Implement inalloca codegen for x86 with the new inalloca design

Calls with inalloca are lowered by skipping all stores for arguments
passed in memory and the initial stack adjustment to allocate argument
memory.

Now the frontend is responsible for the memory layout, and the backend
doesn't have to do any work.  As a result these changes are pretty
minimal.

Reviewers: echristo

Differential Revision: http://llvm-reviews.chandlerc.com/D2637

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@200596 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 54 insertions, 10 deletions

diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index c177038102..55e96a4dea 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -5434,6 +5434,8 @@ void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee,
   int DemoteStackIdx = -100;
 
   if (!CanLowerReturn) {
+    assert(!CS.hasInAllocaArgument() &&
+           "sret demotion is incompatible with inalloca");
     uint64_t TySize = TLI->getDataLayout()->getTypeAllocSize(
                       FTy->getReturnType());
     unsigned Align  = TLI->getDataLayout()->getPrefTypeAlignment(
@@ -7142,8 +7144,18 @@ TargetLowering::LowerCallTo(TargetLowering::CallLoweringInfo &CLI) const {
         Flags.setInReg();
       if (Args[i].isSRet)
         Flags.setSRet();
-      if (Args[i].isByVal) {
+      if (Args[i].isByVal)
         Flags.setByVal();
+      if (Args[i].isInAlloca) {
+        Flags.setInAlloca();
+        // Set the byval flag for CCAssignFn callbacks that don't know about
+        // inalloca.  This way we can know how many bytes we should've allocated
+        // and how many bytes a callee cleanup function will pop.  If we port
+        // inalloca to more targets, we'll have to add custom inalloca handling
+        // in the various CC lowering callbacks.
+        Flags.setByVal();
+      }
+      if (Args[i].isByVal || Args[i].isInAlloca) {
         PointerType *Ty = cast<PointerType>(Args[i].Ty);
         Type *ElementTy = Ty->getElementType();
         Flags.setByValSize(getDataLayout()->getTypeAllocSize(ElementTy));
@@ -7362,8 +7374,18 @@ void SelectionDAGISel::LowerArguments(const Function &F) {
         Flags.setInReg();
       if (F.getAttributes().hasAttribute(Idx, Attribute::StructRet))
         Flags.setSRet();
-      if (F.getAttributes().hasAttribute(Idx, Attribute::ByVal)) {
+      if (F.getAttributes().hasAttribute(Idx, Attribute::ByVal))
         Flags.setByVal();
+      if (F.getAttributes().hasAttribute(Idx, Attribute::InAlloca)) {
+        Flags.setInAlloca();
+        // Set the byval flag for CCAssignFn callbacks that don't know about
+        // inalloca.  This way we can know how many bytes we should've allocated
+        // and how many bytes a callee cleanup function will pop.  If we port
+        // inalloca to more targets, we'll have to add custom inalloca handling
+        // in the various CC lowering callbacks.
+        Flags.setByVal();
+      }
+      if (Flags.isByVal() || Flags.isInAlloca()) {
         PointerType *Ty = cast<PointerType>(I->getType());
         Type *ElementTy = Ty->getElementType();
         Flags.setByValSize(TD->getTypeAllocSize(ElementTy));
diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 6052a48640..82e5ae8534 100644
--- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -75,6 +75,7 @@ void TargetLowering::ArgListEntry::setAttributes(ImmutableCallSite *CS,
   isSRet     = CS->paramHasAttr(AttrIdx, Attribute::StructRet);
   isNest     = CS->paramHasAttr(AttrIdx, Attribute::Nest);
   isByVal    = CS->paramHasAttr(AttrIdx, Attribute::ByVal);
+  isInAlloca = CS->paramHasAttr(AttrIdx, Attribute::InAlloca);
   isReturned = CS->paramHasAttr(AttrIdx, Attribute::Returned);
   Alignment  = CS->getParamAlignment(AttrIdx);
 }
diff --git a/lib/IR/Mangler.cpp b/lib/IR/Mangler.cpp
index 5a099738d5..bc1362dc80 100644
--- a/lib/IR/Mangler.cpp
+++ b/lib/IR/Mangler.cpp
@@ -65,8 +65,8 @@ static void AddFastCallStdCallSuffix(raw_ostream &OS, const Function *F,
   for (Function::const_arg_iterator AI = F->arg_begin(), AE = F->arg_end();
        AI != AE; ++AI) {
     Type *Ty = AI->getType();
-    // 'Dereference' type in case of byval parameter attribute
-    if (AI->hasByValAttr())
+    // 'Dereference' type in case of byval or inalloca parameter attribute.
+    if (AI->hasByValOrInAllocaAttr())
       Ty = cast<PointerType>(Ty)->getElementType();
     // Size should be aligned to DWORD boundary
     ArgWords += ((TD.getTypeAllocSize(Ty) + 3)/4)*4;
diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp
index be6f13899c..9b842b74e4 100644
--- a/lib/Target/X86/X86FastISel.cpp
+++ b/lib/Target/X86/X86FastISel.cpp
@@ -1910,6 +1910,10 @@ bool X86FastISel::DoSelectCall(const Instruction *I, const char *MemIntName) {
   if (isVarArg && isWin64)
     return false;
 
