Re: [PATCH 11/15] math: Fix acos template for arguments greater than 1

[Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>]

On 27/03/24 16:04, Joseph Myers wrote:
> On Wed, 27 Mar 2024, Adhemerval Zanella Netto wrote:
> 
>> And I think it is unrelated to gcc PR95115 because x86_64/i686 will use
>> and specific sysdeps/i386/fpu/e_acosl.c that explicit does not handle this
>> input case for overflow exceptions.  For shared build this case is
>> handle by w_acosl_compat.c:
>>
>>   if (__builtin_expect (isgreater (fabsl (x), 1.0L), 0)
>>       && _LIB_VERSION != _IEEE_)
>>     {
>>       /* acos(|x|>1) */
>>       feraiseexcept (FE_INVALID);
>>       return __kernel_standard_l (x, x, 201);
>>     }
> 
> The compat code is dealing with the possibility of SVID exceptions, which 
> isn't relevant here.

I meant that __ieee754_acosl in being called in both code paths, but 
the compat code returns early without calling __ieee754_acosl for the
possible problematic code path.

> 
>> And that's why I though following the same logic on template would be
>> better.  But I think maybe we should fix on x86_64 implementation instead.
> 
> Yes, we should fix the x86_64 implementation.
> 
> Such issues in dbl-64, flt-32 or ldbl-128 sources would largely have been 
> fixed (modulo compiler bugs) when we started adding new architectures 
> after new architectures stopped using the compat wrappers - but any issues 
> for ldbl-128ibm, ldbl-96 or architecture-specific sources wouldn't have 
> been detected then.
> 

I will drop this change and rather focus on the original issues on missing
symbols. I will open some bug against the x86 implementation so we keep
track of these issues with static linkage.