LegalizeTypes support for VSETCC. Fixes PR2575.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53938 91177308-0d34-0410-b5e6-96231b3b80d8

lib/CodeGen/SelectionDAG/LegalizeTypes.h

@@ -418,6 +418,7 @@ private:
   SDOperand ScalarizeVecRes_SELECT(SDNode *N);
   SDOperand ScalarizeVecRes_UNDEF(SDNode *N);
   SDOperand ScalarizeVecRes_VECTOR_SHUFFLE(SDNode *N);
+  SDOperand ScalarizeVecRes_VSETCC(SDNode *N);
 
   // Vector Operand Scalarization: <1 x ty> -> ty.
   bool ScalarizeVectorOperand(SDNode *N, unsigned OpNo);
@@ -434,19 +435,19 @@ private:
 
   // Vector Result Splitting: <128 x ty> -> 2 x <64 x ty>.
   void SplitVectorResult(SDNode *N, unsigned OpNo);
+  void SplitVecRes_BinOp(SDNode *N, SDOperand &Lo, SDOperand &Hi);
+  void SplitVecRes_UnOp(SDNode *N, SDOperand &Lo, SDOperand &Hi);
 
-  void SplitVecRes_UNDEF(SDNode *N, SDOperand &Lo, SDOperand &Hi);
+  void SplitVecRes_BIT_CONVERT(SDNode *N, SDOperand &Lo, SDOperand &Hi);
-  void SplitVecRes_LOAD(LoadSDNode *N, SDOperand &Lo, SDOperand &Hi);
   void SplitVecRes_BUILD_PAIR(SDNode *N, SDOperand &Lo, SDOperand &Hi);
-  void SplitVecRes_INSERT_VECTOR_ELT(SDNode *N, SDOperand &Lo, SDOperand &Hi);
-  void SplitVecRes_VECTOR_SHUFFLE(SDNode *N, SDOperand &Lo, SDOperand &Hi);
-
   void SplitVecRes_BUILD_VECTOR(SDNode *N, SDOperand &Lo, SDOperand &Hi);
   void SplitVecRes_CONCAT_VECTORS(SDNode *N, SDOperand &Lo, SDOperand &Hi);
-  void SplitVecRes_BIT_CONVERT(SDNode *N, SDOperand &Lo, SDOperand &Hi);
-  void SplitVecRes_UnOp(SDNode *N, SDOperand &Lo, SDOperand &Hi);
-  void SplitVecRes_BinOp(SDNode *N, SDOperand &Lo, SDOperand &Hi);
   void SplitVecRes_FPOWI(SDNode *N, SDOperand &Lo, SDOperand &Hi);
+  void SplitVecRes_INSERT_VECTOR_ELT(SDNode *N, SDOperand &Lo, SDOperand &Hi);
+  void SplitVecRes_LOAD(LoadSDNode *N, SDOperand &Lo, SDOperand &Hi);
+  void SplitVecRes_UNDEF(SDNode *N, SDOperand &Lo, SDOperand &Hi);
+  void SplitVecRes_VECTOR_SHUFFLE(SDNode *N, SDOperand &Lo, SDOperand &Hi);
+  void SplitVecRes_VSETCC(SDNode *N, SDOperand &Lo, SDOperand &Hi);
 
   // Vector Operand Splitting: <128 x ty> -> 2 x <64 x ty>.
   bool SplitVectorOperand(SDNode *N, unsigned OpNo);

lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

@@ -46,9 +46,10 @@ void DAGTypeLegalizer::ScalarizeVectorResult(SDNode *N, unsigned ResNo) {
   case ISD::FPOWI:          R = ScalarizeVecRes_FPOWI(N); break;
   case ISD::INSERT_VECTOR_ELT: R = ScalarizeVecRes_INSERT_VECTOR_ELT(N); break;
   case ISD::LOAD:           R = ScalarizeVecRes_LOAD(cast<LoadSDNode>(N));break;
-  case ISD::VECTOR_SHUFFLE: R = ScalarizeVecRes_VECTOR_SHUFFLE(N); break;
   case ISD::SELECT:         R = ScalarizeVecRes_SELECT(N); break;
   case ISD::UNDEF:          R = ScalarizeVecRes_UNDEF(N); break;
+  case ISD::VECTOR_SHUFFLE: R = ScalarizeVecRes_VECTOR_SHUFFLE(N); break;
+  case ISD::VSETCC:         R = ScalarizeVecRes_VSETCC(N); break;
 
   case ISD::ADD:
   case ISD::FADD:
@@ -141,6 +142,19 @@ SDOperand DAGTypeLegalizer::ScalarizeVecRes_VECTOR_SHUFFLE(SDNode *N) {
   return GetScalarizedVector(N->getOperand(Op));
 }
 
