From patchwork Fri Jan 15 14:58:54 2016 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Stefan Liebler X-Patchwork-Id: 10394 Received: (qmail 72118 invoked by alias); 15 Jan 2016 14:59:10 -0000 Mailing-List: contact libc-alpha-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: libc-alpha-owner@sourceware.org Delivered-To: mailing list libc-alpha@sourceware.org Received: (qmail 72102 invoked by uid 89); 15 Jan 2016 14:59:10 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-2.6 required=5.0 tests=AWL, BAYES_00, RCVD_IN_DNSWL_LOW, RP_MATCHES_RCVD, SPF_HELO_PASS, SPF_PASS autolearn=ham version=3.3.2 spammy=275, 27, 7, 277, 246 X-HELO: plane.gmane.org To: libc-alpha@sourceware.org From: Stefan Liebler Subject: [PATCH] S/390: Do not raise inexact exception in lrint/lround functions in overflow case. Date: Fri, 15 Jan 2016 15:58:54 +0100 Lines: 264 Message-ID: Mime-Version: 1.0 X-Mozilla-News-Host: news://news.gmane.org:119 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.4.0 Hi, I get some math test-failures on s390 for float/double/ldouble for various lrint/lround functions like: lrint (0x1p64): Exception "Inexact" set lrint (-0x1p64): Exception "Inexact" set lround (0x1p64): Exception "Inexact" set lround (-0x1p64): Exception "Inexact" set ... GCC emits "convert to fixed" instructions for casting floating point values to integer values. These instructions raise invalid and inexact exceptions if the floating point value exceeds the integer type ranges. This patch enables the various FIX_DBL_LONG_CONVERT_OVERFLOW macros in order to avoid a cast from floating point to integer type and raise the invalid exception with feraiseexcept. The ldbl-128 rint/round functions are now using the same logic. Bye Stefan --- ChangeLog: * sysdeps/s390/fix-fp-int-convert-overflow.h: New File. * sysdeps/generic/fix-fp-int-convert-overflow.h (FIX_LDBL_LONG_CONVERT_OVERFLOW, FIX_LDBL_LLONG_CONVERT_OVERFLOW): New define. * sysdeps/arm/fix-fp-int-convert-overflow.h: Likewise. * sysdeps/mips/mips32/fpu/fix-fp-int-convert-overflow.h: Likewise. * sysdeps/ieee754/ldbl-128/s_lrintl.c (__lrintl): Avoid conversions to long int where inexact exceptions could be raised. * sysdeps/ieee754/ldbl-128/s_lroundl.c (__lroundl): Likewise. * sysdeps/ieee754/ldbl-128/s_llrintl.c (__llrintl): Avoid conversions to long long int where inexact exceptions could be raised. * sysdeps/ieee754/ldbl-128/s_llroundl.c (__llroundl): Likewise. commit f0c5fb9ece180de6dcf069910b6fea0bf6f7a3ba Author: Stefan Liebler Date: Thu Jan 14 17:26:25 2016 +0100 S/390: Do not raise inexact exception in lrint/lround functions in overflow case. GCC emits "convert to fixed" instructions for casting floating point values to integer values. These instructions raise invalid and inexact exceptions if the floating point value exceeds the integer type ranges. This patch enables the various FIX_DBL_LONG_CONVERT_OVERFLOW macros in order