b9060

The latest release of llama.cpp (b9060) brings significant SYCL backend enhancements, adding six new operations: FILL, CUMSUM, DIAG, SOLVE_TRI, SSM_SCAN, and GATED_DELTA_NET. These additions allow users to leverage Intel GPUs (and other SYCL-compatible hardware) for more advanced model operations, including state-space model operations (SSM_SCAN) and gated delta networks (GATED_DELTA_NET). The release also includes fixes for the abort crash during test-backend-ops and regenerates the ops.md documentation. Developers contributed from Intel, with commits signed using GPG keys.

The project maintains its massive community adoption (109k stars, 17.9k forks) and provides pre-built binaries for macOS (Apple Silicon with optional KleidiAI, Intel, iOS XCFramework), Linux (x64, arm64, s390x, Vulkan, ROCm 7.2, OpenVINO, SYCL FP32/FP16), Android arm64, and Windows (CPU, CUDA 12/13, Vulkan, SYCL, HIP). The SYCL additions are particularly important for users on Intel hardware, enabling them to run a broader set of operations without relying on CUDA or ROCm. This reduces vendor lock-in and improves portability across heterogeneous computing environments.