theCurrentTemp.enty[i].xflpar[j][3] >> theCurrentTemp.enty[i].xflpar[j][4] >> theCurrentTemp.enty[i].xflpar[j][5];
316 
317  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 11, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
318  }
319 
320  for (j=0; j<4; ++j) {
321 
322  in_file >> theCurrentTemp.enty[i].chi2yavg[j] >> theCurrentTemp.enty[i].chi2ymin[j] >> theCurrentTemp.enty[i].chi2xavg[j] >> theCurrentTemp.enty[i].chi2xmin[j];
323 
324  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 12, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
325  }
326 
327  for (j=0; j<4; ++j) {
328 
329  in_file >> theCurrentTemp.enty[i].yavgc2m[j] >> theCurrentTemp.enty[i].yrmsc2m[j] >> theCurrentTemp.enty[i].chi2yavgc2m[j] >> theCurrentTemp.enty[i].chi2yminc2m[j];
330 
331  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 13, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
332  }
333 
334  for (j=0; j<4; ++j) {
335 
336  in_file >> theCurrentTemp.enty[i].xavgc2m[j] >> theCurrentTemp.enty[i].xrmsc2m[j] >> theCurrentTemp.enty[i].chi2xavgc2m[j] >> theCurrentTemp.enty[i].chi2xminc2m[j];
337 
338  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 14, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
339  }
340 
341  for (j=0; j<4; ++j) {
342 
343  in_file >> theCurrentTemp.enty[i].yavggen[j] >> theCurrentTemp.enty[i].yrmsgen[j] >> theCurrentTemp.enty[i].ygx0gen[j] >> theCurrentTemp.enty[i].ygsiggen[j];
344 
345  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 14a, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
346  }
347 
348  for (j=0; j<4; ++j) {
349 
350  in_file >> theCurrentTemp.enty[i].xavggen[j] >> theCurrentTemp.enty[i].xrmsgen[j] >> theCurrentTemp.enty[i].xgx0gen[j] >> theCurrentTemp.enty[i].xgsiggen[j];
351 
352  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 14b, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
353  }
354 
355  in_file >> theCurrentTemp.enty[i].chi2yavgone >> theCurrentTemp.enty[i].chi2yminone >> theCurrentTemp.enty[i].chi2xavgone >> theCurrentTemp.enty[i].chi2xminone >> theCurrentTemp.enty[i].qmin2
356  >> theCurrentTemp.enty[i].mpvvav >> theCurrentTemp.enty[i].sigmavav >> theCurrentTemp.enty[i].kappavav >> theCurrentTemp.enty[i].r_qMeas_qTrue >> theCurrentTemp.enty[i].spare[0];
357 
358  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 15, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
359 
360  in_file >> theCurrentTemp.enty[i].mpvvav2 >> theCurrentTemp.enty[i].sigmavav2 >> theCurrentTemp.enty[i].kappavav2 >> theCurrentTemp.enty[i].qbfrac[0] >> theCurrentTemp.enty[i].qbfrac[1]
361  >> theCurrentTemp.enty[i].qbfrac[2] >> theCurrentTemp.enty[i].fracyone >> theCurrentTemp.enty[i].fracxone >> theCurrentTemp.enty[i].fracytwo >> theCurrentTemp.enty[i].fracxtwo;
362  // theCurrentTemp.enty[i].qbfrac[3] = 1. - theCurrentTemp.enty[i].qbfrac[0] - theCurrentTemp.enty[i].qbfrac[1] - theCurrentTemp.enty[i].qbfrac[2];
363 
364  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 16, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
365 
366  }
367 
368  // next, loop over all barrel x-angle entries
369 
370  for (k=0; k < theCurrentTemp.head.NTyx; ++k) {
371 
372  for (i=0; i < theCurrentTemp.head.NTxx; ++i) {
373 
374  in_file >> theCurrentTemp.entx[k][i].runnum >> theCurrentTemp.entx[k][i].costrk[0]
375  >> theCurrentTemp.entx[k][i].costrk[1] >> theCurrentTemp.entx[k][i].costrk[2];
376 
377  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 17, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
378 
379  // Calculate the alpha, beta, and cot(beta) for this entry
380 
381  theCurrentTemp.entx[k][i].alpha = static_cast<float>(atan2((double)theCurrentTemp.entx[k][i].costrk[2], (double)theCurrentTemp.entx[k][i].costrk[0]));
382 
383  theCurrentTemp.entx[k][i].cotalpha = theCurrentTemp.entx[k][i].costrk[0]/theCurrentTemp.entx[k][i].costrk[2];
384 
385  theCurrentTemp.entx[k][i].beta = static_cast<float>(atan2((double)theCurrentTemp.entx[k][i].costrk[2], (double)theCurrentTemp.entx[k][i].costrk[1]));
386 
387  theCurrentTemp.entx[k][i].cotbeta = theCurrentTemp.entx[k][i].costrk[1]/theCurrentTemp.entx[k][i].costrk[2];
388 
389  in_file >> theCurrentTemp.entx[k][i].qavg >> theCurrentTemp.entx[k][i].pixmax >> theCurrentTemp.entx[k][i].symax >> theCurrentTemp.entx[k][i].dyone
390  >> theCurrentTemp.entx[k][i].syone >> theCurrentTemp.entx[k][i].sxmax >> theCurrentTemp.entx[k][i].dxone >> theCurrentTemp.entx[k][i].sxone;
391 
392  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 18, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
393 
394  in_file >> theCurrentTemp.entx[k][i].dytwo >> theCurrentTemp.entx[k][i].sytwo >> theCurrentTemp.entx[k][i].dxtwo
395  >> theCurrentTemp.entx[k][i].sxtwo >> theCurrentTemp.entx[k][i].qmin >> theCurrentTemp.entx[k][i].clsleny >> theCurrentTemp.entx[k][i].clslenx;
396  // >> theCurrentTemp.entx[k][i].mpvvav >> theCurrentTemp.entx[k][i].sigmavav >> theCurrentTemp.entx[k][i].kappavav;
397 
398  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 19, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
399 
400  for (j=0; j<2; ++j) {
401 
402  in_file >> theCurrentTemp.entx[k][i].ypar[j][0] >> theCurrentTemp.entx[k][i].ypar[j][1]
403  >> theCurrentTemp.entx[k][i].ypar[j][2] >> theCurrentTemp.entx[k][i].ypar[j][3] >> theCurrentTemp.entx[k][i].ypar[j][4];
404 
405  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 20, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
406  }
407 
408  for (j=0; j<9; ++j) {
409 
410  for (l=0; l<TYSIZE; ++l) {in_file >> theCurrentTemp.entx[k][i].ytemp[j][l];}
411 
412  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 21, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
413  }
414 
415  for (j=0; j<2; ++j) {
416 
417  in_file >> theCurrentTemp.entx[k][i].xpar[j][0] >> theCurrentTemp.entx[k][i].xpar[j][1]
418  >> theCurrentTemp.entx[k][i].xpar[j][2] >> theCurrentTemp.entx[k][i].xpar[j][3] >> theCurrentTemp.entx[k][i].xpar[j][4];
419 
420 
421  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 22, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
422  }
423 
424  qavg_avg = 0.f;
425  for (j=0; j<9; ++j) {
426 
427  for (l=0; l<TXSIZE; ++l) {in_file >> theCurrentTemp.entx[k][i].xtemp[j][l]; qavg_avg += theCurrentTemp.entx[k][i].xtemp[j][l];}
428 
429  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 23, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
430  }
431  theCurrentTemp.entx[k][i].qavg_avg = qavg_avg/9.;
432 
433  for (j=0; j<4; ++j) {
434 
435  in_file >> theCurrentTemp.entx[k][i].yavg[j] >> theCurrentTemp.entx[k][i].yrms[j] >> theCurrentTemp.entx[k][i].ygx0[j] >> theCurrentTemp.entx[k][i].ygsig[j];
436 
437  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 24, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
438  }
439 
440  for (j=0; j<4; ++j) {
441 
442  in_file >> theCurrentTemp.entx[k][i].yflpar[j][0] >> theCurrentTemp.entx[k][i].yflpar[j][1] >> theCurrentTemp.entx[k][i].yflpar[j][2]
443  >> theCurrentTemp.entx[k][i].yflpar[j][3] >> theCurrentTemp.entx[k][i].yflpar[j][4] >> theCurrentTemp.entx[k][i].yflpar[j][5];
444 
445  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 25, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
446  }
447 
448  for (j=0; j<4; ++j) {
449 
450  in_file >> theCurrentTemp.entx[k][i].xavg[j] >> theCurrentTemp.entx[k][i].xrms[j] >> theCurrentTemp.entx[k][i].xgx0[j] >> theCurrentTemp.entx[k][i].xgsig[j];
451 
452  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 26, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
453  }
454 
455  for (j=0; j<4; ++j) {
456 
457  in_file >> theCurrentTemp.entx[k][i].xflpar[j][0] >> theCurrentTemp.entx[k][i].xflpar[j][1] >> theCurrentTemp.entx[k][i].xflpar[j][2]
458  >> theCurrentTemp.entx[k][i].xflpar[j][3] >> theCurrentTemp.entx[k][i].xflpar[j][4] >> theCurrentTemp.entx[k][i].xflpar[j][5];
459 
460  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 27, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
461  }
462 
463  for (j=0; j<4; ++j) {
464 
465  in_file >> theCurrentTemp.entx[k][i].chi2yavg[j] >> theCurrentTemp.entx[k][i].chi2ymin[j] >> theCurrentTemp.entx[k][i].chi2xavg[j] >> theCurrentTemp.entx[k][i].chi2xmin[j];
466 
467  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 28, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
468  }
469 
470  for (j=0; j<4; ++j) {
471 
472  in_file >> theCurrentTemp.entx[k][i].yavgc2m[j] >> theCurrentTemp.entx[k][i].yrmsc2m[j] >> theCurrentTemp.entx[k][i].chi2yavgc2m[j] >> theCurrentTemp.entx[k][i].chi2yminc2m[j];
473 
474  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 29, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
475  }
476 
477  for (j=0; j<4; ++j) {
478 
479  in_file >> theCurrentTemp.entx[k][i].xavgc2m[j] >> theCurrentTemp.entx[k][i].xrmsc2m[j] >> theCurrentTemp.entx[k][i].chi2xavgc2m[j] >> theCurrentTemp.entx[k][i].chi2xminc2m[j];
480 
481  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 30, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
482  }
483 
484  for (j=0; j<4; ++j) {
485 
486  in_file >> theCurrentTemp.entx[k][i].yavggen[j] >> theCurrentTemp.entx[k][i].yrmsgen[j] >> theCurrentTemp.entx[k][i].ygx0gen[j] >> theCurrentTemp.entx[k][i].ygsiggen[j];
487 
488  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 30a, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
489  }
490 
491  for (j=0; j<4; ++j) {
492 
493  in_file >> theCurrentTemp.entx[k][i].xavggen[j] >> theCurrentTemp.entx[k][i].xrmsgen[j] >> theCurrentTemp.entx[k][i].xgx0gen[j] >> theCurrentTemp.entx[k][i].xgsiggen[j];
494 
495  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 30b, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
496  }
497 
498  in_file >> theCurrentTemp.entx[k][i].chi2yavgone >> theCurrentTemp.entx[k][i].chi2yminone >> theCurrentTemp.entx[k][i].chi2xavgone >> theCurrentTemp.entx[k][i].chi2xminone >> theCurrentTemp.entx[k][i].qmin2
499  >> theCurrentTemp.entx[k][i].mpvvav >> theCurrentTemp.entx[k][i].sigmavav >> theCurrentTemp.entx[k][i].kappavav >> theCurrentTemp.entx[k][i].r_qMeas_qTrue >> theCurrentTemp.entx[k][i].spare[0];
500 
501  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 31, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
502 
503  in_file >> theCurrentTemp.entx[k][i].mpvvav2 >> theCurrentTemp.entx[k][i].sigmavav2 >> theCurrentTemp.entx[k][i].kappavav2 >> theCurrentTemp.entx[k][i].qbfrac[0] >> theCurrentTemp.entx[k][i].qbfrac[1]
504  >> theCurrentTemp.entx[k][i].qbfrac[2] >> theCurrentTemp.entx[k][i].fracyone >> theCurrentTemp.entx[k][i].fracxone >> theCurrentTemp.entx[k][i].fracytwo >> theCurrentTemp.entx[k][i].fracxtwo;
505  // theCurrentTemp.entx[k][i].qbfrac[3] = 1. - theCurrentTemp.entx[k][i].qbfrac[0] - theCurrentTemp.entx[k][i].qbfrac[1] - theCurrentTemp.entx[k][i].qbfrac[2];
506 
507  if(in_file.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 32, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
508 
509  }
510  }
511 
512 
513  in_file.close();
514 
515  // Add this template to the store
516 
517  thePixelTemp_.push_back(theCurrentTemp);
518 
520  return true;
521 
522  } else {
523 
524  // If file didn't open, report this
525 
526  LOGERROR("SiPixelTemplate") << "Error opening File" << tempfile << ENDL;
527  return false;
528 
529  }
530 
531 } // TempInit
float qavg_avg
average cluster charge of clusters that are less than qavg (normalize 2-D simple templates) ...
int runnum
&lt; Basic template entry corresponding to a single set of track angles
float xflpar[4][6]
Aqfl-parameterized x-correction in 4 charge bins.
int i
Definition: DBlmapReader.cc:9
float xrms[4]
average x-rms of reconstruction binned in 4 charge bins
#define LOGERROR(x)
float clslenx
cluster x-length in pixels at signal height sxmax/2
float qavg
average cluster charge for this set of track angles (now includes threshold effects) ...
float syone
rms for one pixel y-clusters
float chi2yavgone
average y chi^2 for 1 pixel clusters
float dyone
mean offset/correction for one pixel y-clusters
float sigmavav
&quot;sigma&quot; scale fctor for Vavilov distribution
float fracxtwo
fraction of double pixel sample with xsize = 1
float yavggen[4]
generic algorithm: average y-bias of reconstruction binned in 4 charge bins
float sxmax
average pixel signal for x-projection of cluster
float chi2yminc2m[4]
1st pass chi2 min search: minimum of y chi^2 in 4 charge bins (merged clusters)
float xavgc2m[4]
1st pass chi2 min search: average x-bias of reconstruction binned in 4 charge bins ...
#define TXSIZE
float chi2xavgone
average x chi^2 for 1 pixel clusters
float fracytwo
fraction of double pixel sample with ysize = 1
float yrms[4]
average y-rms of reconstruction binned in 4 charge bins
int NTxx
number of Template x-entries in each slice
float mpvvav2
most probable charge in Vavilov distribution for 2 merged clusters (not actually for larger kappa) ...
float ygx0gen[4]
generic algorithm: average y0 from Gaussian fit binned in 4 charge bins
float xavg[4]
average x-bias of reconstruction binned in 4 charge bins
float lorybias
estimate of y-lorentz bias
float fluence
radiation fluence in n_eq/cm^2
float ss50
1/2 of the single hit dcol threshold in electrons
float dytwo
mean offset/correction for one double-pixel y-clusters
float ysize
pixel size (for future use in upgraded geometry)
float dxone
mean offset/correction for one pixel x-clusters
float pixmax
maximum charge for individual pixels in cluster
float qscale
Charge scaling to match cmssw and pixelav.
float chi2yavgc2m[4]
1st pass chi2 min search: average y chi^2 in 4 charge bins (merged clusters)
float qmin
minimum cluster charge for valid hit (keeps 99.9% of simulated hits)
float chi2xavgc2m[4]
1st pass chi2 min search: average x chi^2 in 4 charge bins (merged clusters)
float xgsiggen[4]
generic algorithm: average sigma_x from Gaussian fit binned in 4 charge bins
float chi2xminc2m[4]
1st pass chi2 min search: minimum of x chi^2 in 4 charge bins (merged clusters)
int ID
&lt; template header structure
float Bfield
Bfield in Tesla.
float cotalpha
cot(alpha) is proportional to cluster length in x and is basis of interpolation
float temperature
detector temperature in deg K
float ytemp[9][21]
templates for y-reconstruction (binned over 1 central pixel)
float yrmsc2m[4]
1st pass chi2 min search: average y-rms of reconstruction binned in 4 charge bins ...
float lorywidth
estimate of y-lorentz width for optimal resolution
float fbin[3]
The QBin definitions in Q_clus/Q_avg.
int templ_version
Version number of the template to ensure code compatibility.
float xpar[2][5]
projected x-pixel uncertainty parameterization
float ygx0[4]
average y0 from Gaussian fit binned in 4 charge bins
float xgx0gen[4]
generic algorithm: average x0 from Gaussian fit binned in 4 charge bins
float symax
average pixel signal for y-projection of cluster
float kappavav2
kappa parameter for Vavilov distribution for 2 merged clusters
float xgsig[4]
average sigma_x from Gaussian fit binned in 4 charge bins
float cotbeta
cot(beta) is proportional to cluster length in y and is basis of interpolation
float beta
beta track angle (defined in CMS CMS IN 2004/014)
#define ENDL
float xrmsc2m[4]
1st pass chi2 min search: average x-rms of reconstruction binned in 4 charge bins ...
float lorxwidth
estimate of x-lorentz width for optimal resolution
float xgx0[4]
average x0 from Gaussian fit binned in 4 charge bins
int j
Definition: DBlmapReader.cc:9
char title[80]
template title
float ygsiggen[4]
generic algorithm: average sigma_y from Gaussian fit binned in 4 charge bins
float fracxone
fraction of sample with xsize = 1
float mpvvav
most probable charge in Vavilov distribution (not actually for larger kappa)
float sxtwo
rms for one double-pixel x-clusters
float r_qMeas_qTrue
ratio of measured to true cluster charge
int Dtype
detector type (0=BPix, 1=FPix)
float chi2yavg[4]
average y chi^2 in 4 charge bins
float alpha
alpha track angle (defined in CMS CMS IN 2004/014)
float xtemp[9][13]
templates for x-reconstruction (binned over 1 central pixel)
SiPixelTemplateHeader head
&lt; template storage structure
float chi2xminone
minimum of x chi^2 for 1 pixel clusters
float ypar[2][5]
projected y-pixel uncertainty parameterization
const std::vector< SiPixelTemplateStore > & thePixelTemp_
float s50
1/2 of the multihit dcol threshold in electrons
#define TYSIZE
float zsize
pixel size (for future use in upgraded geometry)
float Vbias
detector bias potential in Volts
float qbfrac[3]
fraction of sample in qbin = 0-2 (&gt;=3 is the complement)
float xsize
pixel size (for future use in upgraded geometry)
float fracyone
fraction of sample with ysize = 1
float qmin2
tighter minimum cluster charge for valid hit (keeps 99.8% of simulated hits)
float yavg[4]
average y-bias of reconstruction binned in 4 charge bins
float chi2ymin[4]
minimum of y chi^2 in 4 charge bins
float yavgc2m[4]
1st pass chi2 min search: average y-bias of reconstruction binned in 4 charge bins ...
int NTy
number of Template y entries
float dxtwo
mean offset/correction for one double-pixel x-clusters
float lorxbias
estimate of x-lorentz bias
#define LOGINFO(x)
int NTyx
number of Template y-slices of x entries
float yflpar[4][6]
Aqfl-parameterized y-correction in 4 charge bins.
float sxone
rms for one pixel x-clusters
float chi2xmin[4]
minimum of x chi^2 in 4 charge bins
float chi2yminone
minimum of y chi^2 for 1 pixel clusters
float costrk[3]
direction cosines of tracks used to generate this entry
static void postInit(std::vector< SiPixelTemplateStore > &thePixelTemp_)
SiPixelTemplateEntry entx[5][29]
29 Barrel x templates spanning cluster lengths from -6px (-1.125Rad) to +6px (+1.125Rad) in each of 5...
float ygsig[4]
average sigma_y from Gaussian fit binned in 4 charge bins
SiPixelTemplateEntry enty[60]
60 Barrel y templates spanning cluster lengths from 0px to +18px [28 entries for fpix] ...
float sigmavav2
&quot;sigma&quot; scale fctor for Vavilov distribution for 2 merged clusters
float sytwo
rms for one double-pixel y-clusters
float xrmsgen[4]
generic algorithm: average x-rms of reconstruction binned in 4 charge bins
float chi2xavg[4]
average x chi^2 in 4 charge bins
float kappavav
kappa parameter for Vavilov distribution
float xavggen[4]
generic algorithm: average x-bias of reconstruction binned in 4 charge bins
float yrmsgen[4]
generic algorithm: average y-rms of reconstruction binned in 4 charge bins
float clsleny
cluster y-length in pixels at signal height symax/2
bool SiPixelTemplate::pushfile ( const SiPixelTemplateDBObject dbobject,
std::vector< SiPixelTemplateStore > &  thePixelTemp_ 
)
static

This routine initializes the global template structures from an external file template_summary_zpNNNN where NNNN are four digits

Parameters
dbobject- db storing multiple template calibrations

Definition at line 542 of file SiPixelTemplate.cc.

References SiPixelTemplateDBObject::char2float::c, EcalCondDB::db, ENDL, SiPixelTemplateDBObject::char2float::f, i, j, relval_steps::k, prof2calltree::l, LOGERROR, LOGINFO, visualization-live-secondInstance_cfg::m, SiPixelTemplateDBObject::reader(), groupFilesInBlocks::temp, TXSIZE, and TYSIZE.

543 {
544  // Add template stored in external dbobject to theTemplateStore
545 
546  // Local variables
547  int i, j, k, l;
548  float qavg_avg;
549  // const char *tempfile;
550  const int code_version={17};
551 
552  // We must create a new object because dbobject must be a const and our stream must not be
553  auto db(dbobject.reader());
554 
555  // Create a local template storage entry
556  SiPixelTemplateStore theCurrentTemp;
557 
558  // Fill the template storage for each template calibration stored in the db
559  for(int m=0; m<db.numOfTempl(); ++m) {
560 
561  // Read-in a header string first and print it
562 
564  for (i=0; i<20; ++i) {
565  temp.f = db.sVector()[db.index()];
566  theCurrentTemp.head.title[4*i] = temp.c[0];
567  theCurrentTemp.head.title[4*i+1] = temp.c[1];
568  theCurrentTemp.head.title[4*i+2] = temp.c[2];
569  theCurrentTemp.head.title[4*i+3] = temp.c[3];
570  db.incrementIndex(1);
571  }
572  theCurrentTemp.head.title[79] = '\0';
573  LOGINFO("SiPixelTemplate") << "Loading Pixel Template File - " << theCurrentTemp.head.title << ENDL;
574 
575  // next, the header information
576 
577  db >> theCurrentTemp.head.ID >> theCurrentTemp.head.templ_version >> theCurrentTemp.head.Bfield >> theCurrentTemp.head.NTy >> theCurrentTemp.head.NTyx >> theCurrentTemp.head.NTxx
578  >> theCurrentTemp.head.Dtype >> theCurrentTemp.head.Vbias >> theCurrentTemp.head.temperature >> theCurrentTemp.head.fluence >> theCurrentTemp.head.qscale
579  >> theCurrentTemp.head.s50 >> theCurrentTemp.head.lorywidth >> theCurrentTemp.head.lorxwidth >> theCurrentTemp.head.ysize >> theCurrentTemp.head.xsize >> theCurrentTemp.head.zsize;
580 
581  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 0A, no template load" << ENDL; return false;}
582 
583  LOGINFO("SiPixelTemplate") << "Loading Pixel Template File - " << theCurrentTemp.head.title
584  <<" code version = "<<code_version
585  <<" object version "<<theCurrentTemp.head.templ_version
586  << ENDL;
587 
588  if(theCurrentTemp.head.templ_version > 17) {
589  db >> theCurrentTemp.head.ss50 >> theCurrentTemp.head.lorybias >> theCurrentTemp.head.lorxbias >> theCurrentTemp.head.fbin[0] >> theCurrentTemp.head.fbin[1] >> theCurrentTemp.head.fbin[2];
590 
591  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 0B, no template load"
592  << ENDL; return false;}
593  } else {
594  theCurrentTemp.head.ss50 = theCurrentTemp.head.s50;
595  theCurrentTemp.head.lorybias = theCurrentTemp.head.lorywidth/2.f;
596  theCurrentTemp.head.lorxbias = theCurrentTemp.head.lorxwidth/2.f;
597  theCurrentTemp.head.fbin[0] = fbin0;
598  theCurrentTemp.head.fbin[1] = fbin1;
599  theCurrentTemp.head.fbin[2] = fbin2;
600  //std::cout<<" set fbin "<< theCurrentTemp.head.fbin[0]<<" "<<theCurrentTemp.head.fbin[1]<<" "
601  // <<theCurrentTemp.head.fbin[2]<<std::endl;
602  }
603 
604  LOGINFO("SiPixelTemplate")
605  << "Template ID = " << theCurrentTemp.head.ID << ", Template Version "
606  << theCurrentTemp.head.templ_version << ", Bfield = "
607  << theCurrentTemp.head.Bfield<< ", NTy = " << theCurrentTemp.head.NTy << ", NTyx = "
608  << theCurrentTemp.head.NTyx<< ", NTxx = " << theCurrentTemp.head.NTxx << ", Dtype = "
609  << theCurrentTemp.head.Dtype<< ", Bias voltage " << theCurrentTemp.head.Vbias << ", temperature "
610  << theCurrentTemp.head.temperature << ", fluence " << theCurrentTemp.head.fluence
611  << ", Q-scaling factor " << theCurrentTemp.head.qscale
612  << ", 1/2 multi dcol threshold " << theCurrentTemp.head.s50 << ", 1/2 single dcol threshold "
613  << theCurrentTemp.head.ss50<< ", y Lorentz Width " << theCurrentTemp.head.lorywidth
614  << ", y Lorentz Bias " << theCurrentTemp.head.lorybias
615  << ", x Lorentz width " << theCurrentTemp.head.lorxwidth
616  << ", x Lorentz Bias " << theCurrentTemp.head.lorxbias
617  << ", Q/Q_avg fractions for Qbin defs " << theCurrentTemp.head.fbin[0]
618  << ", " << theCurrentTemp.head.fbin[1]
619  << ", " << theCurrentTemp.head.fbin[2]
620  << ", pixel x-size " << theCurrentTemp.head.xsize
621  << ", y-size " << theCurrentTemp.head.ysize << ", zsize " << theCurrentTemp.head.zsize << ENDL;
622 
623  if(theCurrentTemp.head.templ_version < code_version) {
624  LOGINFO("SiPixelTemplate") << "code expects version "<< code_version << " finds "
625  << theCurrentTemp.head.templ_version <<", load anyway " << ENDL;
626  //return false; // dk
627  }
628 
629 
630 #ifdef SI_PIXEL_TEMPLATE_USE_BOOST
631 
632  // next, layout the 1-d/2-d structures needed to store template
633  theCurrentTemp.enty.resize(boost::extents[theCurrentTemp.head.NTy]);
634  theCurrentTemp.entx.resize(boost::extents[theCurrentTemp.head.NTyx][theCurrentTemp.head.NTxx]);
635 
636 #endif
637 
638  // next, loop over all barrel y-angle entries
639 
640  for (i=0; i < theCurrentTemp.head.NTy; ++i) {
641 
642  db >> theCurrentTemp.enty[i].runnum >> theCurrentTemp.enty[i].costrk[0]
643  >> theCurrentTemp.enty[i].costrk[1] >> theCurrentTemp.enty[i].costrk[2];
644 
645  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 1, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
646 
647  // Calculate the alpha, beta, and cot(beta) for this entry
648 
649  theCurrentTemp.enty[i].alpha = static_cast<float>(atan2((double)theCurrentTemp.enty[i].costrk[2], (double)theCurrentTemp.enty[i].costrk[0]));
650 
651  theCurrentTemp.enty[i].cotalpha = theCurrentTemp.enty[i].costrk[0]/theCurrentTemp.enty[i].costrk[2];
652 
653  theCurrentTemp.enty[i].beta = static_cast<float>(atan2((double)theCurrentTemp.enty[i].costrk[2], (double)theCurrentTemp.enty[i].costrk[1]));
654 
655  theCurrentTemp.enty[i].cotbeta = theCurrentTemp.enty[i].costrk[1]/theCurrentTemp.enty[i].costrk[2];
656 
657  db >> theCurrentTemp.enty[i].qavg >> theCurrentTemp.enty[i].pixmax >> theCurrentTemp.enty[i].symax >> theCurrentTemp.enty[i].dyone
658  >> theCurrentTemp.enty[i].syone >> theCurrentTemp.enty[i].sxmax >> theCurrentTemp.enty[i].dxone >> theCurrentTemp.enty[i].sxone;
659 
660  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 2, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
661 
662  db >> theCurrentTemp.enty[i].dytwo >> theCurrentTemp.enty[i].sytwo >> theCurrentTemp.enty[i].dxtwo
663  >> theCurrentTemp.enty[i].sxtwo >> theCurrentTemp.enty[i].qmin >> theCurrentTemp.enty[i].clsleny >> theCurrentTemp.enty[i].clslenx;
664  // >> theCurrentTemp.enty[i].mpvvav >> theCurrentTemp.enty[i].sigmavav >> theCurrentTemp.enty[i].kappavav;
665 
666  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 3, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
667 
668  if(theCurrentTemp.enty[i].qmin <= 0.) {LOGERROR("SiPixelTemplate") << "Error in template ID " << theCurrentTemp.head.ID << " qmin = " << theCurrentTemp.enty[i].qmin << ", run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
669 
670  for (j=0; j<2; ++j) {
671 
672  db >> theCurrentTemp.enty[i].ypar[j][0] >> theCurrentTemp.enty[i].ypar[j][1]
673  >> theCurrentTemp.enty[i].ypar[j][2] >> theCurrentTemp.enty[i].ypar[j][3] >> theCurrentTemp.enty[i].ypar[j][4];
674 
675  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 4, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
676 
677  }
678 
679  for (j=0; j<9; ++j) {
680 
681  for (k=0; k<TYSIZE; ++k) {db >> theCurrentTemp.enty[i].ytemp[j][k];}
682 
683  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 5, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
684  }
685 
686  for (j=0; j<2; ++j) {
687 
688  db >> theCurrentTemp.enty[i].xpar[j][0] >> theCurrentTemp.enty[i].xpar[j][1]
689  >> theCurrentTemp.enty[i].xpar[j][2] >> theCurrentTemp.enty[i].xpar[j][3] >> theCurrentTemp.enty[i].xpar[j][4];
690 
691  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 6, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
692 
693  }
694 
695  qavg_avg = 0.f;
696  for (j=0; j<9; ++j) {
697 
698  for (k=0; k<TXSIZE; ++k) {db >> theCurrentTemp.enty[i].xtemp[j][k]; qavg_avg += theCurrentTemp.enty[i].xtemp[j][k];}
699 
700  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 7, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
701  }
702  theCurrentTemp.enty[i].qavg_avg = qavg_avg/9.;
703 
704  for (j=0; j<4; ++j) {
705 
706  db >> theCurrentTemp.enty[i].yavg[j] >> theCurrentTemp.enty[i].yrms[j] >> theCurrentTemp.enty[i].ygx0[j] >> theCurrentTemp.enty[i].ygsig[j];
707 
708  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 8, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
709  }
710 
711  for (j=0; j<4; ++j) {
712 
713  db >> theCurrentTemp.enty[i].yflpar[j][0] >> theCurrentTemp.enty[i].yflpar[j][1] >> theCurrentTemp.enty[i].yflpar[j][2]
714  >> theCurrentTemp.enty[i].yflpar[j][3] >> theCurrentTemp.enty[i].yflpar[j][4] >> theCurrentTemp.enty[i].yflpar[j][5];
715 
716  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 9, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
717  }
718 
719  for (j=0; j<4; ++j) {
720 
721  db >> theCurrentTemp.enty[i].xavg[j] >> theCurrentTemp.enty[i].xrms[j] >> theCurrentTemp.enty[i].xgx0[j] >> theCurrentTemp.enty[i].xgsig[j];
722 
723  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 10, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
724  }
725 
726  for (j=0; j<4; ++j) {
727 
728  db >> theCurrentTemp.enty[i].xflpar[j][0] >> theCurrentTemp.enty[i].xflpar[j][1] >> theCurrentTemp.enty[i].xflpar[j][2]
729  >> theCurrentTemp.enty[i].xflpar[j][3] >> theCurrentTemp.enty[i].xflpar[j][4] >> theCurrentTemp.enty[i].xflpar[j][5];
730 
731  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 11, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
732  }
733 
734  for (j=0; j<4; ++j) {
735 
736  db >> theCurrentTemp.enty[i].chi2yavg[j] >> theCurrentTemp.enty[i].chi2ymin[j] >> theCurrentTemp.enty[i].chi2xavg[j] >> theCurrentTemp.enty[i].chi2xmin[j];
737 
738  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 12, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
739  }
740 
741  for (j=0; j<4; ++j) {
742 
743  db >> theCurrentTemp.enty[i].yavgc2m[j] >> theCurrentTemp.enty[i].yrmsc2m[j] >> theCurrentTemp.enty[i].chi2yavgc2m[j] >> theCurrentTemp.enty[i].chi2yminc2m[j];
744 
745  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 13, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
746  }
747 
748  for (j=0; j<4; ++j) {
749 
750  db >> theCurrentTemp.enty[i].xavgc2m[j] >> theCurrentTemp.enty[i].xrmsc2m[j] >> theCurrentTemp.enty[i].chi2xavgc2m[j] >> theCurrentTemp.enty[i].chi2xminc2m[j];
751 
752  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 14, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
753  }
754 
755  for (j=0; j<4; ++j) {
756 
757  db >> theCurrentTemp.enty[i].yavggen[j] >> theCurrentTemp.enty[i].yrmsgen[j] >> theCurrentTemp.enty[i].ygx0gen[j] >> theCurrentTemp.enty[i].ygsiggen[j];
758 
759  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 14a, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
760  }
761 
762  for (j=0; j<4; ++j) {
763 
764  db >> theCurrentTemp.enty[i].xavggen[j] >> theCurrentTemp.enty[i].xrmsgen[j] >> theCurrentTemp.enty[i].xgx0gen[j] >> theCurrentTemp.enty[i].xgsiggen[j];
765 
766  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 14b, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
767  }
768 
769 
770  db >> theCurrentTemp.enty[i].chi2yavgone >> theCurrentTemp.enty[i].chi2yminone >> theCurrentTemp.enty[i].chi2xavgone >> theCurrentTemp.enty[i].chi2xminone >> theCurrentTemp.enty[i].qmin2
771  >> theCurrentTemp.enty[i].mpvvav >> theCurrentTemp.enty[i].sigmavav >> theCurrentTemp.enty[i].kappavav >> theCurrentTemp.enty[i].r_qMeas_qTrue >> theCurrentTemp.enty[i].spare[0];
772 
773  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 15, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
774 
775  db >> theCurrentTemp.enty[i].mpvvav2 >> theCurrentTemp.enty[i].sigmavav2 >> theCurrentTemp.enty[i].kappavav2 >> theCurrentTemp.enty[i].qbfrac[0] >> theCurrentTemp.enty[i].qbfrac[1]
776  >> theCurrentTemp.enty[i].qbfrac[2] >> theCurrentTemp.enty[i].fracyone >> theCurrentTemp.enty[i].fracxone >> theCurrentTemp.enty[i].fracytwo >> theCurrentTemp.enty[i].fracxtwo;
777  // theCurrentTemp.enty[i].qbfrac[3] = 1. - theCurrentTemp.enty[i].qbfrac[0] - theCurrentTemp.enty[i].qbfrac[1] - theCurrentTemp.enty[i].qbfrac[2];
778 
779  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 16, no template load, run # " << theCurrentTemp.enty[i].runnum << ENDL; return false;}
780 
781  }
782 
783  // next, loop over all barrel x-angle entries
784 
785  for (k=0; k < theCurrentTemp.head.NTyx; ++k) {
786 
787  for (i=0; i < theCurrentTemp.head.NTxx; ++i) {
788 
789  db >> theCurrentTemp.entx[k][i].runnum >> theCurrentTemp.entx[k][i].costrk[0]
790  >> theCurrentTemp.entx[k][i].costrk[1] >> theCurrentTemp.entx[k][i].costrk[2];
791 
792  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 17, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
793 
794  // Calculate the alpha, beta, and cot(beta) for this entry
795 
796  theCurrentTemp.entx[k][i].alpha = static_cast<float>(atan2((double)theCurrentTemp.entx[k][i].costrk[2], (double)theCurrentTemp.entx[k][i].costrk[0]));
797 
798  theCurrentTemp.entx[k][i].cotalpha = theCurrentTemp.entx[k][i].costrk[0]/theCurrentTemp.entx[k][i].costrk[2];
799 
800  theCurrentTemp.entx[k][i].beta = static_cast<float>(atan2((double)theCurrentTemp.entx[k][i].costrk[2], (double)theCurrentTemp.entx[k][i].costrk[1]));
801 
802  theCurrentTemp.entx[k][i].cotbeta = theCurrentTemp.entx[k][i].costrk[1]/theCurrentTemp.entx[k][i].costrk[2];
803 
804  db >> theCurrentTemp.entx[k][i].qavg >> theCurrentTemp.entx[k][i].pixmax >> theCurrentTemp.entx[k][i].symax >> theCurrentTemp.entx[k][i].dyone
805  >> theCurrentTemp.entx[k][i].syone >> theCurrentTemp.entx[k][i].sxmax >> theCurrentTemp.entx[k][i].dxone >> theCurrentTemp.entx[k][i].sxone;
806 
807  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 18, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
808 
809  db >> theCurrentTemp.entx[k][i].dytwo >> theCurrentTemp.entx[k][i].sytwo >> theCurrentTemp.entx[k][i].dxtwo
810  >> theCurrentTemp.entx[k][i].sxtwo >> theCurrentTemp.entx[k][i].qmin >> theCurrentTemp.entx[k][i].clsleny >> theCurrentTemp.entx[k][i].clslenx;
811  // >> theCurrentTemp.entx[k][i].mpvvav >> theCurrentTemp.entx[k][i].sigmavav >> theCurrentTemp.entx[k][i].kappavav;
812 
813  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 19, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
814 
815  for (j=0; j<2; ++j) {
816 
817  db >> theCurrentTemp.entx[k][i].ypar[j][0] >> theCurrentTemp.entx[k][i].ypar[j][1]
818  >> theCurrentTemp.entx[k][i].ypar[j][2] >> theCurrentTemp.entx[k][i].ypar[j][3] >> theCurrentTemp.entx[k][i].ypar[j][4];
819 
820  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 20, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
821  }
822 
823  for (j=0; j<9; ++j) {
824 
825  for (l=0; l<TYSIZE; ++l) {db >> theCurrentTemp.entx[k][i].ytemp[j][l];}
826 
827  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 21, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
828  }
829 
830 
831 
832  for (j=0; j<2; ++j) {
833 
834  db >> theCurrentTemp.entx[k][i].xpar[j][0] >> theCurrentTemp.entx[k][i].xpar[j][1]
835  >> theCurrentTemp.entx[k][i].xpar[j][2] >> theCurrentTemp.entx[k][i].xpar[j][3] >> theCurrentTemp.entx[k][i].xpar[j][4];
836 
837 
838  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 22, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
839  }
840 
841  qavg_avg = 0.f;
842  for (j=0; j<9; ++j) {
843 
844  for (l=0; l<TXSIZE; ++l) {db >> theCurrentTemp.entx[k][i].xtemp[j][l]; qavg_avg += theCurrentTemp.entx[k][i].xtemp[j][l];}
845 
846  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 23, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
847  }
848  theCurrentTemp.entx[k][i].qavg_avg = qavg_avg/9.;
849 
850  for (j=0; j<4; ++j) {
851 
852  db >> theCurrentTemp.entx[k][i].yavg[j] >> theCurrentTemp.entx[k][i].yrms[j] >> theCurrentTemp.entx[k][i].ygx0[j] >> theCurrentTemp.entx[k][i].ygsig[j];
853 
854  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 24, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
855  }
856 
857  for (j=0; j<4; ++j) {
858 
859  db >> theCurrentTemp.entx[k][i].yflpar[j][0] >> theCurrentTemp.entx[k][i].yflpar[j][1] >> theCurrentTemp.entx[k][i].yflpar[j][2]
860  >> theCurrentTemp.entx[k][i].yflpar[j][3] >> theCurrentTemp.entx[k][i].yflpar[j][4] >> theCurrentTemp.entx[k][i].yflpar[j][5];
861 
862  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 25, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
863  }
864 
865  for (j=0; j<4; ++j) {
866 
867  db >> theCurrentTemp.entx[k][i].xavg[j] >> theCurrentTemp.entx[k][i].xrms[j] >> theCurrentTemp.entx[k][i].xgx0[j] >> theCurrentTemp.entx[k][i].xgsig[j];
868 
869  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 26, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
870  }
871 
872  for (j=0; j<4; ++j) {
873 
874  db >> theCurrentTemp.entx[k][i].xflpar[j][0] >> theCurrentTemp.entx[k][i].xflpar[j][1] >> theCurrentTemp.entx[k][i].xflpar[j][2]
875  >> theCurrentTemp.entx[k][i].xflpar[j][3] >> theCurrentTemp.entx[k][i].xflpar[j][4] >> theCurrentTemp.entx[k][i].xflpar[j][5];
876 
877  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 27, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
878  }
879 
880  for (j=0; j<4; ++j) {
881 
882  db >> theCurrentTemp.entx[k][i].chi2yavg[j] >> theCurrentTemp.entx[k][i].chi2ymin[j] >> theCurrentTemp.entx[k][i].chi2xavg[j] >> theCurrentTemp.entx[k][i].chi2xmin[j];
883 
884  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 28, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
885  }
886 
887  for (j=0; j<4; ++j) {
888 
889  db >> theCurrentTemp.entx[k][i].yavgc2m[j] >> theCurrentTemp.entx[k][i].yrmsc2m[j] >> theCurrentTemp.entx[k][i].chi2yavgc2m[j] >> theCurrentTemp.entx[k][i].chi2yminc2m[j];
890 
891  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 29, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
892  }
893 
894  for (j=0; j<4; ++j) {
895 
896  db >> theCurrentTemp.entx[k][i].xavgc2m[j] >> theCurrentTemp.entx[k][i].xrmsc2m[j] >> theCurrentTemp.entx[k][i].chi2xavgc2m[j] >> theCurrentTemp.entx[k][i].chi2xminc2m[j];
897 
898  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 30, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
899  }
900 
901  for (j=0; j<4; ++j) {
902 
903  db >> theCurrentTemp.entx[k][i].yavggen[j] >> theCurrentTemp.entx[k][i].yrmsgen[j] >> theCurrentTemp.entx[k][i].ygx0gen[j] >> theCurrentTemp.entx[k][i].ygsiggen[j];
904 
905  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 30a, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
906  }
907 
908  for (j=0; j<4; ++j) {
909 
910  db >> theCurrentTemp.entx[k][i].xavggen[j] >> theCurrentTemp.entx[k][i].xrmsgen[j] >> theCurrentTemp.entx[k][i].xgx0gen[j] >> theCurrentTemp.entx[k][i].xgsiggen[j];
911 
912  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 30b, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
913  }
914 
915 
916  db >> theCurrentTemp.entx[k][i].chi2yavgone >> theCurrentTemp.entx[k][i].chi2yminone >> theCurrentTemp.entx[k][i].chi2xavgone >> theCurrentTemp.entx[k][i].chi2xminone >> theCurrentTemp.entx[k][i].qmin2
917  >> theCurrentTemp.entx[k][i].mpvvav >> theCurrentTemp.entx[k][i].sigmavav >> theCurrentTemp.entx[k][i].kappavav >> theCurrentTemp.entx[k][i].r_qMeas_qTrue >> theCurrentTemp.entx[k][i].spare[0];
918 
919  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 31, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
920 
921  db >> theCurrentTemp.entx[k][i].mpvvav2 >> theCurrentTemp.entx[k][i].sigmavav2 >> theCurrentTemp.entx[k][i].kappavav2 >> theCurrentTemp.entx[k][i].qbfrac[0] >> theCurrentTemp.entx[k][i].qbfrac[1]
922  >> theCurrentTemp.entx[k][i].qbfrac[2] >> theCurrentTemp.entx[k][i].fracyone >> theCurrentTemp.entx[k][i].fracxone >> theCurrentTemp.entx[k][i].fracytwo >> theCurrentTemp.entx[k][i].fracxtwo;
923  // theCurrentTemp.entx[k][i].qbfrac[3] = 1. - theCurrentTemp.entx[k][i].qbfrac[0] - theCurrentTemp.entx[k][i].qbfrac[1] - theCurrentTemp.entx[k][i].qbfrac[2];
924 
925  if(db.fail()) {LOGERROR("SiPixelTemplate") << "Error reading file 32, no template load, run # " << theCurrentTemp.entx[k][i].runnum << ENDL; return false;}
926 
927  }
928  }
929 
930 
931  // Add this template to the store
932 
933  thePixelTemp_.push_back(theCurrentTemp);
934 
935  }
937  return true;
938 
939 } // TempInit
int i
Definition: DBlmapReader.cc:9
#define LOGERROR(x)
#define TXSIZE
tuple db
Definition: EcalCondDB.py:151
#define ENDL
int j
Definition: DBlmapReader.cc:9
const std::vector< SiPixelTemplateStore > & thePixelTemp_
#define TYSIZE
#define LOGINFO(x)
static void postInit(std::vector< SiPixelTemplateStore > &thePixelTemp_)
float SiPixelTemplate::qavg ( )
inline

