[clfft] 107/107: updated performance measurement scripts and instruction manual
Ghislain Vaillant
ghisvail-guest at moszumanska.debian.org
Thu Jul 30 18:06:43 UTC 2015
This is an automated email from the git hooks/post-receive script.
ghisvail-guest pushed a commit to branch master
in repository clfft.
commit 3a2ae95f3cdbaa8d0ee66c2d4f6f1c66da43d422
Author: bnataraj <bragadeesh.natarajan at amd.com>
Date: Wed Jul 22 07:01:50 2015 -0500
updated performance measurement scripts and instruction manual
---
src/scripts/perf/fftPerformanceTesting.py | 37 ++--
src/scripts/perf/measurePerformance.py | 327 +++++++++++++++++++-----------
src/scripts/perf/plotPerformance.py | 40 +++-
3 files changed, 261 insertions(+), 143 deletions(-)
diff --git a/src/scripts/perf/fftPerformanceTesting.py b/src/scripts/perf/fftPerformanceTesting.py
index 2d1df69..133062d 100644
--- a/src/scripts/perf/fftPerformanceTesting.py
+++ b/src/scripts/perf/fftPerformanceTesting.py
@@ -91,18 +91,18 @@ def executable(library):
if library == 'clFFT' or library == 'null':
if sys.platform == 'win32':
- exe = 'Client.exe'
+ exe = 'clFFT-client-2.5.0.exe'
elif sys.platform == 'linux2':
- exe = 'Client'
+ exe = 'clFFT-client-2.5.0'
+ elif library == 'cuFFT':
+ if sys.platform == 'win32':
+ exe = 'cuFFT-Client.exe'
+ elif sys.platform == 'linux2':
+ exe = 'cuFFT-Client'
else:
print 'ERROR: unknown library -- cannot determine executable name'
quit()
- if not os.path.isfile(exe):
- error_message = 'ERROR: could not find client named ' + exe
- print error_message
- quit()
-
return exe
def max_mem_available_in_bytes(exe, device):
@@ -115,11 +115,8 @@ def max_mem_available_in_bytes(exe, device):
return int(maxMemoryAvailable.group(0))
def max_problem_size(exe, layout, precision, device):
- if layout == 'ci' or layout == 'cp':
+ if layout == 1 or layout == 1:
numbers_in_one_datapoint = 2
- else:
- print 'max_problem_size(): unknown layout'
- quit()
if precision == 'single':
bytes_in_one_number = 4
@@ -129,18 +126,22 @@ def max_problem_size(exe, layout, precision, device):
print 'max_problem_size(): unknown precision'
quit()
- max_problem_size = max_mem_available_in_bytes(exe, device) / (numbers_in_one_datapoint * bytes_in_one_number)
- max_problem_size = max_problem_size / 16
+ max_problem_size = pow(2,25) #max_mem_available_in_bytes(exe, device)
+ if layout == '5':
+ max_problem_size = pow(2,24) #max_mem_available_in_bytes(exe, device)
+ #max_problem_size=max_problem_size/ (numbers_in_one_datapoint * bytes_in_one_number)
+ #max_problem_size = max_problem_size / 16
return max_problem_size
def maxBatchSize(lengthx, lengthy, lengthz, layout, precision, exe, device):
problemSize = int(lengthx) * int(lengthy) * int(lengthz)
maxBatchSize = max_problem_size(exe, layout, precision, device) / problemSize
- if int(lengthx) == pow(2,16) or int(lengthx) == pow(2,17):
- # special cases in the kernel. extra padding is added in, so we need to shrink the batch size to accommodate
- return str(maxBatchSize/2)
- else:
- return str(maxBatchSize)
+ return str(maxBatchSize)
+ #if int(lengthx) == pow(2,16) or int(lengthx) == pow(2,17):
+ # # special cases in the kernel. extra padding is added in, so we need to shrink the batch size to accommodate
+ # return str(maxBatchSize/2)
+ #else:
+ # return str(maxBatchSize)
def create_ini_file_if_requested(args):
if args.createIniFilename:
diff --git a/src/scripts/perf/measurePerformance.py b/src/scripts/perf/measurePerformance.py
index c0cbc6a..e6cf596 100644
--- a/src/scripts/perf/measurePerformance.py
+++ b/src/scripts/perf/measurePerformance.py
@@ -20,9 +20,11 @@ import subprocess
import itertools
import re#gex
import os
+import math
from threading import Timer, Thread
import thread, time
from platform import system
+import numpy as np
from datetime import datetime
