[llvm] [RISCV] Fix incorrect codegen for Zfa with negated forms of constants in the lookup table (PR #68026)

[Alex Bradbury via llvm-commits]

asb wrote:

It looks like this was the cause of all the test failures I saw with the GCC torture suite and Zfa (not that the torture suite exercises it particularly well - only ~50 tests see changed codegen). I'll do a quick test with a few other simple code bases, but after that I don't think there are blockers on following up on https://github.com/llvm/llvm-project/pull/67964 with a PR to mark Zfa as non-experimental.

I wasn't planning on suggesting this for 17.0.2 as I don't think we have a history or expectation of backporting fixes to experimental extensions (and it might send the wrong message about the degree of support we offer for them), but let me know if you feel it would be worth doing.

https://github.com/llvm/llvm-project/pull/68026