Deprecate context interop for CUDA and HIP

Author: Hugh Delaney

sycl/include/sycl/backend.hpp

@@ -213,6 +213,74 @@ get_native<backend::ext_oneapi_cuda, device>(const device &Obj) {
   return static_cast<backend_return_t<backend::ext_oneapi_cuda, device>>(
       Obj.getNative());
 }
+
+template <>
+__SYCL_DEPRECATED(
+    "Context interop is deprecated for CUDA. If a native context is required,"
+    " use cuDevicePrimaryCtxRetain with a native device")
+inline backend_return_t<backend::ext_oneapi_cuda, context> get_native<
+    backend::ext_oneapi_cuda, context>(const context &Obj) {
+  if (Obj.get_backend() != backend::ext_oneapi_cuda) {
+    throw sycl::exception(make_error_code(errc::backend_mismatch),
+                          "Backends mismatch");
+  }
+  return reinterpret_cast<backend_return_t<backend::ext_oneapi_cuda, context>>(
+      Obj.getNative());
+}
+
+template <>
+__SYCL_DEPRECATED("Context interop is deprecated for CUDA. If a native context"
+                  " is required, use cuDevicePrimaryCtxRetain with a native"
+                  " device")
+inline backend_return_t<backend::ext_oneapi_cuda, context> interop_handle::
+    get_native_context<backend::ext_oneapi_cuda>() const {
+  if (get_backend() != backend::ext_oneapi_cuda) {
+    throw sycl::runtime_error(errc::backend_mismatch, "Backends mismatch",
+                              PI_ERROR_INVALID_OPERATION);
+  }
+#ifndef __SYCL_DEVICE_ONLY__
+  return reinterpret_cast<backend_return_t<backend::ext_oneapi_cuda, context>>(
+      getNativeContext());
+#else
+  return 0;
+#endif
+}
+
+#endif
+#if SYCL_EXT_ONEAPI_BACKEND_HIP
+
+template <>
+__SYCL_DEPRECATED(
+    "Context interop is deprecated for HIP. If a native context is required,"
+    " use hipDevicePrimaryCtxRetain with a native device")
+inline backend_return_t<backend::ext_oneapi_hip, context> get_native<
+    backend::ext_oneapi_hip, context>(const context &Obj) {
+  if (Obj.get_backend() != backend::ext_oneapi_hip) {
+    throw sycl::exception(make_error_code(errc::backend_mismatch),
+                          "Backends mismatch");
+  }
+  return reinterpret_cast<backend_return_t<backend::ext_oneapi_hip, context>>(
+      Obj.getNative());
+}
+
+template <>
+__SYCL_DEPRECATED("Context interop is deprecated for HIP. If a native context"
+                  " is required, use hipDevicePrimaryCtxRetain with a native"
+                  " device")
+inline backend_return_t<backend::ext_oneapi_hip, context> interop_handle::
+    get_native_context<backend::ext_oneapi_hip>() const {
+  if (get_backend() != backend::ext_oneapi_hip) {
+    throw sycl::runtime_error(errc::backend_mismatch, "Backends mismatch",
+                              PI_ERROR_INVALID_OPERATION);
+  }
+#ifndef __SYCL_DEVICE_ONLY__
+  return reinterpret_cast<backend_return_t<backend::ext_oneapi_hip, context>>(
+      getNativeContext());
+#else
+  return 0;
+#endif
+}
+
 #endif
 
 template <backend BackendName, typename DataT, int Dimensions,

sycl/include/sycl/ext/oneapi/experimental/backend/cuda.hpp

@@ -35,6 +35,9 @@ inline __SYCL_EXPORT bool has_native_event(event sycl_event) {
 
 // CUDA context specialization
 template <>
+__SYCL_DEPRECATED(
+    "Context interop is deprecated for CUDA. If a native context is required,"
+    " use cuDevicePrimaryCtxRetain with a native device")
 inline auto get_native<backend::ext_oneapi_cuda, context>(const context &C)
     -> backend_return_t<backend::ext_oneapi_cuda, context> {
   // create a vector to be returned
@@ -51,8 +54,11 @@ inline auto get_native<backend::ext_oneapi_cuda, context>(const context &C)
 
 // Specialisation of interop_handles get_native_context
 template <>
-inline backend_return_t<backend::ext_oneapi_cuda, context>
-interop_handle::get_native_context<backend::ext_oneapi_cuda>() const {
+__SYCL_DEPRECATED(
+    "Context interop is deprecated for CUDA. If a native context is required,"
+    " use cuDevicePrimaryCtxRetain with a native device")
+inline backend_return_t<backend::ext_oneapi_cuda, context> interop_handle::
+    get_native_context<backend::ext_oneapi_cuda>() const {
 #ifndef __SYCL_DEVICE_ONLY__
   return std::vector{reinterpret_cast<CUcontext>(getNativeContext())};
 #else