@@ -33,20 +35,20 @@ from performanceUtility import timeout, log, generate235Radices
IAM = 'FFT'
TIMOUT_VAL = 900 #In seconds
-devicevalues = ['gpu', 'cpu']
+devicevalues = ['g', 'c']
layoutvalues = ['cp', 'ci']
placevalues = ['in', 'out']
precisionvalues = ['single', 'double']
-libraryvalues = ['clFFT']
+libraryvalues = ['clFFT','cuFFT']
pow10 = '1-9,10-90:10,100-900:100,1000-9000:1000,10000-90000:10000,100000-900000:100000,1000000-9000000:1000000'
parser = argparse.ArgumentParser(description='Measure performance of the clFFT library')
parser.add_argument('--device',
- dest='device', default='gpu',
+ dest='device', default='g',
help='device(s) to run on; may be a comma-delimited list. choices are ' + str(devicevalues) + '. (default gpu)')
parser.add_argument('-b', '--batchsize',
dest='batchSize', default='1',
- help='number of FFTs to perform with one invocation of the client. the special value \'max\' may be used to adjust the batch size on a per-transform basis to the maximum problem size possible on the device. may be a range or a comma-delimited list. if a range is entered, you may follow it with \':X\', where X is the stepping of the range (if omitted, it defaults to a stepping of 1). e.g., 1-15 or 12,18 or 7,10-30:10,1050-1054. the special value \'pow10\' expands to \'{}\'. Note that [...]
+ help='number of FFTs to perform with one invocation of the client. the special value \'adapt\' may be used to adjust the batch size on a per-transform basis to the maximum problem size possible on the device. (default 1)'.format(pow10))
parser.add_argument('-a', '--adaptivemax',
dest='constProbSize', default='-1',
help='Max problem size that you want to maintain across the invocations of client with different lengths. This is adaptive and adjusts itself automtically.'.format(pow10))
@@ -59,36 +61,48 @@ parser.add_argument('-y', '--lengthy',
parser.add_argument('-z', '--lengthz',
dest='lengthz', default='1',
help='length(s) of z to test; must be factors of 1, 2, 3, or 5 with clFFT; may be a range or a comma-delimited list. e.g., 16-128 or 1200 or 16,32768 (default 1)')
+parser.add_argument('-reps',
+ dest='reps', default='10',
+ help='Number of repetitions (default 10)')
+parser.add_argument('-prime_factor', '--prime_factor',
+ dest='prime_factor', default='2',
+ help='only test the prime factors within the specified range of lengthx/y/z. Select from 2,3,5, and 7. Example: -prime_factor 2,3')
+parser.add_argument('-test_count', '--test_count',
+ dest='test_count', default='100',
+ help='Number of tests to perform')
parser.add_argument('--problemsize',
- dest='problemsize', default=None,
- help='additional problems of a set size. may be used in addition to lengthx/y/z. each indicated problem size will be added to the list of FFTs to perform. should be entered in AxBxC:D format. A, B, and C indicate the sizes of the X, Y, and Z dimensions (respectively). D is the batch size. All values except the length of X are optional. may enter multiple in a comma-delimited list. e.g., 2x2x2:32768 or 256x256:100,512x512:256')
+ dest='problemsize', default=None)
+# help='additional problems of a set size. may be used in addition to lengthx/y/z. each indicated problem size will be added to the list of FFTs to perform. should be entered in AxBxC:D format. A, B, and C indicate the sizes of the X, Y, and Z dimensions (respectively). D is the batch size. All values except the length of X are optional. may enter multiple in a comma-delimited list. e.g., 2x2x2:32768 or 256x256:100,512x512:256')
parser.add_argument('-i', '--inputlayout',
- dest='inputlayout', default='ci',
- help='may enter multiple in a comma-delimited list. choices are ' + str(layoutvalues) + '. ci = complex interleaved, cp = complex planar (default ci)')
+ dest='inputlayout', default='1',
