Philip N Daly wrote:
> Hi all,
>
> Not specifically real time Linux but does anyone know:
>
> 1. How long (in ns, us, ms) it takes to execute a floating point
> operation in C and x86 assembler on a 500MHz Pentium III?
>
> 2. How long the delay is when reading from a PCI card ... i.e. if
> the operation value = readPCI(channel) is atomic, how long would
> it take to execute that instruction (again in ns, us, ms)?
>
Itīs (as usual ) not that easy to answer ...
First take a look on the best case :
Your PCI is running on 33 MHZ and a read from a PCI-card is a memory
like
read . This
means: the PCI-Controller puts the adress on the bus , and the card
replys the data.
If You have a FIFO on your Card an the data is delivered with no delay ,
it means :
1 Adresss Cycle = 30 ns
1 Read Cycle = 30 ns
(plus CPU Cycle time, ~ 10 ns on a 200 MHZ ( not sure about this ...))
gives 70 ns
But ......
There are serveral circumstances which delays your read cycle:
- Normally , there is a PCI -Bridge between the internal PCI and the
PCI-Bus you plug your cards in . ( add 30 ns)
- if the Bus or the Bridge is occupied by another transfer ( GFX-Card,
HD-Controller) you have to wait till it is finished (add 30 ns)
- if these devices are doing Burst transfers ( one time adress, many
time
data) you have to wait until the burst is finished ( add 30ns *
Burstlenght)
- if the data on your card isnt available , your card will send a
Busy-Cycle to the controller (add 30 ns until the time your card is
finished)
- if your card is more intelligent, it will asume a retry : Means the
bus
is released, and the controller tries to read "a little bit " later (
add
30 ns up to 1 us , depending ot your PCI Controller).
- In worst case, someone other graps the bus meanwhile and doing a
bursttransfer .... look above ... :)
So its the usual way, best case timings are easy calculated, but they
are
pure hypothetic.
In Therory You can get a datatransferrate of 132MB/s ( 33MHz * 32 Bit ),
but in a real world system you have 25 -30 MB/s.
Hope this helps a bit ..
Regards
Oliver
>
> Thanks in advance,
>
> +==================================================================+
> Phil Daly, NOAO/AURA, 950 N. Cherry Avenue, Tucson AZ 85719, U S A
> E-mail: [EMAIL PROTECTED] V-mail: (520) 318 8438 Fax: (520) 318 8360
>
> --- [rtl] ---
> To unsubscribe:
> echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
> echo "unsubscribe rtl <Your_email>" | mail
[EMAIL PROTECTED]
> ----
> For more information on Real-Time Linux see:
> http://www.rtlinux.org/~rtlinux/
--
---------------------------------------------------------------------------
Dipl.Ing. (BA) Oliver Schindler c/o Robert Bosch Multimedia
---------------------------------------------------------------------------
"Gosh that takes me back... or is it forward? That's the trouble with
time travel, you never can tell."
-- Doctor Who "Androids of Tara"
---------------------------------------------------------------------------
--- [rtl] ---
To unsubscribe:
echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
----
For more information on Real-Time Linux see:
http://www.rtlinux.org/~rtlinux/
