ST7 based SQRT algorithm

[vdeokar]

Posted on May 07, 2003 at 17:41

ST7 based SQRT algorithm

[vdeokar]

Posted on October 28, 2002 at 17:59

Would like to know if there is a sqrt algorithm - word, no floating point in assembly. I dont want use the C library function. The execution time of less than 150uS at 16MHz clock

[johan239955_stm1_stmicro]

Posted on November 07, 2002 at 19:25

The enclosed assembly sqrt subroutine computes the square root of word and return the nearest word (floating number not supported). Just extract the assembly code between the 2 pragma in this C file and you should be all set.

Execution time is 187us at Fcpu = 8MHz (maximum ST7 speed). That's the best I have but maybe this subroutine can still be optimized Regards Jojo ________________ Attachments : sqrt.zip : https://st--c.eu10.content.force.com/sfc/dist/version/download/?oid=00Db0000000YtG6&ids=0680X000006I09t&d=%2Fa%2F0X0000000bUZ%2FeZKWGOo.QiTa20RsHFauIPxTIistBgSJsqF388drd6Q&asPdf=false

[johanpauvert1]

Posted on November 08, 2002 at 12:20

Cosmic sqrt function computes the square root of a double variable, returning the nearest double floating value. The problem is the execution time : ~1.55ms.

Do you also have sqrt algorithm in assembly that supports doubles instead of words and get faster result ? Breizh

[johan239955_stm1_stmicro]

Posted on November 08, 2002 at 13:37

I don't have better than the Cosmic one regarding double. If an integer variable sqrt algorithm is fine with you, I also have a sqrt long no floating point in assembly language. Both sqrt word and sqrt long have been included in this new attachment.

FYI : Execution time word = 187us at Fcpu = 8MHz Execution time long = 360us at Fcpu = 8MHz Jojo [ This message was edited by: jojo on 14-11-2002 20:46 ] ________________ Attachments : sqrt.zip : https://st--c.eu10.content.force.com/sfc/dist/version/download/?oid=00Db0000000YtG6&ids=0680X000006I09o&d=%2Fa%2F0X0000000bUa%2FL8nIobjKxQzJmf4lQcy.FJIQrzqKPf2gP3L5JFRK6sg&asPdf=false

[johan239955_stm1_stmicro]

Posted on March 12, 2003 at 15:40

I just got better result with the latest sqrt functions from Cosmic : isqrt (word) and lsqrt (long).

At Fcpu = 8MHz using their modm model, I have : Execution time word = from 17us to 34us Execution time long = from ~22us to 57us (result depends on the input parameter value) It looks like this Cosmic subroutine is the best we have so far. Take care, Jojo

[coor]

Posted on March 19, 2003 at 07:01

try this simple algoritm:

unsigned int sqrt_cpu_newton (unsigned int num) { unsigned temp, div = num; unsigned rslt = num; if (num <= 0) return 0; for(; { temp = num / div + div; div = temp >> 1; div += temp & 1; if (rslt > div) rslt = div; else return rslt; } }

[johan239955_stm1_stmicro]

Posted on April 03, 2003 at 20:05

Pretty simple but still longer than the Cosmic functions. I have :

Execution time word = 190us at Fcpu = 8MHz Take care, Jojo

[daffy_19722]

Posted on April 04, 2003 at 03:23

These isqrt (word) and lsqrt (long) subroutines are not included in the latest Cosmic compiler (4.4d).

Where can we get them ?

[johan239955_stm1_stmicro]

Posted on May 07, 2003 at 17:41

These standard library subroutines are available with the latest Cosmic compiler 4.4e.

Take care, Jojo