+ help=' 1. interleaved (default) 2. planar 3. hermitian interleaved 4. hermitian planar 5. real' )
parser.add_argument('-o', '--outputlayout',
- dest='outputlayout', default='ci',
- help='may enter multiple in a comma-delimited list. choices are ' + str(layoutvalues) + '. ci = complex interleaved, cp = complex planar (default ci)')
-parser.add_argument('-p', '--placeness',
+ dest='outputlayout', default='1',
+ help=' 1. interleaved (default) 2. planar 3. hermitian interleaved 4. hermitian planar 5. real' )
+parser.add_argument('--placeness',
dest='placeness', default='in',
- help='may enter multiple in a comma-delimited list. choices are ' + str(placevalues) + '. in = in place, out = out of place (default in)')
+ help='Choices are ' + str(placevalues) + '. in = in place, out = out of place (default in)')
parser.add_argument('-r', '--precision',
dest='precision', default='single',
- help='may enter multiple in a comma-delimited list. choices are ' + str(precisionvalues) + '. (default single)')
+ help='Choices are ' + str(precisionvalues) + '. (default single)')
parser.add_argument('--library',
dest='library', default='clFFT', choices=libraryvalues,
help='indicates the library to use for testing on this run')
parser.add_argument('--label',
dest='label', default=None,
help='a label to be associated with all transforms performed in this run. if LABEL includes any spaces, it must be in \"double quotes\". note that the label is not saved to an .ini file. e.g., --label cayman may indicate that a test was performed on a cayman card or --label \"Windows 32\" may indicate that the test was performed on Windows 32')
-parser.add_argument('--createini',
- dest='createIniFilename', default=None,
- help='create an .ini file with the given name that saves the other parameters given at the command line, then quit. e.g., \'measureperformance.py -x 2048 --createini my_favorite_setup.ini\' will create an .ini file that will save the configuration for a 2048-datapoint 1D FFT.')
+#parser.add_argument('--createini',
+# dest='createIniFilename', default=None,
+# help='create an .ini file with the given name that saves the other parameters given at the command line, then quit. e.g., \'measureperformance.py -x 2048 --createini my_favorite_setup.ini\' will create an .ini file that will save the configuration for a 2048-datapoint 1D FFT.')
parser.add_argument('--ini',
dest='iniFilename', default=None,
help='use the parameters in the named .ini file instead of the command line parameters.')
parser.add_argument('--tablefile',
dest='tableOutputFilename', default=None,
help='save the results to a plaintext table with the file name indicated. this can be used with plotPerformance.py to generate graphs of the data (default: table prints to screen)')
+parser.add_argument('--prefix',
+ dest='prefix', default='./',
+ help='Path where the library client is located (default current directory)')
args = parser.parse_args()
@@ -145,7 +159,7 @@ def checkTimeOutPut(args):
except:
printLog("ERROR: UNKNOWN Exception - +checkWinTimeOutPut()::executeCommand()")
- currCommandProcess = subprocess.Popen(args, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+ currCommandProcess = subprocess.Popen(args, stdout=subprocess.PIPE, stderr=subprocess.PIPE,shell=True)
thread = Thread(target=executeCommand)
thread.start()
thread.join(TIMOUT_VAL) #wait for the thread to complete
@@ -165,75 +179,85 @@ def checkTimeOutPut(args):
# don't try to create and use an .ini file at the same time (it will open a portal through which demons will emerge)
-if args.iniFilename and args.createIniFilename:
- printLog('ERROR: --ini and --createini are mutually exclusive. Please choose only one.')
- quit()
+#if args.iniFilename and args.createIniFilename:
+# printLog('ERROR: --ini and --createini are mutually exclusive. Please choose only one.')
