def tune_kernels(tasks, measure_option, tuner="gridsearch", early_stopping=None, log_filename="tuning.log", mode=0):
for i, tsk in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
name = "topi_nn_conv2d" if mode == 0 else "topi_x86_conv2d_NCHWc"
print(name)
task = autotvm.task.create(name, args=tsk.args, target=target, template_key="direct")
task.workload = tsk.workload
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(task, pop_size=50)
elif tuner == 'random':
tuner_obj = RandomTuner(task)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
n_trial = 1000 # min(len(task.config_space), 1000)
print("trials=", n_trial)
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial, prefix=prefix),
autotvm.callback.log_to_file(log_filename)])
def tune_kernels(
tasks, measure_option, tuner="gridsearch", early_stopping=None, log_filename="logs/tuning.log"
):
for i, task in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# create tuner
if tunes == 0:
tuner_obj = XGBTuner(task, loss_type="rank")
# print("Using XGBTuner")
elif tunes == 1:
tuner_obj = GATuner(task, pop_size=50)
# print("Using GATuner")
elif tunes == 2:
tuner_obj = RandomTuner(task)
# print("Using Random")
elif tunes == 3:
tuner_obj = GridSearchTuner(task)
# print("Using GridSearch")
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
n_trial = len(task.config_space)
tuner_obj.tune(
n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial, prefix=prefix),
autotvm.callback.log_to_file(log_filename),
],
)
def tune_kernels(tasks,
measure_option,
tuner="xgb",
n_trial=1000,
early_stopping=None,
log_filename="tuning.log"):
for i, tsk in enumerate(reversed(tasks)):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# create tuner
if tuner == "xgb" or tuner == "xgb-rank":
tuner_obj = XGBTuner(tsk, loss_type="rank")
elif tuner == "ga":
tuner_obj = GATuner(tsk, pop_size=100)
elif tuner == "random":
tuner_obj = RandomTuner(tsk)
elif tuner == "gridsearch":
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
tsk_trial = min(n_trial, len(tsk.config_space))
tuner_obj.tune(
n_trial=tsk_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(tsk_trial, prefix=prefix),
autotvm.callback.log_to_file(log_filename),
],
)
def tune_kernels(tasks,
measure_option,
tuner='gridsearch',
early_stopping=None,
log_filename='tuning.log'):
for i, task in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i+1, len(tasks))
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(task, pop_size=50)
elif tuner == 'random':
tuner_obj = RandomTuner(task)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
n_trial=len(task.config_space)
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial, prefix=prefix),
autotvm.callback.log_to_file(log_filename)])
def tune_tasks(tasks,
measure_option,
tuner='xgb',
n_trial=1000,
early_stopping=None,
log_filename='tuning.log',
use_transfer_learning=True,
try_winograd=True):
if try_winograd:
for i in range(len(tasks)):
try: # try winograd template
tsk = autotvm.task.create(tasks[i].name, tasks[i].args,
tasks[i].target,
tasks[i].target_host, 'winograd')
input_channel = tsk.workload[1][1]
if input_channel >= 64:
tasks[i] = tsk
except Exception:
pass
# create tmp log file
tmp_log_file = log_filename + ".tmp"
if os.path.exists(tmp_log_file):
os.remove(tmp_log_file)
for i, tsk in enumerate(reversed(tasks)):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(tsk, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(tsk, pop_size=100)
elif tuner == 'random':
tuner_obj = RandomTuner(tsk)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
if use_transfer_learning:
if os.path.isfile(tmp_log_file):
tuner_obj.load_history(
autotvm.record.load_from_file(tmp_log_file))
# do tuning
tuner_obj.tune(n_trial=min(n_trial, len(tsk.config_space)),
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(tmp_log_file)
])
if not os.path.exists(tmp_log_file):
open(tmp_log_file, 'a')
# pick best records to a cache file
autotvm.record.pick_best(tmp_log_file, log_filename)
os.remove(tmp_log_file)
def tune_kernels(tasks,
builder=autotvm.LocalBuilder(),
runner=autotvm.LocalRunner(number=10,
repeat=1,
min_repeat_ms=1000),
tuner='ga',
early_stopping=None,
log_filename=log_file):
measure_option = autotvm.measure_option(builder, runner)
for i, tsk in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# converting conv2d tasks to conv2d_NCHWc tasks
if tsk.workload:
op_name = tsk.workload[0]
if op_name == 'conv2d':
func_create = 'topi_x86_conv2d_NCHWc'
elif op_name == 'depthwise_conv2d_nchw':
func_create = 'topi_x86_depthwise_conv2d_NCHWc_from_nchw'
else:
raise ValueError(
"Tuning {} is not supported on x86".format(op_name))
task = autotvm.task.create(func_create,
args=tsk.args,
target=target,
template_key='direct')
task.workload = tsk.workload
else:
task = tsk
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(task, pop_size=1000)
elif tuner == 'random':
tuner_obj = RandomTuner(task)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
