#include #include namespace c10 { namespace cuda { const char* get_cuda_check_suffix() noexcept { static char* device_blocking_flag = getenv("CUDA_LAUNCH_BLOCKING"); static bool blocking_enabled = (device_blocking_flag && atoi(device_blocking_flag)); if (blocking_enabled) { return ""; } else { return "\nCUDA kernel errors might be asynchronously reported at some" " other API call,so the stacktrace below might be incorrect." "\nFor debugging consider passing CUDA_LAUNCH_BLOCKING=1."; } } } // namespace cuda } // namespace c10