Update optimization passes to handle inalloca arguments

Summary:
I searched Transforms/ and Analysis/ for 'ByVal' and updated those call
sites to check for inalloca if appropriate.

I added tests for any change that would allow an optimization to fire on
inalloca.

Reviewers: nlewycky

Differential Revision: http://llvm-reviews.chandlerc.com/D2449

llvm-svn: 200281

llvm/lib/Analysis/MemoryBuiltins.cpp

@@ -458,7 +458,7 @@ SizeOffsetType ObjectSizeOffsetVisitor::visitAllocaInst(AllocaInst &I) {
 
 SizeOffsetType ObjectSizeOffsetVisitor::visitArgument(Argument &A) {
   // no interprocedural analysis is done at the moment
-  if (!A.hasByValAttr()) {
+  if (!A.hasByValOrInAllocaAttr()) {
     ++ObjectVisitorArgument;
     return unknown();
   }

llvm/lib/Analysis/ValueTracking.cpp

@@ -311,8 +311,9 @@ void llvm::ComputeMaskedBits(Value *V, APInt &KnownZero, APInt &KnownOne,
   if (Argument *A = dyn_cast<Argument>(V)) {
     unsigned Align = 0;
 
-    if (A->hasByValAttr()) {
-      // Get alignment information off byval arguments if specified in the IR.
+    if (A->hasByValOrInAllocaAttr()) {
+      // Get alignment information off byval/inalloca arguments if specified in
+      // the IR.
       Align = A->getParamAlignment();
     } else if (TD && A->hasStructRetAttr()) {
       // An sret parameter has at least the ABI alignment of the return type.
@@ -2070,9 +2071,9 @@ bool llvm::isKnownNonNull(const Value *V, const TargetLibraryInfo *TLI) {
   // Alloca never returns null, malloc might.
   if (isa<AllocaInst>(V)) return true;
 
-  // A byval argument is never null.
+  // A byval or inalloca argument is never null.
   if (const Argument *A = dyn_cast<Argument>(V))
-    return A->hasByValAttr();
+    return A->hasByValOrInAllocaAttr();
 
   // Global values are not null unless extern weak.
   if (const GlobalValue *GV = dyn_cast<GlobalValue>(V))

llvm/lib/Transforms/IPO/ArgumentPromotion.cpp

@@ -155,7 +155,8 @@ CallGraphNode *ArgPromotion::PromoteArguments(CallGraphNode *CGN) {
     Type *AgTy = cast<PointerType>(PtrArg->getType())->getElementType();
 
     // If this is a byval argument, and if the aggregate type is small, just
-    // pass the elements, which is always safe.
+    // pass the elements, which is always safe.  This does not apply to
+    // inalloca.
     if (PtrArg->hasByValAttr()) {
       if (StructType *STy = dyn_cast<StructType>(AgTy)) {
         if (maxElements > 0 && STy->getNumElements() > maxElements) {
@@ -201,7 +202,7 @@ CallGraphNode *ArgPromotion::PromoteArguments(CallGraphNode *CGN) {
     }
     
     // Otherwise, see if we can promote the pointer to its value.
-    if (isSafeToPromoteArgument(PtrArg, PtrArg->hasByValAttr()))
+    if (isSafeToPromoteArgument(PtrArg, PtrArg->hasByValOrInAllocaAttr()))
       ArgsToPromote.insert(PtrArg);
   }
 
@@ -301,7 +302,8 @@ static void MarkIndicesSafe(const ArgPromotion::IndicesVector &ToMark,
 /// This method limits promotion of aggregates to only promote up to three
 /// elements of the aggregate in order to avoid exploding the number of
 /// arguments passed in.
-bool ArgPromotion::isSafeToPromoteArgument(Argument *Arg, bool isByVal) const {
+bool ArgPromotion::isSafeToPromoteArgument(Argument *Arg,
+                                           bool isByValOrInAlloca) const {
   typedef std::set<IndicesVector> GEPIndicesSet;
 
   // Quick exit for unused arguments
@@ -323,6 +325,9 @@ bool ArgPromotion::isSafeToPromoteArgument(Argument *Arg, bool isByVal) const {
   //
   // This set will contain all sets of indices that are loaded in the entry
   // block, and thus are safe to unconditionally load in the caller.
+  //
+  // This optimization is also safe for InAlloca parameters, because it verifies
+  // that the address isn't captured.
   GEPIndicesSet SafeToUnconditionallyLoad;
 
   // This set contains all the sets of indices that we are planning to promote.
@@ -330,7 +335,7 @@ bool ArgPromotion::isSafeToPromoteArgument(Argument *Arg, bool isByVal) const {
   GEPIndicesSet ToPromote;
 
   // If the pointer is always valid, any load with first index 0 is valid.
-  if (isByVal || AllCallersPassInValidPointerForArgument(Arg))
+  if (isByValOrInAlloca || AllCallersPassInValidPointerForArgument(Arg))
     SafeToUnconditionallyLoad.insert(IndicesVector(1, 0));
 