+SDOperand DAGTypeLegalizer::ScalarizeVecRes_VSETCC(SDNode *N) {
+  MVT NewVT = N->getValueType(0).getVectorElementType();
+  SDOperand LHS = GetScalarizedVector(N->getOperand(0));
+  SDOperand RHS = GetScalarizedVector(N->getOperand(1));
+  LHS = DAG.getNode(ISD::SETCC, TLI.getSetCCResultType(LHS), LHS, RHS,
+                    N->getOperand(2));
+  return
+    DAG.getNode(ISD::SELECT, NewVT, LHS,
+                DAG.getConstant(APInt::getAllOnesValue(NewVT.getSizeInBits()),
+                                NewVT),
+                DAG.getConstant(0ULL, NewVT));
+}
+
 
 //===----------------------------------------------------------------------===//
 //  Operand Vector Scalarization <1 x ty> -> ty.
@@ -251,6 +265,7 @@ void DAGTypeLegalizer::SplitVectorResult(SDNode *N, unsigned ResNo) {
   case ISD::INSERT_VECTOR_ELT: SplitVecRes_INSERT_VECTOR_ELT(N, Lo, Hi); break;
   case ISD::LOAD:           SplitVecRes_LOAD(cast<LoadSDNode>(N), Lo, Hi);break;
   case ISD::VECTOR_SHUFFLE: SplitVecRes_VECTOR_SHUFFLE(N, Lo, Hi); break;
+  case ISD::VSETCC:         SplitVecRes_VSETCC(N, Lo, Hi); break;
 
   case ISD::CTTZ:
   case ISD::CTLZ:
@@ -520,6 +535,19 @@ void DAGTypeLegalizer::SplitVecRes_VECTOR_SHUFFLE(SDNode *N, SDOperand &Lo,
   Hi = DAG.getNode(ISD::BUILD_VECTOR, HiVT, &Ops[0], Ops.size());
 }
 
+void DAGTypeLegalizer::SplitVecRes_VSETCC(SDNode *N, SDOperand &Lo,
+                                          SDOperand &Hi) {
+  MVT LoVT, HiVT;
+  GetSplitDestVTs(N->getValueType(0), LoVT, HiVT);
+
+  SDOperand LL, LH, RL, RH;
+  GetSplitVector(N->getOperand(0), LL, LH);
+  GetSplitVector(N->getOperand(1), RL, RH);
+
+  Lo = DAG.getNode(ISD::VSETCC, LoVT, LL, RL, N->getOperand(2));
+  Hi = DAG.getNode(ISD::VSETCC, HiVT, LH, RH, N->getOperand(2));
+}
+
 
 //===----------------------------------------------------------------------===//
 //  Operand Vector Splitting

test/CodeGen/X86/2008-07-23-VSetCC.ll

@@ -0,0 +1,30 @@
+; RUN: llvm-as < %s | llc -march=x86 -mcpu=pentium
+; PR2575
+
+define void @entry(i32 %m_task_id, i32 %start_x, i32 %end_x) nounwind  {
+	br i1 false, label %bb.nph, label %._crit_edge
+
+bb.nph:		; preds = %bb.nph, %0
+	vicmp sgt <4 x i32> zeroinitializer, < i32 -128, i32 -128, i32 -128, i32 -128 >		; <<4 x i32>>:1 [#uses=1]
+	extractelement <4 x i32> %1, i32 3		; <i32>:2 [#uses=1]
+	lshr i32 %2, 31		; <i32>:3 [#uses=1]
+	trunc i32 %3 to i1		; <i1>:4 [#uses=1]
+	select i1 %4, i32 -1, i32 0		; <i32>:5 [#uses=1]
+	insertelement <4 x i32> zeroinitializer, i32 %5, i32 3		; <<4 x i32>>:6 [#uses=1]
+	and <4 x i32> zeroinitializer, %6		; <<4 x i32>>:7 [#uses=1]
+	bitcast <4 x i32> %7 to <4 x float>		; <<4 x float>>:8 [#uses=1]
+	mul <4 x float> zeroinitializer, %8		; <<4 x float>>:9 [#uses=1]
+	bitcast <4 x float> %9 to <4 x i32>		; <<4 x i32>>:10 [#uses=1]
+	or <4 x i32> %10, zeroinitializer		; <<4 x i32>>:11 [#uses=1]
+	bitcast <4 x i32> %11 to <4 x float>		; <<4 x float>>:12 [#uses=1]
+	mul <4 x float> %12, < float 1.000000e+02, float 1.000000e+02, float 1.000000e+02, float 1.000000e+02 >		; <<4 x float>>:13 [#uses=1]
+	sub <4 x float> %13, < float 1.000000e+02, float 1.000000e+02, float 1.000000e+02, float 1.000000e+02 >		; <<4 x float>>:14 [#uses=1]
+	extractelement <4 x float> %14, i32 3		; <float>:15 [#uses=1]
+	call float @fmaxf( float 0.000000e+00, float %15 )		; <float>:16 [#uses=0]
+	br label %bb.nph
+
+._crit_edge:		; preds = %0
+	ret void
+}
+
+declare float @fmaxf(float, float)