def do_calibration( device, baud, name, cal, base, numer, denom ):
if isnan( cal ):
cal = base
with Emon( device, baud ) as emon:
emon.set( name, cal )
tmp = 25
while True:
infomsg( name, "=", cal )
dmean, derr = 1, 1
while threshold < derr / dmean:
sample = tmp
infomsg( "collecting %d denominator samples" % sample, "@", datetime.now() )
dmean, derr = denom( sample )
infomsg( "sample =", sample, "dmean =", dmean, "derr =", derr )
tmp = int( np.ceil( sample * ( derr / dmean / threshold )**2 ) )
tmp = max( tmp, sample )
infomsg( "collecting %d numerator samples" % sample, "@", datetime.now() )
nmean, nerr = numer( sample )
infomsg( "sample =", sample, "nmean =", nmean, "nerr =", nerr )
if dmean - derr < nmean + nerr and nmean - nerr < dmean + derr:
break
cal *= nmean / dmean
with Emon( device, baud ) as emon:
emon.set( name, cal )
return cal
def mkconfig( test_command ):
with mktemp() as config:
check_call( [ "cp", configfile, config ] )
infomsg( "using test-command:", test_command )
with open( config, 'a' ) as fh:
infomsg( "--test-command", test_command, file = fh )
yield config
def read_wattsup( wu_device, counter, samples ):
cmd = [
os.path.join( root, "bin", "wu.py" ), wu_device,
"--log-style", "external",
"-i",
"-e", counter,
"-o", "/dev/stdout",
"--",
"sleep", str( samples )
]
infomsg( "DEBUG:", *cmd )
p = Popen( cmd, stdout = PIPE )
lines = p.communicate()[ 0 ]
if p.returncode != 0:
raise _CalledProcessError( p.returncode, cmd )
terms = len( counter.split( "," ) )
values = list()
first = True
for line in lines.splitlines():
if not first:
values.append( map( float, line.strip().split( "," )[-terms:] ) )
first = False
mean = np.mean( values, 0 )
err = 1.96 * np.std( values, 0 ) / np.sqrt( len( values ) )
return mean, err
def get_localization_files():
files = glob( os.path.join( datadir, options.localization + ".*" ) )
if len( files ) == 0:
if options.pin_root is None:
if "PIN_ROOT" in os.environ:
options.pin_root = os.environ[ "PIN_ROOT" ]
else:
infomsg(
"you must set --pin-root to generate localization files",
file = sys.stderr
)
exit( 2 )
record.time( "partitioning", check_call, [
os.path.join( root, "bin", "partition-func-localization.py" ),
"--output", os.path.join( datadir, options.localization ),
"--num", str( options.partitions ),
"--pin-root", options.pin_root,
"--source", options.source,
config
] )
files = glob( os.path.join( datadir, options.localization + ".*" ) )
if len( files ) != options.partitions:
infomsg(
"ERROR: expected", options.partitions, "partitions, but got",
len( files ), "parititions"
)
exit( 2 )
return sorted( files )
def get_builder( deltas ):
if options.sources is None:
if not options.compound_edits:
fieldpat = re.compile( r'[a-z]\((\d+),(\d+)\)' )
pending = list( reversed( deltas ) )
deltas = list()
while len( pending ) > 0:
gene = pending.pop()
if gene[ 0 ] == 'a':
deltas.append( gene )
elif gene[ 0 ] == 'd':
deltas.append( gene )
elif gene[ 0 ] == 'r':
m = fieldpat.match( gene )
dst, src = m.group( 1, 2 )
pending += [ 'd(%s)' % dst, 'a(%s,%s)' % ( dst, src ) ]
elif gene[ 0 ] == 's':
m = fieldpat.match( gene )
dst, src = m.group( 1, 2 )
pending += [
'r(%s,%s)' % ( dst, src ),
'r(%s,%s)' % ( src, dst )
]
else:
infomsg( "ERROR: unrecognized gene:", gene )
exit( 1 )
deltas = list( enumerate( deltas ) )
builder = GenomeBuilder( genprog )
else:
print >>sys.stderr, "file-based differences not implemented yet"
exit( 2 )
return deltas, builder
def validateCorrectness(self, outfile):
golden = self.getGolden()
new_golden = False
if not os.path.exists(golden):
if self.options.create_golden:
if not os.path.exists(os.path.dirname(golden)):
os.makedirs(os.path.dirname(golden))
fname = next(iter(outfile))
if os.path.isdir(fname):
check_call(["rsync", "-a", fname + "/", golden])
else:
check_call(["cp", "-p", fname, golden])
new_golden = True
else:
infomsg("ERROR: no golden output to compare against", file=sys.stderr)
infomsg("ERROR: try running with original program and --create-golden", file=sys.stderr)
raise IOError(2, "No such file or directory", golden)
try:
return self.diff(golden, outfile)
except Exception as e:
if new_golden:
check_call(["rm", "-rf", golden])
if isinstance(e, CalledProcessError):
return False
raise
def _test( self, deltas ):
