Fix three bugs:

1) Insufficient memory allocation for scatter_matrix packing (needed to round up to MR/NR).
2) Calling destructor on the wrong pointers in indexed_(dpd_)?varray. Currently these are only used with scalar elements so no real error, but would be a problem if non-trivial elements were used.
3) Modify/capture race condition in indexed_dpd/mult.cxx.

src/external/marray/include/indexed_dpd_varray.hpp

@@ -206,9 +206,9 @@ class indexed_dpd_varray : public indexed_dpd_varray_base<Type, indexed_dpd_varr
         {
             if (storage_.size > 0)
             {
-                for (stride_type i = storage_.size/dense_size();i --> 0;)
+                for (stride_type i = storage_.size;i --> 0;)
                 {
-                    alloc_traits::destroy(storage_, data_[i]);
+                    alloc_traits::destroy(storage_, data_[0]+i);
                 }
                 alloc_traits::deallocate(storage_, data_[0], storage_.size);
                 storage_.size = 0;

src/external/marray/include/indexed_varray.hpp

@@ -142,7 +142,7 @@ class indexed_varray : public indexed_varray_base<Type, indexed_varray<Type, All
             {
                 for (stride_type i = storage_.size;i --> 0;)
                 {
-                    alloc_traits::destroy(storage_, data_[i]);
+                    alloc_traits::destroy(storage_, data_[0]+i);
                 }
                 alloc_traits::deallocate(storage_, data_[0], storage_.size);
                 storage_.size = 0;

src/internal/3t/indexed_dpd/mult.cxx

@@ -941,7 +941,7 @@ void mult_block(const communicator& comm, const config& cfg,
                 {
                     if (indices_C[idx_C].factor == T(0)) return;
 
-                    tasks.visit(idx++,
+                    tasks.visit(idx,
                     [&,idx,idx_A,idx_B,idx_C,next_A,next_B]
                     (const communicator& subcomm)
                     {
@@ -982,6 +982,8 @@ void mult_block(const communicator& comm, const config& cfg,
                             TensorGEMM{}(subcomm, cfg, factor, at, bt, T(1), ct);
                         });
                     });
+
+                    idx++;
                 });
             });
         });

src/nodes/matrify.hpp

@@ -85,8 +85,8 @@ allocate_buffers(len_type MB, len_type NB, Matrify& parent, Child& child,
 
     if (!child.pack_ptr)
     {
-        len_type m = A.length(0);
-        len_type n = A.length(1);
+        len_type m = A.length(0) + (MB-1);
+        len_type n = A.length(1) + (NB-1);
 
         if (comm.master())
         {