Update sycl/doc/extensions/experimental/sycl_ext_oneapi_device_global.asciidoc

Reword spec note to generalize beyond dpc++ compiler

Co-authored-by: John Pennycook <[email protected]>

Author: JackAKirk

sycl/doc/extensions/experimental/sycl_ext_oneapi_device_global.asciidoc

@@ -199,7 +199,7 @@ _T_ is restricted to types that have a trivial destructor. _PropertyListT_ enabl
 
 [NOTE]
 ====
-If _T_ is prepended `const` then in the cuda and hip backends the compiler will use the constant address space for the `device_global` instead of the global address space. This allows for optimizations such as using the Nvidia constant cache, and is equivalent to usage of `$$__constant__$$` in the CUDA and HIP native programming languages.
+If _T_ is `const` then implementations may choose to allocate the `device_global` in a dedicated constant address space as an optimization. When using the {dpcpp} compiler with the CUDA or HIP backend, declaring a `device_global<const T>` is equivalent to declaring a `$$__constant__$$` variable.
 ====
 
 When compiling with {cpp} versions before {cpp}20, _T_ must also have a trivial default constructor. In this case, the allocation of type _T_ for a given `device_global` is zero-initialized on a given device prior to the first access to that `device_global` on that device. For the purposes of this definition an access can be a direct access of the `device_global` in kernel code or a copy to or from that `device_global` enqueued to the given device.