   // First, iterate the entry block and mark loads of (geps of) arguments as
@@ -389,7 +394,7 @@ bool ArgPromotion::isSafeToPromoteArgument(Argument *Arg, bool isByVal) const {
         // TODO: This runs the above loop over and over again for dead GEPs
         // Couldn't we just do increment the UI iterator earlier and erase the
         // use?
-        return isSafeToPromoteArgument(Arg, isByVal);
+        return isSafeToPromoteArgument(Arg, isByValOrInAlloca);
       }
 
       // Ensure that all of the indices are constants.

llvm/lib/Transforms/IPO/DeadArgumentElimination.cpp

@@ -378,7 +378,7 @@ bool DAE::RemoveDeadArgumentsFromCallers(Function &Fn)
        I != E; ++I) {
     Argument *Arg = I;
 
-    if (Arg->use_empty() && !Arg->hasByValAttr())
+    if (Arg->use_empty() && !Arg->hasByValOrInAllocaAttr())
       UnusedArgs.push_back(Arg->getArgNo());
   }
 

llvm/lib/Transforms/IPO/FunctionAttrs.cpp

@@ -414,6 +414,10 @@ determinePointerReadAttrs(Argument *A,
   SmallSet<Use*, 32> Visited;
   int Count = 0;
 
+  // inalloca arguments are always clobbered by the call.
+  if (A->hasInAllocaAttr())
+    return Attribute::None;
+
   bool IsRead = false;
   // We don't need to track IsWritten. If A is written to, return immediately.
 

llvm/lib/Transforms/IPO/IPConstantPropagation.cpp

@@ -135,7 +135,7 @@ bool IPCP::PropagateConstantsIntoArguments(Function &F) {
   for (unsigned i = 0, e = ArgumentConstants.size(); i != e; ++i, ++AI) {
     // Do we have a constant argument?
     if (ArgumentConstants[i].second || AI->use_empty() ||
-        (AI->hasByValAttr() && !F.onlyReadsMemory()))
+        AI->hasInAllocaAttr() || (AI->hasByValAttr() && !F.onlyReadsMemory()))
       continue;
   
     Value *V = ArgumentConstants[i].first;

llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp

@@ -767,10 +767,10 @@ static bool isSafeToEliminateVarargsCast(const CallSite CS,
   if (!CI->isLosslessCast())
     return false;
 
-  // The size of ByVal arguments is derived from the type, so we
+  // The size of ByVal or InAlloca arguments is derived from the type, so we
   // can't change to a type with a different size.  If the size were
   // passed explicitly we could avoid this check.
-  if (!CS.isByValArgument(ix))
+  if (!CS.isByValOrInAllocaArgument(ix))
     return true;
 
   Type* SrcTy =
@@ -1049,6 +1049,9 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
                         typeIncompatible(ParamTy, i + 1), i + 1))
       return false;   // Attribute not compatible with transformed value.
 
+    if (CS.isInAllocaArgument(i))
+      return false;   // Cannot transform to and from inalloca.
+
     // If the parameter is passed as a byval argument, then we have to have a
     // sized type and the sized type has to have the same size as the old type.
     if (ParamTy != ActTy &&

llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

@@ -81,10 +81,14 @@ isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
       if (CS.isCallee(UI))
         continue;
 
+      // Inalloca arguments are clobbered by the call.
+      unsigned ArgNo = CS.getArgumentNo(UI);
+      if (CS.isInAllocaArgument(ArgNo))
+        return false;
+
       // If this is a readonly/readnone call site, then we know it is just a
       // load (but one that potentially returns the value itself), so we can
       // ignore it if we know that the value isn't captured.
-      unsigned ArgNo = CS.getArgumentNo(UI);
       if (CS.onlyReadsMemory() &&
           (CS.getInstruction()->use_empty() || CS.doesNotCapture(ArgNo)))
         continue;

llvm/lib/Transforms/ObjCARC/ObjCARC.h

@@ -308,6 +308,7 @@ static inline bool IsPotentialRetainableObjPtr(const Value *Op) {
   // Special arguments can not be a valid retainable object pointer.
   if (const Argument *Arg = dyn_cast<Argument>(Op))
     if (Arg->hasByValAttr() ||
+        Arg->hasInAllocaAttr() ||
         Arg->hasNestAttr() ||
         Arg->hasStructRetAttr())
       return false;

llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp

@@ -374,8 +374,8 @@ static OverwriteResult isOverwrite(const AliasAnalysis::Location &Later,
     return OverwriteUnknown;
 
   // Check to see if the later store is to the entire object (either a global,
-  // an alloca, or a byval argument).  If so, then it clearly overwrites any
-  // other store to the same object.
+  // an alloca, or a byval/inalloca argument).  If so, then it clearly
+  // overwrites any other store to the same object.
   const DataLayout *TD = AA.getDataLayout();
 
   const Value *UO1 = GetUnderlyingObject(P1, TD),
@@ -742,11 +742,11 @@ bool DSE::handleEndBlock(BasicBlock &BB) {
       DeadStackObjects.insert(I);
   }
 
-  // Treat byval arguments the same, stores to them are dead at the end of the
-  // function.
+  // Treat byval or inalloca arguments the same, stores to them are dead at the
+  // end of the function.
   for (Function::arg_iterator AI = BB.getParent()->arg_begin(),
        AE = BB.getParent()->arg_end(); AI != AE; ++AI)
-    if (AI->hasByValAttr())
+    if (AI->hasByValOrInAllocaAttr())
       DeadStackObjects.insert(AI);
 
   // Scan the basic block backwards

llvm/test/Transforms/ArgumentPromotion/inalloca.ll

@@ -0,0 +1,49 @@
+; RUN: opt %s -argpromotion -scalarrepl -S | FileCheck %s
+
+target datalayout = "E-p:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128"
+
+%struct.ss = type { i32, i32 }
+
+; Argpromote + scalarrepl should change this to passing the two integers by value.
+define internal i32 @f(%struct.ss* inalloca  %s) {
+entry:
+  %f0 = getelementptr %struct.ss* %s, i32 0, i32 0
+  %f1 = getelementptr %struct.ss* %s, i32 0, i32 1
+  %a = load i32* %f0, align 4
+  %b = load i32* %f1, align 4
+  %r = add i32 %a, %b
+  ret i32 %r
+}
+; CHECK-LABEL: define internal i32 @f
+; CHECK-NOT: load
+; CHECK: ret
+
+define i32 @main() {
+entry:
+  %S = alloca %struct.ss
+  %f0 = getelementptr %struct.ss* %S, i32 0, i32 0
+  %f1 = getelementptr %struct.ss* %S, i32 0, i32 1
+  store i32 1, i32* %f0, align 4
+  store i32 2, i32* %f1, align 4
+  %r = call i32 @f(%struct.ss* inalloca %S)
+  ret i32 %r
+}
+; CHECK-LABEL: define i32 @main
+; CHECK-NOT: load
+; CHECK: ret
+
+; Argpromote can't promote %a because of the icmp use.
+define internal i1 @g(%struct.ss* %a, %struct.ss* inalloca %b) nounwind  {
+; CHECK: define internal i1 @g(%struct.ss* %a, %struct.ss* inalloca %b)
+entry:
+  %c = icmp eq %struct.ss* %a, %b
+  ret i1 %c
+}
+
+define i32 @test() {
+entry:
+  %S = alloca %struct.ss
+  %c = call i1 @g(%struct.ss* %S, %struct.ss* inalloca %S)
+; CHECK: call i1 @g(%struct.ss* %S, %struct.ss* inalloca %S)
+  ret i32 0
+}

llvm/test/Transforms/DeadStoreElimination/simple.ll

@@ -105,6 +105,15 @@ define void @test9(%struct.x* byval  %a) nounwind  {
 ; CHECK-NEXT: ret void
 }
 
+; Test for inalloca handling.
+define void @test9_2(%struct.x* inalloca  %a) nounwind  {
+	%tmp2 = getelementptr %struct.x* %a, i32 0, i32 0
+	store i32 1, i32* %tmp2, align 4
+	ret void
+; CHECK-LABEL: @test9_2(
+; CHECK-NEXT: ret void
+}
+
 ; va_arg has fuzzy dependence, the store shouldn't be zapped.
 define double @test10(i8* %X) {
         %X_addr = alloca i8*

llvm/test/Transforms/FunctionAttrs/readattrs.ll

@@ -45,3 +45,9 @@ define void @test6_2(i8** %p, i8* %q) {
   call void @test6_1()
   ret void
 }
+
+; CHECK: define void @test7_1(i32* inalloca nocapture %a)
+; inalloca parameters are always considered written
+define void @test7_1(i32* inalloca %a) {
+  ret void
+}

llvm/test/Transforms/InstCombine/call-cast-target-inalloca.ll

@@ -0,0 +1,22 @@
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+target datalayout = "e-p:32:32"
+target triple = "i686-pc-linux-gnu"
+
+declare void @takes_i32(i32)
+declare void @takes_i32_inalloca(i32* inalloca)
+
+define void @f() {
+; CHECK-LABEL: define void @f()
+  %args = alloca i32
+  call void bitcast (void (i32)* @takes_i32 to void (i32*)*)(i32* inalloca %args)
+; CHECK: call void bitcast
+  ret void
+}
+
+define void @g() {
+; CHECK-LABEL: define void @g()
+  call void bitcast (void (i32*)* @takes_i32_inalloca to void (i32)*)(i32 0)
+; CHECK: call void bitcast
+  ret void
+}