Turn off optimizations on emitted external function. (#1592)

Clang chooses to use a __vectorcall optimization in which
address pointers are vector loaded in the gpu::invoke_primitive.
This results in a segfault when stack alignment is absent.
Since the GPU transformer does not rely on CPU for compute,
we disable the optimizations of the emitted function.

src/ngraph/runtime/gpu/gpu_external_function.cpp

@@ -497,7 +497,7 @@ void runtime::gpu::GPU_ExternalFunction::emit_functions()
 
         m_writer << "extern \"C\" void " << current_function->get_name();
         m_writer << "(void** inputs, void** outputs, "
-                 << "gpu::GPURuntimeContext* ctx)\n";
+                 << "gpu::GPURuntimeContext* ctx) __attribute__ ((optnone))\n";
         m_writer.block_begin();
         {
             m_writer << "m_runtime_context = ctx;\n";