This options compiles LLVM code by using llc to convert it to C, then using gcc to compile this C instead of using LLVM's own code generators. This generally results in better performance (particularly when using --profopt to use gcc's profile-directed optimizations) but takes rather longer and uses more RAM.