# "Passing" behavior is more like the original (slower, more energy).
# "Failing" behavior is more optimized (faster, less energy).
fitness = np.array( self.get_fitness( deltas ) )
if len( fitness ) == 0:
return self.UNRESOLVED
if np.any( fitness == 0 ):
return self.UNRESOLVED
m = np.mean( fitness, axis = 0 )
s = np.std( fitness, axis = 0 )
sqrtn = np.sqrt( fitness.shape[ 0 ] )
for i in range( fitness.shape[ 1 ] ):
infomsg( " ", m[ i ], "+/-", 1.96 * s[ i ] / sqrtn )
for i in range( fitness.shape[ 1 ] ):
if np.ptp( self.optimized[ ::, i ] ) == 0 and \
np.ptp( fitness[ ::, i ] ) == 0 and \
self.optimized[ 0, i ] == fitness[ 0, i ]:
# Optimized and fitness are all the same value, likely because
# we are comparing the optimized variant to itself. This counts
# as a fail, since they are clearly drawn from the same distro.
continue
pval = mannwhitneyu( self.optimized[ ::, i ], fitness[ ::, i ] )[ 1 ]
if pval < options.alpha and m[ i ] < self.mean[ i ]:
return self.PASS
return self.FAIL
def get_repair_log( d ):
if "--seed" in cfg:
return os.path.join( d, "repair.debug.%s" % cfg[ "--seed" ] )
logs = glob( os.path.join( d, "repair.debug.*" ) )
if len( logs ) == 1:
return logs[ 0 ]
infomsg( "ERROR: could not determine repair log", file = sys.stderr )
exit( 2 )
def __init__( self, genprog, builder, deltas ):
DD.__init__( self )
self.builder = builder
self.genprog = genprog
infomsg( "INFO: computing optimized energy usage" )
self.optimized = self.get_fitness( deltas )
self.mean = numpy.mean( self.optimized )
def check_call(cmd, **kwargs):
verbose = kwargs.pop("verbose", False)
counts = set()
def get_iter(iter, arg):
if isinstance(arg, Multitmp):
counts.add(len(arg))
count = len(arg)
return iter(arg)
else:
return repeat(arg)
open_files = list()
def iopen(iterable):
for name in iterable:
if os.path.isdir(name):
yield name
else:
fh = open(name, "w")
open_files.append(fh)
yield fh
args = [get_iter(iter, arg) for arg in cmd]
kw = [(k, get_iter(iopen, v)) for k, v in kwargs.items()]
if len(counts) > 1:
raise ValueError("Multitmp args have different lengths")
e = None
procs = list()
try:
while True:
try:
tmpcmd = [next(i) for i in args]
tmpkw = {k: next(v) for k, v in kw}
if verbose:
infomsg("+", *tmpcmd)
procs.append((Popen(tmpcmd, **tmpkw), tmpcmd))
if len(counts) == 0:
raise StopIteration
except StopIteration:
break
while len(procs) > 0:
p, tmpcmd = procs.pop()
p.wait()
if p.returncode != 0:
e = CalledProcessError(cmd=tmpcmd, returncode=p.returncode)
finally:
for p, _ in procs:
p.kill()
for fh in open_files:
fh.close()
if e is not None:
raise e
def reduce_error(f, alpha, probes=5):
global debug_file
# variance computation adapted from:
# https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance
# correction computation adapted from:
# https://en.wikipedia.org/wiki/Unbiased_estimation_of_standard_deviation
if probes % 2 == 0:
probes += 1
log = debug_file
debug_file = None
n = None
mean = None
M2 = None
errp = np.ones(1)
while np.any(alpha < errp):
infomsg("errp =", errp, ": Attempting", probes, "more probes")
i = 0
while i < probes:
for x in f():
if log is not None:
infomsg(*x, file=log)
yield x
x = np.array(x)
if n is None:
n = np.zeros(len(x))
mean = np.zeros(len(x))
M2 = np.zeros(len(x))
errp = np.zeros(len(x))
if len(errp) < len(x):
infomsg("warning: change in fitness dimnsions!")
