[patch] Fix powerpc 64 alignment problem for lwa instruction

[Andrew Pinski]

> 
> 
> On Nov 7, 2006, at 12:56 PM, Andrew Pinski wrote:
> 
> >> We, at Apple, found a ppc64 code generation problem where lwa_operand
> >> routine didn't check the alignment of a memory operand being 32 bits
> >> aligned or not.
> >>
> >> The following patch has been tested on ppc MacOS with "make all", "--
> >> enable-languages=c,c++,objc,obj-c++", and regression tested with a  
> >> top-
> >> level "make check-gcc" with no regression.
> >>
> >> gcc/ChangeLog:
> >>
> >> 	* gcc/config/rs6000/predicates.md (lwa_operand): Check the  
> >> alignment of
> >> 	a memory operand is 32 bits aligned or not.
> >
> > This is the wrong fix, the memory alignment is not the issue here  
> > but the offset
> > field has to be multiple of 4.  This is according to the ISA  
> > documents.
> >
> > Can you give more information about what is going wrong?  Because as  
> > far as I can
> > tell we check the offset to make sure it is a multiple of 4.
> >  	    || GET_CODE (XEXP (XEXP (inner, 0), 1)) != CONST_INT
> > 	    || INTVAL (XEXP (XEXP (inner, 0), 1)) % 4 == 0));
> >
> >
> > Thanks,
> > Andrew Pinski
> 
> For the following memory operand,
> (gdb) p debug_rtx(op)
> (mem/s/j:SI (lo_sum:DI (reg:DI 121)
>          (const:DI (minus:DI (symbol_ref:DI ("mybox") [flags 0x382]  
> <var_decl 0x41687e80 mybox>)
>                  (symbol_ref:DI ("<pic base>") [flags 0x180])))) [0  
> mybox.left+0 S4 A16])
> 
>      GET_CODE (XEXP (inner, 0)) == LO_SUM
>      GET_CODE (XEXP (XEXP (inner, 0), 1)) == CONST
> 
> The alignment is A16, i.e., 16 bits.

What about rejecting all low_sum instead?  The alignment is still the correct
check. Because we can have alignment of 16 and still have an offset which is
a multiple of 4.
Hmm, the other thing is that constraint m seems wrong for the lwa instruction,
I think it should be changed to Y like the load doubleword case.


Thanks,
Andrew Pinski