Re: [PATCH v2 1/6] [powerpc] fenv_private.h clean up

[Paul E Murphy <murphyp@linux.ibm.com>]

On 9/22/19 12:11 AM, Paul Clarke wrote:
> On 9/20/19 10:36 AM, Paul E Murphy wrote:
>> On 9/19/19 1:46 PM, Paul A. Clarke wrote:
>>> -static __always_inline void
>>> -libc_feholdexcept_ppc (fenv_t *envp)
>>> -{
>>> -  __libc_feholdbits_ppc (envp, _FPU_MASK_NOT_RN_NI, 0LL);
>>> -}
>>> -
>>> -static __always_inline void
>>> -libc_feholdexcept_setround_ppc (fenv_t *envp, int r)
>>> -{
>>> -  __libc_feholdbits_ppc (envp, _FPU_MASK_NOT_RN_NI & _FPU_MASK_RN, r);
>>> -}
>>> -
>>> -static __always_inline void
>>> -libc_fesetround_ppc (int r)
>>> -{
>>> -  __fesetround_inline (r);
>>> -}
>>> -
>>> -static __always_inline int
>>> -libc_fetestexcept_ppc (int e)
>>> -{
>>> -  fenv_union_t u;
>>> -  u.fenv = fegetenv_register ();
>>> -  return u.l & e;
>>> -}
>>> -
>>> -static __always_inline void
>>> -libc_feholdsetround_ppc (fenv_t *e, int r)
>>> -{
>>> -  __libc_feholdbits_ppc (e, _FPU_MASK_TRAPS_RN, r);
>>> -}
>>
>> Are these unused? I still see the macro redirections calling out to them after applying this patch.
> 
> Fair question.  I'll admit I'm not an expert, especially in the corner cases, but here's what I see:
> 
>>> -libc_feholdexcept_ppc
> 
> This shows up in the ieee754 version of s_nearbyint{f}.c, but that is not used in favor of sysdeps/powerpc/fpu/s_nearbyint{f}.c which does not use it.
> 
>>> -libc_feholdexcept_setround_ppc
> 
> This is still used, but my patch doesn't remove it, in spite of how it looks above.  As noted in the Changelog, I moved the code from __libc_feholdbits_ppc into libc_feholdexcept_setround_ppc, which makes the patch look like it was removed.
> 
>>> -libc_fesetround_ppc
>>> -libc_fetestexcept_ppc
> 
> These show up in the ieee754 version of s_fma.c, but there is a powerpc version which uses different code (__builtin_fma) instead.
> 
>>> -libc_feholdsetround_ppc
> 
> This _appears_ to be referenced in some uses of SET_RESTORE_ROUND_GENERIC, but actually gets mapped to libc_feholdsetround_ppc_ctx.
> 
> PC
> 

OK. I agree with your conclusions.

Reviewed-By: Paul E Murphy <murphyp@linux.ibm.com>