Cekirdekler
Cekirdekler is a C multi-device GPGPU compute API for OpenCL kernels. It lets developers treat multipleصول accelerators as a single device through a shared-distributed memory model, simplifying GPGPU programming while exposing detailed device control when needed. Key features: - Implicit multi-device support for CPUs and GPUs, with explicit device selection and combination using a plus operator. - Iterative load balancing between devices, converging workload distribution over multiple runs to minimize completion time. - Pipelining for reads, computes, and writes, using event-based queue management to overlap operations and hide latency. - Inter-device pipelines with double-buffered data movement, allowing multiple devices to run different kernel stages concurrently for non-separable workloads. - Batch computing via task pools, keeping all devices busy with grep y scheduling. - Support for both C arrays and C++ wrapper arrays with automatic zero-copy buffer optimization on shared-memory devices. Cekirdekler re