From 3e336a8754678d7a4ef7348c129e6f6e37523798 Mon Sep 17 00:00:00 2001 From: Ulrich Drepper Date: Mon, 16 Apr 2007 20:41:42 +0000 Subject: [PATCH] [BZ #3306] 2007-03-27 Jakub Jelinek [BZ #3306] * math/math_private.h (math_opt_barrier, math_force_eval): Define. * sysdeps/i386/fpu/math_private.h: New file. * sysdeps/x86_64/fpu/math_private.h: New file. * math/s_nexttowardf.c (__nexttowardf): Use math_opt_barrier and math_force_eval macros. Use "+m" constraint on asm rather than "=m" and "m". * math/s_nextafter.c (__nextafter): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c (__nexttoward): Likewise. * sysdeps/ieee754/flt-32/s_nextafterf.c (__nextafterf): Likewise. * sysdeps/ieee754/ldbl-128/s_nexttoward.c (__nexttoward): Likewise. * sysdeps/ieee754/ldbl-96/s_nexttoward.c (__nexttoward): Likewise. * sysdeps/i386/fpu/s_nextafterl.c (__nextafterl): Use math_opt_barrier and math_force_eval macros. * sysdeps/ieee754/ldbl-128/s_nextafterl.c (__nextafterl): Likewise. * sysdeps/ieee754/ldbl-96/s_nextafterl.c (__nextafterl): Likewise. * sysdeps/i386/fpu/s_nexttoward.c: Include float.h. (__nexttoward): Use math_opt_barrier and math_force_eval macros. Use "+m" constraint on asm rather than "=m" and "m". Only use asm to force double result if FLT_EVAL_METHOD is 2. * sysdeps/i386/fpu/s_nexttowardf.c: Include float.h. (__nexttowardf): Use math_opt_barrier and math_force_eval macros. Use "+m" constraint on asm rather than "=m" and "m". Only use asm to force double result if FLT_EVAL_METHOD is not 0. * sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c: Include float.h. (__nexttowardf): Use math_opt_barrier and math_force_eval macros. If FLT_EVAL_METHOD is not 0, force x to float using asm. * sysdeps/ieee754/ldbl-opt/s_nexttowardfd.c: Include float.h. (__nldbl_nexttowardf): Use math_opt_barrier and math_force_eval macros. If FLT_EVAL_METHOD is not 0, force x to float using asm. * sysdeps/ieee754/ldbl-96/s_nexttowardf.c: Include float.h. (__nexttowardf): Use math_opt_barrier and math_force_eval macros. If FLT_EVAL_METHOD is not 0, force x to float using asm. * math/bug-nextafter.c (zero, inf): New variables. (main): Add new tests. * math/bug-nexttoward.c (zero, inf): New variables. (main): Add new tests. --- ChangeLog | 45 +++++ math/bug-nextafter.c | 255 ++++++++++++++++++++++++++++ math/bug-nexttoward.c | 255 ++++++++++++++++++++++++++++ math/math_private.h | 6 + math/s_nextafter.c | 20 +-- math/s_nexttowardf.c | 21 ++- sysdeps/i386/fpu/math_private.h | 18 ++ sysdeps/i386/fpu/s_nexttoward.c | 25 +-- sysdeps/i386/fpu/s_nexttowardf.c | 25 +-- sysdeps/ieee754/flt-32/s_nextafterf.c | 20 +-- sysdeps/ieee754/ldbl-128/s_nexttoward.c | 23 ++- sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c | 23 ++- sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c | 26 +-- sysdeps/ieee754/ldbl-96/s_nexttoward.c | 23 ++- sysdeps/ieee754/ldbl-96/s_nexttowardf.c | 28 +-- sysdeps/ieee754/ldbl-opt/s_nexttowardfd.c | 28 +-- sysdeps/x86_64/fpu/math_private.h | 21 +++ 17 files changed, 740 insertions(+), 122 deletions(-) create mode 100644 sysdeps/i386/fpu/math_private.h create mode 100644 sysdeps/x86_64/fpu/math_private.h diff --git a/ChangeLog b/ChangeLog index 8463b905cb..f38e61d929 100644 --- a/ChangeLog +++ b/ChangeLog @@ -1,3 +1,48 @@ +2007-03-27 Jakub Jelinek + + [BZ #3306] + * math/math_private.h (math_opt_barrier, math_force_eval): Define. + * sysdeps/i386/fpu/math_private.h: New file. + * sysdeps/x86_64/fpu/math_private.h: New file. + * math/s_nexttowardf.c (__nexttowardf): Use math_opt_barrier and + math_force_eval macros. Use "+m" constraint on asm rather than + "=m" and "m". + * math/s_nextafter.c (__nextafter): Likewise. + * sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c (__nexttoward): + Likewise. + * sysdeps/ieee754/flt-32/s_nextafterf.c (__nextafterf): Likewise. + * sysdeps/ieee754/ldbl-128/s_nexttoward.c (__nexttoward): Likewise. + * sysdeps/ieee754/ldbl-96/s_nexttoward.c (__nexttoward): Likewise. + * sysdeps/i386/fpu/s_nextafterl.c (__nextafterl): Use + math_opt_barrier and math_force_eval macros. + * sysdeps/ieee754/ldbl-128/s_nextafterl.c (__nextafterl): Likewise. + * sysdeps/ieee754/ldbl-96/s_nextafterl.c (__nextafterl): Likewise. + * sysdeps/i386/fpu/s_nexttoward.c: Include float.h. + (__nexttoward): Use math_opt_barrier and + math_force_eval macros. Use "+m" constraint on asm rather than + "=m" and "m". Only use asm to force double result if + FLT_EVAL_METHOD is 2. + * sysdeps/i386/fpu/s_nexttowardf.c: Include float.h. + (__nexttowardf): Use math_opt_barrier and + math_force_eval macros. Use "+m" constraint on asm rather than + "=m" and "m". Only use asm to force double result if + FLT_EVAL_METHOD is not 0. + * sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c: Include float.h. + (__nexttowardf): Use math_opt_barrier and + math_force_eval macros. If FLT_EVAL_METHOD is not 0, force + x to float using asm. + * sysdeps/ieee754/ldbl-opt/s_nexttowardfd.c: Include float.h. + (__nldbl_nexttowardf): Use math_opt_barrier and + math_force_eval macros. If FLT_EVAL_METHOD is not 0, force + x to float using asm. + * sysdeps/ieee754/ldbl-96/s_nexttowardf.c: Include float.h. + (__nexttowardf): Use math_opt_barrier and math_force_eval + macros. If FLT_EVAL_METHOD is not 0, force x to float using asm. + * math/bug-nextafter.c (zero, inf): New variables. + (main): Add new tests. + * math/bug-nexttoward.c (zero, inf): New variables. + (main): Add new tests. + 2007-03-22 Jakub Jelinek [BZ #3427] diff --git a/math/bug-nextafter.c b/math/bug-nextafter.c index 2a967c75bf..1d21841ea6 100644 --- a/math/bug-nextafter.c +++ b/math/bug-nextafter.c @@ -4,6 +4,9 @@ #include #include +float zero = 0.0; +float inf = INFINITY; + int main (void) { @@ -34,6 +37,81 @@ main (void) ++result; } + i = 0; + m = FLT_MIN; + feclearexcept (FE_ALL_EXCEPT); + i = nextafterf (m, i); + if (i < 0 || i >= FLT_MIN) + { + puts ("nextafterf+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterf+ did not underflow"); + ++result; + } + i = 0; + feclearexcept (FE_ALL_EXCEPT); + i = nextafterf (-m, -i); + if (i > 0 || i <= -FLT_MIN) + { + puts ("nextafterf- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterf- did not underflow"); + ++result; + } + i = -INFINITY; + feclearexcept (FE_ALL_EXCEPT); + m = nextafterf (zero, inf); + if (m < 0.0 || m >= FLT_MIN) + { + puts ("nextafterf+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterf+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nextafterf (m, i) != 0.0) + { + puts ("nextafterf+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterf+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + m = nextafterf (copysignf (zero, -1.0), -inf); + if (m > 0.0 || m <= -FLT_MIN) + { + puts ("nextafterf- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterf- did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nextafterf (m, -i) != 0.0) + { + puts ("nextafterf- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterf- did not underflow"); + ++result; + } + double di = INFINITY; double dm = DBL_MAX; feclearexcept (FE_ALL_EXCEPT); @@ -59,5 +137,182 @@ main (void) ++result; } + di = 0; + dm = DBL_MIN; + feclearexcept (FE_ALL_EXCEPT); + di = nextafter (dm, di); + if (di < 0 || di >= DBL_MIN) + { + puts ("nextafter+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafter+ did not underflow"); + ++result; + } + di = 0; + feclearexcept (FE_ALL_EXCEPT); + di = nextafter (-dm, -di); + if (di > 0 || di <= -DBL_MIN) + { + puts ("nextafter- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafter- did not underflow"); + ++result; + } + di = -INFINITY; + feclearexcept (FE_ALL_EXCEPT); + dm = nextafter (zero, inf); + if (dm < 0.0 || dm >= DBL_MIN) + { + puts ("nextafter+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafter+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nextafter (dm, di) != 0.0) + { + puts ("nextafter+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafter+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + dm = nextafter (copysign (zero, -1.0), -inf); + if (dm > 0.0 || dm <= -DBL_MIN) + { + puts ("nextafter- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafter- did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nextafter (dm, -di) != 0.0) + { + puts ("nextafter- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafter- did not underflow"); + ++result; + } + +#ifndef NO_LONG_DOUBLE + long double li = INFINITY; + long double lm = LDBL_MAX; + feclearexcept (FE_ALL_EXCEPT); + if (nextafterl (lm, li) != li) + { + puts ("nextafterl+ failed"); + ++result; + } + if (fetestexcept (FE_OVERFLOW) == 0) + { + puts ("nextafterl+ did not overflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nextafterl (-lm, -li) != -li) + { + puts ("nextafterl failed"); + ++result; + } + if (fetestexcept (FE_OVERFLOW) == 0) + { + puts ("nextafterl- did not overflow"); + ++result; + } + + li = 0; + lm = LDBL_MIN; + feclearexcept (FE_ALL_EXCEPT); + li = nextafterl (lm, li); + if (li < 0 || li >= LDBL_MIN) + { + puts ("nextafterl+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterl+ did not underflow"); + ++result; + } + li = 0; + feclearexcept (FE_ALL_EXCEPT); + li = nextafterl (-lm, -li); + if (li > 0 || li <= -LDBL_MIN) + { + puts ("nextafterl- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterl- did not underflow"); + ++result; + } + li = -INFINITY; + feclearexcept (FE_ALL_EXCEPT); + lm = nextafterl (zero, inf); + if (lm < 0.0 || lm >= LDBL_MIN) + { + puts ("nextafterl+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterl+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nextafterl (lm, li) != 0.0) + { + puts ("nextafterl+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterl+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + lm = nextafterl (copysign (zero, -1.0), -inf); + if (lm > 0.0 || lm <= -LDBL_MIN) + { + puts ("nextafterl- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterl- did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nextafterl (lm, -li) != 0.0) + { + puts ("nextafterl- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nextafterl- did not underflow"); + ++result; + } +#endif + return result; } diff --git a/math/bug-nexttoward.c b/math/bug-nexttoward.c index e306a129c2..ff57e5e3f5 100644 --- a/math/bug-nexttoward.c +++ b/math/bug-nexttoward.c @@ -4,6 +4,9 @@ #include #include +float zero = 0.0; +float inf = INFINITY; + int main (void) { @@ -35,6 +38,81 @@ main (void) ++result; } + fi = 0; + m = FLT_MIN; + feclearexcept (FE_ALL_EXCEPT); + fi = nexttowardf (m, fi); + if (fi < 0 || fi >= FLT_MIN) + { + puts ("nexttowardf+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardf+ did not underflow"); + ++result; + } + fi = 0; + feclearexcept (FE_ALL_EXCEPT); + fi = nexttowardf (-m, -fi); + if (fi > 0 || fi <= -FLT_MIN) + { + puts ("nexttowardf- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardf- did not underflow"); + ++result; + } + fi = -INFINITY; + feclearexcept (FE_ALL_EXCEPT); + m = nexttowardf (zero, inf); + if (m < 0.0 || m >= FLT_MIN) + { + puts ("nexttowardf+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardf+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nexttowardf (m, fi) != 0.0) + { + puts ("nexttowardf+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardf+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + m = nexttowardf (copysignf (zero, -1.0), -inf); + if (m > 0.0 || m <= -FLT_MIN) + { + puts ("nexttowardf- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardf- did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nexttowardf (m, -fi) != 0.0) + { + puts ("nexttowardf- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardf- did not underflow"); + ++result; + } + tl = (long double) DBL_MAX + 1.0e305L; double di = INFINITY; double dm = DBL_MAX; @@ -61,5 +139,182 @@ main (void) ++result; } + di = 0; + dm = DBL_MIN; + feclearexcept (FE_ALL_EXCEPT); + di = nexttoward (dm, di); + if (di < 0 || di >= DBL_MIN) + { + puts ("nexttoward+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttoward+ did not