Re: [vhdl-200x] Regarding Fixed Point Algorithmic User's Guide

On 10/09/2013 11:21 AM, Sharad Sinha wrote:
> Hi Jim,
>
>   I guess that brings up the question if the package itself has been 
> verified and validated carefully and if so what are its limitations in 
> terms of accuracy of results and whether they have been quantified. 
> The associated documentation says: "These algorithms are not 
> exhaustively debugged....use at your own risk..".
>
Did this two ways.   First, I checked various numbers, then I ran a loop 
of random numbers.
>   Besides, I am not sure why I would want to validate a hardware 
> implementation of these functions if I am using pre-verified 
> implementations of these in my overall design which is generally the 
> case in complex designs. I understand that if I do an implementation 
> of my own, then I would like to verify it. But do I really need to 
> verify my implementation this way? If verification is indeed an 
> objective of including such packages, then the documentation should be 
> more rigorous with more insight into how the package code works.
>
Basically, all of the functions in this package call out "real" 
functions.    Since most of the REAL type functions actually call out 
the C (from math.h) versions of these functions, I don't see an issue 
with functionality.

>   The reason I am a bit picky about such parts of packages like these 
> is that it makes me feel that probably VHDL equivalent packages of 
> BLAS, LINPACK etc. would be better. Of course based on how packages 
> are being proposed now, it is possible that we can have packages for 
> statistical calculations and likewise.
>
I like to think of it as "math.h" for fixed and floating point.


-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.