[NFC][NVPTX] Cleanup getPreferredVectorAction()

[justinfargnoli]

v2*16 is a legal type in NVPTX. Thus, this is dead code.

[llvmbot]

@llvm/pr-subscribers-backend-nvptx

Author: Justin Fargnoli (justinfargnoli)

Changes

v2*16 is a legal type in NVPTX. Thus, rely on the default implementation of getPreferredVectorAction().

---

Full diff: https://github.com/llvm/llvm-project/pull/114115.diff

1 Files Affected:

• (modified) llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp (-2)

diff --git a/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp b/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
index 57bc5fe0ac361c..377cf8a722b84e 100644
--- a/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
+++ b/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
@@ -1337,8 +1337,6 @@ NVPTXTargetLowering::getPreferredVectorAction(MVT VT) const {
   if (!VT.isScalableVector() && VT.getVectorNumElements() != 1 &&
       VT.getScalarType() == MVT::i1)
     return TypeSplitVector;
-  if (Isv2x16VT(VT))
-    return TypeLegal;
   return TargetLoweringBase::getPreferredVectorAction(VT);
 }
 

[AlexMaclean]

What are the existing tests that verify the behavior is unchanged here?

[justinfargnoli]

What are the existing tests that verify the behavior is unchanged here?

Any test that uses v2*16 types.

• bf16-instructions.ll
• bf16x2-instructions.ll
• bf16x2-instructions-approx.ll
• convert-sm80.ll
• bswap.ll
• ...