Towards Building Error Resilient GPGPU Applications