[sdcc-devel] [ sdcc-Bugs-1028702 ] PIC16: shift left fails

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

Bugs item #1028702, was opened at 2004-09-15 20:36
Message generated for change (Comment added) made by vrokas
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=1028702&group_id=599

Category: pic16 target
>Group: fixed
>Status: Closed
>Resolution: Fixed
Priority: 5
Submitted By: ccsporters (tecodev)
>Assigned to: Vangelis Rokas (vrokas)
Summary: PIC16: shift left fails

Initial Comment:
The following piece of code fails:

#pragma stack 0x100 60
char test(char target) {
  return (1 << target);
}

int main() {
  return test(4);
}

This is compiled into:
_test:
#LINE   2; shifttest.c  char test(char target) {
        MOVFF   FSR2L, POSTDEC1
        MOVFF   FSR1L, FSR2L
        MOVFF   r0x2E, POSTDEC1
[1]     MOVWF   r0x2E
#LINE   3; shifttest.c  return (1 << target);
[2]     MOVF    0x01, W
[3]     MOVWF   r0x2E
        COMF    r0x2E, W
        RRCF    r0x2E, F

First problem (liveranges?):
[1]: Argument "target" is placed in r0x2E,
[3]: literal (!) 0x01 overwrites our argument...

Second problem:
[2]: a literal 0x01 is to be stored in WREG, but
register 0x01 is copied instead
This problem might be caused by gen.c:8603:
pic16_emitpcode(POC_MOVFW,
        pic16_popGet(AOP(left),offset));
By replacing POC_MOVFW with POC_MOVLW this second
problem disappears. (But if this is correct in general,
we do not know...)

Everything done with SDCC pic16 2.4.4 #840 (Sep 13
2004) (UNIX).

Greetings,
        Raphael Neider & Daniel Winkler


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

>Comment By: Vangelis Rokas (vrokas)
Date: 2004-09-17 03:37

Message:
Logged In: YES 
user_id=770505

Bug fixed in SDCC v.2.4.4 #843 .
Code generator function for shift left did not inherently
support cases where result and right (shift count) shared
the same registers. Problem solved by modifying register
allocator to always allocate different registers for result
and right (and possibly left... but this needs extra
optimization, work for the future! ... )


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

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=1028702&group_id=599