[sdcc-devel] [ sdcc-Bugs-1505956 ] sjmp and conditional jump out of range

[SourceForge.net <no...@so...>]

Bugs item #1505956, was opened at 2006-06-14 13:04
Message generated for change (Comment added) made by maartenbrock
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=1505956&group_id=599

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: None
Group: None
Status: Open
Resolution: None
Priority: 5
Private: No
Submitted By: Frieder Ferlemann (frief)
Assigned to: Nobody/Anonymous (nobody)
Summary: sjmp and conditional jump out of range

Initial Comment:
The code below fails with:

?ASxxxx-Error-<a> in line 139 of sjmp.asm
              <a> machine specific addressing or
addressing mode error
?ASxxxx-Error-<a> in line 217 of sjmp.asm
              <a> machine specific addressing or
addressing mode error


Lines 139 and 217 contain jumps to labels which are not
within reach of the addressing mode:
        jc      00275$
        sjmp    00272$


-----------8<------------------------------------------
#define NOP1 do { _asm nop _endasm; } while (0)
#define NOP2 do { _asm nop nop _endasm; } while (0)
#define NOP4 do { _asm nop nop nop nop _endasm; } while (0)
#define NOP8  do {  NOP4;  NOP4; } while (0)
#define NOP16 do {  NOP8;  NOP8; } while (0)
#define NOP32 do { NOP16; NOP16; } while (0)
#define NOP64 do { NOP32; NOP32; } while (0)


volatile char t;

void main(void)
{
  volatile unsigned char i;

  for( i=0; i<100; i++ )
  {
    t++;
    NOP64;
    NOP32;
    NOP16;
    NOP8;
//    NOP4;
    NOP2;
//    NOP1;
  }

  t++;
}


----------------------------------------------------------------------

>Comment By: Maarten Brock (maartenbrock)
Date: 2008-01-19 20:04

Message:
Logged In: YES 
user_id=888171
Originator: NO

Looked into this again today. It would be easier if the assembler would
reject multiple instructions per line unless separated by a special
separator(e.g. '|'). Andif the preprocessor could insert this separator at
places where it finds a backslash in the macro while inside a _asm/_endasm
pair. Then the above NOP2 would have to be changed into either:
#define NOP2 do { _asm cpl a | cpl a _endasm; } while (0)
or:
#define NOP2 do { _asm \
            cpl a \
            cpl a \
        _endasm; } while (0)

Now what would be a good candidate for the separator? And how many people
rely already on the multiple instructions per line acceptance? Any idea is
welcome.

----------------------------------------------------------------------

Comment By: Maarten Brock (maartenbrock)
Date: 2007-08-14 22:53

Message:
Logged In: YES 
user_id=888171
Originator: NO

The problem here is that one cannot tell the preprocessor to put those
assembly instructions on seperate lines. So for NOP4 it generates:

nop nop nop nop

I expected the assembler to choke on this but it accepts it and generates
four 0x00 bytes correctly.

However the peephole optimizer only tries to read one instruction per line
when calculating the instrcution size (see asmLineNodeFromLineNode() in
src/mcs51/main.c) and treats the second and third nop as operands. The last
one is ignored alltogether. And thus the calculated size for this line is 1
byte instead of 4.

The workaround is obvious: don't put two or more asm instructions in a
macro

#define NOP do { _asm nop _endasm; } while (0)
#define NOP2 do { NOP; NOP; } while (0)
#define NOP4 do { NOP2; NOP2; } while (0)

The real solution would be to skip the inst and its op's in
instructionSize() after the calculation and process the rest of the line.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=1505956&group_id=599