On Wed, 2012-03-21 at 17:58 +0100, R. Diez wrote:

> I cannot think of a good reason why we should invest our (little) time 
> implementing a configuration setting that no-one should be using.

Hi R Diez,

You misunderstand the primary role of Or1ksim. It is not intended to
narrowly model one particular implementation, but to allow architects to
explore a wide range of architectures.

I can think of at least one use case, where I would particularly want to
set explicit values in registers. And that is where I am exploring what
the implications are of an architecture that initializes registers on
reset.

> I think we should change or1ksim's behaviour to initialise all general 
> purpose registers with random values and let all broken tests fail from 
> now on. or1ksim's default settings should encourage writing good tests 
> that run everywhere as per the architecture specification.

Random values in particular are a nightmare default. Most of the time,
whatever happens I want to be able to repeat it. Introducing Heisenbugs
by default is the last thing any sane person should do!

> Fixing such broken test cases is straightforward and usually worthwhile. 
> If anybody has such a huge test suite that it's not worth fixing (who 
> would that be?), then let he write a patch for or1ksim to add such a 
> configuration setting.

I'm all for fixing the test cases. It is the change to the tool itself I
am arguing about.

> I'll be posting here patches to fix or1ksim's test suite in this respect.

Excellent. thanks for your hard work and look forward to seeing the
patches.


Jeremy

-- 
Tel:      +44 (1590) 610184
Cell:     +44 (7970) 676050
SkypeID: jeremybennett
Email:   [email protected]
Web:     www.embecosm.com

_______________________________________________
OpenRISC mailing list
[email protected]
http://lists.openrisc.net/listinfo/openrisc

Reply via email to