[llvm] r233778 - Fix PR23065. Avoid optimizing bitcast of build_vector with constant input to scalar_to_vector.

[Jiangning Liu]

Author: jiangning
Date: Tue Mar 31 20:52:38 2015
New Revision: 233778

URL: http://llvm.org/viewvc/llvm-project?rev=233778&view=rev
Log:
Fix PR23065. Avoid optimizing bitcast of build_vector with constant input to scalar_to_vector.


Added:
    llvm/trunk/test/CodeGen/AArch64/bitcast.ll
Modified:
    llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=233778&r1=233777&r2=233778&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Tue Mar 31 20:52:38 2015
@@ -7024,7 +7024,6 @@ ConstantFoldBITCASTofBUILD_VECTOR(SDNode
 
   // Finally, this must be the case where we are shrinking elements: each input
   // turns into multiple outputs.
-  bool isS2V = ISD::isScalarToVector(BV);
   unsigned NumOutputsPerInput = SrcBitSize/DstBitSize;
   EVT VT = EVT::getVectorVT(*DAG.getContext(), DstEltVT,
                             NumOutputsPerInput*BV->getNumOperands());
@@ -7042,10 +7041,6 @@ ConstantFoldBITCASTofBUILD_VECTOR(SDNode
     for (unsigned j = 0; j != NumOutputsPerInput; ++j) {
       APInt ThisVal = OpVal.trunc(DstBitSize);
       Ops.push_back(DAG.getConstant(ThisVal, DstEltVT));
-      if (isS2V && i == 0 && j == 0 && ThisVal.zext(SrcBitSize) == OpVal)
-        // Simply turn this into a SCALAR_TO_VECTOR of the new type.
-        return DAG.getNode(ISD::SCALAR_TO_VECTOR, SDLoc(BV), VT,
-                           Ops[0]);
       OpVal = OpVal.lshr(DstBitSize);
     }
 

Added: llvm/trunk/test/CodeGen/AArch64/bitcast.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/bitcast.ll?rev=233778&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/AArch64/bitcast.ll (added)
+++ llvm/trunk/test/CodeGen/AArch64/bitcast.ll Tue Mar 31 20:52:38 2015
@@ -0,0 +1,27 @@
+; RUN: llc < %s -mtriple=aarch64--linux-gnu | FileCheck %s
+
+; PR23065: SCALAR_TO_VECTOR implies the top elements 1 to N-1 of the N-element vector are undefined.
+
+define <4 x i16> @foo1(<2 x i32> %a) {
+; CHECK-LABEL: foo1:
+; CHECK:       movi	d0, #0000000000000000
+; CHECK-NEXT:  ret
+
+  %1 = shufflevector <2 x i32> <i32 58712, i32 undef>, <2 x i32> %a, <2 x i32> <i32 0, i32 2>
+; Can't optimize the following bitcast to scalar_to_vector.
+  %2 = bitcast <2 x i32> %1 to <4 x i16>
+  %3 = shufflevector <4 x i16> %2, <4 x i16> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
+  ret <4 x i16> %3
+}
+
+define <4 x i16> @foo2(<2 x i32> %a) {
+; CHECK-LABEL: foo2:
+; CHECK:       movi	d0, #0000000000000000
+; CHECK-NEXT:  ret
+
+  %1 = shufflevector <2 x i32> <i32 712, i32 undef>, <2 x i32> %a, <2 x i32> <i32 0, i32 2>
+; Can't optimize the following bitcast to scalar_to_vector.
+  %2 = bitcast <2 x i32> %1 to <4 x i16>
+  %3 = shufflevector <4 x i16> %2, <4 x i16> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
+  ret <4 x i16> %3
+}