float->double->float conversion with 19991122 snapshot (Linux/Intel)

[Jeffrey A Law]

  In message < 19991202005800.52263@atrey.karlin.mff.cuni.cz >you write:
  > > 
  > > > float
  > > > fdf (float xx)
  > > > {
  > > >   double x = xx;
  > > >   return x;
  > > > }
  > > 
  > > What an amazing mess!
  > > Offhand it looks i386-specific.  Sparc generates a reasonable program.
  > Really cool code :)
  > But it is quite easy to fix. The expander was forcing to reg the wrong regi
  > ster..
  > Interesitng that it didn't showed up in the testsuite.
  > 
  > OK to install?
  > 
  > Thu Dec  2 00:57:00 MET 1999  Jan Hubicka  <hubicka@freesoft.cz>
  > 	* i386.md (extend?f?f2): Force to register input not the output registe
  > r.
I went ahead and installed this with the fixed ChangeLog entry.

Jan -- when you install patches, you should use the current date for the
ChangeLog entry, not the date you submitted the patches.

ie, you checked these patches today, not friday:

Fri Nov 26 10:59:12 CET 1999  Jan Hubicka  <hubicka@freesoft.cz>

        * i386.md (addsi3_cc): Add "binary_operator_ok" to the condition.
        (addsi3_carry): Likewise.
        (sbbsi3_cc): Add "binary_operator_ok" to the condition.
        (sbbsi3_carry): Likewise.
        (mulsi3): Rewrite to expander, ensure that only one operand is memory.
        (mulhi3): Likewise.
        (test?i_1): Ensure that only one operand is memory.
        (conditional move patterns): likewise.
[ ... ]


You don't need to go back and fix the date on the old ChangeLog entries, but
please make sure to use the right date for new entries.
jeff