Re: [PATCH] rs6000: Add "cannot_copy" attribute, use it (PR67788, PR67789)

[Richard Biener <richard.guenther@gmail.com>]

On Fri, Oct 2, 2015 at 3:14 AM, Segher Boessenkool
<segher@kernel.crashing.org> wrote:
> On Fri, Oct 02, 2015 at 10:24:07AM +0930, Alan Modra wrote:
>> On Thu, Oct 01, 2015 at 12:18:08PM -0500, Segher Boessenkool wrote:
>> > On Thu, Oct 01, 2015 at 12:14:44PM +0200, Richard Biener wrote:
>> > > So even if not "easy", can you try?
>> >
>> > I did, and after half a day had a big mess and lots of things failing,
>> > no idea where this was headed, and in the meantime bootstrap still fails
>> > (on affected targets).
>>
>> I had a look too, and while you can revise the load_toc_v4_PIC
>> patterns to use labels emitted the usual way (eg. as in
>> i386.c:ix86_init_large_pic_reg) they tend to wander away from the
>> insn.
>
> Yes, and only "bcl 20,31,$+4" avoids the link stack on recent CPUs
> (bcl 20,31,$+8, which we also use, doesn't).
>
>> I think that could be solved, but these labels which aren't
>> referred to by jump insns get converted to NOTE_INSN_DELETED_LABEL
>> somewhere, and that leads to further pain.
>
> Yes.  You need to make the bcl a jump_insn to the label.  And then
> there is yet more pain.

Sounds like supporting this with a special instruction in the assembler
would be easier then? ...

As for the compile-time hit, yes, calling the hook at all and having an
extra loop over all stmts in cfg_layout_can_duplicate_bb_p.  There are
not many other uses of the hook, so I wonder if it is called everywhere
it has to be called.

Richard.

>
> Segher