[PATCH take 2] middle-end: provide generic implementation for isinf

[Joseph S. Myers]

On Fri, 15 Jun 2007, Kaveh R. GHAZI wrote:

> Although I'm calling this "take 2", this time I'm doing "isinf" instead of
> "finite".  However I intend to address as many of the fpclassify functions
> as possible in later patches if this approach is approved.  I followed
> rth's recommendation in PR target/30652 which was to convert isinf(x) to
> isgreater(fabs(x),MAX), where MAX is FLT_MAX, DBL_MAX or LDBL_MAX
> depending on the type of 'x'.  In order to grab those values, I moved the
> MAX generation code from c-cppbuiltin.c to real.c so that the middle-end
> could call it as well.

If this means that __builtin_isinf is now always expanded inline with an 
expansion appropriate to the type of its argument, this should fix PR 20558.

-- 
Joseph S. Myers
joseph@codesourcery.com