+# quit()
#read in .ini parameters if --ini is used
-if args.iniFilename != None:
- if not os.path.isfile(args.iniFilename):
- printLog("No file with the name \'{}\' exists. Please indicate another filename.".format(args.iniFilename))
- quit()
-
- ini = open(args.iniFilename, 'r')
- iniContents = ini.read()
- iniContents = iniContents.split(';')
- for i in range(0,len(iniContents)):
- line = iniContents.pop()
- line = line.partition(' ')
- parameter = line[0]
- value = line[2]
- value = value.replace('\'','').replace('[','').replace(']','').replace(' ','')
-
- if parameter == 'batchSize':
- args.batchSize = value
- elif parameter == 'constProbSize':
- args.constProbSize = value
- elif parameter == 'lengthx':
- args.lengthx = value
- elif parameter == 'lengthy':
- args.lengthy = value
- elif parameter == 'lengthz':
- args.lengthz = value
- elif parameter == 'problemsize':
- args.problemsize = value
- elif parameter == 'device':
- args.device = value
- elif parameter == 'inputlayout':
- args.inputlayout = value
- elif parameter == 'outputlayout':
- args.outputlayout = value
- elif parameter == 'placeness':
- args.placeness = value
- elif parameter == 'precision':
- args.precision = value
- else:
- printLog('{} corrupted. Please re-create a .ini file with the --createini flag.'.format(args.iniFilename))
- quit()
+#if args.iniFilename != None:
+# if not os.path.isfile(args.iniFilename):
+# printLog("No file with the name \'{}\' exists. Please indicate another filename.".format(args.iniFilename))
+# quit()
+#
+# ini = open(args.iniFilename, 'r')
+# iniContents = ini.read()
+# iniContents = iniContents.split(';')
+# for i in range(0,len(iniContents)):
+# line = iniContents.pop()
+# line = line.partition(' ')
+# parameter = line[0]
+# value = line[2]
+# value = value.replace('\'','').replace('[','').replace(']','').replace(' ','')
+# print"value= ",value
+#
+# if parameter == 'batchSize':
+# args.batchSize = value
+# elif parameter == 'constProbSize':
+# args.constProbSize = value
+# elif parameter == 'lengthx':
+# args.lengthx = value
+# elif parameter == 'reps':
+# args.reps = value
+# elif parameter == 'prime_factor':
+# args.prime_factor = value
+# elif parameter == 'test_count':
+# args.test_count = value
+# elif parameter == 'lengthy':
+# args.lengthy = value
+# elif parameter == 'lengthz':
+# args.lengthz = value
+# elif parameter == 'problemsize':
+# args.problemsize = value
+# elif parameter == 'device':
+# args.device = value
+# elif parameter == 'inputlayout':
+# args.inputlayout = value
+# elif parameter == 'outputlayout':
+# args.outputlayout = value
+# elif parameter == 'placeness':
+# args.placeness = value
+# elif parameter == 'precision':
+# args.precision = value
+# else:
+# printLog('{} corrupted. Please re-create a .ini file with the --createini flag.'.format(args.iniFilename))
+# #quit()
#create ini file if requested
-if args.createIniFilename != None:
- printLog('Creating Ini files')
- if os.path.isfile(args.createIniFilename):
- printLog('A file with the name \'{}\' already exists. Please delete the file or choose another name.'.format(args.createIniFilename))
- quit()
- printLog('Creating Ini file:'+args.createIniFilename+'\n')
- ini = open(args.createIniFilename, 'w')
- ini.write('batchSize {} ;'.format(args.batchSize))
- ini.write('constProbSize {} ;'.format(args.constProbSize))
- ini.write('lengthx {} ;'.format(args.lengthx))
- ini.write('lengthy {} ;'.format(args.lengthy))
- ini.write('lengthz {} ;'.format(args.lengthz))
- ini.write('problemsize {} ;'.format(args.problemsize))
- ini.write('device {} ;'.format(args.device))
