Cool, makes sense. LLVM isn't a compiler, your benchmarks are out of date, I was referring to cross-toolchain portability, and the question was directed at Uber.
LLVM is the backend that Clang is based on. They should be taken together when talking about compilers. I chose to refer to it as LLVM because it's the basis on which Clang is built. And when I refer to performance optimizations, I talk about the whole results. Benchmarks can indicate advantages in specific areas, and in many areas, LLVM matches and sometimes beats GCC. But there are other areas where LLVM still has work to be done to match GCC. (Phoronix tends to have frequent comparisons between the two. Currently GCC 4.8 holds the edge, but if you look at the improvements in LLVM over the years, it's clear which will eventually come out on top.) But most importantly, GCC is the older, more mature software, and if you are building mission-critical code, you always use the more reliable compiler. And with regards to cross-toolchain compatibility, last I checked, LLVM could be swapped with GCC with no difference whatsoever. (Well, there are some commands GCC has that LLVM lacks, but they tend to be more esoteric.) Also, this is a forum, and as such it encourages discussion. Asking a question of one person doesn't mean you won't get explanations from others. But that doesn't stop one from being POLITE about it . . .
Le Bump. http://www.alexstjohn.com/WP/2013/07/22 ... -direct3d/ In interesting blog by the founder of Direct3D.