[llvm] r364295 - [VectorLegalizer] ExpandANY_EXTEND_VECTOR_INREG/ExpandZERO_EXTEND_VECTOR_INREG - widen source vector

[Simon Pilgrim via llvm-commits]

Author: rksimon
Date: Tue Jun 25 04:31:37 2019
New Revision: 364295

URL: http://llvm.org/viewvc/llvm-project?rev=364295&view=rev
Log:
[VectorLegalizer] ExpandANY_EXTEND_VECTOR_INREG/ExpandZERO_EXTEND_VECTOR_INREG - widen source vector

The *_EXTEND_VECTOR_INREG opcodes were relaxed back around rL346784 to support source vector widths that are smaller than the output - it looks like the legalizers were never updated to account for this.

This patch inserts the smaller source vector into an undef vector of the same width of the result before performing the shuffle+bitcast to correctly handle this.

Part of the yak shaving to solve the crashes from rL364264 and rL364272

Modified:
    llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp

Modified: llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp?rev=364295&r1=364294&r2=364295&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp Tue Jun 25 04:31:37 2019
@@ -949,6 +949,19 @@ SDValue VectorLegalizer::ExpandANY_EXTEN
   EVT SrcVT = Src.getValueType();
   int NumSrcElements = SrcVT.getVectorNumElements();
 
+  // *_EXTEND_VECTOR_INREG SrcVT can be smaller than VT - so insert the vector
+  // into a larger vector type.
+  if (SrcVT.bitsLE(VT)) {
+    assert((VT.getSizeInBits() % SrcVT.getScalarSizeInBits()) == 0 &&
+           "ANY_EXTEND_VECTOR_INREG vector size mismatch");
+    NumSrcElements = VT.getSizeInBits() / SrcVT.getScalarSizeInBits();
+    SrcVT = EVT::getVectorVT(*DAG.getContext(), SrcVT.getScalarType(),
+                             NumSrcElements);
+    Src = DAG.getNode(
+        ISD::INSERT_SUBVECTOR, DL, SrcVT, DAG.getUNDEF(SrcVT), Src,
+        DAG.getConstant(0, DL, TLI.getVectorIdxTy(DAG.getDataLayout())));
+  }
+
   // Build a base mask of undef shuffles.
   SmallVector<int, 16> ShuffleMask;
   ShuffleMask.resize(NumSrcElements, -1);
@@ -996,6 +1009,19 @@ SDValue VectorLegalizer::ExpandZERO_EXTE
   EVT SrcVT = Src.getValueType();
   int NumSrcElements = SrcVT.getVectorNumElements();
 
+  // *_EXTEND_VECTOR_INREG SrcVT can be smaller than VT - so insert the vector
+  // into a larger vector type.
+  if (SrcVT.bitsLE(VT)) {
+    assert((VT.getSizeInBits() % SrcVT.getScalarSizeInBits()) == 0 &&
+           "ZERO_EXTEND_VECTOR_INREG vector size mismatch");
+    NumSrcElements = VT.getSizeInBits() / SrcVT.getScalarSizeInBits();
+    SrcVT = EVT::getVectorVT(*DAG.getContext(), SrcVT.getScalarType(),
+                             NumSrcElements);
+    Src = DAG.getNode(
+        ISD::INSERT_SUBVECTOR, DL, SrcVT, DAG.getUNDEF(SrcVT), Src,
+        DAG.getConstant(0, DL, TLI.getVectorIdxTy(DAG.getDataLayout())));
+  }
+
   // Build up a zero vector to blend into this one.
   SDValue Zero = DAG.getConstant(0, DL, SrcVT);