AMD Zen 3 Performance With The Initial "znver3" GCC Compiler Support

With the basic SciMark 2 benchmark that tends to be quite exposed to compiler changes, using the Znver3 build compared to Znver2 yielded about a 3.5% improvement to the performance. Or compared to the generic x86-64 build was a 21% improvement for this synthetic test case.

For many test cases though, the Znver3 performance was basically right in line with Znver2 and the other recent CPU family targets. In the case of the libraw RAW image decoder library, using any of these recent targets over generic x86-64 yields around a 16% boost. This roughly aligns with what we are used to seeing since the base x86-64 instruction set doesn't mandate AVX/AVX2 or other common, modern instruction set extensions.

Znver3 tended to come out in front, but by razor-thin margins. It will be more interesting when AMD publishes an optimized scheduler model and any other improvements to their GCC znver3 compiler code.

With GraphicsMagick is one of the areas where the Intel Haswell/Skylake processors can provide some advantages.