n_trial = len(task.config_space)
print("n_trial", n_trial)
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(log_filename)
])
def tune_tasks(
tasks,
measure_option,
tuner="xgb",
n_trial=1000,
early_stopping=None,
log_filename="tuning.log",
use_transfer_learning=True,
):
# create tmp log file
tmp_log_file = log_filename + ".tmp"
if os.path.exists(tmp_log_file):
os.remove(tmp_log_file)
for i, tsk in enumerate(reversed(tasks)):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# create tuner
if tuner == "xgb" or tuner == "xgb-rank":
tuner_obj = XGBTuner(tsk, loss_type="rank")
elif tuner == "xgb_knob":
tuner_obj = XGBTuner(tsk, loss_type="rank", feature_type="knob")
elif tuner == "xgb_itervar":
tuner_obj = XGBTuner(tsk, loss_type="rank", feature_type="itervar")
elif tuner == "xgb_curve":
tuner_obj = XGBTuner(tsk, loss_type="rank", feature_type="curve")
elif tuner == "ga":
tuner_obj = GATuner(tsk, pop_size=50)
elif tuner == "random":
tuner_obj = RandomTuner(tsk)
elif tuner == "gridsearch":
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
if use_transfer_learning:
if os.path.isfile(tmp_log_file):
tuner_obj.load_history(
autotvm.record.load_from_file(tmp_log_file))
# process tuning
tsk_trial = min(n_trial, len(tsk.config_space))
tuner_obj.tune(
n_trial=tsk_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(tsk_trial, prefix=prefix),
autotvm.callback.log_to_file(tmp_log_file),
],
)
# pick best records to a cache file
autotvm.record.pick_best(tmp_log_file, log_filename)
os.remove(tmp_log_file)
def tune_tasks(
tasks,
measure_option,
tuner="xgb",
n_trial=1000,
early_stopping=None,
log_filename="tuning.log",
use_transfer_learning=True,
):
"""Tune tasks and generate tuning log to file"""
# create tmp log file
tmp_log_file = log_filename + ".tmp"
if os.path.exists(tmp_log_file):
os.remove(tmp_log_file)
for i, tsk in enumerate(reversed(tasks)):
prefix = f"[Task {i + 1:2d}/{len(tasks):2d}] "
# create tuner
if tuner in ("xgb", "sgb-rank"):
tuner_obj = XGBTuner(tsk, loss_type="rank")
elif tuner == "ga":
tuner_obj = GATuner(tsk, pop_size=100)
elif tuner == "random":
tuner_obj = RandomTuner(tsk)
elif tuner == "gridsearch":
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
if use_transfer_learning:
if os.path.isfile(tmp_log_file):
tuner_obj.load_history(
autotvm.record.load_from_file(tmp_log_file))
# do tuning
tsk_trial = min(n_trial, len(tsk.config_space))
tuner_obj.tune(
n_trial=tsk_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(tsk_trial, prefix=prefix),
autotvm.callback.log_to_file(tmp_log_file),
],
)
# pick best records to a cache file
if not os.path.exists(log_filename):
with open(log_filename, "w", encoding="utf-8"):
pass
if os.path.exists(tmp_log_file):
autotvm.record.pick_best(tmp_log_file, log_filename)
os.remove(tmp_log_file)
def tune_tasks(tasks,
measure_option,
tuner='xgb',
n_trial=1000,
early_stopping=None,
log_filename='tuning.log',
use_transfer_learning=True):
# create tmp log file
tmp_log_file = log_filename + ".tmp"
if os.path.exists(tmp_log_file):
os.remove(tmp_log_file)
for i, tsk in enumerate(reversed(tasks)):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(tsk, loss_type='rank', feature_type='curve')
elif tuner == 'xgb_knob':
tuner_obj = XGBTuner(tsk, loss_type='rank', feature_type='curve')
elif tuner == 'ga':
tuner_obj = GATuner(tsk, pop_size=50)
elif tuner == 'random':
tuner_obj = RandomTuner(tsk)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
if use_transfer_learning:
print("use_transfer_learning...")
if os.path.isfile(tmp_log_file):
print("from file " + tmp_log_file)
tuner_obj.load_history(
autotvm.record.load_from_file(tmp_log_file))
# do tuning
#print("INFO: len(tsk.config_space) = " + str(len(tsk.config_space)))
n_trial = min(n_trial, len(tsk.config_space))
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(tmp_log_file)
])
# pick best records to a cache file
autotvm.record.pick_best(tmp_log_file, log_filename)
def tune_kernels(tasks,
measure_option,
tuner=tuner,
n_trial=trials,
early_stopping=None,
log_filename=log_file):
for i, tsk in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# Converting conv2d tasks to conv2d_NCHWc tasks. Do we actually want this?
op_name = tsk.workload[0]
#input_shape = tsk.workload[1][0:-1]
#kernel_shape = tsk.workload[2][0:-1]
#input_channels = input_shape[1]
# Only can convert to NCHWc if input channels is divisible by 8.
#convertible = input_channels % 8 == 0
func_create = tsk.name
#if op_name == 'conv2d':
# func_create = 'topi_x86_conv2d_NCHWc'
#elif op_name == 'depthwise_conv2d_nchw':
# func_create = 'topi_x86_depthwise_conv2d_NCHWc_from_nchw'
print(func_create)
task = autotvm.task.create(func_create,
args=tsk.args,
target=target,
template_key='direct')
task.workload = tsk.workload
# Create tuner.
if tuner == 'xgb' or tuner == 'xbg-rank':
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(task, pop_size=50)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tunder: " + tuner)
