Generate correct Sparc32 ABI compliant code for functions that return a struct.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@126108 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/Sparc/DelaySlotFiller.cpp

@@ -79,6 +79,7 @@ namespace {
     MachineBasicBlock::iterator
     findDelayInstr(MachineBasicBlock &MBB, MachineBasicBlock::iterator slot);
 
+    bool needsUnimp(MachineBasicBlock::iterator I, unsigned &StructSize);
 
   };
   char Filler::ID = 0;
@@ -91,6 +92,7 @@ FunctionPass *llvm::createSparcDelaySlotFillerPass(TargetMachine &tm) {
   return new Filler(tm);
 }
 
+
 /// runOnMachineBasicBlock - Fill in delay slots for the given basic block.
 /// We assume there is only one delay slot per delayed instruction.
 ///
@@ -112,6 +114,13 @@ bool Filler::runOnMachineBasicBlock(MachineBasicBlock &MBB) {
         BuildMI(MBB, ++J, I->getDebugLoc(), TII->get(SP::NOP));
       else
         MBB.splice(++J, &MBB, D);
+      unsigned structSize = 0;
+      if (needsUnimp(I, structSize)) {
+        MachineBasicBlock::iterator J = I;
+        ++J; //skip the delay filler.
+        BuildMI(MBB, ++J, I->getDebugLoc(),
+                TII->get(SP::UNIMP)).addImm(structSize);
+      }
     }
   return Changed;
 }
@@ -287,6 +296,28 @@ bool Filler::isDelayFiller(MachineBasicBlock &MBB,
 {
   if (candidate == MBB.begin())
     return false;
+  if (candidate->getOpcode() == SP::UNIMP)
+    return true;
   const TargetInstrDesc &prevdesc = (--candidate)->getDesc();
   return prevdesc.hasDelaySlot();
 }
+
+bool Filler::needsUnimp(MachineBasicBlock::iterator I, unsigned &StructSize)
+{
+  if (!I->getDesc().isCall())
+    return false;
+
+  unsigned structSizeOpNum = 0;
+  switch (I->getOpcode()) {
+  default: llvm_unreachable("Unknown call opcode.");
+  case SP::CALL: structSizeOpNum = 1; break;
+  case SP::JMPLrr:
+  case SP::JMPLri: structSizeOpNum = 2; break;
+  }
+
+  const MachineOperand &MO = I->getOperand(structSizeOpNum);
+  if (!MO.isImm())
+    return false;
+  StructSize = MO.getImm();
+  return true;
+}

lib/Target/Sparc/SparcISelLowering.cpp

@@ -16,7 +16,9 @@
 #include "SparcISelLowering.h"
 #include "SparcTargetMachine.h"
 #include "SparcMachineFunctionInfo.h"
+#include "llvm/DerivedTypes.h"
 #include "llvm/Function.h"
+#include "llvm/Module.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
@@ -116,6 +118,8 @@ SparcTargetLowering::LowerReturn(SDValue Chain,
     // Guarantee that all emitted copies are stuck together with flags.
     Flag = Chain.getValue(1);
   }
+
+  unsigned RetAddrOffset = 8; //Call Inst + Delay Slot
   // If the function returns a struct, copy the SRetReturnReg to I0
   if (MF.getFunction()->hasStructRetAttr()) {
     SparcMachineFunctionInfo *SFI = MF.getInfo<SparcMachineFunctionInfo>();
@@ -127,11 +131,16 @@ SparcTargetLowering::LowerReturn(SDValue Chain,
     Flag = Chain.getValue(1);
     if (MF.getRegInfo().liveout_empty())
       MF.getRegInfo().addLiveOut(SP::I0);
+    RetAddrOffset = 12; // CallInst + Delay Slot + Unimp
   }
 