underflow"); + ++result; + } + di = 0; + feclearexcept (FE_ALL_EXCEPT); + di = nexttoward (-dm, -di); + if (di > 0 || di <= -DBL_MIN) + { + puts ("nexttoward- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttoward- did not underflow"); + ++result; + } + di = -INFINITY; + feclearexcept (FE_ALL_EXCEPT); + dm = nexttoward (zero, inf); + if (dm < 0.0 || dm >= DBL_MIN) + { + puts ("nexttoward+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttoward+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nexttoward (dm, di) != 0.0) + { + puts ("nexttoward+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttoward+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + dm = nexttoward (copysign (zero, -1.0), -inf); + if (dm > 0.0 || dm <= -DBL_MIN) + { + puts ("nexttoward- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttoward- did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nexttoward (dm, -di) != 0.0) + { + puts ("nexttoward- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttoward- did not underflow"); + ++result; + } + +#ifndef NO_LONG_DOUBLE + long double li = INFINITY; + long double lm = LDBL_MAX; + feclearexcept (FE_ALL_EXCEPT); + if (nexttowardl (lm, li) != li) + { + puts ("nexttowardl+ failed"); + ++result; + } + if (fetestexcept (FE_OVERFLOW) == 0) + { + puts ("nexttowardl+ did not overflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nexttowardl (-lm, -li) != -li) + { + puts ("nexttowardl failed"); + ++result; + } + if (fetestexcept (FE_OVERFLOW) == 0) + { + puts ("nexttowardl- did not overflow"); + ++result; + } + + li = 0; + lm = LDBL_MIN; + feclearexcept (FE_ALL_EXCEPT); + li = nexttowardl (lm, li); + if (li < 0 || li >= LDBL_MIN) + { + puts ("nexttowardl+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardl+ did not underflow"); + ++result; + } + li = 0; + feclearexcept (FE_ALL_EXCEPT); + li = nexttowardl (-lm, -li); + if (li > 0 || li <= -LDBL_MIN) + { + puts ("nexttowardl- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardl- did not underflow"); + ++result; + } + li = -INFINITY; + feclearexcept (FE_ALL_EXCEPT); + lm = nexttowardl (zero, inf); + if (lm < 0.0 || lm >= LDBL_MIN) + { + puts ("nexttowardl+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardl+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nexttowardl (lm, li) != 0.0) + { + puts ("nexttowardl+ failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardl+ did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + lm = nexttowardl (copysign (zero, -1.0), -inf); + if (lm > 0.0 || lm <= -LDBL_MIN) + { + puts ("nexttowardl- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardl- did not underflow"); + ++result; + } + feclearexcept (FE_ALL_EXCEPT); + if (nexttowardl (lm, -li) != 0.0) + { + puts ("nexttowardl- failed"); + ++result; + } + if (fetestexcept (FE_UNDERFLOW) == 0) + { + puts ("nexttowardl- did not underflow"); + ++result; + } +#endif + return result; } diff --git a/math/math_private.h b/math/math_private.h index 572e546831..129646f8c5 100644 --- a/math/math_private.h +++ b/math/math_private.h @@ -332,4 +332,10 @@ extern double __slowexp (double __x); extern double __slowpow (double __x, double __y, double __z); extern void __docos (double __x, double __dx, double __v[]); +#ifndef math_opt_barrier +#define math_opt_barrier(x) \ +({ __typeof (x) __x = x; __asm ("" : "+m" (__x)); __x; }) +#define math_force_eval(x) __asm __volatile ("" : : "m" (x)) +#endif + #endif /* _MATH_PRIVATE_H_ */ diff --git a/math/s_nextafter.c b/math/s_nextafter.c index 9c678b79c5..d2af52dda3 100644 --- a/math/s_nextafter.c +++ b/math/s_nextafter.c @@ -26,7 +26,7 @@ static char rcsid[] = "$NetBSD: s_nextafter.c,v 1.8 1995/05/10 20:47:58 jtc Exp #define nexttoward __internal_nexttoward #include -#include "math_private.h" +#include #include #ifdef __STDC__ @@ -49,9 +49,12 @@ static char rcsid[] = "$NetBSD: s_nextafter.c,v 1.8 1995/05/10 20:47:58 jtc Exp return x+y; if(x==y) return y; /* x=y, return y */ if((ix|lx)==0) { /* x == 0 */ + double