# Do tuning.
n_trial = min(n_trial, len(task.config_space))
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(log_filename)
])
def tune_kernels(tasks,
measure_option,
tuner='gridsearch',
early_stopping=None,
log_filename='tuning.log'):
for i, tsk in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# converting conv2d tasks to conv2d_NCHWc tasks
op_name = tsk.workload[0]
if op_name == 'conv2d':
func_create = 'topi_x86_conv2d_NCHWc'
elif op_name == 'depthwise_conv2d_nchw':
func_create = 'topi_x86_depthwise_conv2d_NCHWc_from_nchw'
else:
raise ValueError(
"Tuning {} is not supported on x86".format(op_name))
task = autotvm.task.create(func_create,
args=tsk.args,
target=target,
template_key='direct')
task.workload = tsk.workload
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(task, pop_size=50)
elif tuner == 'random':
tuner_obj = RandomTuner(task)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
n_trial = len(task.config_space)
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(log_filename)
])
with open('data/task%i.pickle' % i, 'wb') as output:
pickle.dump([task, tuner_obj.cost_model.saved_features], output,
pickle.HIGHEST_PROTOCOL)
print('Saved data/task%i.pickle' % i)
def tune_kernels(tasks,
measure_option,
tuner='gridsearch',
early_stopping=None,
log_filename='tuning.log'):
print('tasks: {}'.format(tasks))
for i, tsk in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# converting conv2d tasks to conv2d_NCHWc tasks
op_name = tsk.workload[0]
if op_name == 'conv2d':
func_create = 'topi_x86_conv2d_NCHWc'
elif op_name == 'depthwise_conv2d_nchw':
func_create = 'topi_x86_depthwise_conv2d_NCHWc_from_nchw'
else:
raise ValueError(
"Tuning {} is not supported on x86".format(op_name))
task = autotvm.task.create(func_create,
args=tsk.args,
target=target,
template_key='direct')
task.workload = tsk.workload
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(task, pop_size=50)
elif tuner == 'random':
tuner_obj = RandomTuner(task)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
n_trial = len(task.config_space)
n_trial = 1 # set small num to quick test
print('n trial {}'.format(n_trial))
# print(task.config_space)
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(log_filename)
])
def tune_kernels(tasks,
measure_option,
tuner='gridsearch',
early_stopping=None,
log_filename='tuning.log'):
for i, tsk in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# converting conv2d tasks to conv2d_NCHWc tasks
# data, kernel are tuples of ("TENSOR", shape, dtype)
data, kernel, strides, padding, layout, dtype = tsk.args
kernel_size = (kernel[1][2], kernel[1][3])
data_plc = tvm.placeholder(data[1], name="data")
kernel_plc = tvm.placeholder(kernel[1], name="kernel")
args = [
data_plc, kernel_plc, kernel[1][0], kernel_size, strides, padding,
layout, layout, dtype
]
args = autotvm.task.nnvm_integration.serialize_args(args)
task = autotvm.task.create("topi_x86_conv2d_NCHWc",
args=args,
target=target)
task.workload = conv_NCHWc_arg_to_workload(data_plc, kernel_plc,
kernel_size, strides,
padding, layout, layout,
dtype)
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(task, pop_size=50)
elif tuner == 'random':
tuner_obj = RandomTuner(task)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
n_trial = len(task.config_space)
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(log_filename)
])
def tune_kernels(tasks,
measure_option,
tuner="gridsearch",
early_stopping=None,
log_filename="tuning.log"):
for i, tsk in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# converting conv2d tasks to conv2d_NCHWc tasks
op_name = tsk.workload[0]
if op_name == "conv2d":
func_create = "topi_x86_conv2d_NCHWc"
elif op_name == "depthwise_conv2d_nchw":
func_create = "topi_x86_depthwise_conv2d_NCHWc_from_nchw"
else:
raise ValueError(
"Tuning {} is not supported on x86".format(op_name))
task = autotvm.task.create(func_create,
args=tsk.args,
target=target,
template_key="direct")
task.workload = tsk.workload
# create tuner
if tuner == "xgb" or tuner == "xgb-rank":
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == "ga":
tuner_obj = GATuner(task, pop_size=50)
elif tuner == "random":
tuner_obj = RandomTuner(task)
elif tuner == "gridsearch":
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
n_trial = len(task.config_space)
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(log_filename)
])
def tune_tasks(target):
target = tvm.target.create(target)
tasks = config_tasks(target)
tuner = tuning_option['tuner']
use_transfer_learning = tuning_option['use_transfer_learning']
n_trial = tuning_option['n_trial']
early_stopping = tuning_option['early_stopping']
measure_option = tuning_option['measure_option']
tmp_log_file = log_file + ".tmp"
if os.path.exists(tmp_log_file):
os.remove(tmp_log_file)
for i, tsk in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(tsk, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(tsk, pop_size=50)
elif tuner == 'random':
tuner_obj = RandomTuner(tsk)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
if use_transfer_learning:
if os.path.isfile(tmp_log_file):
tuner_obj.load_history(
autotvm.record.load_from_file(tmp_log_file))
# Do tuning
tuner_obj.tune(n_trial=min(n_trial, len(tsk.config_space)),
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(tmp_log_file)
])
# Pick best records to a cache file
autotvm.record.pick_best(tmp_log_file, log_file)
os.remove(tmp_log_file)
def tune_tasks(tasks,
measure_option,
tuner='gridsearch',
early_stopping=None,
log_filename='tuning.log'):
tmp_log_file = log_filename + ".tmp"
if os.path.exists(tmp_log_file):
os.remove(tmp_log_file)
for i, task in enumerate(tasks):
print(task)
prefix = "[Task %2d/%2d] " % (i+1, len(tasks))
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == 'xgb_knob':
tuner_obj = XGBTuner(tsk, loss_type='rank', feature_type='knob')
elif tuner == 'ga':
tuner_obj = GATuner(task, pop_size=50)
elif tuner == 'random':
tuner_obj = RandomTuner(task)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
#n_trial=len(task.config_space)
n_trial = 2
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial, prefix=prefix),
autotvm.callback.log_to_file(tmp_log_file)])
# pick best records to a cache file
autotvm.record.pick_best(tmp_log_file, log_filename)
os.remove(tmp_log_file)
def tune(self):
# create tuner
if self.tuner == "xgb" or self.tuner == "xgb-rank":
tuner_obj = XGBTuner(self.task, loss_type="rank")
elif self.tuner == "xgb_knob":
tuner_obj = XGBTuner(selftask, loss_type="rank", feature_type="knob")
elif self.tuner == "ga":
tuner_obj = GATuner(self.task, pop_size=50)
