From mboxrd@z Thu Jan 1 00:00:00 1970 X-Msuck: nntp://news.gmane.org/gmane.linux.lib.musl.general/14706 Path: news.gmane.org!.POSTED.blaine.gmane.org!not-for-mail From: Szabolcs Nagy Newsgroups: gmane.linux.lib.musl.general Subject: Re: [PATCH] math: optimize lrint on 32bit targets Date: Sun, 22 Sep 2019 22:43:35 +0200 Message-ID: <20190922204335.GC22009@port70.net> References: <20190921155234.GA22009@port70.net> Reply-To: musl@lists.openwall.com Mime-Version: 1.0 Content-Type: multipart/mixed; boundary="VUDLurXRWRKrGuMn" Injection-Info: blaine.gmane.org; posting-host="blaine.gmane.org:195.159.176.226"; logging-data="219733"; mail-complaints-to="usenet@blaine.gmane.org" User-Agent: Mutt/1.10.1 (2018-07-13) To: musl@lists.openwall.com Original-X-From: musl-return-14722-gllmg-musl=m.gmane.org@lists.openwall.com Sun Sep 22 22:43:52 2019 Return-path: Envelope-to: gllmg-musl@m.gmane.org Original-Received: from mother.openwall.net ([195.42.179.200]) by blaine.gmane.org with smtp (Exim 4.89) (envelope-from ) id 1iC8i0-000v3R-GI for gllmg-musl@m.gmane.org; Sun, 22 Sep 2019 22:43:52 +0200 Original-Received: (qmail 13415 invoked by uid 550); 22 Sep 2019 20:43:47 -0000 Mailing-List: contact musl-help@lists.openwall.com; run by ezmlm Precedence: bulk List-Post: List-Help: List-Unsubscribe: List-Subscribe: List-ID: Original-Received: (qmail 13396 invoked from network); 22 Sep 2019 20:43:47 -0000 Mail-Followup-To: musl@lists.openwall.com Content-Disposition: inline In-Reply-To: <20190921155234.GA22009@port70.net> Xref: news.gmane.org gmane.linux.lib.musl.general:14706 Archived-At: --VUDLurXRWRKrGuMn Content-Type: text/plain; charset=us-ascii Content-Disposition: inline * Szabolcs Nagy [2019-09-21 17:52:35 +0200]: > this was discussed on irc. did more benchmarks, on i486 branches seem better than setting the sign bit but on arm branch is worse so i keep the original code, just changed the code style (asuint macro instead of union). --VUDLurXRWRKrGuMn Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0001-math-optimize-lrint-on-32bit-targets.patch" >From 67990a5c85fc5db55831f9ddddc58317e5b344b6 Mon Sep 17 00:00:00 2001 From: Szabolcs Nagy Date: Mon, 16 Sep 2019 20:33:11 +0000 Subject: [PATCH] math: optimize lrint on 32bit targets lrint in (LONG_MAX, 1/DBL_EPSILON) and in (-1/DBL_EPSILON, LONG_MIN) is not trivial: rounding to int may be inexact, but the conversion to int may overflow and then the inexact flag must not be raised. (the overflow threshold is rounding mode dependent). this matters on 32bit targets (without single instruction lrint or rint), so the common case (when there is no overflow) is optimized by inlining the lrint logic, otherwise the old code is kept as a fallback. on my laptop an i486 lrint call is asm:10ns, old c:30ns, new c:21ns on a smaller arm core: old c:71ns, new c:34ns on a bigger arm core: old c:27ns, new c:19ns --- src/math/lrint.c | 28 +++++++++++++++++++++++++++- 1 file changed, 27 insertions(+), 1 deletion(-) diff --git a/src/math/lrint.c b/src/math/lrint.c index bdca8b7c..ddee7a0d 100644 --- a/src/math/lrint.c +++ b/src/math/lrint.c @@ -1,5 +1,6 @@ #include #include +#include #include "libm.h" /* @@ -26,7 +27,18 @@ as a double. */ #if LONG_MAX < 1U<<53 && defined(FE_INEXACT) -long lrint(double x) +#include +#include +#if FLT_EVAL_METHOD==0 || FLT_EVAL_METHOD==1 +#define EPS DBL_EPSILON +#elif FLT_EVAL_METHOD==2 +#define EPS LDBL_EPSILON +#endif +#ifdef __GNUC__ +/* avoid stack frame in lrint */ +__attribute__((noinline)) +#endif +static long lrint_slow(double x) { #pragma STDC FENV_ACCESS ON int e; @@ -38,6 +50,20 @@ long lrint(double x) /* conversion */ return x; } + +long lrint(double x) +{ + uint32_t abstop = asuint64(x)>>32 & 0x7fffffff; + uint64_t sign = asuint64(x) & (1ULL << 63); + + if (abstop < 0x41dfffff) { + /* |x| < 0x7ffffc00, no overflow */ + double_t toint = asdouble(asuint64(1/EPS) | sign); + double_t y = x + toint - toint; + return (long)y; + } + return lrint_slow(x); +} #else long lrint(double x) { -- 2.21.0 --VUDLurXRWRKrGuMn--