stdlib: add a shortcut for Array.append(contentsOf:) in case the argument is an Array, too.

stdlib/public/core/Array.swift

@@ -1220,7 +1220,18 @@ extension Array: RangeReplaceableCollection {
       _buffer.count += writtenCount
     }
 
-    if writtenUpTo == buf.endIndex {
+    if _slowPath(writtenUpTo == buf.endIndex) {
+
+      // A shortcut for appending an Array: If newElements is an Array then it's
+      // guaranteed that buf.initialize(from: newElements) already appended all
+      // elements. It reduces code size, because the following code
+      // can be removed by the optimizer by constant folding this check in a
+      // generic specialization.
+      if newElements is [Element] {
+        _internalInvariant(remainder.next() == nil)
+        return
+      }
+
       // there may be elements that didn't fit in the existing buffer,
       // append them in slow sequence-only mode
       var newCount = _getCount()

test/SILOptimizer/array_contentof_opt.swift

@@ -54,3 +54,20 @@ public func testString(_ a: inout [String], s: String) {
 public func dontPropagateContiguousArray(_ a: inout ContiguousArray<UInt8>) {
   a += [4]
 }
+
+// Check if the specialized Array.append<A>(contentsOf:) is reasonably optimized for Array<Int>.
+
+// CHECK-LABEL: sil shared {{.*}}@$sSa6append10contentsOfyqd__n_t7ElementQyd__RszSTRd__lFSi_SaySiGTg5
+
+// There should only be a single call to _createNewBuffer or reserveCapacityForAppend/reserveCapacityImpl.
+
+// CHECK-NOT: apply
+// CHECK: [[F:%[0-9]+]] = function_ref @{{.*(_createNewBuffer|reserveCapacity).*}}
+// CHECK-NEXT: apply [[F]]
+// CHECK-NOT: apply
+
+// The number of basic blocks should not exceed 20 (ideally there are no more than 16 blocks in this function).
+// CHECK-NOT: bb20:
+
+// CHECK: } // end sil function '$sSa6append10contentsOfyqd__n_t7ElementQyd__RszSTRd__lFSi_SaySiGTg5
+