diff --git a/docs/ispc.rst b/docs/ispc.rst
index e3138760..ad364396 100644
--- a/docs/ispc.rst
+++ b/docs/ispc.rst
@@ -2938,23 +2938,28 @@ Function Overloading
 --------------------
 
 Functions can be overloaded by parameter type.  Given multiple definitions
-of a function, ``ispc`` uses the following methods to try to find a match.
-If a single match of a given type is found, it is used; if multiple matches
-of a given type are found, an error is issued.
+of a function, ``ispc`` uses the following model to choose the best function:
+each conversion of two types has its cost. ``ispc`` tries to find conversion
+with the smallest cost. When ``ispc`` can't find any conversion it means that
+this function is not suitable. Then ``ispc`` sums costs for all arguments and
+chooses the function with the smallest final cost. If the chosen function 
+has some arguments which costs are bigger than their costs in other function
+this treats as ambiguous.
+Costs of type conversions placed from small to big:
 
-* All parameter types match exactly.
-* All parameter types match exactly, where any reference-type
-  parameters are considered equivalent to their underlying type.
-* Parameters match with only type conversions that don't risk losing any
-  information (for example, converting an ``int16`` value to an ``int32``
-  parameter value.)
-* Parameters match with only promotions from ``uniform`` to ``varying``
-  types.
-* Parameters match using arbitrary type conversion, without changing
-  variability from ``uniform`` to ``varying`` (e.g., ``int`` to ``float``,
-  ``float`` to ``int``.)
-* Parameters match using arbitrary type conversion, including also changing
-  variability from ``uniform`` to ``varying`` as needed.
+1. Parameter types match exactly.
+2. Function parameter type is reference and parameters match when any reference-type parameter are considered equivalent to their underlying type.
+3. Function parameter type is const-reference and parameters match when any reference-type parameter are considered equivalent to their underlying type ignoring const attributes.
+4. Parameters match exactly, except constant attributes. [NO CONSTANT ATTRIBUTES LATER]
+5. Parameters match exactly, except reference attributes. [NO REFERENCES ATTRIBUTES LATER]
+6. Parameters match with only type conversions that don't risk losing any information (for example, converting an int16 value to an int32 parameter value.)
+7. Parameters match with only promotions from uniform to varying types.
+8. Parameters match using arbitrary type conversion, without changing variability from uniform to varying (e.g., int to float, float to int.)
+9. Parameters match with widening and promotions from uniform to varying types. (combination of "6" and "7")
+10. Parameters match using arbitrary type conversion, including also changing variability from uniform to varying.
+
+* If function parameter type is reference and neither "2" nor "3" aren't suitable, function is not suitable
+* If "10" isn't suitable, function is not suitable
 
 
 Re-establishing The Execution Mask
diff --git a/expr.cpp b/expr.cpp
index cb1f8dcc..57c140cc 100644
--- a/expr.cpp
+++ b/expr.cpp
@@ -8097,23 +8097,6 @@ lGetOverloadCandidateMessage(const std::vector<Symbol *> &funcs,
 }
 
 
-static bool
-lIsMatchToNonConstReference(const Type *callType, const Type *funcArgType) {
-    return (CastType<ReferenceType>(funcArgType) &&
-            (funcArgType->IsConstType() == false) &&
-            Type::Equal(callType, funcArgType->GetReferenceTarget()));
-}
-
-
-static bool
-lIsMatchToNonConstReferenceUnifToVarying(const Type *callType,
-                                         const Type *funcArgType) {
-    return (CastType<ReferenceType>(funcArgType) &&
-            (funcArgType->IsConstType() == false) &&
-            Type::Equal(callType->GetAsVaryingType(),
-                        funcArgType->GetReferenceTarget()));
-}
-
 /** Helper function used for function overload resolution: returns true if
     converting the argument to the call type only requires a type
     conversion that won't lose information.  Otherwise return false.
@@ -8160,31 +8143,6 @@ lIsMatchWithTypeWidening(const Type *callType, const Type *funcArgType) {
 }
 
 
-/** Helper function used for function overload resolution: returns true if
-    the call argument type and the function argument type match if we only
-    do a uniform -> varying type conversion but otherwise have exactly the
-    same type.
- */
-static bool
-lIsMatchWithUniformToVarying(const Type *callType, const Type *funcArgType) {
-    return (callType->IsUniformType() &&
-            funcArgType->IsVaryingType() &&
-            Type::EqualIgnoringConst(callType->GetAsVaryingType(), funcArgType));
-}
-
-
-/** Helper function used for function overload resolution: returns true if
-    we can type convert from the call argument type to the function
-    argument type, but without doing a uniform -> varying conversion.
- */
-static bool
-lIsMatchWithTypeConvSameVariability(const Type *callType,
-                                    const Type *funcArgType) {
-    return (CanConvertTypes(callType, funcArgType) &&
-            (callType->GetVariability() == funcArgType->GetVariability()));
-}
-
-
 /* Returns the set of function overloads that are potential matches, given
    argCount values being passed as arguments to the function call.
  */
@@ -8236,13 +8194,15 @@ int
 FunctionSymbolExpr::computeOverloadCost(const FunctionType *ftype,
                                         const std::vector<const Type *> &argTypes,
                                         const std::vector<bool> *argCouldBeNULL,
-                                        const std::vector<bool> *argIsConstant) {
+                                        const std::vector<bool> *argIsConstant,
+                                        int * cost) {
     int costSum = 0;
 
