Re: Computer programmers' habits in electronics

On Wed, 21 Dec 2005 12:28:17 -0800, the renowned Tim Wescott
<tim@xxxxxxxxxxxxxxxx> wrote:

>>>
>Someone who likes to ask that question told me of an interviewee who got
>as far as saying "I think it involves recursion...".
>
>So I wrote a version that did it using recursive function calls and sent
>it to her. I don't know if I would have gotten the job -- they had
>already made the mistake of hiring me.

Something like this? (ugh)

#include <string.h>

rev2(char * istring, int start, int fin)
{
char t;
t = istring[start];
istring[start] = istring[fin];
istring[fin] = t;
start++; fin--;
if (fin > start) rev2(istring, start, fin);
}


....

/* reverse the string in place */
rev2(mystring, 0, strlen(mystring)-1);

....

>Good thing the PC has a lot of stack space.
>
>For interviewing embedded SW engineers we finally settled on a fairly
>basic scaling problem. We started with a little story problem that
>required the interviewee to find the ratio of a couple of numbers and
>multiply it to a third, then we said "oh, by the way, our floating point
>library is too slow -- do it with integers". The question usually took
>about 40 minutes to explore fully, with some folks never getting it and
>some just glancing at the board and writing down the correct answer.
>
>It's amazing how you can separate the desktop programmers from the
>embedded engineers with that one.


Here's one I did to approximate e^(-x) over the range 0 .. 0.75
(the end product would be written in assembler**, the C is just to
explain).



---

where xf, yf are 32-bit integers.

xf = x * 128;
yf = 32768 - xf + (xf* xf)/65536 - (((xf * xf)/32768
)*xf)/(6*32768) + (((xf*xf)/32768)* ((xf*xf)/32768))/(32768 * 24)
- (((((xf*xf)/32768)* ((xf*xf)/32768))/32768) * xf)/(32768 * 120)
;

The result is ~= exp(-x/256) * 32768 over the range
x = 1 to 192.

yf/32768 agrees with exp(-x/256) within 0.001 over that range.


You can easily extend this to additional terms, although it gets a bit
tedious.

It's based on the textbook Taylor/Maclaurin series

infinity
---
\
exp(x) = > (x^n)/n!
/
---
n= 0


** actually assembler macros, but that's another story

---


Best regards,
Spehro Pefhany
--
"it's the network..." "The Journey is the reward"
speff@xxxxxxxxxxxx Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog Info for designers: http://www.speff.com
.