From patchwork Fri Sep 18 20:01:24 2015 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Joseph Myers X-Patchwork-Id: 8797 Received: (qmail 113250 invoked by alias); 18 Sep 2015 20:01:32 -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 113233 invoked by uid 89); 18 Sep 2015 20:01:31 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-0.9 required=5.0 tests=AWL, BAYES_50, RCVD_IN_DNSWL_LOW, SPF_PASS autolearn=ham version=3.3.2 X-HELO: relay1.mentorg.com Date: Fri, 18 Sep 2015 20:01:24 +0000 From: Joseph Myers To: Subject: Avoid excess range overflowing results from cosh, sinh, lgamma (bug 18980) [committed] Message-ID: User-Agent: Alpine 2.10 (DEB 1266 2009-07-14) MIME-Version: 1.0 Various i386 libm functions return values with excess range and precision; Wilco Dijkstra's patches to make isfinite etc. expand inline cause this pre-existing issue to result in test failures (when e.g. a result that overflows float but not long double gets counted as overflowing for some purposes but not others). This patch addresses those cases arising from functions defined in C, adding a math_narrow_eval macro that forces values to memory to eliminate excess precision if FLT_EVAL_METHOD indicates this is needed, and is a no-op otherwise. I'll convert existing uses of volatile and asm for this purpose to use the new macro later, once i386 has clean test results again (which requires fixes for .S files as well). Tested for x86_64 and x86. Committed. 2015-09-18 Joseph Myers [BZ #18980] * sysdeps/generic/math_private.h: Include . (math_narrow_eval): New macro. [FLT_EVAL_METHOD != 0] (excess_precision): Likewise. * sysdeps/ieee754/dbl-64/e_cosh.c (__ieee754_cosh): Use math_narrow_eval on overflowing return value. * sysdeps/ieee754/dbl-64/e_lgamma_r.c (__ieee754_lgamma_r): Likewise. * sysdeps/ieee754/dbl-64/e_sinh.c (__ieee754_sinh): Likewise. * sysdeps/ieee754/flt-32/e_coshf.c (__ieee754_coshf): Likewise. * sysdeps/ieee754/flt-32/e_lgammaf_r.c (__ieee754_lgammaf_r): Likewise. * sysdeps/ieee754/flt-32/e_sinhf.c (__ieee754_sinhf): Likewise. diff --git a/sysdeps/generic/math_private.h b/sysdeps/generic/math_private.h index 6aea864..35591ee 100644 --- a/sysdeps/generic/math_private.h +++ b/sysdeps/generic/math_private.h @@ -20,6 +20,7 @@ #include #include #include +#include #include /* The original fdlibm code used statements like: @@ -405,6 +406,29 @@ extern long double __lgamma_productl (long double t, long double x, ({ __typeof (x) __x = (x); __asm __volatile__ ("" : : "m" (__x)); }) #endif +/* math_narrow_eval reduces its floating-point argument to the range + and precision of its semantic type. (The original evaluation may + still occur with excess range and precision, so the result may be + affected by double rounding.) */ +#if FLT_EVAL_METHOD == 0 +# define math_narrow_eval(x) (x) +#else +# if FLT_EVAL_METHOD == 1 +# define excess_precision(type) __builtin_types_compatible_p (type, float) +# else +# define excess_precision(type) (__builtin_types_compatible_p (type, float) \ + || __builtin_types_compatible_p (type, \ + double)) +# endif +# define