u; INSERT_WORDS(x,hy&0x80000000,1); /* return +-minsubnormal */ - y = x*x; - if(y==x) return y; else return x; /* raise underflow flag */ + u = math_opt_barrier (x); + u = u*u; + math_force_eval (u); /* raise underflow flag */ + return x; } if(hx>=0) { /* x > 0 */ if(hx>hy||((hx==hy)&&(lx>ly))) { /* x > y, x -= ulp */ @@ -74,15 +77,12 @@ static char rcsid[] = "$NetBSD: s_nextafter.c,v 1.8 1995/05/10 20:47:58 jtc Exp if(hy>=0x7ff00000) { x = x+x; /* overflow */ if (FLT_EVAL_METHOD != 0 && FLT_EVAL_METHOD != 1) - asm ("" : "=m"(x) : "m"(x)); + asm ("" : "+m"(x)); return x; /* overflow */ } - if(hy<0x00100000) { /* underflow */ - y = x*x; - if(y!=x) { /* raise underflow flag */ - INSERT_WORDS(y,hx,lx); - return y; - } + if(hy<0x00100000) { + double u = x*x; /* underflow */ + math_force_eval (u); /* raise underflow flag */ } INSERT_WORDS(x,hx,lx); return x; diff --git a/math/s_nexttowardf.c b/math/s_nexttowardf.c index 1a1026612e..0494d1a4e9 100644 --- a/math/s_nexttowardf.c +++ b/math/s_nexttowardf.c @@ -21,7 +21,7 @@ */ #include -#include "math_private.h" +#include #include #ifdef __STDC__ @@ -45,10 +45,12 @@ return x+y; if((long double) x==y) return y; /* x=y, return y */ if(ix==0) { /* x == 0 */ - float x2; + float u; SET_FLOAT_WORD(x,(u_int32_t)(hy&0x80000000)|1);/* return +-minsub*/ - x2 = x*x; - if(x2==x) return x2; else return x; /* raise underflow flag */ + u = math_opt_barrier (x); + u = u * u; + math_force_eval (u); /* raise underflow flag */ + return x; } if(hx>=0) { /* x > 0 */ if(hy<0||(ix>>23)>(iy>>20)-0x380 @@ -70,15 +72,12 @@ x = x+x; /* overflow */ if (FLT_EVAL_METHOD != 0) /* Force conversion to float. */ - asm ("" : "=m"(x) : "m"(x)); + asm ("" : "+m"(x)); return x; } - if(hy<0x00800000) { /* underflow */ - float x2 = x*x; - if(x2!=x) { /* raise underflow flag */ - SET_FLOAT_WORD(x2,hx); - return x2; - } + if(hy<0x00800000) { + float u = x*x; /* underflow */ + math_force_eval (u); /* raise underflow flag */ } SET_FLOAT_WORD(x,hx); return x; diff --git a/sysdeps/i386/fpu/math_private.h b/sysdeps/i386/fpu/math_private.h new file mode 100644 index 0000000000..a426788ef1 --- /dev/null +++ b/sysdeps/i386/fpu/math_private.h @@ -0,0 +1,18 @@ +#ifndef _MATH_PRIVATE_H + +#define math_opt_barrier(x) \ +({ __typeof(x) __x; \ + __asm ("" : "=t" (__x) : "0" (x)); \ + __x; }) +#define math_force_eval(x) \ +do \ + { \ + if (sizeof (x) <= sizeof (double)) \ + __asm __volatile ("" : : "m" (x)); \ + else \ + __asm __volatile ("" : : "f" (x)); \ + } \ +while (0) + +#include +#endif diff --git a/sysdeps/i386/fpu/s_nexttoward.c b/sysdeps/i386/fpu/s_nexttoward.c index 2bd768e448..9bd86a3724 100644 --- a/sysdeps/i386/fpu/s_nexttoward.c +++ b/sysdeps/i386/fpu/s_nexttoward.c @@ -27,7 +27,8 @@ static char rcsid[] = "$NetBSD: $"; */ #include "math.h" -#include "math_private.h" +#include +#include #ifdef __STDC__ double __nexttoward(double x, long double y) @@ -52,10 +53,12 @@ static char rcsid[] = "$NetBSD: $"; return x+y; if((long double) x==y) return y; /* x=y, return y */ if((ix|lx)==0) { /* x == 0 */ - double x2; + double u; INSERT_WORDS(x,(esy&0x8000)<<16,1); /* return +-minsub */ - x2 = x*x; - if(x2==x) return x2; else return x; /* raise underflow flag */ + u = math_opt_barrier (x); + u = u * u; + math_force_eval (u); /* raise underflow flag */ + return x; } if(hx>=0) { /* x > 0 */ if (esy>=0x8000||((ix>>20)&0x7ff)>iy-0x3c00 @@ -85,16 +88,14 @@ static char rcsid[] = "$NetBSD: $"; hy = hx&0x7ff00000; if(hy>=0x7ff00000) { x = x+x; /* overflow */ - /* Force conversion to double. */ - asm ("" : "=m"(x) : "m"(x)); + if (FLT_EVAL_METHOD != 0 && FLT_EVAL_METHOD != 1) + /* Force conversion to double. */ + asm ("" : "+m"(x)); return x; } - if(hy<0x00100000) { /* underflow */ - double x2 = x*x; - if(x2!