This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]
Other format:	[Raw text]

Re: Preserve label alignment info through dbr

From: Jeff Law <law at redhat dot com>
To: Eric Botcazou <ebotcazou at adacore dot com>
Cc: gcc-patches at gcc dot gnu dot org
Date: Wed, 17 Jul 2013 15:03:32 -0600
Subject: Re: Preserve label alignment info through dbr
References: <1821473 dot SyjjZnTRdx at polaris>

On 07/10/2013 03:37 AM, Eric Botcazou wrote:

Hi,

I was a little surprised to find out that dbr can silently drop the alignment
information for labels computed when the CFG is valid (in compute_alignments).
The pessimization can be significant for loops when the top label needs to be
overaligned, as seen on a private port.  Hence the attached patch.

Tested on SPARC/Solaris and SPARC64/Solaris, any objections?


2013-07-10  Eric Botcazou  <ebotcazou@adacore.com>

	* rtl.h (update_alignments): Declare.
	* final.c (grow_label_align): New function extracted from...
	(shorten_branches): ...here.  Call it.
	(update_alignments): New function.
	* reorg.c (sibling_labels): New variable.
	(get_label_before): Add SIBLING parameter.  If it is non-zero, push
	the new label along with it onto the sibling_labels vector.
	(fill_simple_delay_slots): Adjust call to get_label_before.
	(fill_slots_from_thread): Likewise.
	(relax_delay_slots): Likewise.
	(make_return_insns): Likewise.
	(dbr_schedule): Invoke update_alignment on the sibling_labels vector.

Looks good to me.

jeff

References:
- Preserve label alignment info through dbr
  - From: Eric Botcazou

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]