nsync | Pseudo global configmaps and secrets in kubernetes | Frontend Framework library

tensor Flow heavily uses AVX instruction on x86 platforms. If the binary is compiled with AVX512 that is zmm registers the binary can run on supporting hardware. Hence as per the comments requested to check the instruction set via

1. objdump -M intel -S /usr/lib64/libtensorflow.so.2 | grep -i zmm and
2. print $pc in GDB to isloate the instruction.

Note: as per the update changing from Broadwell (no AVX512) to Skylake (AVX512) has solved the issue.

One answer, I have found:

I have installed my python via anaconda2 and I always tried to find out TF_INC and TF_LIB when I activated my repository source activate . and the could not found any ~/anaconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow *.so files

This time I went out every python environment with the shell command source deactivate and I typed the following command python -c 'import tensorflow as tf; print(tf.sysconfig.get_lib())'

Now, I got a different path: ~/anaconda2/lib/python2.7/site-packages/tensorflow, where the lib libtensorflow_framework.so is located.

Adding -DNDEBUG to the build flags fixes the issue. This workaround is explained in TF issue 17316.

I do not believe so. The preceding sentence ( in my -draft copy ) states:

If neither bit is set on both LR and SC, the LR/SC sequence can be observed to occur before orafter surrounding memory operations from the same RISC-V hart. This can be appropriate ....

What I think they are getting at is the lr/sc will only synchronize with the linked address (ok, cache line), so other memory locations between the lr/sc can remain in a looser order than this one. Stated another way, the lr/sc doesn't have to wait for preceding items in the store buffer to make their way out to RAM unless there is some other fence-like operation pending.

As such, it would be inappropriate for implementing a mutex (as most multi-threaded programs consider the mutex a memory ordering point), but well suited for a counter or stack index.

The most obvious answer would be

Officially, you have to create your C++ project in tensorflow sourcetree, write BUILD file and compile it using bazel. Look at this.

I wrote a blog post: https://tuanphuc.github.io/standalone-tensorflow-cpp/ giving detailed instructions to make a standalone C++ Tensorflow with configuration: - Ubuntu 17.10 - gcc 7.2.0 - tensorflow 1.4.0 - Python 2 or 3 - cmake 3.9.6 - Eigen 3.3.4 - Protobuf (master branch) - Googletest (master branch) - bazel Hope it helps

References:

Packaged TensorFlow C++ library for bazel-independent use

graph.pb.h missing

The problem is that in the CROSSTOOL the include directories that contain mutex header are only added to the C++ command lines. I can see how this came to be, the directory where mutex is is org_linaro_components_toolchain_gcc_5_3_1/arm-linux-gnueabihf/include/c++/5.3.1/mutex, note the c++. If you believe that the directory is also valid for C compilation, the principled fix is to change cxx_flag into compiler_flag in the CROSSTOOL.

You are right in that g++ is not the same as gcc. There is an issue at https://github.com/bazelbuild/bazel/issues/4644. But it's not really related to your question, it's only bazel that doesn't put cxx_flags to C compilation command lines, not gcc. Or am I missing something?

And I'd advise not to depend on Bazel test data crosstools :)

--save_temps only works for the target Bazel is explicitly building, it doesn't work for all transitive dependencies. So you have to find a C++ target that you're interested in and build it directly. Then you'll see a bunch of .ii and .s files.

Side track, you can run your build with --subcommands to see all the actions and command lines that Bazel creates.

Re query, bazel query //tensorflow/... works for me, is it possible that the error is caused by sync-out of sources and ./configure? It looks like a configuration issue. I didn't try building TF with cuda support enabled.

I suspect that the problem is because you specified the encoding for file, even although it's binary. Cassandra driver needs Buffer type that is returned when fs.readFile is called only with file name & handler, without encoding (see fs.readFile docs).