The point of memory_order_consume is to be able to omit the load barrier on architectures with weak memory orderings that don't need them for data-dependent ope

The point of memory_order_consume is to be able to omit the load barrier on architectures with weak memory orderings that don't need them for data-dependent operations on the load. This is basically every architecture other than Alpha (which would still need the barrier with memory_order_consume) and x86 and SPARC (which don't need the barrier with memory_order_acquire). ARM and PPC, for example, fall into this category.