Simplify code for calling a function where CanLowerReturn fails, fixing a small bug...

llvm-svn: 157446

diff --git a/llvm/include/llvm/Target/TargetLowering.h b/llvm/include/llvm/Target/TargetLowering.h
index 29abc19..c6056eb 100644 (file)
--- a/llvm/include/llvm/Target/TargetLowering.h
+++ b/llvm/include/llvm/Target/TargetLowering.h
@@ -2049,8 +2049,7 @@ private:
 /// the offsets, if the return value is being lowered to memory.
 void GetReturnInfo(Type* ReturnType, Attributes attr,
                    SmallVectorImpl<ISD::OutputArg> &Outs,
-                   const TargetLowering &TLI,
-                   SmallVectorImpl<uint64_t> *Offsets = 0);
+                   const TargetLowering &TLI);
 
 } // end llvm namespace
 

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index f1b4d80..b5fec7b 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -5149,9 +5149,8 @@ void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee,
 
   // Check whether the function can return without sret-demotion.
   SmallVector<ISD::OutputArg, 4> Outs;
-  SmallVector<uint64_t, 4> Offsets;
   GetReturnInfo(RetTy, CS.getAttributes().getRetAttributes(),
-                Outs, TLI, &Offsets);
+                Outs, TLI);
 
   bool CanLowerReturn = TLI.CanLowerReturn(CS.getCallingConv(),
                                           DAG.getMachineFunction(),
@@ -5264,7 +5263,13 @@ void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee,
     ComputeValueVTs(TLI, PtrRetTy, PVTs);
     assert(PVTs.size() == 1 && "Pointers should fit in one register");
     EVT PtrVT = PVTs[0];
-    unsigned NumValues = Outs.size();
+
+    SmallVector<EVT, 4> RetTys;
+    SmallVector<uint64_t, 4> Offsets;
+    RetTy = FTy->getReturnType();
+    ComputeValueVTs(TLI, RetTy, RetTys, &Offsets);
+
+    unsigned NumValues = RetTys.size();
     SmallVector<SDValue, 4> Values(NumValues);
     SmallVector<SDValue, 4> Chains(NumValues);
 
@@ -5272,8 +5277,7 @@ void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee,
       SDValue Add = DAG.getNode(ISD::ADD, getCurDebugLoc(), PtrVT,
                                 DemoteStackSlot,
                                 DAG.getConstant(Offsets[i], PtrVT));
-      SDValue L = DAG.getLoad(Outs[i].VT, getCurDebugLoc(), Result.second,
-                              Add,
+      SDValue L = DAG.getLoad(RetTys[i], getCurDebugLoc(), Result.second, Add,
                   MachinePointerInfo::getFixedStack(DemoteStackIdx, Offsets[i]),
                               false, false, false, 1);
       Values[i] = L;
@@ -5284,30 +5288,10 @@ void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee,
                                 MVT::Other, &Chains[0], NumValues);
     PendingLoads.push_back(Chain);
 
-    // Collect the legal value parts into potentially illegal values
-    // that correspond to the original function's return values.
-    SmallVector<EVT, 4> RetTys;
-    RetTy = FTy->getReturnType();
-    ComputeValueVTs(TLI, RetTy, RetTys);
-    ISD::NodeType AssertOp = ISD::DELETED_NODE;
-    SmallVector<SDValue, 4> ReturnValues;
-    unsigned CurReg = 0;
-    for (unsigned I = 0, E = RetTys.size(); I != E; ++I) {
-      EVT VT = RetTys[I];
-      EVT RegisterVT = TLI.getRegisterType(RetTy->getContext(), VT);
-      unsigned NumRegs = TLI.getNumRegisters(RetTy->getContext(), VT);
-
-      SDValue ReturnValue =
-        getCopyFromParts(DAG, getCurDebugLoc(), &Values[CurReg], NumRegs,
-                         RegisterVT, VT, AssertOp);
-      ReturnValues.push_back(ReturnValue);
-      CurReg += NumRegs;
-    }
-
     setValue(CS.getInstruction(),
              DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
                          DAG.getVTList(&RetTys[0], RetTys.size()),
-                         &ReturnValues[0], ReturnValues.size()));
+                         &Values[0], Values.size()));
   }
 