+  SDValue RetAddrOffsetNode = DAG.getConstant(RetAddrOffset, MVT::i32);
+
   if (Flag.getNode())
-    return DAG.getNode(SPISD::RET_FLAG, dl, MVT::Other, Chain, Flag);
+    return DAG.getNode(SPISD::RET_FLAG, dl, MVT::Other, Chain,
-  return DAG.getNode(SPISD::RET_FLAG, dl, MVT::Other, Chain);
+                       RetAddrOffsetNode, Flag);
+  return DAG.getNode(SPISD::RET_FLAG, dl, MVT::Other, Chain, 
+                     RetAddrOffsetNode);
 }
 
 /// LowerFormalArguments - V8 uses a very simple ABI, where all values are
@@ -393,6 +402,7 @@ SparcTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
   SmallVector<SDValue, 8> MemOpChains;
 
   const unsigned StackOffset = 92;
+  bool hasStructRetAttr = false;
   // Walk the register/memloc assignments, inserting copies/loads.
   for (unsigned i = 0, realArgIdx = 0, byvalArgIdx = 0, e = ArgLocs.size();
        i != e;
@@ -433,6 +443,7 @@ SparcTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
       MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
                                          MachinePointerInfo(),
                                          false, false, 0));
+      hasStructRetAttr = true;
       continue;
     }
 
@@ -546,6 +557,8 @@ SparcTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
     InFlag = Chain.getValue(1);
   }
 
+  unsigned SRetArgSize = (hasStructRetAttr)? getSRetArgSize(DAG, Callee):0;
+
   // If the callee is a GlobalAddress node (quite common, every direct call is)
   // turn it into a TargetGlobalAddress node so that legalize doesn't hack it.
   // Likewise ExternalSymbol -> TargetExternalSymbol.
@@ -559,6 +572,8 @@ SparcTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
   SmallVector<SDValue, 8> Ops;
   Ops.push_back(Chain);
   Ops.push_back(Callee);
+  if (hasStructRetAttr)
+    Ops.push_back(DAG.getTargetConstant(SRetArgSize, MVT::i32));
   for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
     unsigned Reg = RegsToPass[i].first;
     if (Reg >= SP::I0 && Reg <= SP::I7)
@@ -600,7 +615,29 @@ SparcTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
   return Chain;
 }
 
+unsigned
+SparcTargetLowering::getSRetArgSize(SelectionDAG &DAG, SDValue Callee) const
+{
+  const Function *CalleeFn = 0;
+  if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
+    CalleeFn = dyn_cast<Function>(G->getGlobal());
+  } else if (ExternalSymbolSDNode *E =
+             dyn_cast<ExternalSymbolSDNode>(Callee)) {
+    const Function *Fn = DAG.getMachineFunction().getFunction();
+    const Module *M = Fn->getParent();
+    CalleeFn = M->getFunction(E->getSymbol());
+  }
 
+  if (!CalleeFn)
+    return 0;
+
+  assert(CalleeFn->hasStructRetAttr() &&
+         "Callee does not have the StructRet attribute.");
+
+  const PointerType *Ty = cast<PointerType>(CalleeFn->arg_begin()->getType());
+  const Type *ElementTy = Ty->getElementType();
+  return getTargetData()->getTypeAllocSize(ElementTy);
+}
 
 //===----------------------------------------------------------------------===//
 // TargetLowering Implementation

lib/Target/Sparc/SparcISelLowering.h

@@ -101,6 +101,8 @@ namespace llvm {
 
     SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) const;
+
+    unsigned getSRetArgSize(SelectionDAG &DAG, SDValue Callee) const;
   };
 } // end namespace llvm
 

lib/Target/Sparc/SparcInstrInfo.td

@@ -124,7 +124,8 @@ def call          : SDNode<"SPISD::CALL", SDT_SPCall,
                            [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
                             SDNPVariadic]>;
 
-def retflag       : SDNode<"SPISD::RET_FLAG", SDTNone,
+def SDT_SPRet     : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
+def retflag       : SDNode<"SPISD::RET_FLAG", SDT_SPRet,
                            [SDNPHasChain, SDNPOptInGlue]>;
 
 def flushw        : SDNode<"SPISD::FLUSHW", SDTNone,
@@ -232,6 +233,9 @@ let hasSideEffects = 1, mayStore = 1 in {
                    [(flushw)]>;
 }
 