average cluster charge for this set of track angles

Definition at line 319 of file SiPixelTemplate.h.

References qavg_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

int SiPixelTemplate::qbin ( int  id,
float  cotalpha,
float  cotbeta,
float  locBz,
float  qclus,
float &  pixmx,
float &  sigmay,
float &  deltay,
float &  sigmax,
float &  deltax,
float &  sy1,
float &  dy1,
float &  sy2,
float &  dy2,
float &  sx1,
float &  dx1,
float &  sx2,
float &  dx2 
)

Interpolate beta/alpha angles to produce an expected average charge. Return int (0-4) describing the charge of the cluster [0: 1.5<Q/Qavg, 1: 1<Q/Qavg<1.5, 2: 0.85<Q/Qavg<1, 3: 0.95Qmin<Q<0.85Qavg, 4: Q<0.95Qmin].

Parameters
id- (input) index of the template to use
cotalpha- (input) the cotangent of the alpha track angle (see CMS IN 2004/014)
cotbeta- (input) the cotangent of the beta track angle (see CMS IN 2004/014)
locBz- (input) the sign of this quantity is used to determine whether to flip cot(beta)<0 quantities from cot(beta)>0 (FPix only) for FPix IP-related tracks, locBz < 0 for cot(beta) > 0 and locBz > 0 for cot(beta) < 0
qclus- (input) the cluster charge in electrons
pixmax- (output) the maximum pixel charge in electrons (truncation value)
sigmay- (output) the estimated y-error for CPEGeneric in microns
deltay- (output) the estimated y-bias for CPEGeneric in microns
sigmax- (output) the estimated x-error for CPEGeneric in microns
deltax- (output) the estimated x-bias for CPEGeneric in microns
sy1- (output) the estimated y-error for 1 single-pixel clusters in microns
dy1- (output) the estimated y-bias for 1 single-pixel clusters in microns
sy2- (output) the estimated y-error for 1 double-pixel clusters in microns
dy2- (output) the estimated y-bias for 1 double-pixel clusters in microns
sx1- (output) the estimated x-error for 1 single-pixel clusters in microns
dx1- (output) the estimated x-bias for 1 single-pixel clusters in microns
sx2- (output) the estimated x-error for 1 double-pixel clusters in microns
dx2- (output) the estimated x-bias for 1 double-pixel clusters in microns
lorywidth- (output) the estimated y Lorentz width
lorxwidth- (output) the estimated x Lorentz width

Definition at line 2267 of file SiPixelTemplate.cc.

References funct::abs(), assert(), edm::hlt::Exception, f, i, cmsHarvester::index, j, and mathSSE::sqrt().

Referenced by PixelCPEGeneric::localPosition(), qbin(), and TrackClusterSplitter::splitCluster().

2270 {
2271  // Interpolate for a new set of track angles
2272 
2273 
2274 // Find the index corresponding to id
2275 
2276  int index = -1;
2277  for( int i=0; i<(int)thePixelTemp_.size(); ++i) {
2278  if(id == thePixelTemp_[i].head.ID) {
2279  index = i;
2280  break;
2281  }
2282  }
2283 
2284 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
2285  if(index < 0 || index >= (int)thePixelTemp_.size()) {
2286  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::qbin can't find needed template ID = " << id << std::endl;
2287  }
2288 #else
2289  assert(index >= 0 && index < (int)thePixelTemp_.size());
2290 #endif
2291 
2292  //
2293 
2294 
2295  auto const & templ = thePixelTemp_[index];
2296 
2297  // Interpolate the absolute value of cot(beta)
2298 
2299  auto acotb = std::abs(cotbeta);
2300 
2301  // qcorrect corrects the cot(alpha)=0 cluster charge for non-zero cot(alpha)
2302 
2303  auto cotalpha0 = thePixelTemp_[index].enty[0].cotalpha;
2304  auto qcorrect=std::sqrt((1.f+cotbeta*cotbeta+cotalpha*cotalpha)/(1.f+cotbeta*cotbeta+cotalpha0*cotalpha0));
2305 
2306  // for some cosmics, the ususal gymnastics are incorrect
2307 
2308  float cotb; bool flip_y;
2309  if(thePixelTemp_[index].head.Dtype == 0) {
2310  cotb = acotb;
2311  flip_y = false;
2312  if(cotbeta < 0.f) {flip_y = true;}
2313  } else {
2314  if(locBz < 0.f) {
2315  cotb = cotbeta;
2316  flip_y = false;
2317  } else {
2318  cotb = -cotbeta;
2319  flip_y = true;
2320  }
2321  }
2322 
2323  // Copy the charge scaling factor to the private variable
2324 
2325  auto qscale = thePixelTemp_[index].head.qscale;
2326 
2327 
2328  /*
2329  lorywidth = thePixelTemp_[index].head.lorywidth;
2330  if(locBz > 0.f) {lorywidth = -lorywidth;}
2331  lorxwidth = thePixelTemp_[index].head.lorxwidth;
2332  */
2333 
2334 
2335  auto Ny = thePixelTemp_[index].head.NTy;
2336  auto Nyx = thePixelTemp_[index].head.NTyx;
2337  auto Nxx = thePixelTemp_[index].head.NTxx;
2338 
2339 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
2340  if(Ny < 2 || Nyx < 1 || Nxx < 2) {
2341  throw cms::Exception("DataCorrupt") << "template ID = " << id_current_ << "has too few entries: Ny/Nyx/Nxx = " << Ny << "/" << Nyx << "/" << Nxx << std::endl;
2342  }
2343 #else
2344  assert(Ny > 1 && Nyx > 0 && Nxx > 1);
2345 #endif
2346 
2347  // next, loop over all y-angle entries
2348 
2349  auto ilow = 0;
2350  auto ihigh = 0;
2351  auto yratio = 0.f;
2352 
2353  {
2354  auto j = std::lower_bound(templ.cotbetaY,templ.cotbetaY+Ny,cotb);
2355  if (j==templ.cotbetaY+Ny) { --j; yratio = 1.f; }
2356  else if (j==templ.cotbetaY) { ++j; yratio = 0.f;}
2357  else { yratio = (cotb - (*(j-1)) )/( (*j) - (*(j-1)) ) ; }
2358 
2359  ihigh = j-templ.cotbetaY;
2360  ilow = ihigh-1;
2361  }
2362 
2363 
2364 
2365  // Interpolate/store all y-related quantities (flip displacements when flip_y)
2366 
2367  dy1 = (1.f - yratio)*thePixelTemp_[index].enty[ilow].dyone + yratio*thePixelTemp_[index].enty[ihigh].dyone;
2368  if(flip_y) {dy1 = -dy1;}
2369  sy1 = (1.f - yratio)*thePixelTemp_[index].enty[ilow].syone + yratio*thePixelTemp_[index].enty[ihigh].syone;
2370  dy2 = (1.f - yratio)*thePixelTemp_[index].enty[ilow].dytwo + yratio*thePixelTemp_[index].enty[ihigh].dytwo;
2371  if(flip_y) {dy2 = -dy2;}
2372  sy2 = (1.f - yratio)*thePixelTemp_[index].enty[ilow].sytwo + yratio*thePixelTemp_[index].enty[ihigh].sytwo;
2373 
2374  auto qavg = (1.f - yratio)*thePixelTemp_[index].enty[ilow].qavg + yratio*thePixelTemp_[index].enty[ihigh].qavg;
2375  qavg *= qcorrect;
2376  auto qmin = (1.f - yratio)*thePixelTemp_[index].enty[ilow].qmin + yratio*thePixelTemp_[index].enty[ihigh].qmin;
2377  qmin *= qcorrect;
2378  auto qmin2 = (1.f - yratio)*thePixelTemp_[index].enty[ilow].qmin2 + yratio*thePixelTemp_[index].enty[ihigh].qmin2;
2379  qmin2 *= qcorrect;
2380 
2381 
2382 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
2383  if(qavg <= 0.f || qmin <= 0.f) {
2384  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::qbin, qavg or qmin <= 0,"
2385  << " Probably someone called the generic pixel reconstruction with an illegal trajectory state" << std::endl;
2386  }
2387 #else
2388  assert(qavg > 0.f && qmin > 0.f);
2389 #endif
2390 
2391  // Scale the input charge to account for differences between pixelav and CMSSW simulation or data
2392 
2393  auto qtotal = qscale*qclus;
2394 
2395  // uncertainty and final corrections depend upon total charge bin
2396  auto fq = qtotal/qavg;
2397  int binq;
2398  // since fbin is undefined I define it here d.k. 6/14
2399  fbin_[0]=fbin0; fbin_[1]=fbin1; fbin_[2]=fbin2;
2400 
2401  if(fq > fbin_[0]) {
2402  binq=0;
2403  //std::cout<<" fq "<<fq<<" "<<qtotal<<" "<<qavg<<" "<<qclus<<" "<<qscale<<" "
2404  // <<fbin_[0]<<" "<<fbin_[1]<<" "<<fbin_[2]<<std::endl;
2405  } else {
2406  if(fq > fbin_[1]) {
2407  binq=1;
2408  } else {
2409  if(fq > fbin_[2]) {
2410  binq=2;
2411  } else {
2412  binq=3;
2413  }
2414  }
2415  }
2416 
2417  auto yavggen =(1.f - yratio)*thePixelTemp_[index].enty[ilow].yavggen[binq] + yratio*thePixelTemp_[index].enty[ihigh].yavggen[binq];
2418  if(flip_y) {yavggen = -yavggen;}
2419  auto yrmsgen =(1.f - yratio)*thePixelTemp_[index].enty[ilow].yrmsgen[binq] + yratio*thePixelTemp_[index].enty[ihigh].yrmsgen[binq];
2420 
2421 
2422 // next, loop over all x-angle entries, first, find relevant y-slices
2423 
2424  auto iylow = 0;
2425  auto iyhigh = 0;
2426  auto yxratio = 0.f;
2427 
2428 
2429  {
2430  auto j = std::lower_bound(templ.cotbetaX,templ.cotbetaX+Nyx,acotb);
2431  if (j==templ.cotbetaX+Nyx) { --j; yxratio = 1.f; }
2432  else if (j==templ.cotbetaX) { ++j; yxratio = 0.f;}
2433  else { yxratio = (acotb - (*(j-1)) )/( (*j) - (*(j-1)) ) ; }
2434 
2435  iyhigh = j-templ.cotbetaX;
2436  iylow = iyhigh -1;
2437  }
2438 
2439 
2440 
2441  ilow = ihigh = 0;
2442  auto xxratio = 0.f;
2443 
2444  {
2445  auto j = std::lower_bound(templ.cotalphaX,templ.cotalphaX+Nxx,cotalpha);
2446  if (j==templ.cotalphaX+Nxx) { --j; xxratio = 1.f; }
2447  else if (j==templ.cotalphaX) { ++j; xxratio = 0.f;}
2448  else { xxratio = (cotalpha - (*(j-1)) )/( (*j) - (*(j-1)) ) ; }
2449 
2450  ihigh = j-templ.cotalphaX;
2451  ilow = ihigh-1;
2452  }
2453 
2454 
2455 
2456  dx1 = (1.f - xxratio)*thePixelTemp_[index].entx[0][ilow].dxone + xxratio*thePixelTemp_[index].entx[0][ihigh].dxone;
2457  sx1 = (1.f - xxratio)*thePixelTemp_[index].entx[0][ilow].sxone + xxratio*thePixelTemp_[index].entx[0][ihigh].sxone;
2458  dx2 = (1.f - xxratio)*thePixelTemp_[index].entx[0][ilow].dxtwo + xxratio*thePixelTemp_[index].entx[0][ihigh].dxtwo;
2459  sx2 = (1.f - xxratio)*thePixelTemp_[index].entx[0][ilow].sxtwo + xxratio*thePixelTemp_[index].entx[0][ihigh].sxtwo;
2460 
2461  // pixmax is the maximum allowed pixel charge (used for truncation)
2462 
2463  pixmx=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index].entx[iylow][ilow].pixmax + xxratio*thePixelTemp_[index].entx[iylow][ihigh].pixmax)
2464  +yxratio*((1.f - xxratio)*thePixelTemp_[index].entx[iyhigh][ilow].pixmax + xxratio*thePixelTemp_[index].entx[iyhigh][ihigh].pixmax);
2465 
2466  auto xavggen = (1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index].entx[iylow][ilow].xavggen[binq] + xxratio*thePixelTemp_[index].entx[iylow][ihigh].xavggen[binq])
2467  +yxratio*((1.f - xxratio)*thePixelTemp_[index].entx[iyhigh][ilow].xavggen[binq] + xxratio*thePixelTemp_[index].entx[iyhigh][ihigh].xavggen[binq]);
2468 
2469  auto xrmsgen = (1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index].entx[iylow][ilow].xrmsgen[binq] + xxratio*thePixelTemp_[index].entx[iylow][ihigh].xrmsgen[binq])
2470  +yxratio*((1.f - xxratio)*thePixelTemp_[index].entx[iyhigh][ilow].xrmsgen[binq] + xxratio*thePixelTemp_[index].entx[iyhigh][ihigh].xrmsgen[binq]);
2471 
2472 
2473 
2474  // Take the errors and bias from the correct charge bin
2475 
2476  sigmay = yrmsgen; deltay = yavggen;
2477 
2478  sigmax = xrmsgen; deltax = xavggen;
2479 
2480  // If the charge is too small (then flag it)
2481 
2482  if(qtotal < 0.95f*qmin) {binq = 5;} else {if(qtotal < 0.95f*qmin2) {binq = 4;}}
2483 
2484  return binq;
2485 
2486 } // qbin
int i
Definition: DBlmapReader.cc:9
float fbin_[3]
The QBin definitions in Q_clus/Q_avg.
assert(m_qm.get())
float yratio()
fractional distance in y between cotbeta templates
float qmin()
minimum cluster charge for valid hit (keeps 99.9% of simulated hits)
float sytwo()
rms for one double-pixel y-clusters
float sxone()
rms for one pixel x-clusters
float qscale()
charge scaling factor
float dxone()
mean offset/correction for one pixel x-clusters
T sqrt(T t)
Definition: SSEVec.h:48
Abs< T >::type abs(const T &t)
Definition: Abs.h:22
int j
Definition: DBlmapReader.cc:9
double f[11][100]
float sxtwo()
rms for one double-pixel x-clusters
float dytwo()
mean offset/correction for one double-pixel y-clusters
const std::vector< SiPixelTemplateStore > & thePixelTemp_
float syone()
rms for one pixel y-clusters
float yxratio()
fractional distance in y between cotalpha templates slices
float qavg()
average cluster charge for this set of track angles
float dyone()
mean offset/correction for one pixel y-clusters
float dxtwo()
mean offset/correction for one double-pixel x-clusters
int id_current_
current id
float xxratio()
fractional distance in x between cotalpha templates
int SiPixelTemplate::qbin ( int  id,
float  cotalpha,
float  cotbeta,
float  qclus 
)

Interpolate beta/alpha angles to produce an expected average charge. Return int (0-4) describing the charge of the cluster [0: 1.5<Q/Qavg, 1: 1<Q/Qavg<1.5, 2: 0.85<Q/Qavg<1, 3: 0.95Qmin<Q<0.85Qavg, 4: Q<0.95Qmin].

