[llvm] [AMDGPU] In VectorLegalizer::Expand, if UnrollVectorOp returns Load, … (PR #88475)

[via llvm-commits]

================
@@ -1159,8 +1159,12 @@ void VectorLegalizer::Expand(SDNode *Node, SmallVectorImpl<SDValue> &Results) {
   }
 
   SDValue Unrolled = DAG.UnrollVectorOp(Node);
-  for (unsigned I = 0, E = Unrolled->getNumValues(); I != E; ++I)
-    Results.push_back(Unrolled.getValue(I));
+  const LoadSDNode *Ld = dyn_cast<LoadSDNode>(Unrolled.getNode());
+  if (Ld)
+    Results.push_back(Unrolled);
+  else
+    for (unsigned I = 0, E = Unrolled->getNumValues(); I != E; ++I)
+      Results.push_back(Unrolled.getValue(I));
----------------
choikwa wrote:

The SDag looks like this
  t39: v2i32,ch = load<(dereferenceable invariant load (s64) from %ir..kernarg.offset, align 4, addrspace 4)> t0, t11, undef:i64
              t37: i32 = extract_vector_elt t39, Constant:i32<0>
            t38: i16 = truncate t37
              t35: i32 = extract_vector_elt t39, Constant:i32<1>
            t36: i16 = truncate t35
          t20: v2i16 = BUILD_VECTOR t38, t36
        t32: v2i32 = any_extend t20
        
It looks like the logic after 'assert(N->getNumValues() == 1 ' is responsible for returning load.

https://github.com/llvm/llvm-project/pull/88475