Execute native Windows K DLSS and validate real transformer weights

tools/windows/probe_kernel.hip

@@ -1,4 +1,13 @@
 // No host SDK, CRT, or device library dependency. Compiled only for gfx1201.
+extern "C" __attribute__((global)) void d4r_ptx_pattern(
+    unsigned int* output, unsigned int count, unsigned int seed)
+{
+    const unsigned int i = __builtin_amdgcn_workgroup_id_x() * 128u +
+                           __builtin_amdgcn_workitem_id_x();
+    if (i < count)
+        output[i] = (i * 1664525u + seed) ^ 0xa5a55a5au;
+}
+
 extern "C" __attribute__((global)) void d4r_probe_pattern(
     unsigned int* output, unsigned int count, unsigned int seed)
 {