Re: [avr-gcc-list] RFC gcc multiple memory space implementation

[Svein E. Seldal]

Hi

My suggestion on how the gcc compiler idealy should work, would look 
like this:

- Introduce a new attribute that you may assign to the type definitions 
of pointers (please note, its not a part of the data storeage, but the 
type of the variable):

        char __attribute__((__progmemptr__)) * cptr;    // **or**
        char pm * cptr;

How the attribute "pm" or "attribute..." is written is really 
irrelevant. We only need the feature in itself. Personally I like the 
first one best, because it fits GCC better IMHO.

        #define PM __attribute__((__progmemptr__))

Note that the attribute is a part of the type:

        char * dptr = cptr;   // Invalid; cptr is a code pointer

For avr-libc this will lead to:

        void puts(const char * str);
        void puts_P(const char PM * str);

        puts( cptr );    // Invalid; cptr is a code pointer
        puts_P( cptr );  // Valid

And finally:

        int temp;
        cptr = &temp;   // Fail, because temp is not located in flash


- Introduce native, automatic accesses to the memoryspaces based on the 
types of the accesses:

        int temp;
        temp = *ptr;   // LPM's the contents of code-adr PTR


One of the major issue concerning implementing this into gcc is the fact 
that gcc has no current mechanisms for setting attributes that are a 
part of the type. GCC has attributes for functions and for data 
storeages, but not for types.

Comments?


Svein