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

diff --git a/docs/LangRef.rst b/docs/LangRef.rst
index 62ce112..07e94a9 100644 (file)
--- a/docs/LangRef.rst
+++ b/docs/LangRef.rst
@@ -792,7 +792,10 @@ Currently, only the following parameter attributes are defined:
     An argument allocation may be used by a call at most once because
     the call may deallocate it.  The ``inalloca`` attribute cannot be
     used in conjunction with other attributes that affect argument
-    storage, like ``inreg``, ``nest``, ``sret``, or ``byval``.
+    storage, like ``inreg``, ``nest``, ``sret``, or ``byval``.  The
+    ``inalloca`` attribute also disables LLVM's implicit lowering of
+    large aggregate return values, which means that frontend authors
+    must lower them with ``sret`` pointers.
 
     When the call site is reached, the argument allocation must have
     been the most recent stack allocation that is still live, or the

diff --git a/include/llvm/Support/CallSite.h b/include/llvm/Support/CallSite.h
index d2c47b5..f4324cb 100644 (file)
--- a/include/llvm/Support/CallSite.h
+++ b/include/llvm/Support/CallSite.h
@@ -268,9 +268,10 @@ public:
            paramHasAttr(ArgNo + 1, Attribute::InAlloca);
   }
 
-  /// @brief Determine if there are any inalloca arguments.
+  /// @brief Determine if there are is an inalloca argument.  Only the last
+  /// argument can have the inalloca attribute.
   bool hasInAllocaArgument() const {
-    return getAttributes().hasAttrSomewhere(Attribute::InAlloca);
+    return paramHasAttr(arg_size(), Attribute::InAlloca);
   }
 
   bool doesNotAccessMemory(unsigned ArgNo) const {

diff --git a/include/llvm/Target/TargetCallingConv.h b/include/llvm/Target/TargetCallingConv.h
index 9cc52a5..758a9c9 100644 (file)
--- a/include/llvm/Target/TargetCallingConv.h
+++ b/include/llvm/Target/TargetCallingConv.h
@@ -42,6 +42,8 @@ namespace ISD {
     static const uint64_t ByValAlignOffs = 7;
     static const uint64_t Split          = 1ULL<<11;
     static const uint64_t SplitOffs      = 11;
+    static const uint64_t InAlloca       = 1ULL<<12; ///< Passed with inalloca
+    static const uint64_t InAllocaOffs   = 12;
     static const uint64_t OrigAlign      = 0x1FULL<<27;
     static const uint64_t OrigAlignOffs  = 27;
     static const uint64_t ByValSize      = 0xffffffffULL<<32; ///< Struct size
@@ -68,6 +70,9 @@ namespace ISD {
     bool isByVal()     const { return Flags & ByVal; }
     void setByVal()    { Flags |= One << ByValOffs; }
 
+    bool isInAlloca()  const { return Flags & InAlloca; }
+    void setInAlloca() { Flags |= One << InAllocaOffs; }
+
     bool isNest()      const { return Flags & Nest; }
     void setNest()     { Flags |= One << NestOffs; }
 

diff --git a/include/llvm/Target/TargetLowering.h b/include/llvm/Target/TargetLowering.h
index e9d8729..b5c7e43 100644 (file)
--- a/include/llvm/Target/TargetLowering.h
+++ b/include/llvm/Target/TargetLowering.h
@@ -609,8 +609,9 @@ public:
     return getValueType(Ty, AllowUnknown).getSimpleVT();
   }
 
-  /// Return the desired alignment for ByVal aggregate function arguments in the
-  /// caller parameter area.  This is the actual alignment, not its logarithm.
+  /// Return the desired alignment for ByVal or InAlloca aggregate function
+  /// arguments in the caller parameter area.  This is the actual alignment, not
+  /// its logarithm.
   virtual unsigned getByValTypeAlignment(Type *Ty) const;
 
   /// Return the type of registers that this ValueType will eventually require.
@@ -1965,12 +1966,13 @@ public:
     bool isSRet     : 1;
     bool isNest     : 1;
     bool isByVal    : 1;
+    bool isInAlloca : 1;
     bool isReturned : 1;
     uint16_t Alignment;
 
