Mail Archives: djgpp/2001/01/18/17:52:12
"Tom St Denis" <stdenis AT compmore DOT net> wrote in message
news:947p8h$vij$1 AT nnrp1 DOT deja DOT com...
> In article <947ao2$ca654$1 AT ID-57378 DOT news DOT dfncis DOT de>,
> "Alexei A. Frounze" <dummy_addressee AT hotmail DOT com> wrote:
> > oh man, you're not understanding the issue at all. do you?
> > okay, you suuggest that I use the following code in order to compute the
> > overhead involved by "push eax/edx":
>
> Sorry I though you were just summing the rdtsc counts then dividing out to
> get an average...
1st I find difference between consequtive RDTSCs. I calculate 16 such
differencies and find the average. It's not about summation, it's about
subtraction.
> You're code is wrong anyways, so if you want to be a meany get your code
> right first!
Nope. It's right, your argument is wrong!
Okay, let me explain once more...
Let's take a look at the source:
-------8<-------
mov ch, 17 ; 17 values for average of 16 periods
l0:
mov ax, [es:6ch]
l1:
cmp ax, [es:6ch]
je l1
rdtsc
push edx
push eax
dec ch
jnz l0
-------8<-------
you claim that those pushes and everything after rdtsc involves overhead
which should be fixed by adding a certain small value to the result of
RDTSC, right?
hold on...
okay, i get those 17 values from RDTSC:
delay
n1 = RDTSC
delay
n2 = RDTSC
delay
n3 = RDTSC
...
delay
n17 = RDTSC
then I evaluate differencies of those n's:
d1 = n17 - n16
d2 = n16 - n15
d3 = n15 - n14
...
d16 = n2 - n1
NOW! if you propose me to add a certain constant to each n (n1...n17) (let's
say) 10, I end up with:
d1 = n17 + 10 - (n16 +10) = n17 - n16
d2 = n16 + 10 - (n15 +10) = n16 - n15
d3 = n15 + 10 - (n14 +10) = n15 - n14
...
d16 = n2 + 10 - (n1 +10) = n2 - n1
which is the same stuff.
NOW, DO YOU STILL DISAGREE AND CLAIM THAT MY CODE IS WRONG???
--
Alexei A. Frounze
alexfru [AT] chat [DOT] ru
frounze [AT] ece [DOT] rochester [DOT] edu
http://alexfru.chat.ru
http://members.xoom.com/alexfru/
http://welcome.to/pmode/
- Raw text -