Re: Please revert b2e843161d

On 6/26/26 09:17, Andrew Turner wrote:
> 
> 
>> On 25 Jun 2026, at 19:32, Steve Kargl <[email protected]> wrote:
>>
>> On 6/25/26 10:57, Andrew Turner wrote:
>>> I don’t expect it to work on amd64 in the general case. Godbolt shows both GCC and Clang generating an infinite loop jumping to its self [1]. It may work on some architecture levels, however I don’t know enough about amd64 to know if it’s safe to enable.
>>> The USE_BUILTIN_* flags aren’t intended on being set by users as you found.
>>
>> The comment in the source code certainly does not alert a
>> user to not use -DUSE_BUILTIN_FMAF (or any of the other
>> USE_BUILTIN_*).
> 
>>
>> Architecture specific code should be in msun/arm/.  It would
>> also seem to me that if this is indeed arm specific then
>> it should be guarded with "#if defined(__ARM) && …”
> 
> Other architectures could use this if they have a simple instruction sequence that the compiler uses. I know about arm64 so am comfortable enabling it there, but not on other architectures I don’t use.
> 
> Andrew

My point is that a compiler will use its builtin if one is available
in lieu of a function in libm.  One does not need to pollute the msun
source code with "#ifdef __builtin_XXX() #else ... #endif".  It simply
clutters comparisons to the libm code in NetBSD, OpenBSD, and Openlibm.
I do recognize that one may need to specify a compiler option such
as -mfma for gcc on amd64 to get a builtin.  That is due to the lack
of an fma instruction in the original amd64 instruction set.

Can you at least add a cautionary comment in the source code that
USE_BUILTIN_FMA[F] have only been tested on arm64 and may lead to
segfaults on other architectures?

-- 
steve