core.int128 alignment is adjusted

[WalterBright]

As alignment is now implemented per #14764 the code generated trying to align everything to 16 bytes when the stack is aligned to less is just terrible. I see no reason for alignment to be more than the stack alignment, so this changes the alignment of Cent to match the stack.

[dlang-bot]

Thanks for your pull request, @WalterBright!

Bugzilla references

Your PR doesn't reference any Bugzilla issue.

If your PR contains non-trivial changes, please reference a Bugzilla issue or create a manual changelog.

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub run digger -- build "master + dmd#14768"

[maxhaton]

Why is it explicitly aligned at all?

[ibuclaw]

Doesn't match gdc or ldc.

[ibuclaw]

Why is it explicitly aligned at all?

Indeed, if a compiler has a specific requirement, then the alignment should be dictated by the compiler, not the library.

As for your question, I expect to match native int128 alignment. Could be a __traits that asks the Target, preferably something that allows querying any sort of intN types.

[WalterBright]

I don't really know what change you are requesting. The compiler is using the predefined versions to tell it what alignment to use.

[WalterBright]

I changed it so the alignment that works best with D is under version (DigitalMars)

Made alignment changes only for DigitalMars

[RazvanN7]

@ibuclaw is this ok?

[ibuclaw]

Dmd would no longer be abi compatible with native int128 as a result of this. If everyone is fine letting dmd to continue shooting itself, then OK.

[WalterBright]

I don't know why the native ABI would require it be aligned on 16 bytes anyway because the instructions to implement don't need it.

Be that as it may, for someone who is trying to interface it with the native ABI, for those instances that need it, align(16) in front of those instances will do the trick. Assuming the alignment PR also gets pulled.

[s-ludwig]

Looks like these last two cases are interchanged - this should be 8 and the else branch below should be 4. However, doesn't X86_64 imply D_SIMD anyway? X86-64 has SSE2 support in its base spec.

[WalterBright]

The layout exactly matches what the dmd compiler does.

[s-ludwig]

The comment above:

is64bit ? 8 : 4

vs.

version (X86_64)
    private enum Cent_alignment = 4;
else
    private enum Cent_alignment = 8;

looks to be in contradiction to me.

[WalterBright]

I see it now. Oops. good catch

[ljmf00]

Why is it explicitly aligned at all?

To be ABI complaint. See System V spec on __int128 C type.

[maxhaton]

Why is it explicitly aligned at all?

To be ABI complaint. See System V spec on __int128 C type.

This isn't compliance should be noted in the file if this is intended.

[WalterBright]

There is no reason for dmd to align this. This PR needs to be pulled.