fabs() intrinsic totally broken in P4, as of 2002-02-11

[Richard Henderson]

On Sun, Feb 17, 2002 at 05:35:23PM +0100, Jan Hubicka wrote:
> The drawback is that we can't represent the constant as FP value as it would
> get promptly normalized,

Err, would it?  ~(-0.0) is a NaN.  Oh, I see, real.c is somewhat
insistent about producing canonical QNaNs.  That's irritating.

> Do you see any way out?

Using unspecs to hide the pieces of the constant generation.
Ugly, but...


r~