Issue description:

SchedulerDAGInstrs::buildSchedGraph ignores dependencies between FixedStack
objects and byval parameters. So loading byval parameters from stack may be
inserted *before* it will be stored, since these operations are treated as
independent.

Fix:
Currently ARMTargetLowering::LowerFormalArguments saves byval registers with
FixedStack MachinePointerInfo. To fix the problem we need to store byval
registers with MachinePointerInfo referenced to first the "byval" parameter.

Also commit adds two new fields to the InputArg structure: Function's argument
index and InputArg's part offset in bytes relative to the start position of
Function's argument. E.g.: If function's argument is 128 bit width and it was
splitted onto 32 bit regs, then we got 4 InputArg structs with same arg index,
but different offset values. 



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@165616 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 50 insertions, 12 deletions

diff --git a/include/llvm/Target/TargetCallingConv.h b/include/llvm/Target/TargetCallingConv.h
index f8cebefb0e..8030d38d73 100644
--- a/include/llvm/Target/TargetCallingConv.h
+++ b/include/llvm/Target/TargetCallingConv.h
@@ -113,9 +113,18 @@ namespace ISD {
     MVT VT;
     bool Used;
 
+    /// Index original Function's argument.
+    unsigned OrigArgIndex;
+
+    /// Offset in bytes of current input value relative to the beginning of
+    /// original argument. E.g. if argument was splitted into four 32 bit
+    /// registers, we got 4 InputArgs with PartOffsets 0, 4, 8 and 12.
+    unsigned PartOffset;
+
     InputArg() : VT(MVT::Other), Used(false) {}
-    InputArg(ArgFlagsTy flags, EVT vt, bool used)
-      : Flags(flags), Used(used) {
+    InputArg(ArgFlagsTy flags, EVT vt, bool used,
+             unsigned origIdx, unsigned partOffs)
+      : Flags(flags), Used(used), OrigArgIndex(origIdx), PartOffset(partOffs) {
       VT = vt.getSimpleVT();
     }
   };
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index dccc2fcf25..cdebff94e5 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -6681,7 +6681,7 @@ void SelectionDAGISel::LowerArguments(const BasicBlock *LLVMBB) {
     ISD::ArgFlagsTy Flags;
     Flags.setSRet();
     EVT RegisterVT = TLI.getRegisterType(*DAG.getContext(), ValueVTs[0]);
-    ISD::InputArg RetArg(Flags, RegisterVT, true);
+    ISD::InputArg RetArg(Flags, RegisterVT, true, 0, 0);
     Ins.push_back(RetArg);
   }
 
@@ -6729,7 +6729,8 @@ void SelectionDAGISel::LowerArguments(const BasicBlock *LLVMBB) {
       EVT RegisterVT = TLI.getRegisterType(*CurDAG->getContext(), VT);
       unsigned NumRegs = TLI.getNumRegisters(*CurDAG->getContext(), VT);
       for (unsigned i = 0; i != NumRegs; ++i) {
-        ISD::InputArg MyFlags(Flags, RegisterVT, isArgValueUsed);
+        ISD::InputArg MyFlags(Flags, RegisterVT, isArgValueUsed,
+                              Idx-1, i*RegisterVT.getStoreSize());
         if (NumRegs > 1 && i == 0)
           MyFlags.Flags.setSplit();
         // if it isn't first piece, alignment must be 1
diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index d4d29f57d2..e6f0ef25d8 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -2534,6 +2534,8 @@ ARMTargetLowering::computeRegArea(CCState &CCInfo, MachineFunction &MF,
 void
 ARMTargetLowering::VarArgStyleRegisters(CCState &CCInfo, SelectionDAG &DAG,
                                         DebugLoc dl, SDValue &Chain,
+                                        const Value *OrigArg,
+                                        unsigned OffsetFromOrigArg,
                                         unsigned ArgOffset) const {
   MachineFunction &MF = DAG.getMachineFunction();
   MachineFrameInfo *MFI = MF.getFrameInfo();
@@ -2561,7 +2563,7 @@ ARMTargetLowering::VarArgStyleRegisters(CCState &CCInfo, SelectionDAG &DAG,
                                     getPointerTy());
 
