Re: query: pthread_cond_timedwait and FPU state loss.

[Stefan Hoffmeister]

: On Fri, 12 Oct 2001 17:27:03 +1000, address@hidden wrote:

>The pthread_cond_timedwait function currently loses the floating
>point control register state (linux/i386 port), 

Which version of glibc is that?

>This seems due to the use of setjmp/longjmp which do 
>not save the FPU state

Eeew. This would imply that the SSE(2) control word state is not safe,
either?

>Should an application expect that the C library could modify the
>floating point control register settings?  The application could save
>and restore the settings across calls to any code linked to the C
>library.  Or is this a bug in the C library?

The C library should transparently handle the FPU control word (IOW, it
shouldn't alter it at all). If it doesn't do that, then IMHO this is a
defect.

I have verified this behaviour with glibc 2.2.2 of SuSE 7.2, BTW.