https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128846

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to [email protected] from comment #6)
> > --- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
> > On Linux:
> >
> >   [Nr] Name              Type             Address           Offset
> >        Size              EntSize          Flags  Link  Info  Align
> > ...
> >   [ 6] .tbss             NOBITS           0000000000000000  00000088
> >        0000000000000288  0000000000000000 WAT       0     0     16
> >
> > so the TLS segment is appropriately aligned.  With -mavx2 alignment 
> > increases:
> >
> >   [ 6] .tbss             NOBITS           0000000000000000  00000088
> >        0000000000000288  0000000000000000 WAT       0     0     32
> >
> > same with -m32 on x86_64-linux.
> 
> This is the x86_64-linux -m64 behaviour.  For i686, I see this instead:

I don't have convenient access to a bare i686, but using -march=i686
shows we don't vectorize on trunk, even with -fno-vect-cost-model,
so that would explain the lower alignment (it's the vectorizer requesting
a alrger alignment).  With -march=i686 -msse2 it gets vectorized and
alignment is sufficient.

> * gcc-16:
> 
>   [18] .tbss
>        NOBITS          0804bef0 002ef0 000288 00   0   0 16
>        [00000403]: WRITE, ALLOC, TLS
> 
> * gcc-17:
> 
>   [18] .tbss
>        NOBITS          0804bef0 002ef0 000288 00   0   0  8
>        [00000403]: WRITE, ALLOC, TLS
> 
> On i386-solaris, I find
> 
> * gcc-16:
> 
>   [26] .tbss
>        NOBITS          08060eb0 000eb0 000288 00   0   0 16
>        [00000403]: WRITE, ALLOC, TLS
> 
> * gcc-17:
> 
>   [26] .tbss
>        NOBITS          08060ed0 000ed0 000288 00   0   0 16
>        [00000403]: WRITE, ALLOC, TLS

So the solaris alignment should be sufficient for movaps %xmm, so how
is it not?

Reply via email to