Øyvind Harboe wrote:
> On Fri, May 15, 2009 at 10:31 PM, Michael Fischer <[email protected]> 
> wrote:
>   
>> Hello List,
>>
>> I have tested the speed from 1606 with the patch from
>> Magnus and the 1787 with a FT2232 under Windows. The
>> target was a STR710 with external RAM.
>>
>> A very old speed table from r657 can be found here:
>> http://www.yagarto.de/projects/jtagspeed/index.html#result
>>
>> Here the FT2232 had a speed in the past about 165KB/sec on a STR710 target.
>> I have repeated the test again, here are the results:
>>
>> -  657: 164KB/sec
>> - 1606: 148KB/sec
>> - 1787: 149KB/sec
>>
>> This is a speed decrease from about 9%
>>     
>
> I.e. there is no difference between 1606 & 1787, which is good :-)
>
> Nicolas Pitre reported *huge* standard deviations
> in performance. Are you seing any significant standard
> deviation in performance?
>
> Also, I have no idea if somehow the speed decrease is due to
> some *necessary* extra overhead in terms of extra operations
> against the CPU between 657/1787.
>
> Interesting though.
>
>   
The difference between r657 and r1606 is worth checking.
After r1606 and when the obvious bugs are fixed there should be no large 
performance difference.

Öywind has built a more or less functional equivalent of the old system. 
The number of function calls
has probably increased a little, but this does not affect performance a lot.
The new structure traces callback handles in a  almost IDENTICAL way to 
the old one BUT it is a separate datastructure
that can be INLINED and removed for synchrounous hosts.

So we get a small, not measurable penalty for USB hosts and a  potential 
improvement for synchronous hosts.

Best regards
Magnus

_______________________________________________
Openocd-development mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/openocd-development

Reply via email to