From patchwork Tue Mar 11 20:59:12 2014 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Joseph Myers X-Patchwork-Id: 42 Return-Path: X-Original-To: siddhesh@wilcox.dreamhost.com Delivered-To: siddhesh@wilcox.dreamhost.com Received: from homiemail-mx20.g.dreamhost.com (caibbdcaaahc.dreamhost.com [208.113.200.72]) by wilcox.dreamhost.com (Postfix) with ESMTP id 76AF43600B0 for ; Tue, 11 Mar 2014 13:59:32 -0700 (PDT) Received: by homiemail-mx20.g.dreamhost.com (Postfix, from userid 14307373) id 1E6CA40837CBC; Tue, 11 Mar 2014 13:59:32 -0700 (PDT) X-Original-To: glibc@patchwork.siddhesh.in Delivered-To: x14307373@homiemail-mx20.g.dreamhost.com Received: from sourceware.org (server1.sourceware.org [209.132.180.131]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by homiemail-mx20.g.dreamhost.com (Postfix) with ESMTPS id E3A2B40837CA2 for ; Tue, 11 Mar 2014 13:59:31 -0700 (PDT) DomainKey-Signature: a=rsa-sha1; c=nofws; d=sourceware.org; h=list-id :list-unsubscribe:list-subscribe:list-archive:list-post :list-help:sender:date:from:to:subject:message-id:mime-version :content-type; q=dns; s=default; b=lsaapYDm0NxfXTjULgJiWvhx1zc6W UOexLmlSwCsQ9ntKtByk9dcbCR6AEXM+hyV1TsT2wMj0SibIdlUxa0sepbpaF1ac coNeV+cw19vNXJLVm1G9E7bsWaOyvoo7sfIWd3QZgB1BjhWjeT3aIISACpJX0EtX VxC6op/pODZX4Q= DKIM-Signature: v=1; a=rsa-sha1; c=relaxed; d=sourceware.org; h=list-id :list-unsubscribe:list-subscribe:list-archive:list-post :list-help:sender:date:from:to:subject:message-id:mime-version :content-type; s=default; bh=h6mvthaEsAXiPVaBzYCL9Vy0hvs=; b=onH 8GfZwm86acvkCFtAgyMXbh7Rx7M+3YUsKk/GwzQvVu7hNhCz2uhxLGp2GtI2+1+4 kdG+kuoS7ZB/5/78G1aiMSD/w8v0cg+yxYess2gyGi3IqtL79vnm6Fo0EM2F3QGR 3cvwJpGbStXHqor5GBR2mYUa5pHBSeQSyuHHvmww= Received: (qmail 13007 invoked by alias); 11 Mar 2014 20:59:23 -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 12941 invoked by uid 89); 11 Mar 2014 20:59:22 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-2.0 required=5.0 tests=AWL, BAYES_00 autolearn=ham version=3.3.2 X-HELO: relay1.mentorg.com Date: Tue, 11 Mar 2014 20:59:12 +0000 From: "Joseph S. Myers" To: Subject: Fix nextafter overflow in non-default rounding modes (bug 16677) Message-ID: MIME-Version: 1.0 X-DH-Original-To: glibc@patchwork.siddhesh.in ISO C requires the result of nextafter to be independent of the rounding mode, even when underflow or overflow occurs. This patch fixes the bug in various nextafter implementations that, having done an overflowing computation to force an overflow exception (correct), they then return the result of that computation rather than an infinity computed some other way (incorrect, when the overflowing result of arithmetic with that sign and rounding mode is finite but the correct result is infinite) - generally by falling through to existing code to return a value that in fact is correct for this case (but was computed by an integer increment and so without generating the exceptions required). Having fixed the bug, the previously deferred conversion of nextafter testing in libm-test.inc to ALL_RM_TEST is also included. Tested x86_64 and x86; also spot-checked results of nextafter tests for powerpc32 and mips64 to test the ldbl-128ibm and ldbl-128 changes. (The m68k change is untested.) 2014-03-11 Joseph Myers [BZ #16677] * math/s_nextafter.c (__nextafter): Do not return value from overflowing computation. * sysdeps/i386/fpu/s_nextafterl.c (__nextafterl): Likewise. * sysdeps/ieee754/flt-32/s_nextafterf.c (__nextafterf): Likewise. * sysdeps/ieee754/ldbl-128/s_nextafterl.c (__nextafterl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c (__nextafterl): Likewise. * sysdeps/m68k/m680x0/fpu/s_nextafterl.c (__nextafterl): Likewise. * math/libm-test.inc (nextafter_test): Use ALL_RM_TEST. diff --git a/math/libm-test.inc b/math/libm-test.inc index 0fe0f69..574654e 100644 --- a/math/libm-test.inc +++ b/math/libm-test.inc @@ -9553,10 +9553,7 @@ static const struct test_ff_f_data nextafter_test_data[] = static void nextafter_test (void) { - - START (nextafter, 1); - RUN_TEST_LOOP_ff_f (nextafter, nextafter_test_data, ); - END; + ALL_RM_TEST (nextafter, 1, nextafter_test_data, RUN_TEST_LOOP_ff_f, END); } diff --git a/math/s_nextafter.c b/math/s_nextafter.c index 7b026f0..28962e5 100644 --- a/math/s_nextafter.c +++ b/math/s_nextafter.c @@ -70,10 +70,8 @@ double __nextafter(double x, double y) } hy = hx&0x7ff00000; if(hy>=0x7ff00000) { - x = x+x; /* overflow */ - if (FLT_EVAL_METHOD != 0 && FLT_EVAL_METHOD != 1) - asm ("" : "+m"(x)); - return x; /* overflow */ + double u = x+x; /* overflow */ + math_force_eval (u); } if(hy<0x00100000) { double u = x*x; /* underflow */ diff --git a/sysdeps/i386/fpu/s_nextafterl.c b/sysdeps/i386/fpu/s_nextafterl.c index bafe743..66d903f 100644 --- a/sysdeps/i386/fpu/s_nextafterl.c +++ b/sysdeps/i386/fpu/s_nextafterl.c @@ -106,7 +106,10 @@ long double __nextafterl(long double x, long double y) } } esy = esx&0x7fff; - if(esy==0x7fff) return x+x; /* overflow */ + if(esy==0x7fff) { + long double u = x + x; /* overflow */ + math_force_eval (u); + } if(esy==0) { long double u = x*x; /* underflow */ math_force_eval (u); /* raise underflow flag */ diff --git a/sysdeps/ieee754/flt-32/s_nextafterf.c b/sysdeps/ieee754/flt-32/s_nextafterf.c index b0de3d9..22e0b3d 100644 --- a/sysdeps/ieee754/flt-32/s_nextafterf.c +++ b/sysdeps/ieee754/flt-32/s_nextafterf.c @@ -57,10 +57,8 @@ float __nextafterf(float x, float y) } hy = hx&0x7f800000; if(hy>=0x7f800000) { - x = x+x; /* overflow */ - if (FLT_EVAL_METHOD != 0) - asm ("" : "+m"(x)); - return x; /* overflow */ + float u = x+x; /* overflow */ + math_force_eval (u); } if(hy<0x00800000) { float u = x*x; /* underflow */ diff --git a/sysdeps/ieee754/ldbl-128/s_nextafterl.c b/sysdeps/ieee754/ldbl-128/s_nextafterl.c index e345bc8..d5eaa1c 100644 --- a/sysdeps/ieee754/ldbl-128/s_nextafterl.c +++ b/sysdeps/ieee754/ldbl-128/s_nextafterl.c @@ -67,7 +67,10 @@ long double __nextafterl(long double x, long double y) } } hy = hx&0x7fff000000000000LL; - if(hy==0x7fff000000000000LL) return x+x;/* overflow */ + if(hy==0x7fff000000000000LL) { + long double u = x + x; /* overflow */ + math_force_eval (u); + } if(hy==0) { long double u = x*x; /* underflow */ math_force_eval (u); /* raise underflow flag */ diff --git a/sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c b/sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c index c050944..30b1540 100644 --- a/sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c +++ b/sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c @@ -66,8 +66,11 @@ long double __nextafterl(long double x, long double y) long double with a 106 bit mantissa, and nextafterl is insane with variable precision. So to make nextafterl sane we assume 106 bit precision. */ - if((hx==0xffefffffffffffffLL)&&(lx==0xfc8ffffffffffffeLL)) - return x+x; /* overflow, return -inf */ + if((hx==0xffefffffffffffffLL)&&(lx==0xfc8ffffffffffffeLL)) { + u = x+x; /* overflow, return -inf */ + math_force_eval (u); + return y; + } if (hx >= 0x7ff0000000000000LL) { u = 0x1.fffffffffffff7ffffffffffff8p+1023L; return u; @@ -93,8 +96,11 @@ long double __nextafterl(long double x, long double y) } return x - u; } else { /* x < y, x += ulp */ - if((hx==0x7fefffffffffffffLL)&&(lx==0x7c8ffffffffffffeLL)) - return x+x; /* overflow, return +inf */ + if((hx==0x7fefffffffffffffLL)&&(lx==0x7c8ffffffffffffeLL)) { + u = x+x; /* overflow, return +inf */ + math_force_eval (u); + return y; + } if ((uint64_t) hx >= 0xfff0000000000000ULL) { u = -0x1.fffffffffffff7ffffffffffff8p+1023L; return u; diff --git a/sysdeps/m68k/m680x0/fpu/s_nextafterl.c b/sysdeps/m68k/m680x0/fpu/s_nextafterl.c index 03c136b..ad77ca4 100644 --- a/sysdeps/m68k/m680x0/fpu/s_nextafterl.c +++ b/sysdeps/m68k/m680x0/fpu/s_nextafterl.c @@ -89,7 +89,10 @@ long double __nextafterl(long double x, long double y) } } esy = esx&0x7fff; - if(esy==0x7fff) return x+x; /* overflow */ + if(esy==0x7fff) { + long double u = x + x; /* overflow */ + math_force_eval (u); + } if(esy==0 && (hx & 0x80000000) == 0) { /* underflow */ y = x*x; math_force_eval (y); /* raise underflow flag */