simulavr-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [Simulavr-devel] Tiny10 Family Support


From: Weddington, Eric
Subject: RE: [Simulavr-devel] Tiny10 Family Support
Date: Sun, 27 Feb 2011 20:27:25 -0700


> -----Original Message-----
> From: Boyapati, Anitha
> Sent: Sunday, February 27, 2011 8:04 PM
> To: Weddington, Eric; Joerg Wunsch; address@hidden
> Subject: RE: [Simulavr-devel] Tiny10 Family Support
> 
> The idea is to run the testsuite (dejagnu or more) per architecture
> choosing some candidate device. I have used avrtest, but it supports only
> avr51 and avr6. In freaks I found another AVR simulator by buserror -
> simavr. Even that lacks tiny10 and full xmega support.

There is *no* simulator, especially an open source simulator, that is going to 
have atiny10 family support, or xmega support. These devices are still so new, 
and open source AVR simulators are still not very mature, that you're not going 
to find this anywhere except for the proprietary simulator being worked on in 
Atmel Norway.
 
> That said I find the design of simulavr intuitive. Adding couple of
> instructions and devices should not be too difficult. But before that I
> would like to understand how robust the current features are. Can we rely
> on its results as far as ISA is concerned? I'll have to explore these.

The important thing is not so much adding instructions and devices, but you 
need to investigate:

- How robust is the interface to GDB, and hence being able to run dejagnu?

- How fast is it to run the entire GCC testsuite, via dejagnu, on a particular 
simulator?

You have to remember how many tens of thousands of tests there are in the GCC 
regression test suite that you have to run. Shaving off seconds here and there 
can add up to hours when running the full test. Vidya can attest to this.

This is the main reason why I recommend using the avrtest code base. It was 
designed to run the GCC regression test suite for the avr (so the interface to 
GDB should be robust), and be small and fast.

If you're planning on increasing the number of devices in a simulator and run 
the GCC regression test suite on even more devices, then it will be extremely 
important to be able to blaze through those tests.

Eric Weddington



reply via email to

[Prev in Thread] Current Thread [Next in Thread]