From 40251fc874bc2c4d6399674a597c5e3caee8c477 Mon Sep 17 00:00:00 2001
From: Yifan Zhao <yifanz16@illinois.edu>
Date: Sat, 23 Jan 2021 23:35:24 -0600
Subject: [PATCH] Handle float64 downstream
---
predtuner/approxapp.py | 6 ++++--
predtuner/torchapp.py | 5 ++---
2 files changed, 6 insertions(+), 5 deletions(-)
diff --git a/predtuner/approxapp.py b/predtuner/approxapp.py
index fbc7eda..c7e9415 100644
--- a/predtuner/approxapp.py
+++ b/predtuner/approxapp.py
@@ -241,7 +241,8 @@ class TunerInterface(MeasurementInterface):
self.pbar = tqdm(total=test_limit, leave=False)
self.app_kwargs = app_kwargs
- objective = ThresholdAccuracyMinimizeTime(tuner_thres)
+ # tune_thres can come in as np.float64 and opentuner doesn't like that
+ objective = ThresholdAccuracyMinimizeTime(float(tuner_thres))
input_manager = FixedInputManager(size=len(self.app.op_knobs))
super(TunerInterface, self).__init__(
args,
@@ -263,7 +264,8 @@ class TunerInterface(MeasurementInterface):
from opentuner.resultsdb.models import Result
cfg = desired_result.configuration.data
- qos, perf = self.app.measure_qos_perf(cfg, False)
+ qos, perf = self.app.measure_qos_perf(cfg, False, **self.app_kwargs)
+ qos, perf = float(qos), float(perf)
# Print a debug message for each config in tuning and keep threshold
self.print_debug_config(qos, perf)
self.pbar.update()
diff --git a/predtuner/torchapp.py b/predtuner/torchapp.py
index 172abb0..052a378 100644
--- a/predtuner/torchapp.py
+++ b/predtuner/torchapp.py
@@ -110,8 +110,7 @@ class TorchApp(ModeledApp, abc.ABC):
end = begin + len(target)
qos = self.tensor_to_qos(tensor_output[begin:end], target)
qoses.append(qos)
- # float64 -> float
- return float(self.combine_qos(np.array(qoses)))
+ return self.combine_qos(np.array(qoses))
return [
LinearPerfModel(self._op_costs, self._knob_speedups),
@@ -137,7 +136,7 @@ class TorchApp(ModeledApp, abc.ABC):
qoses.append(self.tensor_to_qos(outputs, targets))
time_end = time_ns() / (10 ** 9)
qos = self.combine_qos(np.array(qoses))
- return float(qos), time_end - time_begin # float64->float
+ return qos, time_end - time_begin
def __repr__(self) -> str:
class_name = self.__class__.__name__
--
GitLab