     // In computing the cost function, we only worry about the actual
     // argument types--using function default parameter values is free for
     // the purposes here...
     for (int i = 0; i < (int)argTypes.size(); ++i) {
+        cost[i] = 0;
         // The cost imposed by this argument will be a multiple of
         // costScale, which has a value set so that for each of the cost
         // buckets, even if all of the function arguments undergo the next
@@ -8255,51 +8215,105 @@ FunctionSymbolExpr::computeOverloadCost(const FunctionType *ftype,
 
         if (Type::Equal(callType, fargType))
             // Perfect match: no cost
-            costSum += 0;
+            // Step "1" from documentation
+            cost[i] += 0;
         else if (argCouldBeNULL && (*argCouldBeNULL)[i] &&
                  lArgIsPointerType(fargType))
-            // Passing NULL to a pointer-typed parameter is also a no-cost
-            // operation
-            costSum += 0;
+            // Passing NULL to a pointer-typed parameter is also a no-cost operation
+            // Step "1" from documentation
+            cost[i] += 0;
         else {
             // If the argument is a compile-time constant, we'd like to
             // count the cost of various conversions as much lower than the
             // cost if it wasn't--so scale up the cost when this isn't the
             // case..
             if (argIsConstant == NULL || (*argIsConstant)[i] == false)
-                costScale *= 128;
+                costScale *= 512;
 
-            // For convenience, normalize to non-const types (except for
-            // references, where const-ness matters).  For all other types,
-            // we're passing by value anyway, so const doesn't matter.
-            const Type *callTypeNC = callType, *fargTypeNC = fargType;
-            if (CastType<ReferenceType>(callType) == NULL)
-                callTypeNC = callType->GetAsNonConstType();
-            if (CastType<ReferenceType>(fargType) == NULL)
-                fargTypeNC = fargType->GetAsNonConstType();
+            if (CastType<ReferenceType>(fargType)) {
+                // Here we completely handle the case where fargType is reference.
+                if (callType->IsConstType() && !fargType->IsConstType()) {
+                    // It is forbidden to pass const object to non-const reference (cvf -> vfr)
+                    return -1;
+                }
+                if (!callType->IsConstType() && fargType->IsConstType()) {
+                    // It is possible to pass (vf -> cvfr)
+                    // but it is worse than (vf -> vfr) or (cvf -> cvfr)
+                    // Step "3" from documentation
+                    cost[i] += 2 * costScale;
+                }
+                if (!Type::Equal(callType->GetReferenceTarget()->GetAsNonConstType(),
+                                 fargType->GetReferenceTarget()->GetAsNonConstType())) {
+                    // Types under references must be equal completely.
+                    // vd -> vfr or vd -> cvfr are forbidden. (Although clang allows vd -> cvfr case.)
+                    return -1;
+                }
+                // penalty for equal types under reference (vf -> vfr is worse than vf -> vf)
+                // Step "2" from documentation
+                cost[i] += 2 * costScale;
+                continue;
+            }
+            const Type *callTypeNP = callType;
+            if (CastType<ReferenceType>(callType)) {
+                callTypeNP = callType->GetReferenceTarget();
+                // we can treat vfr as vf for callType with some penalty
+                // Step "5" from documentation
+                cost[i] += 2 * costScale;
+            }
 
-            if (Type::Equal(callTypeNC, fargTypeNC))
-                // Exact match (after dealing with references, above)
-                costSum += 1 * costScale;
-            // note: orig fargType for the next two...
-            else if (lIsMatchToNonConstReference(callTypeNC, fargType))
-                costSum += 2 * costScale;
-            else if (lIsMatchToNonConstReferenceUnifToVarying(callTypeNC, fargType))
-                costSum += 4 * costScale;
-            else if (lIsMatchWithTypeWidening(callTypeNC, fargTypeNC))
-                costSum += 8 * costScale;
-            else if (lIsMatchWithUniformToVarying(callTypeNC, fargTypeNC))
-                costSum += 16 * costScale;
-            else if (lIsMatchWithTypeConvSameVariability(callTypeNC, fargTypeNC))
-                costSum += 32 * costScale;
-            else if (CanConvertTypes(callTypeNC, fargTypeNC))
-                costSum += 64 * costScale;
+            // Now we deal with references, so we can normalize to non-const types
+            // because we're passing by value anyway, so const doesn't matter.
+            const Type *callTypeNC = callTypeNP, *fargTypeNC = fargType;
+            callTypeNC = callTypeNP->GetAsNonConstType();
+            fargTypeNC = fargType->GetAsNonConstType();
+
+            // Now we forget about constants and references!
+            if (Type::Equal(callTypeNC, fargTypeNC)) {
+                // The best case: vf -> vf.
+                // Step "4" from documentation
+                cost[i] += 1 * costScale;
+                continue;
+            }
+            if (lIsMatchWithTypeWidening(callTypeNC, fargTypeNC)) {
+                // A little bit worse case: vf -> vd.
+                // Step "6" from documentation
+                cost[i] += 8 * costScale;
+                continue;
+            }
+            if (fargType->IsVaryingType() && callType->IsUniformType()) {
+                // Here we deal with brodcasting uniform to varying.
+                // callType - varying and fargType - uniform is forbidden.
+                if (Type::Equal(callTypeNC->GetAsVaryingType(), fargTypeNC)) {
+                    // uf -> vf is better than uf -> ui or uf -> ud
+                    // Step "7" from documentation
+                    cost[i] += 16 * costScale;
+                    continue;
+                }
+                if (lIsMatchWithTypeWidening(callTypeNC->GetAsVaryingType(), fargTypeNC)) {
+                    // uf -> vd is better than uf -> vi (128 < 128 + 64)
+                    // but worse than uf -> ui (128 > 64)
+                    // Step "9" from documentation
+                    cost[i] += 128 * costScale;
+                    continue;
+                }
+                // 128 + 64 is the max. uf -> vi is the worst case.
+                // Step "10" from documentation
+                cost[i] += 128 * costScale;
+            }
+            if (CanConvertTypes(callTypeNC, fargTypeNC))
+                // two cases: the worst is 128 + 64: uf -> vi and
+                // the only 64: (64 < 128) uf -> ui worse than uf -> vd
+                // Step "8" from documentation
+                cost[i] += 64 * costScale;
             else
                 // Failure--no type conversion possible...
                 return -1;
         }
     }
 
