That feature flag is merely a macro in the Linux kernel source code, and doesn't appear to be exposed to userspace. It is entirely different from the kind of flags under discussion, which are in the return values of the CPUID instruction and available for any program to query.

from the patch:

The enhancement applies to string lengths between 1 and 128 bytes long. Support for fast-short REP MOVSB is enumerated by the CPUID feature flag: CPUID [EAX=7H, ECX=0H).EDX.FAST_SHORT_REP_MOVSB[bit 4] = 1

So there is indeed a CPUID feature flag for fast rep movsb.

Which flag should I check for fast PEXT?

See [1], or [2] and [3].

[1] https://github.com/as-com/varint-simd/blob/f4c4af672403cc683...

[2] https://en.wikipedia.org/wiki/X86_Bit_manipulation_instructi...

[3] https://en.wikipedia.org/wiki/List_of_AMD_CPU_microarchitect...

So... ignore the cpuid bit indicating the instruction is present and look at manufacturer string, which is exactly what intel is being lambasted for here.

No, you don't ignore the CPUID feature bit. You use the CPUID to enable the feature by default, but override that default by blacklisting specific models that you know would have undesirable implementations.

So what you would do for PDEP/PEXT would be

  use_pdep = cpuid.supports_bmi2() && !cpuid.is_amd_before_zen3()

and not

  use_pdep = cpuid.is_intel_haswell_or_later()