MPC5200 bogomips

MPC5200 bogomips

[Stuart Adams]

  We are only seeing 263.78 bogomips on a MPC5200 running
  at 396 MHz.

  Doesn't this seem way to low ?? With a 603e core I'd expect
  1 bogomip per MHz or better.

  The exact same kernel source running on an 8xx PPC gets about
  1 bogomip per MHZ and my 745 CPU board does 2 bogomips
  per MHz ... the bogomip calibration code is the same for all
  PPC architectures so it seem like it should be an apples-to-apples
  comparison.

-- Stuart

Re: MPC5200 bogomips

[Mark Chambers]

----- Original Message ----- 
From: "Stuart Adams" <sja@brightstareng.com>
To: <linuxppc-embedded@ozlabs.org>
Sent: Thursday, January 27, 2005 11:29 AM
Subject: MPC5200 bogomips


> 
>   We are only seeing 263.78 bogomips on a MPC5200 running
>   at 396 MHz.
> 
>   Doesn't this seem way to low ?? With a 603e core I'd expect
>   1 bogomip per MHz or better.
> 
>   The exact same kernel source running on an 8xx PPC gets about
>   1 bogomip per MHZ and my 745 CPU board does 2 bogomips
>   per MHz ... the bogomip calibration code is the same for all
>   PPC architectures so it seem like it should be an apples-to-apples
>   comparison.
> 
> -- Stuart
> 

Ha! You almost sound like you are quoting me from when I first
looked at bogomips on the 5200.  Here's the story:  If you drill
down to where bogomips are calculated you will see that it is
measuring the speed of a dbnz instruction.  The calculation assumes
2 clocks per dbnz.  So the 8xx comes out at 1 BogoMIP/MHz.  The
5200 takes 3 clocks/dbnz in this loop, so you get .67 BogoMIP/MHz.  
Not to worry, it's only one specific scenario.  Overall I've found the 
5200 to be about 1.5 MIPS/MHz compared to 8xx 1 MIPS/MHz.

Mark Chambers