For those whom want to elaborate a library with SSE2 instruction...you'll

find an example which show you how to do...(some clues)...

SSE2 is very interesting, beacause it uses FP number 64 bits and in that fact,
looks like 3DNow instructions of AMD...

I think you all know how to compute ln(x) and the trigo functions...

For instructions SSE and SSE2 go to http://www.cpuid.com

Have a look at this code...below...

http://pageperso.aol.fr/GerardChap/Calc.zip


**********************************************
"The more difficult is not to find the Truth, but to substitute it at the Error..."...
(- Unknow -)
Posted on 2003-03-19 06:18:21 by gerard
You'll find in the Folder below, a list a function ....Sin,Cos,Tan,Atan,Ln,exp...

and it is well-explained how to do...(i.e algorithm)...


Please go to the URL below....

I think you'll find here a good stuff...



-------------------------------------------------------------
http://pageperso.aol.fr/GerardChap/Calculus.zip
Posted on 2003-03-29 02:58:37 by gerard
Good work! :alright:

Some functions in the ZIP file seem to be better in plain FPU instructions, though,either in size or in speed, or both.
Posted on 2003-03-29 06:06:13 by Starless
Here it is....

You'll find in the folder below...all (or almost) you want to know about calculation (computation)of transcendentals functions...

They are more accurates and more quick...than in folder "Calculus1" see previous answer....

I think, I enough well explained the construction of these functions...

I hope, you will understand how is computed Sin,Cos,Exp and Ln (Tan & ArcTan also).....



http://pageperso.aol.fr/GerardChap/Calculus1.zip


---------------------------------------------------------
"Pride permits to overcome the diffilcuty to know."
( Unknown )
Posted on 2003-04-01 23:50:48 by gerard
I just had a very quick glance at the source code for some of the functions. One detail immediately got my attention.

The short REAL format only has 24 significant bits to represent a number, equivalent to approximately 9 significant digits in a decimal number.

When converting from decimal to binary, at most the 12th digit of a number would have an effect on the least significant bit of the resulting binary number. I noticed that in the data section, some REAL4 variables (constants) are initialized with numerical strings containing as many as 32 digits!!!

Would anybody have any explanation for putting so many useless digits in a data definition section?

Apart from that anomaly, that library seems quite powerfull and requires more study by myself. However, I'm not yet convinced that it is more efficient than using the FPU.

Raymond
Posted on 2003-04-02 14:34:05 by Raymond
gerard, nice prog.

Are you sure you need to do all those pxors. I think things would work grand without them.
Posted on 2003-04-03 08:16:49 by Eóin
In this folder (this-one which is mentioned below) "calculus2.zip" you could find all

you wanted to compute tangente, arctangente,arcsine and arccosine....

I think it is enough well-explained...and in that fact you could terminate or build

your own library with SSE2 instructions...

To find them (SSE2 instructions) goto
http://www.cpuid.com and try to elaborate their own syntax

with the help of those of SSE instructions...



http://pageperso.aol.fr/GerardChap/Calculus2.zip


---------------------------------------------------------
"Pride permits to overcome the diffilcuty of knowing."
( Unknow )
Posted on 2003-04-07 10:00:45 by gerard