Error while compiling DLDT on AM5728 Ti ARM

[Berry__Stephen]

I have a Beagleboard X15 running Ubuntu 16.04 LTS and I'm trying to build the DLDT from git source downloaded from:

https://github.com/opencv/dldt.git

I'm able to run the install_dependencies.sh without a problem, but trying to build the inference engine fails shortly into the build:

[  3%] Building CXX object tests/libs/gtest/googlemock/CMakeFiles/gmock_main.dir/__/googletest/src/gtest-all.cc.o
[  4%] Building CXX object samples/common/format_reader/CMakeFiles/format_reader.dir/opencv_wraper.cpp.o
[  5%] Building CXX object src/inference_engine/CMakeFiles/inference_engine_s.dir/graph_transformer.cpp.o
c++: error: unrecognized command line option ‘-msse4.2’
src/inference_engine/CMakeFiles/inference_engine_s.dir/build.make:62: recipe for target 'src/inference_engine/CMakeFiles/inference_engine_s.dir/graph_transformer.cpp.o' failed
make[2]: *** [src/inference_engine/CMakeFiles/inference_engine_s.dir/graph_transformer.cpp.o] Error 1
CMakeFiles/Makefile2:149: recipe for target 'src/inference_engine/CMakeFiles/inference_engine_s.dir/all' failed
make[1]: *** [src/inference_engine/CMakeFiles/inference_engine_s.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....

Any clues as to how I can fix this? Thanks!

[Zhen_Z_Intel]

Hello Stephen, I am afraid you cannot compile DLDT on ARM processor cause our SIMD intrinsic set is not compatible on ARM. The kernel of DLDT accelerating dynamic lib on CPU is using MKLDNN which is optimized with Intel SIMD intrinsic set including SSE4.2, AVX,AVX2, ANX512 and VNNI.

[Berry__Stephen]

Hi Fiona,

My main interest is to be able to use the Movidius compute stick (2) on the ARM platform. I was hoping that I could compile the libraries and drivers necessary just for that accelerator. I tried to turn off the CPU and GPU portions of the build with cmake options but it didn't seem to work. Does the Movidius NCS software still require SIMD instructions?

From the build instructions:

To switch on/off the CPU and GPU plugins, use cmake options -DENABLE_MKL_DNN=ON/OFF and -DENABLE_CLDNN=ON/OFF.

Thanks!