Fix compilation with libc++ (ROCm/HIP part)
--- a/include/hip/hip_runtime.h
+++ b/include/hip/hip_runtime.h
@@ -49,10 +49,6 @@ THE SOFTWARE.
 #include <stdio.h>
 #include <stdlib.h>
 #include <assert.h>
-
-#if __cplusplus > 199711L
-#include <thread>
-#endif
 #endif // !defined(__HIPCC_RTC__)
 
 #include <hip/hip_version.h>
@@ -67,6 +63,12 @@ THE SOFTWARE.
 #endif
 
 #if !defined(__HIPCC_RTC__)
+// libc++ transitively includes thread->string_view->cuda_wrappers/algorithm,
+// which uses __host__, which is included via hip runtime above
+#if __cplusplus > 199711L
+#include <thread>
+#endif
+
 #include <hip/hip_runtime_api.h>
 #include <hip/library_types.h>
 #endif // !defined(__HIPCC_RTC__)