return
if np.all(x == 0):
infomsg("warning: 0 fitness: terminating evaluation")
return
n += 1
delta = x - mean
mean += delta / n
M2 += delta * (x - mean)
i = i + 1
var = M2 / (n - 1)
errp = np.sqrt(var / n) / mean
if np.any(n < 100):
c4 = np.sqrt(2 / (n - 1)) * gamma(n / 2) / gamma((n - 1) / 2)
errp = errp / c4
probes += probes
if log is not None:
for j in range(len(n)):
infomsg(n[j], "probes: relative standard error =", errp[j], file=log)
debug_file = log
def check_cmd( cmd ):
try:
status = call( cmd )
except OSError as e:
infomsg( e.strerror )
exit( 2 )
if status != 0:
if not isinstance( cmd, str ):
cmd = " ".join( cmd )
infomsg( "ERROR:", cmd, file = sys.stderr )
exit( status )
def parse_log( infile, outfile ):
infomsg( "INFO: parsing", infile )
cmd = [
parselog, infile,
"--filter", options.filter,
"--final",
"--no-confidence",
"--csv", outfile
]
if options.stop_after is not None:
cmd += [ "--stop-after", str( options.stop_after ) ]
check_call( cmd )
def __init__( self, genprog, builder, deltas ):
DD.__init__( self )
self.builder = builder
self.genprog = genprog
if options.disable_cache:
self.cache_outcomes = 0
self.duration = 0
infomsg( "INFO: computing optimized energy usage" )
self.optimized = np.array( self.get_fitness( deltas ) )
self.mean = np.mean( self.optimized, axis = 0 )
assert np.all( self.mean > 0 ), "'optimized' variant has 0 fitness!"
def get_builder( deltas ):
if options.sources is None:
if not options.compound_edits:
try:
deltas = lower_genome( deltas )
except ValueError as e:
infomsg( "ERROR:", e.message, file = sys.stderr )
exit( 1 )
deltas = list( enumerate( deltas ) )
builder = GenomeBuilder( genprog )
else:
print >>sys.stderr, "file-based differences not implemented yet"
exit( 2 )
return deltas, builder
def align_functions( f1, f2, name ):
aligned = list()
i, j = 0, 0
failure = None
while i < len( f1 ) and j < len( f2 ):
d1, s1 = f1[ i ]
d2, s2 = f2[ j ]
if instreq( s1, s2 ):
aligned.append( ( d1, d2, s1, s2 ) )
i += 1
j += 1
elif isnop( s1 ):
i += 1
elif isnop( s2 ):
j += 1
else:
failure = MisalignedCodeError(
"unexpected instruction pair:", f1[ i ], f2[ j ]
)
infomsg( "ERROR: unexpected instruction pair:" )
infomsg( " ", i, ":", f1[ i ] )
infomsg( " ", j, ":", f2[ j ] )
infomsg( "in function", name )
i+=1
j+=1
if failure is not None:
raise failure
for k, f in [ ( i, f1 ), ( j, f2 ) ]:
while k < len( f ):
if not isnop( f[ k ][ 1 ] ):
raise MisalignedCodeError(
"unexpected final instruction:", f[ k ][ 1 ]
)
k += 1
return aligned
def calcReplayGain(cls, pieces):
command = ['wvgain', '-a' ]
command.extend(pieces)
infomsg( "calculating replaygain info for wavpack files...")
exitcode = subprocess.call( command,
shell=False,
stdout=subprocess.PIPE,
)
if exitcode != 0:
raise ReplayGainError( "fail to calulate replaygain for wavpack files. ")
def get_localized_results( localfile, index ):
storage = get_storage_dir( index )
fitness_log = "fitness%s.csv" % index
cfg = Config()
cfg.load( config )
run_search = False
if os.path.isdir( storage ):
if not os.path.exists( os.path.join( storage, fitness_log ) ):
infomsg( "no such file:", os.path.join( storage, fitness_log ) )
run_search = True
elif not os.path.exists( fitness_log ):
infomsg( "no such file:", fitness_log )
run_search = True
if run_search:
cmd = [
genprog, config,
"--max-evals", str( options.max_evals ),
"--fitness-log", fitness_log
]
if localfile is not None:
cmd += [
"--fix-scheme", "line",
"--fix-file", localfile,
"--fault-scheme", "line",
"--fault-file", localfile,
]
record.time( "GenProg", check_call, cmd )
def get_repair_log( d ):
if "--seed" in cfg:
return os.path.join( d, "repair.debug.%s" % cfg[ "--seed" ] )
logs = glob( os.path.join( d, "repair.debug.*" ) )
if len( logs ) == 1:
return logs[ 0 ]
infomsg( "ERROR: could not determine repair log", file = sys.stderr )
exit( 2 )
if not os.path.isdir( storage ):
repairlog = get_repair_log( "." )
os.makedirs( storage )
sources = [ fitness_log, "multi.cache", "repair.cache", repairlog ]
check_call( [ "mv" ] + sources + [ storage ] )
repairlog = os.path.join( storage, repairlog )
else:
repairlog = get_repair_log( storage )
return repairlog
def read_emon( device, baud, columns, num_samples ):
with Emon( device, baud ) as emon:
emon.set( "calibrate", 1 )
emon.set( "period", 1000 )
samples = list()
for i in range( num_samples ):
row = next( emon )
infomsg( row )
selected = list()
for col in columns:
selected.append( row[ col ] )
samples.append( selected )
mean = np.mean( samples, 0 )
err = 1.96 * np.std( samples, 0 ) / np.sqrt( len( samples ) )
return mean, err
def rename_by_taginfo(self, scheme=default_scheme):
taginfo = self.extract_taginfo()
filename = "%s%s" % ( eval_scheme(scheme, taginfo),
self.extension,
)
filename_good = normalize_filename(filename)
infomsg ("filename_good: %s => %s" % (self.filename, filename_good))
os.rename(self.filename, filename_good)
self.filename = filename_good
self.basename = os.path.splitext(filename_good)[0]
def get_line_coverage( metrics, asmfuns ):
for fname in asmfiles:
for ( fname, fun ), instrs in assembly_funs( fname ):
if not fun in metrics:
infomsg(
"Warning: no annotations for %s:%s" % ( fname, fun ),
file = sys.stderr
)
continue
try:
lines = list( align_functions( instrs, metrics[ fun ], fun ) )
except MisalignedCodeError as e:
infomsg( "ERROR:", e )
continue
for line, cvg, _, _ in lines:
yield fname, line, cvg
def build_variant( self, genome ):
if len( genome ) == 0:
genome = [ "original" ]
seed = self.config.get( "--config", "0" )
with mktemp() as genome_file:
with open( genome_file, 'w' ) as fh:
print( "--oracle-genome", " ".join( genome ), file = fh )
cmd = [
self.genprog, self.configfile, genome_file,
"--seed", seed,
"--keep-source",
"--no-test-cache",
"--search", "oracle",
"--test-command", "true",
]
keepfiles = [ "000000", "repair.debug." + seed ]
tmpdir = tempfile.mkdtemp( dir = "." )
try:
for fname in keepfiles:
if os.path.exists( fname ):
os.rename( fname, os.path.join( tmpdir, fname ) )
with mktemp() as log:
try:
with open( log, 'w' ) as fh:
check_call( cmd, stdout = fh, stderr = fh )
except CalledProcessError:
with open( log ) as fh:
infomsg( fh.read(), file = sys.stderr )
raise
if os.path.exists( genome_file ):
os.remove( genome_file )
if os.path.exists( "000000/000000" ):
yield "000000/000000"
else:
yield None
finally:
for fname in keepfiles:
if os.path.exists( fname ):
check_call( [ "rm", "-rf", fname ] )
if os.path.exists( os.path.join( tmpdir, fname ) ):
os.rename( os.path.join( tmpdir, fname ), fname )
check_call( [ "rm", "-rf", tmpdir ] )
def calcReplayGain(cls, pieces):
command = ['metaflac', '--add-replay-gain' ]
command.extend(pieces)
return
infomsg( "calculating replaygain info for flac files...")