+    for (int i = 0; i < (int)argTypes.size(); ++i) {
+        costSum = costSum + cost[i];
+    }
     return costSum;
 }
 
@@ -8334,6 +8348,7 @@ FunctionSymbolExpr::ResolveOverloads(SourcePos argPos,
     int bestMatchCost = 1<<30;
     std::vector<Symbol *> matches;
     std::vector<int> candidateCosts;
+    std::vector<int*> candidateExpandCosts;
 
     if (actualCandidates.size() == 0)
         goto failure;
@@ -8343,9 +8358,12 @@ FunctionSymbolExpr::ResolveOverloads(SourcePos argPos,
         const FunctionType *ft =
             CastType<FunctionType>(actualCandidates[i]->type);
         AssertPos(pos, ft != NULL);
+        int * cost = new int[argTypes.size()];
         candidateCosts.push_back(computeOverloadCost(ft, argTypes,
                                                      argCouldBeNULL,
-                                                     argIsConstant));
+                                                     argIsConstant,
+                                                     cost));
+        candidateExpandCosts.push_back(cost);
     }
 
     // Find the best cost, and then the candidate or candidates that have
@@ -8358,8 +8376,28 @@ FunctionSymbolExpr::ResolveOverloads(SourcePos argPos,
     if (bestMatchCost == (1<<30))
         goto failure;
     for (int i = 0; i < (int)candidateCosts.size(); ++i) {
-        if (candidateCosts[i] == bestMatchCost)
+        if (candidateCosts[i] == bestMatchCost) {
+            for (int j = 0; j < (int)candidateCosts.size(); ++j) {
+                for (int k = 0; k < argTypes.size(); k++) {
+                    if (candidateCosts[j] != -1 &&
+                        candidateExpandCosts[j][k] < candidateExpandCosts[i][k]) {
+                        std::vector<Symbol *> temp;
+                        temp.push_back(actualCandidates[i]);
+                        temp.push_back(actualCandidates[j]);
+                        std::string candidateMessage =
+                            lGetOverloadCandidateMessage(temp, argTypes, argCouldBeNULL);
+                        Warning(pos, "call to \"%s\" is ambiguous. "
+                                    "This warning will be turned into error in the next ispc release.\n"
+                                    "Please add explicit cast to arguments to have unambiguous match."
+                                    "\n%s", funName, candidateMessage.c_str());
+                    }
+                }
+            }
             matches.push_back(actualCandidates[i]);
+        }
+    }
+    for (int i = 0; i < (int)candidateExpandCosts.size(); ++i) {
+        delete [] candidateExpandCosts[i];
     }
 
     if (matches.size() == 1) {
diff --git a/expr.h b/expr.h
index b539ff1b..7d209bba 100644
--- a/expr.h
+++ b/expr.h
@@ -635,7 +635,8 @@ private:
     static int computeOverloadCost(const FunctionType *ftype,
                                    const std::vector<const Type *> &argTypes,
                                    const std::vector<bool> *argCouldBeNULL,
-                            const std::vector<bool> *argIsConstant);
+                                   const std::vector<bool> *argIsConstant,
+                                   int * cost);
 
     /** Name of the function that is being called. */
     std::string name;
diff --git a/module.cpp b/module.cpp
index d8f165bc..db378591 100644
--- a/module.cpp
+++ b/module.cpp
@@ -840,7 +840,6 @@ Module::AddFunctionDeclaration(const std::string &name,
     // Set function attributes: we never throw exceptions
     function->setDoesNotThrow();
     if (storageClass != SC_EXTERN_C &&
-        !g->generateDebuggingSymbols &&
         isInline)
 #ifdef LLVM_3_2
         function->addFnAttr(llvm::Attributes::AlwaysInline);
@@ -1273,7 +1272,12 @@ lEmitStructDecl(const StructType *st, std::vector<const StructType *> *emittedSt
     for (int i = 0; i < st->GetElementCount(); ++i) {
         const Type *type = st->GetElementType(i)->GetAsNonConstType();
         std::string d = type->GetCDeclaration(st->GetElementName(i));
-        fprintf(file, "    %s;\n", d.c_str());
+        if (type->IsVaryingType()) {
+          fprintf(file, "    %s[%d];\n", d.c_str(), g->target->getVectorWidth());
+        }
+        else {
+          fprintf(file, "    %s;\n", d.c_str());
+        }
     }
     fprintf(file, "};\n");
     fprintf(file, "#endif\n\n");
diff --git a/tests/array-mixed-unif-vary-indexing-2.ispc b/tests/array-mixed-unif-vary-indexing-2.ispc
index 8143ca29..a885a136 100644
--- a/tests/array-mixed-unif-vary-indexing-2.ispc
+++ b/tests/array-mixed-unif-vary-indexing-2.ispc
@@ -16,7 +16,7 @@ export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
     else
         x[a][b-1] = 1;
 
