[PATCH] D136640: [RISCV][DAG] vslideup/vslidedown with zero offset and undef pass through is a nop

[Craig Topper via Phabricator via llvm-commits]

craig.topper added inline comments.


================
Comment at: llvm/lib/Target/RISCV/RISCVISelLowering.cpp:9888
+  case RISCVISD::VSLIDEUP_VL:
+    // vslidedown.vi undef, src, 0 -> src
+    // vslideup.vi   undef, src, 0 -> src
----------------
craig.topper wrote:
> Is this something we should squash in lowering of insert_vector_elt instead?
These all seem to be coming from RV32 insert_vector_elt of i64. If the index is 0 we can use two tail undisturbed slide1ups with VL=2 if the vector to insert into isn't undef. If it is undef we just the slide1ups.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D136640/new/

https://reviews.llvm.org/D136640