Use "autoEmplace" for performance

std/algorithm.d

@@ -794,7 +794,7 @@ template reduce(fun...) if (fun.length >= 1)
                         result = void;
                     foreach (i, T; result.Types)
                     {
-                        emplace(&result[i], seed);
+                        autoEmplace(result[i], seed);
                     }
                     r.popFront();
                     return reduce(result, r);
@@ -856,7 +856,7 @@ template reduce(fun...) if (fun.length >= 1)
 
                     foreach (i, T; result.Types)
                     {
-                        emplace(&result[i], elem);
+                        autoEmplace(result[i], elem);
                     }
                 }
             }
@@ -1186,7 +1186,7 @@ void uninitializedFill(Range, Value)(Range range, Value filler)
     {
         // Must construct stuff by the book
         for (; !range.empty; range.popFront())
-            emplace(addressOf(range.front), filler);
+            autoEmplace(range.front, filler);
     }
     else
         // Doesn't matter whether fill is initialized or not

std/array.d

@@ -30,22 +30,19 @@ if (isIterable!Range && !isNarrowString!Range && !isInfinite!Range)
     alias ForeachType!Range E;
     static if (hasLength!Range)
     {
-        if(r.length == 0) return null;
+        immutable len = r.length;
+        if(len == 0) return null;
 
         static auto trustedAllocateArray(size_t n) @trusted nothrow
         {
             return uninitializedArray!(Unqual!E[])(n);
         }
-        auto result = trustedAllocateArray(r.length);
+        auto result = trustedAllocateArray(len);
 
         size_t i;
-        static auto trustedGetAddr(T)(ref T t) @trusted nothrow pure
-        {
-            return &t;
-        }
         foreach (e; r)
         {
-            emplace(trustedGetAddr(result[i]), e);
+            autoEmplace(result[i], e);
             ++i;
         }
         return cast(E[])result;
@@ -913,17 +910,16 @@ void insertInPlace(T, U...)(ref T[] array, size_t pos, U stuff)
         }
         array.length += to_insert;
         copyBackwards(array[pos..oldLen], array[pos+to_insert..$]);
-        auto ptr = array.ptr + pos;
         foreach (i, E; U)
         {
             static if (is(E : T)) //ditto
             {
-                emplace(ptr++, stuff[i]);
+                autoEmplace(array[pos++], stuff[i]);
             }
             else
             {
                 foreach (v; stuff[i])
-                    emplace(ptr++, v);
+                    autoEmplace(array[pos++], v);
             }
         }
     }

std/conv.d

@@ -4833,6 +4833,55 @@ unittest
     assert(equal(map!(to!int)(["42", "34", "345"]), [42, 34, 345]));
 }
 
+/+
+package helper: This function is functionally equivalent to the normal
+emplace. However, it only calls the actual emplace if really necessary.
+It's dedicated to improving both code readability and performance.
+This function is mostly a workaround for DMD's bad inliner.
+
+Doing it as a dedicated function helps reduce error-prone boilerplate
+code in caller function. Indeed: While the normal emplace will still
+correctly call '=' when needed, it takes an argument by pointer. DMD
+seems to have a lot of trouble making the inline assignment when there
+is an un-necessary indirection.
+
+eg:
+----
+int i;
+emplace(&i, 5);
+vs
+autoEmplace(i, 5);
+----
+
+Using $(D emplace) over raw assignement or $(D autoEmplace) for trivial
+types in a tight loop $(I can) result in a $(I catastrophic) performance
+penalty.
++/
+package void autoEmplace(T, Args...)(ref T chunk, auto ref Args args)
+{
+    static if (Args.length == 0 && isAssignable!(T) && !hasElaborateAssign!T)
+        chunk = T.init;
+    else static if (Args.length == 1 && isAssignable!(T, Args[0]) && !hasElaborateAssign!T)
+        chunk = args[0];
+    else
+        emplace(&chunk, args);
+}
+
+unittest
+{
+    int i;
+    autoEmplace(i);
+    autoEmplace(i, 5);
+
+    static struct S
+    {
+        void opAssign(S);
+    }
+    S s;
+    autoEmplace(s);
+    autoEmplace(s, S.init);
+}
+
 // Undocumented for the time being
 void toTextRange(T, W)(T value, W writer)
     if (isIntegral!T && isOutputRange!(W, char))

std/functional.d

@@ -368,7 +368,7 @@ template adjoin(F...) if (F.length)
             Tuple!(Head, typeof(.adjoin!(F[1..$])(a)).Types) result = void;
             foreach (i, Unused; result.Types)
             {
-                emplace(&result[i], F[i](a));
+                autoEmplace(result[i], F[i](a));
             }
             return result;
         }

std/typecons.d

@@ -3522,7 +3522,7 @@ if (!is(T == class))
             _store = cast(Impl*) enforce(malloc(Impl.sizeof));
             static if (hasIndirections!T)
                 GC.addRange(&_store._payload, T.sizeof);
-            emplace(&_store._payload, args);
+            autoEmplace(_store._payload, args);
             _store._count = 1;
         }