-    a = min(a, 46);
+    a = min(a, 46.f);
 
     RET[programIndex] = x[3][a];
 }
diff --git a/tests/array-mixed-unif-vary-indexing.ispc b/tests/array-mixed-unif-vary-indexing.ispc
index 96fc0870..bd3e6919 100644
--- a/tests/array-mixed-unif-vary-indexing.ispc
+++ b/tests/array-mixed-unif-vary-indexing.ispc
@@ -10,7 +10,7 @@ export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
         for (uniform int j = 0; j < 47; ++j)
             x[i][j] = 2+b-5;
 
-    a = min(a,46);
+    a = min(a,46.f);
     x[a][b-1] = 0;
     RET[programIndex] = x[2][a];
 }
diff --git a/tests/array-multidim-gather-scatter.ispc b/tests/array-multidim-gather-scatter.ispc
index 1528b070..4fc59f52 100644
--- a/tests/array-multidim-gather-scatter.ispc
+++ b/tests/array-multidim-gather-scatter.ispc
@@ -11,7 +11,7 @@ export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
 
     uniform int index[4] = { 0, 1, 2, 4 };
     float v = index[programIndex & 0x3];
-    x[min(a,39)][v] = 0;
+    x[min(a,39.f)][v] = 0;
     RET[programIndex] = x[v+1][v];
 }
 
diff --git a/tests/func-overload-max.ispc b/tests/func-overload-max.ispc
index 37360030..44b20db7 100644
--- a/tests/func-overload-max.ispc
+++ b/tests/func-overload-max.ispc
@@ -4,7 +4,7 @@ export uniform int width() { return programCount; }
 
 export void f_f(uniform float RET[], uniform float aFOO[]) {
     float a = 1. / aFOO[programIndex]; 
-    RET[programIndex] = max(0, a); 
+    RET[programIndex] = max(0.f, a); 
 }
 
 export void result(uniform float RET[]) {
diff --git a/tests/funcptr-uniform-10.ispc b/tests/funcptr-uniform-10.ispc
index 75fcbfe9..8490778f 100644
--- a/tests/funcptr-uniform-10.ispc
+++ b/tests/funcptr-uniform-10.ispc
@@ -16,7 +16,7 @@ export void f_f(uniform float RET[], uniform float aFOO[]) {
     uniform FuncType func[] = { bar, foo };
     float a = aFOO[programIndex]; 
     float b = aFOO[0]-1; // == 0
-    func[min(a, 0)] = foo;
+    func[min(a, 0.f)] = foo;
     RET[programIndex] = func[0](a, b);
 }
 
diff --git a/tests/goto-4.ispc b/tests/goto-4.ispc
index 97c373ef..9d5bf630 100644
--- a/tests/goto-4.ispc
+++ b/tests/goto-4.ispc
@@ -9,7 +9,7 @@ export void f_f(uniform float RET[], uniform float aFOO[]) {
  encore:
     ++RET[programIndex];
     if (any(a != 0)) {
-        a = max(a-1, 0);
+        a = max(a-1, 0.f);
         goto encore;
     }
 }
diff --git a/tests/max-float-1.ispc b/tests/max-float-1.ispc
index 24b9822d..5674a7bf 100644
--- a/tests/max-float-1.ispc
+++ b/tests/max-float-1.ispc
@@ -6,7 +6,7 @@ export uniform int width() { return programCount; }
 export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
     float a = aFOO[programIndex];
     RET[programIndex] = max(3 * a, 10.f);
-    RET[width()-1] = max(b, 100);
+    RET[width()-1] = max(b, 100.f);
 }
 
 
diff --git a/tests/max-float-2.ispc b/tests/max-float-2.ispc
index f990b102..16937b42 100644
--- a/tests/max-float-2.ispc
+++ b/tests/max-float-2.ispc
@@ -6,7 +6,7 @@ export uniform int width() { return programCount; }
 export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
     float a = aFOO[programIndex];
     RET[programIndex] = max(-10 * (a-3), .1f);
-    RET[width() - 1] = max(-10 * b, 2);
+    RET[width() - 1] = max(-10 * b, 2.f);
 }
 
 export void result(uniform float RET[]) {
diff --git a/tests/min-float-1.ispc b/tests/min-float-1.ispc
index 5b62c5c5..3a5ad5f3 100644
--- a/tests/min-float-1.ispc
+++ b/tests/min-float-1.ispc
@@ -6,7 +6,7 @@ export uniform int width() { return programCount; }
 export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
     float a = aFOO[programIndex];
     RET[programIndex] = min(3 * a, 10.f);
-    RET[width()-1] = min(b, 100);
+    RET[width()-1] = min(b, 100.f);
 }
 
 
diff --git a/tests/min-float-2.ispc b/tests/min-float-2.ispc
index 85c226ca..e099d64b 100644
--- a/tests/min-float-2.ispc
+++ b/tests/min-float-2.ispc
@@ -6,7 +6,7 @@ export uniform int width() { return programCount; }
 export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
     float a = aFOO[programIndex];
     RET[programIndex] = min(-10 * (a-3), .1f);
-    RET[width() - 1] = min(-10 * b, 2);
+    RET[width() - 1] = min(-10 * b, 2.f);
 }
 
