Move avo assembly generator to separate asm module

[mmcloughlin]

Move avo code-generator to an asm sub-module. This avoids the problem of the sha1cd module itself depending on avo.

Fixes #148

[pjbgf]

@mmcloughlin thank you for proposing this PR, that's much appreciated. Please amend the make verify target, so that it points to the new asm/asm.go which should fix the build.

[mmcloughlin]

@mmcloughlin thank you for proposing this PR, that's much appreciated. Please amend the make verify target, so that it points to the new asm/asm.go which should fix the build.

Thanks @pjbgf. I have pushed a new iteration with changes:

• Moved the ubc code generator to its own asm sub-module.
• Renamed the ubc/doc.go to ubc/ubc.go and added the go:generate line there too
• Updated make generate to just call go generate -x ./...
• Removed sed lines that add // +build constraints. I think go:build is sufficiently established that there's no need to keep the old form, unless you want to?

The only remaining thing I'd say is that the shared package doesn't make much sense anymore in this refactor at the moment. To avoid trying to share an internal package from the main module I just copied in the few constants needed by the assembly generator. So perhaps now internal/const.go should just be moved to the top-level package (and made private)?

[mmcloughlin]

Note they can't be named K0, K1, ... anymore since these are registers in avo (AVX-512 mask registers).

[pjbgf]

So perhaps now internal/const.go should just be moved to the top-level package (and made private)?

That makes sense. Overall, the changes LGTM. It is up to you if you prefer making the additional change for the shared package here, or on a follow-up PR.

[pjbgf]

@mmcloughlin I will be merging this as is, thank you once again for proposing this change. The shared package change can be done at a later time.

[mmcloughlin]

Great! Happy to help.

Yes the shared package wasn't a blocking issue.