Issue 1829 - Allow labels as inline asm operands

[Orvid]

https://issues.dlang.org/show_bug.cgi?id=1829

This PR allows labels to be used as an operand for any op-code that would normally accept a memory location. The backend is able to handle it perfectly fine, but for some reason the frontend only lets you use labels as an operand to branching instructions. The use of this is quite simple: It's currently impossible to create a jump-table within inline-asm without creating a huge amount of boilerplate code resulting in a separate function for each and every case, as well as the default case, and the code after the default case, which is not, in any form of the word, maintainable. This PR makes the default segment for labels being resolved CS, however it does still allow you to manually override this segment, just as you would with a normal memory operand. Tests are included.

[yebblies]

This does not currently include a test for this syntax, as I would normally have put it in an inlineasm test file, but that won't be added until PR 3469 is merged.

What?

[Orvid]

That's just me being weird yebblies, instead I'm going to change the name of the test file in that PR, and name the one I'll add for this one inlineasmlabels.

[Orvid]

Just updated the PR again, and corrected segment prefix generation to error when trying to use a segment other than CS with branching instructions, because that is not valid generated code. The CS and DS segment prefixes actually use the same prefixes that the hint branch taken and hint branch not taken prefixes do.

[Orvid]

Alright, provided the tests pass, then this is now ready to be reviewed & merged, as it is now tested correctly.

[Orvid]

So, apparently this fixes the 6-year old issue 1829 which I didn't know existed until it was pointed out to me. I've updated the title and summary to reflect this.

[adamdruppe]

This looks cool to me, I want it.

[p0nce]

Very cool, this allows jump-tables!

[yebblies]

This looks good, but I don't know the iasm code that well.

[ghost]

Since most of you ASM folks seem to want this, I think we're a go.

[ghost]

Auto-merge toggled on

[JohanEngelen]

@Orvid Why are there all these NOPs placed in the fail12635 testcase?

[Orvid]

It looks like it was copy-pasta from https://github.com/dlang/dmd/blob/75bccf2ab2cdcdcc9ce63be62a8511ea304c8403/test/fail_compilation/fail353.d

Eliminating them entirely, or switching them to a couple of actual nops would be a valid solution. fail353 needs them because it's specifically trying to get an overflow for a rel8 encoded instruction, but fail12635 has no such constraints.