exitcode = subprocess.call( command,
shell=False,
stdin=subprocess.PIPE,
stdout=subprocess.PIPE
)
if exitcode != 0:
raise ReplayGainError( "fail to calulate replaygain for flac files. ")
def energy_future():
energy = [ None ] * len( options.plane )
start = list()
for plane in options.plane:
fname = planes[ plane ][ 0 ]
with open( fname ) as fh:
infomsg( "reading from plane:", fname )
start.append( int( next( fh ) ) )
try:
yield energy
finally:
for i, plane in enumerate( options.plane ):
fname = planes[ plane ][ 0 ]
with open( fname ) as fh:
delta = int( next( fh ) ) - start[ i ]
if delta < 0:
delta += planes[ plane ][ 1 ]
energy[ i ] = delta * 1e-6
def _test( self, deltas ):
global cache
genome = map( lambda (_, y): y, deltas )
key = " ".join( genome )
if key in cache:
return cache[ key ]
if len( genome ) == 0:
infomsg( "INFO: genome: original" )
else:
infomsg( "INFO: genome:", *genome )
build = self.genprog.build_variant
with swallow( CalledProcessError, build, genome ) as exe:
if exe is None:
fitness = "compile error"
if len( deltas ) == self.num_edits:
result = self.FAIL
else:
result = self.UNRESOLVED
else:
try:
result = self.PASS
for fitness in self.genprog.run_test( exe ):
if any( [ f == 0 for f in fitness ] ):
result = self.FAIL
except CalledProcessError:
fitness = "test failure"
result = self.FAIL
infomsg( " ", fitness, "=", result )
if not options.disable_cache:
cache[ key ] = result
return result
def split (self, cuefile, cmd_args ):
format = cmd_args.format if cmd_args.format else "flac"
scheme = cmd_args.scheme if cmd_args.scheme else default_scheme
target = getLossLessAudio("xyz." + format)
self.check_decodable()
target.check_encodable()
try :
cuesheet = parsecuefile( cuefile)
except NoCuedataError as e:
# when no cuefile is not available
infomsg( e.message)
infomsg("trying embeded cuesheet...")
cuedata = self.embeded_cuedata()
if not cuedata :
raise NoCuedataError("%s does not contain embeded cuedata."
