Update on "[PyTorch] RFC: Add tuple inline storage"

I noticed a bunch of time being spent heap-allocating Tuples in the unpickler. 1-, 2-, and 3-element Tuples are apparently common enough that they get their own bytecode instructions, so I decided to try also giving them their own representation. We store up to 3 IValues inline in `Tuple` rather than doing a second heap allocation for a `std::vector<IValue>`. Differential Revision: [D30592622](https://our.internmc.facebook.com/intern/diff/D30592622/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D30592622/)! [ghstack-poisoned]
pytorch · swolchok · Aug 26, 2021 · Aug 27, 2021 · Aug 27, 2021 · Aug 27, 2021
commit 8496eb9a4499bbd62c019627fef30c2343a5f289
diff --git a/aten/src/ATen/core/ivalue_inl.h b/aten/src/ATen/core/ivalue_inl.h
@@ -262,6 +262,10 @@ struct TORCH_API TupleElements {
   // space in our use case.
   union {
     std::vector<IValue> elementsVector_;
+    // Don't want to declare a std::array because the convenient
+    // iteration and size members are a footgun in this case -- the
+    // actual size of the array may be smaller than 3!
+    // NOLINTNEXTLINE(modernize-avoid-c-arrays)
     IValue elementsInline_[3];
   };