Parameters
id- (input) index of the template to use
cotalpha- (input) the cotangent of the alpha track angle (see CMS IN 2004/014)
cotbeta- (input) the cotangent of the beta track angle (see CMS IN 2004/014)
locBz- (input) the sign of this quantity is used to determine whether to flip cot(beta)<0 quantities from cot(beta)>0 (FPix only) for FPix IP-related tracks, locBz < 0 for cot(beta) > 0 and locBz > 0 for cot(beta) < 0
qclus- (input) the cluster charge in electrons
pixmax- (output) the maximum pixel charge in electrons (truncation value)
sigmay- (output) the estimated y-error for CPEGeneric in microns
deltay- (output) the estimated y-bias for CPEGeneric in microns
sigmax- (output) the estimated x-error for CPEGeneric in microns
deltax- (output) the estimated x-bias for CPEGeneric in microns
sy1- (output) the estimated y-error for 1 single-pixel clusters in microns
dy1- (output) the estimated y-bias for 1 single-pixel clusters in microns
sy2- (output) the estimated y-error for 1 double-pixel clusters in microns
dy2- (output) the estimated y-bias for 1 double-pixel clusters in microns
sx1- (output) the estimated x-error for 1 single-pixel clusters in microns
dx1- (output) the estimated x-bias for 1 single-pixel clusters in microns
sx2- (output) the estimated x-error for 1 double-pixel clusters in microns
dx2- (output) the estimated x-bias for 1 double-pixel clusters in microns Interpolate beta/alpha angles to produce an expected average charge. Return int (0-4) describing the charge of the cluster [0: 1.5<Q/Qavg, 1: 1<Q/Qavg<1.5, 2: 0.85<Q/Qavg<1, 3: 0.95Qmin<Q<0.85Qavg, 4: Q<0.95Qmin].
id- (input) index of the template to use
cotalpha- (input) the cotangent of the alpha track angle (see CMS IN 2004/014)
cotbeta- (input) the cotangent of the beta track angle (see CMS IN 2004/014)
qclus- (input) the cluster charge in electrons

Definition at line 2531 of file SiPixelTemplate.cc.

References f, and qbin().

2532 {
2533  // Interpolate for a new set of track angles
2534 
2535  // Local variables
2536  float pixmx, sigmay, deltay, sigmax, deltax, sy1, dy1, sy2, dy2, sx1, dx1, sx2, dx2, locBz; // lorywidth, lorxwidth;
2537  locBz = -1.f;
2538  if(cotbeta < 0.f) {locBz = -locBz;}
2539  return SiPixelTemplate::qbin(id, cotalpha, cotbeta, locBz, qclus, pixmx, sigmay, deltay, sigmax, deltax,
2540  sy1, dy1, sy2, dy2, sx1, dx1, sx2, dx2); // , lorywidth, lorxwidth);
2541 
2542 } // qbin
int qbin(int id, float cotalpha, float cotbeta, float locBz, float qclus, float &pixmx, float &sigmay, float &deltay, float &sigmax, float &deltax, float &sy1, float &dy1, float &sy2, float &dy2, float &sx1, float &dx1, float &sx2, float &dx2)
double f[11][100]
int SiPixelTemplate::qbin ( int  id,
float  cotbeta,
float  qclus 
)

Interpolate beta/alpha angles to produce an expected average charge. Return int (0-4) describing the charge of the cluster [0: 1.5<Q/Qavg, 1: 1<Q/Qavg<1.5, 2: 0.85<Q/Qavg<1, 3: 0.95Qmin<Q<0.85Qavg, 4: Q<0.95Qmin].

Parameters
id- (input) index of the template to use
cotbeta- (input) the cotangent of the beta track angle (see CMS IN 2004/014)
qclus- (input) the cluster charge in electrons

Definition at line 2551 of file SiPixelTemplate.cc.

References f, and qbin().

2552 {
2553 // Interpolate for a new set of track angles
2554 
2555 // Local variables
2556  float pixmx, sigmay, deltay, sigmax, deltax, sy1, dy1, sy2, dy2, sx1, dx1, sx2, dx2, locBz; //, lorywidth, lorxwidth;
2557  const float cotalpha = 0.f;
2558  locBz = -1.f;
2559  if(cotbeta < 0.f) {locBz = -locBz;}
2560  return SiPixelTemplate::qbin(id, cotalpha, cotbeta, locBz, qclus, pixmx, sigmay, deltay, sigmax, deltax,
2561  sy1, dy1, sy2, dy2, sx1, dx1, sx2, dx2); // , lorywidth, lorxwidth);
2562 
2563 } // qbin
int qbin(int id, float cotalpha, float cotbeta, float locBz, float qclus, float &pixmx, float &sigmay, float &deltay, float &sigmax, float &deltax, float &sy1, float &dy1, float &sy2, float &dy2, float &sx1, float &dx1, float &sx2, float &dx2)
double f[11][100]
void SiPixelTemplate::qbin_dist ( int  id,
float  cotalpha,
float  cotbeta,
float  qbin_frac[4],
float &  ny1_frac,
float &  ny2_frac,
float &  nx1_frac,
float &  nx2_frac 
)

Interpolate beta/alpha angles to produce estimated errors for fastsim

Parameters
id- (input) index of the template to use
cotalpha- (input) the cotangent of the alpha track angle (see CMS IN 2004/014)
cotbeta- (input) the cotangent of the beta track angle (see CMS IN 2004/014)
qbin_frac[4]- (output) the integrated probability for qbin=0, 0+1, 0+1+2, 0+1+2+3 (1.)
ny1_frac- (output) the probability for ysize = 1 for a single-size pixel
ny2_frac- (output) the probability for ysize = 1 for a double-size pixel
nx1_frac- (output) the probability for xsize = 1 for a single-size pixel
nx2_frac- (output) the probability for xsize = 1 for a double-size pixel

Definition at line 2779 of file SiPixelTemplate.cc.

References assert(), edm::hlt::Exception, f, i, and cmsHarvester::index.

Referenced by SiPixelGaussianSmearingRecHitConverterAlgorithm::smearHit().

2781 {
2782  // Interpolate for a new set of track angles
2783 
2784  // Local variables
2785  int i;
2786  int ilow, ihigh, iylow, iyhigh, Ny, Nxx, Nyx, index;
2787  float yratio, yxratio, xxratio;
2788  float acotb, cotb;
2789  float qfrac[4];
2790  //bool flip_y;
2791 
2792  // Find the index corresponding to id
2793 
2794  index = -1;
2795  for(i=0; i<(int)thePixelTemp_.size(); ++i) {
2796 
2797  if(id == thePixelTemp_[i].head.ID) {
2798 
2799  index = i;
2800 // id_current_ = id;
2801  break;
2802  }
2803  }
2804 
2805 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
2806  if(index < 0 || index >= (int)thePixelTemp_.size()) {
2807  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::temperrors can't find needed template ID = " << id << std::endl;
2808  }
2809 #else
2810  assert(index >= 0 && index < (int)thePixelTemp_.size());
2811 #endif
2812 
2813  //
2814 
2815  // Interpolate the absolute value of cot(beta)
2816 
2817  acotb = fabs((double)cotbeta);
2818  cotb = cotbeta;
2819 
2820 
2821  // for some cosmics, the ususal gymnastics are incorrect
2822 
2823 // if(thePixelTemp_[index].head.Dtype == 0) {
2824  cotb = acotb;
2825  //flip_y = false;
2826  //if(cotbeta < 0.f) {flip_y = true;}
2827 // } else {
2828 // if(locBz < 0.f) {
2829 // cotb = cotbeta;
2830 // flip_y = false;
2831 // } else {
2832 // cotb = -cotbeta;
2833 // flip_y = true;
2834 // }
2835 // }
2836 
2837  // Copy the charge scaling factor to the private variable
2838 
2839  Ny = thePixelTemp_[index].head.NTy;
2840  Nyx = thePixelTemp_[index].head.NTyx;
2841  Nxx = thePixelTemp_[index].head.NTxx;
2842 
2843 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
2844  if(Ny < 2 || Nyx < 1 || Nxx < 2) {
2845  throw cms::Exception("DataCorrupt") << "template ID = " << id_current_ << "has too few entries: Ny/Nyx/Nxx = " << Ny << "/" << Nyx << "/" << Nxx << std::endl;
2846  }
2847 #else
2848  assert(Ny > 1 && Nyx > 0 && Nxx > 1);
2849 #endif
2850 
2851  // next, loop over all y-angle entries
2852 
2853  ilow = 0;
2854  yratio = 0.f;
2855 
2856  if(cotb >= thePixelTemp_[index].enty[Ny-1].cotbeta) {
2857 
2858  ilow = Ny-2;
2859  yratio = 1.f;
2860 
2861  } else {
2862 
2863  if(cotb >= thePixelTemp_[index].enty[0].cotbeta) {
2864 
2865  for (i=0; i<Ny-1; ++i) {
2866 
2867  if( thePixelTemp_[index].enty[i].cotbeta <= cotb && cotb < thePixelTemp_[index].enty[i+1].cotbeta) {
2868 
2869  ilow = i;
2870  yratio = (cotb - thePixelTemp_[index].enty[i].cotbeta)/(thePixelTemp_[index].enty[i+1].cotbeta - thePixelTemp_[index].enty[i].cotbeta);
2871  break;
2872  }
2873  }
2874  }
2875  }
2876 
2877  ihigh=ilow + 1;
2878 
2879  // Interpolate/store all y-related quantities (flip displacements when flip_y)
2880  ny1_frac = (1.f - yratio)*thePixelTemp_[index].enty[ilow].fracyone + yratio*thePixelTemp_[index].enty[ihigh].fracyone;
2881  ny2_frac = (1.f - yratio)*thePixelTemp_[index].enty[ilow].fracytwo + yratio*thePixelTemp_[index].enty[ihigh].fracytwo;
2882 
2883  // next, loop over all x-angle entries, first, find relevant y-slices
2884 
2885  iylow = 0;
2886  yxratio = 0.f;
2887 
2888  if(acotb >= thePixelTemp_[index].entx[Nyx-1][0].cotbeta) {
2889 
2890  iylow = Nyx-2;
2891  yxratio = 1.f;
2892 
2893  } else if(acotb >= thePixelTemp_[index].entx[0][0].cotbeta) {
2894 
2895  for (i=0; i<Nyx-1; ++i) {
2896 
2897  if( thePixelTemp_[index].entx[i][0].cotbeta <= acotb && acotb < thePixelTemp_[index].entx[i+1][0].cotbeta) {
2898 
2899  iylow = i;
2900  yxratio = (acotb - thePixelTemp_[index].entx[i][0].cotbeta)/(thePixelTemp_[index].entx[i+1][0].cotbeta - thePixelTemp_[index].entx[i][0].cotbeta);
2901  break;
2902  }
2903  }
2904  }
2905 
2906  iyhigh=iylow + 1;
2907 
2908  ilow = 0;
2909  xxratio = 0.f;
2910 
2911  if(cotalpha >= thePixelTemp_[index].entx[0][Nxx-1].cotalpha) {
2912 
2913  ilow = Nxx-2;
2914  xxratio = 1.f;
2915 
2916  } else {
2917 
2918  if(cotalpha >= thePixelTemp_[index].entx[0][0].cotalpha) {
2919 
2920  for (i=0; i<Nxx-1; ++i) {
2921 
2922  if( thePixelTemp_[index].entx[0][i].cotalpha <= cotalpha && cotalpha < thePixelTemp_[index].entx[0][i+1].cotalpha) {
2923 
2924  ilow = i;
2925  xxratio = (cotalpha - thePixelTemp_[index].entx[0][i].cotalpha)/(thePixelTemp_[index].entx[0][i+1].cotalpha - thePixelTemp_[index].entx[0][i].cotalpha);
2926  break;
2927  }
2928  }
2929  }
2930  }
2931 
2932  ihigh=ilow + 1;
2933 
2934  for(i=0; i<3; ++i) {
2935  qfrac[i]=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index].entx[iylow][ilow].qbfrac[i] + xxratio*thePixelTemp_[index].entx[iylow][ihigh].qbfrac[i])
2936  +yxratio*((1.f - xxratio)*thePixelTemp_[index].entx[iyhigh][ilow].qbfrac[i] + xxratio*thePixelTemp_[index].entx[iyhigh][ihigh].qbfrac[i]);
2937  }
2938  nx1_frac = (1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index].entx[iylow][ilow].fracxone + xxratio*thePixelTemp_[index].entx[iylow][ihigh].fracxone)
2939  +yxratio*((1.f - xxratio)*thePixelTemp_[index].entx[iyhigh][ilow].fracxone + xxratio*thePixelTemp_[index].entx[iyhigh][ihigh].fracxone);
2940  nx2_frac = (1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index].entx[iylow][ilow].fracxtwo + xxratio*thePixelTemp_[index].entx[iylow][ihigh].fracxtwo)
2941  +yxratio*((1.f - xxratio)*thePixelTemp_[index].entx[iyhigh][ilow].fracxtwo + xxratio*thePixelTemp_[index].entx[iyhigh][ihigh].fracxtwo);
2942 
2943 
2944 
2945  qbin_frac[0] = qfrac[0];
2946  qbin_frac[1] = qbin_frac[0] + qfrac[1];
2947  qbin_frac[2] = qbin_frac[1] + qfrac[2];
2948  qbin_frac[3] = 1.f;
2949  return;
2950 
2951 } // qbin
int i
Definition: DBlmapReader.cc:9
assert(m_qm.get())
float yratio()
fractional distance in y between cotbeta templates
double f[11][100]
const std::vector< SiPixelTemplateStore > & thePixelTemp_
float yxratio()
fractional distance in y between cotalpha templates slices
int id_current_
current id
float xxratio()
fractional distance in x between cotalpha templates
float SiPixelTemplate::qmin ( )
inline

minimum cluster charge for valid hit (keeps 99.9% of simulated hits)

Definition at line 334 of file SiPixelTemplate.h.

References qmin_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::qmin ( int  i)
inline

minimum cluster charge for valid hit (keeps 99.9% or 99.8% of simulated hits)

Definition at line 335 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, qmin2_, and qmin_.

float SiPixelTemplate::qscale ( )
inline

charge scaling factor

Definition at line 321 of file SiPixelTemplate.h.

References qscale_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::r_qMeas_qTrue ( )
inline

ratio of measured to true cluster charge

Definition at line 514 of file SiPixelTemplate.h.

References r_qMeas_qTrue_.

float SiPixelTemplate::s50 ( )
inline
float SiPixelTemplate::sigmavav ( )
inline

"sigma" scale fctor for Vavilov distribution

Definition at line 506 of file SiPixelTemplate.h.

References sigmavav_.

float SiPixelTemplate::sigmavav2 ( )
inline

"sigma" scale fctor for 2-cluster Vavilov distribution

Definition at line 509 of file SiPixelTemplate.h.

References sigmavav2_.

bool SiPixelTemplate::simpletemplate2D ( float  xhit,
float  yhit,
std::vector< bool > &  ydouble,
std::vector< bool > &  xdouble,
float  template2d[13+2][21+2] 
)

Make simple 2-D templates from track angles set in interpolate and hit position.

Parameters
xhit- (input) x-position of hit relative to the lower left corner of pixel[1][1] (to allow for the "padding" of the two-d clusters in the splitter)
yhit- (input) y-position of hit relative to the lower left corner of pixel[1][1]
ydouble- (input) STL vector of 21 element array to flag a double-pixel starting at cluster[1][1]
xdouble- (input) STL vector of 11 element array to flag a double-pixel starting at cluster[1][1]
template2d- (output) 2d template of size matched to the cluster. Input must be zeroed since charge is added only.

Definition at line 2963 of file SiPixelTemplate.cc.

References funct::abs(), any(), SimplePixel::btype, BXM3, BYM3, f, i, SimplePixel::i, j, SimplePixel::j, list(), bookConverter::max, SimplePixel::s, mathSSE::sqrt(), SimplePixel::x, and SimplePixel::y.

Referenced by SiPixelTemplateSplit::PixelTempSplit(), and TrackClusterSplitter::splitCluster().

