[sdcc-devel] duplicate assignments to pointers

[Scott D. <sc...@da...>]

I came across a bizarre behavior that I'm having trouble reconciling.
Consider this innocent piece of code:

/* pointer to char arithmetic */

void pc_add(void)
{

  if(*cP0)
    failures++;

  *cP0 = *cP0 + 1;
  if(*cP0 != 1)
    failures++;

}

All it does is check to see if a the value to which cP0 points is zero or
not and then increments that value and checks to see if the increment was
successful. This is part of a (yet un-commit'd PIC regression test).

When compiled for the PIC port, I noticed that the pointer cPO is
referenced in two consecutive iCode assignments. For example, in the
dumpraw0 debug log (obtained with --dumpall) I notice:

----------------------------------------------------------------
a.c(14:0:11:0:0)	 _iffalse_0($2) :
a.c(16:0:12:0:0)		iTemp5 [k9 lr0:0 so:0]{ ia1 re0 rm0
nos0}{char * generic } = @[_cP0 [k2 lr0:0 so:0]{ ia1 re0 rm0 nos0}{char *
generic }]
a.c(16:0:13:0:0)		iTemp6 [k10 lr0:0 so:0]{ ia1 re0 rm0
nos0}{char} = @[_cP0 [k2 lr0:0 so:0]{ ia1 re0 rm0 nos0}{char * generic }]

In other words, _cP0 is copied to iTemp5 AND iTemp6. The only difference
is that iTemp5 is a "char * generic" while iTemp6 is a "char".

So I thought this was perhaps just some carry over from an 8051
idiosynchracy. But if you look at the mcs51 assembly output you see this:

;	a.c 16
	mov	r2,_cP0
	mov	r3,(_cP0 + 1)
	mov	r4,(_cP0 + 2)
	mov	dpl,_cP0
	mov	dph,(_cP0 + 1)
	mov	b,(_cP0 + 2)
	lcall	__gptrget
; Peephole 185   changed order of increment (acc incremented also!)
	inc  a
; Peephole 191   removed redundant mov
	mov  r5,a
	mov  dpl,r2
	mov  dph,r3
	mov  b,r4
	lcall	__gptrput


I suppose this code is perfectly correct. But why bother stashing _cP0
into r2,r3, r4, AND dpl, dph, b. Isn't it adequate to simply reload
dpl,dph,b with _cP0 for the call to __gptrput? (Incidently, gptrput and
gptrget do not affect the contents of dpl and dph; thus the second loading
of dpl and dph is not necessary).

Scott