KiRAT - Kiel Real-time Application Toolkit

Processor Abstraction Layer (PAL)

The Processor Abstraction Layer (PAL) provides a comprehensive set of mathematical, signal-processing, vector, matrix, and utility functions that can be used throughout KiRAT. Its main purpose is to make algorithm development both fast and hardware-independent. Instead of implementing frequently required numerical operations repeatedly, developers can use a consistent set of PAL functions with clearly defined interfaces. This significantly reduces development effort and allows new algorithms to be implemented and tested very quickly.

A key feature of the PAL concept is that the same function can have multiple processor-specific implementations. For each PAL function, a portable ANSI C reference implementation is available as a common baseline. Depending on the target platform, this implementation can be replaced internally by highly optimized versions, for example based on Intel Integrated Performance Primitives (Intel IPP) or dedicated implementations for digital signal processors (DSPs) and other processor architectures.

The application code itself does not have to be modified when changing the underlying hardware platform. The desired PAL backend is selected at compile time using an appropriate compiler switch. The PAL then automatically maps the generic function calls to the implementation optimized for the selected processor.

This separation between the algorithm and its hardware-specific implementation offers two major advantages.

• First, algorithms remain portable and easy to maintain, since the same source code can be used on different processing platforms.

• Second, computationally intensive operations can benefit directly from processor-specific instructions, SIMD extensions, optimized memory access, and highly tuned numerical libraries.

Depending on the operation and target architecture, this can result in a substantial acceleration of the overall processing, without requiring changes to the actual signal-processing algorithm.

The PAL therefore combines the convenience of a high-level numerical function library with the efficiency of hardware-specific optimization. Developers can concentrate on the mathematical and algorithmic aspects of their application, while the PAL provides an efficient implementation for the selected processing platform.