This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: fix for execute/950511-1 failure on ppc-eabi


> (reg/v:SI 83)
> and
> (addressof:SI (reg/v:DI 84) 82 0x30037480)
> don't qualify as rtx_equal_for_memref_p.
> 
> Note that the SET might be conditionally executed.

I was mistaking thinking that alias analysis had noticed that
(reg/v:SI 83) had been set to the addressof (aka canon_rtx
and reg_known_value).  I'll get the simulator running and
play with this some more.  It'd be nice to notice that an
addressoff doesn't conflict with a stack variable.  BTW,
I noticed that you changed things so that BLKmode addressofs
don't conflict.

  1) When does a BLKmode addressof get created?

  2) The routine tends to be very conservative regarding
     BLKmode references (your change invalidates the
     comment at the top:

       If XSIZE or YSIZE is zero, we do not know the amount of memory being
       referenced (the reference was BLKmode), so make the most pessimistic
       assumptions.

     ) ... I feel a little uneasy with the notion that they don't
     conflict in the case of addressofs.  I take it that conflicts
     of this type can only happen in illegal code?

-- John
-------------------------------------------------------------------------
|   Feith Systems  |   Voice: 1-215-646-8000  |  Email: john@feith.com  |
|    John Wehle    |     Fax: 1-215-540-5495  |                         |
-------------------------------------------------------------------------


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]