gstlal_inspiral_flopulator

#!/usr/bin/env python
#
# Copyright (C) 2014  Chad Hanna, Miguel Fernandez
#
# This program is free software; you can redistribute it and/or modify it
# under the terms of the GNU General Public License as published by the
# Free Software Foundation; either version 2 of the License, or (at your
# option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General
# Public License for more details.
#
# You should have received a copy of the GNU General Public License along
# with this program; if not, write to the Free Software Foundation, Inc.,
# 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.

## @file
# A program to compute the LLOID filtering costs of an SVD bank as produced by gstlal_svd_bank


import numpy
import sys
from gstlal import svd_bank

#FIXME hack to deal with pre ER5 / post ER5 bank file formats, remove when not an issue
try:
    banks = svd_bank.read_banks(sys.argv[1])
except AttributeError:
    banks = [svd_bank.read_bank(sys.argv[1])]

totalMFLOPS=0
averageMFLOPS=0
totalMT = 0

for i, bank in enumerate(banks):

    print ""
    rT = [f.rate for f in bank.bank_fragments]
    r = numpy.array(sorted(list(set(rT))))
    rT = numpy.array(rT)

    UT = numpy.array([f.mix_matrix.shape[0] for f in bank.bank_fragments])
    MT = [f.mix_matrix.shape[1] for f in bank.bank_fragments][0]
    NT = numpy.array([(f.end - f.start) * f.rate for f in bank.bank_fragments])

    print "\nSUB BANK %d" % i
    print "--->\tUnique sampling rates: ",r
    print "--->\tSampling rate for a given time slice: ",rT
    print "--->\tTotal SVD filters for a given time slice: ",UT
    print "--->\tNumber of SVD filter samples: ",NT
    print "--->\tTotal real templates (e.g. twice number of complex templates): ",MT

    # Convolution of a 16 sample filter requires a multiply-add per sample point of data for each sample of the filter for each physical template
    resample = (r * 16 * 2 * MT).sum()

    # Convolution of a NT sample filter requires a multiply-add per sample point of data for each sample of the filter for each svd template
    filter = (NT * rT * UT * 2).sum()

    reconstruct = (MT * UT * rT * 2).sum()

    add = (rT * MT).sum()

    # get FLOPs per *complex* template (note the divide by 2)
    totalMT += MT / 2
    print "--->\tMFLOPS from resampling: ", resample / 1000.**2
    print "--->\tMFLOPS from filtering: ", filter / 1000.**2
    print "--->\tMFLOPS from reconstruction: ", reconstruct / 1000.**2
    print "--->\tMFLOPS from addition: ", add / 1000.**2

    MFLOPs = resample / 1000.**2  + filter / 1000.**2 + reconstruct / 1000.**2 + add / 1000.**2
    totalMFLOPS += MFLOPs

    print "--->\tTotal MFLOPS: ", MFLOPs
    print "--->\tMFLOPS per complex template: ", MFLOPs / (MT / 2.)
    
averageMFLOPS = totalMFLOPS / totalMT
print "--\nAverage MFLOPS per template:", averageMFLOPS