[llvm] [RISC-V][MC] Accept an absolute variable value as a CSR number (PR #67377)

[Alex Bradbury via llvm-commits]

asb wrote:

> I'm also not a huge fan of the introduced ambiguity, but I confirm that GNU Binutils have the same behavior as the one proposed in this patch by selecting the CSR first. So, I'd be up to add a test to exhibit that behavior and accept this change since being able to run the riscv-arch-test test-suite would be valuable to all LLVM based RISCV toolchains in my opinion.

I agree, let's add a test case to document the behaviour but otherwise match binutils here.

https://github.com/llvm/llvm-project/pull/67377