def test_tuning_cpu():
ir_mod = tvm.parser.fromtext(
textwrap.dedent("""
#[version = "0.0.5"]
def @main(%a : Tensor[(1, 3, 32, 32), float32], %b : Tensor[(3, 3, 5, 5), float32]) {
nn.conv2d(%a, %b, data_layout="NCHW", kernel_layout="OIHW")
}
"""))
tasks = autotvm.task.relay_integration.extract_from_program(
ir_mod, {}, tvm.target.create("llvm"))
assert len(tasks) == 1, f"Extracted != 1 task from program: {tasks!r}"
task = tasks[0]
measure_option = autotvm.measure_option(autotvm.LocalBuilder(),
autotvm.LocalRunner())
results = []
tuner = RandomTuner(task)
tuner.tune(
n_trial=20,
measure_option=measure_option,
callbacks=(lambda _tuner, _inputs, rs: results.extend(rs),
),
)
assert len(results) == 20
successful_results = [
r for r in results if r.error_no == autotvm.MeasureErrorNo.NO_ERROR
]
assert len(
successful_results) > 0, f"No successful tuning runs: {results!r}"
def runner(target):
# init task
task, target = get_sample_task(target, None)
logging.info("task config space: %s", task.config_space)
measure_option = autotvm.measure_option(autotvm.LocalBuilder(),
autotvm.LocalRunner())
results = []
tuner = RandomTuner(task)
tuner.tune(
n_trial=20,
measure_option=measure_option,
callbacks=(lambda _tuner, _inputs, rs: results.extend(rs),
),
)
assert len(results) == 20
successful_results = [
r for r in results if r.error_no == autotvm.MeasureErrorNo.NO_ERROR
# We filter records before building if we know they won't work ahead of time.
# We can't guarantee we get one good record so we count these as success too
or r.error_no == autotvm.MeasureErrorNo.INSTANTIATION_ERROR
]
assert len(
successful_results) > 0, f"No successful tuning runs: {results!r}"
def test_tuning_gpu(target, ctx):
# init task
task, target = get_sample_task(target, None)
logging.info("task config space: %s", task.config_space)
measure_option = autotvm.measure_option(autotvm.LocalBuilder(),
autotvm.LocalRunner())
results = []
tuner = RandomTuner(task)
tuner.tune(
n_trial=20,
measure_option=measure_option,
callbacks=(lambda _tuner, _inputs, rs: results.extend(rs),
),
)
assert len(results) == 20
successful_results = [
r for r in results if r.error_no == autotvm.MeasureErrorNo.NO_ERROR
]
assert len(
successful_results) > 0, f"No successful tuning runs: {results!r}"
def test_tuning(target, ctx):
# init task
task, target = get_sample_task(target, None)
logging.info("%s", task.config_space)
measure_option = autotvm.measure_option(autotvm.LocalBuilder(),
autotvm.LocalRunner())
tuner = RandomTuner(task)
tuner.tune(n_trial=20, measure_option=measure_option)
def check(target, target_host):
ctx = tvm.context(target, 0)
if not ctx.exist:
logging.info("Skip test because %s is not available" % target)
return
# init task
task, target = get_sample_task(target, target_host)
logging.info("%s", task.config_space)
measure_option = autotvm.measure_option(autotvm.LocalBuilder(),
autotvm.LocalRunner())
tuner = RandomTuner(task)
tuner.tune(n_trial=20, measure_option=measure_option)
def check(target, target_host):
ctx = tvm.context(target, 0)
if not ctx.exist:
logging.info("Skip test because %s is not available" % target)
return
# init task
task, target = get_sample_task(target, target_host)
logging.info("%s", task.config_space)
measure_option = autotvm.measure_option(
autotvm.LocalBuilder(),
autotvm.LocalRunner())
tuner = RandomTuner(task)
tuner.tune(n_trial=20, measure_option=measure_option)
def check(target, target_host):
ctx = tvm.context(target, 0)
if not ctx.exist:
logging.info("Skip test because %s is not available" % target)
return
# init task
task, target = get_sample_task(target, target_host)
logging.info("%s", task.config_space)
measure_option = autotvm.measure_option(mode='local',
timeout=4,
number=2)
tuner = RandomTuner(task)
tuner.tune(n_trial=10, measure_option=measure_option)
def runner(target, dev):
task, target = get_sample_task(target, None)
logging.info("task config space: %s", task.config_space)
# Note: we use the MockedLocalBuilder here instead of autotvm.LocalBuilder()
measure_option = autotvm.measure_option(MockedLocalBuilder(),
autotvm.LocalRunner())
results = []
tuner = RandomTuner(task)
tuner.tune(
n_trial=1,
measure_option=measure_option,
callbacks=(lambda _tuner, _inputs, rs: results.extend(rs),
),
)
assert len(results) == 1
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),
],
)