diff options
Diffstat (limited to 'libm')
-rw-r--r-- | libm/float_wrappers.c | 63 |
1 files changed, 20 insertions, 43 deletions
diff --git a/libm/float_wrappers.c b/libm/float_wrappers.c index 9e8256780..4ea924201 100644 --- a/libm/float_wrappers.c +++ b/libm/float_wrappers.c @@ -26,24 +26,20 @@ /* For the time being, do _NOT_ implement these functions * that are defined by SuSv3 */ -#if 0 -float exp2f(float); -float fmaf(float, float, float); -float fmaxf(float, float); -float fminf(float, float); -float fdimf(float, float); -long long llrintf(float); -long long llroundf(float); -long lroundf(float); -float log2f(float); -long lrintf(float); -float nexttowardf(float, long double); -float remquof(float, float, int *); -float roundf(float); -float scalblnf(float, long); -float truncf(float); -float tgammaf(float); -#endif +#undef L_exp2f /*float exp2f(float);*/ +#undef L_fdimf /*float fdimf(float, float);*/ +#undef L_fmaf /*float fmaf(float, float, float);*/ +#undef L_fmaxf /*float fmaxf(float, float);*/ +#undef L_fminf /*float fminf(float, float);*/ +#undef L_llrintf /*long long llrintf(float);*/ +#undef L_log2f /*float log2f(float);*/ +#undef L_lrintf /*long lrintf(float);*/ +#undef L_nearbyintf /*float nearbyintf(float);*/ +#undef L_nexttowardf /*float nexttowardf(float, long double);*/ +#undef L_remquof /*float remquof(float, float, int *);*/ +#undef L_scalblnf /*float scalblnf(float, long);*/ +#undef L_tgammaf /*float tgammaf(float);*/ +#undef L_truncf /*float truncf(float);*/ /* Implement the following, as defined by SuSv3 */ #if 0 @@ -71,16 +67,18 @@ float hypotf(float, float); int ilogbf(float); float ldexpf(float, int); float lgammaf(float); +long long llroundf(float); float log10f(float); float log1pf(float); float logbf(float); float logf(float); +long lroundf(float); float modff(float, float *); -float nearbyintf(float); float nextafterf(float, float); float powf(float, float); float remainderf(float, float); float rintf(float); +float roundf(float); float scalbnf(float, int); float sinf(float); float sinhf(float); @@ -89,6 +87,7 @@ float tanf(float); float tanhf(float); #endif + #ifdef L_acosf float acosf (float x) { @@ -201,14 +200,12 @@ float erff (float x) #endif -#if 0 #ifdef L_exp2f float exp2f (float x) { return (float) exp2( (double)x ); } #endif -#endif #ifdef L_expf @@ -235,14 +232,12 @@ float fabsf (float x) #endif -#if 0 #ifdef L_fdimf float fdimf (float x, float y) { return (float) fdim( (double)x, (double)y ); } #endif -#endif #ifdef L_floorf @@ -253,7 +248,6 @@ float floorf (float x) #endif -#if 0 #ifdef L_fmaf float fmaf (float x, float y, float z) { @@ -276,7 +270,6 @@ float fminf (float x, float y) return (float) fmin( (double)x, (double)y ); } #endif -#endif #ifdef L_fmodf @@ -327,7 +320,6 @@ float lgammaf (float x) #endif -#if 0 #ifdef L_llrintf long long llrintf (float x) { @@ -342,7 +334,7 @@ long long llroundf (float x) return (float) llround( (double)x ); } #endif -#endif + #ifdef L_log10f float log10f (float x) @@ -360,14 +352,12 @@ float log1pf (float x) #endif -#if 0 #ifdef L_log2f float log2f (float x) { return (float) log2( (double)x ); } #endif -#endif #ifdef L_logbf @@ -386,7 +376,6 @@ float logf (float x) #endif -#if 0 #ifdef L_lrintf long lrintf (float x) { @@ -401,7 +390,6 @@ long lroundf (float x) return (float) lround( (double)x ); } #endif -#endif #ifdef L_modff @@ -416,14 +404,12 @@ float modff (float x, float *iptr) #endif -#if 0 #ifdef L_nearbyintf float nearbyintf (float x) { return (float) nearbyint( (double)x ); } #endif -#endif #ifdef L_nextafterf @@ -434,14 +420,13 @@ float nextafterf (float x, float y) #endif -#if 0 #ifdef L_nexttowardf float nexttowardf (float x, long double y) { return (float) nexttoward( (double)x, (double)y ); } #endif -#endif + #ifdef L_powf float powf (float x, float y) @@ -459,14 +444,12 @@ float remainderf (float x, float y) #endif -#if 0 #ifdef L_remquof float remquof (float x, float y, int *quo) { return (float) remquo( (double)x, (double)y, quo ); } #endif -#endif #ifdef L_rintf @@ -477,7 +460,6 @@ float rintf (float x) #endif -#if 0 #ifdef L_roundf float roundf (float x) { @@ -492,7 +474,6 @@ float scalblnf (float x, long exp) return (float) scalbln( (double)x, exp ); } #endif -#endif #ifdef L_scalbnf @@ -543,7 +524,6 @@ float tanhf (float x) #endif -#if 0 #ifdef L_tgammaf float tgammaf (float x) { @@ -558,6 +538,3 @@ float truncf (float x) return (float) trunc( (double)x ); } #endif -#endif - - |