Compiling fluid inference library

[luotao1]

For compiling fluid inference static and shared library, there are two questions:

1. Do we need an option like WITH_C_API?
2. Can we build a single static and shared library at first, and then spill them?

[Xreki]

Do we need an option like WITH_C_API?

I think we don't need. It should be always built.

Can we build a single static and shared library at first, and then spill them?

It's OK. Please note that the static library should not contain any binary code of third-party libraries.

[luotao1]

TODO List

• Create an independent project, like capi dense example to verify the link to libpaddle_fluid.a or libpaddle_fluid.so
• Compile one static library, which should not contain any binary code of third-party libraries.
• Compile one shared library.
• Dismantle one library into several.
• Make Install.

[wangkuiyi]

I have a question: do we have to install header files and libraries into a special directory structure? What would be the problem for our users to include/link to them if we leave them where they were (in the source directory and the build directory)?