[sdcc-devel] [ sdcc-Bugs-542397 ] Compiler produces bad 'add a,a'.

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

Bugs item #542397, was opened at 2002-04-11 10:36
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=542397&group_id=599

Category: msc51(8051) target
>Group: fixed
>Status: Closed
Resolution: Fixed
Priority: 5
Submitted By: Nobody/Anonymous (nobody)
>Assigned to: Bernhard Held (bernhardheld)
Summary: Compiler produces bad 'add a,a'.

Initial Comment:
Hello!

When compiling the file src.c with the following 
contents, is can not be assembled, because the 
compiler produced the forbidden instruction
"add a,a" (instead of "add a,acc").

The assembler says:

?ASxxxx-Error-<a> in line 287 of src.asm
              <a> machine specific addressing or 
addressing mode error
removing src.rel

Is is also interesting, that the asm code produced by 
the compiler also contains a valid instruction "add 
a,acc" just a few lines above the "add a,a"!

src.c:

#include <at89s8252.h>

char foo(char x_byte)
  {
    SBUF= x_byte;
    while (!TI);
    TI= 0;

    return(x_byte);
  }

void main(void)
  {
    foo('0'+TI*2+RI);
  }

Compiler is SDCC V2.3.0, Windows binaries.

Attached files (in ZIP) are: src.c and src.asm

Greetings,
Patrick Gehriger

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

>Comment By: Bernhard Held (bernhardheld)
Date: 2003-01-04 10:50

Message:
Logged In: YES 
user_id=203539

Fixed by Wim in 2.3.2 

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

Comment By: Wim Lewis (wiml)
Date: 2002-09-19 02:40

Message:
Logged In: YES 
user_id=110010

I went through the 8051 register allocator & code generator
the other day and found where this was happening. I've
committed the change to CVS.

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

Comment By: Wim Lewis (wiml)
Date: 2002-09-14 09:07

Message:
Logged In: YES 
user_id=110010

Fixing sdcc to generate syntactically valid assembly for
this is easy (I will submit a patch). However, this unmasks
a deeper problem; the assembly output is incorrect and
passes the wrong value to foo().

The problem is the expression ('0'+TI*2+RI) when TI and RI
are sbit variables. The aopGet(AOP_CRY) emits some code
which puts the value of the bit register into the
accumulator. However, the register allocator doesn't seem to
know that this will happen, and has stored an intermediate
result in the accumulator as well. The intermediate result
gets clobbered by the value fetched from RI.

I don't think I understand sdcc's register allocator well
enough to fix this.

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

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=542397&group_id=599