Reland https://github.com/intel/intel-xpu-backend-for-triton/commit/15c5e5519ff129104956ef0a726656f9e5cb4359 that is temporarily reverted in https://github.com/intel/intel-xpu-backend-for-triton/pull/2595 by https://github.com/intel/intel-xpu-backend-for-triton/commit/77f98f00658988d59626b50bf946159b3a3ad5f3.