[sdcc-devel] [ sdcc-Bugs-1757671 ] Floating point comparison problem.

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

Bugs item #1757671, was opened at 2007-07-20 18:40
Message generated for change (Comment added) made by maartenbrock
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=1757671&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: z80 port
Group: None
>Status: Closed
>Resolution: Fixed
Priority: 5
Private: No
Submitted By: Philipp Krause (spth)
>Assigned to: Maarten Brock (maartenbrock)
Summary: Floating point comparison problem.

Initial Comment:
I have a problem in the following function:

struct c
{
	float r;
	float i;
};


void clamp(struct c *l)
{
	if(l->r > 240.0f)
		l->r = 240.0f;
	if(l->r < 1.0f)
		l->r = 1.0f;
}

The function seems to work for l->r > 0, but the results are wrong for l->r < 0.

strangely

void clamp(struct c *l)
{
	if(l->r < 1.0f)
		l->r = 1.0f;
}

works as it should.

Philipp

I compiled using
sdcc --std-c99 -mz80

sdcc --version
SDCC : mcs51/gbz80/z80/avr/ds390/pic16/pic14/TININative/xa51/ds400/hc08 2.7.2 #4861 (Jun 26 2007) (UNIX)


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

>Comment By: Maarten Brock (maartenbrock)
Date: 2010-08-08 14:54

Message:
The test passes on test-z80, but then other regression tests fail. I'll
file another bug report for that.

I've fixed the ucsim z80 simulator to set the flags correctly and fixed
src/z80/gen.c in SDCC 2.9.7 #5918.

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

Comment By: Philipp Krause (spth)
Date: 2010-08-07 17:55

Message:
There's always the old simulator: Do a make test-z80 instead of make
test-ucz80. IMO the old simulator is much less buggy.

Philipp


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

Comment By: Maarten Brock (maartenbrock)
Date: 2010-08-07 16:00

Message:
If I add the following code below the release label in genCmp() it still
fails to compare (long)0xBF80 0000 <= (long)0x3F80 0000 correctly in the
regression test.

release:
  tlbl = newiTempLabel (NULL);
  /* check for overflow */
  emit2 ("jp PO,!tlabel", tlbl->key + 100);
  emit2 ("xor a,!immedbyte", 0x80);
  emitLabel (tlbl->key + 100);

I'm beginning to doubt whether the simulator sets the P/V flag incorrectly
for this overflow  or that I misunderstand the Z80 behaviour.

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

Comment By: Maarten Brock (maartenbrock)
Date: 2010-08-07 12:00

Message:
The problem is not in the library but in the z80 code generation for signed
comparison.

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

Comment By: Philipp Krause (spth)
Date: 2007-07-20 18:52

Message:
Logged In: YES 
user_id=564030
Originator: YES

It seems the f(l->r < 1.0f) is correct, but sdcc generates code that
thinks that (l->r > 240.0f) is true for negative l->r.


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

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=1757671&group_id=599