Re: PATCH: PR target/39942: Nonoptimal code - leaveq; xchg %ax,%ax; retq

[Uros Bizjak <ubizjak at gmail dot com>]

H.J. Lu wrote:

> > > ix86_avoid_jump_misspredicts use "align" pattern to pad to a 16byte
> > > boundary. "align" pattern uses ASM_OUTPUT_MAX_SKIP_ALIGN. For Linux,
> > > it
> > >       
> > See my http://gcc.gnu.org/ml/gcc-patches/2009-05/msg00702.html
> > patch, I think ASM_OUTPUT_MAX_SKIP_ALIGN should be fixed anyway
> > to not do the extra .p2align 3 if MAX_SKIP is less than 7.
> > ASM_OUTPUT_MAX_SKIP_PAD doesn't hurt, though I think it only rarely
> > will make a difference. ïAnyway, I think it makes no sense to require
> > it duplicated in all the headers, just define the new macro
> > in linux.h and x86-64.h and in i386.md just use ASM_OUTPUT_MAX_SKIP_PAD
> > if it exists, otherwise fall back to ASM_OUTPUT_MAX_SKIP_ALIGN.
> >
> >     
>
> Here is the updated patch.  OK for trunk?
>   

> 2009-05-15  H.J. Lu  <hongjiu.lu@intel.com>
>
> 	PR target/39942
> 	* config/i386/i386.c (ix86_avoid_jump_misspredicts): Replace
> 	gen_align with gen_pad.
>
> 	* config/i386/i386.md (align): Renamed to ...
> 	(pad): This.  Replace ASM_OUTPUT_MAX_SKIP_ALIGN with
> 	ASM_OUTPUT_MAX_SKIP_PAD.
>
> 	* config/i386/i386.h (ASM_OUTPUT_MAX_SKIP_PAD): New.
> 	* config/i386/x86-64.h (ASM_OUTPUT_MAX_SKIP_PAD): Likewise.
>   

OK for mainline; see bellow for a small comment.

> +#ifdef HAVE_GAS_MAX_SKIP_P2ALIGN
> +#undef  ASM_OUTPUT_MAX_SKIP_PAD
> +#define ASM_OUTPUT_MAX_SKIP_PAD(FILE, LOG, MAX_SKIP)			\
> +  if ((LOG) != 0) {							\
> +    if ((MAX_SKIP) == 0) fprintf ((FILE), "\t.p2align %d\n", (LOG));	\
> +    else fprintf ((FILE), "\t.p2align %d,,%d\n", (LOG), (MAX_SKIP));	\
> +  }

Please follow GNU codding standard in the code above (in both instances).

Thanks,
Uros.