Re: [avr-gcc-list] [discussion] lpm support in avr-gcc

[Svein E. Seldal]

Denis Chertykov wrote:
> IMHO: It's impossible for avr port without changes in generic GCC.
> So, It's a question to GCC maintainers.
> Even Linus has asked about something like this.

Is there applied any pressure to get this feature implemented? I mean, 
if this is a crucial feature that lacks in gcc, shouldn't it be 
implemented? Is really absolutely all gcc targets using non-harvard 
architecture? I know that this problem has been solved by placing eeprom 
and data memory using very high addresses, but this can't be a optimal 
solution in the long rung, can it?

For avr-gcc this lack of feature imposes several implications on the 
code compiled by avr-gcc:

o Strings cannot be placed in flash without declaring its storeage 
first. I.e. const static PROGMEM string[] = ...

o Duplicate strings implemented with PSTR() is not optimized away 
(merged) because they are declared as static.

o Multidimensional stucts and information in flash is very cumbersome. 
Examples: stuct containing a const string or stuct containing pointers 
to other const structs (all stored in flash).

o The _LPM command is volatile and cant be optimized away

o When accessing flash you *always* need to cast the pointer -> unsafe.

I'm not trying to be critical, but I'm trying to throw in arguments to 
be able to apply pressure on this issue.


Svein

avr-gcc-list at http://avr1.org