Jie Zhang

`cmake/Cuda.cmake` file needs update for CUDA 10

确保你正确编译了 protobuf 库，并用对应的 protoc 工具处理了 caffe.proto 文件

@zhly0 consider to use profiler, see this [doc](https://github.com/luoyetx/mini-caffe/blob/master/profile.md)

我自己分析来看也确实差不多 😂

静态分配还可以进一步做，将多个 blob 的内存拼接起来组成大内存，在同一块大内存上分配给小 blob，不过这个实现起来比较麻烦，而且不一定有最优解

What's your cudnn version? I think the newest version is cudnn6, but I didn't test with it. Only cudnn5.1 is tested.

@mxmxlwlw Sorry, I don't use matlab. Would you like to add the interface?

@GarrickLin would you like to send a pr?

@GarrickLin you can link caffe.lib (release version) for both debug and release in your project. Thus, you need to compile release version first.

You need to link pthread, add link flags in tests.cmake