[Bug debug/48928] [4.7 Regression] ICE: in decimal_to_decnumber, at dfp.c:113 with -O -g and decimal float

[joseph at codesourcery dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48928

--- Comment #3 from joseph at codesourcery dot com <joseph at codesourcery dot com> 2011-05-09 09:56:56 UTC ---
On Mon, 9 May 2011, jakub at gcc dot gnu.org wrote:

> Ugh, dfp is complete mess, in many places in the folder, middle-end and
> optimizers dconst{1,2,m1,half} are used even for decimal types/modes, but those
> real formats are binary, not decimal.  The following patch just accepts the
> status quo and handles those 4 standard constants specially instead of ICEing
> on them.

There should probably be a separate PR filed for the underlying problem of 
using those values for decimal floating-point - given that all these 
constants have multiple decimal floating-point representations (with 
different quantum exponents), a decimal floating-point expert ought to 
look at all the uses to figure out what is actually correct in terms of 
getting the right quantum for results, and I doubt the code gets this 
right at present.