[Bug target/51345] [avr] Devices with 8-bit SP need their own multilib(s)

["gjl at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51345

Georg-Johann Lay <gjl at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
         Resolution|FIXED                       |
   Target Milestone|4.7.0                       |4.7.1

--- Comment #6 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2012-03-24 21:06:42 UTC ---
Reopened:

The current (4.7.0) behaviour is:

-mtiny-stack acts as multilib-option for 16-bit SP targets in avr2 and avr25,
i.e. with that switch will trigger multi-lib from ./tiny-stack resp.
./avr25/tiny-stack

The problem is that the tiny-stack multilibs mix 16-bit SP targets that are
treated as 8-bit SP, and targets with 8-bit SP.

The currect implementation assumes SP.H = 0 in tiny-stack mlibs in order to
properly initialize the frame pointer, e.g. in __prologue_saves.

This is no good for 16-bit SP targets.

Solution:

Restore the behaviour for the 16-bit SP targets as was, i.e. -mtiny-stack does
not affect multilib selection for these targets, but if specified on the
command line, the resulting code will only change SP.L and assume SP.H never
changes.

For the 8-bit SP targets, -mtiny-stack should remain as in 4.7.0, i.e. used
internally to trigger tiny-stack multilibs. Specifying -mtiny-stack on the
command line will be redundant for these targets and have no effect.