elif self.tuner == "random":
tuner_obj = RandomTuner(self.task)
elif self.tuner == "gridsearch":
tuner_obj = GridSearchTuner(self.task)
else:
raise ValueError("Invalid tuner: " + tuner)
def _search_best_config():
def _callback(_, inputs, results):
for inp, result in zip(inputs, results):
new_latency = result.costs[0] if result.error_no ==0 else 1e9
if self.record == None or self.best_latency > new_latency:
self.record = autotvm.record.encode(inp, result)
self.best_config = inp.config.to_json_dict()['entity']
self.best_latency = new_latency
return _callback
# do tuning
task_trial = min(self.n_trial, len(self.task.config_space))
tuner_obj.tune(
n_trial=task_trial,
early_stopping=self.early_stopping,
measure_option=self.measure_option,
callbacks=[
autotvm.callback.progress_bar(task_trial),
_search_best_config(),
],
)
print(self.record)
kernel_db = KernelDB()
kernel_db.write_task(self)
del kernel_db
def tune_kernels(
tasks,
measure_option,
tuner,
n_trial,
early_stopping,
log_filename,
use_transfer_learning,
):
for i, tsk in enumerate(reversed(tasks)):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# create tuner
if tuner == "random" or n_trial >= len(tsk.config_space):
tuner_obj = RandomTuner(tsk)
elif tuner == "xgb" or tuner == "xgb-rank":
tuner_obj = XGBTuner(tsk, loss_type="rank")
# use history data to pre-train the cost model
if use_transfer_learning:
if os.path.isfile(log_filename):
tuner_obj.load_history(
autotvm.record.load_from_file(log_filename))
elif tuner == "ga":
tuner_obj = GATuner(tsk, pop_size=100)
elif tuner == "gridsearch":
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
tsk_trial = min(n_trial, len(tsk.config_space))
tuner_obj.tune(
n_trial=tsk_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(tsk_trial, prefix=prefix),
autotvm.callback.log_to_file(log_filename),
],
)
def tune_kernels(tasks,
measure_option,
tuner='gridsearch',
early_stopping=None,
log_filename='tuning.log'):
for i, tsk in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# converting conv2d tasks to conv2d_NCHWc tasks
task = autotvm.task.create("topi_x86_conv2d_NCHWc",
args=tsk.args,
target=target,
template_key='direct')
task.workload = tsk.workload
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(task, pop_size=50)
elif tuner == 'random':
tuner_obj = RandomTuner(task)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
n_trial = len(task.config_space)
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(log_filename)
])
def tune_tasks(tasks,
measure_option,
tuner='xgb',
n_trial=500,
early_stopping=None,
log_filename='tuning.log',
use_transfer_learning=True,
try_winograd=True):
for i in range(len(tasks)):
print(tasks[i].args)
data, kernel, padding, stride, temp, layout, dtype = tasks[i].args
new_args = (data, kernel, padding, stride, temp, layout, dtype)
block_factor = 4
CO, CI, KH, KW = kernel[1]
if CO % block_factor == 0 and CI % block_factor == 0:
new_task = autotvm.task.create(tasks[i].name, new_args,
tasks[i].target,
tasks[i].target_host, 'fp16')
tasks[i] = new_task
if args.pretuned is not None:
pretuned_ctx = autotvm.apply_history_best(args.pretuned)
_tasks = []
for task in tasks:
if pretuned_ctx._query_inside(target, task.workload) is None:
_tasks.append(task)
else:
print('Ignoring {}'.format(task))
tasks = _tasks
print(tasks)
# create tmp log file
tmp_log_file = log_filename + ".tmp"
if os.path.exists(tmp_log_file):
os.remove(tmp_log_file)
for i, tsk in enumerate(reversed(tasks)):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(tsk, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(tsk, pop_size=100)
elif tuner == 'random':
tuner_obj = RandomTuner(tsk)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
if use_transfer_learning:
if os.path.isfile(tmp_log_file):
tuner_obj.load_history(
autotvm.record.load_from_file(tmp_log_file))
# do tuning
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(tmp_log_file)
])
# pick best records to a cache file
autotvm.record.pick_best(tmp_log_file, log_filename)
os.remove(tmp_log_file)
def tune_tasks(tasks,
measure_option,
tuner='xgb',
n_trial=1000,
early_stopping=None,
log_filename='tuning.log',
use_transfer_learning=True):
# create tmp log file
tmp_log_file = log_filename + ".tmp"
#if os.path.exists(tmp_log_file):
# os.remove(tmp_log_file)
check_tmp_his = True
ind = 0
'''if check_tmp_his:
length = 0
if os.path.isfile(tmp_log_file):
lines = list(open(tmp_log_file))
length = len(lines)
else:
check_tmp_his = False'''
tasks = prune_old_tasks(tasks,log_filename)
for i, tsk in enumerate(reversed(tasks)):
prefix = "[Task %2d/%2d] " %(i+1, len(tasks))
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(tsk, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(tsk, pop_size=100)
elif tuner == 'random':
tuner_obj = RandomTuner(tsk)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
if use_transfer_learning:
if os.path.isfile(tmp_log_file):
tuner_obj.load_history(autotvm.record.load_from_file(tmp_log_file))
# do tuning
tsk_trial = min(n_trial, len(tsk.config_space))
'''print(ind)
if check_tmp_his:
if ind < length:
ret = autotvm.record.decode(lines[ind])
inp, _ = ret
if inp.task.workload == tsk.workload:
if (ind + tsk_trial - 1) < length:
ind = (ind + tsk_trial - 1)
ret_end = autotvm.record.decode(lines[ind])
inp_end , _= ret_end
if inp.task.workload == inp_end.task.workload:
ind = ind + 1
continue
else:
check_tmp_his = False
tuner_obj.load_history(autotvm.record.load_from_file(tmp_log_file))
elif (ind + tsk_trial - 1) == length:
ind = (ind + tsk_trial - 1)
ret_end = autotvm.record.decode(lines[ind])
inp_end , _= ret_end
if inp.task.workload == inp_end.task.workload:
ind = ind + 1
check_tmp_his = False
tuner_obj.load_history(autotvm.record.load_from_file(tmp_log_file))
continue
else:
check_tmp_his = False
tuner_obj.load_history(autotvm.record.load_from_file(tmp_log_file))
else:
cmtd = length - ind
ind = length - 1
ret_end = autotvm.record.decode(lines[ind])
inp_end , _= ret_end
if inp.task.workload == inp_end.task.workload:
ind = ind + 1
check_tmp_his = False
tuner_obj.load_history(autotvm.record.load_from_file(tmp_log_file))
tsk_trial = tsk_trial - cmtd
else:
check_tmp_his = False
tuner_obj.load_history(autotvm.record.load_from_file(tmp_log_file))'''
with tempfile.NamedTemporaryFile() as tmp_task_log_file:
tuner_obj.tune(n_trial=tsk_trial,early_stopping=early_stopping,measure_option=measure_option,
callbacks=[autotvm.callback.progress_bar(tsk_trial, prefix=prefix),autotvm.callback.log_to_file(tmp_log_file)])
with open(tmp_log_file, 'a') as tmp_log:
