https://github.com/michalpaszkowski approved this pull request. Thank you for the patch and patience! LGTM! Confirming also that there are no OpenCL CTS regressions and no significant changes in the performance. https://github.com/llvm/llvm-project/pull/82786