* tuning para, * testing on v100 * add fp16 * remove deprecated tensor descriptor * sync with miopen * update build script Co-authored-by: Jing Zhang <jizhan@amd.com>