FIX: SM4 GFNI needs AVX2

[reneme]

This module uses AVX2 along with GFNI and should note that in its info.txt. Not sure we want to do that in the long-term, though. Perhaps better to implement this as a special case in configure.py, to disable GFNI when AVX2 was explicitly disabled with --disable-avx2? Are there other examples of such dependencies that need some sort of generalization, perhaps?

[coveralls]

coverage: 91.238% (-1.2%) from 92.433%
when pulling 07d0556 on Rohde-Schwarz:fix/no_avx2
into 6471abd on randombit:master.

[randombit]

Are there other examples of such dependencies that need some sort of generalization, perhaps

Probably yeah. VAES for instance is defined as VAES+AVX2 bits.

We already handle disabling these at runtime in CPUID cleanly (if you use BOTAN_CLEAR_CPUID=avx2 it clears GFNI, VAES, etc also) but do not have similar logic for build time.

[randombit]

Addresses the immediate issue so fine with me.

I’ll think about what to do here in the general case.

[randombit]

I realize now this doesn’t actually address the real issue

We should have a <dep> on simd_avx2. Otherwise with or without this isa bit, if you do configure.py —disable-modules=simd_avx2 this will fail to build.

[reneme]

Makes sense. Otherwise it won't find the avx-related helper headers that are part of simd_avx2, I reckon. Updated.