tmp_log.write(tmp_task_log_file.read().decode('utf8'))
# pick best records to a cache file
autotvm.record.pick_best(tmp_log_file, log_filename)
os.remove(tmp_log_file)
def tvm_generic(N,
H,
W,
C,
kernel_size,
K,
stride=1,
padding=0,
dilation=1,
groups=1,
number=100,
dev=0,
timeout=4,
target="llvm",
trials=100):
data_shape = (N, C, H, W)
data = relay.var("data", shape=data_shape, dtype="float32")
kernel_size = (kernel_size, kernel_size)
stride = (stride, stride)
padding = (padding, padding)
body = layers.conv2d(data=data,
channels=K,
kernel_size=kernel_size,
strides=stride,
padding=padding,
name="conv2d")
op = relay.Function(relay.ir_pass.free_vars(body), body)
sym, params = create_workload(op)
tasks = autotvm.task.extract_from_program(op,
target=target,
params=params,
ops=(relay.op.nn.conv2d, ))
tuning_option = {
"log_filename":
"tvm_baseline_{}.log".format(
(N, C, H, W, K, kernel_size, stride, padding, dilation, groups)),
"tuner":
"xgb",
"early_stopping":
30,
"measure_option":
autotvm.measure_option(
builder=autotvm.LocalBuilder(timeout=timeout),
runner=autotvm.LocalRunner(number=number,
repeat=1,
timeout=timeout,
min_repeat_ms=150),
# runner=autotvm.RPCRunner(
# '1080ti', # change the device key to your key
# '0.0.0.0', 9190,
# number=20, repeat=3, timeout=4, min_repeat_ms=150)
),
}
log_filename = tuning_option["log_filename"]
tuner = tuning_option["tuner"]
early_stopping = tuning_option["early_stopping"]
measure_option = tuning_option["measure_option"]
# only support one task
assert len(tasks) == 1
for i, task in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(task, pop_size=100)
elif tuner == 'random':
tuner_obj = RandomTuner(task)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
# do tuning
n_trial = trials
length = len(task.config_space)
print("config space length=", length)
# tuner_obj.tune(n_trial=min(n_trial, length),
# early_stopping=early_stopping,
# measure_option=measure_option,
# callbacks=[
# autotvm.callback.progress_bar(n_trial, prefix=prefix),
# autotvm.callback.log_to_file(log_filename)])
if not os.path.exists(log_filename):
raise RuntimeError(
"the log file {} doesn't exists".format(log_filename))
with autotvm.apply_history_best(log_filename):
with relay.build_config(opt_level=3):
graph, lib, params = relay.build_module.build(op,
target=target,
params=params)
ctx = tvm.device(str(target), 0)
data_tvm = tvm.nd.array(
(np.random.uniform(size=data_shape)).astype("float32"))
module = runtime.create(graph, lib, ctx)
module.set_input("data", data_tvm)
module.set_input(**params)
# evaluate
ftimer = module.module.time_evaluator("run",
ctx,
number=number,
repeat=1)
prof_res = np.array(ftimer().results) * 1e3
return prof_res
def tune_kernels(
tasks,
measure_top_n,
measure_option,
tuner="random",
early_stopping=None,
n_trial=5000,
log_filename="tuning.log",
):
"""Tune kernels with the ranking model."""
remeasure_option = None
if tuner == "round":
# Setup another measure option for final remeasurment.
remeasure_option = autotvm.measure_option(
builder=LocalBuilder(),
runner=measure_option["runner"].local_runner,
)
assert isinstance(measure_option["runner"], RankModelRunner)
for i, task in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
callbacks = []
if task.name not in measure_option["runner"].models:
print("%s %s not covered by cost models" % (prefix, task.name))
continue
# create tuner
if tuner == "round":
tuner_obj = RoundTuner(task, n_cfg=measure_top_n)
callbacks = [rank_progress(n_trial, prefix=prefix)
] # Use different callbacks.
else:
if tuner in ("xgb", "xgb-rank"):
tuner_obj = XGBTuner(task, loss_type="rank")
elif tuner == "ga":
tuner_obj = GATuner(task, pop_size=50)
elif tuner == "random":
tuner_obj = RandomTuner(task)
elif tuner == "gridsearch":
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
callbacks = [
autotvm.callback.progress_bar(n_trial, prefix=prefix),
autotvm.callback.log_to_file(log_filename),
]
tic = time.time()
# do tuning
tuner_obj.tune(
n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=callbacks,
)
# Round tuner needs an extra measurement step to get the real throughputs.
if tuner == "round":
top_cfgs = tuner_obj.get_top_rank_cfgs(1)
measure_batch = create_measure_batch(task, remeasure_option)
inputs = [
MeasureInput(task.target, task, config) for config in top_cfgs
]
sys.stderr.write("{} Measure Top {} Configs".format(
prefix, len(inputs)))
results = measure_batch(inputs)
best_idx, best_flops = max(
[(idx, i.task.flop / np.mean(r.costs) /
1e9 if r.error_no == 0 else 0)
for idx, (i, r) in enumerate(zip(inputs, results))],
key=lambda x: x[1],
)
sys.stderr.write(" | Best %.2f GFLOPS at Top %d | %.2fs\n" %
(best_flops, best_idx, time.time() - tic))
autotvm.callback.log_to_file(log_filename)(None, inputs, results)
def tune_tasks(tasks,
measure_option,
tuner='xgb',
n_trial=1000,
early_stopping=500,
log_filename='tuning.log',
use_transfer_learning=True,
try_winograd=True):
if try_winograd:
for i in range(len(tasks)):
try: # try winograd template
tsk = autotvm.task.create(tasks[i].name, tasks[i].args,
tasks[i].target,
tasks[i].target_host, 'winograd')
input_channel = tsk.workload[1][1]
if input_channel >= 64:
tasks[i] = tsk
except Exception:
pass
# create tmp log file
tmp_log_file = log_filename + ".tmp"
# if os.path.exists(tmp_log_file):
# os.remove(tmp_log_file)
for i, tsk in enumerate(reversed(tasks)):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(tsk, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(tsk, pop_size=100)
elif tuner == 'random':
tuner_obj = RandomTuner(tsk)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
if use_transfer_learning:
if os.path.isfile(tmp_log_file):
tuner_obj.load_history(
autotvm.record.load_from_file(tmp_log_file))
# do tuning
n_trial = 1000
n_trial = min(n_trial, len(tsk.config_space))
res = []
for i in range(len(tsk.config_space)):
res.append(tuner_obj.space.get(i + 1))
with open("res.txt", "w", encoding='utf-8') as f:
for line in res:
f.write(str(line) + '\n')
f.close()
#任务单独保存
# import shutil
# shutil.copyfile('res.txt', 'res'+prefix+'.txt')
early_stopping = len(tsk.config_space) // 2
print("n_trial=", n_trial)
print("early_stopping=", early_stopping)
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(tmp_log_file)
])
np.savetxt("feas" + prefix + ".txt",
tuner_obj.cost_model.feas,
fmt='%s',
delimiter=' ')
np.savetxt("x_train" + prefix + ".txt",
tuner_obj.cost_model.x_train,
fmt='%s',
delimiter=' ')
np.savetxt("y_train" + prefix + ".txt",
tuner_obj.cost_model.y_train,
fmt='%s',
delimiter=' ')
#他做到了同算子--》空间维度相同的迁移学习,那么实际上resnet18--->resnet50?