+def UNIMP : F2_1<0b000, (outs), (ins i32imm:$val),
+                "unimp $val", []>;
+
 // FpMOVD/FpNEGD/FpABSD - These are lowered to single-precision ops by the 
 // fpmover pass.
 let Predicates = [HasNoV9] in {  // Only emit these in V8 mode.
@@ -292,11 +296,13 @@ let usesCustomInserter = 1, Uses = [FCC] in {
 // Section A.3 - Synthetic Instructions, p. 85
 // special cases of JMPL:
 let isReturn = 1, isTerminator = 1, hasDelaySlot = 1, isBarrier = 1 in {
-  let rd = O7.Num, rs1 = G0.Num, simm13 = 8 in
+  let rd = O7.Num, rs1 = G0.Num in
-    def RETL: F3_2<2, 0b111000, (outs), (ins), "retl", [(retflag)]>;
+    def RETL: F3_2<2, 0b111000, (outs), (ins i32imm:$val),
+                   "jmp %o7+$val", [(retflag simm13:$val)]>;
 
-  let rd = I7.Num, rs1 = G0.Num, simm13 = 8 in
+  let rd = I7.Num, rs1 = G0.Num in
-    def RET: F3_2<2, 0b111000, (outs), (ins), "ret", []>;
+    def RET: F3_2<2, 0b111000, (outs), (ins i32imm:$val),
+                  "jmp %i7+$val", []>;
 }
 
 // Section B.1 - Load Integer Instructions, p. 90

test/CodeGen/SPARC/2011-01-19-DelaySlot.ll

@@ -7,7 +7,7 @@ entry:
 ; CHECK: test
 ; CHECK: call bar
 ; CHECK-NOT: nop
-; CHECK: ret
+; CHECK: jmp
 ; CHECK-NEXT: restore
   %0 = tail call i32 @bar(i32 %a) nounwind
   ret i32 %0
@@ -18,7 +18,7 @@ entry:
 ; CHECK:      test_jmpl
 ; CHECK:      call
 ; CHECK-NOT:  nop
-; CHECK:      ret
+; CHECK:      jmp
 ; CHECK-NEXT: restore
   %0 = tail call i32 %f(i32 %a, i32 %b) nounwind
   ret i32 %0
@@ -47,7 +47,7 @@ bb:                                               ; preds = %entry, %bb
 
 bb5:                                              ; preds = %bb, %entry
   %a_addr.1.lcssa = phi i32 [ %a, %entry ], [ %a_addr.0, %bb ]
-;CHECK:      ret
+;CHECK:      jmp
 ;CHECK-NEXT: restore
   ret i32 %a_addr.1.lcssa
 }

test/CodeGen/SPARC/2011-01-22-SRet.ll

@@ -7,7 +7,7 @@ entry:
 ;CHECK: make_foo
 ;CHECK: ld [%fp+64], {{.+}}
 ;CHECK: or {{.+}}, {{.+}}, %i0
-;CHECK: ret
+;CHECK: jmp %i7+12
   %0 = getelementptr inbounds %struct.foo_t* %agg.result, i32 0, i32 0
   store i32 %a, i32* %0, align 4
   %1 = getelementptr inbounds %struct.foo_t* %agg.result, i32 0, i32 1
@@ -22,6 +22,7 @@ entry:
 ;CHECK: test
 ;CHECK: st {{.+}}, [%sp+64]
 ;CHECK: make_foo
+;CHECK: unimp 12
   %f = alloca %struct.foo_t, align 8
   call void @make_foo(%struct.foo_t* noalias sret %f, i32 10, i32 20, i32 30) nounwind
   %0 = getelementptr inbounds %struct.foo_t* %f, i32 0, i32 0