VSIPL DSP Libraries
The Vector, Signal, and Image Processing Library (VSIPL) is a standardised set of functions and an open application programming interface (API) that provides portable computational middleware for signal and image processing applications. It was developed by a team of hardware manufacturers, software developers, academics, and US government research centres. The API is designed to be totally processor neutral while allowing extensive processor-specific optimisations beneath the surface.
Scope of the VSIPL API:
- support functions
- scalar functions (mathematical functions, complex arithmetic)
- elementwise functions (mathematical functions, arithmetic, comparison, selection, random number generation)
- signal processing functions (FFT, windows, filters, convolution, correlation, histogram)
- linear algebra functions: (vector and matrix products, matrix decompositions, equation solvers).
See https://www.omg.org/spec/VSIPL/ for more details.
We provide highly optimised VSIPL libraries for:
- ARM A53, A57, and A72
- PPC T2080
- PPC T2081
- PPC T4240
- Intel AVX512
- Intel Phi
- Intel AVX2
- Intel AVX
- Intel SSE
The VSIPL API includes definitions of:
- Core Lite
- Core
- Full
- Double Precision Full
The table below gives links to the Reference Manuals, Quick Reference Guides, and Content Lists of each VSIPL specification:
| Spec | Manual | Guide | Contents |
| Core Lite | Manual | Guide | Contents |
| Core | Manual | Guide | Contents |
| Full | Manual | Guide | Contents |
| DFull | Manual | Guide | Contents |
Multi-threaded versions for shared memory systems are also available - the parallelism is completely transparent to the end-user and no changes are required to existing VSIPL code.
For benchmarks over a range of platforms please click on the links below:
VSIPL Benchmark Report 1: PPC T2080, Intel Haswell & Broadwell, and ARM A72 LX2160A.
VSIPL Benchmark Report 2: ARM A53, A57, and A72.
VSIPL Benchmark Report 3: Intel SkylakeX.
VSIPL Spreadsheet: Intel SkylakeX.
