From patchwork Thu Sep 24 23:44:30 2015 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Joseph Myers X-Patchwork-Id: 8846 Received: (qmail 6426 invoked by alias); 24 Sep 2015 23:44:40 -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 6218 invoked by uid 89); 24 Sep 2015 23:44:40 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-2.3 required=5.0 tests=AWL, BAYES_00, RCVD_IN_DNSWL_LOW, SPF_PASS autolearn=ham version=3.3.2 X-HELO: relay1.mentorg.com Date: Thu, 24 Sep 2015 23:44:30 +0000 From: Joseph Myers To: Subject: Fix hypot missing underflows (bug 18803) [committed] Message-ID: User-Agent: Alpine 2.10 (DEB 1266 2009-07-14) MIME-Version: 1.0 Similar to various other bugs in this area, hypot functions can fail to raise the underflow exception when the result is tiny and inexact but one or more low bits of the intermediate result that is scaled down (or, in the i386 case, converted from a wider evaluation format) are zero. This patch forces the exception in a similar way to previous fixes. Note that this issue cannot arise for implementations of hypotf using double (or wider) for intermediate evaluation (if hypotf should underflow, that means the double square root is being computed of some number of the form N*2^-298, for 0 < N < 2^46, which is exactly represented as a double, and whatever the rounding mode such a square root cannot have a mantissa with all zeroes after the initial 23 bits). Thus no changes are made to hypotf implementations in this patch, only to hypot and hypotl. Tested for x86_64, x86, mips64 and powerpc. Committed. (auto-libm-test-out diffs omitted below.) 2015-09-24 Joseph Myers [BZ #18803] * sysdeps/i386/fpu/e_hypot.S: Use DEFINE_DBL_MIN. (MO): New macro. (__ieee754_hypot) [PIC]: Load PIC register. (__ieee754_hypot): Use DBL_NARROW_EVAL_UFLOW_NONNEG instead of DBL_NARROW_EVAL. * sysdeps/ieee754/dbl-64/e_hypot.c (__ieee754_hypot): Use math_check_force_underflow_nonneg in case where result might be tiny. * sysdeps/ieee754/ldbl-128/e_hypotl.c (__ieee754_hypotl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_hypotl.c (__ieee754_hypotl): Likewise. * sysdeps/ieee754/ldbl-96/e_hypotl.c (__ieee754_hypotl): Likewise. * sysdeps/powerpc/fpu/e_hypot.c (__ieee754_hypot): Likewise. * math/auto-libm-test-in: Add more tests of hypot. * math/auto-libm-test-out: Regenerated. diff --git a/math/auto-libm-test-in b/math/auto-libm-test-in index 8dea2b4..2403216 100644 --- a/math/auto-libm-test-in +++ b/math/auto-libm-test-in @@ -1955,6 +1955,13 @@ hypot 0x0.ffffffp-16382 0x0.fp-16383 no-test-inline hypot 0x0.ffffffp-16382 0x0.fp-16386 no-test-inline hypot 0 min_subnorm no-test-inline +hypot 0x1p-127 0x1p-149 +hypot 0x1p-1023 0x1p-1074 +hypot 0x1p-970 0x1p-1074 +hypot 0x1p-16383 0x1p-16445 no-test-inline +hypot 0x1p-16384 0x1p-16446 no-test-inline +hypot 0x1p-16383 0x1p-16494 no-test-inline + hypot -0x1.fa7deap+0 0x1.a761bab383ac8p+0 