Re: [microblaze-uclinux] microblaze performance

[John Williams <jwilliams@xxxxxxxxxxxxxx>]

Hi Rudi,

Rudolf Usselmann wrote:

> I was playing around a bit with the cache size and trying to
> find a trade off of cache size vs BRAM usage. In my applications,
> it seems that I never have enough BRAMs, so I try to justify
> very carefully where and how many BRAMs I use.

> Here are some of my findings. The tests where performed on a
> Virtex II Pro 20, soc running at 66.66 MHz, the first column
> is the size of icache/dcache.
> 
> 16k/16k, 5M dhrystone: 42030.9 dhrystone/sec  (24 extra BRAMs)
>  8k/8k,  5M dhrystone: 41830.5 dhrystone/sec  (12 extra BRAMs)
>  4k/4k,  5M dhrystone: 41480.0 dhrystone/sec  ( 9 extra BRAMs)
>  2k/2k,  5M dhrystone: 19957.7 dhrystone/sec  ( 8 extra BRAMs)
>  0k/0k,  5M dhrystone:  4815.4 dhrystone/sec  ( 0 extra BRAMs)

This is a great test to perform, thanks for sharing your results.  I 
guess some things to try would be hit the microblaze with a ping flood 
before running dhrystone, see how the cache size impacts the performance 
as well.

Also, maybe running a couple of instances of dhrystone at the same time, 
to see what happens under a processor load > 1.

> Looks like 4K/4K is a optimum setting. Bogo MIPS stayed pretty
> much constant at 32.97 regardless of cache size (which I guess
> does make sense).

Yes.  bogomips is just a tight delay calibration loop, so any non-zero 
cache will tend to yield a big impact.

Cheers,

John


___________________________
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/