diff --git a/proto-cuda/nvrtc/worker.cpp b/proto-cuda/nvrtc/worker.cpp
index 08dab887c..10eaa476b 100644
--- a/proto-cuda/nvrtc/worker.cpp
+++ b/proto-cuda/nvrtc/worker.cpp
@@ -328,8 +328,10 @@ static bool rtcCompile(Ctx& c, const std::string& src, const char* name, const s
         if (r != NVRTC_SUCCESS) { err = "nvrtcAddNameExpression " + e + ": " + c.rtc.getErrorString(r); c.rtc.destroyProgram(&prog); return false; }
     }
     std::string archOpt = "--gpu-architecture=" + c.archOpt;
-    const char* opts[2] = { archOpt.c_str(), "--std=c++17" };
-    r = c.rtc.compileProgram(prog, 2, opts);
+    // -default-device: NVRTC rejects unannotated functions as host code (nvcc treats them as host and discards them);
+    // the pack headers (program.h, memhard.h) carry plain inline helpers, so every unannotated function is device code here.
+    const char* opts[3] = { archOpt.c_str(), "--std=c++17", "-default-device" };
+    r = c.rtc.compileProgram(prog, 3, opts);
     {
         size_t logSize = 0;
         if (c.rtc.getProgramLogSize(prog, &logSize) == NVRTC_SUCCESS && logSize > 1) {