From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S934453AbYD1OFw (ORCPT ); Mon, 28 Apr 2008 10:05:52 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S933301AbYD1OFo (ORCPT ); Mon, 28 Apr 2008 10:05:44 -0400 Received: from www.tglx.de ([62.245.132.106]:56811 "EHLO www.tglx.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933246AbYD1OFn (ORCPT ); Mon, 28 Apr 2008 10:05:43 -0400 Date: Mon, 28 Apr 2008 16:04:47 +0200 (CEST) From: Thomas Gleixner To: Linus Torvalds cc: Harvey Harrison , Ingo Molnar , Andrew Morton , LKML , David Miller Subject: Re: [PATCH] bitops: simplify generic bit finding functions In-Reply-To: Message-ID: References: <1209327591.14173.74.camel@brick> User-Agent: Alpine 1.10 (LFD 962 2008-03-14) MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Sun, 27 Apr 2008, Linus Torvalds wrote: > On Sun, 27 Apr 2008, Harvey Harrison wrote: > > > > No need for a sentinal if we explicitly catch the no bits/all bits set > > cases, make it clear they are special cases returning size/BITS_PER_LONG. > > These things need to be re-done anyway. David already complained that the > thing inlines to something much much too big. The delta between the inlined version including the constant optimizations and the original non inlined version is exactly 1400 bytes on a sparc64 defconfig build. > Let's just make it not be an inline at all. Which makes the whole constant optimization moot. How about making the optimization controlled by a config switch so arch maintainers can decide whether they want to enable the constant optimization or unconditionally call the lib function ? See patch below. It gives back the 1400 bytes on SPARC64 and other platforms that have no instruction for find bit. Thanks, tglx ----------> Subject: bitops: optional bitops mapsize optimization on config switch From: Thomas Gleixner Date: Mon, 28 Apr 2008 00:22:06 +0200 The mapsize optimizations which were moved from x86 to the generic code in commit 64970b68d2b3ed32b964b0b30b1b98518fde388e increased the binary size on non x86 architectures. Make the optimization depend on a config switch so architecture maintainers can decide. Signed-off-by: Thomas Gleixner Acked-by: Ingo Molnar --- arch/x86/Kconfig.cpu | 1 + include/linux/bitops.h | 6 ++++-- 2 files changed, 5 insertions(+), 2 deletions(-) Index: linux-2.6/arch/x86/Kconfig.cpu =================================================================== --- linux-2.6.orig/arch/x86/Kconfig.cpu +++ linux-2.6/arch/x86/Kconfig.cpu @@ -282,6 +282,7 @@ config X86_CPU def_bool y select GENERIC_FIND_FIRST_BIT select GENERIC_FIND_NEXT_BIT + select GENERIC_FIND_BIT_MAPSIZE_OPTIMIZE config X86_GENERIC bool "Generic x86 support" Index: linux-2.6/include/linux/bitops.h =================================================================== --- linux-2.6.orig/include/linux/bitops.h +++ linux-2.6/include/linux/bitops.h @@ -190,6 +190,7 @@ static __always_inline unsigned long find_next_bit(const unsigned long *addr, unsigned long size, unsigned long offset) { +#ifdef CONFIG_GENERIC_FIND_BIT_MAPSIZE_OPTIMIZE unsigned long value; /* Avoid a function call if the bitmap size is a constant */ @@ -209,7 +210,7 @@ find_next_bit(const unsigned long *addr, value = (*addr) & ((~0ul) << offset); return (value == 0) ? BITS_PER_LONG : __ffs(value); } - +#endif /* size is not constant or too big */ return __find_next_bit(addr, size, offset); } @@ -227,6 +228,7 @@ static __always_inline unsigned long find_next_zero_bit(const unsigned long *addr, unsigned long size, unsigned long offset) { +#ifdef CONFIG_GENERIC_FIND_BIT_MAPSIZE_OPTIMIZE unsigned long value; /* Avoid a function call if the bitmap size is a constant */ @@ -246,7 +248,7 @@ find_next_zero_bit(const unsigned long * value = (~(*addr)) & ((~0ul) << offset); return (value == 0) ? BITS_PER_LONG : __ffs(value); } - +#endif /* size is not constant or too big */ return __find_next_zero_bit(addr, size, offset); }