=x) { /* raise underflow flag */ - INSERT_WORDS(x2,hx,lx); - return x2; - } + if(hy<0x00100000) { + double u = x*x; /* underflow */ + math_force_eval (u); /* raise underflow flag */ } INSERT_WORDS(x,hx,lx); return x; diff --git a/sysdeps/i386/fpu/s_nexttowardf.c b/sysdeps/i386/fpu/s_nexttowardf.c index 3fbe53c338..25f70e4f4d 100644 --- a/sysdeps/i386/fpu/s_nexttowardf.c +++ b/sysdeps/i386/fpu/s_nexttowardf.c @@ -19,7 +19,8 @@ static char rcsid[] = "$NetBSD: $"; #endif #include "math.h" -#include "math_private.h" +#include +#include #ifdef __STDC__ float __nexttowardf(float x, long double y) @@ -44,10 +45,12 @@ static char rcsid[] = "$NetBSD: $"; return x+y; if((long double) x==y) return y; /* x=y, return y */ if(ix==0) { /* x == 0 */ - float x2; + float u; SET_FLOAT_WORD(x,((esy&0x8000)<<16)|1);/* return +-minsub*/ - x2 = x*x; - if(x2==x) return x2; else return x; /* raise underflow flag */ + u = math_opt_barrier (x); + u = u * u; + math_force_eval (u); /* raise underflow flag */ + return x; } if(hx>=0) { /* x > 0 */ if(esy>=0x8000||((ix>>23)&0xff)>iy-0x3f80 @@ -69,16 +72,14 @@ static char rcsid[] = "$NetBSD: $"; hy = hx&0x7f800000; if(hy>=0x7f800000) { x = x+x; /* overflow */ - /* Force conversion to float. */ - asm ("" : "=m"(x) : "m"(x)); + if (FLT_EVAL_METHOD != 0) + /* Force conversion to float. */ + asm ("" : "+m"(x)); return x; } - if(hy<0x00800000) { /* underflow */ - float x2 = x*x; - if(x2!=x) { /* raise underflow flag */ - SET_FLOAT_WORD(x2,hx); - return x2; - } + if(hy<0x00800000) { + float u = x*x; /* underflow */ + math_force_eval (u); /* raise underflow flag */ } SET_FLOAT_WORD(x,hx); return x; diff --git a/sysdeps/ieee754/flt-32/s_nextafterf.c b/sysdeps/ieee754/flt-32/s_nextafterf.c index e1568e24c9..600c14621f 100644 --- a/sysdeps/ieee754/flt-32/s_nextafterf.c +++ b/sysdeps/ieee754/flt-32/s_nextafterf.c @@ -18,7 +18,7 @@ static char rcsid[] = "$NetBSD: s_nextafterf.c,v 1.4 1995/05/10 20:48:01 jtc Exp #endif #include "math.h" -#include "math_private.h" +#include #include #ifdef __STDC__ @@ -40,9 +40,12 @@ static char rcsid[] = "$NetBSD: s_nextafterf.c,v 1.4 1995/05/10 20:48:01 jtc Exp return x+y; if(x==y) return y; /* x=y, return y */ if(ix==0) { /* x == 0 */ + float u; SET_FLOAT_WORD(x,(hy&0x80000000)|1);/* return +-minsubnormal */ - y = x*x; - if(y==x) return y; else return x; /* raise underflow flag */ + u = math_opt_barrier (x); + u = u*u; + math_force_eval (u); /* raise underflow flag */ + return x; } if(hx>=0) { /* x > 0 */ if(hx>hy) { /* x > y, x -= ulp */ @@ -61,15 +64,12 @@ static char rcsid[] = "$NetBSD: s_nextafterf.c,v 1.4 1995/05/10 20:48:01 jtc Exp if(hy>=0x7f800000) { x = x+x; /* overflow */ if (FLT_EVAL_METHOD != 0) - asm ("" : "=m"(x) : "m"(x)); + asm ("" : "+m"(x)); return x; /* overflow */ } - if(hy<0x00800000) { /* underflow */ - y = x*x; - if(y!=x) { /* raise underflow flag */ - SET_FLOAT_WORD(y,hx); - return y; - } + if(hy<0x00800000) { + float u = x*x; /* underflow */ + math_force_eval (u); /* raise underflow flag */ } SET_FLOAT_WORD(x,hx); return x; diff --git a/sysdeps/ieee754/ldbl-128/s_nexttoward.c b/sysdeps/ieee754/ldbl-128/s_nexttoward.c index 553e401973..178505c58f 100644 --- a/sysdeps/ieee754/ldbl-128/s_nexttoward.c +++ b/sysdeps/ieee754/ldbl-128/s_nexttoward.c @@ -26,7 +26,7 @@ static char rcsid[] = "$NetBSD: $"; */ #include "math.h" -#include "math_private.h" +#include #include #ifdef __STDC__ @@ -53,10 +53,12 @@ static char rcsid[] = "$NetBSD: $"; return x+y; if((long double) x==y) return y; /* x=y, return y */ if((ix|lx)==0) { /* x == 0 */ - double x2; + double u; INSERT_WORDS(x,(u_int32_t)((hy>>32)&0x80000000),1);/* return +-minsub */ - x2 = x*x; - if(x2==x) return x2; else return x; /* raise underflow flag */ + u = math_opt_barrier (x); + u = u * u; + math_force_eval (u); /* raise underflow flag */ + return x; } if(hx>=0) { /* x > 0 */ if (hy<0||(ix>>20)>(iy>>48)-0x3c00 @@ -87,16 +89,13 @@ static char rcsid[] = "$NetBSD: $"; if(hy>=0x7ff00000) { x = x+x; /* overflow */ if (FLT_EVAL_METHOD != 0 && FLT_EVAL_METHOD != 1) - /* Force conversion to float. */ - asm ("" : "=m"(x) : "m"(x)); + /* Force conversion to double. */ + asm ("" : "+m"(x)); return x; } - if(hy<0x00100000) { /* underflow */ - double x2 = x*x; - if(x2!