From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751993AbcE1T6l (ORCPT ); Sat, 28 May 2016 15:58:41 -0400 Received: from science.sciencehorizons.net ([71.41.210.147]:30034 "HELO ns.sciencehorizons.net" rhost-flags-OK-FAIL-OK-OK) by vger.kernel.org with SMTP id S1751787AbcE1T6J (ORCPT ); Sat, 28 May 2016 15:58:09 -0400 From: George Spelvin To: Linus Torvalds , lkml Cc: "J . Bruce Fields" , George Spelvin , Yoshinori Sato , uclinux-h8-devel@lists.sourceforge.jp Subject: [PATCH v3 10/10] h8300: Add Date: Sat, 28 May 2016 15:57:23 -0400 Message-Id: <1464465443-25305-11-git-send-email-linux@sciencehorizons.net> X-Mailer: git-send-email 2.8.1 In-Reply-To: <1464465443-25305-1-git-send-email-linux@sciencehorizons.net> References: <1464465443-25305-1-git-send-email-linux@sciencehorizons.net> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org This will improve the performance of hash_32() and hash_64(), but due to complete lack of multi-bit shift instructions on H8, performance will still be bad in surrounding code. Designing H8-specific hash algorithms to work around that is a separate project. (But if the maintainers would like to get in touch...) Signed-off-by: George Spelvin Cc: Yoshinori Sato Cc: uclinux-h8-devel@lists.sourceforge.jp --- arch/h8300/Kconfig | 1 + arch/h8300/include/asm/hash.h | 53 +++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 54 insertions(+) create mode 100644 arch/h8300/include/asm/hash.h diff --git a/arch/h8300/Kconfig b/arch/h8300/Kconfig index 986ea84c..6c583dbb 100644 --- a/arch/h8300/Kconfig +++ b/arch/h8300/Kconfig @@ -20,6 +20,7 @@ config H8300 select HAVE_KERNEL_GZIP select HAVE_KERNEL_LZO select HAVE_ARCH_KGDB + select HAVE_ARCH_HASH config RWSEM_GENERIC_SPINLOCK def_bool y diff --git a/arch/h8300/include/asm/hash.h b/arch/h8300/include/asm/hash.h new file mode 100644 index 00000000..04cfbd2b --- /dev/null +++ b/arch/h8300/include/asm/hash.h @@ -0,0 +1,53 @@ +#ifndef _ASM_HASH_H +#define _ASM_HASH_H + +/* + * The later H8SX models have a 32x32-bit multiply, but the H8/300H + * and H8S have only 16x16->32. Since it's tolerably compact, this is + * basically an inlined version of the __mulsi3 code. Since the inputs + * are not expected to be small, it's also simplfied by skipping the + * early-out checks. + * + * (Since neither CPU has any multi-bit shift instructions, a + * shift-and-add version is a non-starter.) + * + * TODO: come up with an arch-specific version of the hashing in fs/namei.c, + * since that is heavily dependent on rotates. Which, as mentioned, suck + * horribly on H8. + */ + +#if defined(CONFIG_CPU_H300H) || defined(CONFIG_CPU_H8S) + +#define HAVE_ARCH__HASH_32 1 + +/* + * Multiply by k = 0x61C88647. Fitting this into three registers requires + * one extra instruction, but reducing register pressure will probably + * make that back and then some. + * + * GCC asm note: %e1 is the high half of operand %1, while %f1 is the + * low half. So if %1 is er4, then %e1 is e4 and %f1 is r4. + * + * This has been designed to modify x in place, since that's the most + * common usage, but preserve k, since hash_64() makes two calls in + * quick succession. + */ +static inline u32 __attribute_const__ __hash_32(u32 x) +{ + u32 temp; + + asm( "mov.w %e1,%f0" + "\n mulxu.w %f2,%0" /* klow * xhigh */ + "\n mov.w %f0,%e1" /* The extra instruction */ + "\n mov.w %f1,%f0" + "\n mulxu.w %e2,%0" /* khigh * xlow */ + "\n add.w %e1,%f0" + "\n mulxu.w %f2,%1" /* klow * xlow */ + "\n add.w %f0,%e1" + : "=&r" (temp), "=r" (x) + : "%r" (GOLDEN_RATIO_32), "1" (x)); + return x; +} + +#endif +#endif /* _ASM_HASH_H */ -- 2.8.1