standalone/plotting/makeBenchmarkPlots.py

0001 import os.path, glob, sys
0002 import ROOT
0003 import array
0004 import math
0005
0006 # N.B.: Consult ./xeon_scripts/benchmark-cmssw-ttbar-fulldet-build.sh for info on nTHs, nVUs, and text file names
0007
0008 def run():
0009     # command line input
0010     arch   = sys.argv[1] # SNB, KNL, SKL-SP
0011     sample = sys.argv[2]
0012     build  = sys.argv[3] # BH, STD, CE, FV
0013     isVU   = sys.argv[4] # 'true' or 'false': if no argument passed, will not do VU plots
0014     isTH   = sys.argv[5] # 'true' or 'false': if no argument passed, will not do TH plots
0015
0016     # reopen file for writing
0017     g = ROOT.TFile('benchmark_'+arch+'_'+sample+'.root','update')
0018
0019     # Vectorization data points
0020     vuvals = ['1','2','4','8']
0021     nth = '1'
0022
0023     if  arch == 'KNL' or arch == 'SKL-SP' or arch == 'LNX-G' or arch == 'LNX-S':
0024         vuvals.append('16')
0025         vuvals.append('16int')
0026     elif arch == 'SNB' :
0027         vuvals.append('8int')
0028     else :
0029         print arch,'is not a valid architecture! Exiting...'
0030         sys.exit(0)
0031
0032     # call the make plots function
0033     if isVU == 'true' :
0034         makeplots(arch,sample,build,vuvals,nth,'VU')
0035
0036     # Parallelization datapoints
0037     if arch == 'KNL' :
0038         nvu = '16int'
0039         thvals = ['1','2','4','8','16','32','64','96','128','160','192','224','256']
0040     elif arch == 'SNB' :
0041         nvu = '8int'
0042         thvals = ['1','2','4','6','8','12','16','20','24']
0043     elif arch == 'SKL-SP' :
0044         nvu = '16int'
0045         thvals = ['1','2','4','8','16','32','48','64']
0046     elif arch == 'LNX-G' :
0047         nvu = '16int'
0048         thvals = ['1','2','4','8','16','32','48','64']
0049     elif arch == 'LNX-S' :
0050         nvu = '16int'
0051         thvals = ['1','2','4','8','16','32','48','64']
0052     else :
0053         print arch,'is not a valid architecture! Exiting...'
0054         sys.exit(0)
0055
0056     # call the make plots function
0057     if isTH == 'true' :
0058         makeplots(arch,sample,build,thvals,nvu,'TH')
0059
0060     g.Write()
0061     g.Close()
0062
0063 def makeplots(arch,sample,build,vals,nC,text):
0064     # position in logs
0065     if   build == 'BH'  : pos = 8
0066     elif build == 'STD' : pos = 11
0067     elif build == 'CE'  : pos = 14
0068     elif build == 'FV'  : pos = 17
0069     else :
0070         print build,'is not a valid test! Exiting...'
0071         sys.exit(0)
0072
0073     # time
0074     print arch,sample,build,text
0075
0076     # define tgraphs vs absolute time and speedup
0077     g_time    = ROOT.TGraphErrors(len(vals)-1)
0078     g_speedup = ROOT.TGraphErrors(len(vals)-1)
0079
0080     # make separate plot for intrinsics measurement
0081     if text is 'VU' :
0082         g_time_int    = ROOT.TGraphErrors(1)
0083         g_speedup_int = ROOT.TGraphErrors(1)
0084
0085     point = 0
0086     for val in vals :
0087         if    val is '16int': xval = 16.0
0088         elif  val is '8int' : xval = 8.0
0089         else                : xval = float(val)
0090
0091         # array of time values
0092         yvals = array.array('d');
0093
0094         # always skip the first event
0095         firstFound = False
0096
0097         # open the correct log file, store times into temp file
0098         if   text is 'VU' : os.system('grep Matriplex log_'+arch+'_'+sample+'_'+build+'_NVU'+val+'_NTH'+nC +'.txt >& log_'+arch+'_'+sample+'_'+build+'_'+text+'.txt')
0099         elif text is 'TH' : os.system('grep Matriplex log_'+arch+'_'+sample+'_'+build+'_NVU'+nC +'_NTH'+val+'.txt >& log_'+arch+'_'+sample+'_'+build+'_'+text+'.txt')
0100         else :
0101             print 'VU or TH are the only options for extra text! Exiting...'
