To complement the LLVM 3.4 SVN compiler benchmarks from yesterday that were looking at the impact of using the SLP Vectorizer that's soon to be enabled by default for some optimization levels, here are some more LLVM Clang compiler development benchmarks. This time around are fresh benchmarks of the open-source C/C++ compiler when trying out the different compiler optimization levels, including -O0, -O1, -O2, -Os, -O3, and -Ofast.
From the LLVM, Clang, and Compiler-RT Subversion repositories as of this weekend, a fresh optimized build of LLVM/Clang 3.4 (no assertions, enable optimizations) was used for testing the -O0, -O1, -O2, -Os, -O3, and -Ofast optimization levels. Most Phoronix readers are well versed on the different optimization levels, but the higher the number, the more optimizations that are applied by the compiler. The -Os level is for optimizing the binary for size while -Ofast is the highest standard optimization level and relies upon the potentially unsafe fast-math.
These straightforward compiler benchmarks were done from a ThinkPad W510 laptop with Intel Core i7 720QM quad-core processor with Hyper Threading. Ubuntu 13.10 was running on the system with the Linux 3.11 kernel. All of the compiler benchmarks were done through the Phoronix Test Suite software.