 export void result(uniform float RET[]) {
diff --git a/tests/ref-overloads.ispc b/tests/ref-overloads.ispc
new file mode 100644
index 00000000..961af36d
--- /dev/null
+++ b/tests/ref-overloads.ispc
@@ -0,0 +1,93 @@
+/////////////////////////VARYING
+//int A011(varying float f){return 0;}  int A011(varying float& f){return 1;}       int A011(varying const float& f){return 2;}
+//int A011(varying double f){return 3;} int A011(varying int f){return 4;}          int A011(uniform float f){return 5;}
+//int A011(uniform float& f){return 6;} int A011(uniform const float& f){return 7;} int A011(uniform double f){return 8;}         int A011(uniform int f){return 9;}
+
+int A013(varying const float& f){return 2;} int A013(varying double f){return 3;}  int A013(varying int f){return 4;}
+int A013(uniform float f){return 5;} int A013(uniform float& f){return 6;}  int A013(uniform const float& f){return 7;}
+int A013(uniform double f){return 8;} int A013(uniform int f){return 9;}
+
+int A014(varying double f){return 3;}  int A014(varying int f){return 4;} 
+int A014(uniform float f){return 5;} int A014(uniform float& f){return 6;}  int A014(uniform const float& f){return 7;} 
+int A014(uniform double f){return 8;} int A014(uniform int f){return 9;}
+
+int A015(varying int f){return 4;} int A015(uniform float f){return 5;}
+int A015(uniform float& f){return 6;}  int A015(uniform const float& f){return 7;} 
+int A015(uniform double f){return 8;} int A015(uniform int f){return 9;}
+
+int A032(varying float& f){return 1;} int A032(varying double f){return 3;} int A032(varying int f){return 4;} 
+int A032(uniform float f){return 5;}  int A032(uniform float& f){return 6;} int A032(uniform const float& f){return 7;}
+int A032(uniform double f){return 8;} int A032(uniform int f){return 9;}
+
+int A033(varying int f){return 4;}
+int A033(uniform float f){return 5;}  int A033(uniform float& f){return 6;} int A033(uniform const float& f){return 7;}
+int A033(uniform double f){return 8;} int A033(uniform int f){return 9;}
+
+/////////////////////////UNIFORM
+
+int A051(varying float f){return 0;}        int A051(varying float& f){return 1;} int A051(varying const float& f){return 2;}
+int A051(varying double f){return 3;}       int A051(varying int f){return 4;}
+int A051(uniform const float& f){return 7;} int A051(uniform double f){return 8;} int A051(uniform int f){return 9;}
+
+int A052(varying float f){return 0;}        int A052(varying float& f){return 1;} int A052(varying const float& f){return 2;}
+int A052(varying double f){return 3;}       int A052(varying int f){return 4;}
+int A052(uniform double f){return 8;}       int A052(uniform int f){return 9;}
+
+int A053(varying float f){return 0;}        int A053(varying float& f){return 1;} int A053(varying const float& f){return 2;}
+int A053(varying double f){return 3;}       int A053(varying int f){return 4;}    int A053(uniform int f){return 9;} 
+
+int A054(varying float& f){return 1;}       int A054(varying const float& f){return 2;}
+int A054(varying double f){return 3;}       int A054(varying int f){return 4;}  int A054(uniform int f){return 9;}
+
+int A055(varying float& f){return 1;}       int A055(varying const float& f){return 2;}
+int A055(varying double f){return 3;}       int A055(varying int f){return 4;}
+
+int A056(varying float& f){return 1;}       int A056(varying const float& f){return 2;} int A056(varying int f){return 4;}
+
+
+export uniform int width() { return programCount; }
+
+
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying float vf = 5;
+    varying const float vcf = 5;
+    varying float& vfr = vf;
+    varying const float& vcfr = vf;
+    uniform float uf = 5;
+    uniform const float ucf = 5;
+    uniform float& ufr = uf;
+    uniform const float& ucfr = uf;
+
+    if (A013(vf) != 2) {RET[programIndex] = 1; return;}
+    if (A014(vf) != 3) {RET[programIndex] = 2; return;}
+    if (A015(vf) != 4) {RET[programIndex] = 3; return;}
+
+    if (A013(vfr) != 2) {RET[programIndex] = 4; return;}
+    if (A014(vfr) != 3) {RET[programIndex] = 5; return;}
+    if (A015(vfr) != 4) {RET[programIndex] = 6; return;}
+
+    if (A032(vcf) != 3) {RET[programIndex] = 7; return;}
+    if (A033(vcf) != 4) {RET[programIndex] = 8; return;}
+
+    if (A032(vcfr) != 3) {RET[programIndex] = 9; return;} 
+    if (A033(vcfr) != 4) {RET[programIndex] = 10; return;} 
+
+    if (A051(uf) != 7) {RET[programIndex] = 11; return;}
+    if (A052(uf) != 8) {RET[programIndex] = 12; return;}
+    if (A053(uf) != 0) {RET[programIndex] = 13; return;}
+    if (A054(uf) != 9) {RET[programIndex] = 14; return;}
+    if (A055(uf) != 3) {RET[programIndex] = 15; return;}
+    if (A056(uf) != 4) {RET[programIndex] = 16; return;}
+
+    if (A051(ufr) != 7) {RET[programIndex] = 17; return;}
