Refactor the __device__ versions of memset and memcpy to be less awkward i.e. not return nullptr as opposed to the destination pointer (it can only be assumed it was done for maximum confusion) and actually unroll as they claim to. Change all of the {to, from}Symbol functions to use hipModuleGetGlobal, as opposed to hc::accelerator::get_symbol_address which is no longer valid with module based dispatch.

include/hip/hcc_detail/program_state.hpp

@@ -35,6 +35,24 @@ THE SOFTWARE.
 struct ihipModuleSymbol_t;
 using hipFunction_t = ihipModuleSymbol_t*;
 
+namespace std
+{
+    template<>
+    struct hash<hsa_agent_t> {
+        size_t operator()(hsa_agent_t x) const
+        {
+            return hash<decltype(x.handle)>{}(x.handle);
+        }
+    };
+}
+
+inline
+constexpr
+bool operator==(hsa_agent_t x, hsa_agent_t y)
+{
+    return x.handle == y.handle;
+}
+
 namespace hip_impl
 {
     struct Kernel_descriptor {
@@ -50,6 +68,8 @@ namespace hip_impl
         }
     };
 
+    const std::unordered_map<
+        hsa_agent_t, std::vector<hsa_executable_t>>& executables();
     const std::unordered_map<
         std::uintptr_t,
         std::vector<std::pair<hsa_agent_t, Kernel_descriptor>>>& functions();