
Math
Library 
200+
functions
Handoptimized
Simple calls
Parallel processing
FFT
DSP functions
Vector functions
Matrix math
Transcendental
Utilities 



DSP
Math Library 
The DSP Math Library is an extensive and
evergrowing collection of standard subroutines for signal processing, matrix arithmetic,
and general purpose vector processing. All algorithms have been hand coded in S4 assembler
to achieve maximum performance and throughput. All functions have a simple C calling
sequence and predictable execution time.


DSP
Programming 
Programming using the DSP Math Library is as
simple as choosing the function to be performed and calling it from your AX program. Most
functions that are needed for signal processing are in the DSP Math Library, and many
other applications are also supported. Standard functions such as FFTs, simple vector
arithmetic, square root, transcendentals, polynomials, convolutions, and matrix functions
are available. More esoteric functions are also included, such as polyphase filtering,
peak finding, frequency and phase shifting, and combination functions for efficient
chained vector arithmetic. All but the most trivial DSP functions sustain 60% to 85%
of peak CPU efficiency – better than any other commercially available vector
processor. A full list of functions, calling sequences, and execution times
can be found in the documentation section of this web
site.


FFTs
Are Our Specialty 
The DSP Math Library includes functions to
perform forward and inverse 1D and 2D Fast Fourier Transforms on Real and Complex data.
Our 1D functions easily handle FFT sizes from 32 words up to 32 billion words on
combinations of powers of 2, 3, and 5. Multiple short FFTs can be executed with a single
call, reducing overhead, and a windowing function can be combined with the FFT with no
additional hardware CPU cycles or latency. The S4 CPU, the TM66 swiFFT chip, is
optimized for the radix16 butterfly algorithm. For example, a 4096point CFFT involves
three radix16 passes, and the dual 50 MHz TM66 chips on an S4 node each deliver a complex
result every clock cycle. Thus, a 4K Complex FFT completes in just 4096 points x 3 passes
/ 2 chips = 6144 cycles or 123 microseconds.


