Thread: Optimizing by eliminating translation units

Hi everyone,

I know that MSVC is capable of looking into other translation units while compiling one translation unit, to enable optimization. I'm told that GCC can only inline code and not more than this. Is it a valid optimization technique, then, for a program that requires it, that right before you create the release executable, you merge all your translation units into one big ugly .c file, so the compiler can have as much information as possible about all other functions while compiling?

Richard

I believe it should be tested, rather than having an opinion rendered.

Wait, I don't mean that you would keep your code like that. I mean, let's say you're developing a game and your source is in 1000 *.c files and 1000 *.h files. Right when it's time to make the executable that is going on the installation DVD, you create a temporary .c file, and put everything into one big file, send it through your compiler (will take ages), and then the resulting executable file is lean and mean.

Then you go back to your 2000 files and keep squashing bugs during the life cycle of the program. See what I mean? For some modern games, they try very hard to get an extra few frames per second. I wonder if this would work.

Recent versions of GCC can do whole-program optimization just like MSVC. Look into -flto and -fwhopr : LinkTimeOptimization - GCC Wiki. It's worth something like 5% in SPEC, which is actually pretty decent for such a well optimized-for benchmark suite.