j0 -1.0 diff --git a/sysdeps/i386/fpu/e_hypot.S b/sysdeps/i386/fpu/e_hypot.S index 5323fde..e69ea09 100644 --- a/sysdeps/i386/fpu/e_hypot.S +++ b/sysdeps/i386/fpu/e_hypot.S @@ -20,8 +20,19 @@ #include #include +DEFINE_DBL_MIN + +#ifdef PIC +# define MO(op) op##@GOTOFF(%edx) +#else +# define MO(op) op +#endif + .text ENTRY(__ieee754_hypot) +#ifdef PIC + LOAD_PIC_REG (dx) +#endif fldl 4(%esp) // x fxam fnstsw @@ -38,7 +49,7 @@ ENTRY(__ieee754_hypot) fmul %st(0) // x * x : y * y faddp // x * x + y * y fsqrt - DBL_NARROW_EVAL + DBL_NARROW_EVAL_UFLOW_NONNEG 2: ret // We have to test whether any of the parameters is Inf. diff --git a/sysdeps/ieee754/dbl-64/e_hypot.c b/sysdeps/ieee754/dbl-64/e_hypot.c index 5cbfcbe..f142c45 100644 --- a/sysdeps/ieee754/dbl-64/e_hypot.c +++ b/sysdeps/ieee754/dbl-64/e_hypot.c @@ -149,7 +149,9 @@ __ieee754_hypot (double x, double y) t1 = 1.0; GET_HIGH_WORD (high, t1); SET_HIGH_WORD (t1, high + (k << 20)); - return t1 * w; + w *= t1; + math_check_force_underflow_nonneg (w); + return w; } else return w; diff --git a/sysdeps/ieee754/ldbl-128/e_hypotl.c b/sysdeps/ieee754/ldbl-128/e_hypotl.c index 01444cf..80e5e38 100644 --- a/sysdeps/ieee754/ldbl-128/e_hypotl.c +++ b/sysdeps/ieee754/ldbl-128/e_hypotl.c @@ -130,7 +130,9 @@ __ieee754_hypotl(long double x, long double y) t1 = 1.0L; GET_LDOUBLE_MSW64(high,t1); SET_LDOUBLE_MSW64(t1,high+(k<<48)); - return t1*w; + w *= t1; + math_check_force_underflow_nonneg (w); + return w; } else return w; } strong_alias (__ieee754_hypotl, __hypotl_finite) diff --git a/sysdeps/ieee754/ldbl-128ibm/e_hypotl.c b/sysdeps/ieee754/ldbl-128ibm/e_hypotl.c index 3b07a47..c68dac0 100644 --- a/sysdeps/ieee754/ldbl-128ibm/e_hypotl.c +++ b/sysdeps/ieee754/ldbl-128ibm/e_hypotl.c @@ -125,7 +125,11 @@ __ieee754_hypotl(long double x, long double y) w = __ieee754_sqrtl(a1*b1-(w*(-w)-(a1*b2+a2*b))); } if(k!=0) - return w*kld; + { + w *= kld; + math_check_force_underflow_nonneg (w); + return w; + } else return w; } diff --git a/sysdeps/ieee754/ldbl-96/e_hypotl.c b/sysdeps/ieee754/ldbl-96/e_hypotl.c index d3152f9..ee3a070 100644 --- a/sysdeps/ieee754/ldbl-96/e_hypotl.c +++ b/sysdeps/ieee754/ldbl-96/e_hypotl.c @@ -132,7 +132,9 @@ long double __ieee754_hypotl(long double x, long double y) t1 = 1.0; GET_LDOUBLE_EXP(exp,t1); SET_LDOUBLE_EXP(t1,exp+k); - return t1*w; + w *= t1; + math_check_force_underflow_nonneg (w); + return w; } else return w; } strong_alias (__ieee754_hypotl, __hypotl_finite) diff --git a/sysdeps/powerpc/fpu/e_hypot.c b/sysdeps/powerpc/fpu/e_hypot.c index 98e4ae6..a3b9d3d 100644 --- a/sysdeps/powerpc/fpu/e_hypot.c +++ b/sysdeps/powerpc/fpu/e_hypot.c @@ -116,7 +116,9 @@ __ieee754_hypot (double x, double y) { x *= two1022; y *= two1022; - return __ieee754_sqrt (x * x + y * y) / two1022; + double ret = __ieee754_sqrt (x * x + y * y) / two1022; + math_check_force_underflow_nonneg (ret); + return ret; } else {