math_narrow_eval(x) \ + ({ \ + __typeof (x) math_narrow_eval_tmp = (x); \ + if (excess_precision (__typeof (math_narrow_eval_tmp))) \ + __asm__ ("" : "+m" (math_narrow_eval_tmp)); \ + math_narrow_eval_tmp; \ + }) +#endif + /* The standards only specify one variant of the fenv.h interfaces. But at least for some architectures we can be more efficient if we diff --git a/sysdeps/ieee754/dbl-64/e_cosh.c b/sysdeps/ieee754/dbl-64/e_cosh.c index af3910d..52a5d50 100644 --- a/sysdeps/ieee754/dbl-64/e_cosh.c +++ b/sysdeps/ieee754/dbl-64/e_cosh.c @@ -83,6 +83,6 @@ __ieee754_cosh (double x) return x * x; /* |x| > overflowthresold, cosh(x) overflow */ - return huge * huge; + return math_narrow_eval (huge * huge); } strong_alias (__ieee754_cosh, __cosh_finite) diff --git a/sysdeps/ieee754/dbl-64/e_lgamma_r.c b/sysdeps/ieee754/dbl-64/e_lgamma_r.c index ea8a9b4..da158cb 100644 --- a/sysdeps/ieee754/dbl-64/e_lgamma_r.c +++ b/sysdeps/ieee754/dbl-64/e_lgamma_r.c @@ -296,7 +296,7 @@ __ieee754_lgamma_r(double x, int *signgamp) r = (x-half)*(t-one)+w; } else /* 2**58 <= x <= inf */ - r = x*(__ieee754_log(x)-one); + r = math_narrow_eval (x*(__ieee754_log(x)-one)); /* NADJ is set for negative arguments but not otherwise, resulting in warnings that it may be used uninitialized although in the cases where it is used it has always been diff --git a/sysdeps/ieee754/dbl-64/e_sinh.c b/sysdeps/ieee754/dbl-64/e_sinh.c index c99d283..291bfad 100644 --- a/sysdeps/ieee754/dbl-64/e_sinh.c +++ b/sysdeps/ieee754/dbl-64/e_sinh.c @@ -89,6 +89,6 @@ __ieee754_sinh (double x) } /* |x| > overflowthresold, sinh(x) overflow */ - return x * shuge; + return math_narrow_eval (x * shuge); } strong_alias (__ieee754_sinh, __sinh_finite) diff --git a/sysdeps/ieee754/flt-32/e_coshf.c b/sysdeps/ieee754/flt-32/e_coshf.c index dedda47..7b22375 100644 --- a/sysdeps/ieee754/flt-32/e_coshf.c +++ b/sysdeps/ieee754/flt-32/e_coshf.c @@ -58,6 +58,6 @@ __ieee754_coshf (float x) if(ix>=0x7f800000) return x*x; /* |x| > overflowthresold, cosh(x) overflow */ - return huge*huge; + return math_narrow_eval (huge*huge); } strong_alias (__ieee754_coshf, __coshf_finite) diff --git a/sysdeps/ieee754/flt-32/e_lgammaf_r.c b/sysdeps/ieee754/flt-32/e_lgammaf_r.c index 424c4e7..45a62c0 100644 --- a/sysdeps/ieee754/flt-32/e_lgammaf_r.c +++ b/sysdeps/ieee754/flt-32/e_lgammaf_r.c @@ -232,7 +232,7 @@ __ieee754_lgammaf_r(float x, int *signgamp) r = (x-half)*(t-one)+w; } else /* 2**26 <= x <= inf */ - r = x*(__ieee754_logf(x)-one); + r = math_narrow_eval (x*(__ieee754_logf(x)-one)); /* NADJ is set for negative arguments but not otherwise, resulting in warnings that it may be used uninitialized although in the cases where it is used it has always been diff --git a/sysdeps/ieee754/flt-32/e_sinhf.c b/sysdeps/ieee754/flt-32/e_sinhf.c index 17c2219..a24fa0c 100644 --- a/sysdeps/ieee754/flt-32/e_sinhf.c +++ b/sysdeps/ieee754/flt-32/e_sinhf.c @@ -59,6 +59,6 @@ __ieee754_sinhf(float x) } /* |x| > overflowthresold, sinh(x) overflow */ - return x*shuge; + return math_narrow_eval (x*shuge); } strong_alias (__ieee754_sinhf, __sinhf_finite)