   // Assign order to nodes here. If the call does not produce a result, it won't

diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 5c9dc47..56b1719 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -997,13 +997,11 @@ unsigned TargetLowering::getVectorTypeBreakdown(LLVMContext &Context, EVT VT,
 /// TODO: Move this out of TargetLowering.cpp.
 void llvm::GetReturnInfo(Type* ReturnType, Attributes attr,
                          SmallVectorImpl<ISD::OutputArg> &Outs,
-                         const TargetLowering &TLI,
-                         SmallVectorImpl<uint64_t> *Offsets) {
+                         const TargetLowering &TLI) {
   SmallVector<EVT, 4> ValueVTs;
   ComputeValueVTs(TLI, ReturnType, ValueVTs);
   unsigned NumValues = ValueVTs.size();
   if (NumValues == 0) return;
-  unsigned Offset = 0;
 
   for (unsigned j = 0, f = NumValues; j != f; ++j) {
     EVT VT = ValueVTs[j];
@@ -1026,8 +1024,6 @@ void llvm::GetReturnInfo(Type* ReturnType, Attributes attr,
 
     unsigned NumParts = TLI.getNumRegisters(ReturnType->getContext(), VT);
     EVT PartVT = TLI.getRegisterType(ReturnType->getContext(), VT);
-    unsigned PartSize = TLI.getTargetData()->getTypeAllocSize(
-                        PartVT.getTypeForEVT(ReturnType->getContext()));
 
     // 'inreg' on function refers to return value
     ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
@@ -1042,10 +1038,6 @@ void llvm::GetReturnInfo(Type* ReturnType, Attributes attr,
 
     for (unsigned i = 0; i < NumParts; ++i) {
       Outs.push_back(ISD::OutputArg(Flags, PartVT, /*isFixed=*/true));
-      if (Offsets) {
-        Offsets->push_back(Offset);
-        Offset += PartSize;
-      }
     }
   }
 }

diff --git a/llvm/lib/Target/X86/X86FastISel.cpp b/llvm/lib/Target/X86/X86FastISel.cpp
index 1b62357..07d0e76 100644 (file)
--- a/llvm/lib/Target/X86/X86FastISel.cpp
+++ b/llvm/lib/Target/X86/X86FastISel.cpp
@@ -1549,9 +1549,8 @@ bool X86FastISel::DoSelectCall(const Instruction *I, const char *MemIntName) {
 
   // Check whether the function can return without sret-demotion.
   SmallVector<ISD::OutputArg, 4> Outs;
-  SmallVector<uint64_t, 4> Offsets;
   GetReturnInfo(I->getType(), CS.getAttributes().getRetAttributes(),
-                Outs, TLI, &Offsets);
+                Outs, TLI);
   bool CanLowerReturn = TLI.CanLowerReturn(CS.getCallingConv(),
                                           *FuncInfo.MF, FTy->isVarArg(),
                                           Outs, FTy->getContext());

diff --git a/llvm/test/CodeGen/X86/bigstructret.ll b/llvm/test/CodeGen/X86/bigstructret.ll
index 633995d..eacbf32 100644 (file)
--- a/llvm/test/CodeGen/X86/bigstructret.ll
+++ b/llvm/test/CodeGen/X86/bigstructret.ll
@@ -1,12 +1,15 @@
-; RUN: llc < %s -march=x86 -o %t
-; RUN: grep "movl      .24601, 12(%ecx)" %t
-; RUN: grep "movl      .48, 8(%ecx)" %t
-; RUN: grep "movl      .24, 4(%ecx)" %t
-; RUN: grep "movl      .12, (%ecx)" %t
+; RUN: llc < %s -march=x86 | FileCheck %s
 
 %0 = type { i32, i32, i32, i32 }
+%1 = type { i1, i1, i1, i32 }
 
-define internal fastcc %0 @ReturnBigStruct() nounwind readnone {
+; CHECK: ReturnBigStruct
+; CHECK: movl $24601, 12(%ecx)
+; CHECK: movl  $48, 8(%ecx)
+; CHECK: movl  $24, 4(%ecx)
+; CHECK: movl  $12, (%ecx)
+
+define fastcc %0 @ReturnBigStruct() nounwind readnone {
 entry:
   %0 = insertvalue %0 zeroinitializer, i32 12, 0
   %1 = insertvalue %0 %0, i32 24, 1
@@ -15,3 +18,29 @@ entry:
   ret %0 %3
 }
 
+; CHECK: ReturnBigStruct2
+; CHECK: movl  $48, 4(%ecx)
+; CHECK: movb  $1, 2(%ecx)
+; CHECK: movb  $1, 1(%ecx)
+; CHECK: movb  $0, (%ecx)
+
+define fastcc %1 @ReturnBigStruct2() nounwind readnone {
+entry:
+  %0 = insertvalue %1 zeroinitializer, i1 false, 0
+  %1 = insertvalue %1 %0, i1 true, 1
+  %2 = insertvalue %1 %1, i1 true, 2
+  %3 = insertvalue %1 %2, i32 48, 3
+  ret %1 %3
+}
+
+; CHECK: CallBigStruct2
+; CHECK: leal  16(%esp), {{.*}}
+; CHECK: call{{.*}}ReturnBigStruct2
+; CHECK: subl  $4, %esp
+; CHECK: movl  20(%esp), %eax
+define fastcc i32 @CallBigStruct2() nounwind readnone {
+entry:
+  %0 = call %1 @ReturnBigStruct2()
+  %1 = extractvalue %1 %0, 3
+  ret i32 %1
+}