[sdcc-devel] [ sdcc-Bugs-1816470 ] miscalculation of literals

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

Bugs item #1816470, was opened at 2007-10-19 13:49
Message generated for change (Comment added) made by maartenbrock
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=1816470&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: C-Front End
Group: None
Status: Open
Resolution: None
Priority: 5
Private: No
Submitted By: Stas Sergeev (stsp)
>Assigned to: Maarten Brock (maartenbrock)
Summary: miscalculation of literals

Initial Comment:
Hi.

It seems sdcc is bad on calculating the literals.
The following program (attached) returns 0xf4 on
sdcc, and 0xf3 on gcc.

---
int main()
{
    return (unsigned char)(0x100 - 4000000 / 307200.0 + .5);
}
---

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

>Comment By: Maarten Brock (maartenbrock)
Date: 2007-10-21 12:27

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

I also thought in that direction, but 0.5 on 256 is too large an error for
the mantissa to be the culprit.

I turns out that SDCC inserted a cast to unsigned char before 0x100
turning it into 0. This is ok if both operands are integer types but not
when one is a float.

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

Comment By: Stas Sergeev (stsp)
Date: 2007-10-19 16:38

Message:
Logged In: YES 
user_id=501371
Originator: YES

Very unlikely because the actual result of that
calculation is 243.48, which is not even above
243.5. While 0xf4==244. Such a huge errors are
clearly unacceptable and have nothing to do with
the limited precision.
Also note that there were no run-time calculations,
no code dealing with floats was generated by sdcc.
That expression is evaluated entirely compile-time.
So the sizeof(float) that applies, is the one the
sdcc was compiled with. And since it was compiled
by gcc...

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

Comment By: Patryk (patryks)
Date: 2007-10-19 16:15

Message:
Logged In: YES 
user_id=1788180
Originator: NO

Maybe differences in floating point precision between SDCC and gcc? I mean
target's sizeof(float).

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

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=100599&aid=1816470&group_id=599