|
|
SuperSpeed* math software products provide compact, economical, and easy-to-use routines that give high level floating point functionality, without hardware contraints. SuperSpeed math provides performance exceeding the fastest conventional floating point software.
Log Point SuperSpeed Soft CoProcessors are compact, high performance numerical processors that constitute the first general purpose exponential floating point (efp) computational environment. The new character of efp computation gives Log Point math products unrivaled performance and flexibility.
Unlike conventional floating point, efp is based upon the combination of a fully logarithmic data format and ultra high speed logarithmic and exponential data transformations. As a result, many tough nonlinear computations are simplified and enhanced.
With the new efp data format, numerical computation is still directly equivalent to conventional floating point (cfp) using the same size data format to provide virtually identical precision and dynamic range. This equivalence allows existing cfp algorithms to achieve significantly improved performance without modification and also provides ample opportunity for further enhancement.
ROMable re-entrant code forms include: C source, optimized assembler source and linkable object modules; libraries compatible with Microsoft Visual C, Borland C/C++, Metaware High C, Watcom C/C++, and GNU Tools. Soft CoProcessors are thoroughly tested to deliver uniform precision with comprehensively documented timing and error characteristics.
Execution times for the Soft CoProcessor in the 3-D plot are for C source. In optimized assembler source, all functions would be faster than C source by 30% or more. If the application is always within dynamic range, further optimization can be obtained (multiply/divide would take 0.015 usec, square root would take 0.030 usec, other function execution times would be reduced by 30% or more). |
