RE: [Sdcc-user] The Z80 port, optimizations

[David H. <ac...@lo...>]

OK Michael,

You are right, that is quite a bit better looking code.  I do like the
stack initialization / finalization code for subroutines that are used
for this compiler (a big complaint with the *old* Hi-Tech compiler; 
   0000 DD E5                50         push    ix
   0002 DD 21 00 00          51         ld      ix,#0
   0006 DD 39                52         add     ix,sp

and the comparison between 'int's is pretty tight as well; 
   0063 FD 21r00s00         149         ld      iy,#_timing_j_1_1
	< here IX is the stack ptr. and we are comparing to a value on
the stack >
   0060 DD 7E 04            148         ld      a,4(ix)
   0067 FD 96 00            150         sub     a,0(iy)
   006A DD 7E 05            151         ld      a,5(ix)
   006D FD 9E 01            152         sbc     a,1(iy)
   0070 DAr87s00            153         jp      c,00111$

The compiler wisely used register colouring to get rid of one variable
(unsigned char), and allocated space for the other int.  I see from the
code that indeed this temporary variable gets put into the data segment,
something I like a lot better than using the stack for variables in
subroutines of real small stack systems:
 17     .area _DATA
   0000                      18 _timing_j_1_1:
   0000                      19         .ds 2

<BTW for readers seeing this thread anew, the 'C' code that this asm is
taken from is:

void timing( unsigned int timeval )
    {
    static unsigned int j;
    unsigned char i, k;
    for ( j = 1; j <= timeval; j++ )
        for ( i = 1; i <= 7; i++ ) {
            if ( interrupted )
                k = 1;
            }
    }				:-)	>
	
The question all this brings to mind is: Is there a way to control what
segment these temporary variables are assigned from?  As I can see, this
goes into the DATA segment.  And I can't think of a better area, but a
nagging doubt comes to mind that warns me that w/out overlaying some of
this data area, I could be getting into trouble.  I *know* there are
issues with this, and Paul was bringing up one solution, but if there
were a way to allow the coder to specify what data area could be safely
overlayed with another in subroutine calls, we *may* be able to save
data space allocation...


-David
Ni-Wumpf ltd.