+    if (A052(ufr) != 8) {RET[programIndex] = 18; return;}
+    if (A053(ufr) != 0) {RET[programIndex] = 19; return;}
+    if (A054(ufr)  != 9) {RET[programIndex] = 20; return;}
+    if (A055(ufr)  != 3) {RET[programIndex] = 21; return;}
+    if (A056(ufr)  != 4) {RET[programIndex] = 22; return;}
+
+   RET[programIndex] = 0;
+}
+
+export void result(uniform float RET[]) { RET[programIndex] = 0; }
diff --git a/tests/ref-overloads1.ispc b/tests/ref-overloads1.ispc
new file mode 100644
index 00000000..114eb011
--- /dev/null
+++ b/tests/ref-overloads1.ispc
@@ -0,0 +1,41 @@
+void A1(float f)               {}
+void A2(float& f)              {}
+void A3(const float& f)        {} 
+void A4(double f)              {}
+void A5(int f)                 {}
+void A6(uniform float f)       {}
+void A7(uniform float& f)      {}
+void A8(uniform const float& f){}
+void A9(uniform double f)      {}
+void A0(uniform int f)         {}
+
+
+
+export uniform int width() { return programCount; }
+
+
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying float vf = 5;
+    varying const float vcf = 5;
+    varying float& vfr = vf;
+    varying const float& vcfr = vf;
+    uniform float uf = 5;
+    uniform const float ucf = 5;
+    uniform float& ufr = uf;
+    uniform const float& ucfr = uf;
+
+    A1(vf); A1(vcf); A1(vfr); A1(vcfr); A1(uf); A1(ucf); A1(ufr); A1(ucfr);
+    A2(vf);          A2(vfr);
+    A3(vf); A3(vcf); A3(vfr); A3(vcfr);
+    A4(vf); A4(vcf); A4(vfr); A4(vcfr); A4(uf); A4(ucf); A4(ufr); A4(ucfr);
+    A5(vf); A5(vcf); A5(vfr); A5(vcfr); A5(uf); A5(ucf); A5(ufr); A5(ucfr);
+                                        A6(uf); A6(ucf); A6(ufr); A6(ucfr);
+                                        A7(uf);          A7(ufr);
+                                        A8(uf); A8(ucf); A8(ufr); A8(ucfr);
+                                        A9(uf); A9(ucf); A9(ufr); A9(ucfr);
+                                        A0(uf); A0(ucf); A0(ufr); A0(ucfr); 
+    RET[programIndex] = 1;
+}
+
+export void result(uniform float RET[]) { RET[programIndex] = 1; }
diff --git a/tests/ref-overloads2.ispc b/tests/ref-overloads2.ispc
new file mode 100644
index 00000000..a8a7152a
--- /dev/null
+++ b/tests/ref-overloads2.ispc
@@ -0,0 +1,54 @@
+void A01(varying float f){} void A01(varying float& f){} void A01(varying const float& f){} void A01(varying double f){} void A01(varying int f){}
+                            void A02(varying float& f){} void A02(varying const float& f){} void A02(varying double f){} void A02(varying int f){}
+void A03(varying float f){}                              void A03(varying const float& f){} void A03(varying double f){} void A03(varying int f){}
+void A04(varying float f){} void A04(varying float& f){}                                    void A04(varying double f){} void A04(varying int f){}
+void A05(varying float f){} void A05(varying float& f){} void A05(varying const float& f){}                              void A05(varying int f){}
+void A06(varying float f){} void A06(varying float& f){} void A06(varying const float& f){} void A06(varying double f){}
+
+void A07(varying float f){} void A07(varying float& f){} void A07(varying const float& f){} void A07(varying double f){} void A07(varying int f){}
+void A08(varying float f){} void A08(varying float& f){} void A08(varying const float& f){} void A08(varying double f){} void A08(varying int f){}
+void A09(varying float f){} void A09(varying float& f){} void A09(varying const float& f){} void A09(varying double f){} void A09(varying int f){}
+void A10(varying float f){} void A10(varying float& f){} void A10(varying const float& f){} void A10(varying double f){} void A10(varying int f){}
+void A11(varying float f){} void A11(varying float& f){} void A11(varying const float& f){} void A11(varying double f){} void A11(varying int f){}
+
+void A01(uniform float f){} void A01(uniform float& f){} void A01(uniform const float& f){} void A01(uniform double f){} void A01(uniform int f){}
+void A02(uniform float f){} void A02(uniform float& f){} void A02(uniform const float& f){} void A02(uniform double f){} void A02(uniform int f){}
+void A03(uniform float f){} void A03(uniform float& f){} void A03(uniform const float& f){} void A03(uniform double f){} void A03(uniform int f){}
+void A04(uniform float f){} void A04(uniform float& f){} void A04(uniform const float& f){} void A04(uniform double f){} void A04(uniform int f){}
+void A05(uniform float f){} void A05(uniform float& f){} void A05(uniform const float& f){} void A05(uniform double f){} void A05(uniform int f){}
+void A06(uniform float f){} void A06(uniform float& f){} void A06(uniform const float& f){} void A06(uniform double f){} void A06(uniform int f){}
+
