[Sdcc-user] Memory type and data locations

[Russel W. <ru...@ru...>]

By way of Bug Report 454266, Toni R=E4s=E4nen and myself have been having a
short discussion about the way in which SDCC deals with array data
initialization.  I had thought that was the wrong forum for the debate
and Toni stated this explicitly, so I am writing this email.

Discussion started from the following program:

    char xdata x [] =3D=20
    {
      'a', 'b', 'c'
    };
   =20
    int main(){
    }
   =20
which generates the code:

   000D 90s00r00             79 	mov	dptr,#_x
   0010 74 61                80 	mov	a,#0x61
   0012 F0                   81 	movx	@dptr,a
   0013 90s00r01             82 	mov	dptr,#(_x + 0x0001)
   0016 74 62                83 	mov	a,#0x62
   0018 F0                   84 	movx	@dptr,a
   0019 90s00r02             85 	mov	dptr,#(_x + 0x0002)
   001C 74 63                86 	mov	a,#0x63
   001E F0                   87 	movx	@dptr,a

I was making the point that to use assignment to initialize xdata seems
unreasonable since the data values and the code have to be stored in
code memory and then transferred to xdata whereas using the Intel Hex
Dowload file (or S-Record dowload file) the values could be initialized
without recourse to code.

If the array is put into the code segment, or made const, then the
compiler uses initialization rather than assignment but puts the
information in the code segment rather than xdata.

The problem I have is that the 8051 based system I am involved with has
RAM, EEPROM and ROM in the xdata space.  Clearly I cannot use assignemnt
for the ROM which is where most of the xdata is.  ROM here is not real
ROM but flash silicon so it is actually OTP memory.  This means it is
entirely appropriate for the code to hold the initialization data (as
opposed to doing separate creation as would be done for ROM) but it
seems SDCC has no mechanisms for dealing with the fact that xdata is not
of homogeneous form, i.e. not all RAM.

I hope Toni sees this as a fair restatement of the problem.  Now it is
really open for debate.  Is this a problem unique to me or is it
something others are having problems with?  Have others found
techniques?  (I was wodnering whether some form of linker script
mechanism =E0 la GCC would be appropriate.)

--=20
Russel.
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
Dr Russel Winder               +44 20 7585 2200
41 Buckmaster Road             +44 7770 465 077
London SW11 1EN, UK            ru...@ru...