+  // Don't know about inalloca yet.
+  if (CS.hasInAllocaArgument())
+    return false;
+
   // Fast-isel doesn't know about callee-pop yet.
   if (X86::isCalleePop(CC, Subtarget->is64Bit(), isVarArg,
                        TM.Options.GuaranteedTailCallOpt))
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 1b04cf46fc..bfe93877e4 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -2584,9 +2584,21 @@ X86TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       X86Info->setTCReturnAddrDelta(FPDiff);
   }
 
+  unsigned NumBytesToPush = NumBytes;
+  unsigned NumBytesToPop = NumBytes;
+
+  // If we have an inalloca argument, all stack space has already been allocated
+  // for us and be right at the top of the stack.  We don't support multiple
+  // arguments passed in memory when using inalloca.
+  if (!Outs.empty() && Outs.back().Flags.isInAlloca()) {
+    NumBytesToPush = 0;
+    assert(ArgLocs.back().getLocMemOffset() == 0 &&
+           "an inalloca argument must be the only memory argument");
+  }
+
   if (!IsSibcall)
-    Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true),
-                                 dl);
+    Chain = DAG.getCALLSEQ_START(
+        Chain, DAG.getIntPtrConstant(NumBytesToPush, true), dl);
 
   SDValue RetAddrFrIdx;
   // Load return address for tail calls.
@@ -2603,10 +2615,14 @@ X86TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   const X86RegisterInfo *RegInfo =
     static_cast<const X86RegisterInfo*>(getTargetMachine().getRegisterInfo());
   for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
+    // Skip inalloca arguments, they have already been written.
+    ISD::ArgFlagsTy Flags = Outs[i].Flags;
+    if (Flags.isInAlloca())
+      continue;
+
     CCValAssign &VA = ArgLocs[i];
     EVT RegVT = VA.getLocVT();
     SDValue Arg = OutVals[i];
-    ISD::ArgFlagsTy Flags = Outs[i].Flags;
     bool isByVal = Flags.isByVal();
 
     // Promote the value if needed.
@@ -2873,8 +2889,9 @@ X86TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   SmallVector<SDValue, 8> Ops;
 
   if (!IsSibcall && isTailCall) {
-    Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
-                           DAG.getIntPtrConstant(0, true), InFlag, dl);
+    Chain = DAG.getCALLSEQ_END(Chain,
+                               DAG.getIntPtrConstant(NumBytesToPop, true),
+                               DAG.getIntPtrConstant(0, true), InFlag, dl);
     InFlag = Chain.getValue(1);
   }
 
@@ -2931,7 +2948,7 @@ X86TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   // Returns a flag for retval copy to use.
   if (!IsSibcall) {
     Chain = DAG.getCALLSEQ_END(Chain,
-                               DAG.getIntPtrConstant(NumBytes, true),
+                               DAG.getIntPtrConstant(NumBytesToPop, true),
                                DAG.getIntPtrConstant(NumBytesForCalleeToPop,
                                                      true),
                                InFlag, dl);