[4/5] x86 long double: Consider pseudo numbers as signaling
Commit Message
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
Comments
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
@@ -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);
new file mode 100644
@@ -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)