2964 {
2965 
2966  // Local variables
2967 
2968  float x0, y0, xf, yf, xi, yi, sf, si, s0, qpix, slopey, slopex, ds;
2969  int i, j, jpix0, ipix0, jpixf, ipixf, jpix, ipix, nx, ny, anx, any, jmax, imax;
2970  float qtotal;
2971  // double path;
2972  std::list<SimplePixel> list;
2973  std::list<SimplePixel>::iterator listIter, listEnd;
2974 
2975  // Calculate the entry and exit points for the line charge from the track
2976 
2977  x0 = xhit - 0.5*zsize_*cota_current_;
2978  y0 = yhit - 0.5*zsize_*cotb_current_;
2979 
2980  jpix0 = floor(x0/xsize_)+1;
2981  ipix0 = floor(y0/ysize_)+1;
2982 
2983  if(jpix0 < 0 || jpix0 > BXM3) {return false;}
2984  if(ipix0 < 0 || ipix0 > BYM3) {return false;}
2985 
2986  xf = xhit + 0.5*zsize_*cota_current_ + lorxwidth_;
2987  yf = yhit + 0.5*zsize_*cotb_current_ + lorywidth_;
2988 
2989  jpixf = floor(xf/xsize_)+1;
2990  ipixf = floor(yf/ysize_)+1;
2991 
2992  if(jpixf < 0 || jpixf > BXM3) {return false;}
2993  if(ipixf < 0 || ipixf > BYM3) {return false;}
2994 
2995 // total charge length
2996 
2997  sf = std::sqrt((xf-x0)*(xf-x0) + (yf-y0)*(yf-y0));
2998  if((xf-x0) != 0.f) {slopey = (yf-y0)/(xf-x0);} else { slopey = 1.e10;}
2999  if((yf-y0) != 0.f) {slopex = (xf-x0)/(yf-y0);} else { slopex = 1.e10;}
3000 
3001 // use average charge in this direction
3002 
3003  qtotal = qavg_avg_;
3004 
3005  SimplePixel element;
3006  element.s = sf;
3007  element.x = xf;
3008  element.y = yf;
3009  element.i = ipixf;
3010  element.j = jpixf;
3011  element.btype = 0;
3012  list.push_back(element);
3013 
3014  // nx is the number of x interfaces crossed by the line charge
3015 
3016  nx = jpixf - jpix0;
3017  anx = abs(nx);
3018  if(anx > 0) {
3019  if(nx > 0) {
3020  for(j=jpix0; j<jpixf; ++j) {
3021  xi = xsize_*j;
3022  yi = slopey*(xi-x0) + y0;
3023  ipix = (int)(yi/ysize_)+1;
3024  si = std::sqrt((xi-x0)*(xi-x0) + (yi-y0)*(yi-y0));
3025  element.s = si;
3026  element.x = xi;
3027  element.y = yi;
3028  element.i = ipix;
3029  element.j = j;
3030  element.btype = 1;
3031  list.push_back(element);
3032  }
3033  } else {
3034  for(j=jpix0; j>jpixf; --j) {
3035  xi = xsize_*(j-1);
3036  yi = slopey*(xi-x0) + y0;
3037  ipix = (int)(yi/ysize_)+1;
3038  si = std::sqrt((xi-x0)*(xi-x0) + (yi-y0)*(yi-y0));
3039  element.s = si;
3040  element.x = xi;
3041  element.y = yi;
3042  element.i = ipix;
3043  element.j = j;
3044  element.btype = 1;
3045  list.push_back(element);
3046  }
3047  }
3048  }
3049 
3050  ny = ipixf - ipix0;
3051  any = abs(ny);
3052  if(any > 0) {
3053  if(ny > 0) {
3054  for(i=ipix0; i<ipixf; ++i) {
3055  yi = ysize_*i;
3056  xi = slopex*(yi-y0) + x0;
3057  jpix = (int)(xi/xsize_)+1;
3058  si = std::sqrt((xi-x0)*(xi-x0) + (yi-y0)*(yi-y0));
3059  element.s = si;
3060  element.x = xi;
3061  element.y = yi;
3062  element.i = i;
3063  element.j = jpix;
3064  element.btype = 2;
3065  list.push_back(element);
3066  }
3067  } else {
3068  for(i=ipix0; i>ipixf; --i) {
3069  yi = ysize_*(i-1);
3070  xi = slopex*(yi-y0) + x0;
3071  jpix = (int)(xi/xsize_)+1;
3072  si = std::sqrt((xi-x0)*(xi-x0) + (yi-y0)*(yi-y0));
3073  element.s = si;
3074  element.x = xi;
3075  element.y = yi;
3076  element.i = i;
3077  element.j = jpix;
3078  element.btype = 2;
3079  list.push_back(element);
3080  }
3081  }
3082  }
3083 
3084  imax = std::max(ipix0, ipixf);
3085  jmax = std::max(jpix0, jpixf);
3086 
3087  // Sort the list according to the distance from the initial point
3088 
3089  list.sort();
3090 
3091  // Look for double pixels and adjust the list appropriately
3092 
3093  for(i=1; i<imax; ++i) {
3094  if(ydouble[i-1]) {
3095  listIter = list.begin();
3096  if(ny > 0) {
3097  while(listIter != list.end()) {
3098  if(listIter->i == i && listIter->btype == 2) {
3099  listIter = list.erase(listIter);
3100  continue;
3101  }
3102  if(listIter->i > i) {
3103  --(listIter->i);
3104  }
3105  ++listIter;
3106  }
3107  } else {
3108  while(listIter != list.end()) {
3109  if(listIter->i == i+1 && listIter->btype == 2) {
3110  listIter = list.erase(listIter);
3111  continue;
3112  }
3113  if(listIter->i > i+1) {
3114  --(listIter->i);
3115  }
3116  ++listIter;
3117  }
3118  }
3119  }
3120  }
3121 
3122  for(j=1; j<jmax; ++j) {
3123  if(xdouble[j-1]) {
3124  listIter = list.begin();
3125  if(nx > 0) {
3126  while(listIter != list.end()) {
3127  if(listIter->j == j && listIter->btype == 1) {
3128  listIter = list.erase(listIter);
3129  continue;
3130  }
3131  if(listIter->j > j) {
3132  --(listIter->j);
3133  }
3134  ++listIter;
3135  }
3136  } else {
3137  while(listIter != list.end()) {
3138  if(listIter->j == j+1 && listIter->btype == 1) {
3139  listIter = list.erase(listIter);
3140  continue;
3141  }
3142  if(listIter->j > j+1) {
3143  --(listIter->j);
3144  }
3145  ++listIter;
3146  }
3147  }
3148  }
3149  }
3150 
3151  // The list now contains the path lengths of the line charge in each pixel from (x0,y0). Cacluate the lengths of the segments and the charge.
3152 
3153  s0 = 0.f;
3154  listIter = list.begin();
3155  listEnd = list.end();
3156  for( ;listIter != listEnd; ++listIter) {
3157  si = listIter->s;
3158  ds = si - s0;
3159  s0 = si;
3160  j = listIter->j;
3161  i = listIter->i;
3162  if(sf > 0.f) { qpix = qtotal*ds/sf;} else {qpix = qtotal;}
3163  template2d[j][i] += qpix;
3164  }
3165 
3166  return true;
3167 
3168 } // simpletemplate2D
float cota_current_
current cot alpha
int i
Definition: DBlmapReader.cc:9
int btype
type of boundary (0=end, 1 = x-boundary, 2 = y-boundary)
Definition: SimplePixel.h:22
bool any(const std::vector< T > &v, const T &what)
Definition: ECalSD.cc:34
#define BXM3
float y
y coordinate of boundary intersection
Definition: SimplePixel.h:19
float cotb_current_
current cot beta
T sqrt(T t)
Definition: SSEVec.h:48
float lorywidth_
Lorentz y-width (sign corrected for fpix frame)
float s
distance from track entry
Definition: SimplePixel.h:17
Abs< T >::type abs(const T &t)
Definition: Abs.h:22
int j
Definition: DBlmapReader.cc:9
int j
y index of traversed pixel
Definition: SimplePixel.h:21
double f[11][100]
float ysize_
Pixel y-size.
#define BYM3
float xsize_
Pixel x-size.
float qavg_avg_
average of cluster charge less than qavg
float x
x coordinate of boundary intersection
Definition: SimplePixel.h:18
int i
x index of traversed pixel
Definition: SimplePixel.h:20
float lorxwidth_
Lorentz x-width.
float zsize_
Pixel z-size (thickness)
How EventSelector::AcceptEvent() decides whether to accept an event for output otherwise it is excluding the probing of A single or multiple positive and the trigger will pass if any such matching triggers are PASS or EXCEPTION[A criterion thatmatches no triggers at all is detected and causes a throw.] A single negative with an expectation of appropriate bit checking in the decision and the trigger will pass if any such matching triggers are FAIL or EXCEPTION A wildcarded negative criterion that matches more than one trigger in the trigger list("!*","!HLTx*"if it matches 2 triggers or more) will accept the event if all the matching triggers are FAIL.It will reject the event if any of the triggers are PASS or EXCEPTION(this matches the behavior of"!*"before the partial wildcard feature was incorporated).Triggers which are in the READY state are completely ignored.(READY should never be returned since the trigger paths have been run
float SiPixelTemplate::ss50 ( )
inline

1/2 of the pixel threshold signal in electrons

Definition at line 323 of file SiPixelTemplate.h.

References ss50_.

float SiPixelTemplate::sxmax ( )
inline

average pixel signal for x-projection of cluster

Definition at line 329 of file SiPixelTemplate.h.

References sxmax_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::sxone ( )
inline

rms for one pixel x-clusters

Definition at line 331 of file SiPixelTemplate.h.

References sxone_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::sxtwo ( )
inline

rms for one double-pixel x-clusters

Definition at line 333 of file SiPixelTemplate.h.

References sxtwo_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::symax ( )
inline

average pixel signal for y-projection of cluster

Definition at line 324 of file SiPixelTemplate.h.

References symax_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::syone ( )
inline

rms for one pixel y-clusters

Definition at line 326 of file SiPixelTemplate.h.

References syone_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::sytwo ( )
inline

rms for one double-pixel y-clusters

Definition at line 328 of file SiPixelTemplate.h.

References sytwo_.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

void SiPixelTemplate::temperrors ( int  id,
float  cotalpha,
float  cotbeta,
int  qBin,
float &  sigmay,
float &  sigmax,
float &  sy1,
float &  sy2,
float &  sx1,
float &  sx2 
)

Interpolate beta/alpha angles to produce estimated errors for fastsim

Parameters
id- (input) index of the template to use
cotalpha- (input) the cotangent of the alpha track angle (see CMS IN 2004/014)
cotbeta- (input) the cotangent of the beta track angle (see CMS IN 2004/014)
qBin- (input) charge bin from 0-3
sigmay- (output) the estimated y-error for CPETemplate in microns
sigmax- (output) the estimated x-error for CPETemplate in microns
sy1- (output) the estimated y-error for 1 single-pixel clusters in microns
sy2- (output) the estimated y-error for 1 double-pixel clusters in microns
sx1- (output) the estimated x-error for 1 single-pixel clusters in microns
sx2- (output) the estimated x-error for 1 double-pixel clusters in microns

Definition at line 2580 of file SiPixelTemplate.cc.

References assert(), edm::hlt::Exception, f, i, and cmsHarvester::index.

Referenced by SiPixelGaussianSmearingRecHitConverterAlgorithm::smearHit().

2582 {
2583  // Interpolate for a new set of track angles
2584 
2585  // Local variables
2586  int i;
2587  int ilow, ihigh, iylow, iyhigh, Ny, Nxx, Nyx, index;
2588  float yratio, yxratio, xxratio;
2589  float acotb, cotb;
2590  float yrms, xrms;
2591  //bool flip_y;
2592 
2593 
2594  // Find the index corresponding to id
2595 
2596  index = -1;
2597  for(i=0; i<(int)thePixelTemp_.size(); ++i) {
2598 
2599  if(id == thePixelTemp_[i].head.ID) {
2600 
2601  index = i;
2602  break;
2603  }
2604  }
2605 
2606 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
2607  if(index < 0 || index >= (int)thePixelTemp_.size()) {
2608  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::temperrors can't find needed template ID = " << id << std::endl;
2609  }
2610 #else
2611  assert(index >= 0 && index < (int)thePixelTemp_.size());
2612 #endif
2613 
2614 
2615 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
2616  if(qBin < 0 || qBin > 5) {
2617  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::temperrors called with illegal qBin = " << qBin << std::endl;
2618  }
2619 #else
2620  assert(qBin >= 0 && qBin < 6);
2621 #endif
2622 
2623 // The error information for qBin > 3 is taken to be the same as qBin=3
2624 
2625  if(qBin > 3) {qBin = 3;}
2626  //
2627 
2628  // Interpolate the absolute value of cot(beta)
2629 
2630  acotb = fabs((double)cotbeta);
2631  cotb = cotbeta;
2632 
2633  // for some cosmics, the ususal gymnastics are incorrect
2634 
2635 // if(thePixelTemp_[index].head.Dtype == 0) {
2636  cotb = acotb;
2637  //flip_y = false;
2638  //if(cotbeta < 0.f) {flip_y = true;}
2639 // } else {
2640 // if(locBz < 0.f) {
2641 // cotb = cotbeta;
2642 // flip_y = false;
2643 // } else {
2644 // cotb = -cotbeta;
2645 // flip_y = true;
2646 // }
2647 // }
2648 
2649  // Copy the charge scaling factor to the private variable
2650 
2651  Ny = thePixelTemp_[index].head.NTy;
2652  Nyx = thePixelTemp_[index].head.NTyx;
2653  Nxx = thePixelTemp_[index].head.NTxx;
2654 
2655 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
2656  if(Ny < 2 || Nyx < 1 || Nxx < 2) {
2657  throw cms::Exception("DataCorrupt") << "template ID = " << id_current_ << "has too few entries: Ny/Nyx/Nxx = " << Ny << "/" << Nyx << "/" << Nxx << std::endl;
2658  }
2659 #else
2660  assert(Ny > 1 && Nyx > 0 && Nxx > 1);
2661 #endif
2662 
2663  // next, loop over all y-angle entries
2664 
2665  ilow = 0;
2666  yratio = 0.f;
2667 
2668  if(cotb >= thePixelTemp_[index].enty[Ny-1].cotbeta) {
2669 
2670  ilow = Ny-2;
2671  yratio = 1.f;
2672 
2673  } else {
2674 
2675  if(cotb >= thePixelTemp_[index].enty[0].cotbeta) {
2676 
2677  for (i=0; i<Ny-1; ++i) {
2678 
2679  if( thePixelTemp_[index].enty[i].cotbeta <= cotb && cotb < thePixelTemp_[index].enty[i+1].cotbeta) {
2680 
2681  ilow = i;
2682  yratio = (cotb - thePixelTemp_[index].enty[i].cotbeta)/(thePixelTemp_[index].enty[i+1].cotbeta - thePixelTemp_[index].enty[i].cotbeta);
2683  break;
2684  }
2685  }
2686  }
2687  }
2688 
2689  ihigh=ilow + 1;
2690 
2691  // Interpolate/store all y-related quantities (flip displacements when flip_y)
2692 
2693  sy1 = (1.f - yratio)*thePixelTemp_[index].enty[ilow].syone + yratio*thePixelTemp_[index].enty[ihigh].syone;
2694  sy2 = (1.f - yratio)*thePixelTemp_[index].enty[ilow].sytwo + yratio*thePixelTemp_[index].enty[ihigh].sytwo;
2695  yrms=(1.f - yratio)*thePixelTemp_[index].enty[ilow].yrms[qBin] + yratio*thePixelTemp_[index].enty[ihigh].yrms[qBin];
2696 
2697 
2698  // next, loop over all x-angle entries, first, find relevant y-slices
2699 
2700  iylow = 0;
2701  yxratio = 0.f;
2702 
2703  if(acotb >= thePixelTemp_[index].entx[Nyx-1][0].cotbeta) {
2704 
2705  iylow = Nyx-2;
2706  yxratio = 1.f;
2707 
2708  } else if(acotb >= thePixelTemp_[index].entx[0][0].cotbeta) {
2709 
2710  for (i=0; i<Nyx-1; ++i) {
2711 
2712  if( thePixelTemp_[index].entx[i][0].cotbeta <= acotb && acotb < thePixelTemp_[index].entx[i+1][0].cotbeta) {
2713 
2714  iylow = i;
2715  yxratio = (acotb - thePixelTemp_[index].entx[i][0].cotbeta)/(thePixelTemp_[index].entx[i+1][0].cotbeta - thePixelTemp_[index].entx[i][0].cotbeta);
2716  break;
2717  }
2718  }
2719  }
2720 
2721  iyhigh=iylow + 1;
2722 
2723  ilow = 0;
2724  xxratio = 0.f;
2725 
2726  if(cotalpha >= thePixelTemp_[index].entx[0][Nxx-1].cotalpha) {
2727 
2728  ilow = Nxx-2;
2729  xxratio = 1.f;
2730 
2731  } else {
2732 
2733  if(cotalpha >= thePixelTemp_[index].entx[0][0].cotalpha) {
2734 
2735  for (i=0; i<Nxx-1; ++i) {
2736 
2737  if( thePixelTemp_[index].entx[0][i].cotalpha <= cotalpha && cotalpha < thePixelTemp_[index].entx[0][i+1].cotalpha) {
2738 
2739  ilow = i;
2740  xxratio = (cotalpha - thePixelTemp_[index].entx[0][i].cotalpha)/(thePixelTemp_[index].entx[0][i+1].cotalpha - thePixelTemp_[index].entx[0][i].cotalpha);
2741  break;
2742  }
2743  }
2744  }
2745  }
2746 
2747  ihigh=ilow + 1;
2748 
2749  sx1 = (1.f - xxratio)*thePixelTemp_[index].entx[0][ilow].sxone + xxratio*thePixelTemp_[index].entx[0][ihigh].sxone;
2750  sx2 = (1.f - xxratio)*thePixelTemp_[index].entx[0][ilow].sxtwo + xxratio*thePixelTemp_[index].entx[0][ihigh].sxtwo;
2751 
2752  xrms=(1.f - yxratio)*((1.f - xxratio)*thePixelTemp_[index].entx[iylow][ilow].xrms[qBin] + xxratio*thePixelTemp_[index].entx[iylow][ihigh].xrms[qBin])
2753  +yxratio*((1.f - xxratio)*thePixelTemp_[index].entx[iyhigh][ilow].xrms[qBin] + xxratio*thePixelTemp_[index].entx[iyhigh][ihigh].xrms[qBin]);
2754 
2755 
2756 
2757 
2758  // Take the errors and bias from the correct charge bin
2759 
2760  sigmay = yrms;
2761 
2762  sigmax = xrms;
2763 
2764  return;
2765 
2766 } // temperrors
int i
Definition: DBlmapReader.cc:9
assert(m_qm.get())
float yratio()
fractional distance in y between cotbeta templates
float xrms(int i)
average x-rms of reconstruction binned in 4 charge bins
float sytwo()
rms for one double-pixel y-clusters
float sxone()
rms for one pixel x-clusters
float yrms(int i)
average y-rms of reconstruction binned in 4 charge bins
double f[11][100]
float sxtwo()
rms for one double-pixel x-clusters
const std::vector< SiPixelTemplateStore > & thePixelTemp_
float syone()
rms for one pixel y-clusters
float yxratio()
fractional distance in y between cotalpha templates slices
int id_current_
current id
float xxratio()
fractional distance in x between cotalpha templates
void SiPixelTemplate::vavilov2_pars ( double &  mpv,
double &  sigma,
double &  kappa 
)

Interpolate beta/alpha angles to produce Vavilov parameters for the 2-cluster charge distribution

Parameters
mpv- (output) the Vavilov most probable charge (well, not really the most probable esp at large kappa)
sigma- (output) the Vavilov sigma parameter
kappa- (output) the Vavilov kappa parameter [0.01 (Landau-like) < kappa < 10 (Gaussian-like)

Definition at line 3255 of file SiPixelTemplate.cc.

References assert(), edm::hlt::Exception, f, i, and mathSSE::sqrt().

Referenced by SiPixelTemplateSplit::PixelTempSplit().

