[PATCH] D159053: [RISCV] Support select/merge like ops for fp16 vectors when only have Zvfhmin

[Jianjian Guan via Phabricator via llvm-commits]

jacquesguan added inline comments.


================
Comment at: llvm/lib/Target/RISCV/RISCVISelLowering.cpp:13988
+    // For splat_vector to fp16 vectors when we only have zvfhmin, we promote it
+    // to fp_round(vf32 splat_vector(c)). But the constant folder of getNode
+    // would fold fp_round(vf32 splat_vector(c)) back to vf16 splat_vector(c)
----------------
michaelmaitland wrote:
> How does this approach compare to making the constant folder of getNode not do the fold when only have zvfhmin?
I have no idea about how to make the constant folder of getNode not do the fold when only have zvfhmin if we just want to change the RISCV target specific code. Do you have some advice?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D159053/new/

https://reviews.llvm.org/D159053