get the op building on CHAI servers

flags for building ops changed in october 2017: tensorflow/tensorflow#13607

i got the new flags from https://www.tensorflow.org/guide/extend/op#compile_the_op_using_your_system_compiler_tensorflow_binary_installation

compiling gives lots of warnings about alignments being off, but all tests pass

makefile

@@ -1,10 +1,11 @@
-TF_INC :=$(shell python -c 'import tensorflow as tf; print(tf.sysconfig.get_include())')
+TF_CFLAGS := $(shell python -c 'import tensorflow as tf; print(" ".join(tf.sysconfig.get_compile_flags()))')
+TF_LFLAGS := $(shell python -c 'import tensorflow as tf; print(" ".join(tf.sysconfig.get_link_flags()))')
 INC=-I${TF_INC}
 layer_norm_fused_op: register_ops.cc layer_norm_fused_op.h layer_norm_fused_op.cc layer_norm_fused_grad_op.cc layer_norm_fused_op_gpu.cu.cc
-	nvcc -std=c++11 -c -o layer_norm_fused_op_gpu.cu.o layer_norm_fused_op_gpu.cu.cc \
-	$(INC) -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC --expt-relaxed-constexpr -gencode=arch=compute_35,code=sm_35 \
-	-gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_52,code=sm_52 \
-	-gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_61,code=compute_61
-	g++ -std=c++11 -shared -o layer_norm_fused_op.so register_ops.cc layer_norm_fused_op.h \
-	layer_norm_fused_grad_op.cc layer_norm_fused_op.cc layer_norm_fused_op_gpu.cu.o \
-	$(INC) -L /usr/local/cuda/lib64/ -fPIC -lcudart -O2 -DNDEBUG
+        nvcc -ccbin gcc-6 -std=c++11 -c -o layer_norm_fused_op_gpu.cu.o layer_norm_fused_op_gpu.cu.cc \
+        $(TF_CFLAGS) $(TF_LFLAGS) -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC --expt-relaxed-constexpr -gencode=arch=compute_35,code=sm_35 \
+        -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_52,code=sm_52 \
+        -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_61,code=compute_61
+        g++-6 -std=c++11 -shared -D_GLIBCXX_USE_CXX11_ABI=0 -o layer_norm_fused_op.so register_ops.cc layer_norm_fused_op.h \
+        layer_norm_fused_grad_op.cc layer_norm_fused_op.cc layer_norm_fused_op_gpu.cu.o \
+        $(TF_CFLAGS) $(TF_LFLAGS) -L /usr/local/cuda/lib64/ -fPIC -lcudart -O2 -DNDEBUG