Re: [PATCH] Fix ix86_split_long_move collision handling with TLS (PR target/66470)

[Uros Bizjak <ubizjak@gmail.com>]

On Tue, Jun 9, 2015 at 4:26 PM, Jakub Jelinek <jakub@redhat.com> wrote:

>> >> I'm afraid that simple scan loop won't work correctly on x32. There
>> >> are some issues with UNSPEC_TP for this target, so we have to generate
>> >> zero_extend of SImode UNSPEC, e.g.:
>> >>
>> >> (plus:DI (zero_extend:DI (unspec:SI [...] UNSPEC_TP) (reg:DI ...))
>> >>
>> >> as can be seen in get_thread_pointer to construct the address. It
>> >> looks that your loop won't find the UNSPEC_TP tag in the above case.
>> >
>> > You're right, for -m32 it would need to start with
>
> Yeah, I meant -mx32 (which I have no experience with nor spare time for).
>
>> >    rtx *x = &addr;
>> > +  while (GET_CODE (*x) == ZERO_EXTEND
>> > +        || GET_CODE (*x) == AND
>> > +        || GET_CODE (*x) == SUBREG)
>> > +    x = &XEXP (*x, 0);
>>
>> Oh, you can use SImode_address_operand predicate here.
>
> Do I need to loop, or can there be just one SImode_address_operand

IIRC, apart from the whole address, only UNSPEC_TP can be
zero_extended. It is a hardware "feature" (== HW bug) that addr32
doesn't apply to segment registers.

> code?  Do you want to use the iterators (as in the second patch) or not
> (then is
>   if (SImode_address_operand (addr, VOIDmode))
>     x = &XEXP (addr, 0);
> ok)?  Is Pmode always SImode for -mx32, or depending on some switch or

Nope, it depends on -maddress-mode switch, and can be SImode or DImode.

> something?  Would it be acceptable to just guard the changes in the patch
> with !TARGET_X32 and let H.J. deal with that target?  I'm afraid I'm lost
> when to ZERO_EXTEND addr (if needed at all), etc.

If you wish, I can take your patch and take if further. -mx32 is a
delicate beast...

Uros.
>
>         Jakub