Re: [PATCH] x86: Turn off GCC branch probability heuristics

From: Mike Galbraith <umgwanakikbuti@gmail.com>
To: Ingo Molnar <mingo@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Jakub Jelinek <jakub@redhat.com>,
	Denys Vlasenko <dvlasenk@redhat.com>,
	Borislav Petkov <bp@alien8.de>,
	Tim Chen <tim.c.chen@linux.intel.com>,
	Andy Lutomirski <luto@amacapital.net>,
	Jason Low <jason.low2@hp.com>, Brian Gerst <brgerst@gmail.com>,
	Aswin Chandramouleeswaran <aswin@hp.com>,
	"Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
	Davidlohr Bueso <dave@stgolabs.net>,
	Peter Zijlstra <a.p.zijlstra@chello.nl>,
	"H. Peter Anvin" <hpa@zytor.com>,
	LKML <linux-kernel@vger.kernel.org>,
	Peter Zijlstra <peterz@infradead.org>
Subject: Re: [PATCH] x86: Turn off GCC branch probability heuristics
Date: Sun, 12 Apr 2015 09:56:11 +0200	[thread overview]
Message-ID: <1428825371.3479.11.camel@gmail.com> (raw)
In-Reply-To: <20150412054747.GA9062@gmail.com>

On Sun, 2015-04-12 at 07:47 +0200, Ingo Molnar wrote:
> * Linus Torvalds <torvalds@linux-foundation.org> wrote:
> 
> > On Sat, Apr 11, 2015 at 11:57 AM, Thomas Gleixner <
> > tglx@linutronix.de> wrote:
> > > 
> > > I thinks its just the no-guess one:
> > > 
> > >    text        data       dec  patch           reduction
> > > 7563475     1781048  10302987
> > > 7192973     1780024   9931461  no-guess            -4.8%
> > > 7354819     1781048    958464  align-1             -2.7%
> > > 7192973     1780024   9931461  no-guess + align-1  -4.8%
> > 
> > Yeah, a 5% code expansion is a big deal. Sadly, it looks like
> > 'no-guess' also disables our explicit likely/unlikely handling.
> 
> So I spent some time trying to get as much code size reduction as 
> possible via GCC optimization options, and the total savings 
> possible 
> are 10.1%:
> 
>      text           data     bss      dec    filename
>  12566391        1617840 1089536 15273767    vmlinux.vanilla
>  11416805        1617840 1089536 14124181    vmlinux.combo
>  10532552        1596080 1089536 13218168    vmlinux.Os
> 
> (combo patch attached below.)
> 
> The -Os savings are 19% total - but as you mentioned before, it's 
> sometimes achieved through unacceptable techniques.
> 
> Unfortunately I found no other GCC options to achieve what -Os does -
> 
> the missing 9% can purely be achieved via -Os, with no cherry-
> picking 
> possible.
> 
> The other, smaller savings are:
> 
> +        # Reduces vmlinux size by 0.25%:
> +        KBUILD_CFLAGS += -fno-caller-saves
> +
> +        # Reduces vmlinux size by 1.10%:
> +        KBUILD_CFLAGS += -fno-inline-small-functions
> +
> +        # Reduces vmlinux size by about 0.95%:
> +        KBUILD_CFLAGS += -fno-tree-ch
> 
> (each of them has to be double checked to make sure it leads to 
> nothing silly and unacceptable - I just blindly tried to find GCC 
> options that impacted kernel code size.)

Ew, my i4790 really hated this patch.

taskset 0xc pipe-test 1
                                                 avg
733.2    727.1    743.1    746.6    737.1        737 KHz  1.000  -gcc_twiddle
713.7    717.9    715.5    718.0    708.7        714 KHz   .968  +gcc_twiddle

tbench.sh 8 30

3566.14  3560.91  3566.35  3556.57  3549.69  3559.93 MB/S 1.000  -gcc_twiddle
2862.18  2899.51  2888.74  2897.18  2878.63  2885.25 MB/S  .810  +gcc_twiddle

        -Mike