tf_custom_op_makefile

CXX = /usr/bin/g++
NVCC = /usr/local/cuda-9.1/bin/nvcc

CUDA_LIB    = /usr/local/cuda-9.1/lib64
CUDA_INC    = /usr/local/cuda-9.1/include
CUDA_CFLAGS = -I $(CUDA_INC)
CUDA_LFLAGS = -lcudart -L $(CUDA_LIB)

TF_LIB    = $(shell python -c 'import tensorflow as tf; print(tf.sysconfig.get_lib())')
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()))')

# for tf <= 1.4.0
# TF_CFLAGS = -I $(TF_INC) -I $(TF_INC)/external/nsync/public -L $(TF_LIB) -D_GLIBCXX_USE_CXX11_ABI=1
# TF_LFLAGS = -L $(TF_LIB) -ltensorflow_framework

all: tf_sampling_so.so

%.cu:

%.cu.o: %.cu
	$(NVCC) -std=c++11 -c -o $@ $^ $(TF_CFLAGS) -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC -O2 --expt-relaxed-constexpr

%_so.so: %.cpp %_g.cu.o 
	$(CXX) -std=c++11 -shared -o $@ $^ $(TF_CFLAGS) $(CUDA_CFLAGS) $(CUDA_LFLAGS) $(TF_LFLAGS) -fPIC -O2

clean:
	rm -f *.so *.cu.o