- ini.write('inputlayout {} ;'.format(args.inputlayout))
- ini.write('outputlayout {} ;'.format(args.outputlayout))
- ini.write('placeness {} ;'.format(args.placeness))
- ini.write('precision {} ;'.format(args.precision))
- printLog('Created Ini file:'+args.createIniFilename+'\n')
- printLog("=========================MEASURE PERFORMANCE START===========================\n")
- quit()
-
+#if args.createIniFilename != None:
+# printLog('Creating Ini files')
+# if os.path.isfile(args.createIniFilename):
+# printLog('A file with the name \'{}\' already exists. Please delete the file or choose another name.'.format(args.createIniFilename))
+# quit()
+# printLog('Creating Ini file:'+args.createIniFilename+'\n')
+# ini = open(args.createIniFilename, 'w')
+# ini.write('batchSize {} ;'.format(args.batchSize))
+# ini.write('constProbSize {} ;'.format(args.constProbSize))
+# ini.write('lengthx {} ;'.format(args.lengthx))
+# ini.write('lengthy {} ;'.format(args.lengthy))
+# ini.write('lengthz {} ;'.format(args.lengthz))
+# ini.write('prime_factor {} ;'.format(args.prime_factor))
+# ini.write('test_count {} ;'.format(args.test_count))
+# ini.write('reps {} ;'.format(args.reps))
+# ini.write('problemsize {} ;'.format(args.problemsize))
+# ini.write('device {} ;'.format(args.device))
+# ini.write('inputlayout {} ;'.format(args.inputlayout))
+# ini.write('outputlayout {} ;'.format(args.outputlayout))
+# ini.write('placeness {} ;'.format(args.placeness))
+# ini.write('precision {} ;'.format(args.precision))
+# printLog('Created Ini file:'+args.createIniFilename+'\n')
+# printLog("=========================MEASURE PERFORMANCE START===========================\n")
+# quit()
+#
#turn pow10 into its range list
if args.batchSize.count('pow10'):
@@ -246,6 +270,7 @@ args.device = args.device.split(',')
args.lengthx = args.lengthx.split(',')
args.lengthy = args.lengthy.split(',')
args.lengthz = args.lengthz.split(',')
+args.prime_factor = args.prime_factor.split(',')
if args.problemsize:
args.problemsize = args.problemsize.split(',')
args.inputlayout = args.inputlayout.split(',')
@@ -266,11 +291,11 @@ if ( args.batchSize.count('max') or args.batchSize.count('adapt') )and len(args.
# in case of an in-place transform, input and output layouts must be the same (otherwise: *boom*)
-for n in args.placeness:
- if n == 'in' or n == 'inplace':
- if len(args.inputlayout) > 1 or len(args.outputlayout) > 1 or args.inputlayout[0] != args.outputlayout[0]:
- printLog('ERROR: if transformation is in-place, input and output layouts must match')
- quit()
+#for n in args.placeness:
+# if n == 'in' or n == 'inplace':
+# if len(args.inputlayout) > 1 or len(args.outputlayout) > 1 or args.inputlayout[0] != args.outputlayout[0]:
+# printLog('ERROR: if transformation is in-place, input and output layouts must match')
+# quit()
# check for valid values in precision
for n in args.precision:
@@ -393,7 +418,7 @@ if args.library == 'clFFT':
-if not os.path.isfile(executable(args.library)):
+if not os.path.isfile(args.prefix+executable(args.library)):
printLog("ERROR: Could not find client named {0}".format(executable(args.library)))
quit()
@@ -477,6 +502,37 @@ args.lengthy = Range(args.lengthy, 'l').expanded
args.lengthz = Range(args.lengthz, 'l').expanded
+def create_prime_factors(args,input_list):
+ powers2=[1]
+ powers3=[1]
+ powers5=[1]
+ powers7=[1]
+ if '2' in args.prime_factor:
+ powers2+=[2**x for x in range(1,int(math.floor(math.log(max(input_list),2)+1)))]
+ if '3' in args.prime_factor:
+ powers3+=[3**x for x in range(1,int(math.floor(math.log(max(input_list),3)+1)))]
+ if '5' in args.prime_factor:
