Date: Tue, 2 Dec 2003 17:19:40 -0800 (PST) From: Peter Wemm <peter@FreeBSD.org> To: Perforce Change Reviews <perforce@freebsd.org> Subject: PERFORCE change 43332 for review Message-ID: <200312030119.hB31JexD091510@repoman.freebsd.org>
next in thread | raw e-mail | index | archive | help
http://perforce.freebsd.org/chv.cgi?CH=43332 Change 43332 by peter@peter_overcee on 2003/12/02 17:19:39 replace the handrolled ffs() with __builtin_ffs(), which gcc does a better job of with instruction sets >= pentiumpro. For example, it uses the conditional move instructions to avoid branches. Affected files ... .. //depot/projects/hammer/sys/amd64/include/cpufunc.h#13 edit Differences ... ==== //depot/projects/hammer/sys/amd64/include/cpufunc.h#13 (text+ko) ==== @@ -129,13 +129,18 @@ static __inline int ffs(int mask) { +#if 0 /* * Note that gcc-2's builtin ffs would be used if we didn't declare * this inline or turn off the builtin. The builtin is faster but * broken in gcc-2.4.5 and slower but working in gcc-2.5 and later * versions. */ - return (mask == 0 ? mask : (int)bsfl((u_int)mask) + 1); + return (mask == 0 ? mask : (int)bsfl((u_int)mask) + 1); +#else + /* Actually, the above is way out of date. The builtins use cmov etc */ + return (__builtin_ffs(mask)); +#endif } #define HAVE_INLINE_FLS
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200312030119.hB31JexD091510>