[Sdcc-user] Peephole 132 ...

[Michael S. <msc...@ba...>]

I had just traped into the following problem with the peephole 132

i have a source that looks like that

switch( a )
case a1: do something break;
	switch( b )
	case b1:  do something break;
	case B2:  do something break;
	case b3:  do something break;
	break; // no default ! 
case a2: do something break;
case a3: do something break;
case a4: do something break;

now my problem was that case b1: was never executed but should be.
it took my hours of hacking, debugging and hacking ...
until i looked into the lst file ..

   0750 90s00r00           2198 	mov	dptr,#_HostCmd
   0753 E0                 2199 	movx	a,@dptr
   0754 FA                 2200 	mov	r2,a
   0755 8A*00              2201 	mov	_HostCommand_tmp_uchar_1_1,r2
                    0757   2202 	C$wsi_128kflash.c$712$2$2 ==.
                           2203 ;	wsi_128kflash.c 712
   0757 C2 95              2204 	clr	_P1_5
                    0759   2205 	C$wsi_128kflash.c$713$2$2 ==.
                           2206 ;	wsi_128kflash.c 713
   0759 D2 95              2207 	setb	_P1_5
                    075B   2208 	C$wsi_128kflash.c$714$2$2 ==.
                           2209 ;	wsi_128kflash.c 714
   075B BA 41 03           2210 	cjne	r2,#0x41,00417$
   075E 02s09rCC           2211 	ljmp	00210$
   0761                    2212 00417$:
   0761 BA 42 02           2213 	cjne	r2,#0x42,00418$
                           2214 ; Peephole 132   changed ljmp to sjmp
   0764 80 81              2215 	sjmp 00179$
   0766                    2216 00418$:
   0766 BA 43 03           2217 	cjne	r2,#0x43,00419$
   0769 02s08r45           2218 	ljmp	00183$

some lines later ...

   07DE                    2276 00438$:
   07DE BA 57 03           2277 	cjne	r2,#0x57,00439$
   07E1 02s09rCC           2278 	ljmp	00210$
   07E4                    2279 00439$:
   07E4 02s09rCF           2280 	ljmp	00222$
                    07E7   2281 	C$wsi_128kflash.c$724$3$17 ==.
                           2282 ;	wsi_128kflash.c 724
   07E7                    2283 00179$:
   07E7 C2 95              2284 	clr	_P1_5
                    07E9   2285 	C$wsi_128kflash.c$725$3$17 ==.
                           2286 ;	wsi_128kflash.c 725

now line 2215 is the above case b1 and the lst file looks like the cases
are re-ordered.
the first thing i noticed was that for this case a peephole has changed
ljmp to sjmp
my first idea was that this could be the reason and the instruction is
wrong.

so i modified my sources a bit

switch( a )
case a1: do something break;
	switch( b )
-->	default : waste time; break; // no need for that
	case b1:  do something break;
	case B2:  do something break;
	case b3:  do something break;
	break; // no default ! 
case a2: do something break;

notice that default case before the not executed case.

with this extra, not needed, code my sources works as expected the case
is executed.
I quick view into the lst file shows that the sjmp is now a ljmp.

so could it be that 
  0764 80 81              2215 	sjmp 00179$
  07E7                    2283 00179$:
is wrong ?
0x7E7-0x764 - 2byte code = 0x81 
but doesn't 0x81 mean a jump backwards to 0x6E3 or so ? this would
explain something ?
i am not shure but i thought that it is only possible to do a short jump
-128 .. +127
and 0x81 is negative ....


Ah, one last thing to mention. The sdcc-built is the same as the one i
have sent to the OKR. i made it from the cvs sources august 28. But
earlier versions also showed that.

Dipl.-Ing. (FH) Michael Schmitt
Baumer Ident GmbH
Entwicklung / Development Department
Hertzstr. 10
D-69469 Weinheim
Deutschland / Germany
Tel. +49 (0) 6201 9957 - 30
Fax. +49 (0) 6201 9957 - 99
E-Mail : msc...@ba...
Web: http://www.baumerident.com