+ powers5+=[5**x for x in range(1,int(math.floor(math.log(max(input_list),5)+1)))]
+ if '7' in args.prime_factor:
+ powers7+=[7**x for x in range(1,int(math.floor(math.log(max(input_list),7)+1)))]
+
+
+ xlist=[]
+ for i in powers2:
+ for j in powers3:
+ for k in powers5:
+ for l in powers7:
+ dummy=int(i)*int(j)*int(k)*int(l)
+ if(dummy<=max(input_list)) and (dummy>=min(input_list)):
+ xlist.append(dummy)
+
+ xlist=sorted(xlist)
+ xlist=xlist[:int(args.test_count)] #snafu
+ return xlist
+
+args.lengthx=create_prime_factors(args,args.lengthx)
+args.lengthy=create_prime_factors(args,args.lengthy)
+args.lengthz=create_prime_factors(args,args.lengthz)
#expand problemsizes ('XxYxZ:batch')
#print "args.problemsize--1-->", args.problemsize
@@ -489,15 +545,21 @@ if args.problemsize and args.problemsize[0] != 'None':
#create the problem size combinations for each run of the client
-problem_size_combinations = itertools.product(args.lengthx, args.lengthy, args.lengthz, args.batchSize)
+# A: This part creats a product of all possible combinations. Too many cases in 2/3D
+#problem_size_combinations = itertools.product(args.lengthx, args.lengthy, args.lengthz, args.batchSize)
+#problem_size_combinations = list(itertools.islice(problem_size_combinations, None))
-problem_size_combinations = list(itertools.islice(problem_size_combinations, None))
+if args.lengthy[0]==1:
+ args.lengthy=[1]*len(args.lengthx)
+if args.lengthz[0]==1:
+ args.lengthz=[1]*len(args.lengthx)
-#print "args.problemsize--2-->", args.problemsize
+dummy=[args.batchSize[0]]*len(args.lengthx)
+problem_size_combinations=zip(args.lengthx,args.lengthy,args.lengthz,dummy)
+#print "args.problemsize--2-->", args.problemsize
#add manually entered problem sizes to the list of FFTs to crank out
manual_test_combinations = []
-
if args.problemsize and args.problemsize[0] != 'None':
for n in args.problemsize:
x = []
@@ -536,6 +598,16 @@ test_combinations = itertools.product(problem_size_combinations, args.device, ar
test_combinations = list(itertools.islice(test_combinations, None))
test_combinations = [TestCombination(params[0][0], params[0][1], params[0][2], params[0][3], params[1], params[2], params[3], params[4], params[5], args.label) for params in test_combinations]
+if args.iniFilename != None:
+ array=np.genfromtxt(args.iniFilename, names=True, delimiter=',', dtype=None)
+ test_combinations = [TestCombination(params[0],params[1], params[2], params[3], params[4],params[5],params[6],params[7],params[8],args.label) for params in array]
+
+#print("lenghtx= ",test_combinations[0].x)
+#print("lenghty= ",test_combinations[0].y)
+#print("lenghtz= ",test_combinations[0].z)
+#print("placeness= ",test_combinations[0].placeness)
+
+
#turn each test combination into a command, run the command, and then stash the gflops
result = [] # this is where we'll store the results for the table
@@ -544,7 +616,10 @@ result = [] # this is where we'll store the results for the table
#open output file and write the header
if args.tableOutputFilename == None:
- args.tableOutputFilename = 'results' + datetime.now().isoformat().replace(':','.') + '.txt'
+ if args.library == 'cuFFT':
+ args.tableOutputFilename = 'cuFFT_' + 'x_'+ str(args.lengthx[0]) + '_y_'+str(args.lengthy[0])+'_z_'+str(args.lengthz[0])+'_'+str(args.precision[0]) +'_'+datetime.now().isoformat().replace(':','.') + '.txt'
+ elif args.library=='clFFT':
+ args.tableOutputFilename = 'clFFT_' + 'x_'+ str(args.lengthx[0]) + '_y_'+str(args.lengthy[0])+'_z_'+str(args.lengthz[0])+'_'+str(args.precision[0])+ '_'+datetime.now().isoformat().replace(':','.') + '.txt'
else:
if os.path.isfile(args.tableOutputFilename):