+                            void A07(uniform float& f){} void A07(uniform const float& f){} void A07(uniform double f){} void A07(uniform int f){}
+void A08(uniform float f){}                              void A08(uniform const float& f){} void A08(uniform double f){} void A08(uniform int f){}
+void A09(uniform float f){} void A09(uniform float& f){}                                    void A09(uniform double f){} void A09(uniform int f){}
+void A10(uniform float f){} void A10(uniform float& f){} void A10(uniform const float& f){}                              void A10(uniform int f){}
+void A11(uniform float f){} void A11(uniform float& f){} void A11(uniform const float& f){} void A11(uniform double f){}
+
+export uniform int width() { return programCount; }
+
+
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying float vf = 5;
+    varying const float vcf = 5;
+    varying float& vfr = vf;
+    varying const float& vcfr = vf;
+    uniform float uf = 5;
+    uniform const float ucf = 5;
+    uniform float& ufr = uf;
+    uniform const float& ucfr = uf;
+
+    A01(vf); A01(vcf); A01(vfr); A01(vcfr); A01(uf); A01(ucf); A01(ufr); A01(ucfr);
+    A02(vf); A02(vcf); A02(vfr); A02(vcfr); A02(uf); A02(ucf); A02(ufr); A02(ucfr);
+    A03(vf); A03(vcf); A03(vfr); A03(vcfr); A03(uf); A03(ucf); A03(ufr); A03(ucfr);
+    A04(vf); A04(vcf); A04(vfr); A04(vcfr); A04(uf); A04(ucf); A04(ufr); A04(ucfr);
+    A05(vf); A05(vcf); A05(vfr); A05(vcfr); A05(uf); A05(ucf); A05(ufr); A05(ucfr);
+    A06(vf); A06(vcf); A06(vfr); A06(vcfr); A06(uf); A06(ucf); A06(ufr); A06(ucfr);
+    A07(vf); A07(vcf); A07(vfr); A07(vcfr); A07(uf); A07(ucf); A07(ufr); A07(ucfr);
+    A08(vf); A08(vcf); A08(vfr); A08(vcfr); A08(uf); A08(ucf); A08(ufr); A08(ucfr);
+    A09(vf); A09(vcf); A09(vfr); A09(vcfr); A09(uf); A09(ucf); A09(ufr); A09(ucfr);
+    A10(vf); A10(vcf); A10(vfr); A10(vcfr); A10(uf); A10(ucf); A10(ufr); A10(ucfr);
+    RET[programIndex] = 1;
+}
+
+export void result(uniform float RET[]) { RET[programIndex] = 1; }
diff --git a/tests/ref-overloads3.ispc b/tests/ref-overloads3.ispc
new file mode 100644
index 00000000..b9529dd5
--- /dev/null
+++ b/tests/ref-overloads3.ispc
@@ -0,0 +1,99 @@
+int A01(varying float f){return 0;} int A01(varying float& f){return 1;} int A01(varying const float& f){return 2;} int A01(varying double f){return 3;} int A01(varying int f){return 4;}
+                                    int A02(varying float& f){return 1;} int A02(varying const float& f){return 2;} int A02(varying double f){return 3;} int A02(varying int f){return 4;}
+int A03(varying float f){return 0;}                                      int A03(varying const float& f){return 2;} int A03(varying double f){return 3;} int A03(varying int f){return 4;}
+int A04(varying float f){return 0;} int A04(varying float& f){return 1;}                                            int A04(varying double f){return 3;} int A04(varying int f){return 4;}
+
+int A07(varying float f){return 0;} int A07(varying float& f){return 1;} int A07(varying const float& f){return 2;} int A07(varying double f){return 3;} int A07(varying int f){return 4;}
+int A08(varying float f){return 0;} int A08(varying float& f){return 1;} int A08(varying const float& f){return 2;} int A08(varying double f){return 3;} int A08(varying int f){return 4;}
+int A09(varying float f){return 0;} int A09(varying float& f){return 1;} int A09(varying const float& f){return 2;} int A09(varying double f){return 3;} int A09(varying int f){return 4;}
+
+int A01(uniform float f){return 5;} int A01(uniform float& f){return 6;} int A01(uniform const float& f){return 7;} int A01(uniform double f){return 8;} int A01(uniform int f){return 9;}
+int A02(uniform float f){return 5;} int A02(uniform float& f){return 6;} int A02(uniform const float& f){return 7;} int A02(uniform double f){return 8;} int A02(uniform int f){return 9;}
+int A03(uniform float f){return 5;} int A03(uniform float& f){return 6;} int A03(uniform const float& f){return 7;} int A03(uniform double f){return 8;} int A03(uniform int f){return 9;}
+int A04(uniform float f){return 5;} int A04(uniform float& f){return 6;} int A04(uniform const float& f){return 7;} int A04(uniform double f){return 8;} int A04(uniform int f){return 9;}
+
+                                    int A07(uniform float& f){return 6;} int A07(uniform const float& f){return 7;} int A07(uniform double f){return 8;} int A07(uniform int f){return 9;}
+int A08(uniform float f){return 5;}                                      int A08(uniform const float& f){return 7;} int A08(uniform double f){return 8;} int A08(uniform int f){return 9;}
+int A09(uniform float f){return 5;} int A09(uniform float& f){return 6;}                                            int A09(uniform double f){return 8;} int A09(uniform int f){return 9;}
+
