[PATCH] D24021: Fix code-gen crash on Power9 when lowering insert_vector_elt node with variable index (PR30189)

[Nemanja Ivanovic via llvm-commits]

nemanjai created this revision.
nemanjai added reviewers: hfinkel, kbarton, amehsan.
nemanjai added a subscriber: llvm-commits.
nemanjai set the repository for this revision to rL LLVM.
Herald added a subscriber: nemanjai.

When I implemented the XXINSERTW instruction I forgot to account for variable insert indices. This patch fixes PR30189.

Repository:
  rL LLVM

https://reviews.llvm.org/D24021

Files:
  lib/Target/PowerPC/PPCISelLowering.cpp
  lib/Target/PowerPC/PPCISelLowering.h
  test/CodeGen/PowerPC/p9-xxinsertw-xxextractuw.ll

Index: test/CodeGen/PowerPC/p9-xxinsertw-xxextractuw.ll
===================================================================
--- test/CodeGen/PowerPC/p9-xxinsertw-xxextractuw.ll
+++ test/CodeGen/PowerPC/p9-xxinsertw-xxextractuw.ll
@@ -968,3 +968,25 @@
   %vecins = shufflevector <4 x float> %a, <4 x float> %a, <4 x i32> <i32 0, i32 1, i32 2, i32 6>
   ret <4 x float> %vecins
 }
+define <4 x float> @insertVarF(<4 x float> %a, float %f, i32 %el) {
+entry:
+; CHECK-LABEL: insertVarF
+; CHECK: stxsspx 1,
+; CHECK: lxvd2x
+; CHECK-BE-LABEL: insertVarF
+; CHECK-BE: stxsspx 1,
+; CHECK-BE: lxvw4x
+  %vecins = insertelement <4 x float> %a, float %f, i32 %el
+  ret <4 x float> %vecins
+}
+define <4 x i32> @insertVarI(<4 x i32> %a, i32 %i, i32 %el) {
+entry:
+; CHECK-LABEL: insertVarI
+; CHECK: stwx
+; CHECK: lxvd2x
+; CHECK-BE-LABEL: insertVarI
+; CHECK-BE: stwx
+; CHECK-BE: lxvw4x
+  %vecins = insertelement <4 x i32> %a, i32 %i, i32 %el
+  ret <4 x i32> %vecins
+}
Index: lib/Target/PowerPC/PPCISelLowering.h
===================================================================
--- lib/Target/PowerPC/PPCISelLowering.h
+++ lib/Target/PowerPC/PPCISelLowering.h
@@ -828,6 +828,7 @@
     SDValue LowerSRA_PARTS(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) const;
Index: lib/Target/PowerPC/PPCISelLowering.cpp
===================================================================
--- lib/Target/PowerPC/PPCISelLowering.cpp
+++ lib/Target/PowerPC/PPCISelLowering.cpp
@@ -665,9 +665,10 @@
       addRegisterClass(MVT::v2i64, &PPC::VRRCRegClass);
       addRegisterClass(MVT::v1i128, &PPC::VRRCRegClass);
     }
+
     if (Subtarget.hasP9Vector()) {
-      setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4i32, Legal);
-      setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4f32, Legal);
+      setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4i32, Custom);
+      setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4f32, Custom);
     }
   }
 
@@ -7861,6 +7862,17 @@
   return DAG.getLoad(Op.getValueType(), dl, Store, FIdx, MachinePointerInfo());
 }
 
+SDValue PPCTargetLowering::LowerINSERT_VECTOR_ELT(SDValue Op,
+                                                  SelectionDAG &DAG) const {
+  assert(Op.getOpcode() == ISD::INSERT_VECTOR_ELT &&
+         "Should only be called for ISD::INSERT_VECTOR_ELT");
+  ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(2));
+  // We have legal lowering for constant indices but not for variable ones.
+  if (C)
+    return Op;
+  return SDValue();
+}
+
 SDValue PPCTargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,
                                                    SelectionDAG &DAG) const {
   SDLoc dl(Op);
@@ -8263,6 +8275,7 @@
   case ISD::SCALAR_TO_VECTOR:   return LowerSCALAR_TO_VECTOR(Op, DAG);
   case ISD::SIGN_EXTEND_INREG:  return LowerSIGN_EXTEND_INREG(Op, DAG);
   case ISD::EXTRACT_VECTOR_ELT: return LowerEXTRACT_VECTOR_ELT(Op, DAG);
+  case ISD::INSERT_VECTOR_ELT:  return LowerINSERT_VECTOR_ELT(Op, DAG);
   case ISD::MUL:                return LowerMUL(Op, DAG);
 
   // For counter-based loop handling.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D24021.69653.patch
Type: text/x-patch
Size: 3500 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160830/9c4a70b2/attachment.bin>