     SmallVector<SDValue, 4> MemOps;
-    for (; firstRegToSaveIndex < 4; ++firstRegToSaveIndex) {
+    for (unsigned i = 0; firstRegToSaveIndex < 4; ++firstRegToSaveIndex, ++i) {
       const TargetRegisterClass *RC;
       if (AFI->isThumb1OnlyFunction())
         RC = &ARM::tGPRRegClass;
@@ -2572,7 +2574,7 @@ ARMTargetLowering::VarArgStyleRegisters(CCState &CCInfo, SelectionDAG &DAG,
       SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i32);
       SDValue Store =
         DAG.getStore(Val.getValue(1), dl, Val, FIN,
-                 MachinePointerInfo::getFixedStack(AFI->getVarArgsFrameIndex()),
+                     MachinePointerInfo(OrigArg, OffsetFromOrigArg + 4*i),
                      false, false, 0);
       MemOps.push_back(Store);
       FIN = DAG.getNode(ISD::ADD, dl, getPointerTy(), FIN,
@@ -2606,14 +2608,16 @@ ARMTargetLowering::LowerFormalArguments(SDValue Chain,
   CCInfo.AnalyzeFormalArguments(Ins,
                                 CCAssignFnForNode(CallConv, /* Return*/ false,
                                                   isVarArg));
-
+  
   SmallVector<SDValue, 16> ArgValues;
   int lastInsIndex = -1;
-
   SDValue ArgValue;
+  Function::const_arg_iterator CurOrigArg = MF.getFunction()->arg_begin();
+  unsigned CurArgIdx = 0;
   for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
     CCValAssign &VA = ArgLocs[i];
-
+    std::advance(CurOrigArg, Ins[VA.getValNo()].OrigArgIndex - CurArgIdx);
+    CurArgIdx = Ins[VA.getValNo()].OrigArgIndex;
     // Arguments stored in registers.
     if (VA.isRegLoc()) {
       EVT RegVT = VA.getLocVT();
@@ -2709,7 +2713,8 @@ ARMTargetLowering::LowerFormalArguments(SDValue Chain,
           if (Flags.isByVal()) {
             unsigned VARegSize, VARegSaveSize;
             computeRegArea(CCInfo, MF, VARegSize, VARegSaveSize);
-            VarArgStyleRegisters(CCInfo, DAG, dl, Chain, 0);
+            VarArgStyleRegisters(CCInfo, DAG,
+                                 dl, Chain, CurOrigArg, Ins[VA.getValNo()].PartOffset, 0);
             unsigned Bytes = Flags.getByValSize() - VARegSize;
             if (Bytes == 0) Bytes = 1; // Don't create zero-sized stack objects.
             int FI = MFI->CreateFixedObject(Bytes,
@@ -2732,7 +2737,8 @@ ARMTargetLowering::LowerFormalArguments(SDValue Chain,
 
   // varargs
   if (isVarArg)
-    VarArgStyleRegisters(CCInfo, DAG, dl, Chain, CCInfo.getNextStackOffset());
+    VarArgStyleRegisters(CCInfo, DAG, dl, Chain, 0, 0,
+                         CCInfo.getNextStackOffset());
 
   return Chain;
 }
diff --git a/lib/Target/ARM/ARMISelLowering.h b/lib/Target/ARM/ARMISelLowering.h
index 2b8f382c95..b5020c9710 100644
--- a/lib/Target/ARM/ARMISelLowering.h
+++ b/lib/Target/ARM/ARMISelLowering.h
@@ -466,7 +466,10 @@ namespace llvm {
                            SmallVectorImpl<SDValue> &InVals) const;
 
     void VarArgStyleRegisters(CCState &CCInfo, SelectionDAG &DAG,
-                              DebugLoc dl, SDValue &Chain, unsigned ArgOffset)
+                              DebugLoc dl, SDValue &Chain,
+                              const Value *OrigArg,
+                              unsigned OffsetFromOrigArg,
+                              unsigned ArgOffset)
       const;
 
     void computeRegArea(CCState &CCInfo, MachineFunction &MF,
diff --git a/test/CodeGen/ARM/2012-10-04-FixedFrame-vs-byval.ll b/test/CodeGen/ARM/2012-10-04-FixedFrame-vs-byval.ll
new file mode 100644
index 0000000000..478048d096
--- /dev/null
+++ b/test/CodeGen/ARM/2012-10-04-FixedFrame-vs-byval.ll
@@ -0,0 +1,19 @@
+; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi | FileCheck %s
+
+@.str = private unnamed_addr constant [12 x i8] c"val.a = %f\0A\00"
+%struct_t = type { double, double, double }
+@static_val = constant %struct_t { double 1.0, double 2.0, double 3.0 }
+
+declare i32 @printf(i8*, ...)
+
+; CHECK:     test_byval_usage_scheduling:
+; CHECK:       str     r3, [sp, #12]
+; CHECK:       str     r2, [sp, #8]
+; CHECK:       vldr    d16, [sp, #8]
+define void @test_byval_usage_scheduling(i32 %n1, i32 %n2, %struct_t* byval %val) nounwind {
+entry:
+  %a = getelementptr inbounds %struct_t* %val, i32 0, i32 0
+  %0 = load double* %a
+  %call = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([12 x i8]* @.str, i32 0, i32 0), double %0)
+  ret void
+}