# pick best records to a cache file
autotvm.record.pick_best(tmp_log_file, log_filename)
def tune_tasks(mod, params, option):
print('Creating tasks ...')
tasks = autotvm.task.extract_from_program(mod["main"],
target=option['target'],
params=params,
ops=(relay.op.nn.conv2d,
relay.op.nn.dense))
if option['try_winograd']:
for i in range(len(tasks)):
try: # try winograd template
tsk = autotvm.task.create(tasks[i].name,
tasks[i].args,
tasks[i].target,
tasks[i].target_host, 'winograd')
input_channel = tsk.workload[1][1]
if input_channel >= 64:
tasks[i] = tsk
except Exception as err:
print(err)
# create tmp log file
tmp_log_file = option['log_file']
if os.path.exists(tmp_log_file): os.remove(tmp_log_file)
for i, tsk in enumerate(reversed(tasks)):
# converting conv2d tasks to conv2d_NCHWc tasks
# op_name = tsk.workload[0]
# if op_name == 'conv2d':
# func_create = 'topi_x86_conv2d_NCHWc'
# elif op_name == 'depthwise_conv2d_nchw':
# func_create = 'topi_x86_depthwise_conv2d_NCHWc_from_nchw'
# else:
# func_create = tasks[i].name
# task = autotvm.task.create(func_create, args=tsk.args,
# target=target, template_key='direct')
# task.workload = tsk.workload
# tsk = task
prefix = "[Task %2d/%2d] (%s)" %(i+1, len(tasks), tsk.name)
# create tuner
tuner = option['tuner']
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(tsk, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(tsk, pop_size=100)
elif tuner == 'random':
tuner_obj = RandomTuner(tsk)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
if option['use_transfer_learning']:
if os.path.isfile(tmp_log_file):
tuner_obj.load_history(autotvm.record.load_from_file(tmp_log_file))
# do tuning
n_trial = min(option['n_trial'], len(tsk.config_space))
tuner_obj.tune(n_trial=n_trial,
early_stopping=option['early_stopping'],
measure_option=option['measure_option'],
callbacks=[
autotvm.callback.progress_bar(n_trial, prefix=prefix),
autotvm.callback.log_to_file(tmp_log_file)])
if os.path.exists(option['log_best_file']):
os.remove(option['log_best_file'])
autotvm.record.pick_best(option['log_file'], option['log_best_file'])
def tune_model(mod, params, tune_settings, target, model_name):
"""
Tune a model for a specified number of trials along with other tune settings.
Tune settings are specified using a json configuration, as per the TVM tools readme.
"""
early_stopping = tune_settings['early_stopping']
number = tune_settings["number"]
save_path = tune_settings["save_path"]
save_name = tune_settings["save_name"]
repeat = tune_settings["repeat"]
debug = tune_settings.get("debug_gadqn") or False
trials = tune_settings["trials"]
tuner = tune_settings["tuner"]
target = tvm.target.Target(target)
tasks = autotvm.task.extract_from_program(
mod["main"],
target=target,
target_host="llvm",
params=params)
runner = autotvm.LocalRunner(
number=number,
repeat=repeat)
measure_option = autotvm.measure_option(
builder=autotvm.LocalBuilder(build_func="default"), runner=runner)
for i, tsk in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# Create a tuner
if tuner in ("xgb", "xgb-rank"):
tuner_obj = XGBTuner(tsk, loss_type="rank")
elif tuner == "xgb_knob":
tuner_obj = XGBTuner(tsk, loss_type="rank", feature_type="knob")
elif tuner == "ga":
tuner_obj = GATuner(tsk, pop_size=50)
elif tuner == "random":
tuner_obj = RandomTuner(tsk)
elif tuner == "gridsearch":
tuner_obj = GridSearchTuner(tsk)
elif tuner == "ga-dqn" and debug:
tuner_obj = GADQNTunerDebug(tsk)
elif tuner == "ga-dqn":
tuner_obj = GADQNTuner(tsk)
else:
raise ValueError("invalid tuner: %s " % tuner)
abs_path = Path(save_path + save_name).resolve()
abs_path.mkdir(exist_ok=True, parents=True)
abs_path_str = str(abs_path)
tuner_obj.tune(
n_trial=min(trials, len(tsk.config_space)),
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(trials, prefix=prefix),
autotvm.callback.log_to_file(abs_path_str + f"/tuning_record_model={model_name}.json"),
],
)
# Save debug info for rl tuner only
if tuner == "ga-dqn" and debug:
tuner_obj.save_model(save_path, save_name + f"_model={model_name}_layer={i}")
del tuner_obj
def tune_kernels(tasks,
gen_graph_tuner_candidates,
measure_top_n,
measure_option,
tuner='random',
early_stopping=None,
n_trial=5000,
log_filename='tuning.log'):
"""Tune kernels with the ranking model."""
