llama.cpp b11121 Adds SYCL Support for get_rows_back Operation
llama.cpp released automated build b11121, introducing SYCL backend support for the get_rows_back tensor operation. This update enables the operation specifically for 32-bit floating-point (FP32) and 16-bit floating-point (FP16) data precision. This change improves operator compatibility for hardware running on the SYCL ecosystem, such as Intel GPUs, bringing its backend features closer to parity with other execution engines. It helps ensure smooth LLM inference across a broader range of heterogeneous computing hardware. The pull request (#25266) restricts initial get_rows_back support in the SYCL backend strictly to FP32 and FP16 formats, with updated documentation in ops.md. Pre-built binaries were distributed alongside this release for macOS, Linux, Windows, Android, and Snapdragon platforms across multiple backends.
## BACKGROUND
llama.cpp is a high-performance C/C++ library designed for executing large language models locally across various hardware platforms. SYCL is an open, cross-platform C++ programming model maintained by the Khronos Group that enables code to run efficiently on heterogeneous hardware architectures like GPUs and accelerators.