Pass slices as not a struct, enable @restrict for T[]

[thewilsonator]

Exploring the idea of splitting T[] arguments into i(32|64), (T*|ptr) on the call level to enable adding @restrict to T[] arguments.

Currently (one among many) thing that fails is doesn't produce the correct alloca, it produces %a = alloca i64, align 8 and then only stores the length in a_arg_length. I can't seem to figure out where the code that produces the bad alloca comes from.

Still todo, runtime functions (_d_.*), test suite update, ABI, inlineIR, ...
Very much WIP.

Hmm, on second thoughts, it would be far less disruptive to do this only if @restrict is applied.

[JohanEngelen]

I don't know if it is worth changing ABI for this. Another approach would be to add preamble code that reads from the argument with norestrict on the read operation? (into a dummy local variable that can then be used for the rest of the function codegen).

[JohanEngelen]

see also: https://llvm.org/devmtg/2017-02-04/Restrict-Qualified-Pointers-in-LLVM.pdf

[JohanEngelen]

Did anyone ever ask about this issue on LLVM's maillist?

[thewilsonator]

we can limit the ABI changes to just functions that take @restrict, but that would break function pointers...