def __init__(self, args):
# Ensure either a file or directory is specified.
assert args.trace_file != None or args.trace_dir != None
# Maximize perfidx, and cache the fixed inputs.
super(MdriverTuner, self).__init__(args,
objective=MaximizeAccuracy(),
input_manager=FixedInputManager())
def __init__(self, driver, *args, **kwargs):
"""
Create a new measurement interface with knowledge of the
queues to communicate with the ctree jit module.
"""
kwargs['input_manager'] = FixedInputManager()
super(CtreeMeasurementInterface, self).__init__(*args, **kwargs)
self._ctree_driver = driver
def __init__(self, args):
self.program_settings = json.load(open(args.program_settings))
input_manager = FixedInputManager(size=self.program_settings['n'])
objective = ThresholdAccuracyMinimizeTime(self.program_settings['accuracy'])
# pass many settings to parent constructor
super(PetaBricksInterface, self).__init__(
args, program_name=args.program,
program_version=self.file_hash(args.program),
input_manager=input_manager, objective=objective)
def main(args, cfg, jvmOptions):
logging.basicConfig(level=logging.INFO)
manipulator = ConfigurationManipulator()
params = dict(cfg.items() + jvmOptions.items())
#print "\nFeature variables...."
for key in params.keys():
#print "\t", key
manipulator.add_parameter(params.get(key))
mi = StreamJitMI(args, jvmOptions, manipulator, FixedInputManager(),
MinimizeTime())
m = TuningRunMain(mi, args)
m.main()
def main(args, cfg, ss):
logging.basicConfig(level=logging.INFO)
manipulator = ConfigurationManipulator()
params = cfg.getAllParameters()
print "\nFeature variables...."
for key in params.keys():
print "\t", key
manipulator.add_parameter(cfg.getParameter(key))
mi = StreamJitMI(args, ss, manipulator, FixedInputManager(),
MinimizeTime())
m = TuningRunMain(mi, args)
m.main()
def __init__(self, args, stableParams, adjParams, accThresh, tuneRuns,
verifyRuns, inputGen, inputGenParams, runner, spec,
accMetric):
self.stableParams = stableParams
self.adjParams = adjParams
self.accThresh = accThresh
self.tuneRuns = tuneRuns
self.verifyRuns = verifyRuns
self.inputGen = inputGen
self.inputGenParams = inputGenParams
self.runner = runner
self.spec = spec
self.accMetric = accMetric
igParams = inputGenParams(stableParams, 1)
self.inputData = inputGen(*igParams)
AxProfUtil.writeDataToFile(self.inputData, AxProf.defaultInputFileName)
objective = ThresholdAccuracyMinimizeTime(accThresh)
input_manager = FixedInputManager()
super(AxProfTunerInterface, self).__init__(args,
objective=objective,
input_manager=input_manager)
def __init__(self, *args, **kwargs):
kwargs["objective"] = MaximizeAccuracy()
kwargs["input_manager"] = FixedInputManager()
super(ModelTuner, self).__init__(*args, **kwargs)
self.model = Model(self.run_id)
def __init__(self, args): super(Unicorn, self).__init__(args, input_manager=FixedInputManager(), objective=MaximizeAccuracy())
if isinstance(p, sjparameters.sjCompositionParameter):
p.equal_division(seed_config)
elif isinstance(p, sjparameters.sjPermutationParameter):
pass #p.set_value(config, p.seed_value())
else:
seed_config[p.name] = p.value
seed_config['multiplier'] = m
seed_configs.append(seed_config)
# The default bandit, plus our custom techniques.
from opentuner.search import technique, bandittechniques, differentialevolution, evolutionarytechniques, simplextechniques
technique.register(
bandittechniques.AUCBanditMetaTechnique([
sjtechniques.FixedTechnique(seed_configs),
differentialevolution.DifferentialEvolutionAlt(),
evolutionarytechniques.UniformGreedyMutation(),
evolutionarytechniques.NormalGreedyMutation(mutation_rate=0.3),
simplextechniques.RandomNelderMead(),
sjtechniques.ForceRemove(),
sjtechniques.ForceFuse(),
sjtechniques.ForceUnbox(),
sjtechniques.ForceEqualDivision(),
sjtechniques.CrossSocketBeforeHyperthreadingAffinity(),
],
name="StreamJITBandit"))
mi = StreamJITMI(args, cfg, jvm_options, manipulator, FixedInputManager(),
MinimizeTime())
m = TuningRunMain(mi, args)
m.main()
def main(self):
"""
Main function
"""
argparser = argparse.ArgumentParser(description='Standalone HLS tuner')
argparser.add_argument('src_dir',
help='Directory with sources',
type=os.path.abspath)
argparser.add_argument('pragma_file',
help='Pragma specification file',
type=os.path.abspath)
argparser.add_argument('output_dir',
help='Output directory',
type=os.path.abspath)
argparser.add_argument('--overwrite',
action='store_true',
help='Overwrite existing tuning runs')
argparser.add_argument('--append',
action='store_true',
help='Append new tuning run to existing runs')
argparser.add_argument('--technique',
action='store',
nargs='*',
help="Search technique",
default=['AUCBanditMetaTechniqueA'])
argparser.add_argument('--parallelism',
type=int,
default=4,
help='Number of builds running in parallel')
argparser.add_argument('--use_prebuilt',
action='store_true',
help='Use prebuilt kernel (for debugging)')
argparser.add_argument('--max_luts', type=int, help='LUT constraint')
argparser.add_argument('--max_regs',
type=int,
help='Register constraint')
argparser.add_argument('--max_dsps', type=int, help='DSP constraint')
argparser.add_argument('--max_brams', type=int, help='BRAM constraint')
argparser.add_argument('--iters',
type=int,
help='Number of test iterations')
argparser.add_argument(
'--no_cleanup',
action='store_true',
help='Do not remove output of successful builds')
args = argparser.parse_args()
self.prepare_output_dir(args)
self.init_logging(args.output_dir)
new_argparser = argparse.ArgumentParser(parents=opentuner.argparsers())
new_args = new_argparser.parse_args('', args)
new_args.database = args.output_dir + '/HLS_tuner.db'
if args.iters is not None:
new_args.test_limit = args.iters
log.info('Search algorithm: ' + str(new_args.technique))
if args.max_luts or args.max_regs or args.max_dsps or args.max_brams:
objective = ThresholdAreaMinimizeTime(args.max_luts, args.max_regs,
args.max_dsps,
args.max_brams)
else:
objective = MinimizeTime()
input_manager = FixedInputManager()
VivadoHLSInterface.main(new_args,
tuner_root=tuner_root,
objective=objective,
input_manager=input_manager)