% self.filename )
cuesheet = parsecuedata( conv2unicode(cuedata) )
infomsg( "splitting audio chunk: %s..." % self.filename)
pieces = shnsplit(self.filename, cuesheet.breakpoints(), format)
target.tag_pieces(pieces, cuesheet)
target.calcReplayGain(pieces)
target.rename_pieces(pieces, scheme)
def get_localization( cmd, asmfiles, repeat = 1, smooth = None ):
asmfuns = defaultdict( dict )
for fname in asmfiles:
for ( fname, fun ), instrs in assembly_funs( fname ):
asmfuns[ fun ][ fname ] = instrs
accum = defaultdict( lambda: 0.0 )
stdout = sys.stdout
stderr = sys.stderr
with open( "/dev/null", 'w' ) as fh:
for i in range( repeat ):
if i > 0:
infomsg( "profiling run", i + 1 )
metrics = profile( args, stdout = stdout, stderr = stderr )
for fname, line, cvg in get_line_coverage( metrics, asmfuns, smooth ):
accum[ fname, line ] += cvg
stdout = fh
stderr = fh
localization = list()
for fname, line in sorted( accum ):
localization.append( ( fname, line, accum[ fname, line ] / repeat ) )
return localization
def build( self, genome ):
if len( genome ) == 0:
infomsg( "INFO: genome: original" )
else:
infomsg( "INFO: genome:", *get_genes( genome ) )
try:
with self.genprog.build_variant( get_genes( genome ) ) as exe:
yield exe
except CalledProcessError as e:
infomsg( "ERROR:", e )
yield None
def get_minimized_fitness( genome ):
infomsg( "getting minimized genome for", genome )
if not os.path.exists( genome + ".min" ):
record.time( "minimize.py", check_call, [
os.path.join( root, "bin", "minimize.py" ), genprog, config,
"--genome-file", genome,
"--cache", genome + ".cache",
"--save-binary", genome + ".bin",
"--save-source", genome + ".src",
"--save-genome", genome + ".min"
] )
if options.regenerate is not None:
cfg = Config()
cfg.load( config )
cmd = cfg[ "--test-command" ]
cmd = cmd.replace( "__EXE_NAME__", genome + ".bin" )
if options.regenerate == "wall":
cmd += " -j 1 --wall --repeat 100 --no-limit"
with record.context( "fitness eval" ):
fitnesses = list()
for i in range( options.rows ):
with mktemp() as fitnessfile:
# use call instead of check_call because test scripts always
# return non-zero status
tmp = cmd.replace( "__FITNESS_FILE__", fitnessfile )
call( [ "sh", "-c", tmp ] )
with open( fitnessfile ) as fh:
for line in fh:
value = float( line.split()[ 0 ] )
infomsg( " ", value )
fitnesses.append( value )
break
else:
infomsg(
"ERROR: no fitness for", genome + ".bin",
file = sys.stderr
)
exit( 2 )
return fitnesses
else:
with open( genome + ".min" ) as fh:
key = " ".join( fh.readlines() )
key = " ".join( key.split() )
d = shelve.open( genome + ".cache" )
return d[ key ]
def run(self, root, argv=sys.argv):
parser = self.getParser()
self.addCommonOptions(parser)
self.options, args = parser.parse_args(args=argv[1:])
try:
self.checkArgs(parser, args)
except ValueError:
return 0 if len(args) == 0 else 1
if find_executable(self.exe) is None:
raise ValueError("%s: command not found" % self.exe)
self.exe = find_executable(self.exe)
if "/" not in self.exe:
self.exe = os.path.join(".", self.exe)
metrics = list()
if self.options.emon is not None:
metrics.append(EmonMetric(root, self.options.emon))
if self.options.model:
metrics.append(ModelMetric(root))
if len(self.options.rapl) > 0:
metrics.append(RaplMetric(root, self.options.rapl))
if self.options.time:
metrics.append(TimeMetric(root))
if self.options.wu is not None:
metrics.append(WuMetric(root, self.options.wu))
try:
if self.options.cpu is None and all([metric.inParallel() for metric in metrics]):
fitness = self.getParallelFitness(root, metrics)
else:
jobs = self.options.jobs
self.options.jobs = 1
fitness = list()
for i in range(jobs):
result = self.getParallelFitness(root, metrics)
while len(fitness) < len(result):
fitness.append(list())
for dim, values in zip(fitness, result):
dim.extend(values)
self.options.jobs = jobs
except IOError as e:
exit(e.errno)
results = [list() for i in range(self.options.jobs)]
for dim in fitness:
if len(dim) != self.options.jobs:
results = [0]
break
for i, x in enumerate(dim):
results[i].append(x)
if self.options.csv is not None:
with open(self.options.csv, "a") as fh:
writer = csv.writer(fh)
for row in results:
writer.writerow(["%g" % y for y in row])
with open(self.fitnessfile, "w") as fh:
for row in results:
infomsg(*["%g" % y for y in row], file=fh)