I have spent some time this week gathering detailed performance numbers on the compile time of both clang and llvm-gcc compared to gcc on two benchmarks, the Objective-C sample application Sketch and 176.gcc from SPEC INT 2000.
These numbers highlight some areas where we are can probably gain performance in Clang. For example, our LLVM IR generation is substantially slower than llvm-gcc’s (which should be doing a similar amount of work). Additionally, it is clear that we still have some work to do if we intend to match gcc’s precompiled header performance.
If you are interested, check it out at http://clang.llvm.org/performance.html