Compilation error: math_function_generated_im2col.cu.o

[luotao1]

/home/luotao02/Paddle/paddle/fluid/framework/dim.h(161): error: Within a __device__/__global__ function, only __shared__ variables may be marked "static"

/home/luotao02/Paddle/paddle/fluid/framework/dim.h(187): error: Within a __device__/__global__ function, only __shared__ variables may be marked "static"

2 errors detected in the compilation of "/home/luotao02/.tmp/tmpxft_00003712_00000000-7_im2col.cpp1.ii".
CMake Error at math_function_generated_im2col.cu.o.cmake:260 (message):
  Error generating file
  /home/luotao02/Paddle/build/paddle/fluid/operators/math/CMakeFiles/math_function.dir//./math_function_generated_im2col.cu.o


make[2]: *** [paddle/fluid/operators/math/CMakeFiles/math_function.dir/math_function_generated_im2col.cu.o] Error 1
make[1]: *** [paddle/fluid/operators/math/CMakeFiles/math_function.dir/all] Error 2
make: *** [all] Error 2

This error is similar to tensorflow/tensorflow#10258.
The reason is that On CUDA versions previous to 8.0, only shared variables could be declared as static in the device code.