xref: /petsc/src/benchmarks/streams/process.py (revision 5b6bfdb9644f185dbf5e5a09b808ec241507e1e7)

#!/usr/bin/env python
#!/bin/env python
#
#    Computers speed up of Streams benchmark results generated by make streams and plots
#
#    matplotlib can switch between different backends hence this needs to be run
#    twice to first generate a file and then display a window
#
from __future__ import print_function
import os
#
def process(fileoutput = 1):
  import re
  ff = open('scaling.log')
  data = ff.read()
  ff.close()

  s = data.split('\n')
  ss = []
  for i in s:
    if not i.startswith('MPI rank'):
      ss.append(i)
  data = '\n'.join(ss)
  hosts  = {}
  triads = {}
  speedups = {}
  match = data.split('Number of MPI processes ')
  for i in match:
    if i:
      fields = i.split('\n')
      size = int(fields[0].split()[0])
      hosts[size] = fields[0].split()[3:]
      triads[size] = float(fields[1].split()[1])

  if len(hosts) < 2: return

  ff = open('scaling.log','a')
  if fileoutput: print('np  speedup')
  if fileoutput: ff.write('np  speedup\n')
  for sizes in hosts:
    speedups[sizes] = triads[sizes]/triads[1]
    if fileoutput: print(sizes,round(triads[sizes]/triads[1],2))
    if fileoutput: ff.write(str(sizes)+' '+str(round(triads[sizes]/triads[1],2))+'\n')

  if fileoutput: print("Estimation of possible speedup of MPI programs based on Streams benchmark.")
  if fileoutput: ff.write("Estimation of possible speedup of MPI programs based on Streams benchmark.\n")

  if fileoutput:
    import re
    last = max(hosts.keys())
    lasthosts = hosts[last]
    for i in range(0,len(lasthosts)):
      lasthosts[i] = re.sub(r"Process [0-9]*", "", lasthosts[i])
    ulasthosts = list(set(lasthosts))
    print("It appears you have "+str(len(ulasthosts))+" node(s)")
    ff.write("It appears you have "+str(len(ulasthosts))+" node(s)\n")

    if len(ulasthosts) < 1:
      testhosts = []
      for i in range(0,len(lasthosts)):
        testhosts.append(ulasthosts[i % len(ulasthosts)])
      if testhosts == lasthosts:
        print("   distributed in a round robin order")
        ff.write("   distributed in a round robin order\n")
      else:
        print("   NOT distributed in a round robin order")
        ff.write("   NOT distributed in a round robin order\n")

  try:
    import matplotlib
  except:
    print("Unable to open matplotlib to plot speedup")
    return

  try:
    if fileoutput: matplotlib.use('Agg')
    import matplotlib.pyplot as plt
  except:
    print("Unable to open matplotlib to plot speedup")
    return

  try:
    fig, ax1 = plt.subplots()
    plt.title('MPI Perfect and Streams Speedup')
    ax2 = ax1.twinx()
    ax1.set_autoscaley_on(False)

    # make sure that actual bandwidth values (as opposed to perfect speedup) takes
    # at least a third of the y axis
    ymax = min(max(hosts.keys()), 3*max(triads.values())/min(triads.values()) - 2)

    ax1.set_xlim([min(hosts.keys()),max(hosts.keys())])
    ax1.set_ylim([min(hosts.keys()),ymax])
    ax1.set_xlabel('Number of MPI processes')
    ax1.set_ylabel('Memory Bandwidth Speedup')
    ax1.plot(hosts.keys(),hosts.keys(),'b',hosts.keys(),speedups.values(),'r-o')
    ax2.set_autoscaley_on(False)
    ax2.set_xlim([min(hosts.keys()),max(hosts.keys())])
    ax2.set_ylim([min(triads.values())/1000.,min(triads.values())*ymax/1000.])
    ax2.set_ylabel("Achieved Bandwidth. Gigabytes per Second")

    plt.show()
    if fileoutput: plt.savefig('scaling.png')
    if fileoutput: print("See graph in the file src/benchmarks/streams/scaling.png")
    if fileoutput: ff.write("See graph in the file src/benchmarks/streams/scaling.png\n")
  except Exception as e:
    if fileoutput: print("Unable to plot speedup to a file")
    else: print("Unable to display speedup plot")
    return

  ff.close()

#
#
if __name__ ==  '__main__':
  import sys
  process(len(sys.argv)-1)