Sourcery VSIPL++ 1.4

Features and notable improvements in this release include:

Binary Packages for Power GNU/Linux Systems

Pre-compiled binary packages of Sourcery VSIPL++ for Power GNU/Linux systems running Fedora Core 7 are now available.

Optimizations

Sourcery VSIPL++ now implements return-block optimization (RBO). This optimization creates more opportunities for Sourcery VSIPL++ to fuse operations, providing improved performance for calculations making use of FFTs and other complex operations.

Sourcery VSIPL++ uses underlying high-performance math libraries, such as Mercury's SAL or Intel's IPP, for many more operations.

Sourcery VSIPL++ uses SIMD operations for more expressions.

The API has been extended to include optimized implementations of threshold, axpy, maxval, and minval.

Performance of computations involving sparse matrices and tensors has been improved.

Technology Previews

Preliminary support for the Cell/B.E. Processor is included in this release. A complete, optimized release for Cell/B.E. is planned for September 2008.

Preliminary support for stencil operations, i.e., operations which compute values for elements of a matrix based on neighboring elements, is included in this release.