inetbot web crawler
Main  |  Get access to the repository  |  API  |  The robot  |  Publications  |  Usenet Groups  |  Plainweb  | 
 inetbot - Groups (beta)

Current group: comp.sys.arm

Benchmarking of the ARMulator model

Benchmarking of the ARMulator model  
Saurav Malla
 Re: Benchmarking of the ARMulator model  
William Munns
From:Saurav Malla
Subject:Benchmarking of the ARMulator model
Date:28 Dec 2004 06:14:47 -0800
If I simulate a particular ARM system having ARM926EJ-S as the core
and other peripheral system, how much does the execution time of a
benckmarking algo differs on the simulator from the actual board.

Currently I have tested a benchmarking algo on the simulator and
clocked the time, it comes out around 18seconds (on average, its not
always the same).
and for the actual H/W I multiplied the total clock cycles it took to
execute the algo with the speed of the processor and it came out to be
80milliseconds.
This total clock cycles can be viewed in the debugger's internals of
the AXD debugger.

Can you tell me why this vast difference in execution time?
If this result is absurd then is there a mistake that i am making.

Thanks
Saurav
From:William Munns
Subject:Re: Benchmarking of the ARMulator model
Date:29 Dec 2004 13:12:17 GMT
armSavvy@gmail.com (Saurav Malla) wrote in
news:beca3ee0.0412280614.cb0d853@posting.google.com:

> If I simulate a particular ARM system having ARM926EJ-S as
> the core and other peripheral system, how much does the
> execution time of a benckmarking algo differs on the
> simulator from the actual board.
>
> Currently I have tested a benchmarking algo on the
> simulator and clocked the time, it comes out around
> 18seconds (on average, its not always the same).
> and for the actual H/W I multiplied the total clock cycles
> it took to execute the algo with the speed of the processor
> and it came out to be 80milliseconds.
> This total clock cycles can be viewed in the debugger's
> internals of the AXD debugger.
>
> Can you tell me why this vast difference in execution time?
> If this result is absurd then is there a mistake that i am
> making.

Assuming everything else is set up correctly (memory wait states
and caching being the most important). Then your problem is
almost definatly because you are using 'real time' rather than
'simulated' and 4000000 (for 4 Mhz). If you use 'real time' it
does not calculate the clock, but uses the PC clock.

HTH
Will
   

Copyright © 2006 inetbot   -   All rights reserved