Re: PATCH [mainline] Fix for long double (TARGET_LONG_DOUBLE_128) on ppc-darwin

[Geoff Keating <geoffk at geoffk dot org>]

On 30/07/2004, at 4:01 PM, Fariborz Jahanian wrote:

> On Jul 30, 2004, at 2:17 PM, Geoffrey Keating wrote:
>
> > Fariborz Jahanian <fjahanian@apple.com> writes:
> >
> > > 7e.com)
> >
> > I think the second part of this patch won't work.  If you get into
> > rs6000_split_multireg_move with a non-offsettable address, there's no
> > code you can generate.  I'll look into this a bit more and come up
> > with a patch.
>
> Second part was added after change in the first part. In the following 
> test case:
>
> static long long foo()
> {
>   static long long rat;
>   return rat;
> }
>
> Following pattern is now being generated:
>
> (insn 9 8 10 (set:SI (reg:SI 121)
>         (plus:SI (reg:SI 31 r31)
>             (high:SI (const:SI (minus:SI (symbol_ref:SI ("!D__*rat.0") 
> [flags 0x2] <var_decl 0x41757380 rat>)
>                         (symbol_ref:SI ("<pic base>"))))))) -1 (nil)
>     (nil))
>
> (insn 10 9 11 (set:SI (reg/f:SI 120)
>         (lo_sum:SI (reg:SI 121)
>             (const:SI (minus:SI (symbol_ref:SI ("!D__*rat.0") [flags 
> 0x2] <var_decl 0x41757380 rat>)
>                     (symbol_ref:SI ("<pic base>")))))) -1 (nil)
>     (expr_list:REG_EQUAL (symbol_ref:SI ("!D__*rat.0") [flags 0x2] 
> <var_decl 0x41757380 rat>)
>         (nil)))

I think you copied-and-pasted the wrong RTL, this RTL and code don't 
match.

> And following code is generated which looks correct to me and works in 
> test case (and bootstrap of gcc):
>
>        addis r2,r10,ha16(rat.0-"L00000000001$pb")
>         la r2,lo16(rat.0-"L00000000001$pb")(r2)
>         lfd f0,0(r2)