[PowerPC] Fix the assert of ISD::SIGN_EXTEND_INREG when type is v2i16 and v2i8

For ISD::SIGN_EXTEND_INREG operation of v2i16 and v2i8 types will cause assert because they are registered as custom operation.
So that the type legalization phase will enter the custom hook, which do not handle ISD::SIGN_EXTEND_INREG operation and fall throw into unreachable assert.

Patch By: wuzish (Zixuan Wu)
Differential Revision: https://reviews.llvm.org/D52449

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@344109 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/PowerPC/PPCISelLowering.cpp b/lib/Target/PowerPC/PPCISelLowering.cpp
index f4fd8e1..1fd5018 100644 (file)
--- a/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -792,13 +792,6 @@ PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
       setOperationAction(ISD::FP_TO_SINT, MVT::v2i64, Legal);
       setOperationAction(ISD::FP_TO_UINT, MVT::v2i64, Legal);
 
-      // Vector operation legalization checks the result type of
-      // SIGN_EXTEND_INREG, overall legalization checks the inner type.
-      setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i64, Legal);
-      setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i32, Legal);
-      setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i16, Custom);
-      setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i8, Custom);
-
       setOperationAction(ISD::FNEG, MVT::v4f32, Legal);
       setOperationAction(ISD::FNEG, MVT::v2f64, Legal);
       setOperationAction(ISD::FABS, MVT::v4f32, Legal);
@@ -9123,30 +9116,6 @@ SDValue PPCTargetLowering::LowerATOMIC_CMP_SWAP(SDValue Op,
   return DAG.getMemIntrinsicNode(NodeTy, dl, Tys, Ops, MemVT, MMO);
 }
 
-SDValue PPCTargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op,
-                                                  SelectionDAG &DAG) const {
-  SDLoc dl(Op);
-  // For v2i64 (VSX), we can pattern patch the v2i32 case (using fp <-> int
-  // instructions), but for smaller types, we need to first extend up to v2i32
-  // before doing going farther.
-  if (Op.getValueType() == MVT::v2i64) {
-    EVT ExtVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
-    if (ExtVT != MVT::v2i32) {
-      Op = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, Op.getOperand(0));
-      Op = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl, MVT::v4i32, Op,
-                       DAG.getValueType(EVT::getVectorVT(*DAG.getContext(),
-                                        ExtVT.getVectorElementType(), 4)));
-      Op = DAG.getNode(ISD::BITCAST, dl, MVT::v2i64, Op);
-      Op = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl, MVT::v2i64, Op,
-                       DAG.getValueType(MVT::v2i32));
-    }
-
-    return Op;
-  }
-
-  return SDValue();
-}
-
 SDValue PPCTargetLowering::LowerSCALAR_TO_VECTOR(SDValue Op,
                                                  SelectionDAG &DAG) const {
   SDLoc dl(Op);
@@ -9586,7 +9555,6 @@ SDValue PPCTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::VECTOR_SHUFFLE:     return LowerVECTOR_SHUFFLE(Op, DAG);
   case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
   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);

diff --git a/lib/Target/PowerPC/PPCInstrVSX.td b/lib/Target/PowerPC/PPCInstrVSX.td
index 1d991c5..b1cfbc7 100644 (file)
--- a/lib/Target/PowerPC/PPCInstrVSX.td
+++ b/lib/Target/PowerPC/PPCInstrVSX.td
@@ -1040,18 +1040,6 @@ def : Pat<(v2f64 (bitconvert v1i128:$A)),
 def : Pat<(v1i128 (bitconvert v2f64:$A)),
           (COPY_TO_REGCLASS $A, VRRC)>;
 
-// sign extension patterns
-// To extend "in place" from v2i32 to v2i64, we have input data like:
-// | undef | i32 | undef | i32 |
-// but xvcvsxwdp expects the input in big-Endian format:
-// | i32 | undef | i32 | undef |
-// so we need to shift everything to the left by one i32 (word) before
-// the conversion.
-def : Pat<(sext_inreg v2i64:$C, v2i32),
-          (XVCVDPSXDS (XVCVSXWDP (XXSLDWI $C, $C, 1)))>;
-def : Pat<(v2f64 (sint_to_fp (sext_inreg v2i64:$C, v2i32))),
-          (XVCVSXWDP (XXSLDWI $C, $C, 1))>;
-
 def : Pat<(v2f64 (PPCsvec2fp v4i32:$C, 0)),
           (v2f64 (XVCVSXWDP (v2i64 (XXMRGHW $C, $C))))>;
 def : Pat<(v2f64 (PPCsvec2fp v4i32:$C, 1)),

diff --git a/test/CodeGen/PowerPC/2018-09-19-sextinreg-vector-crash.ll b/test/CodeGen/PowerPC/2018-09-19-sextinreg-vector-crash.ll
new file mode 100644 (file)
index 0000000..4eb6555
--- /dev/null
+++ b/test/CodeGen/PowerPC/2018-09-19-sextinreg-vector-crash.ll
@@ -0,0 +1,24 @@
+; RUN: llc -verify-machineinstrs < %s -mtriple=powerpc64le-unknown-unknown -mcpu=pwr8 | FileCheck %s
+
+; Ensure this does not crash
+
+define <2 x i8> @test1(<2 x i8> %a) {
+  %1 = shl nuw <2 x i8> %a, <i8 7, i8 7>
+  %2 = ashr exact <2 x i8> %1, <i8 7, i8 7>
+  ret <2 x i8> %2
+}
+; CHECK-LABEL: @test1
+; CHECK: vspltisb [[REG1:[0-9]+]], 7
+; CHECK: vslb [[REG2:[0-9]+]], 2, [[REG1]]
+; CHECK: vsrab [[REG3:[0-9]+]], [[REG2]], [[REG1]]
+
+define <2 x i16> @test2(<2 x i16> %a) {
+  %1 = shl nuw <2 x i16> %a, <i16 15, i16 15>
+  %2 = ashr exact <2 x i16> %1, <i16 15, i16 15>
+  ret <2 x i16> %2
+}
+
+; CHECK-LABEL: @test2
+; CHECK: vspltish [[REG1:[0-9]+]], 15
+; CHECK: vslh [[REG2:[0-9]+]], 2, [[REG1]]
+; CHECK: vsrah [[REG3:[0-9]+]], [[REG2]], [[REG1]]