From: Ingo Molnar <mingo@kernel.org>
Date: Sun, 17 May 2015 05:56:54 +0000 (+0200)
Subject: x86: Pack loops tightly as well
X-Git-Url: http://git.lede-project.org./?a=commitdiff_plain;h=52648e83c9a6b9f7fc3dd272d4d10175e93aa62a;p=openwrt%2Fstaging%2Fblogic.git

x86: Pack loops tightly as well

Packing loops tightly (-falign-loops=1) is beneficial to code size:

     text        data    bss     dec              filename
 12566391        1617840 1089536 15273767         vmlinux.align.16-byte
 12224951        1617840 1089536 14932327         vmlinux.align.1-byte
 11976567        1617840 1089536 14683943         vmlinux.align.1-byte.funcs-1-byte
 11903735        1617840 1089536 14611111         vmlinux.align.1-byte.funcs-1-byte.loops-1-byte

Which reduces the size of the kernel by another 0.6%, so the
the total combined size reduction of the alignment-packing
patches is ~5.5%.

The x86 decoder bandwidth and caching arguments laid out in:

  be6cb02779ca ("x86: Align jump targets to 1-byte boundaries")

apply to loop alignment as well.

Furtermore, modern CPU uarchs have a loop cache/buffer that
is a L0 cache before even any uop cache, covering a few
dozen most recently executed instructions.

This loop cache generally does not have the 16-byte alignment
restrictions of the uop cache.

Now loop alignment can still be beneficial if:

 - a loop is cache-hot and its surroundings are not.

 - if the loop is so cache hot that the instruction
   flow becomes x86 decoder bandwidth limited

But loop alignment is harmful if:

 - a loop is cache-cold

 - a loop's surroundings are cache-hot as well

 - two cache-hot loops are close to each other

 - if the loop fits into the loop cache

 - if the code flow is not decoder bandwidth limited

and I'd argue that the latter five scenarios are much
more common in the kernel, as our hottest loops are
typically:

 - pointer chasing: this should fit into the loop cache
   in most cases and is typically data cache and address
   generation limited

 - generic memory ops (memset, memcpy, etc.): these generally
   fit into the loop cache as well, and are likewise data
   cache limited.

So this patch packs loop addresses tightly as well.

Acked-by: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Aswin Chandramouleeswaran <aswin@hp.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jason Low <jason.low2@hp.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Link: http://lkml.kernel.org/r/20150410123017.GB19918@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---

diff --git a/arch/x86/Makefile b/arch/x86/Makefile
index ca17e5fcd125..57996ee840dd 100644
--- a/arch/x86/Makefile
+++ b/arch/x86/Makefile
@@ -80,6 +80,9 @@ else
         # Align jump targets to 1 byte, not the default 16 bytes:
         KBUILD_CFLAGS += -falign-jumps=1
 
+        # Pack loops tightly as well:
+        KBUILD_CFLAGS += -falign-loops=1
+
         # Don't autogenerate traditional x87 instructions
         KBUILD_CFLAGS += $(call cc-option,-mno-80387)
         KBUILD_CFLAGS += $(call cc-option,-mno-fp-ret-in-387)