|
Parallelism
and kernel software
We
have developed a scalable kernel for a parallel system based on Ariel
boards equipped with four Texas Instruments TMS320C40 DSP chips. The
system is a MIMD machine featuring ring topology and
"pipeline" load sharing.
We
write optimized code
We
are experienced C and C++ programmers, as well as assembler code
developers. Nowadays optimized C compilers for general purpose CPUs have
made assembler programming almost unnecessary; but in the case of
strongly pipelined DSP, C language can introduce huge overloads. We are
able - for TI C30 and C40 and for AD SHARC - to write optimized code up
to ten times in performance - a desirable feature for real time
applications and cost control.
Complex
algorithms
We
manage time-frequency transforms (DFT, FFT, Winograd, Spectrogram, Gabor-like,
wavelet based multiresolution analysis and synthesis), non-linear
filters (Chebitcheff, Laguerre), as well as time and frequency warping
techniques. |