hi,

i need to compute a/sqrt(x) in floating point arithmetic.

if this is done using reciprocal square root, does it introduces

errors .. if all operations are done with 'infinitely precise rounding'

and same no of bits in fractional part?

if yes .. how much precise Rsqrt(x) shud be to do this properly?

thanks,

t0ny

The rounding might be different, but it shouldn't be any more in error.Quote:> hi,

> i need to compute a/sqrt(x) in floating point arithmetic.

> if this is done using reciprocal square root, does it introduces

> errors .. if all operations are done with 'infinitely precise rounding'

> and same no of bits in fractional part?

> if yes .. how much precise Rsqrt(x) shud be to do this properly?

If you are that worried about error, maybe you shouldn't be using floating

point?

-- glen

hi, thanks

that will be great ..Quote:> The rounding might be different, but it shouldn't be any more in error.

have to use erroneous fp ;..Quote:> If you are that worried about error, maybe you shouldn't be using floating point?

i was looking for a way to equate the results (and so errors) of

possible methods, if the error is not more than normal method that'll

be enough..

how this shud be analyzed ?

thanks again ..

1. sqrt function doesn't use FPU sqrt instruction ?

Does sqrt (in libm.a or libmfast.a) use the FPU coprocessor

square root instruction (see page B-53 in the MIPS R4000 User's Manual)

or does it compute it manually ?

I disassembled the executable of a simple program that computes the

square root of a number using sqrt() and found no trace of the

instruction.

Whats up here ? Somehow I feel like I'm being cheated.

System is Indy R4400, IRIX 5.2. languaage is Ada.

___________________________________________________________________________

Tom Haberlandt | Reality's a horde of mice nibbling away in the

___________________________________________________________________________

"I don't speak for my employer." - Woof !

___________________________________________________________________________

3. sqrt(sqrt(-1)) and m-files - maybe a bug?

4. attaching a download to my page

5. DSP Trick: sqrt() via reciprocal sqrt() a la Newton-Raphson

6. ** Computer Buying Guide **

7. Looking for server task which can create a multiple domain N&A from several N&A′s

9. Sqrt pb

10. C, IEEE float, implementation of sqrt/exp/ln/sin...

11. Need 1/sqrt(x) for simple machine

12. incremental algorithm for a/(b+c*sqrt(x)) ?

3 post • Page:**1** of **1**