Unlike GCC, clang could optimize away alloca() and thus the
ALLOCATE_MEM_ON_STACK() instrumentation. To make it harder, let us
invoke a non-inline function on the entire allocated buffer.