3257 {
3258  // Local variables
3259  int i;
3260  int ilow, ihigh, Ny;
3261  float yratio, cotb, cotalpha0, arg;
3262 
3263  // Interpolate in cotbeta only for the correct total path length (converts cotalpha, cotbeta into an effective cotbeta)
3264 
3265  cotalpha0 = thePixelTemp_[index_id_].enty[0].cotalpha;
3266  arg = cotb_current_*cotb_current_ + cota_current_*cota_current_ - cotalpha0*cotalpha0;
3267  if(arg < 0.f) arg = 0.f;
3268  cotb = std::sqrt(arg);
3269 
3270  // Copy the charge scaling factor to the private variable
3271 
3272  Ny = thePixelTemp_[index_id_].head.NTy;
3273 
3274 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
3275  if(Ny < 2) {
3276  throw cms::Exception("DataCorrupt") << "template ID = " << id_current_ << "has too few entries: Ny = " << Ny << std::endl;
3277  }
3278 #else
3279  assert(Ny > 1);
3280 #endif
3281 
3282  // next, loop over all y-angle entries
3283 
3284  ilow = 0;
3285  yratio = 0.f;
3286 
3287  if(cotb >= thePixelTemp_[index_id_].enty[Ny-1].cotbeta) {
3288 
3289  ilow = Ny-2;
3290  yratio = 1.f;
3291 
3292  } else {
3293 
3294  if(cotb >= thePixelTemp_[index_id_].enty[0].cotbeta) {
3295 
3296  for (i=0; i<Ny-1; ++i) {
3297 
3298  if( thePixelTemp_[index_id_].enty[i].cotbeta <= cotb && cotb < thePixelTemp_[index_id_].enty[i+1].cotbeta) {
3299 
3300  ilow = i;
3301  yratio = (cotb - thePixelTemp_[index_id_].enty[i].cotbeta)/(thePixelTemp_[index_id_].enty[i+1].cotbeta - thePixelTemp_[index_id_].enty[i].cotbeta);
3302  break;
3303  }
3304  }
3305  }
3306  }
3307 
3308  ihigh=ilow + 1;
3309 
3310  // Interpolate Vavilov parameters
3311 
3312  mpvvav2_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].mpvvav2 + yratio*thePixelTemp_[index_id_].enty[ihigh].mpvvav2;
3313  sigmavav2_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].sigmavav2 + yratio*thePixelTemp_[index_id_].enty[ihigh].sigmavav2;
3314  kappavav2_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].kappavav2 + yratio*thePixelTemp_[index_id_].enty[ihigh].kappavav2;
3315 
3316  // Copy to parameter list
3317 
3318  mpv = (double)mpvvav2_;
3319  sigma = (double)sigmavav2_;
3320  kappa = (double)kappavav2_;
3321 
3322  return;
3323 
3324 } // vavilov2_pars
float cota_current_
current cot alpha
int i
Definition: DBlmapReader.cc:9
float kappavav2()
kappa parameter for 2-cluster Vavilov distribution
assert(m_qm.get())
float yratio()
fractional distance in y between cotbeta templates
float mpvvav2()
most probable charge in 2-cluster Vavilov distribution (not actually for larger kappa) ...
float cotb_current_
current cot beta
A arg
Definition: Factorize.h:36
float kappavav2_
kappa parameter for 2-cluster Vavilov distribution
T sqrt(T t)
Definition: SSEVec.h:48
float sigmavav2_
&quot;sigma&quot; scale fctor for 2-cluster Vavilov distribution
double f[11][100]
float sigmavav2()
&quot;sigma&quot; scale fctor for 2-cluster Vavilov distribution
const std::vector< SiPixelTemplateStore > & thePixelTemp_
static const G4double kappa
int index_id_
current index
float mpvvav2_
most probable charge in 2-cluster Vavilov distribution (not actually for larger kappa) ...
int id_current_
current id
void SiPixelTemplate::vavilov_pars ( double &  mpv,
double &  sigma,
double &  kappa 
)

Interpolate beta/alpha angles to produce Vavilov parameters for the charge distribution

Parameters
mpv- (output) the Vavilov most probable charge (well, not really the most probable esp at large kappa)
sigma- (output) the Vavilov sigma parameter
kappa- (output) the Vavilov kappa parameter [0.01 (Landau-like) < kappa < 10 (Gaussian-like)

Definition at line 3177 of file SiPixelTemplate.cc.

References assert(), edm::hlt::Exception, f, i, and mathSSE::sqrt().

Referenced by SiPixelTemplateReco::PixelTempReco2D().

3179 {
3180  // Local variables
3181  int i;
3182  int ilow, ihigh, Ny;
3183  float yratio, cotb, cotalpha0, arg;
3184 
3185 // Interpolate in cotbeta only for the correct total path length (converts cotalpha, cotbeta into an effective cotbeta)
3186 
3187  cotalpha0 = thePixelTemp_[index_id_].enty[0].cotalpha;
3188  arg = cotb_current_*cotb_current_ + cota_current_*cota_current_ - cotalpha0*cotalpha0;
3189  if(arg < 0.f) arg = 0.f;
3190  cotb = std::sqrt(arg);
3191 
3192 // Copy the charge scaling factor to the private variable
3193 
3194  Ny = thePixelTemp_[index_id_].head.NTy;
3195 
3196 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
3197  if(Ny < 2) {
3198  throw cms::Exception("DataCorrupt") << "template ID = " << id_current_ << "has too few entries: Ny = " << Ny << std::endl;
3199  }
3200 #else
3201  assert(Ny > 1);
3202 #endif
3203 
3204 // next, loop over all y-angle entries
3205 
3206  ilow = 0;
3207  yratio = 0.f;
3208 
3209  if(cotb >= thePixelTemp_[index_id_].enty[Ny-1].cotbeta) {
3210 
3211  ilow = Ny-2;
3212  yratio = 1.f;
3213 
3214  } else {
3215 
3216  if(cotb >= thePixelTemp_[index_id_].enty[0].cotbeta) {
3217 
3218  for (i=0; i<Ny-1; ++i) {
3219 
3220  if( thePixelTemp_[index_id_].enty[i].cotbeta <= cotb && cotb < thePixelTemp_[index_id_].enty[i+1].cotbeta) {
3221 
3222  ilow = i;
3223  yratio = (cotb - thePixelTemp_[index_id_].enty[i].cotbeta)/(thePixelTemp_[index_id_].enty[i+1].cotbeta - thePixelTemp_[index_id_].enty[i].cotbeta);
3224  break;
3225  }
3226  }
3227  }
3228  }
3229 
3230  ihigh=ilow + 1;
3231 
3232 // Interpolate Vavilov parameters
3233 
3234  mpvvav_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].mpvvav + yratio*thePixelTemp_[index_id_].enty[ihigh].mpvvav;
3235  sigmavav_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].sigmavav + yratio*thePixelTemp_[index_id_].enty[ihigh].sigmavav;
3236  kappavav_ = (1.f - yratio)*thePixelTemp_[index_id_].enty[ilow].kappavav + yratio*thePixelTemp_[index_id_].enty[ihigh].kappavav;
3237 
3238 // Copy to parameter list
3239 
3240 
3241  mpv = (double)mpvvav_;
3242  sigma = (double)sigmavav_;
3243  kappa = (double)kappavav_;
3244 
3245  return;
3246 
3247 } // vavilov_pars
float cota_current_
current cot alpha
float mpvvav_
most probable charge in Vavilov distribution (not actually for larger kappa)
int i
Definition: DBlmapReader.cc:9
float mpvvav()
most probable charge in Vavilov distribution (not actually for larger kappa)
assert(m_qm.get())
float yratio()
fractional distance in y between cotbeta templates
float cotb_current_
current cot beta
A arg
Definition: Factorize.h:36
float sigmavav_
&quot;sigma&quot; scale fctor for Vavilov distribution
float kappavav()
kappa parameter for Vavilov distribution
T sqrt(T t)
Definition: SSEVec.h:48
double f[11][100]
float kappavav_
kappa parameter for Vavilov distribution
const std::vector< SiPixelTemplateStore > & thePixelTemp_
float sigmavav()
&quot;sigma&quot; scale fctor for Vavilov distribution
static const G4double kappa
int index_id_
current index
int id_current_
current id
float SiPixelTemplate::xavg ( int  i)
inline

average x-bias of reconstruction binned in 4 charge bins

Definition at line 375 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and xavg_.

Referenced by SiPixelTemplateReco::PixelTempReco2D().

float SiPixelTemplate::xavgc2m ( int  i)
inline

1st pass chi2 min search: average x-bias of reconstruction binned in 4 charge bins

Definition at line 459 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and xavgc2m_.

Referenced by SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::xflcorr ( int  binq,
float  qflx 
)

Return interpolated x-correction for input charge bin and qflx

Parameters
binq- (input) charge bin [0-3]
qflx- (input) (Q_f-Q_l)/(Q_f+Q_l) for this cluster

Definition at line 1739 of file SiPixelTemplate.cc.

References assert(), edm::hlt::Exception, and f.

Referenced by SiPixelTemplateReco::PixelTempReco2D().

1741 {
1742  // Interpolate using quantities already stored in the private variables
1743 
1744  // Local variables
1745  float qfl, qfl2, qfl3, qfl4, qfl5, dx;
1746 
1747  // Make sure that input is OK
1748 
1749 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1750  if(binq < 0 || binq > 3) {
1751  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::xflcorr called with binq = " << binq << std::endl;
1752  }
1753 #else
1754  assert(binq >= 0 && binq < 4);
1755 #endif
1756 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1757  if(fabs((double)qflx) > 1.) {
1758  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::xflcorr called with qflx = " << qflx << std::endl;
1759  }
1760 #else
1761  assert(fabs((double)qflx) <= 1.);
1762 #endif
1763 
1764 // Define the maximum signal to allow before de-weighting a pixel
1765 
1766  qfl = qflx;
1767 
1768  if(qfl < -0.9f) {qfl = -0.9f;}
1769  if(qfl > 0.9f) {qfl = 0.9f;}
1770 
1771 // Interpolate between the two polynomials
1772 
1773  qfl2 = qfl*qfl; qfl3 = qfl2*qfl; qfl4 = qfl3*qfl; qfl5 = qfl4*qfl;
1774  dx = (1.f - yxratio_)*((1.f-xxratio_)*(xflparll_[binq][0]+xflparll_[binq][1]*qfl+xflparll_[binq][2]*qfl2+xflparll_[binq][3]*qfl3+xflparll_[binq][4]*qfl4+xflparll_[binq][5]*qfl5)
1775  + xxratio_*(xflparlh_[binq][0]+xflparlh_[binq][1]*qfl+xflparlh_[binq][2]*qfl2+xflparlh_[binq][3]*qfl3+xflparlh_[binq][4]*qfl4+xflparlh_[binq][5]*qfl5))
1776  + yxratio_*((1.f-xxratio_)*(xflparhl_[binq][0]+xflparhl_[binq][1]*qfl+xflparhl_[binq][2]*qfl2+xflparhl_[binq][3]*qfl3+xflparhl_[binq][4]*qfl4+xflparhl_[binq][5]*qfl5)
1777  + xxratio_*(xflparhh_[binq][0]+xflparhh_[binq][1]*qfl+xflparhh_[binq][2]*qfl2+xflparhh_[binq][3]*qfl3+xflparhh_[binq][4]*qfl4+xflparhh_[binq][5]*qfl5));
1778 
1779  return dx;
1780 
1781 } // End xflcorr
assert(m_qm.get())
float xflparlh_[4][6]
Aqfl-parameterized x-correction in 4 charge bins for smaller cotbeta, larger cotalpha.
double f[11][100]
float xflparhh_[4][6]
Aqfl-parameterized x-correction in 4 charge bins for larger cotbeta, cotalpha.
float xflparll_[4][6]
Aqfl-parameterized x-correction in 4 charge bins for smaller cotbeta, cotalpha.
float xxratio_
fractional distance in x between cotalpha templates
float xflparhl_[4][6]
Aqfl-parameterized x-correction in 4 charge bins for larger cotbeta, smaller cotalpha.
float yxratio_
fractional distance in y between x-slices of cotalpha templates
float SiPixelTemplate::xgsig ( int  i)
inline

average sigma_x from Gaussian fit binned in 4 charge bins

Definition at line 396 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and xgsig_.

float SiPixelTemplate::xgx0 ( int  i)
inline

average x0 from Gaussian fit binned in 4 charge bins

Definition at line 389 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and xgx0_.

float SiPixelTemplate::xrms ( int  i)
inline

average x-rms of reconstruction binned in 4 charge bins

Definition at line 382 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and xrms_.

Referenced by SiPixelTemplateReco::PixelTempReco2D().

float SiPixelTemplate::xrmsc2m ( int  i)
inline

1st pass chi2 min search: average x-rms of reconstruction binned in 4 charge bins

Definition at line 466 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and xrmsc2m_.

Referenced by SiPixelTemplateSplit::PixelTempSplit().

void SiPixelTemplate::xsigma2 ( int  fxpix,
int  lxpix,
float  sxthr,
float  xsum[13+4],
float  xsig2[13+4] 
)

Return vector of x errors (squared) for an input vector of projected signals Add large Q scaling for use in cluster splitting.

Parameters
fxpix- (input) index of the first real pixel in the projected cluster (doesn't include pseudopixels)
lxpix- (input) index of the last real pixel in the projected cluster (doesn't include pseudopixels)
sxthr- (input) maximum signal before de-weighting
xsum- (input) 11-element vector of pixel signals
xsig2- (output) 11-element vector of x errors (squared)

Definition at line 1578 of file SiPixelTemplate.cc.

References assert(), BHX, BXM2, ENDL, edm::hlt::Exception, f, i, and LOGERROR.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelTemplateSplit::PixelTempSplit().

1580 {
1581  // Interpolate using quantities already stored in the private variables
1582 
1583  // Local variables
1584  int i;
1585  float sigi, sigi2, sigi3, sigi4, yint, sxmax, x0, qscale;
1586 
1587  // Make sure that input is OK
1588 
1589 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1590  if(fxpix < 2 || fxpix >= BXM2) {
1591  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::xsigma2 called with fxpix = " << fxpix << std::endl;
1592  }
1593 #else
1594  assert(fxpix > 1 && fxpix < BXM2);
1595 #endif
1596 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1597  if(lxpix < fxpix || lxpix >= BXM2) {
1598  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::xsigma2 called with lxpix/fxpix = " << lxpix << "/" << fxpix << std::endl;
1599  }
1600 #else
1601  assert(lxpix >= fxpix && lxpix < BXM2);
1602 #endif
1603 
1604 // Define the maximum signal to use in the parameterization
1605 
1606  sxmax = sxmax_;
1607  if(sxmax_ > sxparmax_) {sxmax = sxparmax_;}
1608 
1609 // Evaluate pixel-by-pixel uncertainties (weights) for the templ analysis
1610 
1611  for(i=fxpix-2; i<=lxpix+2; ++i) {
1612  if(i < fxpix || i > lxpix) {
1613 
1614 // Nearest pseudopixels have uncertainties of 50% of threshold, next-nearest have 10% of threshold
1615 
1616  xsig2[i] = s50_*s50_;
1617  } else {
1618  if(xsum[i] < sxmax) {
1619  sigi = xsum[i];
1620  qscale = 1.f;
1621  } else {
1622  sigi = sxmax;
1623  qscale = xsum[i]/sxmax;
1624  }
1625  sigi2 = sigi*sigi; sigi3 = sigi2*sigi; sigi4 = sigi3*sigi;
1626 
1627 // First, do the cotbeta interpolation
1628 
1629  if(i <= BHX) {
1630  yint = (1.f-yratio_)*
1631  (xparly0_[0][0]+xparly0_[0][1]*sigi+xparly0_[0][2]*sigi2+xparly0_[0][3]*sigi3+xparly0_[0][4]*sigi4)
1632  + yratio_*
1633  (xparhy0_[0][0]+xparhy0_[0][1]*sigi+xparhy0_[0][2]*sigi2+xparhy0_[0][3]*sigi3+xparhy0_[0][4]*sigi4);
1634  } else {
1635  yint = (1.f-yratio_)*
1636  (xparly0_[1][0]+xparly0_[1][1]*sigi+xparly0_[1][2]*sigi2+xparly0_[1][3]*sigi3+xparly0_[1][4]*sigi4)
1637  + yratio_*
1638  (xparhy0_[1][0]+xparhy0_[1][1]*sigi+xparhy0_[1][2]*sigi2+xparhy0_[1][3]*sigi3+xparhy0_[1][4]*sigi4);
1639  }
1640 
1641 // Next, do the cotalpha interpolation
1642 
1643  if(i <= BHX) {
1644  xsig2[i] = (1.f-xxratio_)*
1645  (xparl_[0][0]+xparl_[0][1]*sigi+xparl_[0][2]*sigi2+xparl_[0][3]*sigi3+xparl_[0][4]*sigi4)
1646  + xxratio_*
1647  (xparh_[0][0]+xparh_[0][1]*sigi+xparh_[0][2]*sigi2+xparh_[0][3]*sigi3+xparh_[0][4]*sigi4);
1648  } else {
1649  xsig2[i] = (1.f-xxratio_)*
1650  (xparl_[1][0]+xparl_[1][1]*sigi+xparl_[1][2]*sigi2+xparl_[1][3]*sigi3+xparl_[1][4]*sigi4)
1651  + xxratio_*
1652  (xparh_[1][0]+xparh_[1][1]*sigi+xparh_[1][2]*sigi2+xparh_[1][3]*sigi3+xparh_[1][4]*sigi4);
1653  }
1654 
1655 // Finally, get the mid-point value of the cotalpha function
1656 
1657  if(i <= BHX) {
1658  x0 = xpar0_[0][0]+xpar0_[0][1]*sigi+xpar0_[0][2]*sigi2+xpar0_[0][3]*sigi3+xpar0_[0][4]*sigi4;
1659  } else {
1660  x0 = xpar0_[1][0]+xpar0_[1][1]*sigi+xpar0_[1][2]*sigi2+xpar0_[1][3]*sigi3+xpar0_[1][4]*sigi4;
1661  }
1662 
1663 // Finally, rescale the yint value for cotalpha variation
1664 
1665  if(x0 != 0.f) {xsig2[i] = xsig2[i]/x0 * yint;}
1666  xsig2[i] *=qscale;
1667  if(xsum[i] > sxthr) {xsig2[i] = 1.e8;}
1668  if(xsig2[i] <= 0.f) {LOGERROR("SiPixelTemplate") << "neg x-error-squared, id = " << id_current_ << ", index = " << index_id_ <<
1669  ", cot(alpha) = " << cota_current_ << ", cot(beta) = " << cotb_current_ << ", sigi = " << sigi << ENDL;}
1670  }
1671  }
1672 
1673  return;
1674 
1675 } // End xsigma2
float cota_current_
current cot alpha
int i
Definition: DBlmapReader.cc:9
#define LOGERROR(x)
float xparhy0_[2][5]
projected x-pixel uncertainty parameterization for larger cotbeta (central alpha) ...
assert(m_qm.get())
float sxmax_
average pixel signal for x-projection of cluster
float cotb_current_
current cot beta
float s50_
1/2 of the pixel threshold signal in adc units
#define BXM2
float qscale()
charge scaling factor
float sxparmax_
maximum pixel signal for parameterization of x uncertainties
#define ENDL
#define BHX
double f[11][100]
float xpar0_[2][5]
projected x-pixel uncertainty parameterization for central cotalpha
float xparh_[2][5]
projected x-pixel uncertainty parameterization for larger cotalpha
float xparl_[2][5]
projected x-pixel uncertainty parameterization for smaller cotalpha
float sxmax()
average pixel signal for x-projection of cluster
float xparly0_[2][5]
projected x-pixel uncertainty parameterization for smaller cotbeta (central alpha) ...
float xxratio_
fractional distance in x between cotalpha templates
int index_id_
current index
int id_current_
current id
float yratio_
fractional distance in y between cotbeta templates
float SiPixelTemplate::xsize ( )
inline
void SiPixelTemplate::xtemp ( int  fxbin,
int  lxbin,
float  xtemplate[41][13+4] 
)

Return interpolated y-template in single call

Parameters
fxbin- (input) index of first bin (0-40) to fill
fxbin- (input) index of last bin (0-40) to fill
xtemplate- (output) a 41x11 output buffer

Definition at line 1869 of file SiPixelTemplate.cc.

References assert(), BXM1, BXM2, BXSIZE, edm::hlt::Exception, i, and j.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelGaussianSmearingRecHitConverterAlgorithm::smearHit().

