[PATCH] D112102: [RISCV] Reduce the number of RISCV vector builtins by an order of magnitude.

Hsiangkai Wang via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Wed Oct 20 04:08:52 PDT 2021


HsiangKai added inline comments.


================
Comment at: llvm/test/CodeGen/RISCV/rvv/vmerge-rv32.ll:1358
+; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, ta, mu
+; CHECK-NEXT:    vmerge.vvm v8, v8, v9, v0
+; CHECK-NEXT:    ret
----------------
frasercrmck wrote:
> HsiangKai wrote:
> > vmerge.vvm is for integer vectors, doesn't it? Why does it work for floating point vector types?
> It's just a `vselect` in llvm terms, so doesn't care what its (vector) inputs are. The only reason we need `vfmerge` is when one operand is a scalar floating-point register.
Got it. I think I am misled by the title "Vector Integer Merge Instructions" in the spec. Thanks.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112102/new/

https://reviews.llvm.org/D112102



More information about the llvm-commits mailing list