Re: [sdcc-devel] More on pointers

[Sandeep D. <sa...@dd...>]

Michael,

Welcome to the wonderful world of register packing.. this is
a term I use to describe the phase for reducing temporaries (in a 
processor specific manner)..think of them as peephole
optimizations on the intermediate code..in your case for example
the iCode sequence would probably look something like

itempX[b c] := iTempY[d e];
iTempX[b c] = iTempX[b c] + 1;
iTempY[d e] := iTempX[b c];

Here register packing would reduce the code to
iTempY := iTempY + 1;

if there is no other usage of iTempX.. take a look
@ packRegistersForAssign in mcs51/gen.c.. in a tree walking
code generator (like burg - as in lcc) this phase would
have been much simpler ...

Couple of important things here to determie usage & definitions
the MACRO ..OP_DEFS & OP_USES yeild the definition & usage
bitvectors respectively.. bitVectnBitsOn(OP_DEFS(operand type*))
will yield number of definitions for this iTemp..similarly
bitVectnBitsOn(OP_USES(operand type*)) will give you the number
of uses.. to check if an operand is used in an iCode or not
if (bitVectBitValue(OP_USES(operand type*),iCode type*->key))

then there are several other useful things you could do with these
bit vectors... to see if two operands are used in the same icode
if (bitVectBitsInCommon(OP_USES(op1),OP_USES(op2));

Hope this information helps..

Sandeep
> 
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> It's messier than I thought :)  The first optimisation seems to be pointer
> fetch/inc dependent.  This code:
> 
> Is:                                             Should be:
> - ----------------------------------------      --------------------
> ; genAssign
>         ld      -4(ix),e        ; tmp = d       -4(ix) = de
>         ld      -3(ix),d
> ; genPlus
> ; genPlusIncr
>         ld      c,e                             inc de
>         ld      b,d                             (ie really bad !)
>         inc     bc
> ; genAssign
>         ld      e,c
>         ld      d,b
> ; genAssign
>         ld      l,-2(ix) ; 0                    hl = -2(ix)
>         ld      h,-1(ix) ; 1
> ; genPlus
> ; genPlusIncr
>         ld      c,-2(ix) ; 0                    inc -2(ix)
>         ld      b,-1(ix) ; 1                    jr
>         inc     bc                              inc -1(ix)
> ; genAssign
>         ld      -2(ix),c
>         ld      -1(ix),b
> ; genPointerGet
>         ld      a,(hl)                          a = (hl)
>         ld      c,a
> ; genAssign (pointer)
>         ld      l,-4(ix) ; 0                    hl = -4(ix)
>         ld      h,-3(ix) ; 1
>         ld      (hl),c                          ld (hl),a
> ; genGoto
>         jp      00101$                          jp
> 
> is generated from part of a standard memcpy:
>         *d++ = *s++
> 
> - -- Michael
> 
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.0.0 (GNU/Linux)
> Comment: For info see http://www.gnupg.org
> 
> iD8DBQE4o5qLUejL3SuzxEgRAkYnAJwJT9BQU4nj1QbS/BGzqJPPHbxieACeKUrm
> CBfr/S6MAFK+fDBkDUzFAlA=
> =ER8C
> -----END PGP SIGNATURE-----
> 
> _______________________________________________
> sdcc-devel mailing list
> sdc...@li...
> http://lists.sourceforge.net/mailman/listinfo/sdcc-devel

-- 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Sandeep Dutta                        Phone:  650-356-5417
Diab-SDS                             Fax:    650-356-5490
323 Vintage Park Drive               Email:  sa...@dd...
Foster City, CA  94404               URL:    www.ddi.com    
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~