1871 {
1872  // Retrieve already interpolated quantities
1873 
1874  // Local variables
1875  int i, j;
1876 
1877  // Verify that input parameters are in valid range
1878 
1879 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1880  if(fxbin < 0 || fxbin > 40) {
1881  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::xtemp called with fxbin = " << fxbin << std::endl;
1882  }
1883 #else
1884  assert(fxbin >= 0 && fxbin < 41);
1885 #endif
1886 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1887  if(lxbin < 0 || lxbin > 40) {
1888  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::xtemp called with lxbin = " << lxbin << std::endl;
1889  }
1890 #else
1891  assert(lxbin >= 0 && lxbin < 41);
1892 #endif
1893 
1894 // Build the x-template, the central 25 bins are here in all cases
1895 
1896  for(i=0; i<9; ++i) {
1897  for(j=0; j<BXSIZE; ++j) {
1898  xtemplate[i+16][j]=xtemp_[i][j];
1899  }
1900  }
1901  for(i=0; i<8; ++i) {
1902  xtemplate[i+8][BXM1] = 0.f;
1903  for(j=0; j<BXM1; ++j) {
1904  xtemplate[i+8][j]=xtemp_[i][j+1];
1905  }
1906  }
1907  for(i=1; i<9; ++i) {
1908  xtemplate[i+24][0] = 0.f;
1909  for(j=0; j<BXM1; ++j) {
1910  xtemplate[i+24][j+1]=xtemp_[i][j];
1911  }
1912  }
1913 
1914 // Add more bins if needed
1915 
1916  if(fxbin < 8) {
1917  for(i=0; i<8; ++i) {
1918  xtemplate[i][BXM2] = 0.f;
1919  xtemplate[i][BXM1] = 0.f;
1920  for(j=0; j<BXM2; ++j) {
1921  xtemplate[i][j]=xtemp_[i][j+2];
1922  }
1923  }
1924  }
1925  if(lxbin > 32) {
1926  for(i=1; i<9; ++i) {
1927  xtemplate[i+32][0] = 0.f;
1928  xtemplate[i+32][1] = 0.f;
1929  for(j=0; j<BXM2; ++j) {
1930  xtemplate[i+32][j+2]=xtemp_[i][j];
1931  }
1932  }
1933  }
1934 
1935  return;
1936 
1937 } // End xtemp
int i
Definition: DBlmapReader.cc:9
#define BXSIZE
#define BXM1
assert(m_qm.get())
#define BXM2
int j
Definition: DBlmapReader.cc:9
float xtemp_[9][13+4]
templates for x-reconstruction (binned over 5 central pixels)
void SiPixelTemplate::xtemp3d ( int  i,
int  j,
std::vector< float > &  xtemplate 
)

Return interpolated 3d x-template in single call

Parameters
i,j- (input) template indices
xtemplate- (output) a boost 3d array containing two sets of temlate indices and the combined pixel signals

Definition at line 2223 of file SiPixelTemplate.cc.

References BXSIZE, i, j, and relval_steps::k.

Referenced by SiPixelTemplateSplit::PixelTempSplit().

2225 {
2226  // Sum two 2-d templates to make the 3-d template
2227  if(i >= 0 && i < nxbins_ && j <= i) {
2228  for(int k=0; k<BXSIZE; ++k) {
2229  xtemplate[k]=temp2dx_[i][k]+temp2dx_[j][k];
2230  }
2231  } else {
2232  for(int k=0; k<BXSIZE; ++k) {
2233  xtemplate[k]=0.;
2234  }
2235  }
2236 
2237  return;
2238 
2239 } // End xtemp3d
int i
Definition: DBlmapReader.cc:9
#define BXSIZE
boost::multi_array< float, 2 > temp2dx_
2d-primitive for spltting 3-d template
int j
Definition: DBlmapReader.cc:9
float nxbins_
number of bins in each dimension of the x-splitting template
void SiPixelTemplate::xtemp3d_int ( int  nxpix,
int &  nxbins 
)

Make interpolated 3d x-template (stored as class variables)

Parameters
nxpix- (input) number of pixels in cluster (needed to size template)
nxbins- (output) number of bins needed for each template projection

Definition at line 2144 of file SiPixelTemplate.cc.

References assert(), BXM1, BXM3, BXSIZE, diffTreeTool::diff, edm::hlt::Exception, i, j, and relval_steps::k.

Referenced by SiPixelTemplateSplit::PixelTempSplit().

2146 {
2147  // Retrieve already interpolated quantities
2148 
2149  // Local variables
2150  int i, j, k;
2151  int ioff0, ioffp, ioffm;
2152 
2153  // Verify that input parameters are in valid range
2154 
2155 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
2156  if(nxpix < 1 || nxpix >= BXM3) {
2157  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::xtemp3d called with nxpix = " << nxpix << std::endl;
2158  }
2159 #else
2160  assert(nxpix > 0 && nxpix < BXM3);
2161 #endif
2162 
2163 // Calculate the size of the shift in pixels needed to span the entire cluster
2164 
2165  float diff = fabsf(nxpix - clslenx_)/2.f + 1.f;
2166  int nshift = (int)diff;
2167  if((diff - nshift) > 0.5f) {++nshift;}
2168 
2169 // Calculate the number of bins needed to specify each hit range
2170 
2171  nxbins_ = 9 + 16*nshift;
2172 
2173 // Create a 2-d working template with the correct size
2174 
2175  temp2dx_.resize(boost::extents[nxbins_][BXSIZE]);
2176 
2177 // The 9 central bins are copied from the interpolated private store
2178 
2179  ioff0 = 8*nshift;
2180 
2181  for(i=0; i<9; ++i) {
2182  for(j=0; j<BXSIZE; ++j) {
2183  temp2dx_[i+ioff0][j]=xtemp_[i][j];
2184  }
2185  }
2186 
2187 // Add the +- shifted templates
2188 
2189  for(k=1; k<=nshift; ++k) {
2190  ioffm=ioff0-k*8;
2191  for(i=0; i<8; ++i) {
2192  for(j=0; j<k; ++j) {
2193  temp2dx_[i+ioffm][BXM1-j] = 0.f;
2194  }
2195  for(j=0; j<BXSIZE-k; ++j) {
2196  temp2dx_[i+ioffm][j]=xtemp_[i][j+k];
2197  }
2198  }
2199  ioffp=ioff0+k*8;
2200  for(i=1; i<9; ++i) {
2201  for(j=0; j<k; ++j) {
2202  temp2dx_[i+ioffp][j] = 0.f;
2203  }
2204  for(j=0; j<BXSIZE-k; ++j) {
2205  temp2dx_[i+ioffp][j+k]=xtemp_[i][j];
2206  }
2207  }
2208  }
2209 
2210  nxbins = nxbins_;
2211 
2212  return;
2213 
2214 } // End xtemp3d_int
int i
Definition: DBlmapReader.cc:9
#define BXSIZE
#define BXM3
#define BXM1
assert(m_qm.get())
boost::multi_array< float, 2 > temp2dx_
2d-primitive for spltting 3-d template
int j
Definition: DBlmapReader.cc:9
float clslenx_
x-cluster length of smaller interpolated template in pixels
float nxbins_
number of bins in each dimension of the x-splitting template
float xtemp_[9][13+4]
templates for x-reconstruction (binned over 5 central pixels)
float SiPixelTemplate::xxratio ( )
inline

fractional distance in x between cotalpha templates

Definition at line 346 of file SiPixelTemplate.h.

References xxratio_.

float SiPixelTemplate::yavg ( int  i)
inline

average y-bias of reconstruction binned in 4 charge bins

Definition at line 347 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and yavg_.

Referenced by SiPixelTemplateReco::PixelTempReco2D().

float SiPixelTemplate::yavgc2m ( int  i)
inline

1st pass chi2 min search: average y-bias of reconstruction binned in 4 charge bins

Definition at line 431 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and yavgc2m_.

Referenced by SiPixelTemplateSplit::PixelTempSplit().

float SiPixelTemplate::yflcorr ( int  binq,
float  qfly 
)

Return interpolated y-correction for input charge bin and qfly

Parameters
binq- (input) charge bin [0-3]
qfly- (input) (Q_f-Q_l)/(Q_f+Q_l) for this cluster

Definition at line 1687 of file SiPixelTemplate.cc.

References assert(), edm::hlt::Exception, and f.

Referenced by SiPixelTemplateReco::PixelTempReco2D().

1689 {
1690  // Interpolate using quantities already stored in the private variables
1691 
1692  // Local variables
1693  float qfl, qfl2, qfl3, qfl4, qfl5, dy;
1694 
1695  // Make sure that input is OK
1696 
1697 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1698  if(binq < 0 || binq > 3) {
1699  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::yflcorr called with binq = " << binq << std::endl;
1700  }
1701 #else
1702  assert(binq >= 0 && binq < 4);
1703 #endif
1704 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1705  if(fabs((double)qfly) > 1.) {
1706  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::yflcorr called with qfly = " << qfly << std::endl;
1707  }
1708 #else
1709  assert(fabs((double)qfly) <= 1.);
1710 #endif
1711 
1712 // Define the maximum signal to allow before de-weighting a pixel
1713 
1714  qfl = qfly;
1715 
1716  if(qfl < -0.9f) {qfl = -0.9f;}
1717  if(qfl > 0.9f) {qfl = 0.9f;}
1718 
1719 // Interpolate between the two polynomials
1720 
1721  qfl2 = qfl*qfl; qfl3 = qfl2*qfl; qfl4 = qfl3*qfl; qfl5 = qfl4*qfl;
1722  dy = (1.f-yratio_)*(yflparl_[binq][0]+yflparl_[binq][1]*qfl+yflparl_[binq][2]*qfl2+yflparl_[binq][3]*qfl3+yflparl_[binq][4]*qfl4+yflparl_[binq][5]*qfl5)
1723  + yratio_*(yflparh_[binq][0]+yflparh_[binq][1]*qfl+yflparh_[binq][2]*qfl2+yflparh_[binq][3]*qfl3+yflparh_[binq][4]*qfl4+yflparh_[binq][5]*qfl5);
1724 
1725  return dy;
1726 
1727 } // End yflcorr
assert(m_qm.get())
double f[11][100]
float yflparl_[4][6]
Aqfl-parameterized y-correction in 4 charge bins for smaller cotbeta.
float yflparh_[4][6]
Aqfl-parameterized y-correction in 4 charge bins for larger cotbeta.
float yratio_
fractional distance in y between cotbeta templates
float SiPixelTemplate::ygsig ( int  i)
inline

average sigma_y from Gaussian fit binned in 4 charge bins

Definition at line 368 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and ygsig_.

float SiPixelTemplate::ygx0 ( int  i)
inline

average y0 from Gaussian fit binned in 4 charge bins

Definition at line 361 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and ygx0_.

float SiPixelTemplate::yratio ( )
inline

fractional distance in y between cotbeta templates

Definition at line 344 of file SiPixelTemplate.h.

References yratio_.

float SiPixelTemplate::yrms ( int  i)
inline

average y-rms of reconstruction binned in 4 charge bins

Definition at line 354 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and yrms_.

Referenced by SiPixelTemplateReco::PixelTempReco2D().

float SiPixelTemplate::yrmsc2m ( int  i)
inline

1st pass chi2 min search: average y-rms of reconstruction binned in 4 charge bins

Definition at line 438 of file SiPixelTemplate.h.

References assert(), edm::hlt::Exception, i, and yrmsc2m_.

Referenced by SiPixelTemplateSplit::PixelTempSplit().

void SiPixelTemplate::ysigma2 ( int  fypix,
int  lypix,
float  sythr,
float  ysum[21+4],
float  ysig2[21+4] 
)
void SiPixelTemplate::ysigma2 ( float  qpixel,
int  index,
float &  ysig2 
)

Return y error (squared) for an input signal and yindex Add large Q scaling for use in cluster splitting.

Parameters
qpixel- (input) pixel charge
index- (input) y-index index of pixel
ysig2- (output) square error

Definition at line 1513 of file SiPixelTemplate.cc.

References assert(), BHY, BYM2, ENDL, edm::hlt::Exception, f, and LOGERROR.

1515 {
1516  // Interpolate using quantities already stored in the private variables
1517 
1518  // Local variables
1519  float sigi, sigi2, sigi3, sigi4, symax, qscale, err2;
1520 
1521  // Make sure that input is OK
1522 
1523 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1524  if(index < 2 || index >= BYM2) {
1525  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::ysigma2 called with index = " << index << std::endl;
1526  }
1527 #else
1528  assert(index > 1 && index < BYM2);
1529 #endif
1530 
1531  // Define the maximum signal to use in the parameterization
1532 
1533  symax = symax_;
1534  if(symax_ > syparmax_) {symax = syparmax_;}
1535 
1536  // Evaluate pixel-by-pixel uncertainties (weights) for the templ analysis
1537 
1538  if(qpixel < symax) {
1539  sigi = qpixel;
1540  qscale = 1.f;
1541  } else {
1542  sigi = symax;
1543  qscale = qpixel/symax;
1544  }
1545  sigi2 = sigi*sigi; sigi3 = sigi2*sigi; sigi4 = sigi3*sigi;
1546  if(index <= BHY) {
1547  err2 = (1.f-yratio_)*
1548  (yparl_[0][0]+yparl_[0][1]*sigi+yparl_[0][2]*sigi2+yparl_[0][3]*sigi3+yparl_[0][4]*sigi4)
1549  + yratio_*
1550  (yparh_[0][0]+yparh_[0][1]*sigi+yparh_[0][2]*sigi2+yparh_[0][3]*sigi3+yparh_[0][4]*sigi4);
1551  } else {
1552  err2 = (1.f-yratio_)*
1553  (yparl_[1][0]+yparl_[1][1]*sigi+yparl_[1][2]*sigi2+yparl_[1][3]*sigi3+yparl_[1][4]*sigi4)
1554  + yratio_*
1555  (yparh_[1][0]+yparh_[1][1]*sigi+yparh_[1][2]*sigi2+yparh_[1][3]*sigi3+yparh_[1][4]*sigi4);
1556  }
1557  ysig2 =qscale*err2;
1558  if(ysig2 <= 0.f) {LOGERROR("SiPixelTemplate") << "neg y-error-squared, id = " << id_current_ << ", index = " << index_id_ <<
1559  ", cot(alpha) = " << cota_current_ << ", cot(beta) = " << cotb_current_ << ", sigi = " << sigi << ENDL;}
1560 
1561  return;
1562 
1563 } // End ysigma2
float cota_current_
current cot alpha
#define LOGERROR(x)
float symax()
average pixel signal for y-projection of cluster
assert(m_qm.get())
float cotb_current_
current cot beta
float qscale()
charge scaling factor
#define ENDL
#define BYM2
double f[11][100]
float yparl_[2][5]
projected y-pixel uncertainty parameterization for smaller cotbeta
#define BHY
float symax_
average pixel signal for y-projection of cluster
float yparh_[2][5]
projected y-pixel uncertainty parameterization for larger cotbeta
int index_id_
current index
float syparmax_
maximum pixel signal for parameterization of y uncertainties
int id_current_
current id
float yratio_
fractional distance in y between cotbeta templates
float SiPixelTemplate::ysize ( )
inline
void SiPixelTemplate::ytemp ( int  fybin,
int  lybin,
float  ytemplate[41][21+4] 
)

Return interpolated y-template in single call

Parameters
fybin- (input) index of first bin (0-40) to fill
fybin- (input) index of last bin (0-40) to fill
ytemplate- (output) a 41x25 output buffer

Definition at line 1791 of file SiPixelTemplate.cc.

References assert(), BYM1, BYM2, BYSIZE, edm::hlt::Exception, i, and j.

Referenced by SiPixelTemplateReco::PixelTempReco2D(), and SiPixelGaussianSmearingRecHitConverterAlgorithm::smearHit().

1793 {
1794  // Retrieve already interpolated quantities
1795 
1796  // Local variables
1797  int i, j;
1798 
1799  // Verify that input parameters are in valid range
1800 
1801 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1802  if(fybin < 0 || fybin > 40) {
1803  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::ytemp called with fybin = " << fybin << std::endl;
1804  }
1805 #else
1806  assert(fybin >= 0 && fybin < 41);
1807 #endif
1808 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
1809  if(lybin < 0 || lybin > 40) {
1810  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::ytemp called with lybin = " << lybin << std::endl;
1811  }
1812 #else
1813  assert(lybin >= 0 && lybin < 41);
1814 #endif
1815 
1816 // Build the y-template, the central 25 bins are here in all cases
1817 
1818  for(i=0; i<9; ++i) {
1819  for(j=0; j<BYSIZE; ++j) {
1820  ytemplate[i+16][j]=ytemp_[i][j];
1821  }
1822  }
1823  for(i=0; i<8; ++i) {
1824  ytemplate[i+8][BYM1] = 0.f;
1825  for(j=0; j<BYM1; ++j) {
1826  ytemplate[i+8][j]=ytemp_[i][j+1];
1827  }
1828  }
1829  for(i=1; i<9; ++i) {
1830  ytemplate[i+24][0] = 0.f;
1831  for(j=0; j<BYM1; ++j) {
1832  ytemplate[i+24][j+1]=ytemp_[i][j];
1833  }
1834  }
1835 
1836 // Add more bins if needed
1837 
1838  if(fybin < 8) {
1839  for(i=0; i<8; ++i) {
1840  ytemplate[i][BYM2] = 0.f;
1841  ytemplate[i][BYM1] = 0.f;
1842  for(j=0; j<BYM2; ++j) {
1843  ytemplate[i][j]=ytemp_[i][j+2];
1844  }
1845  }
1846  }
1847  if(lybin > 32) {
1848  for(i=1; i<9; ++i) {
1849  ytemplate[i+32][0] = 0.f;
1850  ytemplate[i+32][1] = 0.f;
1851  for(j=0; j<BYM2; ++j) {
1852  ytemplate[i+32][j+2]=ytemp_[i][j];
1853  }
1854  }
1855  }
1856 
1857  return;
1858 
1859 } // End ytemp
int i
Definition: DBlmapReader.cc:9
#define BYSIZE
assert(m_qm.get())
float ytemp_[9][21+4]
templates for y-reconstruction (binned over 5 central pixels)
#define BYM1
int j
Definition: DBlmapReader.cc:9
#define BYM2
void SiPixelTemplate::ytemp3d ( int  i,
int  j,
std::vector< float > &  ytemplate 
)

Return interpolated 3d y-template in single call

Parameters
i,j- (input) template indices
ytemplate- (output) a boost 3d array containing two sets of temlate indices and the combined pixel signals

Definition at line 2120 of file SiPixelTemplate.cc.

References BYSIZE, i, j, and relval_steps::k.

Referenced by SiPixelTemplateSplit::PixelTempSplit().

2122 {
2123  // Sum two 2-d templates to make the 3-d template
2124  if(i >= 0 && i < nybins_ && j <= i) {
2125  for(int k=0; k<BYSIZE; ++k) {
2126  ytemplate[k]=temp2dy_[i][k]+temp2dy_[j][k];
2127  }
2128  } else {
2129  for(int k=0; k<BYSIZE; ++k) {
2130  ytemplate[k]=0.;
2131  }
2132  }
2133 
2134  return;
2135 
2136 } // End ytemp3d
int i
Definition: DBlmapReader.cc:9
#define BYSIZE
int j
Definition: DBlmapReader.cc:9
boost::multi_array< float, 2 > temp2dy_
2d-primitive for spltting 3-d template
float nybins_
number of bins in each dimension of the y-splitting template
void SiPixelTemplate::ytemp3d_int ( int  nypix,
int &  nybins 
)

Make interpolated 3d y-template (stored as class variables)

Parameters
nypix- (input) number of pixels in cluster (needed to size template)
nybins- (output) number of bins needed for each template projection

Definition at line 2042 of file SiPixelTemplate.cc.

