Forums

IAR EWB V3 CLIB IEEE float support?

Started by John Speth December 1, 2005
Hi group-

Page 7 of the MSP430 C/C++ ref guide for IAR EWB V3 says "CLIB... does 
not fully support floating-point numbers in IEEE-754 format".

Does anyone know what the exceptions are to IEEE-754 support using CLIB?

Thanks, John Speth.

Incidentally, IAR, I sent a "request assistance" using the form on the

IAR web site asking the same question.  My acknowledgement email tells 
me that my request was denied.  Here's part of the reply message:

"553
    5.7.1 Hello [213.80.101.18], for MAIL FROM address
    <john.speth@john...> access is denied by the policy analysis
    functions. (in reply to MAIL FROM command)"

(I X'ed out my domain name for spam reasons. It's "coherent".)  I
don't 
know if the message made it or not!!!





Beginning Microcontrollers with the MSP430

John, 

> Page 7 of the MSP430 C/C++ ref guide for IAR
EWB V3 says 
> "CLIB... does not fully support floating-point numbers in
> IEEE-754 format".
> 
> Does anyone know what the exceptions are to IEEE-754 support 
> using CLIB?

I would guess that it does not support subnormals and all four founding
modes of the 754 standard.  I would also guess that signed zeroes are
probably unsupported.  And I'd hazard a guess that it does support some
form of Inf/NaN, but probably not the full signalling/non-signalling
requirements.

Supporting the full set of IEEE-754 (aka IEC 60559) is rather expensive
for those systems that don't need it.

--
Paul Curtis, Rowley Associates Ltd  http://www.rowley.co.uk
CrossWorks for MSP430, ARM, AVR and now MAXQ processors

>> Supporting the full set of IEEE-754 (aka IEC 60559) is rather
expensive 
>> for those systems that don't need it.
Just the paper specification from the IEEE is expensive although it's
only a 
few pages!

----- Original Message ----- 
From: "Paul Curtis" <plc@plc@...>
To: <msp430@msp4...>
Sent: Thursday, December 01, 2005 4:26 PM
Subject: RE: [msp430] IAR EWB V3 CLIB IEEE float support?


> John,
>
>> Page 7 of the MSP430 C/C++ ref guide for IAR EWB V3 says
>> "CLIB... does not fully support floating-point numbers in
>> IEEE-754 format".
>>
>> Does anyone know what the exceptions are to IEEE-754 support
>> using CLIB?
>
> I would guess that it does not support subnormals and all four founding
> modes of the 754 standard.  I would also guess that signed zeroes are
> probably unsupported.  And I'd hazard a guess that it does support some
> form of Inf/NaN, but probably not the full signalling/non-signalling
> requirements.
>
> Supporting the full set of IEEE-754 (aka IEC 60559) is rather expensive
> for those systems that don't need it.
>
> --
> Paul Curtis, Rowley Associates Ltd  http://www.rowley.co.uk
> CrossWorks for MSP430, ARM, AVR and now MAXQ processors
>
>
>
> .
>
>
> Yahoo! Groups Links
>
>
>
>
>
>
>
> 


Hi John!

> Page 7 of the MSP430 C/C++ ref guide for IAR
EWB V3 says "CLIB... does 
> not fully support floating-point numbers in IEEE-754 format".
> 
> Does anyone know what the exceptions are to IEEE-754 support using CLIB?

Both the CLib and DLib share the same low-level floating-point 
implementation. Basically, it has most of the features that you would 
want, including support for NaN:s, infinities, negative zero, correct 
rounding etc.

The things that are not included are subnormal numbers, multiple 
rounding modes, and signalling NaN:s (*).

The things that do differ between CLib and DLib are all the higher-level 
function, such at "sin". The variants in the CLib library are not
aware 
of the special cases such as NaN and Inf. In addition, the DLib library 
contains a lot of functions introduced with the C99 standard.




    (*) The C99 standard says the following about signalling NaN:s, in 
footnote 301 on page 444:

    "Since NaNs created by IEC 60559 operations are always quite, quite 
NaNs (along with inifinities) are sufficient for closure of the
arithmetic."


> Incidentally, IAR, I sent a "request
assistance" using the form on the 
> IAR web site asking the same question.

Thanks, I will look into it.

     -- Anders Lindgren, IAR Systems
-- 
Disclaimer: Opinions expressed in this posting are strictly my own and
not necessarily those of my employer.