[PATCH 4/8] math: Fix UB on sinhf

[Carlos O'Donell]

On 4/25/25 4:54 PM, Adhemerval Zanella wrote:
> The left shift overflows for 'int', use uint64_t instead.  It syncs
> with CORE-MATH commit bbfabd99.
> 
> Checked on aarch64-linux-gnu, x86_64-linux-gnu, and i686-linux-gnu.

LGTM.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>

> ---
>   sysdeps/ieee754/flt-32/e_sinhf.c | 6 +++---
>   1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/sysdeps/ieee754/flt-32/e_sinhf.c b/sysdeps/ieee754/flt-32/e_sinhf.c
> index 754b84abe0..6c8c1db404 100644
> --- a/sysdeps/ieee754/flt-32/e_sinhf.c
> +++ b/sysdeps/ieee754/flt-32/e_sinhf.c
> @@ -3,7 +3,7 @@
>   Copyright (c) 2022-2024 Alexei Sibidanov.
>   
>   The original version of this file was copied from the CORE-MATH
> -project (file src/binary32/sinh/sinhf.c, revision 572ecec).
> +project (file src/binary32/sinh/sinhf.c, revision bbfabd99).

OK.

>   
>   Permission is hereby granted, free of charge, to any person obtaining a copy
>   of this software and associated documentation files (the "Software"), to deal
> @@ -83,8 +83,8 @@ __ieee754_sinhf (float x)
>     double h2 = h * h;
>     int64_t jp = asuint64 (ia + 0x1.8p52);
>     int64_t jm = -jp;
> -  double sp = asdouble (TB[jp & 31] + ((jp >> 5) << 52));
> -  double sm = asdouble (TB[jm & 31] + ((jm >> 5) << 52));
> +  double sp = asdouble (TB[jp & 31] + ((uint64_t)(jp >> 5) << 52));
> +  double sm = asdouble (TB[jm & 31] + ((uint64_t)(jm >> 5) << 52));

OK. Both cast to uint64_t.

>     double te = C[0] + h2 * C[2];
>     double to = (C[1] + h2 * C[3]);
>     double rp = sp * (te + h * to);

OK.

commit bbfabd993a71b049c210b0febfd06d18369fadc1
Author: Paul Zimmermann <Paul.Zimmermann@inria.fr>
Date:   Mon Mar 17 12:29:20 2025 +0100

     various fixes to avoid undefined behavior for binary32 functions
     
     (contributed by Ganesh Ajjanagadde <gajjanag@alum.mit.edu>
      see https://sympa.inria.fr/sympa/arc/core-math/2025-03/msg00007.html)


-- 
Cheers,
Carlos.