Message ID | 20201215141339.2684384-5-siddhesh@sourceware.org |
---|---|
State | New |
Headers | show |
Series | x86 pseudo-normal numbers | expand |
On 15/12/2020 11:13, Siddhesh Poyarekar via Libc-alpha wrote: > Implement a separate x86 version of issignalingl that returns true for > pseudo-normal numbers. > > Also drop comment from the generic version of s_issignalingl since it > is no longer true or relevant. > --- > sysdeps/ieee754/ldbl-96/s_issignalingl.c | 2 -- > sysdeps/x86/fpu/s_issignalingl.c | 39 ++++++++++++++++++++++++ > 2 files changed, 39 insertions(+), 2 deletions(-) > create mode 100644 sysdeps/x86/fpu/s_issignalingl.c > > diff --git a/sysdeps/ieee754/ldbl-96/s_issignalingl.c b/sysdeps/ieee754/ldbl-96/s_issignalingl.c > index ec542ad468..2aa0ffae0e 100644 > --- a/sysdeps/ieee754/ldbl-96/s_issignalingl.c > +++ b/sysdeps/ieee754/ldbl-96/s_issignalingl.c > @@ -34,8 +34,6 @@ __issignalingl (long double x) > hxi ^= 0x40000000; > /* If lxi != 0, then set any suitable bit of the significand in hxi. */ > hxi |= (lxi | -lxi) >> 31; > - /* We do not recognize a pseudo NaN as sNaN; they're invalid on 80387 and > - later. */ > /* We have to compare for greater (instead of greater or equal), because x's > significand being all-zero designates infinity not NaN. */ > return ((exi & 0x7fff) == 0x7fff) && (hxi > 0xc0000000); > diff --git a/sysdeps/x86/fpu/s_issignalingl.c b/sysdeps/x86/fpu/s_issignalingl.c We have multiple internal defines that try to use the common implementation instead of pushing for arch-specific ones (for instance nan-high-order-bit.h, fix-int-fp-convert-zero.h, fix-fp-int-convert-overflow.h, etc). Maybe it would be simpler to just add a new header (nan-handle-pseudo-number.h or something related) and adapt the generic implementation. > new file mode 100644 > index 0000000000..df273b881d > --- /dev/null > +++ b/sysdeps/x86/fpu/s_issignalingl.c > @@ -0,0 +1,39 @@ > +/* Test for signaling NaN - x86 version. > + Copyright (C) 2020 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + <https://www.gnu.org/licenses/>. */ > + > +#include <math.h> > +#include <math_private.h> > +#include <nan-high-order-bit.h> > +#include "sysdeps/x86/fpu/isnanl_common.h" > + > +int > +__issignalingl (long double x) > +{ > + uint32_t exi, hxi, lxi, nan, pn; > + GET_LDOUBLE_WORDS (exi, hxi, lxi, x); > +#if HIGH_ORDER_BIT_IS_SET_FOR_SNAN > +# error not implemented > +#else > + nan = x86_isnanl (exi, hxi, lxi); > + pn = (~hxi & 0x80000000) >> 31; > + hxi = (~hxi & 0x40000000) >> 30; > + > + return nan & (pn | hxi); > +#endif > +} > +libm_hidden_def (__issignalingl) >
On 12/23/20 1:43 AM, Adhemerval Zanella via Libc-alpha wrote: > > > On 15/12/2020 11:13, Siddhesh Poyarekar via Libc-alpha wrote: >> Implement a separate x86 version of issignalingl that returns true for >> pseudo-normal numbers. >> >> Also drop comment from the generic version of s_issignalingl since it >> is no longer true or relevant. >> --- >> sysdeps/ieee754/ldbl-96/s_issignalingl.c | 2 -- >> sysdeps/x86/fpu/s_issignalingl.c | 39 ++++++++++++++++++++++++ >> 2 files changed, 39 insertions(+), 2 deletions(-) >> create mode 100644 sysdeps/x86/fpu/s_issignalingl.c >> >> diff --git a/sysdeps/ieee754/ldbl-96/s_issignalingl.c b/sysdeps/ieee754/ldbl-96/s_issignalingl.c >> index ec542ad468..2aa0ffae0e 100644 >> --- a/sysdeps/ieee754/ldbl-96/s_issignalingl.c >> +++ b/sysdeps/ieee754/ldbl-96/s_issignalingl.c >> @@ -34,8 +34,6 @@ __issignalingl (long double x) >> hxi ^= 0x40000000; >> /* If lxi != 0, then set any suitable bit of the significand in hxi. */ >> hxi |= (lxi | -lxi) >> 31; >> - /* We do not recognize a pseudo NaN as sNaN; they're invalid on 80387 and >> - later. */ >> /* We have to compare for greater (instead of greater or equal), because x's >> significand being all-zero designates infinity not NaN. */ >> return ((exi & 0x7fff) == 0x7fff) && (hxi > 0xc0000000); >> diff --git a/sysdeps/x86/fpu/s_issignalingl.c b/sysdeps/x86/fpu/s_issignalingl.c > > We have multiple internal defines that try to use the common implementation > instead of pushing for arch-specific ones (for instance nan-high-order-bit.h, > fix-int-fp-convert-zero.h, fix-fp-int-convert-overflow.h, etc). Maybe it > would be simpler to just add a new header (nan-handle-pseudo-number.h or > something related) and adapt the generic implementation. OK I'll try to do that. Thanks, Siddhesh
diff --git a/sysdeps/ieee754/ldbl-96/s_issignalingl.c b/sysdeps/ieee754/ldbl-96/s_issignalingl.c index ec542ad468..2aa0ffae0e 100644 --- a/sysdeps/ieee754/ldbl-96/s_issignalingl.c +++ b/sysdeps/ieee754/ldbl-96/s_issignalingl.c @@ -34,8 +34,6 @@ __issignalingl (long double x) hxi ^= 0x40000000; /* If lxi != 0, then set any suitable bit of the significand in hxi. */ hxi |= (lxi | -lxi) >> 31; - /* We do not recognize a pseudo NaN as sNaN; they're invalid on 80387 and - later. */ /* We have to compare for greater (instead of greater or equal), because x's significand being all-zero designates infinity not NaN. */ return ((exi & 0x7fff) == 0x7fff) && (hxi > 0xc0000000); diff --git a/sysdeps/x86/fpu/s_issignalingl.c b/sysdeps/x86/fpu/s_issignalingl.c new file mode 100644 index 0000000000..df273b881d --- /dev/null +++ b/sysdeps/x86/fpu/s_issignalingl.c @@ -0,0 +1,39 @@ +/* Test for signaling NaN - x86 version. + Copyright (C) 2020 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <https://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_private.h> +#include <nan-high-order-bit.h> +#include "sysdeps/x86/fpu/isnanl_common.h" + +int +__issignalingl (long double x) +{ + uint32_t exi, hxi, lxi, nan, pn; + GET_LDOUBLE_WORDS (exi, hxi, lxi, x); +#if HIGH_ORDER_BIT_IS_SET_FOR_SNAN +# error not implemented +#else + nan = x86_isnanl (exi, hxi, lxi); + pn = (~hxi & 0x80000000) >> 31; + hxi = (~hxi & 0x40000000) >> 30; + + return nan & (pn | hxi); +#endif +} +libm_hidden_def (__issignalingl)