TensorFlow Build Error "error: cannot convert 'int8x8_t to 'uint32x2_t"

Hi,

I am currently trying to build TensorFlow 1.14 using Bazel for Jetson AGX Xavier. However, it keeps throwing the following error:

./tensorflow/lite/kernels/internal/optimized/depthwiseconv_uint8_3x3_filter.h:37:64: error: cannot convert 'int8x8_t {aka __vector(8) signed char}' to 'uint32x2_t {aka __vector(2) unsigned int}' for argument '2' to 'void vst1_lane_u32(uint32_t*, uint32x2_t, int)'
   vst1_lane_u32(reinterpret_cast<uint32_t*>(dst), reg, lane_num)
                                                                ^

Also when I try trace back to the beginning of the error, it says the following:

note: use -flax-vector-conversions to permit conversions between vectors with differing element types or numbers of subparts

However, I am already using the -flax-vector-conversions and -fomit-frame-pointer flags inside the tensorflow/lite/tools/make/targets/aarch64_makefile.inc file and with my bazel build command.

bazel build --copt="-flax-vector-conversions" --copt="-fomit-frame-pointer" --cxxopt="-flax-vector-conversions" --cxxopt="-fomit-frame-pointer" --config=monolithic --config=nonccl --config=cuda --cxxopt=-D_GLIBCXX_USE_CXX11_ABI=0 --verbose_failures     tensorflow/tools/pip_package:build_pip_package     //tensorflow:libtensorflow_cc.so     //tensorflow:libtensorflow_framework.so

I know there is a pip package now and no need to build it from source anymore but I have made a couple of changes on the source code to try out another application with TF.

Hi,

You need some changes to enable the TensorFlow branch for ARM system.
Please check this topic for more information:
https://devtalk.nvidia.com/default/topic/1055131/jetson-agx-xavier/building-tensorflow-1-13-on-jetson-xavier/

Thanks.

I have applied exactly all the steps. But it throws the same error.

Hi,

This looks like a known issue of TensorFlow and is fixed already.
Would you mind to check this patch to see if helps?
https://github.com/tensorflow/tensorflow/issues/28277#issuecomment-517469318

Thanks.