review fixes: portable sizeof gate, explicit plan-move, alignment intent

- arena_sizeof_invariant_suite: drop platform-specific absolute baselines
  (328/16/248 were Apple-arm64/libc++ only); keep relative
  ImplLayoutAllocator == ImplLayoutMaster invariant + monostate static_asserts.
- cont_engine: reset arena_plan_ after std::move into op_ so later reads
  see "no plan" rather than a moved-from optional.
- arena_kernels: one-line intent note on trivial kernels' tight packing.

Author: zhihao-deng

src/TiledArray/expressions/cont_engine.h

@@ -306,6 +306,11 @@ class ContEngine : public BinaryEngine<Derived> {
                       outer_size(left_indices_), outer_size(right_indices_),
                       total_perm, this->element_nonreturn_op_,
                       std::move(this->arena_plan_));
+        // Plan ownership transferred to op_; mark carrier slot empty so any
+        // later use of arena_plan_ reads as "no plan" rather than moved-from.
+        if constexpr (!std::is_same_v<arena_plan_storage_t, std::monostate>) {
+          this->arena_plan_.reset();
+        }
       }
       trange_ = ContEngine_::make_trange(outer_perm);
       shape_ = ContEngine_::make_shape(outer_perm);
@@ -337,6 +342,11 @@ class ContEngine : public BinaryEngine<Derived> {
                       outer_size(left_indices_), outer_size(right_indices_),
                       total_perm, this->element_nonreturn_op_,
                       std::move(this->arena_plan_));
+        // Plan ownership transferred to op_; mark carrier slot empty so any
+        // later use of arena_plan_ reads as "no plan" rather than moved-from.
+        if constexpr (!std::is_same_v<arena_plan_storage_t, std::monostate>) {
+          this->arena_plan_.reset();
+        }
       }
       trange_ = ContEngine_::make_trange();
       shape_ = ContEngine_::make_shape();

src/TiledArray/tensor/arena_kernels.h

@@ -47,6 +47,7 @@ OuterTensor arena_trivial_unary(const SrcOuterTensor& src, FillOp&& fill_op) {
   auto shape_fn = [&src](std::size_t ord) -> decltype(auto) {
     return src.data()[ord].range();
   };
+  // Elementwise kernels: pack tight (no cross-cell GEMM to amortize 128B pad).
   ArenaPlan p = plan(N_cells, shape_fn, sizeof(elem_t), alignof(elem_t));
   auto arena = std::make_shared<Arena>();
   if (p.total_bytes > 0) arena->reserve(p.total_bytes, false);
@@ -80,6 +81,7 @@ OuterTensor arena_trivial_binary(const LeftTensor& left, const RightTensor& righ
   auto shape_fn = [&left](std::size_t ord) -> decltype(auto) {
     return left.data()[ord].range();
   };
+  // Elementwise kernels: pack tight (no cross-cell GEMM to amortize 128B pad).
   ArenaPlan p = plan(N_cells, shape_fn, sizeof(elem_t), alignof(elem_t));
   auto arena = std::make_shared<Arena>();
   if (p.total_bytes > 0) arena->reserve(p.total_bytes, false);

tests/arena_sizeof_invariant_suite.cpp

@@ -2,7 +2,6 @@
 
 #include "TiledArray/tensor.h"
 #include "TiledArray/tensor/arena_einsum.h"
-#include "TiledArray/tile_op/contract_reduce.h"
 #include "TiledArray/util/function.h"
 #include "tiledarray.h"
 #include "unit_test_config.h"
@@ -20,10 +19,6 @@ using PlainLeft = TA::Tensor<double>;
 using PlainRight = TA::Tensor<double>;
 using PlainScalar = double;
 
-using PlainContractReduceBase =
-    TA::detail::ContractReduceBase<PlainResult, PlainLeft, PlainRight,
-                                   PlainScalar>;
-
 using PlainArenaPlanStorage =
     TA::detail::arena_plan_storage_t<PlainResult, PlainLeft, PlainRight>;
 
@@ -47,11 +42,6 @@ struct ImplLayoutAllocator {
   TA_NO_UNIQUE_ADDRESS PlainArenaPlanStorage arena_plan_;
 };
 
-/// Sizes captured against master; re-baseline if the toolchain changes.
-constexpr std::size_t kTensorDoubleSizeMaster = 328;
-constexpr std::size_t kContractReduceBaseSizeMaster = 16;
-constexpr std::size_t kImplLayoutSizeMaster = 248;
-
 static_assert(std::is_same_v<PlainArenaPlanStorage, std::monostate>,
               "plain-tensor arena_plan_storage_t must be std::monostate");
 
@@ -62,18 +52,7 @@ static_assert(sizeof(ImplLayoutAllocator) == sizeof(ImplLayoutMaster),
 
 BOOST_AUTO_TEST_SUITE(arena_sizeof_invariant_suite, TA_UT_LABEL_SERIAL)
 
-BOOST_AUTO_TEST_CASE(tensor_double_sizeof_matches_master) {
-  BOOST_CHECK_EQUAL(sizeof(TA::Tensor<double>), kTensorDoubleSizeMaster);
-}
-
-BOOST_AUTO_TEST_CASE(contract_reduce_base_sizeof_matches_master) {
-  BOOST_CHECK_EQUAL(sizeof(PlainContractReduceBase),
-                    kContractReduceBaseSizeMaster);
-}
-
 BOOST_AUTO_TEST_CASE(impl_layout_no_unique_address_invariant) {
-  BOOST_CHECK_EQUAL(sizeof(ImplLayoutMaster), kImplLayoutSizeMaster);
-  BOOST_CHECK_EQUAL(sizeof(ImplLayoutAllocator), kImplLayoutSizeMaster);
   BOOST_CHECK_EQUAL(sizeof(ImplLayoutAllocator), sizeof(ImplLayoutMaster));
 }