=x) { /* raise underflow flag */ - INSERT_WORDS(x2,hx,lx); - return x2; - } + if(hy<0x00100000) { + double u = x*x; /* underflow */ + math_force_eval (u); /* raise underflow flag */ } INSERT_WORDS(x,hx,lx); return x; diff --git a/sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c b/sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c index 3335100592..e2f6521f57 100644 --- a/sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c +++ b/sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c @@ -26,7 +26,7 @@ static char rcsid[] = "$NetBSD: $"; */ #include "math.h" -#include "math_private.h" +#include #include #include @@ -55,10 +55,12 @@ static char rcsid[] = "$NetBSD: $"; return x+y; if((long double) x==y) return y; /* x=y, return y */ if((ix|lx)==0) { /* x == 0 */ - double x2; + double u; INSERT_WORDS(x,(u_int32_t)((hy>>32)&0x80000000),1);/* return +-minsub */ - x2 = x*x; - if(x2==x) return x2; else return x; /* raise underflow flag */ + u = math_opt_barrier (x); + u = u * u; + math_force_eval (u); /* raise underflow flag */ + return x; } if(hx>=0) { /* x > 0 */ if (hy<0||(ix>>20)>(iy>>52) @@ -89,16 +91,13 @@ static char rcsid[] = "$NetBSD: $"; if(hy>=0x7ff00000) { x = x+x; /* overflow */ if (FLT_EVAL_METHOD != 0 && FLT_EVAL_METHOD != 1) - /* Force conversion to float. */ - asm ("" : "=m"(x) : "m"(x)); + /* Force conversion to double. */ + asm ("" : "+m"(x)); return x; } - if(hy<0x00100000) { /* underflow */ - double x2 = x*x; - if(x2!=x) { /* raise underflow flag */ - INSERT_WORDS(x2,hx,lx); - return x2; - } + if(hy<0x00100000) { + double u = x*x; /* underflow */ + math_force_eval (u); /* raise underflow flag */ } INSERT_WORDS(x,hx,lx); return x; diff --git a/sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c b/sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c index a9373ff822..cf655fad16 100644 --- a/sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c +++ b/sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c @@ -19,8 +19,9 @@ static char rcsid[] = "$NetBSD: $"; #endif #include "math.h" -#include "math_private.h" +#include #include +#include #ifdef __STDC__ float __nexttowardf(float x, long double y) @@ -46,10 +47,12 @@ static char rcsid[] = "$NetBSD: $"; return x+y; if((long double) x==y) return y; /* x=y, return y */ if(ix==0) { /* x == 0 */ - float x2; + float u; SET_FLOAT_WORD(x,(u_int32_t)((hy>>32)&0x80000000)|1);/* return +-minsub*/ - x2 = x*x; - if(x2==x) return x2; else return x; /* raise underflow flag */ + u = math_opt_barrier (x); + u = u * u; + math_force_eval (u); /* raise underflow flag */ + return x; } if(hx>=0) { /* x > 0 */ if(hy<0||(ix>>23)>(iy>>52)-0x380 @@ -69,13 +72,16 @@ static char rcsid[] = "$NetBSD: $"; } } hy = hx&0x7f800000; - if(hy>=0x7f800000) return x+x; /* overflow */ + if(hy>=0x7f800000) { + x = x+x; /* overflow */ + if (FLT_EVAL_METHOD != 0) + /* Force conversion to float. */ + asm ("" : "+m"(x)); + return x; + } if(hy<0x00800000) { /* underflow */ - float x2 = x*x; - if(x2!=x) { /* raise underflow flag */ - SET_FLOAT_WORD(x2,hx); - return x2; - } + float u = x*x; + math_force_eval (u); /* raise underflow flag */ } SET_FLOAT_WORD(x,hx); return x; diff --git a/sysdeps/ieee754/ldbl-96/s_nexttoward.c b/sysdeps/ieee754/ldbl-96/s_nexttoward.c index 7945cb5cb1..e30438482c 100644 --- a/sysdeps/ieee754/ldbl-96/s_nexttoward.c +++ b/sysdeps/ieee754/ldbl-96/s_nexttoward.c @@ -26,7 +26,7 @@ static char rcsid[] = "$NetBSD: $"; */ #include "math.h" -#include "math_private.h" +#include #include #ifdef __STDC__ @@ -50,10 +50,12 @@ static char rcsid[] = "$NetBSD: $"; return x+y; if((long double) x==y) return y; /* x=y, return y */ if((ix|lx)==0) { /* x == 0 */ - double x2; + double u; INSERT_WORDS(x,(esy&0x8000)<<16,1); /* return +-minsub */ - x2 = x*x; - if(x2==x) return x2; else return x; /* raise underflow flag */ + u = math_opt_barrier (x); + u = u * u; + math_force_eval (u); /* raise underflow flag */ + return x; } if(hx>=0) { /* x > 0 */ if (esy>=0x8000||((ix>>20)&0x7ff)>iy-0x3c00 @@ -84,16 +86,13 @@ static char rcsid[] = "$NetBSD: $"; if(hy>=0x7ff00000) { x = x+x; /* overflow */ if (FLT_EVAL_METHOD != 0 && FLT_EVAL_METHOD != 1) - /* Force conversion to float. */ - asm ("" : "=m"(x) : "m"(x)); + /* Force conversion to double. */ + asm ("" : "+m"(x)); return x; } - if(hy<0x00100000) { /* underflow */ - double x2 = x*x; - if(x2!