Irix6 casting long double to double yields bogus results

Irix6 casting long double to double yields bogus results

[Kaveh R. Ghazi]

Alex,

I've been running some tests on irix6 floating point and noticed a
problem.  Casting from long double to double seems to be broken now.
Thoughts?

		Thanks,
		--Kaveh

%cat foo.c
#include <stdio.h>

int main()
{
  long double a = 1234.5678L, b = 1234.5678L;
  double c = 1234.5678;

  printf ("%f %Lf %f\n", (double) a, b, c);
  return 0;
}
%gcc -v
Reading specs from /var/tmp/gcc-testing/branch/build/gcc/specs
Configured with: ../egcc-3.3-CVS20030129/configure --verbose
  --enable-languages=c,objc,c++,f77,java
Thread model: single
gcc version 3.3 20030129 (prerelease)
%gcc -Wformat foo.c
%./a.out
1834810029.000000 1234.567800 1234.567800

Re: Irix6 casting long double to double yields bogus results

[Alexandre Oliva]

[-- Attachment #1: Type: text/plain, Size: 566 bytes --]

On Jan 29, 2003, "Kaveh R. Ghazi" <ghazi@caip.rutgers.edu> wrote:

> I've been running some tests on irix6 floating point and noticed a
> problem.  Casting from long double to double seems to be broken now.

I doubt it ever worked with software-emulated 128-bit long doubles :-(
I hate the convoluted logic we use to decide whether to declare the
__make_[sdt]f functions.  I'm checking this obvious patch in for now,
but it would be really nice to clean it up and consolidate the several
duplicate declarations into more logical conditions.  Permission to
proceed?


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: long-double-to-double.patch --]
[-- Type: text/x-patch, Size: 685 bytes --]

Index: gcc/ChangeLog
from  Alexandre Oliva  <aoliva@redhat.com>

	* config/fp-bit.h (__make_dp): Declare if TMODES.

Index: gcc/config/fp-bit.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/fp-bit.h,v
retrieving revision 1.9
diff -u -p -r1.9 fp-bit.h
--- gcc/config/fp-bit.h 26 Jan 2003 10:06:57 -0000 1.9
+++ gcc/config/fp-bit.h 30 Jan 2003 04:17:06 -0000
@@ -521,6 +521,7 @@ extern TFtype df_to_tf (DFtype);
 #endif /* ! FLOAT */
 
 #ifdef TMODES
+extern DFtype __make_dp (fp_class_type, unsigned int, int, UDItype);
 extern TFtype __make_tp (fp_class_type, unsigned int, int, UTItype);
 #ifdef TFLOAT
 #if defined(L_tf_to_sf)

[-- Attachment #3: Type: text/plain, Size: 289 bytes --]


-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                 aoliva@{redhat.com, gcc.gnu.org}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist                Professional serial bug killer

Re: Irix6 casting long double to double yields bogus results

[Kaveh R. Ghazi]

 > From: Alexandre Oliva <aoliva@redhat.com>
 > 
 > On Jan 29, 2003, "Kaveh R. Ghazi" <ghazi@caip.rutgers.edu> wrote:
 > 
 > > I've been running some tests on irix6 floating point and noticed a
 > > problem.  Casting from long double to double seems to be broken now.
 > 
 > I doubt it ever worked with software-emulated 128-bit long doubles :-(
 > I hate the convoluted logic we use to decide whether to declare the
 > __make_[sdt]f functions.  I'm checking this obvious patch in for now,
 > 
 > 	* config/fp-bit.h (__make_dp): Declare if TMODES.
 > 
 > RCS file: /cvs/gcc/gcc/gcc/config/fp-bit.h,v
 > retrieving revision 1.9
 > diff -u -p -r1.9 fp-bit.h
 > --- gcc/config/fp-bit.h 26 Jan 2003 10:06:57 -0000 1.9
 > +++ gcc/config/fp-bit.h 30 Jan 2003 04:17:06 -0000
 > @@ -521,6 +521,7 @@ extern TFtype df_to_tf (DFtype);
 >  #endif /* ! FLOAT */
 >  
 >  #ifdef TMODES
 > +extern DFtype __make_dp (fp_class_type, unsigned int, int, UDItype);
 >  extern TFtype __make_tp (fp_class_type, unsigned int, int, UTItype);
 >  #ifdef TFLOAT
 >  #if defined(L_tf_to_sf)


Ah ha!  This also fixes two warnings in tp-bit.c

tp-bit.c:1588: warning: implicit declaration of function `__make_dp'
tp-bit.c:1588: warning: implicit declaration of function `__make_dp'

(There are still two other tp-bit.c warnings remaining.)

tp-bit.c:1392: warning: no previous prototype for `__fixunstfsi'
tp-bit.c:1392: warning: no previous prototype for `__fixunstfsi'

Right now we don't turn on -Werror for target files like libgcc2.c
etc, but I'm going to eventually turn it on to catch things like
this... :-)

Thanks for the quick fix!

		--Kaveh
--
Kaveh R. Ghazi			ghazi@caip.rutgers.edu

Re: Irix6 casting long double to double yields bogus results

[Alexandre Oliva]

On Jan 30, 2003, "Kaveh R. Ghazi" <ghazi@caip.rutgers.edu> wrote:

> (There are still two other tp-bit.c warnings remaining.)

> tp-bit.c:1392: warning: no previous prototype for `__fixunstfsi'
> tp-bit.c:1392: warning: no previous prototype for `__fixunstfsi'

This should fix them.  Care to give them a round of testing?

--- fp-bit.c.~1.39.~	2003-01-26 08:06:25.000000000 -0200
+++ fp-bit.c	2003-02-03 11:24:39.000000000 -0200
@@ -1377,7 +1381,7 @@ float_to_si (FLO_type arg_a)
 #endif /* L_sf_to_si || L_df_to_si */
 
 #if defined(L_sf_to_usi) || defined(L_df_to_usi) || defined(L_tf_to_usi)
-#if defined US_SOFTWARE_GOFAST || defined(L_tf_to_usi)
+#if defined US_SOFTWARE_GOFAST
 /* While libgcc2.c defines its own __fixunssfsi and __fixunsdfsi routines,
    we also define them for GOFAST because the ones in libgcc2.c have the
    wrong names and I'd rather define these here and keep GOFAST CYG-LOC's


As for explaining what's going on: I was out of my mind when I forced
the function to be defined for L_tf_to_usi, GCC just doesn't use it.
The other variants are only used for GOFAST compatibility.  We might
as well go ahead and take out L_tf_to_usi entirely from Makefile.in
and config/fp-bit.[ch], but there's no good reason to do so, and some
day GCC might be able to use float_to_unsigned anyway, so we might as
well leave it there.

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                 aoliva@{redhat.com, gcc.gnu.org}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist                Professional serial bug killer