Add Compiler-RT for 128bit integer support

[jkthorne]

This is extracted from github.com//pull/7093.

This implements all the expected functions 128bit integer support.

[jkthorne]

how can I make this a draft?

[bcardiff]

You can’t. Draft can only be created and be marked as ready. No worries

[jkthorne]

@bcardiff can you take a look at this when you get a chance?

[asterite]

this PR is for Int128 but there is already a port of mulodi4 that is used to have Int64 overflow in 32 bits
compiler-rt is needed for 32 bit machines

But in #8283

Meanwhile, we are near to drop 32bits packages (and CI). LLVM 8 is not even available in their apt repo for 32 bits

So all the work we (as a community) put into compiler-rt go into:

• Int128, for which there are no use cases and nobody asked it (I added it in the past because it seemed easy, but I was wrong)
• Some operations in 32 bits machines, which will soon not be supported anymore

That's why I question why we need to continue putting effort in these things. It's not that I don't find this PR useful in its own, but given the context where it could be applied it seems to provide very little value.

[asterite]

Regarding the many specs, I don't think we need one spec per operation. We can put all specs inside a single it block... if having multiple specs is a concern (I don't think it poses any problems).

[jkthorne]

I was doing a project with Int128 integers a while ago and found that this was uncompleted.
I would like 128bit integer support in crystal and it seems like the majority if not all the work so far has gone into supporting 32 bit machines.

[asterite]

@wontruefree What did you need Int128 in your project for?

[jkthorne]

@asterite I was integrating with a cpp library that was using Int128. I was looking into slices and other technical solutions but in the end I discontinued the project. I picked this up because I thought it was interesting and I thought it would help us get to 1.0.

[asterite]

Sorry for my interruption and discussion. If we can get this merged and support 32 bits and int128, I think this is the best result. Please go on!

[RX14]

Why are there so many commented out tests?

[jkthorne]

they are broken. The main problem is __udivmodti4 but it is hard to debug and I work on one failure at a time. I have gotten enough help from these PRs that I can close them and reopen them when I am closer if that helps.

[RX14]

If this is WIP, then could you please indicate that in the title?