=x) { /* raise underflow flag */ - INSERT_WORDS(x2,hx,lx); - return x2; - } + if(hy<0x00100000) { + double u = x*x; /* underflow */ + math_force_eval (u); /* raise underflow flag */ } INSERT_WORDS(x,hx,lx); return x; diff --git a/sysdeps/ieee754/ldbl-96/s_nexttowardf.c b/sysdeps/ieee754/ldbl-96/s_nexttowardf.c index a1c38b5d4c..6357975ffc 100644 --- a/sysdeps/ieee754/ldbl-96/s_nexttowardf.c +++ b/sysdeps/ieee754/ldbl-96/s_nexttowardf.c @@ -18,7 +18,8 @@ static char rcsid[] = "$NetBSD: $"; #endif #include "math.h" -#include "math_private.h" +#include +#include #ifdef __STDC__ float __nexttowardf(float x, long double y) @@ -41,10 +42,12 @@ static char rcsid[] = "$NetBSD: $"; return x+y; if((long double) x==y) return y; /* x=y, return y */ if(ix==0) { /* x == 0 */ - float x2; + float u; SET_FLOAT_WORD(x,((esy&0x8000)<<16)|1);/* return +-minsub*/ - x2 = x*x; - if(x2==x) return x2; else return x; /* raise underflow flag */ + u = math_opt_barrier (x); + u = u * u; + math_force_eval (u); /* raise underflow flag */ + return x; } if(hx>=0) { /* x > 0 */ if(esy>=0x8000||((ix>>23)&0xff)>iy-0x3f80 @@ -64,13 +67,16 @@ static char rcsid[] = "$NetBSD: $"; } } hy = hx&0x7f800000; - if(hy>=0x7f800000) return x+x; /* overflow */ - if(hy<0x00800000) { /* underflow */ - float x2 = x*x; - if(x2!=x) { /* raise underflow flag */ - SET_FLOAT_WORD(x2,hx); - return x2; - } + if(hy>=0x7f800000) { + x = x+x; /* overflow */ + if (FLT_EVAL_METHOD != 0) + /* Force conversion to float. */ + asm ("" : "+m"(x)); + return x; + } + if(hy<0x00800000) { + float u = x*x; /* underflow */ + math_force_eval (u); /* raise underflow flag */ } SET_FLOAT_WORD(x,hx); return x; diff --git a/sysdeps/ieee754/ldbl-opt/s_nexttowardfd.c b/sysdeps/ieee754/ldbl-opt/s_nexttowardfd.c index d52526f719..68027f26fa 100644 --- a/sysdeps/ieee754/ldbl-opt/s_nexttowardfd.c +++ b/sysdeps/ieee754/ldbl-opt/s_nexttowardfd.c @@ -20,7 +20,10 @@ * Special cases: */ +#include +#include #include +#include float __nldbl_nexttowardf(float x, double y); @@ -39,10 +42,12 @@ float __nldbl_nexttowardf(float x, double y) return x+y; if((double) x==y) return y; /* x=y, return y */ if(ix==0) { /* x == 0 */ - float x2; + float u; SET_FLOAT_WORD(x,(u_int32_t)(hy&0x80000000)|1);/* return +-minsub*/ - x2 = x*x; - if(x2==x) return x2; else return x; /* raise underflow flag */ + u = math_opt_barrier (x); + u = u * u; + math_force_eval (u); /* raise underflow flag */ + return x; } if(hx>=0) { /* x > 0 */ if(hy<0||(ix>>23)>(iy>>20)-0x380 @@ -60,13 +65,16 @@ float __nldbl_nexttowardf(float x, double y) hx += 1; } hy = hx&0x7f800000; - if(hy>=0x7f800000) return x+x; /* overflow */ - if(hy<0x00800000) { /* underflow */ - float x2 = x*x; - if(x2!=x) { /* raise underflow flag */ - SET_FLOAT_WORD(x2,hx); - return x2; - } + if(hy>=0x7f800000) { + x = x+x; /* overflow */ + if (FLT_EVAL_METHOD != 0) + /* Force conversion to float. */ + asm ("" : "+m"(x)); + return x; + } + if(hy<0x00800000) { + float u = x*x; /* underflow */ + math_force_eval (u); /* raise underflow flag */ } SET_FLOAT_WORD(x,hx); return x; diff --git a/sysdeps/x86_64/fpu/math_private.h b/sysdeps/x86_64/fpu/math_private.h new file mode 100644 index 0000000000..4febcbb5ec --- /dev/null +++ b/sysdeps/x86_64/fpu/math_private.h @@ -0,0 +1,21 @@ +#ifndef _MATH_PRIVATE_H + +#define math_opt_barrier(x) \ +({ __typeof(x) __x; \ + if (sizeof (x) <= sizeof (double)) \ + __asm ("" : "=x" (__x) : "0" (x)); \ + else \ + __asm ("" : "=t" (__x) : "0" (x)); \ + __x; }) +#define math_force_eval(x) \ +do \ + { \ + if (sizeof (x) <= sizeof (double)) \ + __asm __volatile ("" : : "x" (x)); \ + else \ + __asm __volatile ("" : : "f" (x)); \ + } \ +while (0) + +#include +#endif -- 2.11.4.GIT