remeasure_option = None
if tuner == 'round':
# Setup another measure option for final remeasurment.
remeasure_option = autotvm.measure_option(
builder=LocalBuilder(),
runner=measure_option['runner'].local_runner,
)
assert isinstance(measure_option['runner'], RankModelRunner)
best_results = []
for i, task in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
callbacks = []
if task.name in [
'dense_small_batch.cuda', 'conv2d_cudnn.cuda',
'dense_cublas.cuda', 'dense_large_batch.cuda',
'conv2d_transpose_nchw.cuda', 'dense_tensorcore.cuda'
]:
# Ignore these four tasks
continue
if task.name not in measure_option['runner'].models:
print('not covered by cost models')
continue
# create tuner
if tuner == 'round':
tuner_obj = RoundTuner(task, n_cfg=measure_top_n)
callbacks = [rank_progress(n_trial, prefix=prefix)
] # Use different callbacks.
else:
if tuner in ('xgb', 'xgb-rank'):
tuner_obj = XGBTuner(task, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(task, pop_size=50)
elif tuner == 'random':
tuner_obj = RandomTuner(task)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(task)
else:
raise ValueError("Invalid tuner: " + tuner)
callbacks = [
autotvm.callback.progress_bar(n_trial, prefix=prefix),
autotvm.callback.log_to_file(log_filename)
]
tic = time.time()
# do tuning
tuner_obj.tune(n_trial=n_trial,
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=callbacks)
# Round tuner needs an extra measurement step to get the real throughputs.
if tuner == 'round':
max_n_layout = 20 if gen_graph_tuner_candidates else 1
top_cfgs = tuner_obj.get_top_rank_cfgs(max_n_layout)
measure_batch = create_measure_batch(task, remeasure_option)
inputs = [
MeasureInput(task.target, task, config) for config in top_cfgs
]
sys.stderr.write('{} Measure Top {} Configs'.format(
prefix, len(inputs)))
results = measure_batch(inputs)
best_idx, best_flops = max(
[(idx, i.task.flop / np.mean(r.costs) /
1e9 if r.error_no == 0 else 0)
for idx, (i, r) in enumerate(zip(inputs, results))],
key=lambda x: x[1])
best_results.append((task.workload, best_idx, best_flops))
sys.stderr.write(' | Best %.2f GFLOPS at Top %d | %.2fs\n' %
(best_flops, best_idx, time.time() - tic))
autotvm.callback.log_to_file(log_filename)(None, inputs, results)
return best_results
def autotvm_tune(func, params, target):
"""
Parameters:
----------
func : relay.expr.Function
params : dict of str to numpy array
target : tvm.target.Target
ops : List of relay.op
"""
# Array of autotvm.task.Task
tasks = autotvm.task.extract_from_program(func,
target=target,
params=params,
ops=(relay.op.nn.conv2d, ))
# Check tasks.
for i in range(len(tasks)):
op_name = tasks[i].workload[0]
if op_name == 'conv2d':
func_create = 'topi_x86_conv2d_NCHWc'
elif op_name == 'depthwise_conv2d_nchw':
func_create = 'topi_x86_depthwise_conv2d_NCHWc_from_nchw'
else:
raise ValueError(
"Tuning {} is not supported on x86".format(op_name))
print("[Create Task %2d/%2d (%s, %s) ] " %
(i + 1, len(tasks), tasks[i].name, tasks[i].workload[0]))
tsk = autotvm.task.create(func_create,
args=tasks[i].args,
target=tasks[i].target,
template_key='direct')
tsk.workload = tasks[i].workload
tasks[i] = tsk
# turning option.
tuner = 'xgb'
n_trial = 100
early_stopping = None
log_filename = 'tuning.log'
use_transfer_learning = True
measure_option = autotvm.measure_option(
builder=autotvm.LocalBuilder(timeout=10),
runner=autotvm.LocalRunner(number=10, repeat=1, min_repeat_ms=1000))
# create tmp log file
tmp_log_file = log_filename + ".tmp"
if os.path.exists(tmp_log_file):
os.remove(tmp_log_file)
for i, tsk in enumerate(reversed(tasks)):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(tsk, loss_type='rank')
elif tuner == 'ga':
tuner_obj = GATuner(tsk, pop_size=100)
elif tuner == 'random':
tuner_obj = RandomTuner(tsk)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
if use_transfer_learning:
if os.path.isfile(tmp_log_file):
tuner_obj.load_history(
autotvm.record.load_from_file(tmp_log_file))
# do tuning
tuner_obj.tune(n_trial=min(n_trial, len(tsk.config_space)),
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(tmp_log_file)
])
# pick best records to a cache file
autotvm.record.pick_best(tmp_log_file, log_filename)
os.remove(tmp_log_file)
def tune_tasks(
tasks: List[autotvm.task.Task],
log_file: str,
measure_option: autotvm.measure_option,
tuner: str,
trials: int,
early_stopping: Optional[int] = None,
tuning_records: Optional[str] = None,
):
"""Tune a list of tasks and output the history to a log file.
Parameters
----------
tasks : list
A list of autotvm.Tasks to tune.
log_file : str
A file to output the tuning history, in JSON.
measure_option : autotvm.measure_option
Options to build and run a tuning task.
tuner : str
Which tuner to use.
trials : int
The maximum number of tuning trials to perform.
early_stopping : int, optional
The minimum number of tuning trials to perform.
This will be equal to 'trials' if not specified.
tuning_records: str, optional
Path to the file produced by the tuning, to be used during
tuning.
