Re: [Buildroot] [PATCH] package/foot: support soft-float architecture

[Thomas Petazzoni <thomas.petazzoni@bootlin.com>]

Hello Thomas,

On Tue, 04 Feb 2025 12:27:31 +0100
Thomas Bonnefille <thomas.bonnefille@bootlin.com> wrote:

> Architecture using soft float don't support the math FE_* exceptions.
> This commit adds a patch to conditionnally build the FE_* exception
> handling.

Are you sure it's related to soft-float? For example, ARC has floating
point, but does support some of those FE exceptions.

See for example:

# z3 supports arch for which libc fenv.h provides all four macros:
# FE_DOWNWARD, FE_TONEAREST, FE_TOWARDZERO, FE_UPWARD
# See for example in glibc https://sourceware.org/git/glibc.git
# git grep -E '^[[:space:]]*#[[:space:]]*define[[:space:]]+FE_(TONEAREST|UPWARD|DOWNWARD|TOWARDZERO)' sysdeps/
config BR2_PACKAGE_Z3_ARCH_SUPPORTS
        bool
        default y if BR2_aarch64 || BR2_aarch64_be
        default y if BR2_arceb || BR2_arcle
        default y if BR2_arm || BR2_armeb
        default y if BR2_i386
        default y if BR2_m68k
        # BR2_microblaze has only FE_TONEAREST
        default y if BR2_mips || BR2_mipsel || BR2_mips64 || BR2_mips64el
        default y if BR2_or1k
        default y if BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
        default y if BR2_riscv
        default y if BR2_s390x
        # BR2_sh has only FE_{TONEAREST,TOWARDZERO}
        default y if BR2_sparc || BR2_sparc64
        default y if BR2_x86_64
        # BR2_xtensa supports only uclibc which does not have fenv.h

We have this sort of stuff in a few places. Maybe we need to factorize
that?


> +diff --git a/box-drawing.c b/box-drawing.c
> +index 1c613051..e5f67ca7 100644
> +--- a/box-drawing.c
> ++++ b/box-drawing.c
> +@@ -1468,8 +1468,11 @@ draw_box_drawings_light_arc(struct buf *buf, char32_t wc)
> +         double x = circle_hemisphere * sqrt(c_r2 - (y - c_y) * (y - c_y)) + c_x;
> + 
> +         /* See math_error(7) */
> +-        if (errno != 0 ||
> +-            fetestexcept(FE_INVALID | FE_DIVBYZERO | FE_OVERFLOW | FE_UNDERFLOW))
> ++        if (errno != 0
> ++        #if defined (FE_INVALID) && defined (FE_DIVBYZERO) && defined (FE_OVERFLOW) && defined (FE_UNDERFLOW)
> ++            || fetestexcept(FE_INVALID | FE_DIVBYZERO | FE_OVERFLOW | FE_UNDERFLOW)
> ++        #endif

Is it OK to just compile out this code? I mean yes it fixes the build
issues, but is that correct from a runtime perspective?

Thomas
-- 
Thomas Petazzoni, co-owner and CEO, Bootlin
Embedded Linux and Kernel engineering and training
https://bootlin.com
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot