commit df96819daf5555cc4e3ecef96d09b0397da8ccae
author Eli Friedman <eli.friedman@gmail.com> Thu May 26 00:10:27 2011 +0000
committer Eli Friedman <eli.friedman@gmail.com> Thu May 26 00:10:27 2011 +0000
tree 5ea50f0cb8bd965287f30f7d31b8b5954a870a18
parent 25d5745cc5db8ce0da4ed5e3b20587800ae15b65

Skip extra copy from aggregate where it isn't necessary; rdar://problem/8139919 .  This shouldn't make much of a difference at -O3, but should substantially reduce the number of generated memcpy's at -O0.

Originally r130717, but was backed out due to an ObjC regression.

llvm-svn: 132102

clang/lib/CodeGen/CGCall.cpp

diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index 4c9f3d4..712ae89 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -1189,6 +1189,15 @@
     return args.add(EmitReferenceBindingToExpr(E, /*InitializedDecl=*/0),
                     type);
 
+  if (hasAggregateLLVMType(type) && isa<ImplicitCastExpr>(E) &&
+      cast<CastExpr>(E)->getCastKind() == CK_LValueToRValue) {
+    LValue L = EmitLValue(cast<CastExpr>(E)->getSubExpr());
+    assert(L.isSimple());
+    args.add(RValue::getAggregate(L.getAddress(), L.isVolatileQualified()),
+             type, /*NeedsCopy*/true);
+    return;
+  }
+
   args.add(EmitAnyExprToTemp(E), type);
 }
 
@@ -1254,6 +1263,10 @@
                             Alignment, I->Ty);
         else
           StoreComplexToAddr(RV.getComplexVal(), Args.back(), false);
+      } else if (I->NeedsCopy && !ArgInfo.getIndirectByVal()) {
+        Args.push_back(CreateMemTemp(I->Ty));
+        EmitAggregateCopy(Args.back(), RV.getAggregateAddr(), I->Ty,
+                          RV.isVolatileQualified());
       } else {
         Args.push_back(RV.getAggregateAddr());
       }

clang/lib/CodeGen/CGCall.h

diff --git a/clang/lib/CodeGen/CGCall.h b/clang/lib/CodeGen/CGCall.h
index 3f600c0..160a62e 100644
--- a/clang/lib/CodeGen/CGCall.h
+++ b/clang/lib/CodeGen/CGCall.h
@@ -47,8 +47,9 @@
   struct CallArg {
     RValue RV;
     QualType Ty;
-    CallArg(RValue rv, QualType ty)
-    : RV(rv), Ty(ty)
+    bool NeedsCopy;
+    CallArg(RValue rv, QualType ty, bool needscopy)
+    : RV(rv), Ty(ty), NeedsCopy(needscopy)
     { }
   };
 
@@ -57,8 +58,8 @@
   class CallArgList :
     public llvm::SmallVector<CallArg, 16> {
   public:
-    void add(RValue rvalue, QualType type) {
-      push_back(CallArg(rvalue, type));
+    void add(RValue rvalue, QualType type, bool needscopy = false) {
+      push_back(CallArg(rvalue, type, needscopy));
     }
   };