Hi,
I am not able to compile android-nn-driver 21.08, here is the error I get:

ArmnnPreparedModel_1_2.cpp:722:31: error: no matching function for call to 'MicrosecondsDuration'
        timing.timeOnDevice = MicrosecondsDuration(timeTaken.second, timeTaken.first);
                              ^~~~~~~~~~~~~~~~~~~~
ArmnnPreparedModel_1_2.cpp:38:15: note: candidate function not viable: no known conversion from 'time_point<std::chrono::system_clock, [...]>' to 'time_point<std::chrono::steady_clock, [...]>' for 1st argument
unsigned long MicrosecondsDuration(TimePoint endPoint, TimePoint startPoint)

At line 700 we can see that timeTaken is of type armnn::InferenceTimingPair which is defined such as:

using HighResolutionClock = std::chrono::high_resolution_clock::time_point;
using InferenceTimingPair = std::pair<HighResolutionClock, HighResolutionClock>;

inside armnn v21.08

while timepoint is defined as

using TimePoint = std::chrono::steady_clock::time_point;

inside ArmnnPreparedModel_1_2.cpp line 31

I am not a c++ expert, so maybe there is a magical option I can use to make this work? I am cross compiling using clang

I can work an a patch if needed.

Thanks

Hi ARM-software,

I'm using android-nn-driver well until Android P.
However, it is not possible to build in the Android Q recently.
Do you have any plans to support the Android Q? When?

Regards
HJ

Hi @MatthewARM ,

Is root access needed to integrate Android NN driver?
Does Android O 8.1 device with ARM processor (say Huawei P20) come with Android NN driver pre-installed?

Thanks!

I tried to run tflitedemo example and found the message,

ArmnnDriver: ConvertConv2d: not supported by armnn

Will you have the plan to add it?

I notice that the Andorid 14 request the min version of NN-service is 1.4, and require AIDL interface.
I would like to know the release plan for the version supported Android 14.
thanks a lot

I'm trying to build the android-nn-driver by cloning it and pushing to ANDROID_ROOT following the integration guide. Currently the steps to generate the driver from the below step is unclear.
Update the Android build environment to add the Arm NN driver. This ensures that the driver service is built and copied to the system/vendor/bin/hw directory in the Android image. To update the build environment, add to the contents of the variable PRODUCT_PACKAGES within the device-specific makefile that is located in the <ANDROID_ROOT>/device// directory. This file is normally called device.mk
I'm seeing the device.mk files & its dependent mk files in build of cloned repo (git clone https://android.googlesource.com/platform/manifest). Not clear about the linkage between the ADB shell and the android google source cloned repo.
device/ directory is in android google source clone & vendor/ directory is in adb shell ANDROID_ROOT. Please explain the linkage between these two.

Can someone explain/provide the clear steps to build the android-nn-driver using the android emulator & ADB shell.

Hi ARM-software,

Is the support for android NN hal v1.2 in the next release?
Then, what's your schedule?

Regards
HJ

result has reserve model.operations.size() before the loop, if we are still pushing to result, will the size of result and model.operations be the same?

Hi，
2. Mali OpenCL driver integrated into the Android source tree
How to do it？？

I am building [email protected], pushing it to device and running it manually through adb shell. When I try to run I am getting error:

09-24 04:00:22.692 9611 9611 I Manager : Found interface armnn
09-24 04:00:22.692 3235 3235 W hwservicemanager: getTransport: Cannot find entry [email protected]::IDevice/armnn in either framework or device manifest.

Any suggestion how to fix it?

In 18.05 release note, there is
"In general, the amount of RAM needed for a loaded network has been reduced by 20-30% compared to release 18.03."

And in 18.08 release note, there is
"This release of Arm NN integrates the latest Compute Library and adds improvements to thread-safety, memory consumption and overall performance."

I got information about memory usage in issue:
#8

But when I tried in 18.05, it looks like more than 2 * model size.
What version should I use to get better memory usage?

How we measure memory needed when armnn processing a model?
According to model' batch number, input tensor or number of layers?

pradeep.k@PradeepK-LaptopU:~/.../boost_1_64_0$ ./b2 install --user-config=$HOME/armnn-devenv/boost/user-config.jam \

toolset=gcc-arm link=static cxxflags=-fPIC --with-filesystem
--with-test --with-log --with-program_options -j16
/home/pradeep.k/armnn-devenv/boost/boost_1_64_0/tools/build/src/tools/gcc.jam:84: in gcc.init from module gcc
error: toolset gcc initialization:
error: provided command 'aarch64-linux-android-clang++' not found
error: initialized from /home/pradeep.k/armnn-devenv/boost/user-config.jam:1
/home/pradeep.k/armnn-devenv/boost/boost_1_64_0/tools/build/src/build/toolset.jam:43: in toolset.using from module toolset
/home/pradeep.k/armnn-devenv/boost/boost_1_64_0/tools/build/src/build/project.jam:1052: in using from module project-rules
/home/pradeep.k/armnn-devenv/boost/user-config.jam:1: in modules.load from module user-config
/home/pradeep.k/armnn-devenv/boost/boost_1_64_0/tools/build/src/build-system.jam:249: in load-config from module build-system
/home/pradeep.k/armnn-devenv/boost/boost_1_64_0/tools/build/src/build-system.jam:385: in load-configuration-files from module build-system
/home/pradeep.k/armnn-devenv/boost/boost_1_64_0/tools/build/src/build-system.jam:524: in load from module build-system
/home/pradeep.k/armnn-devenv/boost/boost_1_64_0/tools/build/src/kernel/modules.jam:295: in import from module modules
/home/pradeep.k/armnn-devenv/boost/boost_1_64_0/tools/build/src/kernel/bootstrap.jam:139: in boost-build from module
/home/pradeep.k/armnn-devenv/boost/boost_1_64_0/boost-build.jam:17: in module scope from module

Hi ARM-software,

I have a question about ARM HAL driver ([email protected]).
(I am new to the Android build system. So my question would be trivial.)
I refered to #2 and #4.

I have a galaxy s10+ device and I'd like to run tf-lite with ARM GPU by using android-nn-api path. In order to accelerate ARM GPU by using android-nn-api on smart phones, do I have to build this android-nn-driver in manual? (Isn't there prebuilt android-nn-driver?)
I'm also wondering it's possible to copy this driver([email protected]) to the existing target without flashing full image.

Regards,
Dongjin.

Hi ARMer,

I have a built image, and if I don't build-in the nn driver and just build service and *.so, then put them to my fomer image specific directory. run it manually, will it work?

Why ask this?

• Because I try to run it in emulator, this emulator is from android studio. I don't want to change the emulator image to myself built.

What about my try?

• fail. T_T.

Regards,

Are there perhaps ways to build on vulkan or just plain OpenGL?

Thanks

In getType() at:

they always return DeviceType::CPU, whereas they should return 'Other' (for a heterogenous interface) or check the service startup flags to find out which backend is preferred (e.g. GPU).

We should have a cmake for this project and allow pure linux to run and build a libneutralnetwork.so directly. tensroflow lite nnapi delegate will dlopen libnutralnetwork.so in pure linux environment (non-android)

There is no need for a duplicated and separated delegate for tensorflow lite.
https://github.com/ARM-software/armnn/tree/branches/armnn_20_11/delegate