     ArgListEntry() : isSExt(false), isZExt(false), isInReg(false),
-      isSRet(false), isNest(false), isByVal(false), isReturned(false),
-      Alignment(0) { }
+      isSRet(false), isNest(false), isByVal(false), isInAlloca(false),
+      isReturned(false), Alignment(0) { }
 
     void setAttributes(ImmutableCallSite *CS, unsigned AttrIdx);
   };

diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index c177038..55e96a4 100644 (file)
--- 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 6052a48..82e5ae8 100644 (file)
--- 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 5a09973..bc1362d 100644 (file)
--- 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 be6f138..9b842b7 100644 (file)
--- 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 1b04cf4..bfe9387 100644 (file)
--- 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);

diff --git a/test/CodeGen/X86/inalloca-ctor.ll b/test/CodeGen/X86/inalloca-ctor.ll
new file mode 100644 (file)
index 0000000..f81e967
--- /dev/null
+++ b/test/CodeGen/X86/inalloca-ctor.ll
@@ -0,0 +1,34 @@
+; RUN: llc < %s -mtriple=i686-pc-win32 | FileCheck %s
+
+%Foo = type { i32, i32 }
+
+%frame = type { %Foo, i32, %Foo }
+
+declare void @f(%frame* inalloca %a)
+
+declare void @Foo_ctor(%Foo* %this)
+
+define void @g() {
+entry:
+  %args = alloca %frame, inalloca
+  %c = getelementptr %frame* %args, i32 0, i32 2
+; CHECK: movl    $20, %eax
+; CHECK: calll   __chkstk
+; CHECK: movl %esp,
+  call void @Foo_ctor(%Foo* %c)
+; CHECK: leal 12(%{{.*}}),
+; CHECK: subl $4, %esp
+; CHECK: calll _Foo_ctor
+; CHECK: addl $4, %esp
+  %b = getelementptr %frame* %args, i32 0, i32 1
+  store i32 42, i32* %b
+; CHECK: movl $42,
+  %a = getelementptr %frame* %args, i32 0, i32 0
+  call void @Foo_ctor(%Foo* %a)
+; CHECK: subl $4, %esp
+; CHECK: calll _Foo_ctor
+; CHECK: addl $4, %esp
+  call void @f(%frame* inalloca %args)
+; CHECK: calll   _f
+  ret void
+}

diff --git a/test/CodeGen/X86/inalloca-invoke.ll b/test/CodeGen/X86/inalloca-invoke.ll
new file mode 100644 (file)
index 0000000..ac530ca
--- /dev/null
+++ b/test/CodeGen/X86/inalloca-invoke.ll
@@ -0,0 +1,54 @@
+; RUN: llc < %s -mtriple=i686-pc-win32 | FileCheck %s
+
+%Iter = type { i32, i32, i32 }
+
+%frame.reverse = type { %Iter, %Iter }
+
+declare void @llvm.stackrestore(i8*)
+declare i8* @llvm.stacksave()
+declare void @begin(%Iter* sret)
+declare void @plus(%Iter* sret, %Iter*, i32)
+declare void @reverse(%frame.reverse* inalloca align 4)
+
+define i32 @main() {
+  %temp.lvalue = alloca %Iter
+  br label %blah
+
+blah:
+  %inalloca.save = call i8* @llvm.stacksave()
+  %rev_args = alloca %frame.reverse, inalloca, align 4
+  %beg = getelementptr %frame.reverse* %rev_args, i32 0, i32 0
+  %end = getelementptr %frame.reverse* %rev_args, i32 0, i32 1
+
+; CHECK:  calll   __chkstk
+; CHECK:  movl    %[[beg:[^,]*]], %esp
+; CHECK:  leal    12(%[[beg]]), %[[end:[^ ]*]]
+
+  call void @begin(%Iter* sret %temp.lvalue)
+; CHECK:  calll _begin
+
+  invoke void @plus(%Iter* sret %end, %Iter* %temp.lvalue, i32 4)
+          to label %invoke.cont unwind label %lpad
+
+;  Uses end as sret param.
+; CHECK:  movl %[[end]], (%esp)
+; CHECK:  calll _plus
+
+invoke.cont:
+  call void @begin(%Iter* sret %beg)
+
+; CHECK:  movl %[[beg]],
+; CHECK:  calll _begin
+
+  invoke void @reverse(%frame.reverse* inalloca align 4 %rev_args)
+          to label %invoke.cont5 unwind label %lpad
+
+invoke.cont5:                                     ; preds = %invoke.cont
+  call void @llvm.stackrestore(i8* %inalloca.save)
+  ret i32 0
+
+lpad:                                             ; preds = %invoke.cont, %entry
+  %lp = landingpad { i8*, i32 } personality i8* null
+          cleanup
+  unreachable
+}

diff --git a/test/CodeGen/X86/inalloca-stdcall.ll b/test/CodeGen/X86/inalloca-stdcall.ll
new file mode 100644 (file)
index 0000000..93ac451
--- /dev/null
+++ b/test/CodeGen/X86/inalloca-stdcall.ll
@@ -0,0 +1,26 @@
+; RUN: llc < %s -mtriple=i686-pc-win32 | FileCheck %s
+
+%Foo = type { i32, i32 }
+
+declare x86_stdcallcc void @f(%Foo* inalloca %a)
+declare x86_stdcallcc void @i(i32 %a)
+
+define void @g() {
+  %b = alloca %Foo, inalloca
+; CHECK: movl    $8, %eax
+; CHECK: calll   __chkstk
+; CHECK: movl   %[[REG:[^,]*]], %esp
+  %f1 = getelementptr %Foo* %b, i32 0, i32 0
+  %f2 = getelementptr %Foo* %b, i32 0, i32 1
+  store i32 13, i32* %f1
+  store i32 42, i32* %f2
+; CHECK: movl    $13, (%[[REG]])
+; CHECK: movl    $42, 4(%[[REG]])
+  call x86_stdcallcc void @f(%Foo* inalloca %b)
+; CHECK: calll   _f@8
+; CHECK-NOT: %esp
+; CHECK: subl $4, %esp
+; CHECK: calll   _i@4
+  call x86_stdcallcc void @i(i32 0)
+  ret void
+}

diff --git a/test/CodeGen/X86/inalloca.ll b/test/CodeGen/X86/inalloca.ll
new file mode 100644 (file)
index 0000000..84d694a
--- /dev/null
+++ b/test/CodeGen/X86/inalloca.ll
@@ -0,0 +1,65 @@
+; RUN: llc < %s -mtriple=i686-pc-win32 | FileCheck %s
+
+%Foo = type { i32, i32 }
+
+declare void @f(%Foo* inalloca %b)
+
+define void @a() {
+; CHECK-LABEL: _a:
+entry:
+  %b = alloca %Foo, inalloca
+; CHECK: movl    $8, %eax
+; CHECK: calll   __chkstk
+; CHECK: movl   %[[REG:[^,]*]], %esp
+  %f1 = getelementptr %Foo* %b, i32 0, i32 0
+  %f2 = getelementptr %Foo* %b, i32 0, i32 1
+  store i32 13, i32* %f1
+  store i32 42, i32* %f2
+; CHECK: movl    $13, (%[[REG]])
+; CHECK: movl    $42, 4(%[[REG]])
+  call void @f(%Foo* inalloca %b)
+; CHECK: calll   _f
+  ret void
+}
+
+declare void @inreg_with_inalloca(i32 inreg %a, %Foo* inalloca %b)
+
+define void @b() {
+; CHECK-LABEL: _b:
+entry:
+  %b = alloca %Foo, inalloca
+; CHECK: movl    $8, %eax
+; CHECK: calll   __chkstk
+; CHECK: movl   %[[REG:[^,]*]], %esp
+  %f1 = getelementptr %Foo* %b, i32 0, i32 0
+  %f2 = getelementptr %Foo* %b, i32 0, i32 1
+  store i32 13, i32* %f1
+  store i32 42, i32* %f2
+; CHECK: movl    $13, (%[[REG]])
+; CHECK: movl    $42, 4(%[[REG]])
+  call void @inreg_with_inalloca(i32 inreg 1, %Foo* inalloca %b)
+; CHECK: movl    $1, %eax
+; CHECK: calll   _inreg_with_inalloca
+  ret void
+}
+
+declare x86_thiscallcc void @thiscall_with_inalloca(i8* %a, %Foo* inalloca %b)
+
+define void @c() {
+; CHECK-LABEL: _c:
+entry:
+  %b = alloca %Foo, inalloca
+; CHECK: movl    $8, %eax
+; CHECK: calll   __chkstk
+; CHECK: movl   %[[REG:[^,]*]], %esp
+  %f1 = getelementptr %Foo* %b, i32 0, i32 0
+  %f2 = getelementptr %Foo* %b, i32 0, i32 1
+  store i32 13, i32* %f1
+  store i32 42, i32* %f2
+; CHECK: movl    $13, (%[[REG]])
+; CHECK: movl    $42, 4(%[[REG]])
+  call x86_thiscallcc void @thiscall_with_inalloca(i8* null, %Foo* inalloca %b)
+; CHECK: xorl    %ecx, %ecx
+; CHECK: calll   _thiscall_with_inalloca
+  ret void
+}