oldname = args.tableOutputFilename
@@ -558,17 +633,17 @@ printLog('table header---->'+ str(tableHeader))
table = open(args.tableOutputFilename, 'w')
table.write(tableHeader + '\n')
table.flush()
-
if args.constProbSize == -1:
- args.constProbSize = maxBatchSize(1, 1, 1, args.inputlayout[0], args.precision[0], executable(args.library), '--' + args.device[0])
+ args.constProbSize = maxBatchSize(1, 1, 1, args.inputlayout[0], args.precision[0], executable(args.library), '-' + args.device[0])
args.constProbSize = int(args.constProbSize)
printLog('Total combinations = '+str(len(test_combinations)))
vi = 0
-#test_combinations = test_combinations[825:830]
for params in test_combinations:
+ if vi>=int(args.test_count):
+ break
vi = vi+1
printLog("")
printLog('preparing command: '+ str(vi))
@@ -576,28 +651,18 @@ for params in test_combinations:
lengthx = str(params.x)
lengthy = str(params.y)
lengthz = str(params.z)
+ inlayout=str(params.inlayout)
+ outlayout=str(params.outlayout)
+ prefix=str(args.prefix)
+
if params.batchsize == 'max':
- batchSize = maxBatchSize(lengthx, lengthy, lengthz, params.inlayout, params.precision, executable(args.library), '--' + device)
+ batchSize = maxBatchSize(lengthx, lengthy, lengthz, params.inlayout, params.precision, executable(args.library), '-' + device)
elif params.batchsize == 'adapt':
batchSize = str(args.constProbSize/(int(lengthx)*int(lengthy)*int(lengthz)))
else:
batchSize = str(params.batchsize)
- if params.inlayout == 'complexinterleaved' or params.inlayout == 'ci':
- inputlayout = '1'
- elif params.inlayout == 'complexplanar' or params.inlayout == 'cp':
- inputlayout = '2'
- else:
- printLog('ERROR: invalid value for input layout when assembling client command')
-
- if params.outlayout == 'complexinterleaved' or params.outlayout == 'ci':
- outputlayout = '1'
- elif params.outlayout == 'complexplanar' or params.outlayout == 'cp':
- outputlayout = '2'
- else:
- printLog('ERROR: invalid value for output layout when assembling client command')
-
if params.placeness == 'inplace' or params.placeness == 'in':
placeness = ''
elif params.placeness == 'outofplace' or params.placeness == 'out':
@@ -615,20 +680,39 @@ for params in test_combinations:
#set up arguments here
if args.library == 'clFFT':
- arguments = [executable(args.library),
- '--' + device,
+ arguments = [prefix+ executable(args.library),
+ '-' + device,
'-x', lengthx,
'-y', lengthy,
'-z', lengthz,
'--batchSize', batchSize,
- '--inLayout', inputlayout,
- '--outLayout', outputlayout,
+ '--inLayout', inlayout,
+ '--outLayout',outlayout,
placeness,
precision,
- '-p', '10']
+ '-p', args.reps]
+ elif args.library == 'cuFFT':
+ if inlayout[0]=='1' and outlayout[0]=='1':
+ cuFFT_type='1'
+ elif inlayout[0]=='3' and outlayout[0]=='5':
+ cuFFT_type='3'
+ elif inlayout[0]=='5' and outlayout[0]=='3':
+ cuFFT_type='2'
+ else:
+ print"Wrong input/outputlayout. Only C2C/R2C/C2R are supported for Cuda"
+ exit()
+ arguments=[prefix+'cuFFT-Client',
+ '-x', lengthx,
+ '-y', lengthy,
+ '-z', lengthz,
+ '-b', batchSize,
+ '-p',args.reps,
+ '-d',str(int(args.precision[0]=='double')),
+ '-type',cuFFT_type]
writeline = True
try:
+ arguments=' '.join(arguments)
printLog('Executing Command: '+str(arguments))
output = checkTimeOutPut(arguments)
output = output.split(os.linesep);
@@ -656,10 +740,19 @@ for params in test_combinations:
if writeline:
try:
- output = itertools.ifilter( lambda x: x.count('Gflops'), output)
+ if args.library == 'cuFFT':
+ output = itertools.ifilter( lambda x: x.count('gflops'), output)
+ else:
+ output = itertools.ifilter( lambda x: x.count('Gflops'), output)
