Re: [PATCH] math: x86: Use prefix for FP_INIT_ROUNDMODE

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

On 01/09/22 13:49, H.J. Lu wrote:
> On Thu, Sep 1, 2022 at 9:42 AM H.J. Lu <hjl.tools@gmail.com> wrote:
>>
>> On Thu, Sep 1, 2022 at 6:03 AM Adhemerval Zanella via Libc-alpha
>> <libc-alpha@sourceware.org> wrote:
>>>
>>> Not all compilers support the inline asm prefix '%v' to emit the avx
>>> instruction if AVX is enable.  Use a prefix instead.
>>>
>>> Checked on x86_64-linux-gnu and i686-linux-gnu.
>>> ---
>>>  sysdeps/x86/fpu/sfp-machine.h | 8 +++++++-
>>>  1 file changed, 7 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/sysdeps/x86/fpu/sfp-machine.h b/sysdeps/x86/fpu/sfp-machine.h
>>> index 5892f4f5fe..006eebe257 100644
>>> --- a/sysdeps/x86/fpu/sfp-machine.h
>>> +++ b/sysdeps/x86/fpu/sfp-machine.h
>>> @@ -39,9 +39,15 @@ typedef unsigned int UTItype __attribute__ ((mode (TI)));
>>>
>>>  # define FP_RND_MASK           0x6000
>>>
>>> +# ifdef __AVX__
>>> +#  define AVX_INSN_PREFIX      "v"
>>> +# else
>>> +#  define AVX_INSN_PREFIX      ""
>>> +# endif
>>> +
>>>  # define FP_INIT_ROUNDMODE                                     \
>>>    do {                                                         \
>>> -    __asm__ __volatile__ ("%vstmxcsr\t%0" : "=m" (_fcw));      \
>>> +    __asm__ __volatile__ (AVX_INSN_PREFIX "stmxcsr\t%0" : "=m" (_fcw)); \
>>>    } while (0)
>>>  #else
>>>  # define _FP_W_TYPE_SIZE       32
>>> --
>>> 2.34.1
>>>
>>
>> Should we use __builtin_ia32_stmxcsr instead?
>>
> 
> __builtin_ia32_stmxcsr requires extra instructions.
> 

I don't have a strong option, although in theory builtins would play nice in
general code generation.