Optimize builds for AMD Zen Platform

[4-FLOSS-Free-Libre-Open-Source-Software]

This enables -march=znver1 (MZEN), -march=znver2 (MZEN2), -march=znver3 (MZEN3) or -march=znver4 (MZEN4) to be set for the kernel's make process, which will optimize the kernel's binary for the target CPU architecture.

• Optimize builds for AMD Zen Platform

[taalojarvi]

JFYI I maintain a fork of Locietta's Xanmod kernel for AMD Zen platforms.. Apart from very few changes from my side, its close to 1:1 in case you need to test builds.

[Andarwinux]

I suspect this may not make sense at all, kbuild removes the increased ISA capabilities of march via CFLAGS and the kernel doesn't use autovec, so march-specific SIMD scheduling tuning doesn't make sense either.

[4-FLOSS-Free-Libre-Open-Source-Software]

in case you need to test builds.

Test builds can be found in artifacts:

• Artifact bzImage-zen has been successfully uploaded! Final size is 19737380 bytes. Artifact ID is 2217251658

• Artifact download URL: https://github.com/Locietta/xanmod-kernel-WSL2/actions/runs/11948058105/artifacts/2217251658

• Artifact bzImage-zen2 has been successfully uploaded! Final size is 19737024 bytes. Artifact ID is 2217251661

• Artifact download URL: https://github.com/Locietta/xanmod-kernel-WSL2/actions/runs/11948058105/artifacts/2217251661

• Artifact bzImage-zen3 has been successfully uploaded! Final size is 19737931 bytes. Artifact ID is 2217251257

• Artifact download URL: https://github.com/Locietta/xanmod-kernel-WSL2/actions/runs/11948058105/artifacts/2217251257

• Artifact bzImage-zen4 has been successfully uploaded! Final size is 19737353 bytes. Artifact ID is 2217251670

• Artifact download URL: https://github.com/Locietta/xanmod-kernel-WSL2/actions/runs/11948058105/artifacts/2217251670

I suspect this may not make sense at all, kbuild removes the increased ISA capabilities of march via CFLAGS

Good hint. The kernel build system (kbuild) may override or strip out certain flags. Perhaps we need benchmarks to weigh the potential benefits?

and the kernel doesn't use autovec, so march-specific SIMD scheduling tuning doesn't make sense either.

so the kernel does not heavily rely on SIMD (Single Instruction, Multiple Data) instructions, and many of the optimizations provided by -march should be more relevant to user-space applications.

[Locietta]

The -march build feature in Xanmod (and many other x86v3 kernels) is enabled by graysky's kernel compiler patch, they do have a few benchmarks there. There isn't significant difference between x86v3 and zen3 build though.

x86v4 (with AVX-512) doesn't make much sense for kernel either, but xanmod upstream ships x86v4 build anyway, so I just follow their choice. I'm also thinking about dropping skylake build, as I don't like too many arch-specific builds that most people not using. I don't feel like to add 4 more such builds...

Maybe I can just add scoop manifests for taalojarvi's kernel at https://github.com/Locietta/sniffer.

[4-FLOSS-Free-Libre-Open-Source-Software]

Maybe I can just add scoop manifests for taalojarvi's kernel at https://github.com/Locietta/sniffer.

Those forked builds are far too different from yours? Personally, I would prefer to stay with locietta-WSL2-xanmod.

[Andarwinux]

Some advanced architectures might not be affected by some of the older vulnerabilities, so it might still make sense to disable some of the unnecessary mitigations for those specific kernel, but this could bring extra maintenance burdens and security risks.