GCC 9 Compiler Tuning Benchmarks On Intel Skylake AVX-512
In applications that are already hand-tuned and optimized for particular microarchitectures, there isn't much to gain out of the compiler flags.
Google's libvpx vpxenc encoder did get boosted by the Skylake-AVX512 targeting.
LTO did help the performance with Himeno.
Understandably, the more optimization passes that need to run, the longer the compilation process.