X-Authentication-Warning: delorie.com: mail set sender to djgpp-workers-bounces using -f From: Kbwms AT aol DOT com Message-ID: <1a5.19d80d7e.2caaeac4@aol.com> Date: Tue, 30 Sep 2003 10:18:44 EDT Subject: Re: Bug in modfl() in libc.a To: djgpp-workers AT delorie DOT com MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="part1_1a5.19d80d7e.2caaeac4_boundary" X-Mailer: 8.0 for Windows sub 6015 Reply-To: djgpp-workers AT delorie DOT com --part1_1a5.19d80d7e.2caaeac4_boundary Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit In a message dated 9/29/2003 2:14:47 PM Eastern Standard Time, peuha AT cc DOT helsinki DOT fi writes: > >modf(+-x, iptr) returns a result with the same sign as x. > >modf(+-Inf , iptr) returns +-0 and stores +-Inf in the object pointed to by > >iptr. > >modf(NaN, iptr) stores a NaN in the object pointed to by iptr (and returns > a > >NaN). > [snip] > >My interpretation is that the result returned for +-Inf should have the > same > >sign, i.e.: +0 for +Inf, -0 for -Inf. > > My previous patch would also return +0 for negative finite integers and > NaNs. This should do better (but it's completely untested): > > Index: modfl.S > =================================================================== > RCS file: /cvs/djgpp/djgpp/src/libc/ansi/math/modfl.S,v > retrieving revision 1.1 > diff -c -r1.1 modfl.S > *** modfl.S 25 Nov 1994 19:59:40 -0000 1.1 > --- modfl.S 29 Sep 2003 10:08:25 -0000 > *************** > *** 17,22 **** > --- 17,40 ---- > fldcw -2(%ebp) > fld %st(0) > fstpt (%eax) > + fucom %st(1) > + fnstsw -2(%ebp) > + movw -2(%ebp),%dx > + andb $0x44,%dh > + xorb $0x40,%dh > + jnz 0f > + fxam > + fnstsw -2(%ebp) > + fstp %st(0) > + fstp %st(0) > + fldz > + movw -2(%ebp),%dx > + andb $0x02,%dh > + jz 1f > + fchs > + jmp 1f > + 0: > fsubrp %st,%st(1) > + 1: > leave > ret > > Here is the code as patched. It causes a SIGSEGV signal and General Protection Fault when executed. KB Williams --part1_1a5.19d80d7e.2caaeac4_boundary Content-Type: text/html; charset="US-ASCII" Content-Transfer-Encoding: quoted-printable In a message dated 9/29/2003 2:14:47 PM Eastern Standard= Time, peuha AT cc DOT helsinki DOT fi writes:

>modf(+-x, iptr) returns a result with the same sign as=20= x.
>modf(+-Inf , iptr) returns +-0 and stores +-Inf in the object pointed to= by
>iptr.
>modf(NaN, iptr) stores a NaN in the object pointed to by iptr (and retur= ns a
>NaN).
[snip]
>My interpretation is that the result returned for +-Inf should have the=20= same
>sign, i.e.: +0 for +Inf, -0 for -Inf.

My previous patch would also return +0 for negative finite integers and
NaNs.  This should do better (but it's completely untested):

Index: modfl.S
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
RCS file: /cvs/djgpp/djgpp/src/libc/ansi/math/modfl.S,v
retrieving revision 1.1
diff -c -r1.1 modfl.S
*** modfl.S   25 Nov 1994 19:59:40 -0000   1.1
--- modfl.S   29 Sep 2003 10:08:25 -0000
***************
*** 17,22 ****
--- 17,40 ----
   fldcw -2(%ebp)
   fld %st(0)
   fstpt (%eax)
+   fucom %st(1)
+   fnstsw -2(%ebp)
+   movw -2(%ebp),%dx
+   andb $0x44,%dh
+   xorb $0x40,%dh
+   jnz 0f
+   fxam
+   fnstsw -2(%ebp)
+   fstp %st(0)
+   fstp %st(0)
+   fldz
+   movw -2(%ebp),%dx
+   andb $0x02,%dh
+   jz 1f
+   fchs
+   jmp 1f
+ 0:
   fsubrp %st,%st(1)
+ 1:
   leave
   ret



Here is the code as patched.  It causes a SIGSEGV signal and General Pr= otection Fault when executed.


KB Williams --part1_1a5.19d80d7e.2caaeac4_boundary--