[PATCH, libiberty] Fix (biased) exponent = 0 case in floatformat_to_double

[Julian Brown <julian@codesourcery.com>]

[-- Attachment #1: Type: text/plain, Size: 1049 bytes --]

Hi,

This patch fixes a case in floatformat_to_double, as used in binutils 
for disassembling floats for m68k (and soon ARM also). The failure mode 
was as follows: the exponent was tested for zero after biasing, which is 
true for floating point values 1.0 <= value < 2.0. Such values took a 
code path intended for denormals, with the effect of causing them to 
always produce the value 1.0 (or, perhaps, very close to 1.0).

This patch only processes values as denormal when special_exponent is true.

Tested with "make check" on CSL's current binutils branch, with targets 
arm-none-eabi and --enable-targets=all (the ARM version includes a new 
testcase which exercises some floating-point values).

The patch applies cleanly on GCC head's libiberty too, though I've not 
explicitly tested it there. I suspect it'll also apply on binutils head.

OK to apply on src and gcc head, or is there any more testing I should do?

Cheers,

Julian

ChangeLog (libiberty):

     * floatformat.c (floatformat_to_double): Fix (biased) exponent=0
     case.

[-- Attachment #2: libiberty-floatformat-1 --]
[-- Type: text/plain, Size: 841 bytes --]

Index: floatformat.c
===================================================================
RCS file: /cvs/src/src/libiberty/floatformat.c,v
retrieving revision 1.19
diff -c -p -r1.19 floatformat.c
*** floatformat.c	12 Feb 2006 15:54:25 -0000	1.19
--- floatformat.c	24 Apr 2006 17:47:42 -0000
*************** floatformat_to_double (const struct floa
*** 394,400 ****
  
        /* Handle denormalized numbers.  FIXME: What should we do for
  	 non-IEEE formats?  */
!       if (exponent == 0 && mant != 0)
  	dto += ldexp ((double)mant,
  		      (- fmt->exp_bias
  		       - mant_bits
--- 394,400 ----
  
        /* Handle denormalized numbers.  FIXME: What should we do for
  	 non-IEEE formats?  */
!       if (special_exponent && exponent == 0 && mant != 0)
  	dto += ldexp ((double)mant,
  		      (- fmt->exp_bias
  		       - mant_bits