We develop and market optimised software libraries for signal and image processing and scientific/engineering applications. The libraries take full advantage of the single instruction, multiple data (SIMD/vector) capabilities of current processors, and important routines are hand-tuned (often at the assembler level) to keep the pipeline running at full capacity. Where multicore versions of the processors are available, the libraries are multithreaded to yield additional performance with no user effort.
We currently support the following libraries:
VSIPL provides a portable, open standard API which is widely used in DSP applications.
CSIPL offers access to the same facilities as the VSIPL library through our own C-language
API, which we call CSIPL. This is popular for applications which need DSP library facilities, but are not VSIPL-based.
The VSIPL Image library API is an adjunct standard to VSIPL. We provide an optimised and multithreaded implementation.
FFTW is a highly optimised version of the widely used public domain FFT library. Our implementation provides major performance advantages compared with the open source version.
Veclib provides a wide range of low level optimised vector facilities.
The libraries are available for
PowerPC 74xx with Altivec unit (G4)
MIPS with paired-single and 3D extensions
Intel Processors:
Our Intel libraries are multithreaded and will run on any Intel SSE processor.
Versions with processor-specific optimisations are available as follows:
Processor family Linux VXWorks
Atom Now Now
Core 2 duo Now Now
Core 2 Quad Now Now
Core i7 Now Now
and for a range of operating systems. For details of availability see the individual brochures.
Our technology is used by a number of major DSP board manufacturers - why not join them? We can provide custom-optimised implementations of the libraries for your boards. If you need libraries to conform to your in-house API our technology provides a cost effective route.