0102             exit
0103
0104         # open temp file, store event times into yvals
0105         with open('log_'+arch+'_'+sample+'_'+build+'_'+text+'.txt') as f :
0106             for line in f :
0107                 if 'Matriplex' not in line : continue
0108                 if 'Total' in line : continue
0109                 if not firstFound :
0110                     firstFound = True
0111                     continue
0112                 lsplit = line.split()
0113                 yvals.append(float(lsplit[pos]))
0114
0115         # Compute mean and uncertainty on mean from yvals
0116         sum = 0.;
0117         for yval in range(0,len(yvals)):
0118             sum = sum + yvals[yval]
0119         if len(yvals) > 0 :
0120             mean = sum/len(yvals)
0121         else :
0122             mean = 0
0123         emean = 0.;
0124         for yval in range(0,len(yvals)):
0125             emean = emean + ((yvals[yval] - mean) * (yvals[yval] - mean))
0126         if len(yvals) > 1 :
0127             emean = math.sqrt(emean / (len(yvals) - 1))
0128             emean = emean/math.sqrt(len(yvals))
0129         else :
0130             emean = 0
0131
0132         # Printout value for good measure
0133         print val,mean,'+/-',emean
0134
0135         # store intrinsics val into separate plot
0136         if 'int' not in val :
0137             g_time.SetPoint(point,xval,mean)
0138             g_time.SetPointError(point,0,emean)
0139             point = point+1
0140         else :
0141             g_time_int.SetPoint(0,xval,mean)
0142             g_time_int.SetPointError(0,0,emean)
0143
0144     # always write out the standard plot
0145     g_time.Write('g_'+build+'_'+text+'_time')
0146
0147     # write out separate intrinsics plot
0148     if text is 'VU' :
0149         g_time_int.Write('g_'+build+'_'+text+'_time_int')
0150
0151     # Speedup calculation
0152     xval0 = array.array('d',[0])
0153     yval0 = array.array('d',[0])
0154     yerr0 = array.array('d',[0])
0155
0156     # Get first point to divide by
0157     g_time.GetPoint(0,xval0,yval0)
0158     yerr0.append(g_time.GetErrorY(0))
0159
0160     point = 0
0161     for val in vals :
0162         # set up inputs
0163         xval = array.array('d',[0])
0164         yval = array.array('d',[0])
0165         yerr = array.array('d',[0])
0166
0167         # get standard plots from standard plot
0168         if 'int' not in val :
0169             g_time.GetPoint(point,xval,yval)
0170             yerr.append(g_time.GetErrorY(point))
0171         else :
0172             g_time_int.GetPoint(0,xval,yval)
0173             yerr.append(g_time_int.GetErrorY(0))
0174
0175         speedup  = 0.
0176         espeedup = 0.
0177         if yval[0] > 0. and yval0[0] > 0. :
0178             speedup  = yval0[0]/yval[0]
0179             espeedup = speedup * math.sqrt(math.pow(yerr0[0]/yval0[0],2) + math.pow(yerr[0]/yval[0],2))
0180
0181         # store in the correct plot
0182         if 'int' not in val :
0183             g_speedup.SetPoint(point,xval[0],speedup)
0184             g_speedup.SetPointError(point,0,espeedup)
0185             point = point+1
0186         else :
0187             g_speedup_int.SetPoint(0,xval[0],speedup)
0188             g_speedup_int.SetPointError(0,0,espeedup)
0189
0190     # always write out the standard plot
0191     g_speedup.Write('g_'+build+'_'+text+'_speedup')
0192
0193     # write out separate intrinsics plot
0194     if text is 'VU' :
0195         g_speedup_int.Write('g_'+build+'_'+text+'_speedup_int')
0196
0197     # all done
0198     return
0199
0200 if __name__ == "__main__":
0201     run()