Make benchmark android app debuggable.
Also added guide for operator level profiling on a Android device. PiperOrigin-RevId: 298985442 Change-Id: Ie3231348849f444daa25ea53678b7cfa514a0f66
This commit is contained in:
Juho Ha
TensorFlower Gardener
parent
0e41e3b246
commit
786c819ea8
@ -21,22 +21,21 @@
|
||||
<!-- Necessary for loading custom models from disk. -->
|
||||
<uses-permission android:name="android.permission.READ_EXTERNAL_STORAGE"/>
|
||||
|
||||
<!-- Target SDK 21 (<23) to avoid the need for requesting storage
|
||||
permissions. This APK will almost always be used from the command-line
|
||||
anyway, and be expicitly installed by the developer. -->
|
||||
<uses-sdk
|
||||
android:minSdkVersion="21"
|
||||
android:targetSdkVersion="21" />
|
||||
android:minSdkVersion="23"
|
||||
android:targetSdkVersion="23" />
|
||||
|
||||
<application>
|
||||
<application
|
||||
android:debuggable="true">
|
||||
<!-- This Activity runs the TensorFlow Lite benchmark at creation, using
|
||||
a provided set of arguments, then immediately terminates. -->
|
||||
<activity android:name="org.tensorflow.lite.benchmark.BenchmarkModelActivity"
|
||||
android:screenOrientation="portrait"
|
||||
android:label="TFLite Benchmark"
|
||||
android:theme="@android:style/Theme.NoDisplay"
|
||||
android:exported="true"
|
||||
android:noHistory="true" />
|
||||
<activity
|
||||
android:name=".BenchmarkModelActivity"
|
||||
android:screenOrientation="portrait"
|
||||
android:label="TFLite Benchmark"
|
||||
android:theme="@android:style/Theme.NoDisplay"
|
||||
android:exported="true"
|
||||
android:noHistory="true" />
|
||||
</application>
|
||||
|
||||
</manifest>
|
||||
|
||||
@ -34,8 +34,10 @@ bazel build -c opt \
|
||||
(2) Connect your phone. Install the benchmark APK to your phone with adb:
|
||||
|
||||
```
|
||||
adb install -r -d bazel-bin/tensorflow/lite/tools/benchmark/android/benchmark_model.apk
|
||||
adb install -r -d -g bazel-bin/tensorflow/lite/tools/benchmark/android/benchmark_model.apk
|
||||
```
|
||||
Note: Make sure to install with "-g" option to grant the permission for reading
|
||||
extenal storage.
|
||||
|
||||
(3) Push the compute graph that you need to test.
|
||||
|
||||
@ -50,9 +52,10 @@ and can be appended to the `args` string alongside the required `--graph` flag
|
||||
args key).
|
||||
|
||||
```
|
||||
adb shell am start -S -n \
|
||||
org.tensorflow.lite.benchmark/org.tensorflow.lite.benchmark.BenchmarkModelActivity \
|
||||
--es args '"--graph=/data/local/tmp/mobilenet_quant_v1_224.tflite --num_threads=4"'
|
||||
adb shell am start -S \
|
||||
-n org.tensorflow.lite.benchmark/.BenchmarkModelActivity \
|
||||
--es args '"--graph=/data/local/tmp/mobilenet_quant_v1_224.tflite \
|
||||
--num_threads=4"'
|
||||
```
|
||||
|
||||
(5) The results will be available in Android's logcat, e.g.:
|
||||
@ -62,3 +65,69 @@ adb logcat | grep "Average inference"
|
||||
|
||||
... tflite : Average inference timings in us: Warmup: 91471, Init: 4108, Inference: 80660.1
|
||||
```
|
||||
|
||||
## To trace Tensorflow Lite internals including operator invocation
|
||||
|
||||
The steps described here follows the method of
|
||||
https://developer.android.com/topic/performance/tracing/on-device. Refer to the
|
||||
page for more detailed information.
