[llvm] [SelectionDAG][RISCV] Promote VECREDUCE_{FMAX,FMIN,FMAXIMUM,FMINIMUM} (PR #128800)

[Jim Lin via llvm-commits]

================

----------------
tclin914 wrote:

`llvm.vector.reduce.fadd.nxv1f16` in vreductions-fp-sdnode.ll is unable to be promoted/expanded when just zvfhmin is enabled. That's why I create a new file with `-f16` suffix for `vector.reduce.{FMAX,FMIN,FMAXIMUM,FMINIMUM}` , which can correctly promote f16 reductions to f32 (but `vector.reduce.fadd`  cannot). Are there better solutions for this? Thanks.

https://github.com/llvm/llvm-project/pull/128800