[Buildroot] [PATCH] package/glibc: fix signbit build issue with gcc < 6.x

Romain Naour romain.naour at gmail.com
Mon Oct 16 19:34:14 UTC 2017


Hi All,

Le 15/10/2017 à 23:57, Romain Naour a écrit :
> As reported in [1], mesa3d package fail to build with a x86_64 toolchain
> using gcc 5.5.0 and glibc 2.26.
> 
> Add a new patch providing signbit() functions for C++ that does not use
> __MATH_TG macro.
> 
> [1] https://sourceware.org/bugzilla/show_bug.cgi?id=22296

I withdraw this patch since it's not the correct fix [1]

[1] https://sourceware.org/ml/libc-alpha/2017-10/msg00703.html

Best regards,
Romain

> 
> Signed-off-by: Romain Naour <romain.naour at gmail.com>
> ---
>  ...-version-of-signbit-that-does-not-use-__M.patch | 65 ++++++++++++++++++++++
>  1 file changed, 65 insertions(+)
>  create mode 100644 package/glibc/0007-Provide-a-C-version-of-signbit-that-does-not-use-__M.patch
> 
> diff --git a/package/glibc/0007-Provide-a-C-version-of-signbit-that-does-not-use-__M.patch b/package/glibc/0007-Provide-a-C-version-of-signbit-that-does-not-use-__M.patch
> new file mode 100644
> index 0000000..ea7a4e7
> --- /dev/null
> +++ b/package/glibc/0007-Provide-a-C-version-of-signbit-that-does-not-use-__M.patch
> @@ -0,0 +1,65 @@
> +From 2a735f52ec321f06b2af4b4265160d54260e5c44 Mon Sep 17 00:00:00 2001
> +From: Romain Naour <romain.naour at gmail.com>
> +Date: Sun, 15 Oct 2017 20:04:24 +0200
> +Subject: [PATCH] Provide a C++ version of signbit that does not use __MATH_TG
> + (bug 22296)
> +
> +The macro __MATH_TG contains the logic to select between long double and
> +_Float128, when these types are ABI-distinct.  This logic relies on
> +__builtin_types_compatible_p, which is not available in C++ mode.
> +
> +On the other hand, C++ function overloading provides the means to
> +distinguish between the floating-point types.  The overloading
> +resolution will match the correct parameter regardless of type
> +qualifiers, i.e.: const and volatile.
> +
> +This is the same fix as for issignaling [1] but the issue appear
> +only with gcc < 6.x. Since GCC 6.0, __builtin_signbit is type-generic,
> +so signbit() is defined without __MATH_TG.
> +
> +Tested for x86_64.
> +
> +	* math/math.h [defined __cplusplus] (signbit): Provide a C++
> +	definition for signbit that does not rely on __MATH_TG,
> +	since __MATH_TG uses __builtin_types_compatible_p, which is only
> +	available in C mode.
> +
> +[1] a16e8bc08edca84d507715c66d6cddbbc7ed3b62
> +
> +Signed-off-by: Romain Naour <romain.naour at gmail.com>
> +---
> +patch status: submited upstream
> +https://sourceware.org/ml/libc-alpha/2017-10/msg00654.html
> +---
> + math/math.h | 15 +++++++++++++++
> + 1 file changed, 15 insertions(+)
> +
> +diff --git a/math/math.h b/math/math.h
> +index 2b216c6..0013ad9 100644
> +--- a/math/math.h
> ++++ b/math/math.h
> +@@ -418,6 +418,21 @@ enum
> + /* Return nonzero value if sign of X is negative.  */
> + # if __GNUC_PREREQ (6,0)
> + #  define signbit(x) __builtin_signbit (x)
> ++# elif defined __cplusplus
> ++   /* In C++ mode, __MATH_TG cannot be used, because it relies on
> ++      __builtin_types_compatible_p, which is a C-only builtin.  On the
> ++      other hand, overloading provides the means to distinguish between
> ++      the floating-point types.  The overloading resolution will match
> ++      the correct parameter (regardless of type qualifiers (i.e.: const
> ++      and volatile).  */
> ++extern "C++" {
> ++inline int signbit (float __val) { return __signbitf (__val); }
> ++inline int signbit (double __val) { return __signbit (__val); }
> ++inline int signbit (long double __val) { return __signbitl (__val); }
> ++#  if __HAVE_DISTINCT_FLOAT128
> ++inline int signbit (_Float128 __val) { return __signbitf128 (__val); }
> ++#  endif
> ++} /* extern C++ */
> + # elif __GNUC_PREREQ (4,0)
> + #  define signbit(x) __MATH_TG ((x), __builtin_signbit, (x))
> + # else
> +-- 
> +2.9.5
> +
> 




More information about the buildroot mailing list