+
output = list(itertools.islice(output, None))
thisResult = re.search('\d+\.*\d*e*-*\d*$', output[-1])
+ if args.library == 'cuFFT':
+ thisResult = re.search('[-+]?\d*\.\d+|\d+$', output[-1])
thisResult = float(thisResult.group(0))
+
+ if args.library=='clFFT' and params.inlayout=='5':
+ thisResult=thisResult/2;
thisResult = (params.x, params.y, params.z, batchSize, params.device, params.inlayout, params.outlayout, params.placeness, params.precision, params.label, thisResult)
outputRow = ''
diff --git a/src/scripts/perf/plotPerformance.py b/src/scripts/perf/plotPerformance.py
index 3976512..0fe14a5 100644
--- a/src/scripts/perf/plotPerformance.py
+++ b/src/scripts/perf/plotPerformance.py
@@ -41,7 +41,7 @@ def plotGraph(dataForAllPlots, title, plottype, plotkwargs, xaxislabel, yaxislab
display a pretty graph
"""
dh.write('Making graph\n')
- colors = ['k','y','m','c','r','b','g']
+ colors = ['k','y','m','c','b','r','g']
#plottype = 'plot'
for thisPlot in dataForAllPlots:
getattr(pylab, plottype)(thisPlot.xdata, thisPlot.ydata,
@@ -150,18 +150,35 @@ def plotFromDataFile():
else: # small numbers on x-axis
args.xaxisscale = 'linear'
+ if args.yaxisscale == None:
+ args.yaxisscale = 'linear'
+
+ plotkwargs = {}
if args.xaxisscale == 'linear':
- plotkwargs = {}
plottype = 'plot'
elif args.xaxisscale == 'log2':
plottype = 'semilogx'
- plotkwargs = {'basex':2}
+ if (args.yaxisscale=='log2'):
+ plottype = 'loglog'
+ plotkwargs = {'basex':2,'basey':2}
+ elif (args.yaxisscale=='log10'):
+ plottype = 'loglog'
+ plotkwargs = {'basex':2,'basey':10}
+ elif (args.yaxisscale=='linear'):
+ plottype = 'semilogx'
+ plotkwargs = {'basex':2}
elif args.xaxisscale == 'log10':
plottype = 'semilogx'
- plotkwargs = {'basex':10}
+ if (args.yaxisscale=='log2'):
+ plottype = 'loglog'
+ plotkwargs = {'basex':10,'basey':2}
+ elif (args.yaxisscale=='log10'):
+ plottype = 'loglog'
+ plotkwargs = {'basex':10,'basey':10}
else:
print 'ERROR: invalid value for x-axis scale'
quit()
+
plots = set(getattr(row, multiplePlots) for row in data)
@@ -171,7 +188,7 @@ def plotFromDataFile():
self.xdata = inxdata
self.ydata = inydata
- dataForAllPlots = []
+ dataForAllPlots=[]
for plot in plots:
dataForThisPlot = itertools.ifilter( lambda x: getattr(x, multiplePlots) == plot, data)
dataForThisPlot = list(itertools.islice(dataForThisPlot, None))
@@ -216,9 +233,12 @@ def plotFromDataFile():
"""
display a pretty graph
"""
- colors = ['k','y','m','c','r','b','g']
-
- for thisPlot in dataForAllPlots:
+ colors = ['k','y','m','c','b','g','r']
+ def getkey(item):
+ return str(item.label)
+ dataForAllPlots.sort(key=getkey)
+ #for thisPlot in sorted(dataForAllPlots,key=getkey):
+ for thisPlot in sorted(dataForAllPlots,key=getkey):
getattr(pylab, plottype)(thisPlot.xdata, thisPlot.ydata, '{}.-'.format(colors.pop()), label=thisPlot.label, **plotkwargs)
if len(dataForAllPlots) > 1:
@@ -287,6 +307,10 @@ parser.add_argument('--x_axis_scale',
dest='xaxisscale', default=None, choices=['linear','log2','log10'],
help='the desired scale for the graph\'s x-axis. if nothing is specified,\
it will be selected automatically')
+parser.add_argument('--y_axis_scale',
+ dest='yaxisscale', default=None, choices=['linear','log2','log10'],
+ help='the desired scale for the graph\'s y-axis. if nothing is specified,\
+ linear will be selected')
parser.add_argument('--y_axis_label',
dest='yaxislabel', default=None,
help='the desired label for the graph\'s y-axis. if YAXISLABEL contains any\
--
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/debian-science/packages/clfft.git
More information about the debian-science-commits
mailing list