to avoid a cast from floating point to integer type and raise the invalid exception with feraiseexcept. The ldbl-128 rint/round functions are now using the same logic. diff --git a/sysdeps/arm/fix-fp-int-convert-overflow.h b/sysdeps/arm/fix-fp-int-convert-overflow.h index fcc2199..c5ec50a 100644 --- a/sysdeps/arm/fix-fp-int-convert-overflow.h +++ b/sysdeps/arm/fix-fp-int-convert-overflow.h @@ -25,8 +25,10 @@ see ). */ #define FIX_FLT_LLONG_CONVERT_OVERFLOW 1 #define FIX_DBL_LLONG_CONVERT_OVERFLOW 1 +#define FIX_LDBL_LLONG_CONVERT_OVERFLOW 0 #define FIX_FLT_LONG_CONVERT_OVERFLOW 0 #define FIX_DBL_LONG_CONVERT_OVERFLOW 0 +#define FIX_LDBL_LONG_CONVERT_OVERFLOW 0 #endif /* fix-fp-int-convert-overflow.h */ diff --git a/sysdeps/generic/fix-fp-int-convert-overflow.h b/sysdeps/generic/fix-fp-int-convert-overflow.h index 73884a6..fb68114 100644 --- a/sysdeps/generic/fix-fp-int-convert-overflow.h +++ b/sysdeps/generic/fix-fp-int-convert-overflow.h @@ -27,5 +27,7 @@ #define FIX_FLT_LLONG_CONVERT_OVERFLOW 0 #define FIX_DBL_LONG_CONVERT_OVERFLOW 0 #define FIX_DBL_LLONG_CONVERT_OVERFLOW 0 +#define FIX_LDBL_LONG_CONVERT_OVERFLOW 0 +#define FIX_LDBL_LLONG_CONVERT_OVERFLOW 0 #endif /* fix-fp-int-convert-overflow.h */ diff --git a/sysdeps/ieee754/ldbl-128/s_llrintl.c b/sysdeps/ieee754/ldbl-128/s_llrintl.c index 445cde5..84fc576 100644 --- a/sysdeps/ieee754/ldbl-128/s_llrintl.c +++ b/sysdeps/ieee754/ldbl-128/s_llrintl.c @@ -24,6 +24,7 @@ #include #include +#include static const long double two112[2] = { @@ -91,6 +92,12 @@ __llrintl (long double x) feraiseexcept (t == LLONG_MIN ? FE_INEXACT : FE_INVALID); return LLONG_MIN; } + else if (FIX_LDBL_LLONG_CONVERT_OVERFLOW && x != (long double) LLONG_MIN) + { + feraiseexcept (FE_INVALID); + return sx == 0 ? LLONG_MAX : LLONG_MIN; + } + #endif return (long long int) x; } diff --git a/sysdeps/ieee754/ldbl-128/s_llroundl.c b/sysdeps/ieee754/ldbl-128/s_llroundl.c index e5dd145..bfc81cc 100644 --- a/sysdeps/ieee754/ldbl-128/s_llroundl.c +++ b/sysdeps/ieee754/ldbl-128/s_llroundl.c @@ -23,7 +23,7 @@ #include #include - +#include long long int __llroundl (long double x) @@ -78,7 +78,14 @@ __llroundl (long double x) FE_INVALID must be raised and the return value is unspecified. */ #ifdef FE_INVALID - if (x <= (long double) LLONG_MIN - 0.5L) + if (FIX_LDBL_LLONG_CONVERT_OVERFLOW + && !(sign == -1 && x > (long double) LLONG_MIN - 0.5L)) + { + feraiseexcept (FE_INVALID); + return sign == 1 ? LLONG_MAX : LLONG_MIN; + } + else if (!FIX_LDBL_LLONG_CONVERT_OVERFLOW + && x <= (long double) LLONG_MIN - 0.5L) { /* If truncation produces LLONG_MIN, the cast will not raise the exception, but may raise "inexact". */ diff --git a/sysdeps/ieee754/ldbl-128/s_lrintl.c b/sysdeps/ieee754/ldbl-128/s_lrintl.c index ff4780a..23f828f 100644 --- a/sysdeps/ieee754/ldbl-128/s_lrintl.c +++ b/sysdeps/ieee754/ldbl-128/s_lrintl.c @@ -24,6 +24,7 @@ #include #include +#include static