RE: [Simulavr-devel] avrtest and simulavr vs simulavrxx

[Weddington, Eric]

> -----Original Message-----
> From: Joel Sherrill [mailto:address@hidden 
> Sent: Tuesday, March 10, 2009 4:12 PM
> To: Weddington, Eric
> Cc: address@hidden
> Subject: Re: [Simulavr-devel] avrtest and simulavr vs simulavrxx
> 
> That makes sense.  Once my "magic port" and time limit
> changes are committed we will need to create another
> dejagnu/boards file.  I assume you can run avr-elf gcc tests
> easily.

Mmm, somewhat.
 
> I noticed avrtest only supports two avr device names avr51
> and avr6.  I assume these are the cores/multilib names.

Yes. They are called "architectures" in the port, and they reference features + 
instruction sets. They are arbitrary names though with no intended relationship 
to actual core revisions.

> How do these map into the simulavrxx devices?
> 
> AT90S4433
> AT90S8515
> ATMEGA128

Btw, you can always reference these in the gas user manual, under "AVR 
Options", for the -mmcu=<device> flag.

atmega128 = avr51
at90s8515 = avr2
at90s4433 = avr2

The atmega128 is pretty much used as the "golden standard" for AVR. The avr6 
architecture was added to avrtest by Tristan Gingold. The AT90* devices are 
older devices.

 
> Do we want "synthethic" CPU models for testing which comprise
> the right core and the phony avrtest magic ports?

I would rather test to a specific device, than to any kind of architecture 
type. At this point, I don't think we guarantee that one can correctly compile 
an AVR application using just the architecture name (e.g. avr6), though in 
theory it is supposed to work. If you do select an architecture name, you don't 
pull in any I/O device header file from avr-libc, hence no access to any I/O 
registers.