Extensive Benchmarks Looking At AMD Znver1 GCC 9 Performance, EPYC Compiler Tuning
To no surprise, the newer compiler releases due take longer time to compile due to adding more optimization passes and other features that end up generally consuming more time to compile code with hopes of yielding faster performance for the resulting binary.
The C-Ray ray-tracer saw a regression on GCC 7 but now GCC 9 is seeing similar performance to GCC 6 and 8 series.
In total I ended up running dozens of benchmarks across these tested compiler configurations, but with most of them being uneventful compared to GCC8, here is a look at the geometric mean for all of that data of the test profiles successfully tested across all compiler combinations. This end composite shows the GCC 9 performance to be in line with GCC 8 and for Znver1 tuning to be about 2% faster using this geometric mean data. GCC 6 does come across as faster than the newer compilers due to the multiple GraphicsMagick tests carried out where there has been a significant performance regression since GCC 7. At least aside from GraphicsMagick, there weren't any other significant GCC performance regressions to note.
Next in this article is looking closer at the various optimization options for AMD EPYC on GCC 9.