Building ngraph-bridge fails on command 'make -j8'

[zap-wizard]

I build ngraph-bridge with following command:
python build_ngtf.py --use_prebuilt_tensorflow --build_plaidml_backend

It runs for few hours, but then suddenly fails:

[ 96%] Linking CXX shared library ../../libcpu_backend.dylib
[ 96%] Built target cpu_backend
make: *** [all] Error 2
Traceback (most recent call last):
  File "build_ngtf.py", line 525, in <module>
    main()
  File "build_ngtf.py", line 402, in main
    build_ngraph(build_dir, ngraph_src_dir, ngraph_cmake_flags, verbosity)
  File "/Users/username/ngraph-bridge/tools/build_utils.py", line 93, in build_ngraph
    command_executor(cmd, verbose=True)
  File "/Users/username/ngraph-bridge/tools/build_utils.py", line 60, in command_executor
    raise Exception("Error running command: " + cmd)
Exception: Error running command: make -j8

What the command make -j8 should do? Why is it missing the target parameter?

I'm using MacOS X High Sierra 10.13.16

[zap-wizard]

Digging little further into the problem, I have now edited my build_utils.py file, so that its easier to debug.

Added Popen and PIPE to be imported from subprocess.
line=24

from subprocess import check_output, call, Popen, PIPE

Added the additional error message ( " Error: " + Popen(shlex.split(cmd), stdout=PIPE, stderr=PIPE).communicate()[1].decode('utf-8') ).
line=60

if (call(shlex.split(cmd), stdout=stdout, stderr=stderr) != 0):
        raise Exception("Error running command: " + cmd + " Error: " + Popen(shlex.split(cmd), stdout=PIPE, stderr=PIPE).communicate()[1].decode('utf-8'))

So now we can see why the command make -j8 fails.

Running the same command as before:
python build_ngtf.py --use_prebuilt_tensorflow --build_plaidml_backend

Again, it ran for a while, then got this error message:

[ 96%] Linking CXX shared library ../../libcpu_backend.dylib
[ 96%] Built target cpu_backend
make: *** [all] Error 2
Traceback (most recent call last):
  File "build_ngtf.py", line 525, in <module>
    main()
  File "build_ngtf.py", line 402, in main
    build_ngraph(build_dir, ngraph_src_dir, ngraph_cmake_flags, verbosity)
  File "/Users/username/ngraph-bridge/tools/build_utils.py", line 93, in build_ngraph
    command_executor(cmd, verbose=True)
  File "/Users/username/ngraph-bridge/tools/build_utils.py", line 60, in command_executor
    raise Exception("Error running command: " + cmd + " Error: " + Popen(shlex.split(cmd), stdout=PIPE, stderr=PIPE).communicate()[1].decode('utf-8'))
Exception: Error running command: make -j8 Error: /Users/username/ngraph-bridge/build_cmake/ngraph/src/ngraph/runtime/plaidml/plaidml_translate.cpp:105:13: warning: enumeration value 'PLAIDML_DATA_BFLOAT16' not handled in switch [-Wswitch]
    switch (dt)
            ^
/Users/username/ngraph-bridge/build_cmake/ngraph/src/ngraph/runtime/plaidml/plaidml_translate.cpp:105:13: note: add missing switch cases
    switch (dt)
            ^
/Users/username/ngraph-bridge/build_cmake/ngraph/src/ngraph/runtime/plaidml/plaidml_translate.cpp:124:1: error: control may reach end of non-void function [-Werror,-Wreturn-type]
}
^
1 warning and 1 error generated.
make[2]: *** [src/ngraph/runtime/plaidml/CMakeFiles/plaidml_backend.dir/plaidml_translate.cpp.o] Error 1
make[1]: *** [src/ngraph/runtime/plaidml/CMakeFiles/plaidml_backend.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....
make: *** [all] Error 2

It looks like the problem is with plaidml...

Any help is appreciated!

[zap-wizard]

Now that I know this is a plaidml problem, I found out that this issue is a duplicate of #447. Also it seems like there's an solution there 😄