Reland 9b750186115b04267de6bc10d38476557bad0a53 that is temporarily reverted in https://github.com/intel/intel-xpu-backend-for-triton/pull/5522 by bccf48a0e00cc43343e89f3dd950580bc800acb5. CI: https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/19547076240/job/55968560616