On Mon, Sep 24, 2007 at 04:05:45PM +0200, Andreas Schwab wrote: > Daniel Jacobowitz <[EMAIL PROTECTED]> writes: > > > Glibc's test-float failed on my qemu testing. I tracked it down to > > these routines: if you count the bits carefully, you'll see that > > 0x7FC00000 sets the quiet NaN bit (on most hardware - signalling NaN > > in the MIPS case); so does a.high >> 41, which copies it from the > > original NaN. I think this routine should not force a quiet or > > signalling NaN, but just preserve the input NaN's signalling-ness. > > You may need to make sure that at least one mantissa bit is set.
I think - but am not quite sure - that this will already be the case because the NaN will have come from some other NaN representation. Good point, though - we can easily set the next bit, as long as we don't fiddle with the signalling bit. -- Daniel Jacobowitz CodeSourcery