[clang] [llvm] [LoongArch] Support sc.q instruction for 128bit cmpxchg operation (PR #116771)

[via cfe-commits]

https://github.com/heiher commented:

In addition, I propose adding test cases for atomic ops based on the native `cmpxchg128`. Due to the lack of the `ll.q` instruction, even `atomic128_read` needs to be implemented using `cmpxchg128 (ll.d + ld.d + sc.q)`. Alternatively, a vector 128-bit load can be used if `lsx` is available.

https://github.com/llvm/llvm-project/pull/116771