Re: Monotonic clock

On my M1 iMac, mach_timebase_info returns the ratio 125/3. Multiply ticks by this ratio to get nanosec. Ticks are returned from mach_continuous_time().

And common sense would dictate that, indeed, ticks are arriving at a rate of 1 per ≈42 ns, or about 24 MHz rate. To be the inverse would demand that the realtime clock receives ticks at a rate of 42 GHz, which seems wholly unreasonable, given that the CPU clocks at around 4 GHz or lower.


> On Oct 9, 2025, at 08:14, David McClain (as dbm at refined-audiometrics dot com) <[email protected]> wrote:
> 
> Better yet, here is the direct code in Mac header file IOLib.h:
> 
> /*!
>  * @group Raw Mach Time API
>  * In general prefer to use the <time.h> API clock_gettime_nsec_np(3), which
>  * deals in the same clocks (and more) in ns units.
>  * Conversion of ns to (resp. from) tick units as returned by the mach time APIs
>  * is performed by division (resp. multiplication) with the fraction returned by
>  * mach_timebase_info().
>  */
> 
> struct mach_timebase_info {
> 	uint32_t numer;
> 	uint32_t denom;
> };
> 
> typedef struct mach_timebase_info *mach_timebase_info_t;
> typedef struct mach_timebase_info  mach_timebase_info_data_t;
> 
> /*!
>  * @function mach_timebase_info
>  * Returns fraction to multiply a value in mach tick units with to convert it
>  * to nanoseconds.
>  *
>  * @param info	Timebase fraction out parameter.
>  * @result		KERN_SUCCESS if `info` was filled in.
>  */
> kern_return_t
> mach_timebase_info(mach_timebase_info_t info);
> 
> 
> 
>> On Oct 9, 2025, at 08:08, David McClain (as dbm at refined-audiometrics dot com) <[email protected]> wrote:
>> 
>> I could be mistaken, but from my reading of the Apple docs, it appears that he has the ratio upside down:
>> 
>> https://github.com/Shinmera/precise-time/blob/main/darwin.lisp
>> 
>> 
>> (define-constant MONOTONIC-TIME-UNITS-PER-SECOND
>>     (cffi:with-foreign-objects ((tb :uint32 2))
>>       (if (= 0 (cffi:foreign-funcall "mach_timebase_info" :pointer tb :int))
>>           (let ((ticks-to-nanos (/ (cffi:mem-aref tb :uint32 1) (cffi:mem-aref tb :uint32 0))))
>>             (floor (* 1000000000 ticks-to-nanos)))
>>           (error "Failed to get time scale for monotonic time."))))
>> 
>> On my M1 iMac, mach_timebase_info returns the ratio 125/3. info numerator is at index 0, and denominator is at index 1.
>> 
>> 
>> 
>>> On Oct 9, 2025, at 07:39, Yacin Nadji (as yacin at defmacro dot cc) <[email protected]> wrote:
>>> 
>>>> On Oct 9, 2025, at 5:11 AM, Eugene Zaikonnikov (as ez at norphonic dot com) <[email protected]> wrote:
>>>> 
>>>> Hello,
>>>> 
>>>> Just wonder before I do it with FFI, is there already any support for
>>>> reading monotonic clock? Or perhaps a library that covers LispWorks too?
>>> 
>>> isn’t that what get-internal-real-time does or am i mistaken? if i am or you want a library and more precision maybe check out shinmera’s precise time <https://shinmera.github.io/precise-time/> library. 
>>> 
>>> cheers
>>> 
>>> yacin
>> 
>