title: Cross-Compiler's access to memory

question:

 does CData contain code and data or only constant data?

answer:

 (Elizabeth Rather)
 CData is intended as code space.

 (Stephen Pelc)
 CDATA spaces refer to code areas, basically read/execute.
 IDATA refers to read/write space that is automagically
 initialised.
 UDATA refers to read/write space that is NOT initialised
 at cross compile time.

 The intention on Harvard targets (e.g. AVR and 8051) is
 that CDATA describes the type of the CODE area, and UDATA
 and IDATA refer to the RAM space. If you need a third
 address space you'll have to invent an xDATA name for
 yourself.

 In a machine with a single adress space all three types
 live in the same address space.

 (mlg)
 So the best interpretation for CData is: "code and constant data".

question:

 What memory access words assess what address spaces?

answer:

> "cross @" means "the version of @ used by host to write to its
> image of the target memory"
>
> "target @" means "the version of @ used by the target system"


 (Stephen Pelc)
 I have added P for perhaps, H for Harvard and S for
 single address space machines. In the MPE
 cross compilers, @C and !C exist in the compiler in all
 architectures, but are mapped onto @ and ! for single
 address space machines. The use of @ and ! to refer to
 code space (even if CDATA is selected) is IMHO at the very
 least an ambiguous condition for a Harvard target.

             Exists?    CData    IData     UData
 cross @       y        S        y         P
 cross !       y        S        y         P
 cross @C      H/P      H        n         n
 cross !C      H/P      H        n         n
 target @      y        S        y         y
 target !      y        S        y         y
 target @C     y        y        n         n
 target !C     H/P      P        n         n

 Note that some Harvard CPUs, e.g. Z8, do allow you
 to write to code space and hence !C and friends are
 useful in both host and target.

 (Elizabeth Rather)
 I agree with Stephen's interpretations

 (mlg's note: H means "yes for Harvard", S means "yes for single
 address space machines", H/P means "yes for Harvard, perhaps
 for single address space machines".)


question:

 On a Harvard processor, code and data memory may use
 the same range of addresses (e.g. $0000..$FFFF), but
 belong to different spaces.

 Does the cross-compiler need to map these two
 different spaces (code and data) onto a single development
 space?

answer:

 No. A SECTION describes the memory map of the TARGET, not
 where you put it during development on the HOST. The table
 above permits you full access to both Harvard spaces, since
 you must use @C and !C and friends to access data in CDATA
 space.

 The point about SECTION is that it allows you to describe
 several areas of each type of memory.

 [...]
 [To avoid confusion, remember that] the comma words ARE affected
 by the xDATA switching but that @ and ! are NOT. This is necessary
 so that you can build tables using CREATE, e.g.

 CDATA CREATE MyTable  \ -- codeaddr
   5 c, 7 c, ...
 IDATA CREATE ATable   \ -- dataaddr
   7 c, 9 c, ...

 (Elizabeth Rather)
 I agree with Stephen's interpretations