References assert(), BYM1, BYM3, BYSIZE, diffTreeTool::diff, edm::hlt::Exception, i, j, and relval_steps::k.

Referenced by SiPixelTemplateSplit::PixelTempSplit().

2044 {
2045 
2046  // Retrieve already interpolated quantities
2047 
2048  // Local variables
2049  int i, j, k;
2050  int ioff0, ioffp, ioffm;
2051 
2052  // Verify that input parameters are in valid range
2053 
2054 #ifndef SI_PIXEL_TEMPLATE_STANDALONE
2055  if(nypix < 1 || nypix >= BYM3) {
2056  throw cms::Exception("DataCorrupt") << "SiPixelTemplate::ytemp3d called with nypix = " << nypix << std::endl;
2057  }
2058 #else
2059  assert(nypix > 0 && nypix < BYM3);
2060 #endif
2061 
2062  // Calculate the size of the shift in pixels needed to span the entire cluster
2063 
2064  float diff = fabsf(nypix - clsleny_)/2. + 1.f;
2065  int nshift = (int)diff;
2066  if((diff - nshift) > 0.5f) {++nshift;}
2067 
2068  // Calculate the number of bins needed to specify each hit range
2069 
2070  nybins_ = 9 + 16*nshift;
2071 
2072  // Create a 2-d working template with the correct size
2073 
2074  temp2dy_.resize(boost::extents[nybins_][BYSIZE]);
2075 
2076  // The 9 central bins are copied from the interpolated private store
2077 
2078  ioff0 = 8*nshift;
2079 
2080  for(i=0; i<9; ++i) {
2081  for(j=0; j<BYSIZE; ++j) {
2082  temp2dy_[i+ioff0][j]=ytemp_[i][j];
2083  }
2084  }
2085 
2086  // Add the +- shifted templates
2087 
2088  for(k=1; k<=nshift; ++k) {
2089  ioffm=ioff0-k*8;
2090  for(i=0; i<8; ++i) {
2091  for(j=0; j<k; ++j) {
2092  temp2dy_[i+ioffm][BYM1-j] = 0.f;
2093  }
2094  for(j=0; j<BYSIZE-k; ++j) {
2095  temp2dy_[i+ioffm][j]=ytemp_[i][j+k];
2096  }
2097  }
2098  ioffp=ioff0+k*8;
2099  for(i=1; i<9; ++i) {
2100  for(j=0; j<k; ++j) {
2101  temp2dy_[i+ioffp][j] = 0.f;
2102  }
2103  for(j=0; j<BYSIZE-k; ++j) {
2104  temp2dy_[i+ioffp][j+k]=ytemp_[i][j];
2105  }
2106  }
2107  }
2108 
2109  nybins = nybins_;
2110  return;
2111 
2112 } // End ytemp3d_int
int i
Definition: DBlmapReader.cc:9
float clsleny_
y-cluster length of smaller interpolated template in pixels
#define BYSIZE
assert(m_qm.get())
float ytemp_[9][21+4]
templates for y-reconstruction (binned over 5 central pixels)
#define BYM1
int j
Definition: DBlmapReader.cc:9
#define BYM3
boost::multi_array< float, 2 > temp2dy_
2d-primitive for spltting 3-d template
float nybins_
number of bins in each dimension of the y-splitting template
float SiPixelTemplate::yxratio ( )
inline

fractional distance in y between cotalpha templates slices

Definition at line 345 of file SiPixelTemplate.h.

References yxratio_.

float SiPixelTemplate::zsize ( )
inline

pixel z-size or thickness (microns)

Definition at line 513 of file SiPixelTemplate.h.

References zsize_.

Member Data Documentation

float SiPixelTemplate::abs_cotb_
private

absolute value of cot beta

Definition at line 527 of file SiPixelTemplate.h.

float SiPixelTemplate::chi2xavg_[4]
private

average x chi^2 in 4 charge bins

Definition at line 581 of file SiPixelTemplate.h.

Referenced by chi2xavg().

float SiPixelTemplate::chi2xavgc2m_[4]
private

1st pass chi2 min search: average x-chisq for merged clusters

Definition at line 589 of file SiPixelTemplate.h.

Referenced by chi2xavgc2m().

float SiPixelTemplate::chi2xavgone_
private

average x chi^2 for 1 pixel clusters

Definition at line 593 of file SiPixelTemplate.h.

Referenced by chi2xavgone().

float SiPixelTemplate::chi2xmin_[4]
private

minimum of x chi^2 in 4 charge bins

Definition at line 582 of file SiPixelTemplate.h.

Referenced by chi2xmin().

float SiPixelTemplate::chi2xminc2m_[4]
private

1st pass chi2 min search: minimum x-chisq for merged clusters

Definition at line 590 of file SiPixelTemplate.h.

Referenced by chi2xminc2m().

float SiPixelTemplate::chi2xminone_
private

minimum of x chi^2 for 1 pixel clusters

Definition at line 594 of file SiPixelTemplate.h.

Referenced by chi2xminone().

float SiPixelTemplate::chi2yavg_[4]
private

average y chi^2 in 4 charge bins

Definition at line 579 of file SiPixelTemplate.h.

Referenced by chi2yavg().

float SiPixelTemplate::chi2yavgc2m_[4]
private

1st pass chi2 min search: average y-chisq for merged clusters

Definition at line 585 of file SiPixelTemplate.h.

Referenced by chi2yavgc2m().

float SiPixelTemplate::chi2yavgone_
private

average y chi^2 for 1 pixel clusters

Definition at line 591 of file SiPixelTemplate.h.

Referenced by chi2yavgone().

float SiPixelTemplate::chi2ymin_[4]
private

minimum of y chi^2 in 4 charge bins

Definition at line 580 of file SiPixelTemplate.h.

Referenced by chi2ymin().

float SiPixelTemplate::chi2yminc2m_[4]
private

1st pass chi2 min search: minimum y-chisq for merged clusters

Definition at line 586 of file SiPixelTemplate.h.

Referenced by chi2yminc2m().

float SiPixelTemplate::chi2yminone_
private

minimum of y chi^2 for 1 pixel clusters

Definition at line 592 of file SiPixelTemplate.h.

Referenced by chi2yminone().

float SiPixelTemplate::clslenx_
private

x-cluster length of smaller interpolated template in pixels

Definition at line 552 of file SiPixelTemplate.h.

Referenced by clslenx().

float SiPixelTemplate::clsleny_
private

y-cluster length of smaller interpolated template in pixels

Definition at line 551 of file SiPixelTemplate.h.

Referenced by clsleny().

float SiPixelTemplate::cota_current_
private

current cot alpha

Definition at line 525 of file SiPixelTemplate.h.

Referenced by SiPixelTemplate().

float SiPixelTemplate::cotb_current_
private

current cot beta

Definition at line 526 of file SiPixelTemplate.h.

Referenced by SiPixelTemplate().

float SiPixelTemplate::dxone_
private

mean offset/correction for one pixel x-clusters

Definition at line 546 of file SiPixelTemplate.h.

Referenced by dxone().

float SiPixelTemplate::dxtwo_
private

mean offset/correction for one double-pixel x-clusters

Definition at line 548 of file SiPixelTemplate.h.

Referenced by dxtwo().

float SiPixelTemplate::dyone_
private

mean offset/correction for one pixel y-clusters

Definition at line 540 of file SiPixelTemplate.h.

Referenced by dyone().

float SiPixelTemplate::dytwo_
private

mean offset/correction for one double-pixel y-clusters

Definition at line 542 of file SiPixelTemplate.h.

Referenced by dytwo().

float SiPixelTemplate::fbin_[3]
private

The QBin definitions in Q_clus/Q_avg.

Definition at line 613 of file SiPixelTemplate.h.

Referenced by fbin().

int SiPixelTemplate::id_current_
private

current id

Definition at line 523 of file SiPixelTemplate.h.

Referenced by SiPixelTemplate().

int SiPixelTemplate::index_id_
private

current index

Definition at line 524 of file SiPixelTemplate.h.

Referenced by SiPixelTemplate().

float SiPixelTemplate::kappavav2_
private

kappa parameter for 2-cluster Vavilov distribution

Definition at line 601 of file SiPixelTemplate.h.

Referenced by kappavav2().

float SiPixelTemplate::kappavav_
private

kappa parameter for Vavilov distribution

Definition at line 598 of file SiPixelTemplate.h.

Referenced by kappavav().

float SiPixelTemplate::lorxbias_
private

Lorentz x-bias.

Definition at line 605 of file SiPixelTemplate.h.

Referenced by lorxbias().

float SiPixelTemplate::lorxwidth_
private

Lorentz x-width.

Definition at line 603 of file SiPixelTemplate.h.

Referenced by lorxwidth().

float SiPixelTemplate::lorybias_
private

Lorentz y-bias.

Definition at line 604 of file SiPixelTemplate.h.

Referenced by lorybias().

float SiPixelTemplate::lorywidth_
private

Lorentz y-width (sign corrected for fpix frame)

Definition at line 602 of file SiPixelTemplate.h.

Referenced by lorywidth().

float SiPixelTemplate::mpvvav2_
private

most probable charge in 2-cluster Vavilov distribution (not actually for larger kappa)

Definition at line 599 of file SiPixelTemplate.h.

Referenced by mpvvav2().

float SiPixelTemplate::mpvvav_
private

most probable charge in Vavilov distribution (not actually for larger kappa)

Definition at line 596 of file SiPixelTemplate.h.

Referenced by mpvvav().

float SiPixelTemplate::nxbins_
private

number of bins in each dimension of the x-splitting template

Definition at line 611 of file SiPixelTemplate.h.

float SiPixelTemplate::nybins_
private

number of bins in each dimension of the y-splitting template

Definition at line 610 of file SiPixelTemplate.h.

float SiPixelTemplate::pixmax_
private

maximum pixel charge

Definition at line 534 of file SiPixelTemplate.h.

Referenced by pixmax().

float SiPixelTemplate::qavg_
private

average cluster charge for this set of track angles

Definition at line 533 of file SiPixelTemplate.h.

Referenced by qavg().

float SiPixelTemplate::qavg_avg_
private

average of cluster charge less than qavg

Definition at line 609 of file SiPixelTemplate.h.

float SiPixelTemplate::qmin2_
private

tighter minimum cluster charge for valid hit (keeps 99.8% of simulated hits)

Definition at line 595 of file SiPixelTemplate.h.

Referenced by qmin().

float SiPixelTemplate::qmin_
private

minimum cluster charge for valid hit (keeps 99.9% of simulated hits)

Definition at line 550 of file SiPixelTemplate.h.

Referenced by qmin().

float SiPixelTemplate::qscale_
private

charge scaling factor

Definition at line 535 of file SiPixelTemplate.h.

Referenced by qscale().

float SiPixelTemplate::r_qMeas_qTrue_
private

ratio of measured to true cluster charges

Definition at line 612 of file SiPixelTemplate.h.

Referenced by r_qMeas_qTrue().

float SiPixelTemplate::s50_
private

1/2 of the pixel threshold signal in adc units

Definition at line 536 of file SiPixelTemplate.h.

Referenced by s50().

float SiPixelTemplate::sigmavav2_
private

"sigma" scale fctor for 2-cluster Vavilov distribution

Definition at line 600 of file SiPixelTemplate.h.

Referenced by sigmavav2().

float SiPixelTemplate::sigmavav_
private

"sigma" scale fctor for Vavilov distribution

Definition at line 597 of file SiPixelTemplate.h.

Referenced by sigmavav().

float SiPixelTemplate::ss50_
private

1/2 of the pixel threshold signal in adc units

Definition at line 537 of file SiPixelTemplate.h.

Referenced by ss50().

bool SiPixelTemplate::success_
private

true if cotalpha, cotbeta are inside of the acceptance (dynamically loaded)

Definition at line 528 of file SiPixelTemplate.h.

float SiPixelTemplate::sxmax_
private

average pixel signal for x-projection of cluster

Definition at line 544 of file SiPixelTemplate.h.

Referenced by sxmax().

float SiPixelTemplate::sxone_
private

rms for one pixel x-clusters

Definition at line 547 of file SiPixelTemplate.h.

Referenced by sxone().

float SiPixelTemplate::sxparmax_
private

maximum pixel signal for parameterization of x uncertainties

Definition at line 545 of file SiPixelTemplate.h.

float SiPixelTemplate::sxtwo_
private

rms for one double-pixel x-clusters

Definition at line 549 of file SiPixelTemplate.h.

Referenced by sxtwo().

float SiPixelTemplate::symax_
private

average pixel signal for y-projection of cluster

Definition at line 538 of file SiPixelTemplate.h.

Referenced by symax().

float SiPixelTemplate::syone_
private

rms for one pixel y-clusters

Definition at line 541 of file SiPixelTemplate.h.

Referenced by syone().

float SiPixelTemplate::syparmax_
private

maximum pixel signal for parameterization of y uncertainties

Definition at line 539 of file SiPixelTemplate.h.

float SiPixelTemplate::sytwo_
private

rms for one double-pixel y-clusters

Definition at line 543 of file SiPixelTemplate.h.

Referenced by sytwo().

boost::multi_array<float,2> SiPixelTemplate::temp2dx_
private

2d-primitive for spltting 3-d template

Definition at line 615 of file SiPixelTemplate.h.

boost::multi_array<float,2> SiPixelTemplate::temp2dy_
private

2d-primitive for spltting 3-d template

Definition at line 614 of file SiPixelTemplate.h.

const std::vector< SiPixelTemplateStore >& SiPixelTemplate::thePixelTemp_
private

Definition at line 620 of file SiPixelTemplate.h.

float SiPixelTemplate::xavg_[4]
private

average x-bias of reconstruction binned in 4 charge bins

Definition at line 571 of file SiPixelTemplate.h.

Referenced by xavg().

float SiPixelTemplate::xavgc2m_[4]
private

1st pass chi2 min search: average x-bias of reconstruction binned in 4 charge bins

Definition at line 587 of file SiPixelTemplate.h.

Referenced by xavgc2m().

float SiPixelTemplate::xflparhh_[4][6]
private

Aqfl-parameterized x-correction in 4 charge bins for larger cotbeta, cotalpha.

Definition at line 578 of file SiPixelTemplate.h.

float SiPixelTemplate::xflparhl_[4][6]
private

Aqfl-parameterized x-correction in 4 charge bins for larger cotbeta, smaller cotalpha.

Definition at line 577 of file SiPixelTemplate.h.

float SiPixelTemplate::xflparlh_[4][6]
private

Aqfl-parameterized x-correction in 4 charge bins for smaller cotbeta, larger cotalpha.

Definition at line 576 of file SiPixelTemplate.h.

float SiPixelTemplate::xflparll_[4][6]
private

Aqfl-parameterized x-correction in 4 charge bins for smaller cotbeta, cotalpha.

Definition at line 575 of file SiPixelTemplate.h.

float SiPixelTemplate::xgsig_[4]
private

sigma from Gaussian fit binned in 4 charge bins

Definition at line 574 of file SiPixelTemplate.h.

Referenced by xgsig().

float SiPixelTemplate::xgx0_[4]
private

average x0 from Gaussian fit binned in 4 charge bins

Definition at line 573 of file SiPixelTemplate.h.

Referenced by xgx0().

float SiPixelTemplate::xpar0_[2][5]
private

projected x-pixel uncertainty parameterization for central cotalpha

Definition at line 561 of file SiPixelTemplate.h.

float SiPixelTemplate::xparh_[2][5]
private

projected x-pixel uncertainty parameterization for larger cotalpha

Definition at line 563 of file SiPixelTemplate.h.

float SiPixelTemplate::xparhy0_[2][5]
private

projected x-pixel uncertainty parameterization for larger cotbeta (central alpha)

Definition at line 557 of file SiPixelTemplate.h.

float SiPixelTemplate::xparl_[2][5]
private

projected x-pixel uncertainty parameterization for smaller cotalpha

Definition at line 562 of file SiPixelTemplate.h.

float SiPixelTemplate::xparly0_[2][5]
private

projected x-pixel uncertainty parameterization for smaller cotbeta (central alpha)

Definition at line 556 of file SiPixelTemplate.h.

float SiPixelTemplate::xrms_[4]
private

average x-rms of reconstruction binned in 4 charge bins

Definition at line 572 of file SiPixelTemplate.h.

Referenced by xrms().

float SiPixelTemplate::xrmsc2m_[4]
private

1st pass chi2 min search: average x-rms of reconstruction binned in 4 charge bins

Definition at line 588 of file SiPixelTemplate.h.

Referenced by xrmsc2m().

float SiPixelTemplate::xsize_
private

Pixel x-size.

Definition at line 606 of file SiPixelTemplate.h.

Referenced by xsize().

float SiPixelTemplate::xtemp_[9][13+4]
private

templates for x-reconstruction (binned over 5 central pixels)

Definition at line 564 of file SiPixelTemplate.h.

float SiPixelTemplate::xxratio_
private

fractional distance in x between cotalpha templates

Definition at line 560 of file SiPixelTemplate.h.

Referenced by xxratio().

float SiPixelTemplate::yavg_[4]
private

average y-bias of reconstruction binned in 4 charge bins

Definition at line 565 of file SiPixelTemplate.h.

Referenced by yavg().

float SiPixelTemplate::yavgc2m_[4]
private

1st pass chi2 min search: average y-bias of reconstruction binned in 4 charge bins

Definition at line 583 of file SiPixelTemplate.h.

Referenced by yavgc2m().

float SiPixelTemplate::yflparh_[4][6]
private

Aqfl-parameterized y-correction in 4 charge bins for larger cotbeta.

Definition at line 570 of file SiPixelTemplate.h.

float SiPixelTemplate::yflparl_[4][6]
private

Aqfl-parameterized y-correction in 4 charge bins for smaller cotbeta.

Definition at line 569 of file SiPixelTemplate.h.

float SiPixelTemplate::ygsig_[4]
private

average sigma_y from Gaussian fit binned in 4 charge bins

Definition at line 568 of file SiPixelTemplate.h.

Referenced by ygsig().

float SiPixelTemplate::ygx0_[4]
private

average y0 from Gaussian fit binned in 4 charge bins

Definition at line 567 of file SiPixelTemplate.h.

Referenced by ygx0().

float SiPixelTemplate::yparh_[2][5]
private

projected y-pixel uncertainty parameterization for larger cotbeta

Definition at line 555 of file SiPixelTemplate.h.

float SiPixelTemplate::yparl_[2][5]
private

projected y-pixel uncertainty parameterization for smaller cotbeta

Definition at line 554 of file SiPixelTemplate.h.

float SiPixelTemplate::yratio_
private

fractional distance in y between cotbeta templates

Definition at line 553 of file SiPixelTemplate.h.

Referenced by yratio().

float SiPixelTemplate::yrms_[4]
private

average y-rms of reconstruction binned in 4 charge bins

Definition at line 566 of file SiPixelTemplate.h.

Referenced by yrms().

float SiPixelTemplate::yrmsc2m_[4]
private

1st pass chi2 min search: average y-rms of reconstruction binned in 4 charge bins

Definition at line 584 of file SiPixelTemplate.h.

Referenced by yrmsc2m().

float SiPixelTemplate::ysize_
private

Pixel y-size.

Definition at line 607 of file SiPixelTemplate.h.

Referenced by ysize().

float SiPixelTemplate::ytemp_[9][21+4]
private

templates for y-reconstruction (binned over 5 central pixels)

Definition at line 558 of file SiPixelTemplate.h.

float SiPixelTemplate::yxratio_
private

fractional distance in y between x-slices of cotalpha templates

Definition at line 559 of file SiPixelTemplate.h.

Referenced by yxratio().

float SiPixelTemplate::zsize_
private

Pixel z-size (thickness)

Definition at line 608 of file SiPixelTemplate.h.

Referenced by zsize().