[patch] MIPS/gas: load/store address overflow on binutils 2.14

[Maciej W. Rozycki]

On Fri, 12 Sep 2003, Thiemo Seufer wrote:

> >  It looks like a good idea, except that I'd do sign extension differently
> > -- to let gcc know what we really mean.
> 
> Huh? gcc provides input for the assembler, it will never see what
> happens here.

 Well, gcc is definitely going to see gas code -- it won't hurt to try to
optimize it, either. 

> >  Like this:
> > 
> >   if (! dbl && (! (ep->X_add_number & ~(offsetT) 0xffffffff))
> > 		&& ! (-ep->X_add_number & ~(offsetT) 0xffffffff)))
> >     ep->X_add_number = (ep->X_add_number << 32) >> 32;
> 
> IIRC X_add_number is unsigned.

 Well, offsetT seems to be defined as signed: either "bfd_signed_vma", or,
for legacy code, "long".

> > This might need checking for BFD64 though.  The second check of
> > ep->X_add_number is needed to handle cases like "lw $2,-0xf0000000" (try
> > it out!). 
> 
> What's the meaning of a negative address?

 I'd like to see all valid numbers handled consistently.  While
"-0xf0000000" may look strange itself, a result of a more complicated
expression may also be negative and we should preserve the commutative
property of addition, i.e. we should not force people to reorder
operations to avoid unary minuses.

  Maciej

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +