[clang] [llvm] [RISCV] Add support for quadruple-precision floating point ABIs (PR #195166)

[Gábor Spaits via cfe-commits]

spaits wrote:

If we disable the `addRegisterClass` the quad-prec fp registers in ISelLowering, then we will basically have the quad-prec. ABI string, but fp128 will be passed as soft float. That is bad.

If we enable it. It crashes. That's even worse.

To add these ABIs the correct way, tested/stable on not just GISel, but on SelDAG to I will have to add some minimal code gen for Q. To be more precise minimal code gen:
- so quad-prec. fp + non quad-prec. ABI will work together
- Addressing the needs of `llvm/test/CodeGen/RISCV/calling-conv-*` tests. (That is going to be `fptosi fp128` codegen).


https://github.com/llvm/llvm-project/pull/195166