Re: [microblaze-uclinux] OT?, Performance

[Goran Bilski <Goran.Bilski@xxxxxxxxxx>]

emanuel stiebler wrote:

> Goran Bilski wrote:
>
>> A VAX 780 could do 1757 iterations/s.
>
>
> It is also called dhrystones per second
>
>> This is what is called 1 DMIPS.
>
>
> Or VAX MIPS
>
>>  From the numbers that I have seen in the thread here, MicroBlaze 
>> could do 5035 iterations/s => 2.8 DMIPS which would happen when 
>> running without caches on a slow external memory.
>
>
> Slow memory ? I thought the MB vanilla has DDRAM ?
> And still, around 30 times slower than on the xilinx web ?

Every access takes around 12 clock cycles,
A load instruction would take around 24 clock cycles to execute. (12 
clock cycles to fetch the instruction and 12 clock cycles to do the read 
of data).
Compare this with running from LMB where the load instruction takes 2 
clock cycles.
One thing that also makes a difference for the Dhrystone numbers is if 
the HW divider has been enabled for MicroBlaze.
It usually improves the Dhrystone numbers with 10%.

>
>> Someone else turn on the caches and got 56529.1 iterations/s => 
>> (56529.1/1757) DMIPS => 32 DMIPS which is more what to expect from 
>> external memory.
>
>
> Sorry, Göran, but you didn't notice, that this was for the VirtexIIPro,
> not for the MicroBlaze. And as the 2VP7 runs at 100 MHz, the expectation
> would be around 150 DMIPS, as the VirtexIIPro has 32 KByte cache already
> (16K icache, 16k dcache) 

He only said that it was on V2P7, it can be either a MicroBlaze or PPC.
I suspect from the numbers that it's a MicroBlaze since this would be 
the expected numbers from a MicroBlaze at 100 MHz and with SDRAM memories.
It it was PowerPC it would have had a higher value than that.

He also specified it in another email:
 >>I'm currently using EDK 6.2
 >>MB 2.10a
 >>opb_v20 1.10b

 >>microblaze runs at 100Mhz

>
>
> So, in both cases we have a pretty impressive difference to the
> web numbers (but we still don't have a value for the MicroBlaze with 
> caches on)
>
>> The DMIPS number that xilinx has reported is when running from 
>> internal memories.
>
>
> So at least the VirtexIIPro chips should reach it, but they don't :(
>
> Cheers,
> emanuel
>
> ___________________________
> microblaze-uclinux mailing list
> microblaze-uclinux@itee.uq.edu.au
> Project Home Page : http://www.itee.uq.edu.au/~jwilliams/mblaze-uclinux
> Mailing List Archive : 
> http://www.itee.uq.edu.au/~listarch/microblaze-uclinux/
>


___________________________
microblaze-uclinux mailing list
microblaze-uclinux@itee.uq.edu.au
Project Home Page : http://www.itee.uq.edu.au/~jwilliams/mblaze-uclinux
Mailing List Archive : http://www.itee.uq.edu.au/~listarch/microblaze-uclinux/