[PATCH 5/6] include: Add __rdtsc and __rdtscp intrinsics.

[Rémi Bernon]

On 3/10/21 4:00 PM, Rémi Bernon wrote:
> On 3/10/21 3:57 PM, Jacek Caban wrote:
>> Hi Rémi,
>>
>> On 10.03.2021 12:22, Rémi Bernon wrote:
>>> Signed-off-by: Rémi Bernon <rbernon at codeweavers.com>
>>> ---
>>>   include/msvcrt/intrin.h | 30 ++++++++++++++++++++++++++++++
>>>   1 file changed, 30 insertions(+)
>>>
>>> diff --git a/include/msvcrt/intrin.h b/include/msvcrt/intrin.h
>>> index 781c6fac823..061866fe63b 100644
>>> --- a/include/msvcrt/intrin.h
>>> +++ b/include/msvcrt/intrin.h
>>> @@ -20,6 +20,36 @@ static inline void __cpuid(int info[4], int ax)
>>>   {
>>>       return __cpuidex(info, ax, 0);
>>>   }
>>> +
>>> +#ifdef __i386
>>> +static inline unsigned __int64 __rdtsc(void)
>>> +{
>>> +    unsigned __int64 a;
>>> +    __asm__ ("rdtsc" : "=A" (a));
>>> +    return a;
>>> +}
>>> +
>>> +static inline unsigned __int64 __rdtscp(unsigned int *aux)
>>> +{
>>> +    unsigned __int64 a;
>>> +    __asm__ ("rdtscp" : "=A" (a), "=c" (*aux));
>>> +    return a;
>>> +}
>>> +#elif defined __amd64
>>> +static inline unsigned __int64 __rdtsc(void)
>>> +{
>>> +    unsigned __int64 a, d;
>>> +    __asm__ ("rdtsc" : "=a" (a), "=d" (d));
>>> +    return (d << 32) | a;
>>> +}
>>> +
>>> +static inline unsigned __int64 __rdtscp(unsigned int *aux)
>>> +{
>>> +    unsigned __int64 a, d;
>>> +    __asm__ ("rdtscp" : "=a" (a), "=d" (d), "=c" (*aux));
>>> +    return (d << 32) | a;
>>> +}
>>> +#endif
>>>   #endif
>>
>>
>> I think you meant __i386__ and __x86_64__ for guards. If guards were 
>> right, you'd get a redefinition because __rdtsc is a compiler builtin. 
>> It seems to me that all we need for __rdtsc is to provide MSVC-style 
>> declaration in winnt.h, like it's done in WinSDK.
>>
>>
>> Thanks,
>>
>> Jacek
>>
> 
> Yeah I'm not sure where I got these guards from, but I think the test 
> failed to compile, so I assumed MinGW missed these intrinsics? I'll 
> double check.

So I can confirm that although the guard aren't the usual ones there's 
no __rdtsc(p) intrinsics on MinGW and the test fails to compile without 
these (and I don't get redefinition errors with them).

Then I agree that these are probably MSVC intrinsics and may need to be 
guarded for that case, but they are still needed for GCC (and possibly 
clang?).
-- 
Rémi Bernon <rbernon at codeweavers.com>