Re: [avr-libc-dev] [bug #21623] boot.h: Use the "z" register constraint

[Wouter van Gulik]

Shaun Jackman schreef:
> Follow-up Comment #2, bug #21623 (project avr-libc):
>
> Definitely remove r30 and r31 from the clobber. The spm instruction doesn't
> clobber the Z register, since it has the same value going out as coming in. If
> it did modify the Z register, it should be listed with a read/write
> constraint, as in "+z" or possibly "+&z" if it's an early clobber.
>
> Do you have a small test case for the EEPROM bug you mentioned? I've had no
> problem, and I'm using both gcc and avr-libc from head. I use both the EEPROM
> API and my flash patch with no trouble. I'm certainly not against more testing
> though.

The EEPROM problem arises when there are many (IIRC at least 7) 16 bit 
registers used and need to be saved over a call then gcc fails 
allocating space for the 16 bit vars. For more details look at this bug 
report http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31644

In your case however it is totally different. Here the spm instruction 
needs the Z register. The datasheets (for the Atmega324p and Atmeg8535) 
states that: "Once a programming operation is initiated, the address is 
latched and the Z-pointer can be used for other operations." So it's 
save to assume it's not modified.

In the EEPROM read word function Z was used as return value but it was 
not strictly necessary, any arbitrary 16 bit reg was ok. Somehow using Z 
as output of a assembler piece is causing trouble when having many local 
values.

HTH,

Wouter