[PATCH] Cygwin: Adapt math functions to use 64bit long double on aarch64

[Corinna Vinschen]

Hi Thirumalai,

Ping?


On Mar 11 21:08, Corinna Vinschen wrote:
> the long double math functions are taken from mingw-w64 and only
> slightly changed to fit into Cygwin.
> 
> Ideally, all changes to make aarch64 work, too, are also taken
> right from mingw-w64.
> 
> Are the below changes the only ones required to make long double
> arithmetic work on aarch64?  Are they in line with the upstream
> changes in mingw-w64?
> 
> 
> Thanks,
> Corinna
> 
> 
> On Feb 24 08:38, Thirumalai Nagalingam wrote:
> > Hi,
> > 
> > On Cygwin AArch64, long double has the same representation and precision
> > as double (64-bit), unlike x86 extended precision.
> > 
> > This patch updates math functions to correctly handle this case by avoiding
> > assumptions about extended precision in nextafterl and related functions.
> > It also updates rintl to use the generic implementation on AArch64 and
> > adjusts constants in cephes_mconf.h and lgammal.c accordingly.
> > 
> > Thanks & regards
> > Thiru
> > 
> > In-lined patch:
> > ---
> >  winsup/cygwin/math/cephes_mconf.h | 4 ++--
> >  winsup/cygwin/math/lgammal.c      | 4 ++--
> >  winsup/cygwin/math/nextafterl.c   | 4 ++++
> >  winsup/cygwin/math/rintl.c        | 2 +-
> >  4 files changed, 9 insertions(+), 5 deletions(-)
> > 
> > diff --git a/winsup/cygwin/math/cephes_mconf.h b/winsup/cygwin/math/cephes_mconf.h
> > index 832fae0df..654de88bf 100644
> > --- a/winsup/cygwin/math/cephes_mconf.h
> > +++ b/winsup/cygwin/math/cephes_mconf.h
> > @@ -66,7 +66,7 @@ extern double __QNAN;
> >  #endif
> > 
> >  /*long double*/
> > -#if defined(__arm__) || defined(_ARM_)
> > +#if defined(__arm__) || defined(_ARM_) || defined(__aarch64__)
> >  #define MAXNUML        1.7976931348623158E308
> >  #define MAXLOGL        7.09782712893383996843E2
> >  #define MINLOGL        -7.08396418532264106224E2
> > @@ -84,7 +84,7 @@ extern double __QNAN;
> >  #define PIL    3.1415926535897932384626L
> >  #define PIO2L  1.5707963267948966192313L
> >  #define PIO4L  7.8539816339744830961566E-1L
> > -#endif /* defined(__arm__) || defined(_ARM_) */
> > +#endif /* defined(__arm__) || defined(_ARM_)  || defined(__aarch64__) */
> > 
> >  #define isfinitel isfinite
> >  #define isinfl isinf
> > diff --git a/winsup/cygwin/math/lgammal.c b/winsup/cygwin/math/lgammal.c
> > index 022a16acf..961eec280 100644
> > --- a/winsup/cygwin/math/lgammal.c
> > +++ b/winsup/cygwin/math/lgammal.c
> > @@ -198,11 +198,11 @@ static uLD C[] = {
> > 
> >  /* log( sqrt( 2*pi ) ) */
> >  static const long double LS2PI  =  0.91893853320467274178L;
> > -#if defined(__arm__) || defined(_ARM_)
> > +#if defined(__arm__) || defined(_ARM_) || defined(__aarch64__)
> >  #define MAXLGM 2.035093e36
> >  #else
> >  #define MAXLGM 1.04848146839019521116e+4928L
> > -#endif /* defined(__arm__) || defined(_ARM_) */
> > +#endif /* defined(__arm__) || defined(_ARM_) || defined(__aarch64__) */
> > 
> >  /* Logarithm of gamma function */
> >  /* Reentrant version */
> > diff --git a/winsup/cygwin/math/nextafterl.c b/winsup/cygwin/math/nextafterl.c
> > index b1e479a95..80c9c3c4d 100644
> > --- a/winsup/cygwin/math/nextafterl.c
> > +++ b/winsup/cygwin/math/nextafterl.c
> > @@ -16,6 +16,9 @@
> >  long double
> >  nextafterl (long double x, long double y)
> >  {
> > +#if defined(__aarch64__) && (LDBL_MANT_DIG == DBL_MANT_DIG)
> > +  return (long double) nexttoward (x, y);
> > +# else
> >    union {
> >        long double ld;
> >        struct {
> > @@ -63,6 +66,7 @@ nextafterl (long double x, long double y)
> >      u.parts.mantissa |=  normal_bit;
> > 
> >    return u.ld;
> > +# endif /* defined(__aarch64__) */
> >  }
> > 
> >  /* nexttowardl is the same function with a different name.  */
> > diff --git a/winsup/cygwin/math/rintl.c b/winsup/cygwin/math/rintl.c
> > index 9ec159d17..1e30de069 100644
> > --- a/winsup/cygwin/math/rintl.c
> > +++ b/winsup/cygwin/math/rintl.c
> > @@ -9,7 +9,7 @@ long double rintl (long double x) {
> >    long double retval = 0.0L;
> >  #if defined(_AMD64_) || defined(__x86_64__) || defined(_X86_) || defined(__i386__)
> >    __asm__ __volatile__ ("frndint;": "=t" (retval) : "0" (x));
> > -#elif defined(__arm__) || defined(_ARM_)
> > +#elif defined(__arm__) || defined(_ARM_)|| defined(__aarch64__)
> >      retval = rint(x);
> >  #endif
> >    return retval;
> > --
> > 
>