Hard code endianness

[brycekahle]

This results in a reduction of the instructions necessary for endian-related operations. The wall clock time difference for a single call is miniscule (picoseconds).

An example, with the ByteOrder.Uint32 function:
Runtime instructions: https://godbolt.org/z/7vW1Pfnso
Buildtime instructions: https://godbolt.org/z/baxW1Wb7W

The function call on amd64 changes from:

MOVQ    "".NativeEndian(SB), CX
MOVQ    "".NativeEndian+8(SB), DX
MOVQ    64(CX), CX
MOVQ    AX, BX
MOVL    $4, DI
MOVQ    DX, AX
MOVQ    CX, DX
MOVQ    DI, CX
CALL    DX

to

XCHGL   AX, AX

The difference on arm64 is equivalent, but uses different specific instructions.

[lmb]

Interesting, thanks!

[lmb]

I'm getting the following when trying to pull your branch:

$ git pull [email protected]:DataDog/ebpf.git bryce.kahle/endian-build-time
From github.com:DataDog/ebpf
 * branch            bryce.kahle/endian-build-time -> FETCH_HEAD
fatal: refusing to merge unrelated histories

Similar result when checking out via github CLI, it only shows me a single commit with all source code smushed into one blob. Did you do any unorthodox git mangling on your repo that could cause this?

[brycekahle]

Did you do any unorthodox git mangling on your repo that could cause this?

I don't believe so. Did you try gh pr checkout 649?

[lmb]

I tried pushing a fixup, but it seems like you have maintainer edits disabled? Anyways, I removed the generic fallback and added you good writeup to the commit message: 1a5ff41 Let me know if you want to force push that into your branch or if I should make a new PR.

[brycekahle]

like you have maintainer edits disabled?

Apparently this isn't supported for org-owned forks at the moment.

Let me know if you want to force push that into your branch or if I should make a new PR.

Whatever is easier for you, is fine with me.

[lmb]

See #662. Can you use a personal fork in the future?

[brycekahle]

Can you use a personal fork in the future?

Yeah. That shouldn't be a problem.