Created by: jeng1220

PR types

Bug fixes

PR changes

Others

Describe

Because Paddle uses dlopen to load TensorRT at runtime, the TensoRT version might not be as same as the version that built with Paddle. In original implementation, Paddle always throws warning message " The Paddle lib links the XXX version TensorRT, make sure the runtime TensorRT you are using is no less than this verion, otherwise, there might be Segfault!" However, it is still unclear to user.

This patch adds getInferLibVersion to check which version of TensorRT is loaded by dlopen at runtime. If versions are not consistent. Paddle will throw "This Paddle lib was built with TensorRT XXX, but this system is using TensorRT YYY. It is recommended to use a consistent TensorRT version. Otherwise, there might be Segfault!"