Embeding shader source to code or use external file

In the professional world shaders are one of many assets, and shaders are in their own files.

When built for shipping shaders get processed and bundled together with all the other assets, meaning they are small pieces of a very large file.

I thought it was unexpected that you mention slow load on mobile devices. Are you working on a mobile game?

Not a mobile game, but a mobile/desktop OpenGL app. On older devices, when shaders are inside assets (on Android) unpack and load takes longer time.

I did both of what you described while I added "normal" shaders to a resource package, I also added them to code as plain text for some little build in feature (Debug-Drawing in GL).

I would recommend to ship them as external resource because packages could be maintained easier, you otherwise would need to rebuild your source when changing a shader and that may propably take longer compile times, also for bigger projects, your assembly file will be less in size as when you add the shader source to code.

Yes, GLSL takes a "relative" long time to compile because it is plain text so your devices graphics chip needs first to get all the text transferred before it is able to split, tokenize and parse it before it is able to translate the source into a GPU ready assembly program. You could not do anything very straight forward here because GLSL takes plain text source but you could cache shaders as binary output on disk and reuse it if possible. This is a lot faster because anything is already processed but the binary is platform and vendor dependent so you wont be able to ship that without a guarantee to work.

Access time is also something you should optimize. Make your assets a package and align that to the disk cache so you will reduce I/O-Buffer usage. In C++ you are also able to reserve page memory for your files (Memory Mapped Files) using CreateFileMapping on Windows and mmap on Linux/Unix (Android is Unix). This will enable your program to tell the OS to load a specific part of the file into a memory page (thats why you should align/padd your file) and increase access time by several milliseconds. You could also access two or more files (as long as they fit into the "chunks" you loaded) at once, even in an asynchronous multithreaded way. (This are the tricks that AAA games use on console)

Presently I compile my shaders (HLSL) straight to a header file and include them in my projects. The additional compile time is negligible (in my case), with the entire project taking eight seconds to build an exe.