"""
if not tasks:
logger.warning("there were no tasks found to be tuned")
return
if not early_stopping:
early_stopping = trials
for i, tsk in enumerate(tasks):
prefix = "[Task %2d/%2d] " % (i + 1, len(tasks))
# Create a tuner
if tuner in ("xgb", "xgb-rank"):
tuner_obj = XGBTuner(tsk, loss_type="rank")
elif tuner == "xgb_knob":
tuner_obj = XGBTuner(tsk, loss_type="rank", feature_type="knob")
elif tuner == "ga":
tuner_obj = GATuner(tsk, pop_size=50)
elif tuner == "random":
tuner_obj = RandomTuner(tsk)
elif tuner == "gridsearch":
tuner_obj = GridSearchTuner(tsk)
else:
raise TVMCException("invalid tuner: %s " % tuner)
# If transfer learning is being used, load the existing results
if tuning_records and os.path.exists(tuning_records):
logger.info("loading tuning records from %s", tuning_records)
start_time = time.time()
tuner_obj.load_history(
autotvm.record.load_from_file(tuning_records))
logging.info("loaded history in %.2f sec(s)",
time.time() - start_time)
tuner_obj.tune(
n_trial=min(trials, len(tsk.config_space)),
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(trials, prefix=prefix),
autotvm.callback.log_to_file(log_file),
],
)
def tune_tasks(tasks,
only_parse,
measure_option,
tuner='xgb',
n_trial=1000,
early_stopping=None,
log_filename=None,
use_transfer_learning=True,
try_nchwc=True,
try_winograd=False,
try_spatial_pack_depthwise=False):
# create tmp log file
tmp_log_file = log_filename + ".tmp"
if not use_transfer_learning:
if os.path.exists(tmp_log_file):
os.remove(tmp_log_file)
else:
# select actual best logs
if not os.path.exists(tmp_log_file):
os.mknod(tmp_log_file)
autotvm.record.pick_best(tmp_log_file, log_filename)
if os.path.exists(tmp_log_file):
# sort out best historic entries
print("Load historic training logs...")
best_context = autotvm.task.ApplyHistoryBest(
autotvm.record.load_from_file(tmp_log_file))
best_tgtkeys = best_context.best_by_targetkey
print("Total tasks: %s" % len(tasks))
if try_nchwc:
for i in range(len(tasks)):
# converting conv2d tasks to conv2d_NCHWc tasks
op_name = tasks[i].workload[0]
if op_name == 'conv2d':
func_create = 'topi_x86_conv2d_NCHWc'
elif op_name == 'depthwise_conv2d_nchw':
func_create = 'topi_x86_depthwise_conv2d_NCHWc_from_nchw'
try: # try NCHWc template
tsk = autotvm.task.create(func_create, tasks[i].args,
tasks[i].target,
tasks[i].target_host, 'direct')
tsk.workload = tasks[i].workload
print(
"[Override Task %2d/%2d (%s) with NCHWc] {cfg.space: %i -> %i}"
% (i + 1, len(tasks), tsk.workload[0],
len(tasks[i].config_space), len(tsk.config_space)))
tasks[i] = tsk
except Exception:
pass
if try_winograd:
for i in range(0, len(tasks)):
try: # try winograd template
tsk = autotvm.task.create(tasks[i].name, tasks[i].args,
tasks[i].target,
tasks[i].target_host, 'winograd')
tasks.append(tsk)
print(
"[New Task %2d->%2d (%s) winograd] {cfg.space: %i -> %i}" %
(i + 1, len(tasks), tsk.workload[0],
len(tasks[i].config_space), len(tsk.config_space)))
except Exception:
pass
# if we want to use spatial pack for depthwise convolution
if try_spatial_pack_depthwise:
for i in range(len(tasks)):
if tasks[i].name == 'topi_nn_depthwise_conv2d_nchw':
tuner = 'xgb_knob'
tsk = autotvm.task.create(tasks[i].name, tasks[i].args,
tasks[i].target,
tasks[i].target_host,
'contrib_spatial_pack')
tasks.append(tsk)
print(
"[New Task %2d->%2d (%s) contrib_spatial_pack] {cfg.space: %i -> %i}"
% (i + 1, len(tasks), tsk.workload[0],
len(tasks[i].config_space), len(tsk.config_space)))
for i, tsk in enumerate(tasks):
tsk_template = ''
tsk_org_flop = -1
device_name = target.device_name if target.device_name else "cpu"
try:
# compute best historic entry GFLOPS
tsk_template = best_tgtkeys[(device_name,
tsk.workload)][0][2].template_key
tsk_org_cost = np.mean(best_tgtkeys[(device_name,
tsk.workload)][1].costs)
tsk_org_flop = tsk.flop / tsk_org_cost / 1e9
except:
pass
if tsk_org_flop == -1:
org_flop_str = "no history"
else:
org_flop_str = "%.2f GFLOPS /%s" % (tsk_org_flop, tsk_template)
prefix = "[Task %2d/%2d %s|%s] (%s) {%s}" % (
i + 1, len(tasks), tsk.workload[1][:4], tsk.workload[2][:4],
tsk.workload[0], org_flop_str)
if only_parse:
print("%s SKIP tunning" % prefix)
continue
# create tuner
if tuner == 'xgb' or tuner == 'xgb-rank':
tuner_obj = XGBTuner(tsk, loss_type='rank')
elif tuner == 'xgb_knob':
tuner_obj = XGBTuner(tsk, loss_type='rank', feature_type='knob')
elif tuner == 'ga':
tuner_obj = GATuner(tsk, pop_size=50)
elif tuner == 'random':
tuner_obj = RandomTuner(tsk)
elif tuner == 'gridsearch':
tuner_obj = GridSearchTuner(tsk)
else:
raise ValueError("Invalid tuner: " + tuner)
if use_transfer_learning:
if os.path.isfile(tmp_log_file):
tuner_obj.load_history(
autotvm.record.load_from_file(tmp_log_file))
# do tuning
tuner_obj.tune(n_trial=min(n_trial, len(tsk.config_space)),
early_stopping=early_stopping,
measure_option=measure_option,
callbacks=[
autotvm.callback.progress_bar(n_trial,
prefix=prefix),
autotvm.callback.log_to_file(tmp_log_file)
])
# pick best records to a cache file
autotvm.record.pick_best(tmp_log_file, log_filename)
if not use_transfer_learning:
os.remove(tmp_log_file)