+export uniform int width() { return programCount; }
+
+
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying float vf = 5;
+    varying const float vcf = 5;
+    varying float& vfr = vf;
+    varying const float& vcfr = vf;
+    uniform float uf = 5;
+    uniform const float ucf = 5;
+    uniform float& ufr = uf;
+    uniform const float& ucfr = uf;
+
+    if (A01(vf)  != 0) {RET[programIndex] = 1; return;}
+    if (A01(vcf) != 0) {RET[programIndex] = 2; return;}
+    if (A01(vfr) != 1) {RET[programIndex] = 3; return;}
+    if (A01(vcfr)!= 2) {RET[programIndex] = 4; return;}
+    if (A01(uf)  != 5) {RET[programIndex] = 5; return;}
+    if (A01(ucf) != 5) {RET[programIndex] = 6; return;}
+    if (A01(ufr) != 6) {RET[programIndex] = 7; return;}
+    if (A01(ucfr)!= 7) {RET[programIndex] = 8; return;}
+
+    if (A02(vf)  != 1) {RET[programIndex] = 9; return;}
+    if (A02(vcf) != 2) {RET[programIndex] = 10; return;}
+    if (A02(vfr) != 1) {RET[programIndex] = 11; return;}
+    if (A02(vcfr)!= 2) {RET[programIndex] = 12; return;}
+    if (A02(uf)  != 5) {RET[programIndex] = 13; return;}
+    if (A02(ucf) != 5) {RET[programIndex] = 14; return;}
+    if (A02(ufr) != 6) {RET[programIndex] = 15; return;}
+    if (A02(ucfr)!= 7) {RET[programIndex] = 16; return;}
+
+    if (A03(vf)  != 0) {RET[programIndex] = 17; return;}
+    if (A03(vcf) != 0) {RET[programIndex] = 18; return;}
+    if (A03(vfr) != 0) {RET[programIndex] = 19; return;}
+    if (A03(vcfr)!= 2) {RET[programIndex] = 20; return;}
+    if (A03(uf)  != 5) {RET[programIndex] = 21; return;}
+    if (A03(ucf) != 5) {RET[programIndex] = 22; return;}
+    if (A03(ufr) != 6) {RET[programIndex] = 23; return;}
+    if (A03(ucfr)!= 7) {RET[programIndex] = 24; return;}
+
+    if (A04(vf)  != 0) {RET[programIndex] = 25; return;}
+    if (A04(vcf) != 0) {RET[programIndex] = 26; return;}
+    if (A04(vfr) != 1) {RET[programIndex] = 27; return;}
+    if (A04(vcfr)!= 0) {RET[programIndex] = 28; return;}
+    if (A04(uf)  != 5) {RET[programIndex] = 29; return;}
+    if (A04(ucf) != 5) {RET[programIndex] = 30; return;}
+    if (A04(ufr) != 6) {RET[programIndex] = 31; return;}
+    if (A04(ucfr)!= 7) {RET[programIndex] = 32; return;}
+
+    if (A07(vf)  != 0) {RET[programIndex] = 33; return;}
+    if (A07(vcf) != 0) {RET[programIndex] = 34; return;}
+    if (A07(vfr) != 1) {RET[programIndex] = 35; return;}
+    if (A07(vcfr)!= 2) {RET[programIndex] = 36; return;}
+    if (A07(uf)  != 6) {RET[programIndex] = 37; return;}
+    if (A07(ucf) != 7) {RET[programIndex] = 38; return;}
+    if (A07(ufr) != 6) {RET[programIndex] = 39; return;}
+    if (A07(ucfr)!= 7) {RET[programIndex] = 40; return;}
+
+    if (A08(vf)  != 0) {RET[programIndex] = 41; return;}
+    if (A08(vcf) != 0) {RET[programIndex] = 42; return;}
+    if (A08(vfr) != 1) {RET[programIndex] = 43; return;}
+    if (A08(vcfr)!= 2) {RET[programIndex] = 44; return;}
+    if (A08(uf)  != 5) {RET[programIndex] = 45; return;}
+    if (A08(ucf) != 5) {RET[programIndex] = 46; return;}
+    if (A08(ufr) != 5) {RET[programIndex] = 47; return;}
+    if (A08(ucfr)!= 7) {RET[programIndex] = 48; return;}
+
+    if (A09(vf)  != 0) {RET[programIndex] = 49; return;}
+    if (A09(vcf) != 0) {RET[programIndex] = 50; return;}
+    if (A09(vfr) != 1) {RET[programIndex] = 51; return;}
+    if (A09(vcfr)!= 2) {RET[programIndex] = 52; return;}
+    if (A09(uf)  != 5) {RET[programIndex] = 53; return;}
+    if (A09(ucf) != 5) {RET[programIndex] = 54; return;}
+    if (A09(ufr) != 6) {RET[programIndex] = 55; return;}
+    if (A09(ucfr)!= 5) {RET[programIndex] = 56; return;}
+
+    RET[programIndex] = 0;
+}
+
+export void result(uniform float RET[]) { RET[programIndex] = 0; }
diff --git a/tests/typedef-2.ispc b/tests/typedef-2.ispc
index e8117dd4..af71dc6e 100644
--- a/tests/typedef-2.ispc
+++ b/tests/typedef-2.ispc
@@ -19,7 +19,7 @@ export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
     for (uniform int i = 0; i < 16; ++i)
         for (uniform int j = 0; j < 16; ++j)
             bar.foo[i].x[j] = b;
-    RET[programIndex] = bar.foo[min(15, a-1)].x[min(15, a-1)];
+    RET[programIndex] = bar.foo[min(15.f, a-1)].x[min(15.f, a-1)];
 }
 
 export void result(uniform float RET[]) { RET[programIndex] = 5; }
diff --git a/type.cpp b/type.cpp
index b22f9283..c7b59592 100644
--- a/type.cpp
+++ b/type.cpp
@@ -1909,7 +1909,15 @@ StructType::StructType(const std::string &n, const llvm::SmallVector<const Type
 
 const std::string 
 StructType::GetCStructName() const {
-  return lMangleStructName(name, variability);
+  // only return mangled name for varying structs for backwards
+  // compatibility...
+
+  if (variability == Variability::Varying) {
+    return lMangleStructName(name, variability);
+  }
+  else {
+    return GetStructName();
+  }
 }
 
 Variability