Bug#137020: acknowledged by developer (Bug#137020: fixed in glibc 2.3.1-15)

[GOTO Masanori <gotom@debian.or.jp>]

At Fri, 28 Mar 2003 14:54:24 +0100,
Vincent Lefevre wrote:
> 
> On Fri, Mar 28, 2003 at 22:04:26 +0900, GOTO Masanori wrote:
> > I tested as:
> > 
> > #include <stdio.h>
> > #include <fpu_control.h>
> > 
> > main()
> > {
> >         unsigned int a;
> > 
> >         _FPU_GETCW(a);
> >         _FPU_SETCW(a);
> >         printf("%x\n",a);
> > }
> > 
> > Original bug does not say "_FPU_RC_ZERO".  You _have to_ send a
> > test program to report BTS in the next time.
> 
> OK. But note that I did (very shortly) explain the problem in my
> original bug report.
> 
> On Fri, Mar 28, 2003 at 22:19:23 +0900, GOTO Masanori wrote:
> > ... and this bug is really fixed in glibc 2.3.2-1.
> 
> I haven't seen glibc 2.3.2-1.

Ah, I should say:

	"this bug will be fixed in glibc 2.3.2-1 which will be
	 released as the next version of the current glibc 2.3.1-16"

> When you say "fixed", did you test with _FPU_RC_ZERO only or did you
> check the code?

Yes.

> In particular, the following test program would have failed too
> with glibc 2.3.1:
> 
> #include <stdio.h>
> #include <fpu_control.h>
> 
> main()
> {
>   unsigned int d;
> 
>   _FPU_GETCW(d);
>   _FPU_SETCW(d);
>   printf("%x\n",d);
> }
> 
> (This is your test program where 'a' has been replaced by 'd'.)

This problem is caused by fpu_control.h with incomplete definition of 
_FPU_GETCW, _FPU_SETCW.  Rewrite your /usr/include/fpu_control.h
as follows (it's pulled from glibc 2.3.2):

/* Macros for accessing the hardware control word.  */
#define _FPU_GETCW(__cw) ( { \						<-
  union { double d; fpu_control_t cw[2]; } \
    tmp __attribute__ ((__aligned__(8))); \
  __asm__ ("mffs 0; stfd%U0 0,%0" : "=m" (tmp.d) : : "fr0"); \
  (__cw)=tmp.cw[1]; \							<-
  tmp.cw[1]; } )	
#define _FPU_SETCW(__cw) { \						<-
  union { double d; fpu_control_t cw[2]; } \
    tmp __attribute__ ((__aligned__(8))); \
  tmp.cw[0] = 0xFFF80000; /* More-or-less arbitrary; this is a QNaN. */ \
  tmp.cw[1] = __cw; \							<-
  __asm__ ("lfd%U0 0,%0; mtfsf 255,0" : : "m" (tmp.d) : "fr0"); \
}

The point is "__cw".  Please check.

Regards,
-- gotom