llvm

95ba470e - [SYCL] Don't fetch device image and property for small ranges (#22488)

[SYCL] Don't fetch device image and property for small ranges (#22488) Fixes CMPLRLLVM-75858 This PR fixes performance regression after https://github.com/intel/llvm/pull/21979 in kernel submission. The regression was because of us fetching device images to get `idQueriesRange` device image property. However, if a kernel is submitted with global range < `INT_MAX` which is a very common case, we can skip fetching device image and kernel property. This PR improves kernel submission time by ~6% (on PVC) for PyTorch benchmarks. --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>