[llvm] [LegalizeTypes][AMDGPU]: Allow for scalarization of insert_subvector (PR #104236)

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Thu Aug 15 07:54:48 PDT 2024


================
@@ -889,10 +889,13 @@ SDValue DAGTypeLegalizer::ScalarizeVecOp_CONCAT_VECTORS(SDNode *N) {
 /// instead.
 SDValue DAGTypeLegalizer::ScalarizeVecOp_INSERT_SUBVECTOR(SDNode *N,
                                                           unsigned OpNo) {
-  auto Elt = GetScalarizedVector(N->getOperand(OpNo));
-  auto VecOp = N->getOperand(1 - OpNo);
-  return DAG.getNode(ISD::INSERT_VECTOR_ELT, SDLoc(N), VecOp.getValueType(),
-                     VecOp, Elt, N->getOperand(2));
+  // We should not be attempting to scalarize the containing vector
+  assert(OpNo == 1);
+  SDValue Elt = GetScalarizedVector(N->getOperand(OpNo));
+  SDValue ContainingVec = N->getOperand(1 - OpNo);
----------------
arsenm wrote:

Might as well assume 0 if you're asserting above 

https://github.com/llvm/llvm-project/pull/104236


More information about the llvm-commits mailing list