const long double two112[2] = { @@ -120,6 +121,12 @@ __lrintl (long double x) feraiseexcept (t == LONG_MIN ? FE_INEXACT : FE_INVALID); return LONG_MIN; } + else if (FIX_LDBL_LONG_CONVERT_OVERFLOW && x != (long double) LONG_MIN) + { + feraiseexcept (FE_INVALID); + return sx == 0 ? LONG_MAX : LONG_MIN; + } + #endif return (long int) x; } diff --git a/sysdeps/ieee754/ldbl-128/s_lroundl.c b/sysdeps/ieee754/ldbl-128/s_lroundl.c index 34d2264..f032625 100644 --- a/sysdeps/ieee754/ldbl-128/s_lroundl.c +++ b/sysdeps/ieee754/ldbl-128/s_lroundl.c @@ -23,7 +23,7 @@ #include #include - +#include long int __lroundl (long double x) @@ -87,7 +87,14 @@ __lroundl (long double x) FE_INVALID must be raised and the return value is unspecified. */ #ifdef FE_INVALID - if (x <= (long double) LONG_MIN - 0.5L) + if (FIX_LDBL_LONG_CONVERT_OVERFLOW + && !(sign == -1 && x > (long double) LONG_MIN - 0.5L)) + { + feraiseexcept (FE_INVALID); + return sign == 1 ? LONG_MAX : LONG_MIN; + } + else if (!FIX_LDBL_LONG_CONVERT_OVERFLOW + && x <= (long double) LONG_MIN - 0.5L) { /* If truncation produces LONG_MIN, the cast will not raise the exception, but may raise "inexact". */ diff --git a/sysdeps/mips/mips32/fpu/fix-fp-int-convert-overflow.h b/sysdeps/mips/mips32/fpu/fix-fp-int-convert-overflow.h index fb2ec69..b9f4aea 100644 --- a/sysdeps/mips/mips32/fpu/fix-fp-int-convert-overflow.h +++ b/sysdeps/mips/mips32/fpu/fix-fp-int-convert-overflow.h @@ -25,6 +25,7 @@ see ). */ #define FIX_FLT_LLONG_CONVERT_OVERFLOW 1 #define FIX_DBL_LLONG_CONVERT_OVERFLOW 1 +#define FIX_LDBL_LLONG_CONVERT_OVERFLOW 0 /* As of GCC 5 and binutils 2.25, for MIPS I GCC generates calls to assembler macros for conversions from floating point to integer @@ -32,5 +33,6 @@ lose exceptions. */ #define FIX_FLT_LONG_CONVERT_OVERFLOW (__mips == 1) #define FIX_DBL_LONG_CONVERT_OVERFLOW (__mips == 1) +#define FIX_LDBL_LONG_CONVERT_OVERFLOW 0 #endif /* fix-fp-int-convert-overflow.h */ diff --git a/sysdeps/s390/fix-fp-int-convert-overflow.h b/sysdeps/s390/fix-fp-int-convert-overflow.h new file mode 100644 index 0000000..61279ed --- /dev/null +++ b/sysdeps/s390/fix-fp-int-convert-overflow.h @@ -0,0 +1,33 @@ +/* Fix for conversion of floating point to integer overflow. S390 version. + Copyright (C) 2016 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#ifndef FIX_FP_INT_CONVERT_OVERFLOW_H +#define FIX_FP_INT_CONVERT_OVERFLOW_H 1 + +/* GCC emits "convert to fixed" instructions for casting floating point values + to integer values. These instructions raise invalid and inexact exceptions + if the floating point value exceeds the integer type ranges. */ +#define FIX_FLT_LLONG_CONVERT_OVERFLOW 1 +#define FIX_DBL_LLONG_CONVERT_OVERFLOW 1 +#define FIX_LDBL_LLONG_CONVERT_OVERFLOW 1 + +#define FIX_FLT_LONG_CONVERT_OVERFLOW 1 +#define FIX_DBL_LONG_CONVERT_OVERFLOW 1 +#define FIX_LDBL_LONG_CONVERT_OVERFLOW 1 + +#endif /* fix-fp-int-convert-overflow.h */