AMD AOCC 2.3 Squeezing Out Extra Performance For EPYC Over GCC 10, Clang 11

[pyler]

In 2020, GCC produces slower binaries than LLVM.

[AndyChow]

Very very impressive results. Only area that seems weaker is the graphic stuff. Might just need a little tweak. The NCNN results are amazing.

[Paradigm Shifter]

AMD seem to be pulling it all together, slowly but surely. I vaguely remember seeing some early AOCC benchmarks which were atrocious.

[sdack]

For anyone who saw the issue with the L1/L2 cache parameters when using -march=znver2 versus -march=native, do know this issues has been resolved. It was apparently a cosmetic problem. In short, "nothing to see here", in more than one sense.

https://gcc.gnu.org/git/?p=gcc.git;a...30ade58c312acd

[ctopper]

Which AMD-specific instructions will cause that? Typically AMD is a bit behind Intel with respect to supporting various instruction sets (and that's not a criticism). There are definitely system instructions (mostly running only in privileged mode) where they are different, but the question is which instructions that a compiler like gcc will generate will cause an incompatibility.

I could have been more specific - Intel Core 10th/11th gen.

The AMD only instructions I know that the compiler can autogenerate are INSERTQ/EXTRQ from SSE4A.