|
||||
|
||||
(0)-(3) Follow the steps (0)-(3) of [build/install/run](#to_buildinstallrun)
|
||||
section.
|
||||
|
||||
(4) Set up Quick Settings tile for System Tracing app on your device. Follow the
|
||||
[instruction]
|
||||
(https://developer.android.com/topic/performance/tracing/on-device#set-up-tile).
|
||||
The System Tracing tile will be added to the Quick Settings panel, which appears
|
||||
in next figure:
|
||||
|
||||
{ width=40% }
|
||||
|
||||
Optionally, you can set up other configurations for tracing from the app menu.
|
||||
Refer to the [guide]
|
||||
(https://developer.android.com/topic/performance/tracing/on-device#app-menu)
|
||||
for more information.
|
||||
|
||||
(5) Tap the System Tracing tile, which has the label "Record trace". The tile
|
||||
becomes enabled, and a persistent notification appears to notify you that the
|
||||
system is now recording a trace, as shown in next figure:
|
||||
|
||||
{ width=40% }
|
||||
|
||||
(6) Run the benchmark with platform tracing enabled.
|
||||
|
||||
```
|
||||
adb shell am start -S \
|
||||
-n org.tensorflow.lite.benchmark/.BenchmarkModelActivity \
|
||||
--es args '"--graph=/data/local/tmp/mobilenet_quant_v1_224.tflite \
|
||||
--num_threads=4 --enable_op_profiling=true --enable_platform_tracing=true"'
|
||||
```
|
||||
|
||||
(7) Wait until the benchmark finishes. It can be checked from Android log
|
||||
messages, e.g.,
|
||||
|
||||
```
|
||||
adb logcat | grep "Average inference"
|
||||
|
||||
... tflite : Average inference timings in us: Warmup: 91471, Init: 4108, Inference: 80660.1
|
||||
```
|
||||
|
||||
(8) Stop tracing by tapping either the System Tracing tile in the Quick Settings
|
||||
panel or on the System Tracing notification. The system displays a new
|
||||
notification that contains the message "Saving trace". When saving is complete,
|
||||
the system dismisses the notification and displays a third notification,
|
||||
confirming that your trace has been saved and that you're ready to share the
|
||||
system trace, as shown in next figure:
|
||||
|
||||
{ width=40% }
|
||||
|
||||
(9) [Share]
|
||||
(https://developer.android.com/topic/performance/tracing/on-device#share-trace)
|
||||
a trace file, [convert]
|
||||
(https://developer.android.com/topic/performance/tracing/on-device#converting_between_trace_formats)
|
||||
between tracing formats and [create]
|
||||
(https://developer.android.com/topic/performance/tracing/on-device#create-html-report)
|
||||
an HTML report.
|
||||
Note that, the catured tracing file format is either in Perfetto format or in
|
||||
Systrace format depending on the Android version of your device. Select the
|
||||
appropriate method to handle the generated file.
|
||||
|
||||
@ -18,6 +18,7 @@ package org.tensorflow.lite.benchmark;
|
||||
import android.app.Activity;
|
||||
import android.content.Intent;
|
||||
import android.os.Bundle;
|
||||
import android.os.Trace;
|
||||
import android.util.Log;
|
||||
|
||||
/** Main {@code Activity} class for the benchmark app. */
|
||||
@ -37,7 +38,9 @@ public class BenchmarkModelActivity extends Activity {
|
||||
String args = bundle.getString(ARGS_INTENT_KEY_0, bundle.getString(ARGS_INTENT_KEY_1));
|
||||
Log.i(TAG, "Running TensorFlow Lite benchmark with args: " + args);
|
||||
|
||||
Trace.beginSection("TFLite Benchmark Model");
|
||||
BenchmarkModel.run(args);
|
||||
Trace.endSection();
|
||||
|
||||
finish();
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user