Compile time is one of the bottlenecks HFT engineers notice every day. A typical trading system codebase is several million lines of template-heavy C++. A clean rebuild on a 16-core developer box takes anywhere from five minutes (a minimal microservice) to forty (the full order book plus matching engine plus risk plus connectivity). Every dev triggers a clean build several times a week: at branch switches, after dependency bumps, when CI catches something they can’t reproduce locally. Twenty engineers losing fifteen minutes a day to compile waits is a senior engineer’s salary per year, paid in throughput.
That’s the backdrop for an idea that has been quietly circulating on technical blogs for years: the compiler binary itself can be optimized. GCC and Clang are programs like any others. They get compiled by some compiler. If that build process applies profile-guided optimization, link-time optimization, and tunes for the build host’s CPU, the resulting gcc binary runs faster when you invoke it to compile your code. The speedup compounds across every translation unit in your build.
Full article: https://hftuniversity.com/post/your-distro-already-pgo-d-your-compiler-so-what-does-super-gcc-actually-give-you
No posts

Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.